Review Reports

Round 1

Reviewer 1 Report

remotesensing-1637617-peer-review-v1

Review 11 March 2022

Summary

This manuscript analyzes the impact of the COVID-19 lockdown on the drought in the Vidarbha region in central India. The authors take advantage of the large scale temporary change in farmers’ activities during the nation-wide lockdown from March 24 to May 31, 2020 as a test bed to try disentangling the sensitivity of the total water storage to the climate and to policy measures. It must be mentioned that the subject is highly relevant due to the dramatic societal consequence of the repeated droughts in this region – the number of farmer suicides due to crop failures.

The methodology is based on the collection of terrestrial water storage anomalies (TSWA) over this region from the Gravity Recovery and Climate Experiment (GRACE) between April 2002 and August 2020. Other large scale time-series are analyzed: precipitation from the Climate Hazards group Infrared Precipitation with Stations (CHIRPS), evapotranspiration from Global Land Data Assimilation System (GLDAS) with two models but also Normalized Difference Vegetation Index (NDVI). Another track is also followed based on the stochastic forecast of TWSA from November 2019 to May 2020, based on the series up to October 2019 so to test a “business as usual” scenario to be compared with the lockdown affected observation.

General comments

The subject is very relevant, the methodology is consistent, the manuscript is well written using appropriate figures and complemented by a reference list encompassing all the aspects of this study. I have nothing to add concerning the long-term trends. I would just have a suggestion regarding the structure of the paper and bunch of comments/questions about the forecast experiment.

In the Discussion section, the sub-section 4.1 fits to the discussion label. The 4.2 sub-section is about NDVI data and results and discussion. Maybe the data could be introduced earlier. The 4.3 sub-section is mainly presenting the results of the forecast experiment and the discussion begins properly at line 357 –I think. The paper would benefit from a slight re-balance.

The rest of my comments are wandering around the forecast experiment. “The indication of an overall TWSA rise of 3.65 to 19.32 km³” is written 3 times in the manuscript. Therefore, I tried to understand where these values came from. I noticed the “and” in the caption of figure 7 (“Comparison of forecast performances in November 2019 and May 2020”) and I remembered that the domain extended over 3°x3° which I roughly evaluated as 102,410 .km². Then reading on Figure 7 that the best model, STL, overestimated TWSA by ~3.3 cm in November 2019 and underestimated by another 3.3 cm in May 2020, I found a net rise of 6.7 cm corresponding to a volume of 6.3 km³ increase from a business as usual scenario. As for the worst forecast, NNAR, it is underestimating both in November (~7cm) and in May (~16 cm). I came to a volume of about 9.2 km³. This ranges from twice the lower bound of the manuscript to half its upper bound. So, what’s wrong in the above? If we select only the two models scoring best, the rise is even smaller.

A whole battery of statistical scores is used to assess the performances of the 6 forecasting techniques. If I understand well, each test consists in comparing GRACE observations (“Original”) of November 2019 to May 2020 with the corresponding forecast values. Therefore, each test is considering one single realization and “n” in equations 3 to 8 is just n=7. Doing so, the forecast horizons 1 month to 7 months are pooled and the conclusions made about May 2020 actually could benefit from the better skill during the preceding months.

Any stochastic forecasting procedure allows to estimate the uncertainty along with the forecast. This uncertainty should be reported in Figure 7. In an ideal setting (many realizations), this uncertainty would probably increase quickly from the forecast horizon of 1 month (November 2019) to 7 months (May 2020). On Figure 2b and 4, we can see the series have many gaps and the worst is during 2018 between the end of the mission and the begin of the follow on mission. The stochastic forecasting models are run after the gaps have been filled by linear interpolation i.e. with no new information included. This may explain why the forecasts appear so smooth compared to the observed series.

The experiment could be modified by considering a shorter learning period and a shorter forecast horizon. It would be possible to repeat the forecast experiment and estimate the scores separately for the forecast horizons (the farther being May 2020 for one run, April 2020 for another, etc.). Of course, this still far from an ideal setting let alone the fact that those realizations would not be all independent from each other. In conclusion, I think that something has to be done for this experiment; Figure 7 is too weak and the warning that it is difficult to separate out the influence of lockdown on TWSA rise is not enough.

Specific comments

L46 Give the correct number of deaths.

L91 Introduce briefly what a “mascon” block is.

