the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 16 Jul 2024
Identifying irrigated areas using land surface temperature and hydrological modelling: Application to Rhine basin
Abstract. Information about irrigation with relevant spatiotemporal resolution for understanding and modelling irrigation dynamics is important for improved water resources management. However, achieving a frequent and consistent characterization of areas where signals from rain-fed pixels overlap with irrigated pixels has been challenging. Here, we identify irrigated areas using a novel framework that combines hydrological modeling and satellite observations of land surface temperature. We tested the proposed methodology on the Rhine basin covering the period from 2010 to 2019 at a 1 km resolution. The result includes multiyear irrigated maps and irrigation frequency. Temporal analysis reveals that an average of 159 thousand hectares received irrigation at least once during the study period. The proposed methodology can approximate irrigated areas with R2 values of 0.79 and 0.77 for 2013 and 2016 compared to irrigation statistics, respectively. The method approximates irrigated areas in regions with large agricultural holdings better than in regions with small fragmented agricultural holdings, due to binary classification and the choice of spatial resolution. The irrigated areas are mainly identified in the established areas indicated in the existing irrigation maps. A comparison with global datasets reveals different disparities due to spatial resolution, input data, reference period, and processing techniques. From multiyear analysis, it is evident that irrigation extent is positively correlated with precipitation (r = 0.73, p-value = 0.0163) and less with potential evapotranspiration.
- Preprint
(14361 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2024-1929', Anonymous Referee #1, 06 Aug 2024
- AC1: 'Reply on RC1', Devi Purnamasari, 28 Aug 2024
-
RC2: 'Comment on egusphere-2024-1929', Anonymous Referee #2, 08 Aug 2024
The article proposes a method for identifying irrigation areas using differences between modeled and satellite-derived land surface temperature (LST) together with hydrological modeling. The irrigated areas are detected from LST differences using a random forest method. LST is modeled from a surface energy balance using the evapotranspiration (latent heat flux) derived from the wflow_sbm hydrological model. The methodology is implemented in the Rhine basin, and the results are compared against four existing global irrigation maps and regional statistics of irrigated areas. The article's topic is timely, relevant for the HESS journal and holds significant interest for the scientific community. The manuscript is clear, well-written, and offers a broad perspective on the discussions. I think the article should be considered for publication after major revision addressing the issues below.
I believe the use of LST differences for irrigation mapping needs better justification. It raises the question of why not directly use satellite-based retrievals and evapotranspiration (ET) derived from the hydrological model to detect irrigation, instead of reverting to simulate LST from energy balance. In other words, existing methods for estimating actual irrigation could be used to identify irrigated areas simply by masking where irrigation is detected. For example, Olivera-Guerra et al. (2020, https://doi.org/10.1016/j.rse.2019.111627) used the coupling between an energy and water balance model to estimate irrigation, which was evaluated in both non-irrigated and irrigated fields. Although it is argued that errors in ET retrievals may hinder irrigation mapping, the errors involved in both satellite-based and modeled LST are equally significant. Additionally, the use of LST-derived products (e.g., ET, root-zone soil moisture, water stress) in estimating or detecting irrigation should be introduced and discussed in the introduction section, as shown by some studies (Droogers et al. 2010, https://doi.org/10.1016/j.agwat.2010.03.017; Olivera-Guerra et al. 2018, 2020, https://doi.org/10.1016/j.agwat.2018.06.014; Chen et al. 2018, https://doi.org/10.1016/j.rse.2017.10.030). Without this context, the use of LST is presented as the key point and the novelty in estimating irrigation. Therefore, I believe the novelty in using LST to detect irrigated areas should be well justified.
Another important point to deepen is the use of LST in wet condition (humid regions or wet years in the study area). It would be interesting to analyze differences in the classification of irrigated areas in dry and wet years to draw more conclusions about the use of LST in such conditions. For example, differences in LST or ET are more important in dry years (i.e., water-limited regimes) than in wet years (energy-limited regimes), particularly in dry years with the presence of fields where the crop water requirement is fully supplied to avoid water stress. Therefore, irrigated areas would be easier to detect in drier conditions, while more errors are likely in wet conditions (energy-limited regimes).
