Spatio-temporal characteristics and driving factors of the meteorological drought across China based on CMIP6

Drought is a kind of natural disaster, usually having a substantial impact on the national economy, society and ecological environment. Therefore, people attach great importance to drought research. In the context of global warming, droughts have been exacerbated by climate change so far and have led to significant changes in the areas impacted by droughts worldwide (Samaniego et al. 2018). Usually, drought arises from a severe precipitation deficiency and is described in terms of changes in major meteorological factors (such as precipitation and temperature) and anomalies compared to long-term averages (Yang et al. 2020). The Intergovernmental Panel on Climate Change (IPCC) Fifth and Sixth Assessment Reports concluded that the climate system is getting warmer on the global scale (IPCC 2021). Under the effects of climate warming, the evolution and mechanism of drought have become an increasingly important research topic. Therefore, exploring predicted changes in drought characteristics under different emission scenarios will help formulate disaster prevention and mitigation policies.

Usually, there are four major types of droughts: agricultural drought (Heim 2002), socioeconomic drought (Shukla & Wood 2008), hydrological drought (Mishra & Singh 2011) and meteorological droughts (Hayes et al. 2011). Due to the availability of relatively long historical data for observations of meteorological variables, meteorological drought is one of the most extensively studied droughts (Harris et al. 2014). A lot of studies on meteorological drought indicators have been carried out. The most commonly used methods are the Standardized Precipitation Index (SPI) (Mckee et al. 1993), the Standardized Precipitation Evapotranspiration Index (SPEI) (Vicente-Serrano et al. 2010) and the Palmer Drought Severity Index (PDSI) (Palmer 1965). The SPI only considers precipitation and ignores the influence of temperature on meteorological drought. The SPEI responds very differently to precipitation and temperature of different climates and locations and it relies highly on temperature. The PDSI is the index that is most widely applied to meteorological drought and is most suitable to assess the impact of global warming on drought, which incorporates precipitation and temperature. The PDSI can adjust and balance the contributions of precipitation and temperature to local water anomalies for different climate conditions, implying better spatio-temporal consistency and comparability. Meanwhile, Wells et al. (2004) found a new algorithm for calculating PDSI, named the Self-Calibrating PDSI (SC-PDSI). Based on the historical meteorological observation data of a single station, the SC-PDSI can determine the climate parameters applicable to the local area. The SC-PDSI has better spatial and temporal stability and comparability than the earlier PDSI. Therefore, we chose the SC-PDSI to investigate the spatio-temporal characteristics of meteorological drought.

The General Circulation Models (GCMs) are generally used for meteorological drought projection, which is published by the Coupled Model Intercomparison Project (CMIP). CMIP6 is considered advantageous over CMIP3 and CMIP5 in reproducing temperature and precipitation patterns (Bock et al. 2020). In particular, CMIP6 is advantageous in capturing precipitation features (Na et al. 2020). Therefore, we used the precipitation and temperature data in the latest CMIP6 for predicting meteorological drought. Precipitation and temperature, as the data of the SC-PDSI calculation are the main influencing factors of meteorological drought. There have been many studies on the influencing factors of meteorological drought. In the midlands of the UK, researchers found that precipitation has a greater impact on meteorological drought compared to temperature (Rahmani & Fattahi 2021). However, the observed increase in precipitation in Central Europe and the Balkans is not significant; in these regions, temperature drives the increase in meteorological drought severity (Spinoni et al. 2015).

In China, studies on the driving factors of meteorological drought mainly focus on the El Niño Southern Oscillation (ENSO), the Pacific Decadal Oscillation (PDO), the Atlantic Oscillation (AO) and solar activity (Huang et al. 2016; Yue et al. 2022). However, there are few relevant studies on the effects of precipitation and temperature elements on the meteorological drought in China. Precipitation and temperature, as two meteorological factors, also play an important role in meteorological drought. In northeast China, Li et al. (2020) found that the direct effect of precipitation on meteorological drought was greater than that of other meteorological factors and the dominant driving factors were also influenced by seasons. The influence of meteorological factors on meteorological drought is very complex. Hence, considering the complex geographical environment in China, we estimate the contribution of precipitation and temperature to meteorological drought at the watershed scale and analyse the influence of the above two meteorological factors on meteorological drought in China.

