Anthropogenic shift towards higher risk of flash drought over China

Flash droughts refer to a type of droughts that have rapid intensification without sufficient early warning. To date, how will the flash drought risk change in a warming future climate remains unknown due to a diversity of flash drought definition, unclear role of anthropogenic fingerprints, and uncertain socioeconomic development. Here we propose a new method for explicitly characterizing flash drought events, and find that the exposure risk over China will increase by about 23% ± 11% during the middle of this century under a socioeconomic scenario with medium challenge. Optimal fingerprinting shows that anthropogenic climate change induced by the increased greenhouse gas concentrations accounts for 77% ± 26% of the upward trend of flash drought frequency, and population increase is also an important factor for enhancing the exposure risk of flash drought over southernmost humid regions. Our results suggest that the traditional drought-prone regions would expand given the human-induced intensification of flash drought risk.

1. General Comments This manuscript presents a new method to characterise flash drought events (i.e., rapid onset, short duration, co-evolving with a heatwave, severe impacts) and to estimate the exposure risk. This method is applied in China because the propensity of semi-humid and humid regions to endure this kind of hazard. The authors claim that this type of event will increase 30% by the second half of this century under a medium socioeconomic challenge scenario increase. Using the fingerprinting approach, the authors attribute that 80% of the expected changes are induced by the increased greenhouse gas concentrations. The subject covered by this study is a highly relevant research topic having broad scientific and public interest, hence, it is highly suitable for this journal. In the present manuscript, however, there are many key points that have to be clarified before publication.

Specific Comments
• My major concern with this study is the lack of uncertainty estimates for all major results. It is highly recognised in the literature (see the 2018 NCC 1 article authored by the last co-author, ref. nr. 6) that the SMI is highly uncertain due to the large influence of the land surface parameterizations. For this reason is highly recommended to use several LSM to estimate the uncertainty bounds of the SMI estimates (as done in the previous reference). In the present case, the authors only use CLM4.5 (see L.70) to estimate the SMI using CMIP5 forcings. In this way, the authors can estimate the GCM variability but not the LSM variability and hence the results may be overor under-estimated, depending of the bias of the current LSM. Therefore, I strongly suggest the authors to use at least two more LSM for generating a super ensemble. Then, estimate the standard errors of the major results.
• The link with the heatwaves (part of the definition of flash droughts) is not clear based on what us written. Please elaborate (L147, L240ff). How temperature is taken into account for the definition of a flash drought event? Or is not taken? If not, then the introduction is misleading (L35 ff).
• I strongly recommend to use a trend-preserving bias correction (see S. Hempel et al. doi:10.5194/esd-4-219-2013.) instead of the equidistant CDF matching method (L312). The method used my have changed the trends of P and T in the CMIP5 models and hence may alter the results of this study.
• Sampling uncertainty (variability in the model-simulated response) is key for the optimal fingerprinting approach (see Allen and Scott, 2003 10.1007/ s00382-003-0313-9 ). In the present case, I consider that the SMI uncertainty is totally underestimated by the use of a single LSM (CLM4.5). Please investigate, how the inclusion of more LSMs would affect the 80% estimate.
• It would be interesting to know what is the uncertainty contribution of GCMs and LSMs on the flash drought occurrences. This, of course, only can be done by including more LSM in this study.
3. Final Remarks Based on the comments mentioned above and bearing in mind the NCOMMS publishing standards for a research article, I recommend to return this manuscript to the Authors for major revisions. 1

