Machine learning-constrained projection of bivariate hydrological drought magnitudes and socioeconomic risks

Liu, Rutong; Yin, Jiabo; Slater, Louise; Kang, Shengyu; Yang, Yuanhang; Liu, Pan; Guo, Jiali; Gu, Xihui; Volchak, Aliaksandr

doi:https://doi.org/10.5194/hess-2023-181

Preprints

https://doi.org/10.5194/hess-2023-181

Preprints

06 Sep 2023

| 06 Sep 2023

Status: a revised version of this preprint is currently under review for the journal HESS.

Machine learning-constrained projection of bivariate hydrological drought magnitudes and socioeconomic risks

Rutong Liu, Jiabo Yin, Louise Slater, Shengyu Kang, Yuanhang Yang, Pan Liu, Jiali Guo, Xihui Gu, and Aliaksandr Volchak

Abstract. Climate change accelerates the water cycle and alters the spatiotemporal distribution of hydrological variables, thus complicating the projection of future streamflow and hydrological droughts. Although machine learning is increasingly employed for hydrological simulations, few studies have used it to project hydrological droughts, not to mention the bivariate risks of drought duration and severity as well as their socioeconomic effects under climate change. We develop a cascade modeling chain to project future bivariate hydrological drought characteristics in 179 catchments over China, using 5 bias-corrected GCM outputs under three shared socioeconomic pathways, five hydrological models and a deep learning model. We quantify the contribution of various meteorological variables to daily streamflow by using a random forest model, then employ terrestrial water storage anomalies and a standardized runoff index to evaluate recent changes in hydrologic drought. Subsequently, we construct a bivariate framework to jointly model drought duration and severity by using Copula functions and the most likely realization method. Finally, we use this framework to project future risks of hydrological droughts as well as associated exposure of gross domestic product and population. Results show that our hybrid hydrological-deep learning model achieves >0.8 Kling-Gupta efficiency in 161 out of 179 catchments. By the late 21^st century, bivariate drought risk is projected to double over 60 % catchments, mainly located in Southwest China. Our hybrid model also projects substantial GDP and population exposures by increasing bivariate drought risks, suggesting an urgent need to design climate mitigation strategies towards a sustainable development pathway.

Received: 22 Jul 2023 – Discussion started: 06 Sep 2023

Download & links

Preprint (PDF, 9551 KB)

Supplement (199 KB)

Download & links

Rutong Liu, Jiabo Yin, Louise Slater, Shengyu Kang, Yuanhang Yang, Pan Liu, Jiali Guo, Xihui Gu, and Aliaksandr Volchak

Status: final response (author comments only)

RC1:
'Comment on hess-2023-181', Anonymous Referee #1, 04 Oct 2023
In Line 59, I'm not sure about the relationship between these two sentences. Can the author's example support the previous sentence? In my opinion, it seems like it cannot.

In the last paragraph of the Introduction section, it is recommended that the author use the past tense when describing the work done in this paper.

The title of section 5.3 of the article is "Suggestions for drought mitigation in China", but I didn't seem to find any relevant suggestions based on the research results in this section (maybe I misunderstood?). In my opinion, the author presented the research background and significance of the paper to readers in section 5.2, but did not provide reasonable suggestions for reducing drought events in China. I think this is also of interest to readers, and I suggest that the author engage in a detailed discussion of this part of the content.

The author is suggested to abbreviate the content in Section 5.2 and include it in the summary section after the conclusion.

The last sentence of the paper's conclusion, in my opinion, suggests that drought events are just one of the possible disasters that may be exacerbated by climate change, and the author did not discuss the connection between achieving carbon neutrality goals and drought events. Perhaps I did not express myself clearly, but what I meant was that there was no corresponding data in the paper to support the latter half of the sentence.

I noticed that the author used five hydrological models to conduct experiments, but these models all have some limitations to varying degrees. For example, how did the author consider the applicability of the XAJ hydrological model in different climatic regions, and whether the model parameters obtained using the SCE-UA algorithm are universal? Are these parameters effective if spatial heterogeneity is strictly considered? In addition, the author's introduction of these five hydrological models is too brief, and it is suggested that the author provide slightly more detail in this section. Furthermore, can the author try to use hydrological models such as WRF-Hydro, SWAT, or the hydrological modules of land surface process models to conduct experiments in this paper? In my opinion, these general hydrological models would be more convincing.
Citation: https://doi.org/10.5194/hess-2023-181-RC1
- AC1:
  'Reply on RC1', Jiabo Yin, 12 Oct 2023
  We appreciate the Reviewer for your helpful suggestions, which will strengthen the quality of the work. Here are our replies.
  In Line 59, I'm not sure about the relationship between these two sentences. Can the author's example support the previous sentence? In my opinion, it seems like it cannot.
  
