Identifying Optimal Reanalysis and Remote Sensing Data Combinations for Multi-Scale SPEI-Based Drought Assessment in Zhejiang Province, China

Abstract

Accurate drought assessment is crucial for effective regional water resource management. While reanalysis and remote sensing products enable high-resolution drought assessment, their regional application requires rigorous local validation. This study evaluates nine data combinations, pairing three precipitation products with three evapotranspiration products, to identify the optimal combination for robust SPEI estimation and subsequently to investigate the spatiotemporal variations in drought conditions during 1980–2020 in Zhejiang Province, China. The results indicate that the choice of precipitation product is the dominant factor influencing SPEI accuracy, with the combination of CMFD V2.0 precipitation and GLEAM v4.2a evapotranspiration identified as the most reliable for SPEI estimation across multiple timescales (SPEI1/3/6/12). The long-term trend analysis of the SPEI derived from this optimal data combination reveals significant spatiotemporal heterogeneity: temporally, a pronounced “wetter winters, drier springs” seasonal pattern emerges, posing a substantial threat to agricultural water security; spatially, a distinct divergence shows central/northeastern areas wetting while southern/southeastern regions experience a significant drying trend, particularly for long-term hydrological drought (SPEI12). Additionally, the prevalence of light droughts across the province suggests a sustained baseline of water stress. Attribution analysis further demonstrates that precipitation is the dominant driver of droughts across all timescales. This study contributes both a validated, high-resolution data foundation for regional drought assessment and a scientific basis for targeted drought adaptation strategies.

1. Introduction

Drought is one of the most destructive climate-related hazards globally, with profound impacts on water resources, ecosystems, agriculture, and human society [1,2]. Under global climate change, drought events have become increasingly frequent and severe in many regions worldwide, exhibiting greater spatial and temporal heterogeneity, and are projected to intensify further in the coming decades [3]. These facts underscore the urgent need to conduct accurate drought assessments to support effective risk management and adaptation strategies.

To effectively quantify and assess drought, numerous indices have been developed. The early Palmer Drought Severity Index (PDSI) is limited by its computational complexity and poor spatial comparability [4]. Subsequently, the Standardized Precipitation Index (SPI), based solely on precipitation, has become widely used due to its computational simplicity and multi-scalar property [5]. However, in the context of global warming, neglecting the intensified evapotranspiration driven by rising temperatures can result in underestimation of drought severity. To overcome this deficiency, Vicente-Serrano et al. [6] proposed the Standardized Precipitation Evapotranspiration Index (SPEI), whose core advantage lies in its simultaneous capture of changes in both the water supply (precipitation) and water demand (potential evapotranspiration). Numerous studies have since confirmed the superiority of the SPEI in various climatic zones [7,8,9,10]. Moreover, its multi-scalar feature enables the diagnosis of drought propagation, i.e., the evolution from short-term meteorological drought (e.g., SPEI1) to agricultural drought characterized by soil moisture deficits (e.g., SPEI6) and eventually to hydrological drought affecting surface and subsurface water supply (e.g., SPEI12) [11].

Accurate calculation of indices such as the SPEI is highly dependent on high-quality meteorological input data. Traditional approaches rely heavily on ground-based meteorological observations; however, ground stations are often sparsely and unevenly distributed, particularly in areas with complex terrain or in developing countries [12,13]. Although drought indices derived from station observations can effectively reflect drought conditions in the vicinity of the stations, their limited spatial coverage constrains the ability to capture drought patterns across broader regions, especially in regions with high spatial variability or sparse station networks. This may result in an incomplete or even misleading understanding of drought dynamics. Gridded drought information generated from sparse station data through spatial interpolation tends to smooth out critical local drought details and introduce significant uncertainty [14].

To overcome these limitations, meteorological datasets derived from reanalysis and remote sensing products have emerged as valuable alternatives or complements to ground-based observations for drought monitoring and assessment [15,16,17,18]. These products offer high-resolution, spatiotemporally continuous meteorological variables such as precipitation and evapotranspiration, which are essential inputs for drought indices like the SPEI. For instance, products from the Global Precipitation Measurement (GPM) mission, such as Integrated Multi-satellitE Retrievals for GPM (IMERG) [19], and the Tropical Rainfall Measuring Mission (TRMM) [20] provide high-quality precipitation estimates. Reanalysis products, such as the Global Land Data Assimilation System (GLDAS) [21], provide key land surface variables including evapotranspiration. The application of these products in drought studies has become increasingly widespread, with numerous studies confirming their overall effectiveness [16,22,23,24,25,26,27,28]. For example, Alahacoon and Edirisinghe [28] demonstrated that drought indices derived from remote sensing datasets can effectively capture spatial details and have become advanced tools for drought assessment at both global and regional scales. Kim et al. [16] utilized the high-resolution European Reanalysis Atmosphere 5th Generation (ERA5) dataset to derive meteorological drought indices in South Korea and validated their accuracy against observational data for major drought cases.

Nevertheless, despite their proven utility at broader scales, the direct application of reanalysis and remote sensing products to regional drought assessment remains subject to considerable uncertainty. These products rely on different sensor physics, retrieval algorithms, model processes, or data assimilation techniques, leading to significant variability in their performance across different geographical regions and climatic conditions [29,30,31,32,33]. Therefore, region-specific evaluation and validation of the applicability of various reanalysis and remote sensing products for drought assessment are essential, serving as a critical foundation for ensuring the robustness of subsequent analyses of historical drought dynamics at the regional scale [34,35].