Responses to the comments from Reviewer #1
We thank the reviewer for the critical review. The thoughtful comments have helped improve the manuscript. The reviewer's comments are italicized and our responses immediately follow.
In general the manuscript is interesting and novel as it applied optimal fingerprinting to analyze rapid soil moisture depletion. However, the manuscript is also fundamentally flawed because of its lack of consideration for duration of events it coins as "flash droughts". The flash drought methodology improves considerably on that of Lettenmaier (2015, 2016); however, duration of the overall event is still not considered. Indeed, the authors frequently refer to the "short duration" of flash drought, while citing studies from Ford and Labosier (2017)

and Otkin et al. (2018) that specifically argue flash drought is defined by its intensification and duration.
That is, an event is a flash drought if it exhibits 1) rapid intensification and/or onset (i.e., the flash) and 2) sufficient duration to cause some physical or socioeconomic problem (i.e., the drought). The authors use events such as the 2012 central U.S. drought as evidence for the socioeconomic impact of flash drought; however, they fail to mention that the 2012 event lasted for more than 8 weeks. The reason the 2012 event was so pernicious was because it exhibited both 1) rapid onset/intensification, reducing time for preparation AND 2) sufficient intensity and duration to significantly diminish crop productivity and yield. Therefore, it is paramount in flash drought studies that one use both duration and intensification rate to identify these events. I strongly encourage the authors incorporate these changes to their methodology, which will enhance the impact of their findings.

Response:
We would like to thank the reviewer for the positive and constructive comments. We totally agree with the reviewer that "an event is a flash drought if it exhibits 1) rapid intensification and/or onset (i.e., the flash) and 2) sufficient duration to cause some physical or socioeconomic problem (i.e., the drought)", and it is also the motivation for us to revise the definition of Lettenmaier (2015, 2016). In fact, we started our flash drought study by following Mo and Lettenmaier (2015), and identified an increasing trend for China flash drought during past three decades (Wang and Yuan et al., 2016). Then we investigated the causes of the increasing trends and explored the future changes for the flash droughts in China, but found the previous definition (i.e., concurrent heat and dry anomaly) results in many events with too short 2 durations (e.g., 5 days) to have any impact, and it did not explicitly characterize the rapid evolution of flash droughts. Therefore, in this study, we have proposed a definition that can consider both the duration and intensification rate. 1) To clarify the flash drought definition, we have now moved those statements from the Methods section to the beginning of the Results section, and explained the motivation of proposing a new definition in the revised manuscript as follows: "Definition and characteristics of flash drought events. A popular definition for flash drought events in the hydroclimate community is based on the joint distribution of positive temperature anomaly (e.g., heatwave) and soil moisture deficit 13,20,26 , although the ecohydrological community has different opinions from the perspective of flash drought impact 16 . The former essentially defines an event with concurrent heat extreme and dry conditions, but not necessarily a drought event. For a drought event (no matter conventional drought or flash drought), the system should reach a water deficit for a period of time. If a dry anomaly lasts for a very short period (which is common in the previous definitions 13,20,26 ), it may not have any significant impacts on the ecosystem or the society. Therefore, we consider both the rapid decline rate of soil moisture and the dry persistency in this study: 1) when the percentile for pentad (five-days) mean soil moisture is lower than 40%, a flash drought onset is identified if the soil moisture percentile continuously decreases into 20%, with an average decline rate of no less than 5% for each pentad (e.g., June 30-July 9 in Fig. 1); and 2) if the average decline rate is less than 5% per pentad, the flash drought terminates (e.g., July 10-July 14 in Fig. 1). These two stages are regarded as onset and recovery stages of a flash drought event. For the case illustrated in Fig. 1, the duration for this flash drought event is 15 days (3 pentads). As compared with previous concurrent heat and drought conditions 13,20 (blue asterisks in Fig. 1), the new definition reflects both the "flash" (decline rate of soil moisture percentile larger than 5%) and "drought" (soil moisture percentile less than 20%) conditions, while the previous definition may overestimate flash drought frequency with too short durations (usually only 5 days, e.g., June 20-24 in Fig. 1) that does not lead to any impact. Moreover, the severity of flash drought can also be explicitly estimated with the new definition, using the soil moisture percentile deficit during the flash drought event (red shaded area in Fig. 1)." (L75-L98 in the revised manuscript) heat and drought conditions (old; i.e., pentad mean surface air temperature anomaly larger than its standard deviation, and soil moisture percentile lower than 30%) and by the new definition used in this study (new; see text for details), respectively.
2) All analysis in this study DOES consider the duration of flash drought. We have now clarified them in the revised manuscript as follows: "Frequency, duration and severity of flash drought events are calculated similar to those of conventional drought events 25 , but at a higher temporal resolution. During a study period, the frequency is defined as the average number of flash drought events during the growing seasons per year, the mean duration is the average number of days during which an event lasts, and the mean severity is the mean accumulated soil moisture percentile deficits from the threshold of 40%. There is a higher chance for flash droughts over southern China (humid region) than over northern China (semiarid region), according to both the new and old definitions (Figs. 2a-2b).
There are more flash drought events based on the old definition (Fig. 2b), but their corresponding mean durations are very short, mostly around 7 days (Fig. 2d). In contrast, the mean durations of 3) We would like to thank the reviewer for pointing out the causes of the severe impact of the 2012 central USA drought. We have now incorporated them in the Discussion section as follows: "In addition, the rapid onset of flash droughts poses a great challenge for early warning. The reason that the 2012 central USA drought was so pernicious was because it started with a flash drought with a rapid onset and intensification which reduced time for preparation, and the flash drought was followed by a subseasonal drought that lasted for 8 weeks. In this regard, the flash drought occurred at the onset stage of a subseasonal drought 35 . Improving understanding of sub-seasonal predictability with consideration of vegetation-drought interactions and human interventions (e.g., agricultural practices), and developing climate-hydrology-human coupled 6 prediction systems that utilize multiscale memory from the earth system (e.g., oceanic and land processes, or even human activities) would fundamentally increase our capability in managing risk of flash droughts as well as those connected with subseasonal or seasonal droughts for both the society and environment." (L220-L230) 7

