the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 29 Jun 2023
Differentiating between crop and soil effects on soil moisture dynamics
Abstract. There is urgent need for developing sustainable agricultural land use schemes. On the one side, climate change is expected to increase drought risk as well as the frequency of extreme precipitation events in many regions. On the other side crop production has induced increased greenhouse gas emissions and enhanced nutrient and pesticide leaching to groundwater and receiving streams. Consequently, sustainable management schemes require sound knowledge of site-specific soil hydrological processes, accounting explicitly for the interplay between soil heterogeneities and crops. Here we present a powerful diagnostic tool applied to a highly diversified arable field with seven different crops and two management schemes. A principal component analysis was applied to a set of 64 soil moisture time series.
About 97 % of the spatial and temporal variance of the data set was explained by the first five principal components. Meteorological drivers accounted for 72 % of the variance. Another 17 % was attributed to different seasonal behaviour of different crops. The effect of very low soil moisture in deeper layers at the onset of the growing season explained another 4.1 %, and soil texture 2.2 %. The fifth component represented the effect of soil depth (1.7 %). In contrast, neither topography nor weed control had a significant effect on soil moisture. Contrary to common expectations, soil and rooting pattern heterogeneity seemed not to play a major role in this case study.
- Preprint
(867 KB) - Metadata XML
- BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on egusphere-2023-1115', Anonymous Referee #1, 12 Jul 2023
This paper describes the application of the well-known principal component analysis method to disentangle effects of crops and soil properties on soil moisture dynamics using 64 soil moisture time series from an agricultural experiment with differently managed small plots. This study is based on a quite large data set of soil moisture measurements and is tangential to an important topic in environmental research. Unfortunately, the interpretations of the results are partly very speculative and difficult to comprehend. Furthermore, transferability of the results to other areas is very limited, as they are determined by the very specific conditions of the experimental study area. I recommend that the authors turn these weaknesses into strengths by arguing that homogeneous soil properties make it easier to study the effects of crop types on soil water balance. The manuscript is mostly well written but need to be checked by a native speaker. I have listed further limitations in my general and specific comments below.
General comments:
The main goal of this study is to disentangle effects of crops and soil properties on soil moisture dynamics. However, the results cannot be generalized due to the peculiarities of the study area. On the one hand, the large vegetation effect observed in this study is due to very specific small-scale crop management with various crops in one field, which does not occur in regular agricultural systems. On the other hand, the soil texture of the studied plots is very similar, so that the minor soil effects on soil moisture found in this study are not representative for landscapes with more typical soil heterogeneity. The similarity in soil texture might also be the reason for the low influence of soil sensor depth and roots on the soil moisture time series.
For the reasons stated above, the title of the manuscript is not appropriate and should instead reflect the very specific conditions of the study area.
The data of the synthetic time series shown in Figures 4, 6, 8, and 10 as well as their interpretations are difficult to understand. To convince readers that the interpretation is robust, these data need to be explained and justified much better.
This study uses data from an underground LoRa-based sensor network. The authors claim that this system is novel, but information on why it is novel is largely lacking. In addition, the soil moisture time series shows large data gaps. The authors provide some general information about data gaps, but do not go into technical detail (e.g. battery failure, transmission failure, sensor failure etc.), which would be interesting given the novelty of the wireless system.
The authors compare “conventional” with “reduced” cases, but in both cases weeds are being controlled. Therefore, not difference between both cases in terms of soil moisture can be expected.
The measured time series of soil moisture should also be presented in meaningful figures, since these form the basis for the statistical analysis. If the number of figures becomes too large, they can also be presented in an appendix.
Specific comments:
L13-15: Combine sentences.
L42: All cited papers didn’t use TDR, but capacitance probes etc.. These kind of low-cost soil moisture sensors are usually used in wireless sensor network applications (see e.g. Bogena et al., 2022). Therefore, I suggest using the more general term “electromagnetic soil moisture sensors”.
L66: Explain in more detail the novelty of this wireless soil moisture monitoring system (please note that are large number of similar systems already exist, see e.g. Bogena et al., 2022)
L83-84: Explain “yield potential zones”.
L95: The “DriBox” is just the housing for the electronics. Please provide information on the manufacturer of the electronic parts.
L97: Does this mean that you have dug 0.9 m deep trenches for the cables? Please explain the installation of the sensors in more detail.
L104: Why was only data from one drone campaign used in this study? Given the high temporal variability of plant and soil water status, the use of a single snapshot may not be sufficiently representative for the conclusions drawn in this analysis.
L117: What is the accuracy of the soil texture prediction model? Please provide more information on the data processing in the appendix.
L118: What do you mean with “gamma sensor” and how does it reduce uncertainty?
L123: Please describe in more detail the technical problems (e.g. transmission failure etc.).
L125: Could you explain why these sensors show frequent malfunctioning (e.g. do to the sensors itself or do the wireless system)?
L125: Define “short”.
L140-141: Was this the case in this study? Otherwise, delete.
L143: Please explain “local effects”.
L158-160: The interpretation that the first PC shows the control of atmospheric forcing should be better justified. For instance, the time series of scores could be correlated with P-ET time series.
L169-173: Move to "Methods" section and expand explanation (e.g., arbitrary factors).
L174-177: These interpretations of Fig. 4 are not clear to me. Maybe I have too little experience with PCA, but I think that other readers see it similarly and also need more explanation.
L186: The direct use of surface temperature (Ts) may not be a very good proxy for ETa. Typically, energy balance models or the warming rates from diurnal Ts measurements are used to infer ETa from Ts (e.g. Panwar et al., 2019). In addition, it is evident from Table 2 that Ts is strongly anticorrelated with NDVI, indicating that the two variables are not independent.
L193: What is meant by this? The soil map does not show any relevant structures.
L194-195: These interpretations are too speculative.
L206-209: These interpretations are not clear to me. Furthermore, the soil texture in the study area is extremely homogeneous, which is why any interpretation of soil effects seems to me to be exaggerated.
L222-223: This statement is not clear to me. Please explain in more detail.
L239-240: Please explain in more detail how you arrive at 61%.
L253-254: This statement needs to be better justified.
L258-259: Too speculative.
L262-263: Too speculative.
L265-268: These interpretations are implausible because the aforementioned effects on soil organic matter take many years to occur.
L272: In this case crop management shapes the environment.
L285: Figure 9.
L286: It is not clear to me why positive loadings should indicate a damped behavior of soil moisture.
L294: In my opinion, this research is not an indispensable prerequisite for tailored field and crop management. In fact, modern sensor-based agricultural techniques allow for a tailored crop management already (e.g. Chamara et al., 2022).
Figures
Fig. 1: Please add horizontal bars for each patch to the figure to make the vegetation stages of the patches easier to understand. In addition, potential ET should be plotted, which is a better proxy for actual ET then air temperature.
Figs. 3 and 7: Use same color scheme as in Fig. 5 to better differentiate the different sensor depths.
References
Bogena, H.R., A. Weuthen and S. Huisman (2022): Recent developments in wireless soil moisture sensing to support scientific research and agricultural management. Sensors 22: 9792. DOI: 10.3390/s22249792
Chamara, N., Islam, M. D., Bai, G. F., Shi, Y., & Ge, Y. (2022). Ag-IoT for crop and environment monitoring: Past, present, and future. Agricultural systems, 203, 103497.
Panwar, A., Kleidon, A., & Renner, M. (2019). Do surface and air temperatures contain similar imprints of evaporative conditions?. Geophysical Research Letters, 46(7), 3802-3809. https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2019GL082248
Citation: https://doi.org/10.5194/egusphere-2023-1115-RC1 -
AC1: 'Reply on RC1', Kathrin Grahmann, 20 Sep 2023
Reviewer 1
This paper describes the application of the well-known principal component analysis method to disentangle effects of crops and soil properties on soil moisture dynamics using 64 soil moisture time series from an agricultural experiment with differently managed small plots. This study is based on a quite large data set of soil moisture measurements and is tangential to an important topic in environmental research. Unfortunately, the interpretations of the results are partly very speculative and difficult to comprehend. Furthermore, transferability of the results to other areas is very limited, as they are determined by the very specific conditions of the experimental study area. I recommend that the authors turn these weaknesses into strengths by arguing that homogeneous soil properties make it easier to study the effects of crop types on soil water balance. The manuscript is mostly well written but need to be checked by a native speaker. I have listed further limitations in my general and specific comments below.
We would like to thank the reviewer for the thorough review. We did our best to meet the comments and recommendations. We added more explanations and details to support the reader in comprehending the interpretation of the data. We agree that in our study soil texture exhibits little heterogeneity and thus the results allow only limited inferences on soil heterogeneity effects. On the other hand, soil homogeneity is not a necessary prerequisite for application of the principal component analysis, and the approach can be used to assess related effects even when they are small. In addition, the term “soil effects” in the title does not only refer to effects of soil heterogeneity but to effects of increasing damping of hydrological signals with increasing soil depth as well.
General comments
The main goal of this study is to disentangle effects of crops and soil properties on soil moisture dynamics. However, the results cannot be generalized due to the peculiarities of the study area. On the one hand, the large vegetation effect observed in this study is due to very specific small-scale crop management with various crops in one field, which does not occur in regular agricultural systems. On the other hand, the soil texture of the studied plots is very similar, so that the minor soil effects on soil moisture found in this study are not representative for landscapes with more typical soil heterogeneity. The similarity in soil texture might also be the reason for the low influence of soil sensor depth and roots on the soil moisture time series.
We reworked the text to emphasize the peculiarities of the study on the one hand, and the wider applicability of the presented approach on the other hand. In terms of minor soil texture heterogeneity please see above.
For the reasons stated above, the title of the manuscript is not appropriate and should instead reflect the very specific conditions of the study area.
Please see comment above.
The data of the synthetic time series shown in Figures 4, 6, 8, and 10 as well as their interpretations are difficult to understand. To convince readers that the interpretation is robust, these data need to be explained and justified much better.
Additional explanations are added to the Methods and Results section.
In the Methods section, we added to the elaboration of how these Figures are produced and how they can be interpreted: “The scores of the principal components constitute time series. Every observed time series can be presented at arbitrary precision as a combination of various principal components. When the data set consists of time series of the same observable measured at different locations, the first principal component describes the mean behaviour inherent in the data set. Subsequent principal components reflect typical modifications of that mean behaviour at single locations due to different effects. Thus generating synthetic time series as linear combinations of the first PC and another additional PC helps to assign this additional PC to a specific effect. To that end scores of that component have either been added to or subtracted from those of the first component using arbitrarily selected factors. The two resulting graphs show how the respective PC causes deviations from the mean behaviour of the data set.“
In the Results section, we added elaboration on how we interpreted the deviations from the mean behaviour.
This study uses data from an underground LoRa-based sensor network. The authors claim that this system is novel, but information on why it is novel is largely lacking.
More explanation on the novelty is added to the manuscript:
“The novelty of this Internet of underground Things (IouT) soil moisture monitoring network is characterized by its unique on-farm installation environment and the deployment of 180 sensors in up to 90 cm soil depth, allowing for high spatio-temporal resolution wireless data transmission, and enabling conventional farming practices like machinery traffic, tillage and mechanical weeding.”
In addition, the soil moisture time series shows large data gaps. The authors provide some general information about data gaps, but do not go into technical detail (e.g. battery failure, transmission failure, sensor failure etc.), which would be interesting given the novelty of the wireless system.
More details are added to the manuscript: “Transmission failures due to discharged batteries, due to signal disturbances in sinks after rainfall or in patches with a high density of biomass (e.g. maize) and theft of parts of the monitoring system led to data gaps that amounted to 81 out of 257 days of the measuring period.”
The authors compare “conventional” with “reduced” cases, but in both cases weeds are being controlled. Therefore, not difference between both cases in terms of soil moisture can be expected.
We differentiate between “conventional” and “reduced” weed control because mechanical weeding impacts soil structure and could enhance soil evaporation which in turn could results in deeper rooting of the plants in contrast to chemical weed control.
The measured time series of soil moisture should also be presented in meaningful figures, since these form the basis for the statistical analysis. If the number of figures becomes too large, they can also be presented in an appendix.
Additional figures can be provided for the appendix.
Specific comments:
L13-15: Combine sentences.
Adjusted in the manuscript.
L42: All cited papers didn’t use TDR, but capacitance probes etc.. These kind of low-cost soil moisture sensors are usually used in wireless sensor network applications (see e.g. Bogena et al., 2022). Therefore, I suggest using the more general term “electromagnetic soil moisture sensors”.
