Characteristics of Propagation From Meteorological Drought to Hydrological Drought in the Pearl Rive

珠江气象干旱向水文干旱的传播特征

Abstract

Drought events occur more frequently under recent climate change. Generally, meteorological drought is the fuse of hydrological drought; thus, it is important to understand the characteristics of meteorological drought and its propagation to hydrological drought for early warning.

Taking the Pearl River Basin (PRB) in China as study area, this study adopted K-means cluster analysis method to divide the PRB into subregions with similar precipitation characteristics. Then, standardized precipitation index and standardized runoff index were used to analyze the characteristics of meteorological drought and hydrological drought, respectively, and the maximum Pearson correlation coefficient was used to determine the drought propagation time (DPT) between these two types of drought. Moreover, the link between meteorological drought and hydrological drought was explored based on continuous wavelet transform and cross wavelet transform. The results revealed that: the PRB has experienced severe meteorological and hydrological droughts since early 2000s, and hydrological drought was more serious than meteorological drought in each of the five subregions in the PRB. The DPTs from meteorological drought to hydrological drought were mainly 2–6 months, and the periodic characteristics of meteorological drought were mainly responsible for those of hydrological drought. Precipitation and runoff could greatly affect the DPT, while the impacts of evapotranspiration and shallow soil moisture on the DPT were not significant. Furthermore, El-Niño Southern Oscillation and Pacific Decadal Oscillation are important factors that affect the DPT from meteorological to hydrological drought in the PRB.

摘要

近年来，干旱事件在气候变化下更加频繁。通常，气象干旱是水文干旱的导火索；因此，了解气象干旱的特征及其向水文干旱的传播对于及早预警至关重要。

本研究以中国珠江流域为研究区域，采用K均值聚类分析方法将珠江流域划分为具有类似降水特征的子区域。然后，利用标准化降水指数和标准化径流指数分别分析了气象干旱和水文干旱的特征，并利用最大的皮尔逊相关系数确定了这两种类型干旱之间的干旱传播时间（DPT）。此外，基于连续小波变换和交叉小波变换探讨了气象干旱与水文干旱之间的联系。结果表明：自21世纪初以来，珠江流域经历了严重的气象和水文干旱，而水文干旱比气象干旱更为严重。从气象干旱到水文干旱的DPT主要是2-6个月，气象干旱的周期特征主要影响了水文干旱的周期特征。降水和径流对DPT的影响很大，而蒸发蒸腾和浅层土壤水分对DPT的影响不显著。此外，厄尔尼诺-南方涛动和太平洋年代振荡是影响气象干旱向水文干旱传播的重要因素。

1. Introduction

Due to global warming, extreme weather events occur more frequently. Compared with other natural hazards, drought is a slow-onset hazard that occurs in large areas over long time periods. Drought can have great impacts on socioeconomic development and ecological stability (King-Okumu et al, 2020; Shi et al, 2018; Varela et al, 2019; Zhou, Shi, Fu, Li, Gan, & Liu, 2020). From the perspective of drought development, drought is mainly caused by long-term precipitation deficit, which may further lead to soil moisture and runoff shortage through water cycle. Therefore, it is essential to enhance drought monitoring and investigate its evolution for disaster prevention and mitigation.

Generally, drought can be divided into four types, which are meteorological drought, hydrological drought, agricultural drought, and socioeconomic drought (American Meteorological Society, 2013). Meteorological drought, which is related to long-term lack of precipitation, is usually the first step of drought occurrence (Yuan et al, 2017). As a basic meteorological variable that can greatly influence the water circulation and hydrological processes, less precipitation may lead to water shortage in rivers, lakes, reservoirs, or even soil; then, hydrological drought occurs. Moreover, agricultural drought and socioeconomic drought are usually associated with the cases that water supply cannot meet the demand for irrigation and socioeconomic development, respectively (Shi et al, 2018). So far, there have been numerous indices to describe different types of drought. For example, Fu et al (2018) used standardized precipitation index (SPI) to analyze the characteristics of meteorological drought in Northeast China. Veettil and Mishra (2020) analyzed the relationship between hydrological drought and climate, watershed, and other factors by using standardized runoff index 

(SRI) at different time scales. Ding et al (2020) analyzed the response of vegetation to agricultural drought in different regions of China using PDSI (Palmer drought severity index). S. N. Liu, Shi, & Sivakumar (2020) used socioeconomic drought index (SEDI) to identify and predict the socioeconomic drought events in the Pearl River Basin (PRB).

Previous studies have focused on a single drought characteristic or the relationship between several types of drought (Fu et al, 2018; F. Wang et al, 2020; J. F. Wu, Liu, et al, 2018). However, due to climate change and human activities, the evolution mechanism of different types of drought, which has become an important part of drought monitoring and prediction, is still not clear and need further research. The process in which one type of drought causes another type of drought is called drought propagation. The studies on drought propagation began around 2001. Stahl (2001) studied the basic characteristics and time of drought propagation. In recent years, a few studies have applied different advanced methods to the study of drought propagation and analyzed a range of different basins. Sattar et al (2018) used a Bayesian model to calculate the probability of hydrological drought after meteorological drought. Y. Xu, Zhang, et al (2019) studied the impacts of human activities on the propagation from meteorological to hydrological drought in northern China and found that human activities accelerated the drought propagation. S. Z. Huang et al (2015) studied the propagation relationship between meteorological drought and agricultural drought in the Weihe River Basin and found that large-scale climate patterns were important factors in affecting drought propagation. M. L. Zhao et al (2019) analyzed the relationship between socioeconomic drought and meteorological drought in the Datong River Basin and found that El-Niño Southern Oscillation (ENSO) had the greatest impact on drought propagation among several climate indices. However, global studies have found that there are various factors that can affect the drought propagation, including watershed properties, climate types and human activities (Apurv & Cai, 2020; Gevaert et al, 2018; J. F. Wu et al, 2017). Therefore, drought propagation usually has regional characteristics (N. C. Chen et al, 2020; R. H. Li, Chen, et al, 2020).

To this end, specific regional studies are still quite necessary. For example, in the PRB, serious drought events occurred in 1963, 1999, 2003, and later years (Q. Huang et al, 2019; S. N. Liu, Shi, Niu, et al, 2020), but there was still no effective early warning. This study aims to investigate the characteristics of propagation from meteorological drought to hydrological drought in the PRB, China. The objectives of this study are: (1) to analyze the characteristics of meteorological drought and hydrological drought in the whole PRB based on standardized indices; (2) to investigate the drought propagation time (DPT) in different subregions of the PRB and analyze the distinctions among different subregions; (3) to quantify the relationship between meteorological drought and hydrological drought at different time scales; and (4) to explore possible influencing factors of drought propagation. The outcomes of this study can provide new avenues to recognize the importance of drought propagation in drought analysis, especially in the context of climate change.

