A Global Multiscale SPEI Dataset under an Ensemble Approach
Round 1
Reviewer 1 Report
1. Please add more informations about why you select Hargreaves-Samani and Thornthwaite to compute PET.
2. Please explain why are you did't compute historical period 1960-1999 based on ESMs/GCMs VS based on WFD.
Author Response
Please see the attachment
Author Response File: 
Author Response.docx
Reviewer 2 Report
The authors presented a global dataset of drought for future simulations using CMIP5 data. The topic seems to be interesting and very relevant to several implications. However, I believe that the manuscript cannot be accepted for publication in its present form for the following reasons:
- The novelty/contribution of this work is questionable recalling the presence of different studies/datasets that evaluate drought characteristics under a future changing climate. See for example:
Philosophical Transactions of the Royal Society A. 2022;380(2238):2021028
Geophysical Research Letters. 2020;47(11):e2020GL087820.
Water. 2019;11(2):312.
Geophysical Research Letters. 2018;45(21):11,913-11,20.
Current Climate Change Reports. 2018;4(2):164-79.
Climatic Change. 2017;144(3):535-48.
Climatic Change. 2017;144(3):519-33.
Regional Environmental Change. 2014;14(5):1907-19.
- I am wondering why the historical dataset was restricted only to 1999, while data is available to 2005.
- The rationale behind validating the historical dataset against a set of products (e.g. CRU, GPCP, etc.) is unclear, recalling that the model data is already bias-corrected. Also, what is the rationale behind using this high number (9) of accuracy metrics, and how the values of these metrics are interpreted? In the same context, it is recommended to use a normalized version of the RMSE, as the conventional RMSE is sensitive to the presence of outliers/extremes in the data, which is common for some air temperatures and precipitation in some regions worldwide.
- Also, I am wondering why the authors have only relied on 6 members of the CMIP5 achieve, while there is a high number of models available. This low sample size can be a source of uncertainty in the obtained results.
- It is recommended to compare the PET results obtained using Hargreaves-Samani against the Thornthwaite method.
- The definition of a drought event is unclear in this work. Have you considered an SPEI value below 0? Or other thresholds? This should be clarified and a clear definition of the drought metrics (e.g. magnitude, intensity, frequency, and duration) should be included.
- It is unclear why the authors adopted precipitation data to log-logistic distribution. I mean that there is always a degree of uncertainty in the calculation of the SPI/SPEI due to the effects of probability distributions and parameter errors. For example, in the calculation of SPI, McKee et al. (1993), the proposers of this index, suggested using the Gamma distribution to fit the cumulative precipitation. However, scholars including Hong et al. (2013), Cindrić et al. (2012), Gabriel and Monica (2015), Vergni et al. (2017) indicated that the applicability of different theoretical distributions in SPI is inconsistent across different regions. Specifically, Guttman et al. (1999) verified that the Pearson Type III distribution is a better universal model in America; Sienz et al. (2012) concluded that the Weibull type distributions give distinctly improved fits compared to the Gamma in Europe; Gabriel and Monica (2015) demonstrated that the Generalized Normal distribution presents the best performance in fitting the precipitation series in Brazil. This is typically the case for SPEI as well.
Journal of Water Resources Research. 2013, 2, 33-41, doi:10.12677/jwrr.2013.21006.
EMS. 2012, GF48: EMS2012-316.
J. Revista Brasileira de Engenharia Agrícola e Ambiental – Agriambi. 2015,
19, 1129-1135, doi:10.1590/1807-1929/agriambi.v19n12p1129-1135.
J. Theoretical and Applied Climatology. 2017, 128, 13-26, doi:10.1007/s00704-015-1685-6.
Journal of the American Water Resources Association. 1999, 35, 311-322, doi:10.1111/j.1752-
1688.1999.tb03592.x.
- Few illustrative examples of the different drought characteristics should be included in the text, particularly for different sub-regions of the world.