Responses to the comments from Reviewer #2
We thank the reviewer for the critical review. The reviewer's comments are italicized and our responses immediately follow.   Figure S2. The same as Figure S1, but for surface air temperature (K). is not the motivation of proposing a new flash drought definition. In this study, we would like to investigate the causes of the increasing trends and explore the future changes for the flash droughts in China. We found the previous definition (i.e., concurrent heat and dry anomaly) results in many events with too short durations (e.g., 5 days) to have any impact, and it did not explicitly characterize the rapid evolution of a flash drought. Therefore, in this study, we have proposed a definition that can consider both the duration and intensification rate. For details, please see our response to the Reviewer #1 above for the motivation of developing a new definition, which has also been included in the revised version of the manuscript.
2) Although the question raised by the reviewer regarding the reason for the different changes in flash droughts over USA and China is not relevant to the main scope of this study, it is an interesting question. So, we have now included a brief discussion in the revised manuscript as follows: "The increasing trend in flash droughts over China is different from that over USA 13 , but this is not due to the definition of flash droughts because our previous studies based on the definition of concurrent heat and dry anomaly also suggests upward trends in China 20 . While rigorous analysis based on comprehensive detection and attribution simulations should be carried out over USA to understand the difference, our preliminary speculation is that the difference may come from two sources: 1) the focus periods and the internal climate variability are different between the studies over USA and China, which would influence the trend analysis; and 2) the hotspots of flash droughts over USA (central Great Plains) experienced an increased soil moisture 13 while those over China (southern China) experienced significantly decreased soil moisture 20 , again during 11 different study periods. The different long-term trends in soil moisture may play an important role in altering flash drought trends, but the variability of hydroclimate variables (e.g., precipitation, temperature and ET) at short time scales should also be investigated in details." (L189-L201) c) Apparently they haven't investigated trends in the historic data? That seems to me to be the place one would start, rather than investigating climate model output. With climate model output, you know things are going to change -that's pretty much a given -but whether the results make any sense is indeterminate if you don't at least do an analysis of the historic record.
Response: No, actually we DID investigate the trends in the historical data. Fig. 4a (Fig. 2a in the previous version) shows the changes of flash droughts during 1961-2005, and we have also compared the CMIP5-driven results (colored lines and shadings) with those driven by observed meteorological forcings (black line). We have also used the historical simulations to select CMIP5 models that are suitable for future projections: "In fact, 13 CMIP5/ALL/LSM simulations were carried out, and 11 CMIP5 models were selected due to their capability in reproducing the upward trend for flash drought frequency over China for CLM4.5 and VIC, while 7 CMIP5 models were selected for NoahMP (see Supplementary Table 1 for details)." (L354-L357)  Response: For the contrast between USA and China, please see our response to your comment #b) above. For the reference, we agree with the reviewer that it is not relevant for future projections. It is related to the uncertainty of drought index where ET parameterization plays an important role. Therefore, we have used the reference to argue the "drought index variety" and revised the manuscript as follows: "The Fifth Assessment Report of the Intergovernmental Panel on Climate Change concluded that there was a low confidence in detecting and attributing human impact on drought changes since the middle of the 20 th century over global land areas due to internal climate variability 1 , data scarcity and drought index variety 2-3 , resulting in large uncertainty in future drought projection 1,4-6 ." (L33-L37) Taken together, my recommendation is that this paper should not be seriously considered by Nature Communications. Frankly, I don't think it should have been sent out for review in the first place. The authors would be far better advised to submit their work to a quality disciplinary journal that has a history of publishing in the area, e.g., JHM or J Clim.

