the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
What determines peat swamp vegetation type in the Central Congo Basin?
Abstract. The Central Congo Basin is home to the largest peat swamp in the tropics. Two major vegetation types overlay the peat: hardwood trees, and palms (mostly the trunkless Raphia laurentii variety), with each dominant in different locations. The cause of the location of these differently composed swamp areas is not understood. We investigated their distribution using a recent vegetation classification across the 165,600 km2 region. Using a regression model we assessed the impacts of elevation, seasonal rainfall and temperature on the presence of each peat vegetation type. We used monthly 0.05° resolution CHIRPS rainfall climatology (CHPclim) and maximum temperature (CHIRTS) data together with 90 m resolution terrain data (MERIT Hydro). Our model was successful in predicting the percentage palm swamp composition when tested using data held back for verification, with R2 ~ 0.79, RMSE = 14.8 %, and p < 0.05 for the largely rain-fed hydrological sub-basins. However, it did not perform well in areas where peatland inundation is controlled by river flooding. We found that palm swamp composition varies primarily with elevation and dry season climatological variables (rainfall and temperature), with additional, significant contributions from the total wet season rainfall and temperature. There are indications of an optimal range of net water availability (the difference between rainfall and actual evapotranspiration, accounting for run-off) for palm swamp dominance, above and below which hardwood swamp dominates. In this study we progress our understanding of the determinants of peat swamp vegetation type in the central Congo Basin. Improved understanding will contribute to assessing how changes in environmental factors, including land-use and climate change impacts, could impact swamp type distribution and carbon fluxes in the future.
- Preprint
(10331 KB) - Metadata XML
-
Supplement
(1340 KB) - BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on egusphere-2022-580', Anonymous Referee #1, 06 Sep 2022
The manuscript concentrates on the Central Congo Basin, aiming at to distinguish the reasons why certain regions of the peat swamps are covered by hardwood trees, and other by palms. This problem is tackled by modelling their known distributions by using elevation and meteorological data. As conclusions, the authors detect a range of water supply which enables palm swamps to exist, whereas outside of the range hardwood trees will dominate.
Generally speaking, Cental Congo Basin or its vegetation types are not particularly familiar for me, but as such the premises, applied data, analyses and conclusions appear to be sound and well justified. Language of the manuscript is also good and requires no particular modifications. In addition, storyline is clear and the text itself reads well, which is not the case for all the manuscripts. Some of the chapters are rather long and detailes, particularly results and discussion. But for someone interested in this specific topic, this may be a gread advantage. Considerations as included in the conclusions are also detailed and sound justified.
I have no major concerns regarding to the manuscript, only a few detailed observations which may deserve to be addressed when revising it:
Row 30: maybe reference to Fig. 1 could be on the row where CC is mentioned for the first time (27)
Row 34: increases the carbon stock; does this refer to situation that Cuvette Centrale wouldn't exist? Wouldn't this be easier to say as a proportion of the total carbon stock?
Row 91: as Crezee et al. (2022) land classification map is a data of high importance in this paper, it would be fair to describe a bit of how it was constructed (as well as acknowledging its potential sources of error, which may also affect on e.g. detected anomalies)
Rows 167-171: I'm not totally convinced of the use of STD in this context; it kind of reflects the uncertainty or inaccuracy of the rainfall estimate, but won't indicate the direction of it. Moreover, high STD may reflect for example a hill or a pit; in the first case it'll probably increase the runoff from the pixel to its neigbours, and in the latter from neighbours to the target pixel. I'm not necessarily suggesting to reject this model term, but use of it is not totally justified, as it won't necessarily indicate any particular tendency per se.
Row 363: what is a "blackwater river"?
Row 419: I'm not sure if "contribute significantly" is the best way to say here; rather, they enable to model the vegetation types at a reasonable accuracyCitation: https://doi.org/10.5194/egusphere-2022-580-RC1 - AC1: 'Reply on RC1', Selena Georgiou, 29 Sep 2022
-
RC2: 'Comment on egusphere-2022-580', Anonymous Referee #2, 14 Sep 2022
General comments:
The manuscript by Georgiou et al addresses the question what controls the distribution of vegetation types in the Congo Basin peat swamp. The peat swamp has only been recently mapped based on field and remote sensing data. I found it very interesting to read the manuscript and think about what the distribution of hardwood trees and palms could tell us about hydrological processes and climatological boundary conditions. The topic is highly relevant given the threats the Congo peatland is currently facing due to regional anthropogenic alterations of the carbon and water cycle and global climate change. However, I do see two fundamental problems in the methodological approach of the manuscript. Addressing those problems may fundamentally change the findings of the manuscript.