1 绪论

由于全球变暖，极端天气事件发生的频率更高。与其他自然灾害相比，干旱是一种缓慢发展的灾害，它在长时间内影响广泛的地区。干旱对社会经济发展和生态稳定性有很大影响（King-Okumu等，2020年; Shi等，2018年; Varela等，2019年; Zhou，Shi，Fu，Li，Gan和Liu，2020年）。从干旱发展的角度来看，干旱主要是由长期降水不足引起的，这可能通过水循环进一步导致土壤湿度和径流不足。因此，加强干旱监测并调查其演变对于灾害预防和减轻至关重要。

通常，干旱可以分为四种类型，即气象干旱、水文干旱、农业干旱和社会经济干旱（美国气象学会，2013年）。气象干旱与长期降水缺乏有关，通常是干旱发生的第一步（Yuan等，2017年）。降水量较少可能导致河流、湖泊、水库甚至土壤中的水源不足，随后发生水文干旱。此外，农业干旱和社会经济干旱通常与供水无法满足灌溉和社会经济发展需求的情况相关联（Shi等，2018年）。到目前为止，已经有许多指数来描述不同类型的干旱。例如，Fu等（2018年）使用标准化降水指数（SPI）分析了中国东北地区气象干旱的特征。Veettil和Mishra（2020年）利用不同时间尺度上的标准化径流指数（SRI）分析了水文干旱与气候、流域等因素的关系。Ding等（2020年）利用帕尔默干旱严重指数（PDSI）分析了中国不同地区植被对农业干旱的响应。刘胜年，Shi和Sivakumar（2020年）利用社会经济干旱指数（SEDI）识别和预测了珠江流域的社会经济干旱事件。

以往的研究主要关注单一干旱特征或几种干旱类型之间的关系（Fu等，2018年;王等，2020年;吴景峰，刘等，2018年）。然而，由于气候变化和人类活动，不同类型干旱的演变机制，这已成为干旱监测和预测的重要组成部分，仍然不清楚，需要进一步研究。一个类型的干旱引起另一类型干旱的过程称为干旱传播。对干旱传播的研究始于2001年左右。Stahl（2001年）研究了干旱传播的基本特征和时间。近年来，一些研究应用了不同的先进方法研究了干旱传播，并分析了一系列不同的流域。Sattar等（2018年）利用贝叶斯模型计算了气象干旱后水文干旱的概率。Y. Xu，张等（2019年）研究了人类活动对中国北方气象干旱传播的影响，并发现人类活动加速了干旱传播。S. Z.黄等（2015年）研究了渭河流域气象干旱与农业干旱之间的传播关系，发现大尺度气候模式是影响干旱传播的重要因素。M. L.赵等（2019年）分析了大同河流域社会经济干旱与气象干旱之间的关系，发现厄尔尼诺南方涛动（ENSO）是影响干旱传播的几个气候指标中影响最大的。然而，全球研究发现有各种各样的因素可以影响干旱传播，包括流域特性、气候类型和人类活动（Apurv&Cai，2020年;Gevaert等，2018年;吴景峰等，2017年）。因此，干旱传播通常具有区域特征（陈能超等，2020年;李如何，陈等，2020年）。

因此，特定区域研究仍然非常必要。例如，在珠江流域，1963年，1999年，2003年和以后的几年发生了严重的干旱事件（黄等，2019年;刘胜年，石，牛等，2020年），但仍然没有有效的预警。本研究旨在调查中国珠江流域气象干旱传播到水文干旱的特征。本研究的目标是：（1）基于标准化指数分析珠江流域整体的气象干旱和水文干旱特征；（2）

调查珠江流域不同子区域的干旱传播时间（DPT），并分析不同子区域之间的差异；（3）量化不同时间尺度上气象干旱与水文干旱之间的关系；以及（4）探讨干旱传播的可能影响因素。本研究的结果可以为认识干旱传播在干旱分析中的重要性，特别是在气候变化背景下，提供新的途径。

在全球变暖的背景下，干旱等极端天气事件的发生频率增加。与其他自然灾害相比，干旱通常是一种缓慢发展的灾害，影响范围广泛，时间跨度长。干旱对社会经济发展和生态稳定性具有重要影响。因此，加强干旱监测和调查其演变对于预防和减轻灾害影响至关重要。

干旱通常可分为气象干旱、水文干旱、农业干旱和社会经济干旱等四种类型。气象干旱是指长期降水不足，通常是干旱发生的首要因素。降水不足可能导致河流、湖泊、水库甚至土壤中的水源减少，从而引发水文干旱。此外，农业干旱和社会经济干旱通常与供水无法满足农业灌溉和社会经济发展需求相关联。

过去的研究主要集中在单一干旱特征或几种干旱类型之间的关系上，而对不同类型干旱的演变机制了解不足。干旱传播是指一种干旱类型引发另一种干旱类型的过程。近年来，一些研究开始关注干旱传播的机制，并尝试应用各种方法进行研究。然而，全球范围内的研究表明，干旱传播受到多种因素的影响，包括流域特性、气候类型和人类活动等。因此，干旱传播通常呈现出区域性特征。

针对特定区域的研究仍然非常必要。例如，在珠江流域，曾发生过多起严重干旱事件，但仍缺乏有效的早期预警系统。因此，本研究旨在深入探究珠江流域气象干旱向水文干旱传播的特征。通过分析珠江流域的气象干旱和水文干旱特征、研究不同子区域的干旱传播时间、量化不同时间尺度上的干旱关系，并探讨可能影响干旱传播的因素，本研究的结果可以为干旱分析提供新的视角和方法，特别是在气候变化的背景下。

2. Materials and Methods

2.1. Study Area and Data

The Pearl River, which flows through six provinces in South China (Figure 1), is the second largest river in terms of water resources in China (S. N. Liu, Shi, Niu, et al, 2020). The PRB consists of three major tributaries, that is, the West River, North River, and East River. The total drainage area of the PRB is 453,690 km2 , and the climate of the PRB is tropical and subtropical. The annual mean temperature of the PRB is 14–22°C and the annual mean precipitation is 1,200–2,200 mm. Precipitation in the PRB is distributed with highly spatial-temporal heterogeneities, for example, 80% of precipitation usually occurs from April to September.

Therefore, floods and droughts may easily occur in the rainy season and dry season, respectively.

Because meteorological drought and hydrological drought are closely related to precipitation and runoff, respectively, this study selects the precipitation and runoff data to calculate the corresponding drought indices. Meanwhile, the long-term datasets are needed to fully understand the spatial and temporal characteristics of drought propagation process. Although station observations are regarded to be relatively reliable at the point where a meteorological station locates, the density of the meteorological station network is generally not high enough to fully describe the spatial distributions of meteorological variables (Shi et al, 2020). From the perspective of spatial coverage, grid-based data sets (Harrigan, et al, 2020; Mishra,et al, 2011) can be considered as a viable substitute or supplement for station observations when conducting drought analysis. Therefore, the precipitation and runoff data provided by European Centre for Medium-Range Weather Forecasts (ECMWF) are used in this study, which are from 1981 to 2019 with a spatial resolution of 0.1°. ERA-5 is the fifth generation of ECMWF global reanalysis, succeeding ERA-Interim and covering the entire globe. ERA5-Land reanalysis data set is obtained through a series of improvements to ERA-5 data set, which makes it more accurate in all types of land applications (Pelosi et al, 2020). Based on this data set, SPI and SRI are calculated at different time scales. In addition, in order to analyze the variations of influencing factors during the process of drought propagation, this study also uses the evapotranspiration and three layers of soil moisture (SM1: 0–7 cm, SM2: 7–28 cm, SM3: 28–100 cm) of ERA5Land reanalysis data set. Moreover, the annual ENSO and Pacific Decadal Oscillation (PDO) data provided by National Oceanic and Atmospheric Administration (NOAA) are used to explore the driving forces of drought propagation.