However, most previous evaluation studies share a key limitation: they typically assess only a single variable relevant to drought estimation, for instance, comparing multiple precipitation products while using a fixed potential evapotranspiration dataset, thereby neglecting the compound effect of uncertainties from both core inputs in the SPEI calculation [36]. Such “one-sided” assessments may lead to a sub-optimal selection of the best data combination. For example, an optimal precipitation product combined with a sub-optimal evapotranspiration product may perform worse than a sub-optimal precipitation product paired with a superior evapotranspiration product. Thus, a critical scientific question remains insufficiently explored: which combination of precipitation and evapotranspiration datasets can most accurately reproduce observed droughts in a specific region?

Addressing this question is of both academic and practical importance in regions like Zhejiang Province. As one of China’s most densely populated and economically dynamic regions, Zhejiang faces significant development-driven pressure on its water resources. Meanwhile, its typical monsoon climate induces high intra- and inter-annual precipitation variability, which places the regional water system in a fragile balance between flood and drought [37]. While considerable attention has been given to flood assessments, drought has received comparatively limited scholarly focus in this region. Given the escalating frequency and severity of droughts in recent years and the scarcity of relevant studies, it is imperative to conduct accurate drought assessments in Zhejiang Province, supported by a rigorously validated and reliable data foundation.

In light of the aforementioned research context, this study is designed to provide a more robust and high-resolution data foundation for drought analysis in Zhejiang Province through a systematic evaluation. An assessment matrix comprising nine data combinations, formed by pairing three precipitation and three evapotranspiration datasets derived from various reanalysis and remote sensing products, was constructed. The specific objectives are (1) to comprehensively evaluate the performance of the nine data combinations in capturing drought conditions across multiple timescales (SPEI1, SPEI3, SPEI6, and SPEI12) and to identify the optimal data combination, and (2) to utilize this optimal combination to construct long-term, high-resolution SPEI datasets to investigate the spatiotemporal variations in drought in Zhejiang. By advancing a more rigorous and multidimensional evaluation of various data combinations, this study seeks to improve the robustness of regional drought assessment based on reanalysis and remote sensing products and to offer a solid scientific basis for regional water resource planning and risk management.

2. Materials and Methods

2.1. Study Area

Zhejiang Province is situated along the southeastern coast of China (27°02′–31°11′ N, 118°01′–123°10′ E), covering a total land area of approximately 105,500 km² (Figure 1). The province exhibits a highly varied topography, transitioning from mountainous highlands in the southwest, through hilly terrain in its central regions, to expansive alluvial plains in the northeast. This geographic complexity exerts significant influence on regional climatic conditions and hydrological processes.

Zhejiang possesses an extensive river network, comprising eight major river basins distributed from south to north: Aojiang, Feiyunjiang, Oujiang, Jiaojiang, Yongjiang, Qiantangjiang, Tiaoxi, and the Grand Canal. Overall, the province benefits from relatively abundant water resources. Climatically, Zhejiang is governed by a subtropical monsoon system, with its climate predominantly categorized as humid subtropical (Cfa) according to the Köppen–Geiger classification [38,39], featuring hot, humid summers and cool, dry winters. The annual mean temperature ranges from 15 °C to 18 °C, while the annual mean precipitation varies between 1100 mm and 2000 mm [40,41]. Despite the overall abundance of precipitation, its intra-annual distribution is markedly uneven, with precipitation heavily concentrated during the plum rain season (May–June) and the typhoon season (July–September) [37]. This pronounced seasonality makes the region particularly prone to seasonal drought events.

In recent decades, the combined effects of climate change and rapid urbanization have intensified the province’s exposure to drought risks. As one of the most economically developed and densely populated provinces in China, Zhejiang exhibits a high dependency on water resources. Droughts pose significant threats to water resource management, ecosystem stability, agricultural production, and socioeconomic development. Therefore, enhancing drought monitoring and assessment capabilities in Zhejiang is of critical practical significance.

2.2. Data

2.2.1. Ground-Based Meteorological Observations

Daily meteorological data from 12 ground-based stations distributed throughout Zhejiang Province were employed in this study to serve as observational references for drought assessment. These data were obtained from the National Climate Center, China Meteorological Administration (http://data.cma.cn). As illustrated in Figure 1, the stations utilized are well dispersed across diverse climatic and topographic zones of Zhejiang Province, ensuring representative capture of the province’s hydro-climatic variability. The collected meteorological variables include the daily precipitation, maximum and minimum air temperatures, relative humidity, wind speed, and solar radiation, covering the period 2003–2018. Potential evapotranspiration (PET), a critical factor for calculating the SPEI, was estimated using the Food and Agriculture Organization of the United Nations (FAO) Penman–Monteith (PM) method [42].

2.2.2. Reanalysis and Remote Sensing Precipitation Datasets

Three reanalysis and remote sensing precipitation products were employed in this study. These products were chosen for their widespread application in hydrometeorological research and methodological heterogeneity, allowing for a comprehensive intercomparison of their performance in SPEI estimation. The following products were employed:

• CMFD V2.0 (China Meteorological Forcing Dataset Version 2.0):
  CMFD V2.0 [43] is a hybrid product jointly developed by the Institute of Tibetan Plateau Research, Chinese Academy of Sciences, and the Department of Earth System Science, Tsinghua University. It integrates ground-based observations with ERA5 reanalysis data and satellite data, leveraging artificial intelligence technology for radiation and precipitation products, to provide a high-precision, high-resolution, and long-time-series meteorological forcing dataset for China. CMFD V2.0 offers seven key meteorological variables, including the precipitation rate, at a 3 h temporal resolution and a 0.1° spatial resolution from 1951 to 2020. The dataset can be downloaded from the National Tibetan Plateau Data Center (https://cstr.cn/18406.11.Atmos.tpdc.302088) (accessed on 1 June 2025).