We agree. We changed it accordingly.
L47: This study uses data from an underground LoRa-based sensor network. The authors claim that this system is novel, but information on why it is novel is largely lacking.
This will be highlighted in the end of the introduction: “The novelty of this Internet of underground Things (IouT) soil moisture monitoring network is characterized by its unique on-farm installation environment and the deployment of 180 sensors in up to 90 cm soil depth, allowing for high spatio-temporal resolution wireless data transmission, and enabling conventional farming practices like machinery traffic, tillage and mechanical weeding.”
L66: Explain in more detail the novelty of this wireless soil moisture monitoring system (please note that are large number of similar systems already exist, see e.g. Bogena et al., 2022)
We thank the reviewer for the literature recommendation of Bogena et al. (2022) which we were not aware of as this manuscript was prepared before the publication of that paper.
The system is novel in terms of installation environment and number of installed sensors. Those wireless Lora systems may have been installed and used in the past in other ecosystems, but to the best of our knowledge we do not know about agricultural systems, and in particular one single field that is equipped with 180 sensors providing the information wirelessly in high temporal resolution and hence allow business as usual machine traffic and tillage. We added this justification in the introduction.
L83-84: Explain “yield potential zones”.
We further explained the experimental design of patchCROP and provided a short information on the cluster analysis that has been carried out to define two different yield potential zones in the field. Details on the clustering method are provided in Donat et al. (2022).
L95: The “DriBox” is just the housing for the electronics. Please provide information on the manufacturer of the electronic parts.
We elaborated the technical section and provided all the hardware details: “In each patch, one Dribox box was equipped with a SDI-12 distributer (serial data interface at 1200 baud rate, TBS04, TekBox, Saigon, Vietnam) connected to six TDR-sensors (TDR310H, Acclima, Meridian, USA) and attached to an outdoor remote terminal unit (RTU) fully LoRaWAN compliant (TBS12B: 4+1 channel analogue to SDI-12 interface for 24 Bit A/D conversion of sensor signals, TekBox, Saigon, Vietnam).”
L97: Does this mean that you have dug 0.9 m deep trenches for the cables? Please explain the installation of the sensors in more detail.
We described the installation process more comprehensively and made clear that the soil pit was only 30 to 40 cm deep whereas the 60 and 90 cm sensors were inserted vertically with previously prepared tunnels and tubes that push the sensor into the soil.
L104: Why was only data from one drone campaign used in this study? Given the high temporal variability of plant and soil water status, the use of a single snapshot may not be sufficiently representative for the conclusions drawn in this analysis.
There are no other thermal data available. However, we can include NDVI data from four additional dates (between March 2021 and July 2021) into additional analyses. Results can be added to the manuscript.
L117: What is the accuracy of the soil texture prediction model? Please provide more information on the data processing in the appendix.
The Geophilus system is a service that was purchased to receive the final texture map. Overdrive and sampling have been carried out by the Geophilus company (https://www.gkb-ev.de/publikationen/eip/geophilus.pdf). The model prediction accuracy was provided including gamma and ERa as covariates to predict clay, silt and sand. The additive log ration (ALR) transformation was applied to clay and sand fractions. The best fit was reached with a with Non-linear regression (exponential) model, having a root mean square error of 1.8% for clay, 5.7% for sand and 4.6% for silt. We added that information to the M&M section.
L118: What do you mean with “gamma sensor” and how does it reduce uncertainty?
The gamma sensor is used to detect the natural gamma radiation emitted by the ground. It is emitted mainly by uranium and thorium particles and thus refelcts the proportion of potassium-rich minerals in the clay and silt fraction. Therefore, the measured gamma activity is proportional to the clay content.Because the γ-radiation is less sensitive to soil moisture than the ERa readings, the ratio between the γ-activity and the ERa of the array with the smallest electrode spacing (investigation depth: 0–0.25 m) represents the influence of the soil water on the ERa readings (Bönecke et al., 2021).
Information on the gamma sensor and a new reference were added.
L123: Please describe in more detail the technical problems (e.g. transmission failure etc.).
Information is now provided in the manuscript: “Transmission failures due to discharged batteries, due to signal disturbances in sinks after rainfall or in patches with a high density of biomass (e.g. maize) and theft of parts of the monitoring system led to data gaps that amounted to 81 out of 257 days of the measuring period.”
L125: Could you explain why these sensors show frequent malfunctioning (e.g. do to the sensors itself or do the wireless system)?
Sensors that showed a particularly high frequency of transmission failures were excluded entirely from the study. Unfortunately, it was not possible to determine the exact reason for the high number of errors for specific sensors. Possible reasons could be: Technical failures of individual sensors; transmission failures between sensor and node box due to e.g. cable damage; overlapping of different effects already described that weaken the RSSI signal. At the latter it must be considered that all sensors at a specific patch are connected to the same node box. Thus, if data from other sensors at the same patch were transmitted, problems with individual sensors are more likely to be the reason for the data gaps than transmission errors between the node box and the gateway.
L125: Define “short”.
Details were added to the manuscript: “Of all 20668 interpolated gaps, 96 % were shorter than two hours, 3 % between two and six hours and 1 % longer than six hours. In 26 cases, the gap exceeded the duration of one day.”
L140-141: Was this the case in this study? Otherwise, delete.
All analysed PC had an eigenvalue greater than one.
L143: Please explain “local effects”.
This part of the methodology was not necessarily important for the manuscript and was therefore deleted.
L158-160: The interpretation that the first PC shows the control of atmospheric forcing should be better justified. For instance, the time series of scores could be correlated with P-ET time series.
The correlation between the scores and the cumulative climatic water balance (P-ETp) is -0.97. The information was added to the manuscript.
L169-173: Move to "Methods" section and expand explanation (e.g., arbitrary factors).
Moved to the Methods section and expanded explanation added in the manuscript (see general comment on Figures 4, 6, 8 and 10).
L174-177: These interpretations of Fig. 4 are not clear to me. Maybe I have too little experience with PCA, but I think that other readers see it similarly and also need more explanation.
We added additional explanations in the Methods and Results sections (see comment above).
L186: The direct use of surface temperature (Ts) may not be a very good proxy for ETa. Typically, energy balance models or the warming rates from diurnal Ts measurements are used to infer ETa from Ts (e.g. Panwar et al., 2019). In addition, it is evident from Table 2 that Ts is strongly anticorrelated with NDVI, indicating that the two variables are not independent.
Diurnal data were not available as the drone images provided only a single snapshot in time. Instead, the spatial pattern of surface temperature was deemed to be related to that of actual evapotranspiration in a monotonic, although not necessarily linear way. Close anti-correlation of the resulting pattern with that of NDVI provided some evidence that this approach was justified.
L193: What is meant by this? The soil map does not show any relevant structures.
We clarified the statement: “Although the affected patches do not correspond to anomalies in the soil map, it is still apparent that the location of the patches roughly follows an east-west direction.”
L194-195: These interpretations are too speculative.
We rephrased to better describe the effect: “The most obvious difference between the orange line (negative loading on PC3) and the blue line (positive loading on PC3) during the first half of the study period is that the latter reaches a maximum of soil moisture after rainfall much earlier compared to the former (Figure 6).”
Thereby, in combination with additional elaboration in the Discussion section, we hope to support the reader in comprehending the interpretation of this PC: “Loadings on the third principal component were not related to crop types. In contrast, a spatial pattern emerged: Only sensors from 0.9 m depth from six adjacent patches exhibited strongly negative loadings (Figure 2) whereas all other sensors showed minor positive or negative loadings. This points to an effect of subsoil substrates, that is higher loam content and consequently higher water holding capacity. That would be consistent with delayed response to seepage fluxes and reduced desiccation in the vegetation period (Figure 6).”
L206-209: These interpretations are not clear to me. Furthermore, the soil texture in the study area is extremely homogeneous, which is why any interpretation of soil effects seems to me to be exaggerated.
The statement has been refined to clarify that we do not refer to the soil as loamy but describe the development over time of the orange graph as behaviour which is typical for loamy soils: “Figure 8 illustrates the effect of the fourth PC on time series. A positive factor would be typical for more sandy soils and for patches with fallow in autumn and winter (blue line). In contrast the orange line depicts behaviour in more loamy soils and for winter crops. The latter line exhibits slightly more delayed responses to rainstorms and subsequent less steep recovery as would be expected for more loamy soils. However, it is not clear how winter crops on the one side and fallow on the other side could induce such a different behaviour.”
L222-223: This statement is not clear to me. Please explain in more detail.
The statement has been re-formulated. We want to express that our analyses revealed various effects of soil texture, soil depth, crops and management.
L239-240: Please explain in more detail how you arrive at 61%.
We added additional explanations: “When not considering the temporal component reflected by PC1 and thus only looking at the spatial variability, 61% of the remaining variance (attributed to PC2 to PC64) is caused by the vegetation effect reflected by PC2.”
L253-254: This statement needs to be better justified.
The scores are time series and reflect the effect size of a particular process represented by the respective PC. The more the scores of a certain PC deviate from zero during single periods, the stronger the respective effect is. Consequently, the development of the time series of PC2 scores – strongly varying and having an amplitude greater than 20 – indicates that the effect of vegetation on total variability varies by time.
L258-259: Too speculative.
We elaborated a little bit more on that but emphasizing that these are very preliminary inferences, based on own observations and similar observations made by other colleagues (e.g., Döring et al., in preparation).
L262-263: Too speculative.
See comment above and following comment.
L265-268: These interpretations are implausible because the aforementioned effects on soil organic matter take many years to occur.
See comment above and reply to general comment of RC2: “The interpretation of the fourth principal component is consistent with own observations and similar observations made by other colleagues (e.g., Döring et al., in preparation). Effects of changing soil organic carbon quantity and quality are assumed to occur only at larger time scales which is closely related to the problem of detecting respective changes within shorter periods. However, that might be more a problem of detectability rather than a sound disproof of the suggested mechanism. We think more research is needed here, including but not being restricted to indirect methods like that used in our studies.”
L272: In this case crop management shapes the environment.
We agree and we adjusted the respective phrase in the manuscript.
L285: Figure 9.
Thank you, of course Figure 9 should be referenced.
L286: It is not clear to me why positive loadings should indicate a damped behavior of soil moisture.
The statement has been elaborated a little bit more: “Loadings on this component are clearly related with depth (Figure 9). Strong positive loadings indicate a strongly damped behaviour of soil moisture time series: The blue line, representing sites with positive loadings on PC5 which is typical for sensors at greater depth (Figure 9) exhibits clearly reduced amplitudes compared to the yellow line, that is, sensors at shallow depth (Figure 9, Figure 10).”
In combination with information on how Figures 4, 6, 8, and 10 are derived and how they can be interpreted, we hope that readers can now follow our interpretations.
L294: In my opinion, this research is not an indispensable prerequisite for tailored field and crop management. In fact, modern sensor-based agricultural techniques allow for a tailored crop management already (e.g. Chamara et al., 2022).
The statement relates to disentangling and quantifying different effects in general, not specifically to the suggested approach. We consider the latter very helpful in addition to modern sensor systems..
Figures
Fig. 1: Please add horizontal bars for each patch to the figure to make the vegetation stages of the patches easier to understand. In addition, potential ET should be plotted, which is a better proxy for actual ET then air temperature.
The figure can be adjusted accordingly.
Figs. 3 and 7: Use same color scheme as in Fig. 5 to better differentiate the different sensor depths.
The figures can be adjusted accordingly.
References
Bogena, H.R., A. Weuthen and S. Huisman (2022): Recent developments in wireless soil moisture sensing to support scientific research and agricultural management. Sensors 22: 9792. DOI: 10.3390/s22249792
Chamara, N., Islam, M. D., Bai, G. F., Shi, Y., & Ge, Y. (2022). Ag-IoT for crop and environment monitoring: Past, present, and future. Agricultural systems, 203, 103497.
Panwar, A., Kleidon, A., & Renner, M. (2019). Do surface and air temperatures contain similar imprints of evaporative conditions?. Geophysical Research Letters, 46(7), 3802-3809. https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2019GL082248
Citation: https://doi.org/10.5194/egusphere-2023-1115-AC1
-
AC1: 'Reply on RC1', Kathrin Grahmann, 20 Sep 2023
-
RC2: 'Comment on egusphere-2023-1115', Tobias L. Hohenbrink, 28 Jul 2023
Summary
In the study “Differentiating between crop and soil effects on soil moisture dynamics” by Helen Scholz et al. 64 soil moisture time series covering eight months are evaluated by a principal component analysis. The data have been measured in three depths at a site in Eastern Germany with a wireless network of TDR sensors. The resulting components were interpreted based on supporting information about (i) precipitation and temperature, (ii) crop rotation, (iii) sand content in the upper 25 cm, and (iv) NDVI and surface temperature. A share of 97 % of total soil moisture variance could be described by the first five components and has been assigned to meteorological conditions (27%), the cropping system (17 %), soil properties (6,3 %), and signal damping (1.7 %).