Response:
We regret that the reviewer did not realize the value of our work, which might be partly due to unclear presentation in our previous version of the manuscript. We hope we have now improved the clarification and addressed the reviewer's comments above.
In fact, we believe we are the first to propose the new definition of flash drought for detection and attribution of flash drought change, and for future projection of flash drought risk, not just in China, but also in the worldwide literatures. We believe our study has both new insights and broad interests that make it qualified for external review by Nature Communications.

Responses to the comments from Reviewer #3
We thank the reviewer for the critical review. The thoughtful comments are valuable and the suggestions make our work more confident. The reviewer's comments are italicized and our responses immediately follow.

Response:
We would like to thank the reviewer for the positive and constructive comments.
Please see our detailed responses below.

Specific Comments
My major concern with this study is the lack of uncertainty estimates for all major results.  Therefore, we would like to keep using ECDF matching method in this study. Response: We agree with reviewer, the new estimation result for the GHG contribution based on multiple CMIP5 models and three LSMs is below: "The best estimates of scaling factors show that only the GHG signal is detectable, with a contribution of 77%±14% to the increase (Fig. 4b). These results are also consistent with the simulations of flash drought frequency, where GHGs play a dominant role in all anthropogenic forcings ( Supplementary Fig. 7)." (L155-L158)  Fig. 4). As compared with the LSM simulations driven by observed meteorological forcings, the CMIP5/ALL/LSM simulations underestimate the national-averaged flash drought frequency by one event per decade, but it is within the uncertainty of LSMs (Supplementary Fig. 5). Without using LSM ensemble simulations, the uncertainty range for the drought frequency might be underestimated by 24%-47% ( Supplementary Fig. 5)." (L114-L124) Figure S4. Uncertainty ranges of the frequency of flash drought events from CMIP5/ALL/LSM simulations. Here, the uncertainty range was quantified by the 1.645 standard deviation of the frequency (events/decade). There are 29, 11, 11 and 7 realizations for ALL/LSMs (a), ALL/CLM4.5 (b), ALL/VIC (c) and ALL/NoahMP (d) simulations (see Table S1). Response: We would like to thank the reviewer for the positive comments. We hope we have addressed them above.
I presented one major concern in my initial review of this manuscript, that the flash drought definition introduced here did not explicitly consider the duration of the event. To their credit, the authors have clarified their definition, but have not addressed the primary issue. I will try to make my concern more clear because I believe there was some miscommunication.
The current definition requires 1) the pentad-average soil moisture remain below the 40th percentile and decreases to at or below the 20th percentile, and 2) the average decline rate is at or more than 5% per pentad. This captures the "flash", but does not explicitly account for duration. Unless I am missing something, an event in which soil moisture declines from the 35th percentile to the 12th percentile in 2 pentads, and then quickly recovers back to the 40th percentile, would be considered a flash drought, no? In that case, the actual duration of the event would be 10 days, which is too few for drought impacts. This is my issue: the definition should also include a minimal time that the soil moisture remain below the (e.g.,) 30th or 40th percentile before the event is termed a flash drought.
Along this line of argument, the authors continue to refer to flash droughts as "short duration" (line 42) and lasting for "less than one month" (line 55), yet these statements are not supported by previous research. The authors state that the average duration of events identified as flash droughts using their methodology is 15 days, which seems like it should be closer to the minimum duration to be considered a drought. Likely, without an explicit duration requirement included in their definition, many very short (i.e., 5-10 day) events are identified as flash droughts, which will skew the statistics and add considerable uncertainty to the results. I implore the authors to include an explicit duration requirement in their definition and revise accordingly. Without this, I cannot recommend this paper for publication. 2) I disagree with ref. 2's argument that developing a new flash drought definition is unnecessary. However, I do think the authors should present a more substantial argument for why a new flash drought definition -beyond the flawed Mo and Lettenmaier methodology -is necessary, and how the new definition accounts for these issues. The authors provide some of this in the introduction, but it needs to be thoroughly expanded.
Reviewer #3 (Remarks to the Author): Comments on the Revised Manuscript Anthropogenic shift towards higher risk of flash drought over China By Yuan et al.
In general I am very satisfied with the rebuttal letter and modifications carried out by the authors. It appears that including many LSMs was crucial to refine the conclusions of the manuscript.  With this minor additions, I consider that this manuscript can be published.