• IMERG V07B (Integrated Multi-satellitE Retrievals for GPM Version 07B):
  As the latest precipitation product under NASA’s Global Precipitation Measurement (GPM) mission, IMERG V07B [19] integrates passive microwave and infrared observations from multiple satellite sensors. The Final Run version incorporates global gauge data during post-processing for bias correction, substantially enhancing its accuracy and making it well suited for hydrometeorological applications. IMERG V07B Final Run provides data at a spatial resolution of 0.1° and multiple temporal resolutions including monthly, daily, and 30 min, spanning from 1998 to the present (with an approximate 3.5-month latency). It is accessible via NASA’s GPM Data Portal (https://gpm.nasa.gov/data/imerg) (accessed on 30 May 2025).

• TMPA 3B42V7 (TRMM Multi-satellite Precipitation Analysis Version 7):
  Developed collaboratively by NASA and the Japan Aerospace Exploration Agency (JAXA), TMPA 3B42V7 [20,44] is a multi-source satellite precipitation product integrating microwave and infrared observations from the TRMM (Tropical Rainfall Measuring Mission) satellite with gauge-based post-processing. Designed for use in tropical and subtropical regions, it delivers 3-hourly/daily/monthly precipitation estimates at a spatial resolution of 0.25°, covering the period from January 1998 to December 2019. TMPA 3B42V7 has been widely adopted in hydrometeorological studies, and it is available through NASA’s GES DISC (https://disc.gsfc.nasa.gov/datasets/TRMM_3B42_Daily_7/summary) (accessed on 30 May 2025).

2.2.3. Reanalysis and Remote Sensing Evapotranspiration Datasets

Similarly, three evapotranspiration (ET) datasets, derived from reanalysis and remote sensing products, were selected for SPEI estimation:

• GLDAS-2.2 (Global Land Data Assimilation System Version 2.2):
  Developed by NASA’s Goddard Space Flight Center, GLDAS-2.2 [21,45] simulates global terrestrial water and energy fluxes by assimilating satellite- and ground-based observations into land surface models. This product provides 24 land surface variables, including evapotranspiration, at a daily temporal resolution and a spatial resolution of 0.25°, spanning from February 2003 to the present. GLDAS-2.2 data are available via NASA’s GES DISC (https://disc.gsfc.nasa.gov/datasets/GLDAS_CLSM025_DA1_D_2.2/summary) (accessed on 30 May 2025).

• GLEAM v4.2a (Global Land Evaporation Amsterdam Model version 4.2a):
  GLEAM v4.2a [46] is a high-resolution global land evaporation and soil moisture dataset led by the Hydro-Climate Extremes Lab (H-CEL) at Ghent University. Utilizing a hybrid modeling framework, this product integrates physical process modeling with machine learning, incorporating eddy covariance observations and a range of satellite and reanalysis inputs. GLEAM v4.2a provides 12 key variables, including potential and actual evapotranspiration, at a daily temporal resolution and 0.1° spatial resolution, covering 1980–2023. The datasets can be accessed from the GLEAM portal (https://www.gleam.eu/) (accessed on 1 June 2025).

• PML-V2 (Penman–Monteith–Leuning Version 2):
  Developed by the Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, PML-V2 [47,48] is based on an enhanced Penman–Monteith–Leuning model that integrates stomatal conductance with photosynthetic processes. This product provides coordinated estimates of evapotranspiration and gross primary productivity. It features an 8-day temporal resolution and a high spatial resolution of 0.05° and spans from July 2002 to August 2019. PML-V2 data are available through the National Tibetan Plateau Data Center (https://cstr.cn/18406.11.Geogra.tpdc.270251) (accessed on 30 May 2025).
  To create a consistent framework for comparison, all reanalysis and remote sensing datasets were resampled to 0.1° spatial resolution and harmonized to a daily temporal resolution. The three precipitation datasets and three evapotranspiration datasets were then cross-combined to form a complete 3 × 3 matrix, yielding nine data combinations for SPEI calculation and performance evaluation (

  Table 1. Matrix of the nine data combinations for SPEI estimation. Each cell represents a unique data combination formed by pairing a precipitation dataset (row) with an evapotranspiration dataset (column).

  	Evapotranspiration Products	GLDAS-2.2 (EI)	GLEAM v4.2a (EII)	PML-V2 (EIII)
  Precipitation Products
  CMFD V2.0 (PI)		PIEI	PIEII	PIEIII
  IMERG V07B Final Run (PII)		PIIEI	PIIEII	PIIEIII
  TMPA 3B42V7 (PIII)		PIIIEI	PIIIEII	PIIIEIII

  ). This full matrix design minimizes subjective bias and enables a comprehensive assessment of the influence of different data pairings on SPEI estimation. It is worth noting that GLEAM v4.2a provides both actual evapotranspiration (ET) and potential evapotranspiration (PET), whereas GLDAS-2.2 and PML-V2 only provide ET. As SPEI computation requires PET, the evaporative stress index (S = ET/PET) derived from GLEAM v4.2a was utilized to convert the ET values from GLDAS-2.2 and PML-V2 into PET, thus ensuring their suitability for SPEI computations.