General commentsObjectives of the study:
The research question addressed in the study (L66-70) is generally relevant and also interesting for the readers of HESS. It should be defined more precisely what exactly is meant by “highly diversified fields” in this study. It might also be unclear at first what “quantify the drivers of soil moisture” really means. The readers might first think about quantifying the individual components of the hydrological water balance by absolute values. However, due to the z-transformation, this cannot be achieved with a PCA. The objectives should be formulated more precisely.
Methods:
PCA of soil moisture time series is a promising approach to identify the dominating factors of soil moisture dynamics and assess the strength of their effects. It is not a new approach, since some very similar studies already exist, where a PCA has been applied to soil moisture time series. However, this should not be a problem for a publication in HESS, because we can still learn a lot from repeating the analyses at new sites. The main methodological problem I see in the study is that extensive and robust data are needed to identify interpretable patterns with the PCA approach, which are important to draw valid conclusions about thematic research questions. Unfortunately, quite limited data were considered in this study.
Analysed Data:
Only a very short period of eight months of soil moisture measurements have been analyzed. These time series additionally contained large data gaps, unfortunately during interesting times: (i) the period during steady rain mid of May, and (ii) the three weeks after the strong rain in July. Unfortunately, the data gaps meet particularly interesting situations where soil moisture information would have been very important to learn about the hydrological functioning at the site. The study would be improved strongly, when soil moisture data for a longer time period could be included. Maybe moisture time series of higher quality have been measured in the subsequent growing period.
The available soil texture information only contains sand contents in the upper 25 cm derived from geoelectric exploration.This information is poorly suited for process interpretations, because the sand content at the TDR-sensor positions varies in a very small range of only 3 % (between 77.9% and 80.7% ,Table 1), which might even be close to the uncertainties of the geoelectrical method. There are a lot of other potential factors determining the soil hydraulic properties (e,g, clay content, bulk density, organic carbon content, etc.), which have not been taken into account in this study. I think that this marginal variance in sand content cannot be used alone to explain the soil moisture patterns identified by principal components. When single components shall be related to soil texture, more texture information from all considered soil depths is needed. Therefore, I highly recommend going back to the field, taking new soil samples (e.g with a small hand auger or a gouge auger) and determining their sand silt and clay contents.
Findings, interpretations and conclusions:
The 1st, 2nd and 5th principal components could be related to reasonable controlling factors and the process interpretations also seem plausible. This does not apply to the third and fourth components. The interpretations of these components are not based on solid data.
I assume that either the information actually needed to interpret these PCs is not available, or that the PCA fails to provide clearly interpretable components here. The weak interpretation of the third and fourth components should be discussed in more detail. In general, there should be more discussion of the suitability of the available data for principal component interpretation.
Minor commentsL30-32: Please, provide some more references for the effects listed.
L33: What is exactly meant by “complexity of the assessment and monitoring ”. What shall be assessed and why?
L47-50: “Soil moisture variograms” are a poor example for “sophisticated data analysis approaches”, because they are very simple. Please rephrase or find another example.
L55-57: The concept of “temporal stability” was introduced by Vachaud (1985) (https://doi.org/10.2136/sssaj1985.03615995004900040006x) which should be acknowledged with a citation. The review by Vanderlinden et al. (2012) (https://doi.org/10.2136/vzj2011.0178) also seems to be a very suitable reference here.
L64: The term “highly diversified fields” should be defined more exactly.
L83-84: What is a “yield potential zone”?
Table 1: What is meant by “treatment”? Readers might think about pest control or soil tillage. Maybe you can find another term.
Table 1: The “highly heterogeneous soils” (L75) are not reflected in the sand content listed in the Table. They vary only in a range of 3%. Therefore, I expect that they cannot explain large parts of the soil moisture variance. The clay content would be much more interesting here.
L94-98: The technical description should be improved. What do the “node boxes do”? How are the TDR sensors connected to the node boxes?
L102: How have the meteorological data been measured?
L111: Which physical variable is meant by “near infrared” and the red band? The intensity? or a relative share?
L124: I really regret (i) that the considered time periods are so short and (ii) that the data gaps occur during the most interesting periods. I see this as one of the biggest problems in this study. Is it possible to extend the period or maybe use other data from the following growing period?
L128-130: Please explain the implications of the z-transformation. Readers have to know that the z-transformation has to be kept in mind when interpreting the scores of a PC.
L140-141: Please rephrase the explanation of the criterion by Kaiser (1960). Eigenvalues greater than one indicate that a PC explains more variance than one input time series can contribute to the total variance of the entire input data set.
L143-145: I don’t understand what has been done here and why. Please provide more information.
L156-161: Please mention in half a sentence why the scores and loadings of the first PC are not shown here in the manuscript.
L183-189: It is very difficult to follow and to understand the effects and potential causal relations that are described here. For example: Soil temperature is negatively correlated with the loadings of PC 2 which in turn indicate a negative (summer crops) and positive (winter crops) correlation between the moisture time series and the scores of PC 2. I am sure that most readers (including me) need a better explanation of these dependencies. They need to be better guided in order not to get lost.
Figure 4: What about harvesting? In August the winter crops (blue line) have constant scores (indicating stopped transpiration after harvesting?) while the scores describing moisture dynamics for summer crops (red line) are still decreasing (ongoing transpiration?). Unfortunately there is a data gap.
190-195: It is hard to follow the description of the third PC. I have the feeling that in the third PC the effects of several factors interact. Perhaps the relevant supporting information to understand PC 3 is simply not known. If the authors are really confident in their interpretation of the third PC, they should describe the relationships more clearly. If they are skeptical, as I am, they should discuss these problems in detail.
L203-205: Are the correlations with the sand contents not shown? As mentioned earlier, I don’t think that the sand content can explain any variance due to its small variation.
L203-209: It is rather difficult to interpret the effects of two different factors (cropping system and sand content of upper 25 cm) in PC 4, which explains only 2.2% of the total variance.
L217: Please check if it should be lupine instead of sunflower.
L222-223: I don’t really know what is meant here. Is redundancy here the correct term?
L232: “quantification of the strength of these effects” might be more precise
L247-250: Please check if Yang et al. (2015) have also z-transformed their data. If not it might be difficult to compare their findings with those of this study.
L265: What do you mean by loamy soils? I think that all soils at the site are sandy soils.
L265-267: Very speculative. I think that an increase of carbon stock happens at larger time scales and can unlikely explain the moisture patterns explained by PC 4.
L274-291: I can imagine that soil texture is an important factor controlling soil moisture dynamics at the investigated site. However, as mentioned before, more information about the depth distribution of soil texture is needed. If it is planned to run the “patchCROP” experiment for longer, it is really worth going back to the field, collecting soil samples at each TDR sensor position in 30, 60, and 90 cm depth and performing a texture analysis.
L296: I agree that it is important to study the interaction of different factors in their effect on soil moisture dynamics. Unfortunately, in these interactions, the patterns identified by a PCA often become blurred, making interpretation difficult with the usually limited supporting information available.
L304-305: I agree, but is that conclusion really founded on the findings of this study? The sentence could also be shifted to the introduction.
L307-309: This paragraph might be shifted to the discussion section.
Citation: https://doi.org/10.5194/egusphere-2023-1115-RC2 -
AC2: 'Reply on RC2', Kathrin Grahmann, 20 Sep 2023
Reviewer 2
Summary
In the study “Differentiating between crop and soil effects on soil moisture dynamics” by Helen Scholz et al. 64 soil moisture time series covering eight months are evaluated by a principal component analysis. The data have been measured in three depths at a site in Eastern Germany with a wireless network of TDR sensors. The resulting components were interpreted based on supporting information about (i) precipitation and temperature, (ii) crop rotation, (iii) sand content in the upper 25 cm, and (iv) NDVI and surface temperature. A share of 97 % of total soil moisture variance could be described by the first five components and has been assigned to meteorological conditions (27%), the cropping system (17 %), soil properties (6,3 %), and signal damping (1.7 %).
Thanks for the comprehensive and in-depth review.
General commentsObjectives of the study:
The research question addressed in the study (L66-70) is generally relevant and also interesting for the readers of HESS. It should be defined more precisely what exactly is meant by “highly diversified fields” in this study. It might also be unclear at first what “quantify the drivers of soil moisture” really means. The readers might first think about quantifying the individual components of the hydrological water balance by absolute values. However, due to the z-transformation, this cannot be achieved with a PCA. The objectives should be formulated more precisely.
We did our best to clarify information on the objectives (Abstract, Introduction) and on the details of the study.
Diversification of agricultural systems can be implemented and reached through spatial and temporal approaches. In patchCROP we combined both and designed a completely new cropping system design with a high level of diversification in terms of crops, soil management zones, field size and land use intensity (in terms of plant protection). The changing soil-hydrological dynamics in complex diversified agricultural systems with increasing heterogeneity and site-specific adjustment of crops, soil types and field management which have hardly been studied so far.
We added to the Methods section the limitations of the analysis of z-transformed data sets regarding absolute values.
Methods:
PCA of soil moisture time series is a promising approach to identify the dominating factors of soil moisture dynamics and assess the strength of their effects. It is not a new approach, since some very similar studies already exist, where a PCA has been applied to soil moisture time series. However, this should not be a problem for a publication in HESS, because we can still learn a lot from repeating the analyses at new sites. The main methodological problem I see in the study is that extensive and robust data are needed to identify interpretable patterns with the PCA approach, which are important to draw valid conclusions about thematic research questions. Unfortunately, quite limited data were considered in this study.
We agree that long and gapless time series would be ideal for any in-depth analysis. However, such data sets are often not available. Fortunatley though PCA can be applied and the results be interpreted despite data gaps. Therefore, we consider the methodology suitable for many real-world monitoring setups.
Analysed Data:
Only a very short period of eight months of soil moisture measurements have been analyzed. These time series additionally contained large data gaps, unfortunately during interesting times: (i) the period during steady rain mid of May, and (ii) the three weeks after the strong rain in July. Unfortunately, the data gaps meet particularly interesting situations where soil moisture information would have been very important to learn about the hydrological functioning at the site. The study would be improved strongly, when soil moisture data for a longer time period could be included. Maybe moisture time series of higher quality have been measured in the subsequent growing period.
We agree in terms of the detrimental long data gap. Still, other important and characteristic time periods of the year were covered, such as the moist winter months with subsequent rain falls in end of January and in February and the dry weeks in June. On the other hand, though, considering longer time series beyond the length of a single cropping period would cause another problem inasmuch as effects of different crops would mix up in the soil moisture readings of single sites. Thus, identification of crop-related effects would hardly be feasible.
The available soil texture information only contains sand contents in the upper 25 cm derived from geoelectric exploration. This information is poorly suited for process interpretations, because the sand content at the TDR-sensor positions varies in a very small range of only 3 % (between 77.9% and 80.7% ,Table 1), which might even be close to the uncertainties of the geoelectrical method. There are a lot of other potential factors determining the soil hydraulic properties (e,g, clay content, bulk density, organic carbon content, etc.), which have not been taken into account in this study. I think that this marginal variance in sand content cannot be used alone to explain the soil moisture patterns identified by principal components. When single components shall be related to soil texture, more texture information from all considered soil depths is needed. Therefore, I highly recommend going back to the field, taking new soil samples (e.g with a small hand auger or a gouge auger) and determining their sand silt and clay contents.
Sand was varying a lot at the field scale between 69.1 and 81.2% at the site, but little within patches. Clay and silt estimates are available from Geophilus and can be further analysed and added to this manuscript.
In the meantime, additional data were provided. They are manual soil auger results until 1 m depth available from project activities in the DFG excellence cluster PhenoRob for eight out of 12 analysed patches. This information can be also incorporated in further analyses.
But even then, we agree that in our study soil texture exhibits little heterogeneity and thus the results allow only limited inferences on soil heterogeneity effects. On the other hand, soil homogeneity is not a necessary prerequisite for application of the principal component analysis, and the approach can be used to assess related effects even when they are small. In addition, the term “soil effects” in the title does not only refer to effects of soil heterogeneity but to effects of increasing damping of hydrological signals with increasing soil depth as well.