In this study, we focus on the evolution and influencing mechanisms of meteorological drought at basin scales over China. The objective was to determine the trend of drought under climate change conditions using the SC-PDSI under four future scenarios (SSP126, SSP245, SSP370 and SSP585) and historical periods. We explored the leading meteorological factors of meteorological drought and the meteorological drought risk of different river basins in China. This study is helpful for understanding and predicting drought trends in different basins around China.

The daily precipitation and temperature observational datasets from 1961 to 2014 over China were obtained from 823 meteorological stations in the National Meteorological Information Center, the China Meteorological Administration (CMA) (http://data.cma.cn/, accessed on 20 May 2021). The meteorological station datasets cover the area ranging from 16.53°N to 53.48°N and 75.18°E to 132.96°E. The CMA has conducted quality control on the temperature data collected at all meteorological stations; the meteorological data of each meteorological station have no obvious mutations and random changes and, therefore, can represent the climate status of the research area. This study established a daily precipitation and temperature observational dataset and the spatial resolution was processed at 10 km ×10 km over China by the bilinear interpolation method (Gustavsson & Peetre 1977).

Based on the observed data, the performances of bias-corrected CMIP6 General Circulation Models (GCMs) for simulating a meteorological drought over China were evaluated. The models are archived at the Program on Climate Model Diagnosis and Intercomparison (PCMDI) website (https://pcmdi.llnl.gov/index.html, accessed on 10 January 2021). Limited by the daily scale CMIP6 data, 15 climate models were selected in this research (Table 1). In this research, data sets were selected as daily data. We first processed daily scale data into monthly scale data and then calculated the meteorological drought index (SC-PDSI). The reason for selecting daily scale data is to lay the foundation for further research and the CMIP6 daily dataset provides more possibilities for drought research, such as hydrological drought and its connection with meteorological drought. The historical period of CMIP6 data covers 54 years, from 1961 to 2014; therefore, the future period is selected to be from 2021 to 2074. Considering future conditions and global warming signals, four Shared Socioeconomic Pathways (SSP) scenarios (SSP126, SSP245, SSP370 and SSP585) with low, medium and high future greenhouse gas emissions were simulated and compared. All model outputs were first bilinearly-interpolated to the same 10 km × 10 km grid as the observations for the different spatial resolutions of different GCMs. In addition to precipitation and temperature data, the soil available water capacity is also required to calculate the SC-PDSI and the soil available water capacity data were obtained from the available water capacity dataset provided by Webb et al. (2000) in globally gridded digital format.

The Equidistant Cumulative Distribution Function (EDCDF) as a bias correction method (Li et al. 2010) was used in this research for bias correction of GCMs. It takes advantage of the difference between the cumulative distribution characteristics of climate elements simulated by GCMs and those regionally measured to correct the deviation of simulated values and effectively capture the extreme values. The EDCDF method improves the inherent errors in climate model data and the limitations of applied interpolation methods to increase the simulation accuracy of climate elements. The EDCDF has good applicability in correcting the deviation of precipitation and temperature in China (Zhang et al. 2021). Based on the CMA dataset, bilinear interpolation and the EDCDF method are combined to downscale and correct the precipitation and temperature of the 15 CMIP6 models (Table 1) in both historical (1961–2014) and future (2021–2074) periods.

The PDSI introduced the concept of ‘Climatically Appropriate for Existing Condition’ (CAFEC) precipitation (Palmer 1965), which means the minimum precipitation required to maintain normal soil moisture levels in a given area. The magnitude of CAFEC precipitation depends on local climatic conditions. The difference between actual precipitation and CAFEC precipitation can reflect the moisture departure in a given area at a given time.

A simple two-layer soil water balance model was used to calculate the actual and probable values of hydrological components, based on meteorological observed precipitation and temperature data. Meanwhile, the Possible Evapotranspiration (PET) was estimated using the Thornthwaite method based on the monthly average temperature. In addition, the PDSI was calculated by the adaptive algorithm proposed by Wells et al. (2004). Table 2 shows the classification of drought according to the Palmer's recommendation.