2.3. Methods

To provide a clear overview of the methodological sequence, Figure 2 presents a framework diagram that summarizes the overall workflow of this study, including data preparation, performance evaluation of different reanalysis and remote sensing data combinations for SPEI estimation, and subsequent analyses of spatiotemporal drought variations in Zhejiang Province.

2.3.1. Standardized Precipitation Evapotranspiration Index (SPEI)

The Standardized Precipitation Evapotranspiration Index (SPEI), developed by Vicente-Serrano et al. [6], is a multi-scalar drought index that accounts for the combined effects of precipitation and potential evapotranspiration on the surface water balance. The calculation of the SPEI begins by determining the monthly difference, $D_i$, between precipitation ($P_i$) and potential evapotranspiration ($PE{T}_i$), where $D_i=P_i-PE{T}_i$. This time series is then aggregated at various timescales of k months (where k is 1, 3, 6, and 12 in this study) to reflect the cumulative water balance over different durations. The aggregated series, $D^k$, is subsequently fitted to a probability distribution to account for the range of possible values. The three-parameter log-logistic distribution is employed owing to its capability to accommodate negative values, which makes it well suited for water balance data. The cumulative distribution function (CDF) is defined as

$$F\left(x\right)={\left[1+{\left({\displaystyle \frac{\alpha }{x-\gamma }}\right)}^{\beta }\right]}^{-1}$$

(1)

where $\alpha$, $\beta$, and $\gamma$ are the scale, shape, and origin parameters, respectively, estimated from the $D^k$ data series using the L-moment method.

The probability derived from the CDF, $\mathrm{P}=1-F\left(x\right)$, is subsequently transformed into a standard normal variable to obtain the final SPEI value. This standardization allows for robust comparisons of drought conditions across different locations and periods. The transformation is expressed as follows:

$$SPEI=W-{\displaystyle \frac{C_0+C_{1}W+C_2{W}^2}{1+d_{1}W+d_2{W}^2+d_3{W}^3}}$$

(2)

where $W=\sqrt{-2\ln P}$ for P ≤ 0.5. For P > 0.5, P is replaced by 1 − P. The constants used in the approximation are $C_0$= 2.515517, $C_1$= 0.802853, $C_2$= 0.010328, $d_1$= 1.432788, $d_2$= 0.189269, and $d_3$= 0.001308.

The derived SPEI values are used to identify drought severity, as classified in

Table 2. Classification of drought severity based on SPEI values.

Drought Severity	$\mathbf{\text{SPEI}}$Values
No drought	$-0.5<\mathrm{SPEI}$
Light drought	$-1.0<\mathrm{SPEI}\le -0.5$
Moderate drought	$-1.5<\mathrm{SPEI}\le 1.0$
Severe drought	$-2.0<\mathrm{SPEI}<-1.5$
Extreme drought	$\mathrm{SPEI}\le -2.0$

[11]. Negative SPEI values indicate varying levels of dryness, while positive values denote wet conditions.

2.3.2. Performance Evaluation

To quantitatively evaluate the performance of the reanalysis- and remote sensing-based SPEI estimations (hereafter RSSPEI) against the observation-based benchmarks (ObsSPEI), a grid-to-point comparison was employed. Specifically, for each of the 12 meteorological stations, the 0.1° grid cell from the reanalysis and remote sensing datasets nearest to the station location was selected. SPEI values extracted from these grid cells were then directly compared to the station-observation-based SPEI, thereby avoiding spatial interpolation of the sparse station data, which could introduce potential biases.

Performance evaluation was conducted at multiple timescales (SPEI1, SPEI3, SPEI6, and SPEI12) over the period 2003–2018. The agreement between the RSSPEI and ObsSPEI was quantified using three commonly adopted statistical indicators: the Nash–Sutcliffe Efficiency (NSE), the Root Mean Square Error (RMSE), and the Pearson correlation coefficient (R). These metrics are calculated as follows:

$$NSE=1-{\displaystyle \frac{{\displaystyle {\sum }_{i=1}^n{(O_i-S_i)}^2}}{{\displaystyle {\sum }_{i=1}^n{(O_i-\stackrel{-}{O})}^2}}}$$

(3)

$$\mathrm{RMSE}=\sqrt{{\displaystyle \frac{1}{n}}({\displaystyle {\sum }_{i=1}^n{(S_i-O_i)}^2})}$$

(4)

$$\mathrm{R}={\displaystyle \frac{{\displaystyle {\sum }_{i=1}^n{(S_i-\stackrel{-}{S})}^2}{(O_i-\stackrel{-}{O})}^2}{\sqrt{{\displaystyle {\sum }_{i=1}^n{(S_i-\stackrel{-}{S})}^2}}\sqrt{{\displaystyle {\sum }_{i=1}^n{({\mathrm{O}}_i-\stackrel{-}{O})}^2}}}}$$

(5)

In these equations, n is the sample size of the data; $S_i$ and $O_i$ are the $i$ th values of the reanalysis- and remote sensing-based SPEI and meteorological-observation-based SPEI, respectively. $\stackrel{-}{S}$ and $\stackrel{-}{O}$ are their respective mean values. The optimal values of the NSE, RMSE, and R are 1.0, 0, and 1.0, respectively.