  Response: Thank you; we have rephrased these sentences as follows:
  China's socioeconomic development, and particularly its agricultural sector, is threatened by the rapid intensification of extreme hazards under climate change (Piao et al., 2010). In recent years, China has been hit by severe drought events which have caused considerable damage to ecosystem productivity and socio-economic growth (Zhai and Zou, 2005; Yin et al., 2023). For instance, one extreme drought in Sichuan Province in 2022 resulted in power shortages and led to economic losses of 669 million dollars. Water shortage is also a key challenge hindering the sustainable development of the North China Plain (Chen and Yang, 2013). Over the period of 1985-2014, drought accounted for about 19% of economic losses among all meteorological hazards (Chen and Sun, 2019). With continued global warming, the economic losses from severe drought events might increase by over ten billion US dollars per year by the late 21st century, underscoring the importance of projecting future droughts over China (Lu et al., 2023).
  In the last paragraph of the Introduction section, it is recommended that the author use the past tense when describing the work done in this paper.
  
  Response: Thank you for your good suggestion. These sentences will be corrected to past tense.
  The title of section 5.3 of the article is "Suggestions for drought mitigation in China", but I didn't seem to find any relevant suggestions based on the research results in this section (maybe I misunderstood?). In my opinion, the author presented the research background and significance of the paper to readers in section 5.2, but did not provide reasonable suggestions for reducing drought events in China. I think this is also of interest to readers, and I suggest that the author engage in a detailed discussion of this part of the content.
  
  Response: Thank you; we will provide more details about the relevant suggestions as follows:
  The Intergovernmental Panel on Climate Change (IPCC) has emphasized that projections of future climate trends can equip policymakers with the scientific insight needed to navigate the challenges of climate change (Pörtner et al., 2022). The results of this study aim to alert policymakers to drought risk in Southwestern China which was recently hit by severe drought events, and the intensity of which is expected to continue growing with climate change. Our findings emphasize the importance of strictly implementing carbon emission reduction initiatives and developing prevention programs to limit future drought losses. Preserving local ecological balance and employing rational use of water resources could be the key to mitigating potential future losses from extreme droughts (Sohn et al., 2016; Chang et al., 2019). Although China already has hydraulic structures with a total water storage capacity of over 7,064 billion m3, existing irrigation facilities need to be expanded to mitigate the challenge of drought under climate change (Xiao-jun et al., 2012; Cai et al., 2015). In addition, it may also be beneficial for policymakers to establish a drought monitoring system to comprehensively assess drought impacts in all potential sectors, which can link the government with research organizations (Wilhite et al., 2007).
  The author is suggested to abbreviate the content in Section 5.2 and include it in the summary section after the conclusion.
  
  Response：Thank you for your suggestion; we will revise accordingly as follows:
  The uncertainty caused by underlying surface conditions and the coupling relationships behind interrelated variables remain unexplained in this study. Revealing interactions among multisource data is important to understand the drivers affecting the water cycle under climate change. Here, only five GCM outputs and one in situ observation dataset were used to drive our HTM models. Including a larger number of GCMs and observational data to assemble a more sophisticated model might be an effective approach to improve the accuracy and reliability of the models. Additionally, due to the heterogeneity of different climatic regions in China, it would be worth expanding the choice of hydrological models (e.g. including the weather research and forecasting model hydrological modeling system (WRF-Hydro), soil and water assessment tool (SWAT) or the hydrological modules of land surface process models) to reduce uncertainties in the projections. Finally, the GDP and population projections are unlikely to reflect future economic development and population migration well, especially governmental choices in terms of immigration and economic policies. It may be better to consider the dynamic impacts of human management on socioeconomic development, to construct a more reliable projection framework.
  
  These contents will also be summarized as follows:
  This study does not sufficiently investigate the revelation of drought hazard drivers and could further investigate the choice of datasets and hydrological models to strengthen the reliability of the simulations in future work.
  The last sentence of the paper's conclusion, in my opinion, suggests that drought events are just one of the possible disasters that may be exacerbated by climate change, and the author did not discuss the connection between achieving carbon neutrality goals and drought events. Perhaps I did not express myself clearly, but what I meant was that there was no corresponding data in the paper to support the latter half of the sentence.
  
  Response: We will rephrase this sentence as follows:
  Our findings demonstrate that China is likely to face higher drought risks in a warmer future, emphasizing the urgency of implementing strategies to reduce carbon emissions.
  I noticed that the author used five hydrological models to conduct experiments, but these models all have some limitations to varying degrees. For example, how did the author consider the applicability of the XAJ hydrological model in different climatic regions, and whether the model parameters obtained using the SCE-UA algorithm are universal? Are these parameters effective if spatial heterogeneity is strictly considered? In addition, the author's introduction of these five hydrological models is too brief, and it is suggested that the author provide slightly more detail in this section. Furthermore, can the author try to use hydrological models such as WRF-Hydro, SWAT, or the hydrological modules of land surface process models to conduct experiments in this paper? In my opinion, these general hydrological models would be more convincing.
  