The wavelet coherence method (Torrence & Webster 1999) is used to analyse the driving meteorological factor of meteorological drought. The wavelet coherence method is usually referred to as the Fourier squared coherency. When comparing two variables, it is used to identify frequency bands that covary between two time series of variables. After the normalized calculation of the wavelet power spectrum, the wavelet coherence is obtained by the square of the absolute value of the wavelet coherence spectrum. This study used the wavelet coherence method to calculate the frequency band and time interval of meteorological variables.

The return period is a method that can respond to the degree of drought risk. The application of the return period has gradually evolved from univariate to multivariate (Hao & Singh 2015). For a multivariate return period, it is usually necessary to construct joint distribution functions.

To analyse meteorological drought characteristics during the historical period (from 1961 to 2014) and future period (from 2021 to 2074), the intensity and duration were selected as two drought measures. We analysed both the evolution patterns of meteorological drought characteristics in different basins in China and the risk assessment of predicted drought intensity and duration.

Drought risk is also an important evaluation indicator for meteorological drought, then we analysed the return period of meteorological drought. We fit the duration and intensity distribution of each basin using five distributions, the Normal, Gamma, Weibull, Generalised Extreme Value (GEV) and Exponential (EXP) distributions. The Kolmogorov–Smirnov test (Massey 1951) is selected to test the applicability of the distribution. The results all passed the 99% confidence test. In the historical period, the optimal fitting distribution is the Weibull distribution for drought duration and intensity. In the future period, the optimal distribution of drought duration is the Weibull, Gamma and GEV distributions and the optimal distribution of intensity is the Weibull distribution. In general, the Weibull distribution is the optimal fitting distribution of the two drought characteristic variables in historical and future periods. Therefore, the Weibull distribution function was used to fit and calculate the duration and intensity of the meteorological drought in all basins.

Overall, there is a large difference in the range of probability variation of drought duration and drought intensity under the two drought return periods. Compared with historical periods, the value of drought intensity in the future period is larger for a certain probability level. Meanwhile, the probability change rates of drought intensity and duration are different and the univariate drought risk analysis is not comprehensive enough. Therefore, instead of focusing on a single characteristic of drought, we should incorporate multiple drought risk evaluation indicators.

Second, we used the copula method to carry out a multifactor drought risk assessment by combining drought duration and intensity. We selected four couple distributions (Clayton copula, Gaussian copula, Gumbel copula and Frank copula) and three evaluation indices (Mean Square Error (MSE)), RMSE and Pearson Correlation Coefficient (R)) to optimize the distribution. The fitting distributions of the above two drought characteristics in nine river basins in China were evaluated. As seen from the evaluation results, the joint distribution effect of the Frank copula function was the best. Therefore, this study applied the Frank copula function as the joint distribution function in all river basins.

Meteorological droughts are influenced by meteorological variables. To explore the influence of meteorological variables on meteorological droughts, we discuss and analyse both precipitation and temperature.

The covarying relationship between temperature and drought intensity is significantly correlated with low emissions in the future period and significant coherence within the 0–4 years band shows an in-phase difference of 10°–90° (Figure 10). The results show that precipitation is the main influencing factor of meteorological drought, with the most significant correlation in the historical period and mainly occurs in the northern basin (B1), the central basins (B2–B4) and the western basins (B8–B9). In the future, the areas with a significant correlation between precipitation and drought intensity are mainly concentrated in northeast basins (B1–B3) and southeast coastal basins (B6–B7). In particular, the correlation is more prominent under the higher emission scenario.

