Review – egusphere-2023-1171

Myers et al :  Seasonal variation in landcover estimates reveals sensitivities and opportunities for environmental models

The authors present results from a study exploring the impact of using temporally resolved land cover data for water quality and hydrological model development. The manuscript reads reasonably well but there are some structural issues that need to be addressed (see specific comments below). My other concerns are relatively minor and are related to the presentation and discussion of results. I believe this manuscript could represent a useful contribution to our understanding of model uncertainty associated with seasonal landcover changes. While, the manuscript, in current form is not suitable for publication, with relatively minor revisions this study would be suitable and of broad interest to the readership of HESS.

Specific comments:

Line 16: Qualify the potential implications on land cover characterization here.

Line 55: It seems like you need a research question/hypothesis linked to the spatiotemporal variability in land cover quantification. Where certain types of catchments more likely to exhibit large seasonal shifts in land cover quantification?

Figure 3. The differences in landcover between seasons seem most pronounced for 2016. This is interesting and isn’t explored in the manuscript. This is important as you are using 2016 (i.e. national landcover database) as the reference and if this was a particularly anomalous year could there be implications for your conclusions? Perhaps think about exploring the drivers of inter annual variability here (e.g. climate drivers).

Table 2. I’m not sure on the relevance of presenting the AIC score here? This is normally used for model selection –wouldn’t the RMSE of the fitted values be a more useful indicator of differences between the land cover quantification methods.

Figure 4. Colour contrast makes it difficult to view the different lines/points. Perhaps consider using a different palette with stronger contrasts? Also this looks like a quadratic relationship rather than linear?

Line 265: Move figure from supplementary material to support discussion of the implications for model parameters. I think this is an important part of the manuscript and should be given more prominence.

Conclusion – avoid excessive referencing to other studies in the conclusion section.

Lines 300 -315: Rather than a bullet point list I suggest you develop a more coherent narrative focused on the implications of your findings and future research directions. This could go before the conclusion section.

Review for manuscript entitled "Seasonal variation in landcover estimates reveals sensitivities and opportunities for environmental models"

In this manuscript, the authors investigate the difference of classifying a land-use land cover (LULC) dataset during the growing vs. non-growing season. They use the Dynamic world dataset for this purpose and find differences in classification of 5 to 10% area in built and tree classes. In other words, there is 5 to 10% more(less) area classified as built(trees) in the non-growing season compared to the growing season. Subsequently, they investigate the effect of this difference on water quantity and quality modelling. For water quality modelling, the authors use both a statistical approach and the hydrologic model SWAT. The latter is also used for water quantity modelling. The selected 37 basins are located in the Eastern US and SWAT is set up for two small scale basins. Overall, the authors find small differences when using the classification during the growing season and non-growing season. The largest difference is found when SWAT parameters calibrated on streamflow and nitrogen yield data using the growing season data are transferred to using the non-growing season data. However, a calibration of SWAT in another catchment shows that similar performance can be achieved using the growing season and non-growing season data. This highlights that model parameters do compensate for differences in LULC maps in SWAT, an aspect that is not discussed in the manuscript. While the topic of LULC classification is very relevant because land cover information is a fundamental requirement for environmental models, the conclusions of this study are very limited and not of general nature because of two reasons. First, the limited spatial coverage only focusing on small basins in the Eastern US. Second, and more importantly, the choice of methods, i.e., using the SWAT model that only accounts for the dominant land cover class (l. 79) within a computational unit. Accounting for dominant land cover is not state-of-the-art. For example, state-of-the-art LSMs like the Community land model v5 use a mosaic approach to represent different land cover types and plant functional types within a grid cell (see also comment on CLM below). These models are also able to have transient land cover, which would account for the differences between growing and non-growing season. SWAT also does not account for relationships between model parameters and land cover. Within hydrologic modelling, regionalization methods have been developed that relate model parameters to land cover, which reduces the reduction in performance when transferring parameters from one land cover map to another (Samaniego et al. 2010). For these reasons, I do not find the results of this study are substantial enough to be interesting to a wider readership and recommend rejection.

The comments below might help the authors to further improve their manuscript in the continuation of their work.

Major comments

Overall, the results and discussion section is very short, only 120 lines! The authors should discuss how the differences in LULC classification that they find here would impact simulations using other methods than the ones used here (i.e., SWAT). To my understanding, the reduction in performance presented in this manuscript using SWAT provide an upper bound and different methods (i.e., CLM5 and methods outlined in Samaniego et al. 2010) would actually be less sensitive to the difference in LULC classification.

Case \#2: Parameters transferred from growing to non-growing season lead to substantial drop in model performance, but how do parameters calibrated to non-growing season perform when using the growing season data? The authors should include this case in their analysis.

L. 242.: The reference to Clark et al. (2015) is misleading. State-of-the-art land-surface models like CLMv5 are able to account for transient land use and land cover change (see chapter 27 in CLM5.0 technical note, downloaded from https://www2.cesm.ucar.edu/models/cesm2/land/CLM50_Tech_Note.pdf). The authors need to discuss how their findings here are relevant for land-surface models that are employing a mosaic approach to represent different land and plant functional types within a grid cell.

Table 4: The performance metrics are often higher in the validation than in the calibration period. Normally, it is expected that there is a substantial drop in model performance between calibration and validation period. The authors should explain this.

Minor comments:

Section 2.5: Not all readers will be familiar with SWAT parameters listed in Table 1. It would be helpful to add a short description in the Appendix.

Table 3: The caption should mention that all results presented here are based on parameters calibrated with the growing season LULC (see l. 213f), even the results shown for the calibration period of the non-growing season. This could be misunderstood otherwise.

Fig. 6: The visual quality of this Figure is low. In Figure 6a to 6c, the markers in the Scatter plots are overlapping and the underlying relationship between the simulated and observed discharge cannot be seen. Also the dpi resolution of the hydrographs is low leading to pixelization that does not allow to appreciate differences between different lines. I also suggest to use a black line instead of a yellow line and make the line widths slightly thicker.

References:

Samaniego, L., R. Kumar, and S. Attinger (2010), Multiscale parameter regionalization of a grid‐based hydrologicmodel at the mesoscale,Water Resour. Res.,46, W05523, doi:10.1029/2008WR007327