2.3.3. Mann–Kendall Trend Analysis

The Mann–Kendall (MK) statistical test is a widely applied nonparametric method for detecting monotonic trends within time series data [49,50]. A principal advantage of this rank-based test is that it does not require a normal data distribution and exhibits low sensitivity to outliers, making it highly suitable for analyzing hydrometeorological data. In this study, the MK test was applied to detect temporal trends in the SPEI across multiple timescales. The MK statistic $S$ is computed as follows:

$$S={\sum }_{i=1}^{n-1}{\sum }_{j=i+1}^n\mathrm{sgn} \left(x_j-x_i\right)$$

(6)

$$\mathrm{sgn} \left(x_j-x_i\right)=\left\{\begin{array}{c}+1, if x_j-x_i>0\\ 0, if x_j-x_i=0\\ -1, if x_j-x_i<0\end{array}\right.$$

(7)

where $x_j$ and $x_i$ are the values of the time series, and n is the length of the series.

For large n, the statistic S is approximately normally distributed with an expected value of zero. The variance of S, accounting for tied ranks, is given by

$$\mathrm{Var} \left(S\right)={\displaystyle \frac{n\left(n-1\right)\left(2n+5\right)-{\sum }_{k=1}^m{t}_k\left(t_k-1\right)\left(2t_k+5\right)}{18}}$$

(8)

where m is the total number of tied groups, and $t_k$ denotes the number of ties in the k-th group.

The standardized test statistic Z is given by

$$Z=\left\{\begin{array}{c}{\displaystyle \frac{S-1}{\sqrt{\mathrm{Var} \left(S\right)}}}, if S>0\\ 0, if S=0\\ {\displaystyle \frac{S+1}{\sqrt{\mathrm{Var} \left(S\right)}}}, if S<0\end{array}\right.$$

(9)

A positive Z value indicates an increasing trend, while a negative Z value indicates a decreasing trend. The statistical significance of the trend is assessed by comparing the absolute value of Z to the standard normal distribution at the chosen significance level. At the 5% level, trends are considered significant when $\left|Z\right|$> 1.96. Specifically, Z > 1.96 denotes a significant increasing trend in the SPEI (i.e., wetting), while Z <−1.96 suggests a significant decreasing trend (i.e., drying). The bigger $\left|Z\right|$ is, the stronger the detected trend is.

2.3.4. Contribution Analysis of Climatic Drivers

To quantify the relative importance of precipitation and potential evapotranspiration (PET) in driving the SPEI, a multiple linear regression (MLR) model was established for each 0.1° grid cell across the study area. For each SPEI timescale (1, 3, 6, and 12 months), the time series of the SPEI was treated as the dependent variable, with the corresponding time series of precipitation and PET serving as the independent variables.

Prior to regression, all time series variables were standardized to a range of 0 to 1 to eliminate the influence of different units and scales. Independent variables were retained in the MLR model only if they met a statistical significance level of p < 0.05. The standardized MLR model is expressed as

$$y_s=A{x}_{1s}+B{x}_{2s}+ϵ$$

(10)

where $y_s$ is the standardized SPEI, $x_{1s}$ and $x_{1s}$ are the standardized precipitation and PET, respectively, A and B are the standardized regression coefficients, and ϵ is the residual error.

Based on the regression coefficients, the relative contribution rate, which represents the relative importance of each driver within the model, was calculated as follows:

$${\eta }_1={\displaystyle \frac{\left|A\right|}{\left|A\right|+\left|B\right|}}$$

(11)

$${\eta }_2={\displaystyle \frac{\left|B\right|}{\left|A\right|+\left|B\right|}}$$

(12)

where ${\eta }_1$ and ${\eta }_2$ are the relative contribution rates of P and PET to the SPEI variation, respectively.

3. Results

3.1. Performance Evaluation of Reanalysis and Remote Sensing Data Combinations for SPEI Estimation

3.1.1. Evaluation at the Station Level

Figure 3 presents the performance of the RSSPEI from nine data combinations (as detailed in Table 1) in replicating the ObsSPEI across the 12 stations at four SPEI timescales (SPEI1, SPEI3, SPEI6, and SPEI12), evaluated by the NSE, RMSE, and R. The results show that combinations incorporating the CMFD V2.0 precipitation dataset (i.e., PIEI, PIEII, PIEIII) consistently outperform others, characterized by higher NSE and R values and a lower RMSE across all SPEI timescales. The IMERG V07B-based combinations (i.e., PIIEI, PIIEII, PIIEIII) follow, while combinations utilizing TMPA 3B42V7 precipitation (i.e., PIIIEI, PIIIEII, PIIIEIII) demonstrate comparatively weaker performance. In addition to their superior accuracy, the CMFD V2.0-based combinations also exhibit greater inter-station stability in SPEI estimation, whereas combinations driven by IMERG V07B and particularly TMPA 3B42V7 display higher variability among stations.

Figure 3 also reveals that the effectiveness of various reanalysis and remote sensing data combinations in estimating the SPEI tends to improve with increasing SPEI timescales. For most data combinations, the estimation for SPEI12 is the most accurate, with the highest values for the NSE and R and the lowest values for the RMSE. This is expected, as longer timescales smooth out short-term data noise and focus on more persistent and stable climatic signals.

Among the nine combinations, PIEII (CMFD V2.0 and GLEAM v4.2a) consistently yields superior performance across all SPEI timescales. Furthermore, the interquartile ranges (i.e., box sizes in Figure 3) for PIEII are relatively compact, reflecting strong robustness and stability across the diverse climatic and geographic conditions of the 12 stations in the study area.