- The text includes some typos and syntax problems that should be fixed.
- It is unclear whether the developed dataset includes any further results (e.g. trend assessment) of the different drought characteristics.
Author Response
Please see the attachment
Author Response File: Author Response.docx
Reviewer 3 Report
This is an interesting work done by the authors and a good contribution to the scientific community. I recommend author gave some additional efforts in the introduction section and cite some topic-related studies to enhance the readership.
-The resolution of data is course and it will be very helpful (not mandatory) for readers and researchers if the authors produce data on 0.25°.
-Did authors check data against any in situ SPEI?
-What are the differences between the current (study) and historical SPEI dataset (in situ based)?
Author Response
Please see the attachment
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
The authors have responded satisfactorily to most of the comments raised in the earlier revision. I have only two main comments:
- TIt is recommended that one table be included in which the authors compare the spatial and temporal characteristics of their drought database, including the selected index, timescale, GCMs, etc., with other available global datasets for future drought, in order to justify the main novel aspects of the current study.
-The validation metrics used to assess the WFD's efficiency in comparison to multiple external air temperature and precipitation datasets are displayed in Figures 3 and 4. Different regions of the world have widely varying WFD results, as shown by the interquartile range of the boxes. Therefore, it is necessary to plot the values of some of these metrics on a map so that areas with superior WFD skill can be highlighted. For the same reason, I'm curious as to why the authors didn't assess how well the WFD represented the various drought characteristics (such as frequency, intensity, and duration) instead of focusing on the input variables (e.g. air temperature, precipitation).
Author Response
Dear Rev 2. Thanks for your valuable comments. In details:
- It is recommended that one table be included in which the authors compare the spatial and temporal characteristics of their drought database, including the selected index, timescale, GCMs, etc., with other available global datasets for future drought, in order to justify the main novel aspects of the current study.
We included Table 1 (renumbering the existing tables), and related references, comparing the main characteristics of the datasets providing SPEI. We preferred to focus on SPEI to reduce the scope as drought indicators are numerous. Although many global datasets on SPEI have been derived to support drought studies also for future time-horizons, besides being among those at the highest horizontal resolution globally (0.5°), the presented dataset brings the novelty of being multi-scale (multiple durations and timings) and adds a further ensemble component (beyond GCMs/ESMs and RCPs) due to the use of different potential evapotranspiration formulations.
- The validation metrics used to assess the WFD's efficiency in comparison to multiple external air temperature and precipitation datasets are displayed in Figures 3 and 4. Different regions of the world have widely varying WFD results, as shown by the interquartile range of the boxes. Therefore, it is necessary to plot the values of some of these metrics on a map so that areas with superior WFD skill can be highlighted.
We added, in the Supplementary Material, two maps about the RMSE of Temperature (T) and Precipitation (P) with a comment in the main text. We selected the RMSE as it is one of the widely used performance indices and, in our study, among the ones with the largest interquartile range (for both T and P). In particular, the maps show the RMSE between the WFD reference dataset and the two fully independent datasets for T (GHCN_CAMS) and P (CRU4.05) (see Table 5). As stated in the User Notes, maps of the different performance indices can be made available.
- For the same reason, I'm curious as to why the authors didn't assess how well the WFD represented the various drought characteristics (such as frequency, intensity, and duration) instead of focusing on the input variables (e.g. air temperature, precipitation).
This choice was guided by the fact that comparing directly the drought indicators could have been a couple of drawbacks: i) we should have relied on different (and thus not physically consistent) sources for P and T time series if we wanted to ensure at least a comparison against SPEI based on datasets fully independent from WFD for both variables (GHCN_CAMS for T and CRU4.05 for P); ii) we would have obtained a good performance for the wrong reason, i.e. simultaneous overestimation or underestimation of P and PET giving comparable time series of P-PET but hiding the bias. We explain this choice also in the main text.