Meteorological drought is influenced by many factors and among the meteorological factors, precipitation and temperature dominate. The results in Section 4.3 show that precipitation appears to have a more pronounced effect on meteorological drought than the temperature in China. There have been many studies on the drivers of meteorological drought in other countries around the world. For example, in the midlands of the UK, Rahmani & Fattahi (2021) found that using nonlinear dynamics and cross-correlation, precipitation has a greater effect on meteorological droughts than temperature. This indicates that meteorological droughts are more sensitive to precipitation fluctuations than to temperature fluctuations. However, whether precipitation is the main driver of a meteorological drought depends on the actual precipitation situation in the study area. For example, Spinoni et al. (2015) found that in Central Europe and the Balkans, precipitation did not increase significantly and temperature drives the increase in meteorological drought severity. However, in Northern and Eastern Europe, although there has been a marked increase in temperature, the significant increase of precipitation leads to a decrease in drought characteristics (Spinoni et al. 2015). Therefore, whether meteorological drought is driven by precipitation or temperature mainly depends on the degree of precipitation decrease or temperature increase. In China, Liu et al. (2016) also found that the Precipitation–Concentration Degree (PCD) is significantly correlated with meteorological drought. Results show that the higher the PCD, the longer the duration of drought and more frequently the drought occurred. Therefore, precipitation has a significant impact on meteorological drought in China. There were a few limitations regarding our analysis on the drivers of meteorological drought, we did not consider other parameters, such as evapotranspiration, wind speed and sunshine. Next, we will consider a variety of meteorological factors and explore their impacts on meteorological drought.

For the temperature variable in China, total uncertainty will initially decrease and then increase in the future (2021–2074), reaching its minimum in approximately 2040. Model uncertainty and scenario uncertainty of temperature impact total fractional uncertainty. This conclusion is consistent with the research results of You et al. (2021). As time progresses, the total fractional uncertainty of precipitation will decrease, which dominates the contributions from the model and internal variability. This feature is also found in the global CMIP6 uncertainty study (Zhang & Chen 2021). This shows that the main contribution of temperature and precipitation uncertainty comes from model uncertainty, among the three sources of uncertainty. Meanwhile, model uncertainty shows clear regional differences. Model uncertainties in northern basins (B1–B4) and western basins (B8–B9) have the greatest impacts. Therefore, in the application of CMIP6 in China, the CMIP6 model should be carefully selected to reduce the simulation error caused by CMIP6. In addition, the correction method of climate model dataset also has a significant impact on model uncertainty. At present, there are many correction methods for climate models and there are certain errors in different correction results. The corrected errors will also be transferred to the model uncertainty. Therefore, in addition to the selection of climate models, the selection of correction methods should also be emphasised.

This research investigated the spatio-temporal evolution and the future risk of meteorological drought over nine river basins in China based on the bias-corrected CMIP6 model outputs. The drought risk was projected based on copula of probabilistic multivariate drought characteristics. The results show a significant increasing trend of drought intensity and duration in the Northwest River, Southwest River and Songliao River Basins under four SSP scenarios (SSP126, SSP245, SSP370 and SSP585). The low and medium emission scenarios have more volatile drought risks than the others. Meanwhile, the driving factors of a meteorological drought are affected by the climatic characteristics and the precipitation is the most important factor in most basins in China.

Although the bias-corrected CMIP6 models perform well in China, model uncertainty is substantial in both temperature and precipitation, which can be transmitted from climate pattern to drought risk. The adoption of GCMs could introduce some uncertainties into drought prediction, we should explore the uncertainties in future drought work and it is a necessary aspect that could be better understood. Quantifying the uncertainty in drought risk from CMIP6 will reduce the impact of uncertainty in the drought risk projection.

This study provided a high-precision dataset of CMIP6 at 10 km spatial resolution, which is conducive to couple GCMs and hydrological models in hydrological process analysis. Focusing on the meteorological drought CMIP6 projections, our study could provide a reference for evaluating the drought recognition ability of CMIP6, especially in different river basins in China. In order to reduce the potential impacts of drought, future trends of drought need to be fully understood to lay the foundation for drought adaptation and mitigation strategies.

The authors express our gratitude to Ming Pan (The Scripps Institution of Oceanography) for helpful discussions, and we thank the China Meteorological Administration (CMA) and the Program on Climate Model Diagnosis and Intercomparison (PCMDI) for making available the meteorological station datasets and CMIP6 models datasets. The authors acknowledge the support of the National Natural Science Foundation of China. We also thank the editors and reviewers for their efforts on this manuscript.

All authors contributed to the study conception and design. Material preparation, data collection and analysis were performed by Mengru Zhang and Linyan Zhang. The first draft of the manuscript was written by Mengru Zhang. All authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.

This work was supported by the General Project of the National Natural Science Foundation of China (Grant No. U2243203, 52079036 and 42071040).

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.