2.2. Methods

In this study, Mann-Kendall (MK) trend analysis method and Sen's slope method are adopted to evaluate the trends of precipitation and runoff at different time scales. K-means cluster analysis and silhouette coefficient (SC) are used to classify the subregions with similar precipitation characteristics. SPI and SRI are used to identify meteorological drought and hydrological drought, respectively. The Pearson correlation coefficient (PCC) is used to identify the DPT, while continuous wavelet transform (CWT) and cross wavelet transform (XWT) are used to explore the link between meteorological drought and hydrological drought.

The flow chart of this study is shown in Figure 2.

2.2.1. Trend Analysis Method

The MK trend analysis method is a nonparametric test method, which is often used to identify the trend of time series. It has been widely used in the field of hydrometeorology (Fu et al, 2018; Zhou, Ding, et al, 2020; Zhou, Shi, Fu, Li, Gan, Liu, & Liu, 2020). Moreover, Sen's slope method is often used together with the MK method to further determine the trend magnitude of time series (Shi et al, 2017; Zhou, Shi, Fu, Li, Gan, & Liu, 2020). In this study, these two methods are used to analyze the variations of precipitation (for meteorological drought) and runoff (for hydrological drought) at different time scales. The specific calculation processes are as follows: For the time series X=\left \{x_1,x_2, .... x_n \right \}    the MK rank statistic S_k can be given by the following formula:

Figure 2. The flow chart of this study

where:

then, the statistic Z can be calculated as:

where

where t is the range of any given node. The normal statistic Z is used to indicate the strength of the trend (Zhou, Shi, Fu, Li, Gan, & Liu,  2020). The positive and negative values of Z represent increasing and  decreasing trends, respectively. When Z value is greater than 1.96, it represents a significant increasing trend. Similarly, when Z is less than −1.96, it represents a significant decreasing trend.

The magnitude of the trend, β, was evaluated by Sen's slope method, which can be calculated as follows:

The positive/negative value of β represents an upward/downward trend.

2.2.2. Drought Regionalization

The topography of the PRB is complicated and greatly affected by monsoon, so the spatial and temporal distributions of precipitation are uneven in this river basin. Therefore, to more reasonably reveal the process of drought propagation caused by precipitation deficit in different locations of the PRB independently, K-means cluster analysis is used to classify the subregions with similar precipitation characteristics.

K-means cluster analysis is suitable for processing large datasets and is a commonly used nonhierarchical clustering algorithm with high computational efficiency (B. J. Liu et al, 2016; Y. Yang et al, 2020). This method reduces the redundant information in the analysis of drought propagation process. The K-means method searches K clusters randomly in each iteration to initialize the center, and then measures the distance between the data point (xij) and the center (cj). The goal is to minimize the specified objective function as follows:

However, the results of K-means cluster analysis can be affected by the initial cluster centers and the number of selected clusters. Therefore, SC, which represents the cohesion of a cluster sample and the degree of separation between a cluster sample and other samples, is usually selected to evaluate whether the number of clusters is reasonable. The specific calculation process is as follows:

where a（ i） is the average distance between the i-th sample and other members in the same cluster, and  , b（i,k） is the mean distance from the i-th sample to the k-th cluster member. The SC values range from −1 to 1. The higher SC value represents the better classification, and the negative SC value means that the clustering result is incorrect. When the average SC value reaches the maximum and the number of negative values is the least, it is the optimal number of clusters.

2.2.3. Drought Indices

This study chooses SPI and SRI to identify meteorological drought and hydrological drought, respectively.

Both indices can quantify drought events at different time scales. The procedure to calculate SPI and SRI are the same in the present study, which includes three steps. First, the cumulative precipitation/runoff series are obtained from the original precipitation/runoff series of a specific time scale. Second, these long-term records are fitted into a probability distribution and the fitted distribution is used to estimate the cumulative probability of time series values. Third, the SPI/SRI value is calculated by converting the cumulative probability to a standard normal distribution. Niu et al (2015) pointed out that precipitation and runoff in the PRB are more in line with the log-normal distribution. Therefore, this study selects the log-normal distribution to fit the distribution of precipitation and runoff in the calculation process. The probability distribution function of the log-normal distribution is shown as follows:

where  is the standard deviation of logarithmic values, and  is the mean of logarithmic values. Once F(x) is calculated, SPI/SRI can be easily obtained as the standardized value of F(x).

3. Results

3.1. Variations of Precipitation and Runoff

Figure 3 shows the temporal and spatial variations of precipitation and runoff in the PRB. In general, the spatial distribution patterns of precipitation and runoff in the PRB were similar. From 1981 to 2019, the variation range of precipitation was from −27.07 mm/a to 2.48 mm/a, and the average changing rate of precipitation was −7.97 mm/a. Precipitation in the western part of the PRB showed an increasing trend, but the growth trend was not significant. In other parts, precipitation showed decreasing trends, among which the trends in the northeastern part of the PRB were significant (see Figures 3a and 3c).

Figure 3b shows that the variation range of runoff was from −26.97 mm/a to 4.18 mm/a, with the average changing rate of −8.85 mm/a. Runoff in most parts showed decreasing trends, while the southern part showed an increasing trend. However, the increasing trends of runoff in most parts were not significant, and the significant decreasing trends were found in the northeastern and western parts of the PRB (see Figure 3d)

Different from the spatial distribution of precipitation, the area where runoff showed a significant decreasing trend was larger. This is because the runoff change was not only affected by precipitation, but also closely related to vegetation conditions and regional human activities (T. S. Li and Xia, 2018; Y. D. Yang et al, 2019).

The western part of the PRB has higher altitude, larger slope, and relatively more complex terrain, which may affect the process of flow concentration. Moreover, the variations and significant spatial differences of precipitation and runoff in the eastern part may be caused by urbanization (Wai et al, 2017). Urbanization can generate a larger number of suspended particles which are usually regarded as the condensation nuclei of water vapor, and thus, change the state of the rain-forming conditions (W. B. Li et al, 2009).

图3显示了鄱阳湖流域（PRB）降水和径流的时空变化。总体而言，鄱阳湖流域降水和径流的空间分布模式相似。从1981年到2019年，降水变化范围为−27.07毫米/年至2.48毫米/年，平均变化率为−7.97毫米/年。鄱阳湖流域西部的降水呈增加趋势，但增长趋势不显著。在其他地区，降水呈减少趋势，其中鄱阳湖流域东北部的趋势显著（见图3a和3c）。

图3b显示，径流的变化范围为−26.97毫米/年至4.18毫米/年，平均变化率为−8.85毫米/年。大多数地区的径流呈减少趋势，而南部地区呈增加趋势。然而，大多数地区径流的增长趋势不显著，而鄱阳湖流域东北部和西部地区的显著减少趋势。

与降水的空间分布不同的是，径流呈显著减少趋势的区域更大。这是因为径流变化不仅受降水影响，还与植被状况和区域人类活动密切相关（T. S. Li和夏，2018年；Y. D. Yang等，2019年）。

鄱阳湖流域西部地势较高，坡度较大，地形相对较复杂，这可能影响了流量集中过程。此外，东部降水和径流的变化和显著空间差异可能是由城市化引起的（Wai等，2017年）。城市化可以产生更多的悬浮颗粒，通常被视为水蒸气的凝结核，从而改变降水形成条件的状态（W. B. Li等，2009年）。

3.2. Classification of Different Subregions

In this study, K-means cluster analysis was used to classify the subregions with similar dry and wet characteristics according to the precipitation, topography, longitude and latitude of the PRB. The SC was used to evaluate the clustering results, as shown in Table 2.