Findings, interpretations and conclusions:
The 1st, 2nd and 5th principal components could be related to reasonable controlling factors and the process interpretations also seem plausible. This does not apply to the third and fourth components. The interpretations of these components are not based on solid data.
I assume that either the information actually needed to interpret these PCs is not available, or that the PCA fails to provide clearly interpretable components here. The weak interpretation of the third and fourth components should be discussed in more detail. In general, there should be more discussion of the suitability of the available data for principal component interpretation.
We elaborated and refined our reasoning in terms of the third and fourth component. We agree that these arguments are far from unequivocal proofs. But we consider it worthwhile to consider even unexpected results. E.g., the interpretation of the fourth principal component is consistent with own observations and similar observations made by other colleagues (e.g., Döring et al., in preparation). Effects of changing soil organic carbon quantity and quality are assumed to occur only at larger time scales which is closely related to the problem of detecting respective changes within shorter periods. However, that might be more a problem of detectability rather than a sound disproof of the suggested mechanism. We think more research is needed here, including but not being restricted to indirect methods like that used in our studies.
Minor commentsL30-32: Please, provide some more references for the effects listed.
Additional references were included:
- Fischer, C., Roscher, C., Jensen, B., Eisenhauer, N., Baade, J., Attinger, S., Scheu, S., Weisser, W. W., Schumacher, J., Hildebrandt, A.: How Do Earthworms, Soil Texture and Plant Composition Affect Infiltration along an Experimental Plant Diversity Gradient in Grassland?, PLos ONE, 9, 6, https://doi.org/10.1371/journal.pone.0098987, 2014.
- Koudahe, K., Allen, S. C., Djaman, K.: Critical review of the impact of cover crops on soil properties, International Soil and Water Conservation Research, 10, 343-354, https://doi.org/10.1016/j.iswcr.2022.03.003, 2022.
- Nunes, M. R., van Es, H. M., Schindelbeck, R., Ristow, A. J., Ryan, M.: No-till and cropping system diversification improve soil health and crop yield, Geoderma, 328, 30-43, https://doi.org/10.1016/j.geoderma.2018.04.031, 2018.
L33: What is exactly meant by “complexity of the assessment and monitoring ”. What shall be assessed and why?
The more independent variables are present in agricultural systems, the higher the demand for frequency and spacings of soil moisture measurement / related data. We revised the phrase.
L47-50: “Soil moisture variograms” are a poor example for “sophisticated data analysis approaches”, because they are very simple. Please rephrase or find another example.
We rephrased the formulation: “Methods include geostatistical analysis (Vereecken et al., 2014) or data driven approaches (Hong et al., 2016).” Examples for more sophisticated approaches will be given in the following sentence.
L55-57: The concept of “temporal stability” was introduced by Vachaud (1985) (https://doi.org/10.2136/sssaj1985.03615995004900040006x) which should be acknowledged with a citation. The review by Vanderlinden et al. (2012) (https://doi.org/10.2136/vzj2011.0178) also seems to be a very suitable reference here.
Thank you for the valuable note, the references were added to the manuscript.
L64: The term “highly diversified fields” should be defined more exactly.
The term has been defined more clearly, making clear that it refers to the multitude of different crops and management schemes within a single arable field (see general comments).
L83-84: What is a “yield potential zone”?
We further explained the experimental design of the experimental field and provided a short information on the cluster analysis that has been carried out to define two different yield potential zones within in the field. A reference is given in the text (Donat et al. 2022).
Table 1: What is meant by “treatment”? Readers might think about pest control or soil tillage. Maybe you can find another term.
We decided to re-name this column to “crop groups”. Crop group A contains winter crops, crop group B contains fallow (in winter), followed by summer crops and crop group C contains cover crops, followed by summer crops.
Table 1: The “highly heterogeneous soils” (L75) are not reflected in the sand content listed in the Table. They vary only in a range of 3%. Therefore, I expect that they cannot explain large parts of the soil moisture variance. The clay content would be much more interesting here.
At the study site the sand content in the upper layer varied between 69 % and 81 %. However, the variability in the analysed patches was indeed low. Information on clay content, which is in the meantime also available for deeper layers of eight out of 12 patches, can be used for further analysis. Results can be added to the manuscript.
L94-98: The technical description should be improved. What do the “node boxes do”? How are the TDR sensors connected to the node boxes?
We elaborated the technical description of the sensor system and provide all hardware details (see reply to RC1).
L102: How have the meteorological data been measured?
This information has been added. We had two meteorological stations on site.
L111: Which physical variable is meant by “near infrared” and the red band? The intensity? or a relative share?
Details on the values used for calculation were added (near infrared as light reflected by vegetation and red as absorbed by vegetation).
L124: I really regret (i) that the considered time periods are so short and (ii) that the data gaps occur during the most interesting periods. I see this as one of the biggest problems in this study. Is it possible to extend the period or maybe use other data from the following growing period?
We agree in terms of the detrimental long data gap. Still, other important and characteristic time periods of the year were covered, such as the moist winter months with subsequent rain falls in end of January and in February and the dry weeks in June. On the other hand, though, considering longer time series beyond the length of a single cropping period would cause another problem inasmuch as effects of different crops would mix up in the soil moisture readings of single sites. Thus, identification of crop-related effects would hardly be feasible.
L128-130: Please explain the implications of the z-transformation. Readers have to know that the z-transformation has to be kept in mind when interpreting the scores of a PC.
We added to the manuscript that due to the z-transformation absolute values of soil moisture and thus absolute changes cannot be interpreted or explained by PCA.
L140-141: Please rephrase the explanation of the criterion by Kaiser (1960). Eigenvalues greater than one indicate that a PC explains more variance than one input time series can contribute to the total variance of the entire input data set.
We gladly replace the original version with the suggestion of the reviewer.
L143-145: I don’t understand what has been done here and why. Please provide more information.
This part of the methodology was not necessarily important for the manuscript and was therefore deleted.
L156-161: Please mention in half a sentence why the scores and loadings of the first PC are not shown here in the manuscript.
Since the loadings on the first PC were all one-directional, the graphic was not shown. However, it can be provided in the appendix.
L183-189: It is very difficult to follow and to understand the effects and potential causal relations that are described here. For example: Soil temperature is negatively correlated with the loadings of PC 2 which in turn indicate a negative (summer crops) and positive (winter crops) correlation between the moisture time series and the scores of PC 2. I am sure that most readers (including me) need a better explanation of these dependencies. They need to be better guided in order not to get lost.
The paragraph has been re-formulated: “As shown in Table 3, the NDVI as a proxy for photosynthesis potential was positively correlated with the loadings. Surface temperature exhibited a negative correlation. On the other hand, the spatial pattern of surface temperature is assumed to be inversely related to that of actual evapotranspiration. Thus, both proxies, NDVI and surface temperature, support the inference that positive loadings on this principal component represent sites with above-average plant activity and root water uptake.”
Figure 4: What about harvesting? In August the winter crops (blue line) have constant scores (indicating stopped transpiration after harvesting?) while the scores describing moisture dynamics for summer crops (red line) are still decreasing (ongoing transpiration?). Unfortunately there is a data gap.
We agree, this effect can be attributed to the earlier harvesting of winter crops. We will add this observation to the description of the Figure.
190-195: It is hard to follow the description of the third PC. I have the feeling that in the third PC the effects of several factors interact. Perhaps the relevant supporting information to understand PC 3 is simply not known. If the authors are really confident in their interpretation of the third PC, they should describe the relationships more clearly. If they are skeptical, as I am, they should discuss these problems in detail.
By providing supplementary explanations for Figures 4, 6, 8, and 10, we hope that our interpretations can be better followed. It should be better illustrated that in Figure 6 two different types of drainage behaviour are shown. Due to the local, non-systematic occurrence of particularly pronounced loadings we attribute this PC to soil properties.
L203-205: Are the correlations with the sand contents not shown? As mentioned earlier, I don’t think that the sand content can explain any variance due to its small variation.
The correlation with sand content of loadings of other loadings were weak and thus not shown (0.18, 0.22, -0.36, -0.26 for PC1, PC2, PC3 and PC5, respectively).
As previously explained, more data on texture are available now for part of the analysed patches and can be used for further analysis.
We also refer to our answer to general comments of the first reviewer: “We agree that in our study soil texture exhibits little heterogeneity and thus the results allow only limited inferences on soil heterogeneity effects. On the other hand, soil homogeneity is not a necessary prerequisite for application of the principal component analysis, and the approach can be used to assess related effects even when they are small.”
L203-209: It is rather difficult to interpret the effects of two different factors (cropping system and sand content of upper 25 cm) in PC 4, which explains only 2.2% of the total variance.
See comment above: “We agree that our arguments are far from unequivocal proofs. But we consider it worthwhile to consider even unexpected results. Our preliminary interpretation of the fourth principal component is consistent with own observations and similar observations made by other colleagues (e.g., Döring et al., in preparation). Effects of changing soil organic carbon quantity and quality are assumed to occur only at larger time scales which is closely related to the problem of detecting respective changes within shorter periods. However, that might be more a problem of detectability rather than a sound disproof of the suggested mechanism. We think more research is needed here, including but not being restricted to indirect methods like that used in our studies.”
L217: Please check if it should be lupine instead of sunflower.
Thank you for the valuable remark. It is indeed lupine.
L222-223: I don’t really know what is meant here. Is redundancy here the correct term?
We revised the wording. We want to express that our analyses revealed various effects of soil texture, soil depth, crops and management.
L232: “quantification of the strength of these effects” might be more precise
We revised the wording into “quantification of the impact of these effects”
L247-250: Please check if Yang et al. (2015) have also z-transformed their data. If not it might be difficult to compare their findings with those of this study.
Since no z-transformed data set was used in the reference and the type of vegetation in the referenced study also differed, we decided not to make a comparison to the results of this study.
L265: What do you mean by loamy soils? I think that all soils at the site are sandy soils.
The phrase has been re-formulated: “According to this component, soil moisture dynamics at the fallow patches resembled more the typical behaviour one would expect for sandy soils, and that of winter crop patches more a more damped behaviour typical for more loamy soils.”
L265-267: Very speculative. I think that an increase of carbon stock happens at larger time scales and can unlikely explain the moisture patterns explained by PC 4.
See comment above:
“We elaborated and refined our reasoning in terms of the third and fourth component. We agree that these arguments are far from unequivocal proofs. But we consider it worthwhile to consider even unexpected results. E.g., the fourth principal component is consistent with own observations and similar observations made by other colleagues (e.g., Döring et al., in preparation). Effects of changing soil organic carbon quantity and quality are assumed to occur only at larger time scales which is closely related to the problem of detecting respective changes within shorter periods. However, that might be more a problem of detectability rather than a sound disproof of the suggested mechanism. We think more research is needed here, including but not being restricted to indirect methods like that used in our studies.”
L274-291: I can imagine that soil texture is an important factor controlling soil moisture dynamics at the investigated site. However, as mentioned before, more information about the depth distribution of soil texture is needed. If it is planned to run the “patchCROP” experiment for longer, it is really worth going back to the field, collecting soil samples at each TDR sensor position in 30, 60, and 90 cm depth and performing a texture analysis.
Soil texture has been determined manually and in the laboratory through research project activities in the DFG excellence cluster PhenoRob. We will be able to use those data for further additional interpretation which can be added to this manuscript as a new additional sampling campaign is not feasible due to long laboratory waiting times.
L296: I agree that it is important to study the interaction of different factors in their effect on soil moisture dynamics. Unfortunately, in these interactions, the patterns identified by a PCA often become blurred, making interpretation difficult with the usually limited supporting information available.
We consider PCA a powerful tool in this regard, although only just another step on the way to develop diagnostic tools for complex real-world systems. We added a corresponding statement: “Principal component analysis is a further step to meet these challenges although not entirely without problems.”
L304-305: I agree, but is that conclusion really founded on the findings of this study? The sentence could also be shifted to the introduction.
The phrasing was revised to highlight the connection between the study and this statement: “In particular, the plant-induced effects on soil hydraulic properties would be worthwhile to be studied in more detail. Knowledge from data-driven approaches can support adequate crop selection as a management option to encounter the increasing drought risk in the study region.”
L307-309: This paragraph might be shifted to the discussion section.