(1) Ground truth data
The manuscript uses as ground truth data the mapping product of Crezee et al. 2022. The supplementary Figure 1 of the paper by Crezee et al. shows the nine remote-sensing products that were used to map peat-associated vegetation, i.e. the ground truth data used here in the work of Georgiou et al. Three of the nine input variables were based on elevation data. The fact that detailed elevation data was already used in the generation of the ground truth data conceptually prohibits that in Georgiou et al. elevation data is again used to build a regression model. In Georgiou et al., it is found that peat swamp vegetation is mainly a function of elevation. Knowing that the ground truth data was already created with elevation data makes this a trivial finding. Any discussion in Georgiou et al. on the influence of elevation-based variables is far-fetched given this fundamental problem of the ground truth data. To analyze the influence of elevation, the authors would need to work with ground truth data that is e.g. solely based on optical and microwave satellite signatures, but not on elevation.
(2) Division into sub-basins and random cross validation
The distribution of hardwood trees and palm shows patterns with clear spatial autocorrelation structure. The authors ignored this structure in their 'random' cross-validation approach at sub-basin scale, and thus seriously underestimated predictive error and likely have built overfitted models with non-causal predictors. For details I refer to the highly cited methodological paper of Roberts et al. 2017 on data structure and cross validation (see below). The derived models at sub-basin scale that use, apart from elevation, many different types of climatological-based variables are therefore highly questionable. The authors would need to show that the proposed climatological variables are reliable in a stratified cross-validation that acknowledges the spatial auto-correlation of the data. I believe that this would require an aggregation of sub-basins into larger regions. Perhaps one model for RoC and one for DRC in which one perhaps e.g. stratify the cross-validation by sub-basins (= not building a model for each sub-basin but building a model for four sub-basins and cross-validate against the fifth). Only variables that survive as reliable predictors in such a stratified cross-validation could be used as basis for an interpretation of optimal vegetation conditions
Detailed comments:
Line 35:
Harmonize use of Pg C and Gt C in the paper.
Line 98:
A useful variable might be the 'topographic wetness index' that combines subbasin area and local slope to estimate ground- and surface water impacts on soil wetness (e.g. Kopecky et al. 2021).
Line 163-164:
Sentence unclear
Line 210:
It's not 'train-test' since “test” data needs to be independent. With a random sampling, test data points are spatially auto-correlated with training points, thus they are not independent.
Line 261:
Also for RoC sub-basins not all show a positive correlation b/w palm fraction and annual rainfall (Roc5 show negative correlation)
Figure 6:
spatial variation of precipiation in RoC is only 100 mm, ~ 6-7%. In this example, it's quite likely that this trend will prove unreliable in a stratified cross validation.
Line 455:
Is there any physiological indication why palms should be less able to tolerate wetness than hardwood trees? Based on the methodological problems of the study, I found the discussion on the optimal water amounts for palms based on the negative correlation of palms with rainfall far-fetched.
Roberts et al. 2017. Cross-validation strategies for data with temporal, spatial, hierarchical, or phylogenetic structure. Ecography 40: 913–929. doi: 10.1111/ecog.02881
Kopecky et al. 2021. Topographic Wetness Index calculation guidelines based on measured soil moisture and plant species composition. Science of the Total Environment. doi.org/10.1016/j.scitotenv.2020.143785
Citation: https://doi.org/10.5194/egusphere-2022-580-RC2 - AC2: 'Reply on RC2', Selena Georgiou, 29 Sep 2022
Status: closed
-
RC1: 'Comment on egusphere-2022-580', Anonymous Referee #1, 06 Sep 2022
The manuscript concentrates on the Central Congo Basin, aiming at to distinguish the reasons why certain regions of the peat swamps are covered by hardwood trees, and other by palms. This problem is tackled by modelling their known distributions by using elevation and meteorological data. As conclusions, the authors detect a range of water supply which enables palm swamps to exist, whereas outside of the range hardwood trees will dominate.
Generally speaking, Cental Congo Basin or its vegetation types are not particularly familiar for me, but as such the premises, applied data, analyses and conclusions appear to be sound and well justified. Language of the manuscript is also good and requires no particular modifications. In addition, storyline is clear and the text itself reads well, which is not the case for all the manuscripts. Some of the chapters are rather long and detailes, particularly results and discussion. But for someone interested in this specific topic, this may be a gread advantage. Considerations as included in the conclusions are also detailed and sound justified.
I have no major concerns regarding to the manuscript, only a few detailed observations which may deserve to be addressed when revising it:
Row 30: maybe reference to Fig. 1 could be on the row where CC is mentioned for the first time (27)
Row 34: increases the carbon stock; does this refer to situation that Cuvette Centrale wouldn't exist? Wouldn't this be easier to say as a proportion of the total carbon stock?
Row 91: as Crezee et al. (2022) land classification map is a data of high importance in this paper, it would be fair to describe a bit of how it was constructed (as well as acknowledging its potential sources of error, which may also affect on e.g. detected anomalies)
Rows 167-171: I'm not totally convinced of the use of STD in this context; it kind of reflects the uncertainty or inaccuracy of the rainfall estimate, but won't indicate the direction of it. Moreover, high STD may reflect for example a hill or a pit; in the first case it'll probably increase the runoff from the pixel to its neigbours, and in the latter from neighbours to the target pixel. I'm not necessarily suggesting to reject this model term, but use of it is not totally justified, as it won't necessarily indicate any particular tendency per se.