Known from Table 2, when the number of clusters was 2, the SC was the largest (i.e., 0.45); however, since the number of negative SC was also the largest (i.e., 267), this number of clusters was not reasonable. When the number of clusters was 5, the SC was the second largest (i.e., 0.44), and the number of negative SC was 0. As a result, this study regarded the number of clusters as 5, and the PRB was classified into five subregions (i.e., subregion Ⅰ, Ⅱ, Ⅲ, Ⅳ, and Ⅴ). B. J. Liu et al (2016) also used K-means cluster analysis to study the extreme precipitation in the PRB, and the results showed that when the number of clusters was 5, it was reasonable. C. Huang et al (2017) analyzed the drought risk of the PRB by using rotated empirical orthogonal function to divide different homogeneous spaces, and the results showed that it was reasonable to divide five subregions. Therefore, the result in this study is basically consistent with those in previous studies. The spatial distribution of each subregion was shown in Figure 4. Subregions Ⅱ and Ⅳ are located in the upper reaches of the West River Basin. Subregions Ⅰ and Ⅲ are located in the middle and lower reaches of the West River Basin. Subregion Ⅴ contains the North River Basin and East River Basin (see Figures 1 and 4). Subregions Ⅰ, Ⅲ, and Ⅴ are characterized by low elevations (see Figure 1) and the average annual precipitation is relatively large. The average annual precipitation is 1828.01 mm, 2,371 mm and 1846.67 mm, respectively.

Compared with subregions Ⅰ, Ⅲ, and Ⅴ, subregions Ⅱ and Ⅳ are characterized by high elevations and the annual precipitation is relatively small. The average annual precipitation is 1353.02 mm and 1678.62 mm, respectively. The PRB is often affected by the warm air from the South Sea, so the precipitation values of subregions Ⅰ, Ⅲ, and Ⅴ are large. Due to the topography, the warm and wet air from the South Sea and the cold air from the north converge in subregion Ⅲ, which leads to the largest annual average precipitation in this region (B. J. Liu et al, 2016). Subregion Ⅴ has been strongly affected by human activities, for example, urbanization may increase precipitation (W. B. Li et al, 2009). Subregions Ⅱ and Ⅳ have relatively smaller annual average precipitation because their higher altitudes can hold back the warm and moist air from the South Sea (B. J. Liu et al, 2016).

在这项研究中，利用K均值聚类分析根据鄱阳湖流域的降水、地形、经度和纬度等特征将具有相似干湿特征的次区域进行分类。

使用了SC来评估聚类结果，如表2所示。

从表2可以得知，当聚类数为2时，SC最大（即0.45）；然而，由于负SC的数量也最大（即267个），因此这个聚类数是不合理的。

当聚类数为5时，SC第二大（即0.44），负SC的数量为0。

因此，本研究将聚类数视为5，并将鄱阳湖流域分为五个次区域（即Ⅰ、Ⅱ、Ⅲ、Ⅳ和Ⅴ区）。

B.J.刘等人（2016）也利用K均值聚类分析研究了鄱阳湖流域的极端降水情况，结果表明当聚类数为5时是合理的。

C.黄等人（2017）通过使用旋转的经验正交函数分析了鄱阳湖流域的干旱风险，结果显示将其划分为五个次区域是合理的。

因此，本研究的结果基本与以前的研究一致。

每个次区域的空间分布如图4所示。Ⅱ区和Ⅳ区位于西江流域的上游。

Ⅰ区和Ⅲ区位于西江流域的中下游。Ⅴ区包含北江流域和东江流域（见图1和图4）。Ⅰ、Ⅲ和Ⅴ区的海拔较低（见图1），年平均降水量相对较大，分别为1828.01毫米、2371毫米和1846.67毫米。

与Ⅰ、Ⅲ和Ⅴ区相比，Ⅱ和Ⅳ区的特点是海拔较高，年降水量相对较小，平均年降水量分别为1353.02毫米和1678.62毫米。鄱阳湖流域经常受到南海暖湿气流的影响，因此Ⅰ、Ⅲ和Ⅴ区的降水量较大。由于地形的影响，来自南海的温暖湿润气流与来自北方的冷空气在Ⅲ区交汇，导致该地区年平均降水量最大（B.J.刘等人，2016）。Ⅴ区受到人类活动的影响较大，例如，城市化可能会增加降水量（W.B.李等人，2009）。Ⅱ和Ⅳ区的年平均降水量相对较小，因为它们较高的海拔可以阻挡来自南海的温暖湿润空气（B.J.刘等人，2016）。

3.3. Spatiotemporal Patterns of Meteorological Drought and Hydrological Drought

Figure 5 shows the variations of meteorological drought and hydrological drought at different time scales in different subregions during the study period. Known from Figure 5, the spatiotemporal patterns of meteorolog 

ical drought and hydrological drought were basically similar. From 1981 to 2000, all the subregions showed the changing characteristics of dry and wet alternations for either meteorological drought or hydrological drought. SPIn and SRIn in each subregion changed little and no obvious wet and dry events occurred. After 2000, both the two types of drought frequently occurred; hydrological drought was more serious than meteorological drought. Based on the data from Pearl River Water Resources Commission (www.pearlwater.gov.

cn/xxcx/szygg), severe droughts occurred in 2003, 2004, 2007, 2009, and 2010, which is consistent with the results shown in Figure 5. In addition, with the increase of time scale (i.e., from 1 to 12 months), the severity of meteorological drought and hydrological drought exacerbated.

Through comparing SPIn and SRIn, it is found that meteorological drought and hydrological drought were closely related in each subregion of the PRB; however, there were still certain differences. For instance, the durations of meteorological drought were generally shorter than those of hydrological drought. In each subregion, the temporal characteristics of meteorological drought and hydrological drought were highly consistent. From a regional perspective, severe droughts in subregions Ⅰ and Ⅴ occurred around 2003, while no significant drought occurred in other subregions; severe droughts in subregions Ⅱ, Ⅲ, and Ⅳ occurred around 2010, while only moderate droughts occurred in subregions Ⅰ and Ⅴ. The development of drought could be affected by topography, climatic conditions, and human activities, so that the spatial characteristics of drought among different subregions could be obvious. For instance, more severe droughts occurred in subregion Ⅴ than in other subregions, which is mainly due to the fact that subregion Ⅴ is located in Guangdong-Hong Kong-Macao Greater Bay Area, one of the regions with the fastest socioeconomic development in China. The rapid increase of water consumption and the excessive exploitation of groundwater has affected the development of drought (Han et al, 2019).