The phrasing was revised to highlight the potential of such analyses as one of the conclusions drawn from this study: “Information from this study will contribute to elucidate management effects as well as to develop both parsimonious and tailored mechanistic models. Findings of this study highly depend on local conditions. However, we consider the presented approach generally applicable to a large range of site conditions. In this regard, principal component analysis of soil moisture time series performed as a powerful diagnostic tool and is highly recommended.”
Citation: https://doi.org/10.5194/egusphere-2023-1115-AC2
-
AC2: 'Reply on RC2', Kathrin Grahmann, 20 Sep 2023
Status: closed
-
RC1: 'Comment on egusphere-2023-1115', Anonymous Referee #1, 12 Jul 2023
This paper describes the application of the well-known principal component analysis method to disentangle effects of crops and soil properties on soil moisture dynamics using 64 soil moisture time series from an agricultural experiment with differently managed small plots. This study is based on a quite large data set of soil moisture measurements and is tangential to an important topic in environmental research. Unfortunately, the interpretations of the results are partly very speculative and difficult to comprehend. Furthermore, transferability of the results to other areas is very limited, as they are determined by the very specific conditions of the experimental study area. I recommend that the authors turn these weaknesses into strengths by arguing that homogeneous soil properties make it easier to study the effects of crop types on soil water balance. The manuscript is mostly well written but need to be checked by a native speaker. I have listed further limitations in my general and specific comments below.
General comments:
The main goal of this study is to disentangle effects of crops and soil properties on soil moisture dynamics. However, the results cannot be generalized due to the peculiarities of the study area. On the one hand, the large vegetation effect observed in this study is due to very specific small-scale crop management with various crops in one field, which does not occur in regular agricultural systems. On the other hand, the soil texture of the studied plots is very similar, so that the minor soil effects on soil moisture found in this study are not representative for landscapes with more typical soil heterogeneity. The similarity in soil texture might also be the reason for the low influence of soil sensor depth and roots on the soil moisture time series.
For the reasons stated above, the title of the manuscript is not appropriate and should instead reflect the very specific conditions of the study area.
The data of the synthetic time series shown in Figures 4, 6, 8, and 10 as well as their interpretations are difficult to understand. To convince readers that the interpretation is robust, these data need to be explained and justified much better.
This study uses data from an underground LoRa-based sensor network. The authors claim that this system is novel, but information on why it is novel is largely lacking. In addition, the soil moisture time series shows large data gaps. The authors provide some general information about data gaps, but do not go into technical detail (e.g. battery failure, transmission failure, sensor failure etc.), which would be interesting given the novelty of the wireless system.
The authors compare “conventional” with “reduced” cases, but in both cases weeds are being controlled. Therefore, not difference between both cases in terms of soil moisture can be expected.
The measured time series of soil moisture should also be presented in meaningful figures, since these form the basis for the statistical analysis. If the number of figures becomes too large, they can also be presented in an appendix.
Specific comments:
L13-15: Combine sentences.
L42: All cited papers didn’t use TDR, but capacitance probes etc.. These kind of low-cost soil moisture sensors are usually used in wireless sensor network applications (see e.g. Bogena et al., 2022). Therefore, I suggest using the more general term “electromagnetic soil moisture sensors”.
L66: Explain in more detail the novelty of this wireless soil moisture monitoring system (please note that are large number of similar systems already exist, see e.g. Bogena et al., 2022)
L83-84: Explain “yield potential zones”.
L95: The “DriBox” is just the housing for the electronics. Please provide information on the manufacturer of the electronic parts.
L97: Does this mean that you have dug 0.9 m deep trenches for the cables? Please explain the installation of the sensors in more detail.
L104: Why was only data from one drone campaign used in this study? Given the high temporal variability of plant and soil water status, the use of a single snapshot may not be sufficiently representative for the conclusions drawn in this analysis.
L117: What is the accuracy of the soil texture prediction model? Please provide more information on the data processing in the appendix.
L118: What do you mean with “gamma sensor” and how does it reduce uncertainty?
L123: Please describe in more detail the technical problems (e.g. transmission failure etc.).
L125: Could you explain why these sensors show frequent malfunctioning (e.g. do to the sensors itself or do the wireless system)?
L125: Define “short”.
L140-141: Was this the case in this study? Otherwise, delete.
L143: Please explain “local effects”.
L158-160: The interpretation that the first PC shows the control of atmospheric forcing should be better justified. For instance, the time series of scores could be correlated with P-ET time series.
L169-173: Move to "Methods" section and expand explanation (e.g., arbitrary factors).
L174-177: These interpretations of Fig. 4 are not clear to me. Maybe I have too little experience with PCA, but I think that other readers see it similarly and also need more explanation.
L186: The direct use of surface temperature (Ts) may not be a very good proxy for ETa. Typically, energy balance models or the warming rates from diurnal Ts measurements are used to infer ETa from Ts (e.g. Panwar et al., 2019). In addition, it is evident from Table 2 that Ts is strongly anticorrelated with NDVI, indicating that the two variables are not independent.
L193: What is meant by this? The soil map does not show any relevant structures.
L194-195: These interpretations are too speculative.
L206-209: These interpretations are not clear to me. Furthermore, the soil texture in the study area is extremely homogeneous, which is why any interpretation of soil effects seems to me to be exaggerated.
L222-223: This statement is not clear to me. Please explain in more detail.
L239-240: Please explain in more detail how you arrive at 61%.
L253-254: This statement needs to be better justified.
L258-259: Too speculative.
L262-263: Too speculative.
L265-268: These interpretations are implausible because the aforementioned effects on soil organic matter take many years to occur.
L272: In this case crop management shapes the environment.
L285: Figure 9.
L286: It is not clear to me why positive loadings should indicate a damped behavior of soil moisture.
L294: In my opinion, this research is not an indispensable prerequisite for tailored field and crop management. In fact, modern sensor-based agricultural techniques allow for a tailored crop management already (e.g. Chamara et al., 2022).
Figures
Fig. 1: Please add horizontal bars for each patch to the figure to make the vegetation stages of the patches easier to understand. In addition, potential ET should be plotted, which is a better proxy for actual ET then air temperature.
Figs. 3 and 7: Use same color scheme as in Fig. 5 to better differentiate the different sensor depths.
References
Bogena, H.R., A. Weuthen and S. Huisman (2022): Recent developments in wireless soil moisture sensing to support scientific research and agricultural management. Sensors 22: 9792. DOI: 10.3390/s22249792
Chamara, N., Islam, M. D., Bai, G. F., Shi, Y., & Ge, Y. (2022). Ag-IoT for crop and environment monitoring: Past, present, and future. Agricultural systems, 203, 103497.
Panwar, A., Kleidon, A., & Renner, M. (2019). Do surface and air temperatures contain similar imprints of evaporative conditions?. Geophysical Research Letters, 46(7), 3802-3809. https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2019GL082248
Citation: https://doi.org/10.5194/egusphere-2023-1115-RC1 -
AC1: 'Reply on RC1', Kathrin Grahmann, 20 Sep 2023
Reviewer 1
This paper describes the application of the well-known principal component analysis method to disentangle effects of crops and soil properties on soil moisture dynamics using 64 soil moisture time series from an agricultural experiment with differently managed small plots. This study is based on a quite large data set of soil moisture measurements and is tangential to an important topic in environmental research. Unfortunately, the interpretations of the results are partly very speculative and difficult to comprehend. Furthermore, transferability of the results to other areas is very limited, as they are determined by the very specific conditions of the experimental study area. I recommend that the authors turn these weaknesses into strengths by arguing that homogeneous soil properties make it easier to study the effects of crop types on soil water balance. The manuscript is mostly well written but need to be checked by a native speaker. I have listed further limitations in my general and specific comments below.
We would like to thank the reviewer for the thorough review. We did our best to meet the comments and recommendations. We added more explanations and details to support the reader in comprehending the interpretation of the data. We agree that in our study soil texture exhibits little heterogeneity and thus the results allow only limited inferences on soil heterogeneity effects. On the other hand, soil homogeneity is not a necessary prerequisite for application of the principal component analysis, and the approach can be used to assess related effects even when they are small. In addition, the term “soil effects” in the title does not only refer to effects of soil heterogeneity but to effects of increasing damping of hydrological signals with increasing soil depth as well.
General comments
The main goal of this study is to disentangle effects of crops and soil properties on soil moisture dynamics. However, the results cannot be generalized due to the peculiarities of the study area. On the one hand, the large vegetation effect observed in this study is due to very specific small-scale crop management with various crops in one field, which does not occur in regular agricultural systems. On the other hand, the soil texture of the studied plots is very similar, so that the minor soil effects on soil moisture found in this study are not representative for landscapes with more typical soil heterogeneity. The similarity in soil texture might also be the reason for the low influence of soil sensor depth and roots on the soil moisture time series.
We reworked the text to emphasize the peculiarities of the study on the one hand, and the wider applicability of the presented approach on the other hand. In terms of minor soil texture heterogeneity please see above.
For the reasons stated above, the title of the manuscript is not appropriate and should instead reflect the very specific conditions of the study area.
Please see comment above.
The data of the synthetic time series shown in Figures 4, 6, 8, and 10 as well as their interpretations are difficult to understand. To convince readers that the interpretation is robust, these data need to be explained and justified much better.
Additional explanations are added to the Methods and Results section.
In the Methods section, we added to the elaboration of how these Figures are produced and how they can be interpreted: “The scores of the principal components constitute time series. Every observed time series can be presented at arbitrary precision as a combination of various principal components. When the data set consists of time series of the same observable measured at different locations, the first principal component describes the mean behaviour inherent in the data set. Subsequent principal components reflect typical modifications of that mean behaviour at single locations due to different effects. Thus generating synthetic time series as linear combinations of the first PC and another additional PC helps to assign this additional PC to a specific effect. To that end scores of that component have either been added to or subtracted from those of the first component using arbitrarily selected factors. The two resulting graphs show how the respective PC causes deviations from the mean behaviour of the data set.“
In the Results section, we added elaboration on how we interpreted the deviations from the mean behaviour.
This study uses data from an underground LoRa-based sensor network. The authors claim that this system is novel, but information on why it is novel is largely lacking.
More explanation on the novelty is added to the manuscript:
“The novelty of this Internet of underground Things (IouT) soil moisture monitoring network is characterized by its unique on-farm installation environment and the deployment of 180 sensors in up to 90 cm soil depth, allowing for high spatio-temporal resolution wireless data transmission, and enabling conventional farming practices like machinery traffic, tillage and mechanical weeding.”
In addition, the soil moisture time series shows large data gaps. The authors provide some general information about data gaps, but do not go into technical detail (e.g. battery failure, transmission failure, sensor failure etc.), which would be interesting given the novelty of the wireless system.
More details are added to the manuscript: “Transmission failures due to discharged batteries, due to signal disturbances in sinks after rainfall or in patches with a high density of biomass (e.g. maize) and theft of parts of the monitoring system led to data gaps that amounted to 81 out of 257 days of the measuring period.”
The authors compare “conventional” with “reduced” cases, but in both cases weeds are being controlled. Therefore, not difference between both cases in terms of soil moisture can be expected.
We differentiate between “conventional” and “reduced” weed control because mechanical weeding impacts soil structure and could enhance soil evaporation which in turn could results in deeper rooting of the plants in contrast to chemical weed control.
The measured time series of soil moisture should also be presented in meaningful figures, since these form the basis for the statistical analysis. If the number of figures becomes too large, they can also be presented in an appendix.
Additional figures can be provided for the appendix.
Specific comments:
L13-15: Combine sentences.
Adjusted in the manuscript.
L42: All cited papers didn’t use TDR, but capacitance probes etc.. These kind of low-cost soil moisture sensors are usually used in wireless sensor network applications (see e.g. Bogena et al., 2022). Therefore, I suggest using the more general term “electromagnetic soil moisture sensors”.
We agree. We changed it accordingly.
L47: This study uses data from an underground LoRa-based sensor network. The authors claim that this system is novel, but information on why it is novel is largely lacking.
This will be highlighted in the end of the introduction: “The novelty of this Internet of underground Things (IouT) soil moisture monitoring network is characterized by its unique on-farm installation environment and the deployment of 180 sensors in up to 90 cm soil depth, allowing for high spatio-temporal resolution wireless data transmission, and enabling conventional farming practices like machinery traffic, tillage and mechanical weeding.”
L66: Explain in more detail the novelty of this wireless soil moisture monitoring system (please note that are large number of similar systems already exist, see e.g. Bogena et al., 2022)
We thank the reviewer for the literature recommendation of Bogena et al. (2022) which we were not aware of as this manuscript was prepared before the publication of that paper.