3.1.2. Overall Agreement Between Remote Sensing-Based and Observation-Based SPEI

To further validate the regional-scale consistency between the RSSPEI and ObsSPEI, data from all 12 stations for the period 2003–2018 were aggregated, and scatter plots were constructed to compare RSSPEI and ObsSPEI values under each of the nine reanalysis and remote sensing data combinations. Figure 4, Figure 5, Figure 6 and Figure 7 present these comparisons for SPEI1, SPEI3, SPEI6, and SPEI12, respectively, with each subplot representing a specific data combination.

In Figure 4, Figure 5, Figure 6 and Figure 7, the degree of clustering around the 1:1 reference line offers a visual assessment of estimation accuracy, complemented by the displayed NSE, RMSE, and R values. The results align well with those from the station-level analysis: combinations based on CMFD V2.0 precipitation exhibit tighter clustering and higher statistical agreement with observed values, followed by IMERG V07B-based combinations, while combinations based on TMPA 3B42V7 display greater dispersion, indicating reduced accuracy and reliability. Notably, for extreme drought conditions (ObsSPEI < −2.0), the reanalysis- and remote sensing-based estimations for SPEI1 and SPEI3 exhibit conspicuous errors. However, these discrepancies markedly diminish for SPEI6 and SPEI12 estimations, likely due to the smoothing effect of longer time integration.

Quantitatively, the PIEII combination (CMFD V2.0 and GLEAM v4.2a) again stands out as the optimal choice for SPEI estimation. Even in its comparatively weakest performance for SPEI3 (Figure 5b), it still maintains excellent efficacy, with an NSE of 0.84, an RMSE of 0.38, and an R of 0.92. For SPEI12 estimation (Figure 7b), it achieves the highest accuracy across all 36 scenarios (four SPEI timescales × nine data combinations), with an NSE of 0.90, RMSE of 0.30, and R value of 0.95. In contrast, the PIIIEIII combination (TMPA 3B42V7 and PML-V2), for example, shows a much weaker performance for SPEI12 (Figure 7i), with an NSE of 0.77, an RMSE of 0.46, and an R of 0.89.

In summary, both the station-level and region-wide assessments consistently and robustly demonstrate the superiority of the PIEII data combination (CMFD V2.0 and GLEAM v4.2a) in reproducing the observed SPEI across multiple timescales in Zhejiang Province. Therefore, it was selected and utilized for the subsequent analysis of the spatial–temporal changes in drought in this region.

3.2. Spatiotemporal Variations in Drought in Zhejiang Province

3.2.1. Spatiotemporal Trends of Multi-Scale SPEI and Their Climatic Drivers

Using the validated high-performance PIEII data combination, the temporal coverage was extended to 1980–2020 (intersection of CMFD V2.0 and GLEAM v4.2a time series) to explore the long-term spatial–temporal changes in drought conditions in Zhejiang Province.

Figure 8 presents the province-averaged SPEI time series at four timescales (SPEI1, SPEI3, SPEI6, and SPEI12) from 1980 to 2020. The results reveal pronounced inter-annual variability. At shorter timescales (SPEI1 and SPEI3), particularly SPEI1, frequent fluctuations between wet and dry conditions are evident, reflecting the influence of short-term meteorological anomalies. At medium- to long-term scales (SPEI6 and SPEI12), multi-year fluctuations become more apparent. A distinct and severe multi-year drought period is observed from approximately 2003 to 2009, especially in the SPEI12 series. This prolonged drought is one of the most significant dry periods recorded in recent decades. In contrast, during 2015–2020, Zhejiang Province is predominantly characterized by wet conditions, despite experiencing alternating dry and wet periods.

Complementing the time series analysis, the MK trend analysis was applied to the SPEI time series across four timescales. Figure 9 presents the spatial distribution of MK Z-scores for SPEI1, SPEI3, SPEI6, and SPEI12. The results reveal pronounced spatial heterogeneity among different SPEI timescales. For the one-month timescale SPEI (SPEI1), most regions exhibit a drying trend (negative Z-scores), with dispersed wetting trends in central and northeastern Zhejiang. However, these trends are generally not statistically significant (i.e., $\left|Z\right|$< 1.96). The spatial distribution of the SPEI3 trend is largely consistent with that of SPEI1 but with intensified signals of change, especially in the southern and southeastern regions, where Z-scores fall below −1.96, indicating significant drying trends. For the six-month timescale SPEI (SPEI 6), a more organized and distinct spatial pattern becomes evident. The areas exhibiting a wetting trend (positive Z-scores) expand, though they remain concentrated in central and northeastern Zhejiang and are not always statistically significant. Conversely, intensified drying persists in the southern and southeastern regions. For the 12-month timescale SPEI (SPEI12), the most coherent and significant trends are observed. A pronounced wetting trend dominates the central and northeastern regions, with a substantial proportion of these areas exhibiting significant wetting signals (Z > 1.96). In contrast, the northwestern, western, southern, and southeastern regions exhibit scattered drying trends, with the southeastern region showing particularly strong and statistically significant drying.

To further capture intra-annual trends in the SPEI across multiple timescales, the MK test was applied on a monthly basis. The spatial distributions of the MK Z-scores for each month are presented in Figure 10, Figure 11, Figure 12 and Figure 13 for SPEI1, SPEI3, SPEI6, and SPEI12, respectively.