图5显示了在研究期间不同子区域的不同时间尺度下气象干旱和水文干旱的变化。

从图5可知，气象干旱和水文干旱的时空模式基本相似。

从1981年到2000年，所有子区域无论是气象干旱还是水文干旱都表现出干湿交替的变化特征。

每个子区域的SPIn和SRIn变化不大，没有明显的干湿事件发生。

2000年后，这两种类型的干旱经常发生；水文干旱比气象干旱更为严重。

根据珠江水利委员会的数据（www.pearlwater.gov.cn/xxcx/szygg），严重干旱发生在2003年、2004年、2007年、2009年和2010年，这与图5中显示的结果一致。

此外，随着时间尺度的增加（即从1到12个月），气象干旱和水文干旱的严重程度加剧。

通过比较SPIn和SRIn，发现珠江流域各个子区域的气象干旱和水文干旱密切相关，但仍存在一定差异。

例如，气象干旱的持续时间通常比水文干旱短。

在每个子区域中，气象干旱和水文干旱的时间特征高度一致。

从区域的角度来看，Ⅰ和Ⅴ子区域的严重干旱发生在2003年左右，而其他子区域没有发生明显的干旱；

Ⅱ、Ⅲ和Ⅳ子区域的严重干旱发生在2010年左右，而Ⅰ和Ⅴ子区域仅发生了中等干旱。

干旱的发展可能受地形、气候条件和人类活动的影响，因此不同子区域之间的干旱空间特征可能明显。

例如，在Ⅴ子区域比其他子区域更频繁地发生严重干旱，主要是因为Ⅴ子区域位于中国经济发展最快的珠江三角洲地区之一，水资源消耗的快速增加和地下水过度开发影响了干旱的发展（Han等，2019年）。

3.4. Propagation from Meteorological Drought to Hydrological Drought

To quantify the characteristics of propagation from meteorological drought to hydrological drought, the MPCC between SPIn and SRI1 was used to represent the DPT. Figure 6 shows the spatial distributions of the MPCC and the DPT from meteorological drought to hydrological drought in the PRB.

Figure  6a indicated that the response relationship between meteorological drought and hydrological drought in the PRB was extraordinarily strong. The MPCC ranges from 0.68 to 0.89, with clear patterns in spatial distribution. The lower MPCC values were mainly found in the coastal areas of subregions Ⅰ and Ⅴ, while the higher MPCC values were mainly found in subregion Ⅲ. This is consistent with the changing characteristics of precipitation and runoff in Figure 3, indicating that the relationship between meteorological drought and hydrological drought was much closer in subregions with larger changing magnitude of precipitation and runoff. Known from Figure 6b, the DPTs in the PRB were mainly between 2 and 6 months, also with clear patterns in spatial distribution. Therefore, this study further analyzed the MPCC and the DPT in each subregion (Figure 7). The strongest response relationship between meteorological drought and hydrological drought was found in subregion Ⅲ (Figure 7a), with the MPCC range of 0.80–0.82. Moreover, the proportion of the grids with the DPT of 2 months in this subregion was 70.3%. The weakest response relationship between meteorological drought and hydrological drought was found in subregion Ⅴ (Figure 7a), with the MPCC range of 0.75–0.80. The DPTs were mainly 2–4 months, and the proportions of the grids with the DPTs of 2, 3, and 4 months in this subregion were 29.2%, 43.8%, and 9.7%, respectively. Subregions Ⅰ, Ⅱ, and Ⅳ also had acceptable response relationships between meteorological drought and hydrological drought, and the DPTs were mainly 3–4, 5–6, and 3–4  months in subregions Ⅰ, Ⅱ, and Ⅳ, respectively.

Moreover, the proportions of the grids with different DPTs were 63.4% for 3 months and 22% for 4 months in subregion Ⅰ, 32.5% for 5 months and 31.4% for 6 months in subregion Ⅱ, and 45% for 3 months and 36.3% for 4 months in subregion Ⅳ.

为了量化气象干旱向水文干旱传播特征，MPCC在SPIn和SRI1之间被用来代表DPT。

图6显示了在PRB中从气象干旱到水文干旱的MPCC和DPT的空间分布。

图6a表明PRB中气象干旱和水文干旱之间的响应关系异常强大。

MPCC范围从0.68到0.89，具有清晰的空间分布模式。

较低的MPCC值主要集中在Ⅰ和Ⅴ亚区的沿海地区，而较高的MPCC值主要集中在Ⅲ亚区。

这与图3中降水和径流变化特征一致，表明降水和径流变化幅度较大的亚区之间的气象干旱和水文干旱关系更为密切。

从图6b可知，PRB中的DPT主要在2到6个月之间，也具有明显的空间分布模式。

因此，本研究进一步分析了每个亚区的MPCC和DPT（图7）。

在亚区Ⅲ中找到了气象干旱和水文干旱之间最强的响应关系（图7a），MPCC范围为0.80-0.82。此外，该亚区具有2个月DPT的网格比例为70.3%。

在亚区Ⅴ中找到了气象干旱和水文干旱之间最弱的响应关系（图7a），MPCC范围为0.75-0.80。DPT主要为2-4个月，该亚区的DPT为2、3和4个月的网格比例分别为29.2%，43.8%和9.7%。Ⅰ、Ⅱ和Ⅳ亚区的气象干旱和水文干旱之间也有可接受的响应关系，而DPT主要分别为3-4个月、5-6个月和3-4个月。

此外，在Ⅰ亚区，DPT不同的网格比例为3个月63.4%，4个月22%；在Ⅱ亚区，DPT为5个月32.5%、6个月31.4%；在Ⅳ亚区，DPT为3个月45%、4个月36.3%。

3.5. Link Between Meteorological Drought and Hydrological Drought

According to the results of Subsection 3.4, the DPT accounting for the maximum proportion of grids in each subregion could be determined, which were 3, 5, 2, 3 and 3 months in subregions Ⅰ, Ⅱ, Ⅲ, Ⅳ, and Ⅴ, respectively. Then, the link between meteorological drought and hydrological drought can be further analyzed based on SPIm (m represents the DPT accounting for with the maximum proportion of grids) and SRI1. CWT was used to determine the periodic characteristics of meteorological drought and hydrological drought in each subregion, and XWT was used to highlight their resonance frequencies and phase shifts in the time-frequency domain.

根据第3.4小节的结果，可以确定在每个子区域中占最大比例的网格的DPT，分别为Ⅰ、Ⅱ、Ⅲ、Ⅳ和Ⅴ区域的月份分别为3、5、2、3和3个月。

然后，可以基于SPIm（m代表占比最大的网格的DPT）和SRI1进一步分析气象干旱和水文干旱之间的联系。

使用CWT确定每个子区域的气象干旱和水文干旱的周期特征，使用XWT突出它们在时间频率领域的共振频率和相位差。

3.5.1. CWT Analysis of SPIm and SRI1

The CWT analysis was conducted at the confidence level of 95%, and Figure 8 shows the corresponding wavelet power spectrum. In this figure, the 95% significance confidence level against red noise is exhibited as a thick contour. Inside the black cone shape contour is the region without an edge effect, and outside the black cone contour is the region with cone of influence. The color bar on the bottom denotes wavelet energy.

In subregion Ⅰ, SPI3 had significant periodicities of 6–16, 12–16, 14–20, and 8–20 months during 1988–1996, 1990–1994, 1998–2002, and 2014–2017, respectively (Figure 8a); while SRI1 had generally similar periodic characteristics to SPI3, with the periodicities of 6–16 months during 1989–1991, 12–16 months during 1991–1994, and 14–18 months during 1998–2002 (Figure 8f). The significant periodicities of SPIm and SRI1 in all the five subregions are listed in Table  3. It is observed that, except for subregion Ⅱ, meteorological drought and hydrological drought in other subregions had similar periodic characteristics (highlighted by bold in Table  3). In subregion Ⅱ, SPI5 had significant periodicities of 4–12, 10–16, 4–12, 20–24, and 5–10 months during 1982–1984, 1985–1988, 1989–1993, 2008–2010, and 2007–2011, respectively; however, SRI1 did not have significant periodic characteristics.