The system is novel in terms of installation environment and number of installed sensors. Those wireless Lora systems may have been installed and used in the past in other ecosystems, but to the best of our knowledge we do not know about agricultural systems, and in particular one single field that is equipped with 180 sensors providing the information wirelessly in high temporal resolution and hence allow business as usual machine traffic and tillage. We added this justification in the introduction.
L83-84: Explain “yield potential zones”.
We further explained the experimental design of patchCROP and provided a short information on the cluster analysis that has been carried out to define two different yield potential zones in the field. Details on the clustering method are provided in Donat et al. (2022).
L95: The “DriBox” is just the housing for the electronics. Please provide information on the manufacturer of the electronic parts.
We elaborated the technical section and provided all the hardware details: “In each patch, one Dribox box was equipped with a SDI-12 distributer (serial data interface at 1200 baud rate, TBS04, TekBox, Saigon, Vietnam) connected to six TDR-sensors (TDR310H, Acclima, Meridian, USA) and attached to an outdoor remote terminal unit (RTU) fully LoRaWAN compliant (TBS12B: 4+1 channel analogue to SDI-12 interface for 24 Bit A/D conversion of sensor signals, TekBox, Saigon, Vietnam).”
L97: Does this mean that you have dug 0.9 m deep trenches for the cables? Please explain the installation of the sensors in more detail.
We described the installation process more comprehensively and made clear that the soil pit was only 30 to 40 cm deep whereas the 60 and 90 cm sensors were inserted vertically with previously prepared tunnels and tubes that push the sensor into the soil.
L104: Why was only data from one drone campaign used in this study? Given the high temporal variability of plant and soil water status, the use of a single snapshot may not be sufficiently representative for the conclusions drawn in this analysis.
There are no other thermal data available. However, we can include NDVI data from four additional dates (between March 2021 and July 2021) into additional analyses. Results can be added to the manuscript.
L117: What is the accuracy of the soil texture prediction model? Please provide more information on the data processing in the appendix.
The Geophilus system is a service that was purchased to receive the final texture map. Overdrive and sampling have been carried out by the Geophilus company (https://www.gkb-ev.de/publikationen/eip/geophilus.pdf). The model prediction accuracy was provided including gamma and ERa as covariates to predict clay, silt and sand. The additive log ration (ALR) transformation was applied to clay and sand fractions. The best fit was reached with a with Non-linear regression (exponential) model, having a root mean square error of 1.8% for clay, 5.7% for sand and 4.6% for silt. We added that information to the M&M section.
L118: What do you mean with “gamma sensor” and how does it reduce uncertainty?
The gamma sensor is used to detect the natural gamma radiation emitted by the ground. It is emitted mainly by uranium and thorium particles and thus refelcts the proportion of potassium-rich minerals in the clay and silt fraction. Therefore, the measured gamma activity is proportional to the clay content.Because the γ-radiation is less sensitive to soil moisture than the ERa readings, the ratio between the γ-activity and the ERa of the array with the smallest electrode spacing (investigation depth: 0–0.25 m) represents the influence of the soil water on the ERa readings (Bönecke et al., 2021).
Information on the gamma sensor and a new reference were added.
L123: Please describe in more detail the technical problems (e.g. transmission failure etc.).
Information is now provided in the manuscript: “Transmission failures due to discharged batteries, due to signal disturbances in sinks after rainfall or in patches with a high density of biomass (e.g. maize) and theft of parts of the monitoring system led to data gaps that amounted to 81 out of 257 days of the measuring period.”
L125: Could you explain why these sensors show frequent malfunctioning (e.g. do to the sensors itself or do the wireless system)?
Sensors that showed a particularly high frequency of transmission failures were excluded entirely from the study. Unfortunately, it was not possible to determine the exact reason for the high number of errors for specific sensors. Possible reasons could be: Technical failures of individual sensors; transmission failures between sensor and node box due to e.g. cable damage; overlapping of different effects already described that weaken the RSSI signal. At the latter it must be considered that all sensors at a specific patch are connected to the same node box. Thus, if data from other sensors at the same patch were transmitted, problems with individual sensors are more likely to be the reason for the data gaps than transmission errors between the node box and the gateway.
L125: Define “short”.
Details were added to the manuscript: “Of all 20668 interpolated gaps, 96 % were shorter than two hours, 3 % between two and six hours and 1 % longer than six hours. In 26 cases, the gap exceeded the duration of one day.”
L140-141: Was this the case in this study? Otherwise, delete.
All analysed PC had an eigenvalue greater than one.
L143: Please explain “local effects”.
This part of the methodology was not necessarily important for the manuscript and was therefore deleted.
L158-160: The interpretation that the first PC shows the control of atmospheric forcing should be better justified. For instance, the time series of scores could be correlated with P-ET time series.
The correlation between the scores and the cumulative climatic water balance (P-ETp) is -0.97. The information was added to the manuscript.
L169-173: Move to "Methods" section and expand explanation (e.g., arbitrary factors).
Moved to the Methods section and expanded explanation added in the manuscript (see general comment on Figures 4, 6, 8 and 10).
L174-177: These interpretations of Fig. 4 are not clear to me. Maybe I have too little experience with PCA, but I think that other readers see it similarly and also need more explanation.
We added additional explanations in the Methods and Results sections (see comment above).
L186: The direct use of surface temperature (Ts) may not be a very good proxy for ETa. Typically, energy balance models or the warming rates from diurnal Ts measurements are used to infer ETa from Ts (e.g. Panwar et al., 2019). In addition, it is evident from Table 2 that Ts is strongly anticorrelated with NDVI, indicating that the two variables are not independent.
Diurnal data were not available as the drone images provided only a single snapshot in time. Instead, the spatial pattern of surface temperature was deemed to be related to that of actual evapotranspiration in a monotonic, although not necessarily linear way. Close anti-correlation of the resulting pattern with that of NDVI provided some evidence that this approach was justified.
L193: What is meant by this? The soil map does not show any relevant structures.
We clarified the statement: “Although the affected patches do not correspond to anomalies in the soil map, it is still apparent that the location of the patches roughly follows an east-west direction.”
L194-195: These interpretations are too speculative.
We rephrased to better describe the effect: “The most obvious difference between the orange line (negative loading on PC3) and the blue line (positive loading on PC3) during the first half of the study period is that the latter reaches a maximum of soil moisture after rainfall much earlier compared to the former (Figure 6).”
Thereby, in combination with additional elaboration in the Discussion section, we hope to support the reader in comprehending the interpretation of this PC: “Loadings on the third principal component were not related to crop types. In contrast, a spatial pattern emerged: Only sensors from 0.9 m depth from six adjacent patches exhibited strongly negative loadings (Figure 2) whereas all other sensors showed minor positive or negative loadings. This points to an effect of subsoil substrates, that is higher loam content and consequently higher water holding capacity. That would be consistent with delayed response to seepage fluxes and reduced desiccation in the vegetation period (Figure 6).”
L206-209: These interpretations are not clear to me. Furthermore, the soil texture in the study area is extremely homogeneous, which is why any interpretation of soil effects seems to me to be exaggerated.
The statement has been refined to clarify that we do not refer to the soil as loamy but describe the development over time of the orange graph as behaviour which is typical for loamy soils: “Figure 8 illustrates the effect of the fourth PC on time series. A positive factor would be typical for more sandy soils and for patches with fallow in autumn and winter (blue line). In contrast the orange line depicts behaviour in more loamy soils and for winter crops. The latter line exhibits slightly more delayed responses to rainstorms and subsequent less steep recovery as would be expected for more loamy soils. However, it is not clear how winter crops on the one side and fallow on the other side could induce such a different behaviour.”
L222-223: This statement is not clear to me. Please explain in more detail.
The statement has been re-formulated. We want to express that our analyses revealed various effects of soil texture, soil depth, crops and management.
L239-240: Please explain in more detail how you arrive at 61%.
We added additional explanations: “When not considering the temporal component reflected by PC1 and thus only looking at the spatial variability, 61% of the remaining variance (attributed to PC2 to PC64) is caused by the vegetation effect reflected by PC2.”
L253-254: This statement needs to be better justified.
The scores are time series and reflect the effect size of a particular process represented by the respective PC. The more the scores of a certain PC deviate from zero during single periods, the stronger the respective effect is. Consequently, the development of the time series of PC2 scores – strongly varying and having an amplitude greater than 20 – indicates that the effect of vegetation on total variability varies by time.
L258-259: Too speculative.
We elaborated a little bit more on that but emphasizing that these are very preliminary inferences, based on own observations and similar observations made by other colleagues (e.g., Döring et al., in preparation).
L262-263: Too speculative.
See comment above and following comment.
L265-268: These interpretations are implausible because the aforementioned effects on soil organic matter take many years to occur.
See comment above and reply to general comment of RC2: “The interpretation of the fourth principal component is consistent with own observations and similar observations made by other colleagues (e.g., Döring et al., in preparation). Effects of changing soil organic carbon quantity and quality are assumed to occur only at larger time scales which is closely related to the problem of detecting respective changes within shorter periods. However, that might be more a problem of detectability rather than a sound disproof of the suggested mechanism. We think more research is needed here, including but not being restricted to indirect methods like that used in our studies.”
L272: In this case crop management shapes the environment.
We agree and we adjusted the respective phrase in the manuscript.
L285: Figure 9.
Thank you, of course Figure 9 should be referenced.
L286: It is not clear to me why positive loadings should indicate a damped behavior of soil moisture.
The statement has been elaborated a little bit more: “Loadings on this component are clearly related with depth (Figure 9). Strong positive loadings indicate a strongly damped behaviour of soil moisture time series: The blue line, representing sites with positive loadings on PC5 which is typical for sensors at greater depth (Figure 9) exhibits clearly reduced amplitudes compared to the yellow line, that is, sensors at shallow depth (Figure 9, Figure 10).”
In combination with information on how Figures 4, 6, 8, and 10 are derived and how they can be interpreted, we hope that readers can now follow our interpretations.
L294: In my opinion, this research is not an indispensable prerequisite for tailored field and crop management. In fact, modern sensor-based agricultural techniques allow for a tailored crop management already (e.g. Chamara et al., 2022).
The statement relates to disentangling and quantifying different effects in general, not specifically to the suggested approach. We consider the latter very helpful in addition to modern sensor systems..
Figures
Fig. 1: Please add horizontal bars for each patch to the figure to make the vegetation stages of the patches easier to understand. In addition, potential ET should be plotted, which is a better proxy for actual ET then air temperature.
The figure can be adjusted accordingly.
Figs. 3 and 7: Use same color scheme as in Fig. 5 to better differentiate the different sensor depths.
The figures can be adjusted accordingly.
References
Bogena, H.R., A. Weuthen and S. Huisman (2022): Recent developments in wireless soil moisture sensing to support scientific research and agricultural management. Sensors 22: 9792. DOI: 10.3390/s22249792
Chamara, N., Islam, M. D., Bai, G. F., Shi, Y., & Ge, Y. (2022). Ag-IoT for crop and environment monitoring: Past, present, and future. Agricultural systems, 203, 103497.
Panwar, A., Kleidon, A., & Renner, M. (2019). Do surface and air temperatures contain similar imprints of evaporative conditions?. Geophysical Research Letters, 46(7), 3802-3809. https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2019GL082248
Citation: https://doi.org/10.5194/egusphere-2023-1115-AC1
-
AC1: 'Reply on RC1', Kathrin Grahmann, 20 Sep 2023
-
RC2: 'Comment on egusphere-2023-1115', Tobias L. Hohenbrink, 28 Jul 2023
Summary
In the study “Differentiating between crop and soil effects on soil moisture dynamics” by Helen Scholz et al. 64 soil moisture time series covering eight months are evaluated by a principal component analysis. The data have been measured in three depths at a site in Eastern Germany with a wireless network of TDR sensors. The resulting components were interpreted based on supporting information about (i) precipitation and temperature, (ii) crop rotation, (iii) sand content in the upper 25 cm, and (iv) NDVI and surface temperature. A share of 97 % of total soil moisture variance could be described by the first five components and has been assigned to meteorological conditions (27%), the cropping system (17 %), soil properties (6,3 %), and signal damping (1.7 %).
General commentsObjectives of the study:
The research question addressed in the study (L66-70) is generally relevant and also interesting for the readers of HESS. It should be defined more precisely what exactly is meant by “highly diversified fields” in this study. It might also be unclear at first what “quantify the drivers of soil moisture” really means. The readers might first think about quantifying the individual components of the hydrological water balance by absolute values. However, due to the z-transformation, this cannot be achieved with a PCA. The objectives should be formulated more precisely.