As shown in Figure 10, there is a pronounced seasonal asymmetry in SPEI1 trends over the past four decades. A wetting trend is observed during the winter months of December and January, as well as in late autumn (November) and early summer (June), with December showing the most evident wetting trend, although it is not statistically significant over the majority of the province. In contrast, a drying tendency occurs in the spring months of March and April and the autumn month of October. February is marked by a spatial dichotomy, with slight wetting in the north and drying in the south. Other months (May, July, August, and September) exhibit more complex and spatially heterogeneous trends, which are generally insignificant. Overall, the SPEI1 trends indicate wetter winters and drier springs in Zhejiang Province over the past four decades.

For SPEI3 (Figure 11), the seasonal patterns become more defined. The wetting trend during winter (December–February) becomes more evident. Concurrently, the drying trend during the spring (March–May) also intensifies across the province. From June to November, the SPEI3 trends are mixed, with wetting dominating summer and drying prevailing in autumn.

The trends for SPEI6 (Figure 12) demonstrate marked spatial variability across Zhejiang Province. During the winter, a widespread wetting trend is observed, particularly across the northern and northeastern regions. In contrast, the west–central and southernmost regions, and a small portion of the southeastern coastal area, exhibit slight drying trends. In spring, drying trends intensify, especially across southern and western Zhejiang. Summer also shows notable drying over most regions, particularly in south–central and eastern coastal regions, with significant trends. During autumn, regions with drying trends diminish in size, while wetting areas expand, though most trends remain statistically insignificant.

SPEI12 trends (Figure 13) exhibit a high degree of spatial consistency across all months. A wetting trend is prevalent across the majority of central and northeastern Zhejiang, whereas the northwestern, southern, and southeastern regions are predominantly characterized by a drying trend. Although the spatial pattern remains relatively stable throughout the year, the extent of the wetting or drying areas fluctuates slightly with the seasons. Overall, the wetting areas in summer and autumn are larger than those in winter and spring. The southeastern coastal region consistently demonstrates a drying trend throughout the year, which is often statistically significant.

To identify the primary drivers of SPEI variability, the relative contributions of precipitation and PET were further quantified, as illustrated in Figure 14. The results show that precipitation consistently exerts the dominant influence on SPEI variations across all timescales. Specifically, for the short-term SPEI1, precipitation accounts for 57–69% of the variations across the province, while PET contributes 31–43%. Similar patterns are observed for SPEI3 and SPEI6, with precipitation explaining 53–64% and PET contributing 36–47%. In contrast, at the 12-month scale (SPEI12), the dominance of precipitation becomes particularly pronounced, with contribution rates reaching 71–86%, compared to only 14–29% for PET. These findings imply that the short- to medium-term SPEI is co-regulated by both precipitation and PET, whereas the long-term SPEI is primarily governed by precipitation in this region.

3.2.2. Spatial Distributions of Drought Prevalence Across Multiple SPEI Timescales

In addition to conducting trend and attribution analyses of the SPEI to elucidate long-term drought changes and their climatic drivers, this study also investigated the spatial patterns of drought prevalence across various severity levels based on the PIEII data combinations (CMFD V2.0 and GLEAM v4.2a). For each 0.1° grid cell, the number of months from 1980 to 2020 that fell within four predefined drought categories (see Table 2) was counted. This count was then divided by the total number of months (492 months during 1980–2020) to determine the occurrence rate for each drought category. It is important to note that this metric represents the percentage of total time a location was subjected to a given drought condition, reflecting the overall temporal exposure to drought rather than the frequency of distinct, independent drought events. Figure 15 illustrates the spatial distribution of these occurrence rates over Zhejiang Province, based on the SPEI at four timescales (SPEI1, SPEI3, SPEI6, and SPEI12).

Overall, light droughts (−1.0 < SPEI ≤ −0.5) occur most frequently, followed by moderate (−1.5 < SPEI ≤ −1.0), severe (−2.0 < SPEI ≤ −1.5), and extreme droughts (SPEI ≤ −2.0). At the one-month timescale (SPEI1), which reflects short-term meteorological drought, light droughts are widespread, with most areas experiencing rates exceeding 15%. Moderate droughts occur more frequently in the south than in the north. Severe droughts are less frequent, generally below 10% in occurrence rate in most areas, and are more localized in the northwest. Extreme droughts are very rare across the entire province.

For SPEI3 (reflecting meteorological drought with emerging agricultural drought signals) and SPEI6 (primarily indicating agricultural drought), the spatial extent of light droughts with an occurrence rate above 15% shrinks, especially in the northern regions. Moderate droughts show a relatively uniform spatial distribution, with rates mostly ranging between 5% and 15% across the province. Notably, the occurrence rates of moderate drought are generally higher under SPEI3 than SPEI6. In contrast, severe droughts occur more frequently under SPEI6, particularly in the southernmost regions. Extreme droughts remain infrequent at both timescales.

At the 12-month timescale (SPEI12), which captures hydrological drought, the occurrence rates of light droughts are higher than at shorter timescales, with hotspots, primarily concentrated in central and western regions, exceeding 20%. Moderate droughts are more prevalent in the southwest and less frequent in the central–northwestern zone, while the opposite is observed for severe droughts. Extreme droughts remain rare at this longer timescale.