CWT分析在95%的置信水平下进行，图8显示了相应的小波功率谱。在该图中，95%的显著性置信水平相对于红色噪声表现为一个粗实线轮廓。在黑色锥形轮廓内部是没有边缘效应的区域，在黑色锥形轮廓外部是有影响区域。底部的色条表示小波能量。

在Ⅰ子区，SPI3在1988年至1996年、1990年至1994年、1998年至2002年和2014年至2017年期间表现出6-16个月、12-16个月、14-20个月和8-20个月的显著周期性（图8a）；而SRI1在整体上具有与SPI3类似的周期特征，在1989年至1991年、1991年至1994年和1998年至2002年期间表现出6-16个月、12-16个月和14-18个月的周期性（图8f）。所有五个子区中SPIm和SRI1的显著周期性列在表3中。观察到，除了Ⅱ子区外，其他子区的气象干旱和水文干旱具有类似的周期特征（在表3中用粗体突出显示）。在Ⅱ子区，SPI5在1982年至1984年、1985年至1988年、1989年至1993年、2008年至2010年和2007年至2011年期间表现出4-12个月、10-16个月、4-12个月、20-24个月和5-10个月的显著周期性；然而，SRI1没有显著的周期特征。

3.5.2. XWT Analysis of SPIm and SRI1

Although SPIm and SRI1 in most subregions had similar periodic characteristics, their periodic structures were different. Figure 9 shows the cross wavelet power spectra of SPIm and SRI1 in different subregions. In this figure, the 95% significance confidence level against red noise is exhibited as a thick contour, and the relative phase relationship is denoted as arrows (antiphase pointing left, in-phase pointing right). Outside the black cone contour is the region with cone of influence. The color bar on the bottom denotes wavelet energy. The phase angle between two series can be used as a method to test their interaction relationship.

When two series are in phase, the phase angle is 0°, and the arrow on the scale diagram points to the right (→). When two series are in antiphase, the phase angle is 180°, and the arrow on the scale diagram points to the left (←). When the phase angle is 90° (↓), the first series leads the second series to 3 months. When the phase angle is −90° (↑), it means that the second series leads the first series to 3 months (Chang et al, 2017; Shuai et al, 2018).

As shown in Figure 9a, the phase arrows mostly point right, indicating that meteorological drought and hydrological drought presented a significant positive correlation, that is, at the time scale of 4–12 months (1982–1985), 4–16 months (1988–1997), 12–24 months (1998–2003), and 20–28 months (2007–2013). Most of the phase angle amplitudes are less than 10° in these periods, indicating that SPI3 and SRI1 are basically in the same phase in subregion Ⅰ, and there is almost no lag time. During 2007–2013, the phase arrows are less than 30°, indicating that the change of SPI3 leaded SRI1 for less than 1 month. That may be due to other factors interfering with the DPT (Han et al, 2019; S. Z. Huang et al, 2015). Although hydrological drought in subregion Ⅱ had no significant periodic characteristics, there was still a close relationship between meteorological drought and hydrologic drought at the time scale of 6–14 months (1981–1983), 10– 16 months (1985–1987), 6–14 months (1988–1990), 6–16 months (1991–1994), 28–32 months (1994–1998), 20–28 months (2007–2012), and 16–20 months (2014–2017) (Figure 9b). In these periods, the phase angles are very small, most of the phase angles is less than 10° and the maximum phase angle is less than 30°, indicating that SPI5 and SRI1 are basically in the same phase in subregion Ⅱ. Similarly, the characteristics of resonance frequency and phase shift for meteorological and hydrological droughts in others subregion could also be obtained. The significant correlation period in all the five subregions are listed in Table 4. Periods that coincide with periodic features in Table 3 are highlighted by bold in Table 4. In general, there was a significant positive correlation between meteorological drought and hydrological drought in phase change,which indicated that the periodic characteristics of meteorological drought were mainly responsible for those of hydrological drought. In addition, the results of XWT analysis proved that there was a stable relationship between meteorological drought and hydrological drought in the PRB, and the DPT determined in this study is reasonable. Thus, the determined DPTs would be of great value for regional drought monitoring and early warning in future.

尽管大多数子区域的SPIm和SRI1具有相似的周期特征，但它们的周期结构是不同的。图9显示了不同子区域中SPIm和SRI1的交叉小波功率谱。在这个图中，针对红色噪声的95%显著性置信水平显示为厚实的轮廓线，相对相位关系表示为箭头（同相指向右，反相指向左）。黑色锥体轮廓线外是影响区域。底部的色条表示小波能量。两个系列之间的相位角可以用作测试它们之间相互关系的方法。

当两个系列同相时，相位角为0°，标尺图上的箭头指向右（→）。当两个系列反相时，相位角为180°，标尺图上的箭头指向左（←）。当相位角为90°（↓）时，第一个系列领先于第二个系列3个月。当相位角为−90°（↑）时，意味着第二个系列领先于第一个系列3个月（Chang et al, 2017; Shuai et al, 2018）。

如图9a所示，相位箭头大多指向右，表明气象干旱和水文干旱呈显著的正相关，即在时间尺度为4-12个月（1982-1985年）、4-16个月（1988-1997年）、12-24个月（1998-2003年）和20-28个月（2007-2013年）。这些时期的大多数相位角幅度小于10°，表明在子区域Ⅰ中，SPI3和SRI1基本上处于相同的相位，并且几乎没有滞后时间。在2007-2013年期间，相位箭头小于30°，表明SPI3的变化领先于SRI1不到1个月。这可能是由于其他因素干扰了DPT（Han et al, 2019; S. Z. Huang et al, 2015）。尽管子区域Ⅱ的水文干旱没有显著的周期特征，但气象干旱和水文干旱之间仍然存在密切的关系，时间尺度为6-14个月（1981-1983年）、10-16个月（1985-1987年）、6-14个月（1988-1990年）、6-16个月（1991-1994年）、28-32个月（1994-1998年）、20-28个月（2007-2012年）和16-20个月（2014-2017年）（图9b）。在这些时期，相位角非常小，大多数相位角小于10°，最大相位角小于30°，表明在子区域Ⅱ中，SPI5和SRI1基本上处于相同的相位。类似地，还可以获得其他子区域气象和水文干旱的共振频率和相位移特征。所有五个子区域的显著相关时期列在表4中。与表3中周期特征相符的时期在表4中以粗体突出显示。总的来说，在相位变化方面，气象干旱与水文干旱之间存在显著的正相关，这表明气象干旱的周期特征主要影响了水文干旱的周期特征。此外，XWT分析的结果证明了PRB中气象干旱和水文干旱之间存在稳定的关系，并且本研究确定的DPT是合理的。因此，确定的DPT对未来区域干旱监测和预警将具有重要价值。

4. Discussion

4.1. Impacts of the Related Meteorological and Hydrological Factors on the DPT

Previous studies have shown that drought could be closely related to precipitation, runoff, evapotranspiration, and soil moisture (Fang et al, 2020; Han et al, 2019). To explore the impacts of these factors on the DPT, this study calculated the multiyear mean values of the related meteorological and hydrological factors corresponding to different DPTs. As mentioned above, the DPTs in the PRB were mainly between 2 and 6 months; therefore, Figure 10 only shows the relationships of these five DPTs (i.e., 2–6 months) with the annual mean precipitation (P), runoff (R), evapotranspiration (ET), and shallow soil moisture at different depths (i.e., SM1: 0–7 cm, SM2: 7–28 cm, SM3: 28–100 cm). Known from Figures 10a and 10b, the DPT was significantly related to the annual mean precipitation and runoff, and areas with higher annual mean precipitation and runoff had shorter DPTs. In contrast, the annual mean evapotranspiration and shallow soil moisture did not change much under different DPTs (Figures 10c–10f); however, there were still variation trends to some extent. For example, areas with greater annual mean evapotranspiration and shallow soil moisture had shorter DPTs. Specifically, such trends became less significant along with the increase of soil depth.