Methods:
PCA of soil moisture time series is a promising approach to identify the dominating factors of soil moisture dynamics and assess the strength of their effects. It is not a new approach, since some very similar studies already exist, where a PCA has been applied to soil moisture time series. However, this should not be a problem for a publication in HESS, because we can still learn a lot from repeating the analyses at new sites. The main methodological problem I see in the study is that extensive and robust data are needed to identify interpretable patterns with the PCA approach, which are important to draw valid conclusions about thematic research questions. Unfortunately, quite limited data were considered in this study.
Analysed Data:
Only a very short period of eight months of soil moisture measurements have been analyzed. These time series additionally contained large data gaps, unfortunately during interesting times: (i) the period during steady rain mid of May, and (ii) the three weeks after the strong rain in July. Unfortunately, the data gaps meet particularly interesting situations where soil moisture information would have been very important to learn about the hydrological functioning at the site. The study would be improved strongly, when soil moisture data for a longer time period could be included. Maybe moisture time series of higher quality have been measured in the subsequent growing period.
The available soil texture information only contains sand contents in the upper 25 cm derived from geoelectric exploration.This information is poorly suited for process interpretations, because the sand content at the TDR-sensor positions varies in a very small range of only 3 % (between 77.9% and 80.7% ,Table 1), which might even be close to the uncertainties of the geoelectrical method. There are a lot of other potential factors determining the soil hydraulic properties (e,g, clay content, bulk density, organic carbon content, etc.), which have not been taken into account in this study. I think that this marginal variance in sand content cannot be used alone to explain the soil moisture patterns identified by principal components. When single components shall be related to soil texture, more texture information from all considered soil depths is needed. Therefore, I highly recommend going back to the field, taking new soil samples (e.g with a small hand auger or a gouge auger) and determining their sand silt and clay contents.
Findings, interpretations and conclusions:
The 1st, 2nd and 5th principal components could be related to reasonable controlling factors and the process interpretations also seem plausible. This does not apply to the third and fourth components. The interpretations of these components are not based on solid data.
I assume that either the information actually needed to interpret these PCs is not available, or that the PCA fails to provide clearly interpretable components here. The weak interpretation of the third and fourth components should be discussed in more detail. In general, there should be more discussion of the suitability of the available data for principal component interpretation.
Minor commentsL30-32: Please, provide some more references for the effects listed.
L33: What is exactly meant by “complexity of the assessment and monitoring ”. What shall be assessed and why?
L47-50: “Soil moisture variograms” are a poor example for “sophisticated data analysis approaches”, because they are very simple. Please rephrase or find another example.
L55-57: The concept of “temporal stability” was introduced by Vachaud (1985) (https://doi.org/10.2136/sssaj1985.03615995004900040006x) which should be acknowledged with a citation. The review by Vanderlinden et al. (2012) (https://doi.org/10.2136/vzj2011.0178) also seems to be a very suitable reference here.
L64: The term “highly diversified fields” should be defined more exactly.
L83-84: What is a “yield potential zone”?
Table 1: What is meant by “treatment”? Readers might think about pest control or soil tillage. Maybe you can find another term.
Table 1: The “highly heterogeneous soils” (L75) are not reflected in the sand content listed in the Table. They vary only in a range of 3%. Therefore, I expect that they cannot explain large parts of the soil moisture variance. The clay content would be much more interesting here.
L94-98: The technical description should be improved. What do the “node boxes do”? How are the TDR sensors connected to the node boxes?
L102: How have the meteorological data been measured?
L111: Which physical variable is meant by “near infrared” and the red band? The intensity? or a relative share?
L124: I really regret (i) that the considered time periods are so short and (ii) that the data gaps occur during the most interesting periods. I see this as one of the biggest problems in this study. Is it possible to extend the period or maybe use other data from the following growing period?
L128-130: Please explain the implications of the z-transformation. Readers have to know that the z-transformation has to be kept in mind when interpreting the scores of a PC.
L140-141: Please rephrase the explanation of the criterion by Kaiser (1960). Eigenvalues greater than one indicate that a PC explains more variance than one input time series can contribute to the total variance of the entire input data set.
L143-145: I don’t understand what has been done here and why. Please provide more information.
L156-161: Please mention in half a sentence why the scores and loadings of the first PC are not shown here in the manuscript.
L183-189: It is very difficult to follow and to understand the effects and potential causal relations that are described here. For example: Soil temperature is negatively correlated with the loadings of PC 2 which in turn indicate a negative (summer crops) and positive (winter crops) correlation between the moisture time series and the scores of PC 2. I am sure that most readers (including me) need a better explanation of these dependencies. They need to be better guided in order not to get lost.
Figure 4: What about harvesting? In August the winter crops (blue line) have constant scores (indicating stopped transpiration after harvesting?) while the scores describing moisture dynamics for summer crops (red line) are still decreasing (ongoing transpiration?). Unfortunately there is a data gap.
190-195: It is hard to follow the description of the third PC. I have the feeling that in the third PC the effects of several factors interact. Perhaps the relevant supporting information to understand PC 3 is simply not known. If the authors are really confident in their interpretation of the third PC, they should describe the relationships more clearly. If they are skeptical, as I am, they should discuss these problems in detail.
L203-205: Are the correlations with the sand contents not shown? As mentioned earlier, I don’t think that the sand content can explain any variance due to its small variation.
L203-209: It is rather difficult to interpret the effects of two different factors (cropping system and sand content of upper 25 cm) in PC 4, which explains only 2.2% of the total variance.
L217: Please check if it should be lupine instead of sunflower.
L222-223: I don’t really know what is meant here. Is redundancy here the correct term?
L232: “quantification of the strength of these effects” might be more precise
L247-250: Please check if Yang et al. (2015) have also z-transformed their data. If not it might be difficult to compare their findings with those of this study.
L265: What do you mean by loamy soils? I think that all soils at the site are sandy soils.
L265-267: Very speculative. I think that an increase of carbon stock happens at larger time scales and can unlikely explain the moisture patterns explained by PC 4.
L274-291: I can imagine that soil texture is an important factor controlling soil moisture dynamics at the investigated site. However, as mentioned before, more information about the depth distribution of soil texture is needed. If it is planned to run the “patchCROP” experiment for longer, it is really worth going back to the field, collecting soil samples at each TDR sensor position in 30, 60, and 90 cm depth and performing a texture analysis.
L296: I agree that it is important to study the interaction of different factors in their effect on soil moisture dynamics. Unfortunately, in these interactions, the patterns identified by a PCA often become blurred, making interpretation difficult with the usually limited supporting information available.
L304-305: I agree, but is that conclusion really founded on the findings of this study? The sentence could also be shifted to the introduction.
L307-309: This paragraph might be shifted to the discussion section.
Citation: https://doi.org/10.5194/egusphere-2023-1115-RC2 -
AC2: 'Reply on RC2', Kathrin Grahmann, 20 Sep 2023
Reviewer 2
Summary
In the study “Differentiating between crop and soil effects on soil moisture dynamics” by Helen Scholz et al. 64 soil moisture time series covering eight months are evaluated by a principal component analysis. The data have been measured in three depths at a site in Eastern Germany with a wireless network of TDR sensors. The resulting components were interpreted based on supporting information about (i) precipitation and temperature, (ii) crop rotation, (iii) sand content in the upper 25 cm, and (iv) NDVI and surface temperature. A share of 97 % of total soil moisture variance could be described by the first five components and has been assigned to meteorological conditions (27%), the cropping system (17 %), soil properties (6,3 %), and signal damping (1.7 %).
Thanks for the comprehensive and in-depth review.
General commentsObjectives of the study:
The research question addressed in the study (L66-70) is generally relevant and also interesting for the readers of HESS. It should be defined more precisely what exactly is meant by “highly diversified fields” in this study. It might also be unclear at first what “quantify the drivers of soil moisture” really means. The readers might first think about quantifying the individual components of the hydrological water balance by absolute values. However, due to the z-transformation, this cannot be achieved with a PCA. The objectives should be formulated more precisely.
We did our best to clarify information on the objectives (Abstract, Introduction) and on the details of the study.
Diversification of agricultural systems can be implemented and reached through spatial and temporal approaches. In patchCROP we combined both and designed a completely new cropping system design with a high level of diversification in terms of crops, soil management zones, field size and land use intensity (in terms of plant protection). The changing soil-hydrological dynamics in complex diversified agricultural systems with increasing heterogeneity and site-specific adjustment of crops, soil types and field management which have hardly been studied so far.
We added to the Methods section the limitations of the analysis of z-transformed data sets regarding absolute values.
Methods:
PCA of soil moisture time series is a promising approach to identify the dominating factors of soil moisture dynamics and assess the strength of their effects. It is not a new approach, since some very similar studies already exist, where a PCA has been applied to soil moisture time series. However, this should not be a problem for a publication in HESS, because we can still learn a lot from repeating the analyses at new sites. The main methodological problem I see in the study is that extensive and robust data are needed to identify interpretable patterns with the PCA approach, which are important to draw valid conclusions about thematic research questions. Unfortunately, quite limited data were considered in this study.
We agree that long and gapless time series would be ideal for any in-depth analysis. However, such data sets are often not available. Fortunatley though PCA can be applied and the results be interpreted despite data gaps. Therefore, we consider the methodology suitable for many real-world monitoring setups.
Analysed Data:
Only a very short period of eight months of soil moisture measurements have been analyzed. These time series additionally contained large data gaps, unfortunately during interesting times: (i) the period during steady rain mid of May, and (ii) the three weeks after the strong rain in July. Unfortunately, the data gaps meet particularly interesting situations where soil moisture information would have been very important to learn about the hydrological functioning at the site. The study would be improved strongly, when soil moisture data for a longer time period could be included. Maybe moisture time series of higher quality have been measured in the subsequent growing period.
We agree in terms of the detrimental long data gap. Still, other important and characteristic time periods of the year were covered, such as the moist winter months with subsequent rain falls in end of January and in February and the dry weeks in June. On the other hand, though, considering longer time series beyond the length of a single cropping period would cause another problem inasmuch as effects of different crops would mix up in the soil moisture readings of single sites. Thus, identification of crop-related effects would hardly be feasible.
The available soil texture information only contains sand contents in the upper 25 cm derived from geoelectric exploration. This information is poorly suited for process interpretations, because the sand content at the TDR-sensor positions varies in a very small range of only 3 % (between 77.9% and 80.7% ,Table 1), which might even be close to the uncertainties of the geoelectrical method. There are a lot of other potential factors determining the soil hydraulic properties (e,g, clay content, bulk density, organic carbon content, etc.), which have not been taken into account in this study. I think that this marginal variance in sand content cannot be used alone to explain the soil moisture patterns identified by principal components. When single components shall be related to soil texture, more texture information from all considered soil depths is needed. Therefore, I highly recommend going back to the field, taking new soil samples (e.g with a small hand auger or a gouge auger) and determining their sand silt and clay contents.
Sand was varying a lot at the field scale between 69.1 and 81.2% at the site, but little within patches. Clay and silt estimates are available from Geophilus and can be further analysed and added to this manuscript.
In the meantime, additional data were provided. They are manual soil auger results until 1 m depth available from project activities in the DFG excellence cluster PhenoRob for eight out of 12 analysed patches. This information can be also incorporated in further analyses.
But even then, we agree that in our study soil texture exhibits little heterogeneity and thus the results allow only limited inferences on soil heterogeneity effects. On the other hand, soil homogeneity is not a necessary prerequisite for application of the principal component analysis, and the approach can be used to assess related effects even when they are small. In addition, the term “soil effects” in the title does not only refer to effects of soil heterogeneity but to effects of increasing damping of hydrological signals with increasing soil depth as well.
Findings, interpretations and conclusions:
The 1st, 2nd and 5th principal components could be related to reasonable controlling factors and the process interpretations also seem plausible. This does not apply to the third and fourth components. The interpretations of these components are not based on solid data.
I assume that either the information actually needed to interpret these PCs is not available, or that the PCA fails to provide clearly interpretable components here. The weak interpretation of the third and fourth components should be discussed in more detail. In general, there should be more discussion of the suitability of the available data for principal component interpretation.
We elaborated and refined our reasoning in terms of the third and fourth component. We agree that these arguments are far from unequivocal proofs. But we consider it worthwhile to consider even unexpected results. E.g., the interpretation of the fourth principal component is consistent with own observations and similar observations made by other colleagues (e.g., Döring et al., in preparation). Effects of changing soil organic carbon quantity and quality are assumed to occur only at larger time scales which is closely related to the problem of detecting respective changes within shorter periods. However, that might be more a problem of detectability rather than a sound disproof of the suggested mechanism. We think more research is needed here, including but not being restricted to indirect methods like that used in our studies.