Luis Samaniego
We thank the reviewer for the critical review. The thoughtful comments have helped improve the manuscript. The reviewer's comments are italicized and our responses immediately follow.
I presented one major concern in my initial review of this manuscript, that the flash drought definition introduced here did not explicitly consider the duration of the event. To their credit, the authors have clarified their definition, but have not addressed the primary issue. I will try to make my concern more clear because I believe there was some miscommunication.
The current definition requires 1) the pentad-average soil moisture remain below the 40th percentile and decreases to at or below the 20th percentile, and 2) the average decline rate is at or more than 5% per pentad. This captures the "flash", but does not explicitly account for duration. Unless I am missing something, an event in which soil moisture declines from the 35th percentile to the 12th percentile in 2 pentads, and then quickly recovers back to the 40th percentile, would be considered a flash drought, no? In that case, the actual duration of the event would be 10 days, which is too few for drought impacts. This is my issue: the definition should also include a minimal time that the soil moisture remain below the (e.g.,) 30th or 40th percentile before the event is termed a flash drought.
Along this line of argument, the authors continue to refer to flash droughts as "short duration" (line 42) and lasting for "less than one month" (line 55), yet these statements are not supported by previous research. The authors state that the average duration of events identified as flash droughts using their methodology is 15 days, which seems like it should be closer to the minimum duration to be considered a drought. Likely, without an explicit duration requirement included in their definition, many very short (i.e., 5-10 day) events are identified as flash droughts, which will skew the statistics and add considerable uncertainty to the results. I implore the authors to include an explicit duration requirement in their definition and revise accordingly. Without this, I cannot recommend this paper for publication.