Figure 11 shows the changing rates of the annual mean precipitation, runoff, and evapotranspiration from 1981 to 2019 in different subregions. The results of shallow soil moisture at different depths were not presented because the changing rates were not large. Known from Figure 11a, the annual mean precipitation in the PRB presented a decreasing trend, and the changing rate in subregion Ⅲ was the largest, while that in subregion Ⅱ was the smallest. The pattern of the annual mean runoff was similar to that of the annual mean precipitation (Figure 11b). As mentioned above, the DPT in subregion Ⅲ was the shortest and that in subregion Ⅱ was the longest (Figure 6); therefore, it is inferred that the changes of precipitation and runoff would be important factors in affecting the DPT in the PRB. In contrast, the annual mean evapotranspira

tion showed an increasing trend, and the changing rate in subregion Ⅴ was the largest (Figures 11c). Tian and Yang (2017) showed that the recent temperature of the PRB presented an increasing trend under global warming. T. S. Li and Xia (2018) pointed out that the vegetation of the PRB also presented an increasing trend. From the perspective of hydrological cycle, due to climate warming and more vegetation, evapotranspiration of the PRB would be enhanced, leading to a downward trend of soil moisture. Along with the decrease of precipitation as input, when there was a long-term precipitation deficit in the PRB, the DPT from meteorological drought to hydrological drought would be shortened.

In addition, subregion Ⅴ had the closest changing rates of precipitation, runoff, and evapotranspiration to those in subregion Ⅲ; however, subregion Ⅴ had longer DPT than that in subregion Ⅲ (Figure 6). This is probably due to the impacts of reservoir operation in subregion Ⅴ (e.g., Xinfengjiang Reservoir) on the propagation from meteorological drought to hydrological drought. Previous studies (Wu et al, 2017; J. F.

Wu, Liu, et al, 2018) showed that, at a smaller time scale, reservoir operation could reduce the sensitivity between meteorological drought and hydrological drought, thereby slowing the drought propagation. Several other studies (Shi et al, 2018; J. F. Wu et al, 2019) proved that Xinfengjiang Reservoir changed the hydrological conditions in the East River basin (mainly in subregion Ⅴ). This could explain why some areas in subregion Ⅴ had greater DPTs than subregion Ⅲ. DPT, drought propagation time.

4.2. Teleconnection Analysis on Large-Scale Climatic Patterns

ET is an important parameter for the mass and energy exchange between atmospheric and terrestrial ecosystems. It also plays a key role in the flow of mass and heat in the global atmospheric system. Thus, it can be used to monitor the change of water and energy from the moisture to the atmosphere (S. Z. Huang et al, 2015). Therefore, large-scale climatic patterns may affect the DPT from meteorological to hydrological drought by affecting ET. Previous studies have shown that different types of drought are closely linked with large-scale climatic patterns such as ENSO and PDO (Han et al, 2019; S. Z. Huang et al, 2015). In this study, only the relationships of these two large-scale climatic patterns (i.e., ENSO and PDO) with ET in each subregion using XWT were analyzed, and the results are shown in Figures 12 and 13.

Figure 12 shows that the influences of ENSO on meteorological drought in each subregion were mainly at the small time scale of 2–4  months, and most areas showed significant negative correlations during 1995–2000. Specifically, in subregions Ⅰ and Ⅴ, significant positive relationships between ENSO and meteorological drought with 2–4 months were found during 1982–1989. In subregion Ⅰ, there was a positive correlation with 12 months at relatively large scale during 1990–1995. Figure 13 shows that the influences of PDO on meteorological drought in each subregion were mainly at the small time scale of 2–4 months during 1995–2002, and most areas showed significant positive correlations. Specifically, in subregions Ⅱ, Ⅲ, and Ⅳ, significant negative relationships between PDO and meteorological drought with 2–3 months 

were found during 2010–2015. In subregion Ⅳ, there was a significant positive correlation with 2–3 month during 2013–2017. Hence, ENSO and PDO both have strong relationships with ET.

The change of ET is closely related to precipitation and temperature (Soni & Syed, 2021). Previous studies have shown that meteorological drought situation in the PRB is increasingly severe, and drought events increase (Deng et al, 2018; C. Huang et al, 2017). ENSO increases the frequency and intensity of extreme precipitation events in the PRB, which makes the region prone to flood and drought events and warming temperatures (Duan et al, 2017). PDO is also an important factor affecting the characteristic of dryness and wetness in the PRB. Due to the phase change of PDO, the precipitation intensity in South China would significantly decrease (C. Huang et al, 2017), and the frequency of extreme precipitation would increase (Y. Zhao et al, 2014). Meanwhile, due to global warming, the temperature in the PRB has increased significantly (Zhang et al, 2018). The hydrological cycle is strengthened with the increase of water vapor feedback (Huntington, 2006). Hence, the changes of ENSO and PDO can greatly affect meteorological factors related to ET, that may be the reason why ENSO and PDO are closely related to ET in Figures 12 and 13. To this end, ENSO and PDO are regarded as important factors that can affect the DPT from meteorological to hydrological drought in the PRB.

5. Conclusions This study investigated the characteristics of propagation from meteorological drought to hydrological drought in the PRB based on ERA5-Land data during 1981–2019. Variations of precipitation and runoff in the whole PRB were analyzed by SPI and SRI, respectively, and five subregions were classified according to the dry and wet characteristics using K-means cluster analysis method. In addition, the DPTs from meteorological drought and hydrological drought and the link between meteorological drought and hydrological drought in each subregion were quantitatively investigated. The main findings of this study are as follows: First, precipitation and runoff in most parts of the PRB showed decreasing trends from 1981 to 2019. Specifically, the PRB has experienced severe meteorological and hydrological droughts since early 2000s, and hydrological drought was more serious than meteorological drought in each of the five subregions in the PRB. With the increase of time scale (i.e., from 1 to 12 months), the severity of meteorological drought and hydrological drought exacerbated.

Second, there was a strong response relationship between meteorological drought and hydrological drought, and the DPTs in the PRB were 2–6 months. The CWT analysis results indicated that most subregions (except for subregion Ⅱ) had similar periodic characteristics in both meteorological drought and hydrological drought. The XWT analysis results indicated that these two types of drought had similar resonance frequency and phase shift characteristics. Therefore, a stable relationship between meteorological drought and hydrological drought was found in the PRB.

Third, precipitation and runoff were the two important factors in influencing the DPT, and subregions with larger precipitation and runoff were more likely to have shorter DPTs. Moreover, ENSO and PDO are regarded as important factors that can affect the DPT from meteorological to hydrological drought in the PRB.