  Response: As the hydrological models might show uncertainties, we employed five different models as candidates to simulate streamflow. In implementing this process, we calibrate our models in each catchment. In other words, the parameters of hydrological models in different catchments are not universal. We will provide more details about the hydrological models in Section 2.3.1 as follows:
  The XAJ (Xinanjiang) model is a hydrological model, which often achieves better performance in humid and semi-humid areas than in arid areas (Zhao, 1992). As the model was developed based on the underlying surface of the Yangtze River Basin in China, it is composed of a three-layer evapotranspiration module with four parameters and separates the runoff into four components (i.e., surface water, groundwater, interflow water and flow routing) (Tian et al., 2013). To date, it is widely reported that the XAJ model usually shows the best accuracy relative to other models in simulating hydrological conditions in China (Hu et al., 2005). However, due to inadequacies in the simulation of arid regions, the results of the XAJ model were not considered as the best option in northern China.
  We use the SCE-UA (Shuffled Complex Evolution) approach to maximize the objective function (i.e., Kling-Gupta efficiency) to optimize these models (Duan et al., 1992). The most complete 20-year observation period is selected to calibrate five models in each watershed. To calibrate the hydrological models, a cross-validation method developed by Arsenault et al. (2017) is used, which employs the odd years of the data to calibrate the models, and the even years to validate them. As catchments are located in different climatic regions, the parameters of the models are calibrated for each catchment, which means that the parameters are not universal. Although the uncertainties shown by the hydrological models are ineradicable, the overall uncertainty is acceptable after optimizing the five hydrological models for each catchment.
  We believe your suggestion of using WRF-Hydro and SWAT models is great, but is beyond the scope of this manuscript. We will discuss the future application of these models in the Discussion section 5.2 as follows.
  Additionally, due to the heterogeneity of different climatic regions in China, it would be worth expanding the choice of hydrological models (e.g. including the weather research and forecasting model hydrological modeling system, soil and water assessment tool or the hydrological modules of land surface process models) to reduce uncertainties in the projections.
  
  Citation: https://doi.org/10.5194/hess-2023-181-AC1
RC2:
'Comment on hess-2023-181', Anonymous Referee #2, 03 Jan 2024
This manuscript assesses the future evolution of hydrological droughts and their socioeconomic implications under a warming climate, proposing urgent strategies to address the climate crisis in China. The main conclusions are well-supported by the results, and the storyline is clear. The topic of bivariate drought aligns well with the scope of HESS. Therefore, I recommend a Minor Revision before acceptance.
Text errors should be addressed:

In line 384, please remove the redundant period.

In line 198, is it supposed to be 'retained' instead of 'retailed'?

In line 278, 'demote' should be revised to 'denote'.

The titles in Figure 10 are suggested to be revised, as they only include severity and duration under SSP1-26 and SSP3-70, which isn't consistent with the elaboration in section 4.3.

In lines 222 and 223, the number of HTMs differs from subsequent elaboration in line 284. The number of HTMs should be corrected.

In lines 235 and 401, tables of drought classification and candidate distributions are suggested to be referenced from the supplement file.

In Figure 6, considering the limited number of stations located in Northwestern China and the use of interpolated methods for calculating sensitivities, does it potentially affect the accuracy of the analysis? The author is suggested to elaborate on the reliability of the results in section 4.2.

In line 375, the author is suggested to revise this conclusion. First, it should be mentioned which type of drought is sensitive to temperature. Second, whether the feedback of drought to temperature is reliable should be discussed, as drought is affected by both hydrological and thermal factors. Univariate sensitivity isn’t a powerful support under global warming.

The author is suggested to add an explanation of which approach was used for the analysis in section 4.2. Is this analysis conducted at spatial, temporal, or spatial-temporal dimensions? More specifically, is the input data for the RF model the multi-year average of each variable from each grid (spatial), or spatial average at each timestep (temporal), or variable for each timestep for each grid (spatial-temporal)?
Citation: https://doi.org/10.5194/hess-2023-181-RC2
- AC2:
  'Reply on RC2', Jiabo Yin, 06 Jan 2024
  Text errors should be addressed:
  
  In line 384, please remove the redundant period.
  
  In line 198, is it supposed to be 'retained' instead of 'retailed'?
  
  In line 278, 'demote' should be revised to 'denote'.
  
  Response: Thank you for your careful examination. These errors will be corrected.
  