According to Lines 301-304, the fact that the model trained with data from a specific year cannot be used to identify irrigated areas for the entire study period could justify the use of existing models for estimating irrigation and consequently detecting irrigated areas, rather than relying on LST differences. Comparing irrigation mapping using LST differences and ET differences should be performed for further analysis. Such analysis would allow for a more robust justification of the use of LST for irrigation mapping.
Lines 421-429. The limitations of LST in humid regions should be discussed. Even though decreased precipitation may lead to reductions in the extent of irrigated areas during the driest years, particularly in semi-arid regions (e.g., Afghanistan and the Ebro basin), this may not necessarily be the case in more humid regions where precipitation amounts are still substantial, such as the Rhine basin. In wet conditions, detection of irrigation using LST becomes more challenging and errors are more likely, leading to potential compensations that hamper the establishment of a clear relationship between precipitation and irrigated areas. Therefore, further evaluations should be carried out. For instance, Appendix B confirms that less precipitation leads to more irrigated areas, as detection is more easily captured by LST and more areas require irrigation.
Other comments
Lines 40-42. I would recommend delving deeper into the irrigation detection in diverse climates, discussing the advantages of using LST in semi-arid to arid regions and the challenges in temperate to humid climates under an energy-limited regime.
Lines 42-44 are not in context with irrigation retrievals in diverse climates.
Line 63. Add references of existing approaches.
Line 80. The wflow_sbm should be previously introduced.
Line 139. PTFs?
Section 2.3. Since the LST module is simply the inversion of the energy balance equations, I would recommend moving most of the equations related to the energy balance (for example, those from lines 160-179 and 186-197) to the appendix. This would give more prominence to the irrigation mapping methodology.
Section 2.4.2. Classification based on visual detection is prone to errors and should be evaluated accordingly. Are there irrigated plots available to assess the classification?
Lines 245-248. The presence of neighboring land cover types (floodplains and forests as mentioned by authors) may also influence agricultural fields. It would be interesting to evaluate their impact on both the classification of irrigated/non-irrigated areas and the LST itself.
Figure 6. Reduce the range of the second y-axis to see more details in LST differences. Change this y-axis label to “Temperature difference”.
Figure 7. Why negative differences are obtained in irrigated crops between observed and modelled LST ? How is that related to possible misclassifications between irrigate/non-irrigated fields. Change the y-axis label to “Temperature difference”.
Figure 8. What are the reasons of the large underestimation of DE24 in 2013 and the overestimation in DE73 in 2016. The year could be added as title to each plot.
Line 359. Recall the hectares of the estimated irrigated areas.
Figure 10. Correct the caption of the figure (a, b and c).
Line 375. east of the border ?
Figure 11. Add the region (Lower, Middle and Rhine valley) as title of each figure and in the caption of the figure.
Figure 12. The period of representation per irrigation map could be add to the tittle of each figure.
Citation: https://doi.org/10.5194/egusphere-2024-1929-RC2 - AC2: 'Reply on RC2', Devi Purnamasari, 08 Oct 2024
Data sets
LST observations from MODIS Zhengming Wan https://lpdaac.usgs.gov
Shortwave radiation LSA-SAF https://datalsasaf.lsasvcs.ipma.pt/PRODUCTS/MSG/MDIDSSF/NETCDF/
Surface albedo LSA-SAF https://datalsasaf.lsasvcs.ipma.pt/PRODUCTS/MSG/MDAL/NETCDF/
Eurostat irrigation data Eurostat https://ec.europa.eu/eurostat/databrowser/view/EF_POIRRIG/default/table?lang=en
Model code and software
Code for processing dataset Devi Purnamasari https://github.com/dvprnmsr/irrigation_paper
Interactive computing environment
Code for processing dataset Devi Purnamasari https://github.com/dvprnmsr/irrigation_paper
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 330 | 101 | 58 | 489 | 8 | 10 |
- HTML: 330
- PDF: 101
- XML: 58
- Total: 489
- BibTeX: 8
- EndNote: 10
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1