Overall, the outcomes of this study can improve our understanding of mechanism of drought propagation in a region and help in more effectively enhancing drought monitoring and prevention, especially in changing future climatic conditions. This, in turn, can lead to a more sustainable planning and management of our water resources.

以前的研究表明，干旱可能与降水、径流、蒸散发和土壤湿度密切相关（Fang等人，2020年；Han等人，2019年）。为了探讨这些因素对干旱传播时间（DPT）的影响，本研究计算了与不同DPT对应的相关气象和水文因素的多年均值。如上所述，珠江流域的DPT主要在2到6个月之间；因此，图10仅显示了这五个DPT（即2-6个月）与年均降水（P）、径流（R）、蒸散发（ET）和不同深度的浅层土壤湿度（即SM1：0-7厘米，SM2：7-28厘米，SM3：28-100厘米）的关系。从图10a和10b可以看出，DPT与年均降水和径流显著相关，年均降水和径流较多的地区DPT较短。相比之下，年均蒸散发和浅层土壤湿度在不同DPT下变化不大（图10c-10f）；但是，仍然存在一定程度的变化趋势。例如，年均蒸散发和浅层土壤湿度较大的地区DPT较短。具体来说，随着土壤深度的增加，这种趋势变得不那么显著。

图11显示了1981年至2019年不同子区域年均降水、径流和蒸散发的变化率。不呈现不同深度浅层土壤湿度的变化率，因为变化率不大。从图11a可以看出，珠江流域的年均降水呈下降趋势，其中Ⅲ号子区的变化率最大，而Ⅱ号子区最小。年均径流的模式与年均降水相似（图11b）。如上所述，Ⅲ号子区的DPT最短，而Ⅱ号子区最长（图6）；因此，推断降水和径流的变化将是影响珠江流域DPT的重要因素。相比之下，年均蒸散发呈增加趋势，其中Ⅴ号子区的变化率最大（图11c）。Tian和Yang（2017年）表明，珠江流域的近期温度呈上升趋势，受全球变暖的影响。T.S.Li和Xia（2018年）指出，珠江流域的植被也呈上升趋势。从水文循环的角度来看，由于气候变暖和更多的植被，珠江流域的蒸散发将增强，导致土壤湿度下降的趋势。随着降水输入的减少，在珠江流域出现长期降水赤字时，从气象干旱到水文干旱的传播时间将缩短。

此外，Ⅴ号子区的降水、径流和蒸散发的变化率与Ⅲ号子区最接近；然而，Ⅴ号子区的DPT比Ⅲ号子区长（图6）。这可能是由于Ⅴ号子区水库运行（例如，新丰江水库）对从气象干旱到水文干旱的传播的影响。以前的研究（Wu等人，2017年；J.F. Wu，Liu等人，2018年）表明，在较小的时间尺度上，水库运行可以减少气象干旱和水文干旱之间的敏感性，从而减缓干旱传播。其他几项研究（Shi等人，2018年；J.F. Wu等人，2019年）证明，新丰江水库改变了东江流域的水文条件（主要在Ⅴ号子区）。这可以解释为什么Ⅴ号子区的某些地区比Ⅲ号子区的DPT更长。 DPT，干旱传播时间。

4.2. 大尺度气候模式的遥相关分析

蒸散发是大气和陆地生态系统之间质量和能量交换的重要参数。它在全球大气系统中的质量和热量流动中起着关键作用。因此，它可以用于监测从土壤到大气中水和能量的变化（S. Z. Huang等人，2015年）。因此，大尺度气候模式可能通过影响蒸散发来影响从气象干旱到水文干旱的DPT。以往的研究表明，不同类型的干旱与ENSO和PDO等大尺度气候模式密切相关（Han等人，2019年；S. Z. Huang等人，2015年）。本研究仅分析了这两种大尺度气候模式（即ENSO和PDO）与每个子区域的蒸散发之间的关系，使用XWT，并且结果显示在图12和图13中。

图12显示，ENSO对每个子区域的气象干旱的影响主要是在2-4个月的小时间尺度上，大多数地区在1995

- 2000年期间显示出显著的负相关性。具体而言，在Ⅰ和Ⅴ号子区，在1982-1989年期间，ENSO和2-4个月的气象干旱之间显示出显著正相关。在Ⅰ号子区，1990-1995年期间存在12个月的相对较大尺度的正相关性。

图13显示，PDO对每个子区域的气象干旱的影响主要集中在1995-2002年的2-4个月小时间尺度上，大多数地区显示出显著的正相关性。具体而言，在Ⅱ、Ⅲ和Ⅳ号子区，在2010-2015年期间，PDO和2-3个月的气象干旱之间显示出显著的负相关性。在Ⅳ号子区，2013-2017年期间存在与2-3个月相关的显著正相关性。因此，ENSO和PDO都与蒸散发有着密切的关系。

蒸散发的变化与降水和温度密切相关（Soni & Syed，2021年）。以往的研究表明，珠江流域的气象干旱情况日益严重，干旱事件增加（Deng等人，2018年；C. Huang等人，2017年）。ENSO增加了珠江流域极端降水事件的频率和强度，使该地区更容易发生洪涝和干旱事件，并导致气温升高（Duan等人，2017年）。PDO也是影响珠江流域干湿特征的重要因素。由于PDO相位变化，华南的降水强度将显著减少（C. Huang等人，2017年），极端降水频率将增加（Y. Zhao等人，2014年）。同时，由于全球变暖，珠江流域的温度显著升高（Zhang等人，2018年）。随着水汽反馈的增强，水文循环得到加强（Huntington，2006年）。因此，ENSO和PDO的变化可以极大地影响与蒸散发相关的气象因素，这可能是为什么图12和图13中ENSO和PDO与蒸散发密切相关的原因。因此，ENSO和PDO被认为是影响珠江流域从气象干旱到水文干旱的DPT的重要因素。

5. 结论

本研究基于1981年至2019年的ERA5-Land数据，调查了珠江流域从气象干旱到水文干旱传播特征。通过SPI和SRI分析了整个珠江流域的降水和径流变化，并利用K-means聚类分析方法将其划分为五个子区域，根据干湿特征进行分类。此外，定量研究了每个子区域的气象干旱和水文干旱之间的联系以及从气象干旱到水文干旱的DPT。本研究的主要发现如下：

首先，从1981年到2019年，珠江流域大部分地区的降水和径流呈下降趋势。具体而言，珠江流域自2000年初以来经历了严重的气象和水文干旱，每个子区域的水文干旱都比气象干旱更严重。随着时间尺度的增加（即从1到12个月），气象干旱和水文干旱的严重程度加剧。

第二，气象干旱与水文干旱之间存在着强烈的响应关系，珠江流域的DPT为2-6个月。CWT分析结果表明，大多数子区域（除了Ⅱ号子区）在气象干旱和水文干旱中具有相似的周期特征。XWT分析结果表明，这两种类型的干旱具有相似的共振频率和相位移特征。因此，在珠江流域发现了气象干旱与水文干旱之间的稳定关系。

第三，降水和径流是影响DPT的两个重要因素，降水和径流较多的子区域更容易具有较短的DPT。此外，ENSO和PDO被认为是影响珠江流域从气象干旱到水文干旱的DPT的重要因素。

总的来说，本研究的结果可以提高我们对某一地区干旱传播机制的理解，并有助于更有效地增强干旱监测和预防，尤其是在未来气候条件发生变化的情况下。这将进一步推动我们更可持续地规划和管理水资源。