Row 363: what is a "blackwater river"?
Row 419: I'm not sure if "contribute significantly" is the best way to say here; rather, they enable to model the vegetation types at a reasonable accuracyCitation: https://doi.org/10.5194/egusphere-2022-580-RC1 - AC1: 'Reply on RC1', Selena Georgiou, 29 Sep 2022
-
RC2: 'Comment on egusphere-2022-580', Anonymous Referee #2, 14 Sep 2022
General comments:
The manuscript by Georgiou et al addresses the question what controls the distribution of vegetation types in the Congo Basin peat swamp. The peat swamp has only been recently mapped based on field and remote sensing data. I found it very interesting to read the manuscript and think about what the distribution of hardwood trees and palms could tell us about hydrological processes and climatological boundary conditions. The topic is highly relevant given the threats the Congo peatland is currently facing due to regional anthropogenic alterations of the carbon and water cycle and global climate change. However, I do see two fundamental problems in the methodological approach of the manuscript. Addressing those problems may fundamentally change the findings of the manuscript.
(1) Ground truth data
The manuscript uses as ground truth data the mapping product of Crezee et al. 2022. The supplementary Figure 1 of the paper by Crezee et al. shows the nine remote-sensing products that were used to map peat-associated vegetation, i.e. the ground truth data used here in the work of Georgiou et al. Three of the nine input variables were based on elevation data. The fact that detailed elevation data was already used in the generation of the ground truth data conceptually prohibits that in Georgiou et al. elevation data is again used to build a regression model. In Georgiou et al., it is found that peat swamp vegetation is mainly a function of elevation. Knowing that the ground truth data was already created with elevation data makes this a trivial finding. Any discussion in Georgiou et al. on the influence of elevation-based variables is far-fetched given this fundamental problem of the ground truth data. To analyze the influence of elevation, the authors would need to work with ground truth data that is e.g. solely based on optical and microwave satellite signatures, but not on elevation.
(2) Division into sub-basins and random cross validation
The distribution of hardwood trees and palm shows patterns with clear spatial autocorrelation structure. The authors ignored this structure in their 'random' cross-validation approach at sub-basin scale, and thus seriously underestimated predictive error and likely have built overfitted models with non-causal predictors. For details I refer to the highly cited methodological paper of Roberts et al. 2017 on data structure and cross validation (see below). The derived models at sub-basin scale that use, apart from elevation, many different types of climatological-based variables are therefore highly questionable. The authors would need to show that the proposed climatological variables are reliable in a stratified cross-validation that acknowledges the spatial auto-correlation of the data. I believe that this would require an aggregation of sub-basins into larger regions. Perhaps one model for RoC and one for DRC in which one perhaps e.g. stratify the cross-validation by sub-basins (= not building a model for each sub-basin but building a model for four sub-basins and cross-validate against the fifth). Only variables that survive as reliable predictors in such a stratified cross-validation could be used as basis for an interpretation of optimal vegetation conditions
Detailed comments:
Line 35:
Harmonize use of Pg C and Gt C in the paper.
Line 98:
A useful variable might be the 'topographic wetness index' that combines subbasin area and local slope to estimate ground- and surface water impacts on soil wetness (e.g. Kopecky et al. 2021).
Line 163-164:
Sentence unclear
Line 210:
It's not 'train-test' since “test” data needs to be independent. With a random sampling, test data points are spatially auto-correlated with training points, thus they are not independent.
Line 261:
Also for RoC sub-basins not all show a positive correlation b/w palm fraction and annual rainfall (Roc5 show negative correlation)
Figure 6:
spatial variation of precipiation in RoC is only 100 mm, ~ 6-7%. In this example, it's quite likely that this trend will prove unreliable in a stratified cross validation.
Line 455:
Is there any physiological indication why palms should be less able to tolerate wetness than hardwood trees? Based on the methodological problems of the study, I found the discussion on the optimal water amounts for palms based on the negative correlation of palms with rainfall far-fetched.
Roberts et al. 2017. Cross-validation strategies for data with temporal, spatial, hierarchical, or phylogenetic structure. Ecography 40: 913–929. doi: 10.1111/ecog.02881
Kopecky et al. 2021. Topographic Wetness Index calculation guidelines based on measured soil moisture and plant species composition. Science of the Total Environment. doi.org/10.1016/j.scitotenv.2020.143785
Citation: https://doi.org/10.5194/egusphere-2022-580-RC2 - AC2: 'Reply on RC2', Selena Georgiou, 29 Sep 2022
Viewed
HTML | XML | Total | Supplement | BibTeX | EndNote | |
---|---|---|---|---|---|---|
641 | 347 | 38 | 1,026 | 101 | 29 | 26 |
- HTML: 641
- PDF: 347
- XML: 38
- Total: 1,026
- Supplement: 101
- BibTeX: 29
- EndNote: 26
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1