Response:
We would like to thank the reviewer for the positive and constructive comments. 2 1) We agree with the reviewer without an explicit duration requirement, many 10-day events are included. We have now revised the definition by specifying the minimum duration of 3-pentad (15 days), and by including those pentads with soil moisture lower than 20 th percentile after the onset of flash drought. The revised definition excludes the 10-day events and reduces the flash drought frequency (Fig. 2a), but the mean duration is now between 20-40 days for most areas in China (Fig. 2c). The revised text and figures are as follows: "Definition and characteristics of flash drought events. To identify a flash drought event, we consider both the rapid decline rate of soil moisture and the dry persistency in this study: 1) the pentad (five-days) mean root-zone (top 1m) soil moisture decreases from above 40 th percentile to 20 th percentile, with an average decline rate of no less than 5% for each pentad (e.g., June 30-July 14 in Fig. 1); 2) if the declined soil moisture rises up to 20 th percentile again, the drought terminates (e.g., July 15-July 19 in Fig. 1); and 3) the drought should last for at least 3 pentads (15 days). The first two criterions describe the onset and recovery stages of a flash drought event.
Although the recovery threshold could increase from 20 th to 40 th percentile, here the 20 th percentile is chosen to exclude many events that can last for more than 3-6 months if the 30 th or 40 th percentile threshold is used, where those events should be regarded as seasonal droughts instead of flash droughts. The third criterion is the minimal time that the soil moisture remains below 40 th percentile, which excludes those events that decrease from above 40 th percentile rapidly down to 20 th percentile within 10 days, but then recover up to 40 th percentile suddenly." (L84-97 in the revised manuscript)  2) The reason for choosing 15 days as the duration threshold is to avoid counting short (10-day) dry spells and to keep those 15-day events that DO have impacts. We have applied the revised definition to the FLUXNET stations to investigate the flash drought impact on vegetation. Figure R1 shows that a 3-pentad flash drought event can have significant adverse effect on gross primary productivity (GPP). In fact, even for a 2-pentad flash drought (which 4 was identified in our previous definition), we can also find cases with adverse vegetation response (not shown). Therefore, the minimum drought duration may vary for different climate and vegetation regimes, which needs further investigation. Here, we use 15 days as the minimum duration that is considered as a drought. "Flash droughts refer to a type of droughts that have rapid intensification and severe impacts without sufficient early warning." (L17-18) 5 "…and find that the exposure risk over China will increase by about 23%±11% (more than 40% and soil moisture deficit 13,20,26 , although the eco-hydrological community has different opinions from the perspective of flash drought impact 16 . The former essentially defines an event with concurrent heat extreme and dry conditions, but not necessarily a drought event. For a drought event (no matter conventional drought or flash drought), the system should reach a water deficit for a period of time. If a dry anomaly lasts for a very short period (which is common in the previous definitions 13,20,26 ), it may not have any significant impacts on the ecosystem or the society. Moreover, the rapid intensification is a also key feature to distinguish flash droughts from conventional droughts in terms of their physical characteristics and impacts, and the identification of the severity and different drought stages facilitates early warning and risk assessment during the evolvement of flash droughts. Here, we propose a new flash drought 8 definition based on soil moisture that can capture both "flash" (rapid intensification of a drought condition, e.g., rapid decline in soil moisture) and "drought" (under a certain soil moisture threshold for a period of time) conditions." (L57-78) 9

Responses to the comments from Reviewer #3
We thank the reviewer for the critical review. The thoughtful comments are valuable and the suggestions make our work more confident. The reviewer's comments are italicized and our responses immediately follow.
In general I am very satisfied with the rebuttal letter and modifications carried out by the authors. It appears that including many LSMs was crucial to refine the conclusions of the manuscript.  With this minor additions, I consider that this manuscript can be published.

Response:
We would like to thank the reviewer for the positive comments. According to the suggested uncertainty estimation method 30 , we have now included a figure to show the contribution of uncertainty for GCM/LSM for the projection of flash drought frequency and duration, and revised the manuscript as follows: "In the flash drought projection, the uncertainty 30 from CMIP5 models is larger than that from LSMs, especially over northern China (Supplementary Fig. 7)." 10 Figure S7. The ratio between the Global Climate Model (GCM; 11 CMIP5 models used in this study) contribution and LSM contribution to the uncertainty ranges for the changes in flash drought frequency and duration shown in Figure S6. The values larger than 1 suggest that the uncertainty from GCMs is larger than that from LSMs.