Minor commentsL30-32: Please, provide some more references for the effects listed.
Additional references were included:
- Fischer, C., Roscher, C., Jensen, B., Eisenhauer, N., Baade, J., Attinger, S., Scheu, S., Weisser, W. W., Schumacher, J., Hildebrandt, A.: How Do Earthworms, Soil Texture and Plant Composition Affect Infiltration along an Experimental Plant Diversity Gradient in Grassland?, PLos ONE, 9, 6, https://doi.org/10.1371/journal.pone.0098987, 2014.
- Koudahe, K., Allen, S. C., Djaman, K.: Critical review of the impact of cover crops on soil properties, International Soil and Water Conservation Research, 10, 343-354, https://doi.org/10.1016/j.iswcr.2022.03.003, 2022.
- Nunes, M. R., van Es, H. M., Schindelbeck, R., Ristow, A. J., Ryan, M.: No-till and cropping system diversification improve soil health and crop yield, Geoderma, 328, 30-43, https://doi.org/10.1016/j.geoderma.2018.04.031, 2018.
L33: What is exactly meant by “complexity of the assessment and monitoring ”. What shall be assessed and why?
The more independent variables are present in agricultural systems, the higher the demand for frequency and spacings of soil moisture measurement / related data. We revised the phrase.
L47-50: “Soil moisture variograms” are a poor example for “sophisticated data analysis approaches”, because they are very simple. Please rephrase or find another example.
We rephrased the formulation: “Methods include geostatistical analysis (Vereecken et al., 2014) or data driven approaches (Hong et al., 2016).” Examples for more sophisticated approaches will be given in the following sentence.
L55-57: The concept of “temporal stability” was introduced by Vachaud (1985) (https://doi.org/10.2136/sssaj1985.03615995004900040006x) which should be acknowledged with a citation. The review by Vanderlinden et al. (2012) (https://doi.org/10.2136/vzj2011.0178) also seems to be a very suitable reference here.
Thank you for the valuable note, the references were added to the manuscript.
L64: The term “highly diversified fields” should be defined more exactly.
The term has been defined more clearly, making clear that it refers to the multitude of different crops and management schemes within a single arable field (see general comments).
L83-84: What is a “yield potential zone”?
We further explained the experimental design of the experimental field and provided a short information on the cluster analysis that has been carried out to define two different yield potential zones within in the field. A reference is given in the text (Donat et al. 2022).
Table 1: What is meant by “treatment”? Readers might think about pest control or soil tillage. Maybe you can find another term.
We decided to re-name this column to “crop groups”. Crop group A contains winter crops, crop group B contains fallow (in winter), followed by summer crops and crop group C contains cover crops, followed by summer crops.
Table 1: The “highly heterogeneous soils” (L75) are not reflected in the sand content listed in the Table. They vary only in a range of 3%. Therefore, I expect that they cannot explain large parts of the soil moisture variance. The clay content would be much more interesting here.
At the study site the sand content in the upper layer varied between 69 % and 81 %. However, the variability in the analysed patches was indeed low. Information on clay content, which is in the meantime also available for deeper layers of eight out of 12 patches, can be used for further analysis. Results can be added to the manuscript.
L94-98: The technical description should be improved. What do the “node boxes do”? How are the TDR sensors connected to the node boxes?
We elaborated the technical description of the sensor system and provide all hardware details (see reply to RC1).
L102: How have the meteorological data been measured?
This information has been added. We had two meteorological stations on site.
L111: Which physical variable is meant by “near infrared” and the red band? The intensity? or a relative share?
Details on the values used for calculation were added (near infrared as light reflected by vegetation and red as absorbed by vegetation).
L124: I really regret (i) that the considered time periods are so short and (ii) that the data gaps occur during the most interesting periods. I see this as one of the biggest problems in this study. Is it possible to extend the period or maybe use other data from the following growing period?
We agree in terms of the detrimental long data gap. Still, other important and characteristic time periods of the year were covered, such as the moist winter months with subsequent rain falls in end of January and in February and the dry weeks in June. On the other hand, though, considering longer time series beyond the length of a single cropping period would cause another problem inasmuch as effects of different crops would mix up in the soil moisture readings of single sites. Thus, identification of crop-related effects would hardly be feasible.
L128-130: Please explain the implications of the z-transformation. Readers have to know that the z-transformation has to be kept in mind when interpreting the scores of a PC.
We added to the manuscript that due to the z-transformation absolute values of soil moisture and thus absolute changes cannot be interpreted or explained by PCA.
L140-141: Please rephrase the explanation of the criterion by Kaiser (1960). Eigenvalues greater than one indicate that a PC explains more variance than one input time series can contribute to the total variance of the entire input data set.
We gladly replace the original version with the suggestion of the reviewer.
L143-145: I don’t understand what has been done here and why. Please provide more information.
This part of the methodology was not necessarily important for the manuscript and was therefore deleted.
L156-161: Please mention in half a sentence why the scores and loadings of the first PC are not shown here in the manuscript.
Since the loadings on the first PC were all one-directional, the graphic was not shown. However, it can be provided in the appendix.
L183-189: It is very difficult to follow and to understand the effects and potential causal relations that are described here. For example: Soil temperature is negatively correlated with the loadings of PC 2 which in turn indicate a negative (summer crops) and positive (winter crops) correlation between the moisture time series and the scores of PC 2. I am sure that most readers (including me) need a better explanation of these dependencies. They need to be better guided in order not to get lost.
The paragraph has been re-formulated: “As shown in Table 3, the NDVI as a proxy for photosynthesis potential was positively correlated with the loadings. Surface temperature exhibited a negative correlation. On the other hand, the spatial pattern of surface temperature is assumed to be inversely related to that of actual evapotranspiration. Thus, both proxies, NDVI and surface temperature, support the inference that positive loadings on this principal component represent sites with above-average plant activity and root water uptake.”
Figure 4: What about harvesting? In August the winter crops (blue line) have constant scores (indicating stopped transpiration after harvesting?) while the scores describing moisture dynamics for summer crops (red line) are still decreasing (ongoing transpiration?). Unfortunately there is a data gap.
We agree, this effect can be attributed to the earlier harvesting of winter crops. We will add this observation to the description of the Figure.
190-195: It is hard to follow the description of the third PC. I have the feeling that in the third PC the effects of several factors interact. Perhaps the relevant supporting information to understand PC 3 is simply not known. If the authors are really confident in their interpretation of the third PC, they should describe the relationships more clearly. If they are skeptical, as I am, they should discuss these problems in detail.
By providing supplementary explanations for Figures 4, 6, 8, and 10, we hope that our interpretations can be better followed. It should be better illustrated that in Figure 6 two different types of drainage behaviour are shown. Due to the local, non-systematic occurrence of particularly pronounced loadings we attribute this PC to soil properties.
L203-205: Are the correlations with the sand contents not shown? As mentioned earlier, I don’t think that the sand content can explain any variance due to its small variation.
The correlation with sand content of loadings of other loadings were weak and thus not shown (0.18, 0.22, -0.36, -0.26 for PC1, PC2, PC3 and PC5, respectively).
As previously explained, more data on texture are available now for part of the analysed patches and can be used for further analysis.
We also refer to our answer to general comments of the first reviewer: “We agree that in our study soil texture exhibits little heterogeneity and thus the results allow only limited inferences on soil heterogeneity effects. On the other hand, soil homogeneity is not a necessary prerequisite for application of the principal component analysis, and the approach can be used to assess related effects even when they are small.”
L203-209: It is rather difficult to interpret the effects of two different factors (cropping system and sand content of upper 25 cm) in PC 4, which explains only 2.2% of the total variance.
See comment above: “We agree that our arguments are far from unequivocal proofs. But we consider it worthwhile to consider even unexpected results. Our preliminary interpretation of the fourth principal component is consistent with own observations and similar observations made by other colleagues (e.g., Döring et al., in preparation). Effects of changing soil organic carbon quantity and quality are assumed to occur only at larger time scales which is closely related to the problem of detecting respective changes within shorter periods. However, that might be more a problem of detectability rather than a sound disproof of the suggested mechanism. We think more research is needed here, including but not being restricted to indirect methods like that used in our studies.”
L217: Please check if it should be lupine instead of sunflower.
Thank you for the valuable remark. It is indeed lupine.
L222-223: I don’t really know what is meant here. Is redundancy here the correct term?
We revised the wording. We want to express that our analyses revealed various effects of soil texture, soil depth, crops and management.
L232: “quantification of the strength of these effects” might be more precise
We revised the wording into “quantification of the impact of these effects”
L247-250: Please check if Yang et al. (2015) have also z-transformed their data. If not it might be difficult to compare their findings with those of this study.
Since no z-transformed data set was used in the reference and the type of vegetation in the referenced study also differed, we decided not to make a comparison to the results of this study.
L265: What do you mean by loamy soils? I think that all soils at the site are sandy soils.
The phrase has been re-formulated: “According to this component, soil moisture dynamics at the fallow patches resembled more the typical behaviour one would expect for sandy soils, and that of winter crop patches more a more damped behaviour typical for more loamy soils.”
L265-267: Very speculative. I think that an increase of carbon stock happens at larger time scales and can unlikely explain the moisture patterns explained by PC 4.
See comment above:
“We elaborated and refined our reasoning in terms of the third and fourth component. We agree that these arguments are far from unequivocal proofs. But we consider it worthwhile to consider even unexpected results. E.g., the fourth principal component is consistent with own observations and similar observations made by other colleagues (e.g., Döring et al., in preparation). Effects of changing soil organic carbon quantity and quality are assumed to occur only at larger time scales which is closely related to the problem of detecting respective changes within shorter periods. However, that might be more a problem of detectability rather than a sound disproof of the suggested mechanism. We think more research is needed here, including but not being restricted to indirect methods like that used in our studies.”
L274-291: I can imagine that soil texture is an important factor controlling soil moisture dynamics at the investigated site. However, as mentioned before, more information about the depth distribution of soil texture is needed. If it is planned to run the “patchCROP” experiment for longer, it is really worth going back to the field, collecting soil samples at each TDR sensor position in 30, 60, and 90 cm depth and performing a texture analysis.
Soil texture has been determined manually and in the laboratory through research project activities in the DFG excellence cluster PhenoRob. We will be able to use those data for further additional interpretation which can be added to this manuscript as a new additional sampling campaign is not feasible due to long laboratory waiting times.
L296: I agree that it is important to study the interaction of different factors in their effect on soil moisture dynamics. Unfortunately, in these interactions, the patterns identified by a PCA often become blurred, making interpretation difficult with the usually limited supporting information available.
We consider PCA a powerful tool in this regard, although only just another step on the way to develop diagnostic tools for complex real-world systems. We added a corresponding statement: “Principal component analysis is a further step to meet these challenges although not entirely without problems.”
L304-305: I agree, but is that conclusion really founded on the findings of this study? The sentence could also be shifted to the introduction.
The phrasing was revised to highlight the connection between the study and this statement: “In particular, the plant-induced effects on soil hydraulic properties would be worthwhile to be studied in more detail. Knowledge from data-driven approaches can support adequate crop selection as a management option to encounter the increasing drought risk in the study region.”
L307-309: This paragraph might be shifted to the discussion section.
The phrasing was revised to highlight the potential of such analyses as one of the conclusions drawn from this study: “Information from this study will contribute to elucidate management effects as well as to develop both parsimonious and tailored mechanistic models. Findings of this study highly depend on local conditions. However, we consider the presented approach generally applicable to a large range of site conditions. In this regard, principal component analysis of soil moisture time series performed as a powerful diagnostic tool and is highly recommended.”
Citation: https://doi.org/10.5194/egusphere-2023-1115-AC2
-
AC2: 'Reply on RC2', Kathrin Grahmann, 20 Sep 2023
Data sets
Dataset of TDR soil moisture data from a LoRaWAN based soil sensing network of a selection of sensors at patchCROP for December 2020 to August 2021 for a principal component analysis Helen Scholz, Kathrin Grahmann https://doi.org/10.4228/zalf-3rsc-6c30
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 313 | 177 | 33 | 523 | 16 | 17 |
- HTML: 313
- PDF: 177
- XML: 33
- Total: 523
- BibTeX: 16
- EndNote: 17
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1