  The titles in Figure 10 are suggested to be revised, as they only include severity and duration under SSP1-26 and SSP3-70, which isn't consistent with the elaboration in section 4.3.
  
  Response: Thank you for your suggestion. The elaboration in the manuscript is correct, and the title error of this figure will be corrected.
  
  In lines 222 and 223, the number of HTMs differs from subsequent elaboration in line 284. The number of HTMs should be corrected.
  
  Response: Thank you for your suggestion. We will correct the number of HTMs in lines 284 to ten.
  
  In lines 235 and 401, tables of drought classification and candidate distributions are suggested to be referenced from the supplement file.
  
  Response: Thank you for your suggestion. We will add citations as follows:
  
  The hydrologic drought classification and ranges indicated by SRI are shown in Table S1.
  
  Based on the maximum Bayesian Information Criterion (BIC), we select the best-performing marginal distributions for duration and severity from seven candidate distributions shown in Table S2, based on historical data for each catchment.
  
  In Figure 6, considering the limited number of stations located in Northwestern China and the use of interpolated methods for calculating sensitivities, does it potentially affect the accuracy of the analysis? The author is suggested to elaborate on the reliability of the results in section 4.2.
  
  Response: Thank you for your suggestion. We will add an explanation as follows:
  
  Due to the sparse number of observation stations in Northwestern China, the reliability of the sensitivity analysis for these regions is lower than that of the dense observed areas.
  
  In line 375, the author is suggested to revise this conclusion. First, it should be mentioned which type of drought is sensitive to temperature. Second, whether the feedback of drought to temperature is reliable should be discussed, as drought is affected by both hydrological and thermal factors. Univariate sensitivity isn’t a powerful support under global warming.
  
  Response: Thank you for your suggestion. We will reshape this conclusion as follows:
  
  The temperature has a positive contribution to streamflow generation in Northeast China, suggesting a potential mitigation for the deficiency of surface flow. However, there is interactive feedback between hydrological and thermal factors that result in an inability to directly assess the impact of temperature on hydrologic droughts.
  
  The author is suggested to add an explanation of which approach was used for the analysis in section 4.2. Is this analysis conducted at spatial, temporal, or spatial-temporal dimensions? More specifically, is the input data for the RF model the multi-year average of each variable from each grid (spatial), or spatial average at each timestep (temporal), or variable for each timestep for each grid (spatial-temporal)?
  
  Response: Thank you for your suggestion. We will add an explanation as follows:
  
  We quantified the sensitivity of seven historical mean meteorological variables (i.e., pr, ps, SH, RH, srlds, srsds, temperature) to monthly streamflow in each grid.
  
  Citation: https://doi.org/10.5194/hess-2023-181-AC2

Rutong Liu, Jiabo Yin, Louise Slater, Shengyu Kang, Yuanhang Yang, Pan Liu, Jiali Guo, Xihui Gu, and Aliaksandr Volchak

Supplement

https://doi.org/10.5194/hess-2023-181-supplement

Rutong Liu, Jiabo Yin, Louise Slater, Shengyu Kang, Yuanhang Yang, Pan Liu, Jiali Guo, Xihui Gu, and Aliaksandr Volchak

Viewed

Total article views: 764 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	Supplement	BibTeX	EndNote
566	168	30	764	111	18	21

HTML: 566
PDF: 168
XML: 30
Total: 764
Supplement: 111
BibTeX: 18
EndNote: 21

Views and downloads (calculated since 06 Sep 2023)

Month	HTML	PDF	XML	Total
Sep 2023	242	57	5	304
Oct 2023	90	35	5	130
Nov 2023	31	5	1	37
Dec 2023	64	16	4	84
Jan 2024	54	13	5	72
Feb 2024	30	15	3	48
Mar 2024	34	20	3	57
Apr 2024	21	7	4	32

Cumulative views and downloads (calculated since 06 Sep 2023)

Month	HTML	PDF	XML	Total
Sep 2023	242	57	5	304
Oct 2023	90	35	5	130
Nov 2023	31	5	1	37
Dec 2023	64	16	4	84
Jan 2024	54	13	5	72
Feb 2024	30	15	3	48
Mar 2024	34	20	3	57
Apr 2024	21	7	4	32

Viewed (geographical distribution)

Total article views: 704 (including HTML, PDF, and XML) Thereof 704 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 25 Apr 2024

Short summary

Climate change accelerates the water cycle and alters the spatiotemporal distribution of hydrological variables, thus complicating the projection of future streamflow and hydrological droughts. We develop a cascade modeling chain to project future bivariate hydrological drought characteristics over China, using 5 bias-corrected GCM outputs under three shared socioeconomic pathways, five hydrological models and a deep learning model.


Total:	0
HTML:	0
PDF:	0
XML:	0