L112-122 It isn’t clear who is doing what. As I understand native pixel are 3°x 3°. They are somhow downscaled to 0.5°x 0.5° by NASA JPL (?). Who is using a priori information (and which information) for doing what? As for “ranging” data and “degree 2 and order 0” coefficients, the reader has no way but read also Watkins et al 2015. The authors used scale factors at each pixel data. I understand at each 0.5°x 0.5° pixel but I would need some clue about what was done. Same for anomalies: are the TSW values up-scaled back to 3°x 3° before removing the long-term mean?

L265 & L298 Both sub-sections have the same title.

L346 considering

Table 1, a negative value of MAE is given for the STL forecast.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

In this manuscript, authors have assessed the replenishment of terrestrial water storage over a drought prone region of India at both short-term and long-term. This study appears to be interesting, but needs few clarifications.

They used GRACE data for TWS estimates, which has only one native grid over the study region. It would be beneficial if authors could support it through few in-situ (well) observations. 

Authors should also provide details of scale factors used for TWS computation.

Sections 4.1 and 4.2 have surprisingly the same title.

It was already shown that rainfall has a lead/lag relation with TWS anomalies, but the long-term changes in precipitation has usually not linked with TWS changes (e.g., Changes in terrestrial water storage versus rainfall and discharges in the Amazon basin; An assessment of terrestrial water storage, rainfall and river discharge over Northern India from satellite data).

Authors should look into irrigation pattern changes and its association with TWS trend.

Why authors have not considered satellite-based ET which has finer spatial resolution?

How to quantify the changes in TWS due to human intervention or environmental/atmospheric processes at both long and short time scales?

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

The authors have replied satisfyingly to my comments. Concerning the forecasts, it took me some effort to catch their point. I think that they should add to the main text some additional explanations given in the response: "Overall TWSA rise of 3.65 to 19.32 km³ is estimated based on May-2020 values only." They should also modify the caption of the (new) Figure 7: Observed (GRACE) and forecast (six stochastic models) of TWSA from November 2019 to May 2020 ... Otherwise we compare the differences between the values of November and May. (Note that this would be a kind of post-processing of the forecasts)

There are some slight corrections:

L46 The “,” are misplaced, I think; do you mean 5.13 million deaths?

L16, … 12 instances of “forecasted” to be changed into “forecast”

L304 considering

L311 ExponenTial

Author Response

Rev 1. Comment 1: The authors have replied satisfyingly to my comments. Concerning the forecasts, it took me some effort to catch their point. I think that they should add to the main text some additional explanations given in the response: "Overall TWSA rise of 3.65 to 19.32 km³ is estimated based on May-2020 values only." They should also modify the caption of the (new) Figure 7: Observed (GRACE) and forecast (six stochastic models) of TWSA from November 2019 to May 2020 ... Otherwise we compare the differences between the values of November and May. (Note that this would be a kind of post-processing of the forecasts)

Reply: Thanks for your careful observation of the manuscript. We believe the manuscript had much improved upon answering to your comments.

Thanks for the concern. We included the additional explanation as: " Our analyses indicate an overall TWSA rise of 3.65 to 19.32 km³ (estimated based on May-2020 values only), which is most likely linked with COVID linked lockdown in the study area."

Following the reviewer's suggestion, we modified the Figure 7 caption as: "Figure 7: Comparison of forecast performances between observed (GRACE) and forecast (six stochastic models) of TWSA from November 2019 to May 2020. The following stochastic models are used: (a) ARIMA (Autoregressive integrated moving average); (b) ETS (Error, Trend, Seasonal or the Expotential Smoothing); (c) NNAR (Neural Network Autoregression); (d) Reg. based (time-series regression based); (e) STL (Seasonal and trend decomposition through Loess smoother) and (f) TBATS (T for trigonometric regressors to model multiple-seasonalities; B for Box-Cox transformations; A for ARMA errors; T for trend; S for seasonality). 95% confidence interval is shown as patterns"

Rev 1. Comment 2: L46 The “,” are misplaced, I think; do you mean 5.13 million deaths?

Reply: We have corrected that. The comma should be only once here: 513,226 deaths. Thanks.

Rev 1. Comment 3: L16, … 12 instances of “forecasted” to be changed into “forecast”.

Reply: Thanks for pointing this out. We have corrected that.

Rev 1. Comment 4: L304 considering

Reply: Thanks, corrected.

Rev 1. Comment 5: L311 ExponenTial

Reply: Thanks, modified.

Reviewer 2 Report

The authors have addressed my comments satisfactorily and the revised manuscript reads better.  I recommend it for acceptance for possible publication.

Author Response

We would like to thank the reviewer for his/her interest in our work and also for the careful consideration of the manuscript.