4. Discussion

4.1. Applicability of Reanalysis and Remote Sensing Data in Regional Drought Assessment

This study conducted a comprehensive evaluation of various combinations of reanalysis and remote sensing datasets for precipitation and evapotranspiration in estimating the SPEI across multiple timescales. A key finding is the predominant influence of precipitation products on SPEI estimation over the evapotranspiration products. As shown in Figure 3, the choice of precipitation product markedly affects the accuracy of the derived SPEI, whereas replacing evapotranspiration products under the same precipitation dataset results in relatively minor differences. This phenomenon likely stems from the structure of the SPEI formulation [6], where precipitation acts as the primary input variable and exerts dominant control on the water balance. In contrast, potential evapotranspiration, as the water output, serves more as a secondary adjustment factor that modulates the magnitude of water deficit. Additionally, precipitation in the study area (i.e., Zhejiang Province) exhibits pronounced spatiotemporal variability due to its complex topography and subtropical monsoon climate. Evapotranspiration, conversely, is a more gradual process governed by multiple factors including energy, water availability, and vegetation physiology, and its variability is typically less pronounced than that of precipitation [51]. Consequently, differences among evapotranspiration products have a relatively limited impact on SPEI estimations. Therefore, the selection of a reliable precipitation product emerges as the critical factor for enhancing the accuracy of drought assessment based on reanalysis and remote sensing data.

Although the combination of CMFD V2.0 and GLEAM v4.2a (PIEII) was validated as the most robust among the nine evaluated combinations, our findings also reveal a shared limitation across all data combinations: there are substantial errors in estimating extreme droughts (SPEI ≤− 2.0) at shorter timescales (SPEI1 and SPEI3), whereas performance improves markedly at longer timescales (SPEI6/12). This can be attributed to the limited responsiveness of reanalysis and remote sensing products to abrupt and severe water deficits, particularly in capturing extreme low-rainfall events or rapid evaporative losses during heatwaves [32,52]. At short timescales, such as SPEI1 and SPEI3, errors in monthly precipitation and PET estimation tend to be directly amplified within the SPEI calculation. In contrast, at the longer timescales of SPEI6 and SPEI12, the accumulation process acts as a low-pass filter, effectively smoothing random or short-term errors and thereby better highlighting the more persistent, long-term climate signal [53]. This finding has crucial implications for practical applications: when using reanalysis and remote sensing data for the detection of short-term extreme droughts, caution must be exercised. It is strongly recommended that such applications be supplemented by ground-based meteorological observations to avoid misjudgment of drought severity.

4.2. Insights from Multi-Scale Spatiotemporal Drought Variations for Adaptive Water Management

Based on the validated, high-performance PIEII data combination, this study conducted a multi-scale analysis of spatiotemporal variations in drought in Zhejiang Province over the past four decades. The province-averaged SPEI time series reveals a significant and persistent drought period from 2003 to 2009. This finding aligns with broader regional evidence showing that southern China, particularly the Yangtze River Basin and its adjacent regions, experienced multiple severe and prolonged droughts during this period [54,55,56]. PIEII’s effectiveness in capturing this multi-year drought validates its reliability for historical drought reconstruction and reinforces its suitability for regional drought assessment.

A key aspect of our methodology is the month-by-month trend analysis of the SPEI, which is crucial for identifying critical intra-annual heterogeneity in drought dynamics. It is through this refined approach that we unveil the complex, scale-dependent, and seasonally structured characteristics of drought evolution, with different regions exhibiting distinct wetting and drying trends at various timescales. The implications of the SPEI trends become clear when linked to specific drought types. The analysis for short-term meteorological and initial agricultural droughts (SPEI1 and SPEI3) reveals a significantly intensified drying signal in spring, indicating a rising risk of water deficit prior to planting, which has major consequences for agricultural planning. For mid-term agricultural drought (SPEI6) and long-term hydrological drought (SPEI12), the trends show a notable spatial divergence. Notably, the southeastern coastal regions show marked drying trends, which correspond to the increasing pressure on water resources exacerbated by rapid urbanization in these areas [57]. This underscores the urgent need for targeted regional drought risk management, with a focus on strengthening irrigation security, optimizing water resource allocation, and enhancing infrastructure resilience.

A distinct seasonal asymmetry also emerges in the drought trends, characterized by a “wetter winters, drier springs” pattern. This is consistent with the early-spring drying trend in Southern China reported by Li et al. [58]. The observed seasonal drought pattern is compelling evidence of the complex manifestation of global climate change at a regional scale and reinforces the need for season-specific drought risk management. In addition, according to the analysis of the spatial patterns of drought prevalence, light droughts are widespread across Zhejiang Province, pointing to an elevated baseline of water stress that may affect long-term agricultural and hydrological stability.

The quantitative attribution analysis of precipitation and PET demonstrates that precipitation is the dominant driver of drought variability across all timescales, with particularly strong control over long-term hydrological drought. This is consistent with previous studies [59,60,61,62]. For instance, Zhang et al. [62] reported that precipitation deficit contributed more than PET excess in triggering flash droughts across most subregions of China, although they also identified PET as a significant factor influencing soil moisture droughts. In line with these findings, our results also highlight the non-negligible role of PET in shaping short- to medium-term droughts. Under the context of global warming and increasing atmospheric water demand, the influence of PET on drought dynamics should not be underestimated.

Furthermore, our results shed light on the characteristics of drought propagation across different timescales. Drought propagation delineates the temporal evolution and transformation of water deficits, originating from meteorological anomalies (meteorological drought) and progressively impacting agricultural systems (agricultural drought) and hydrological regimes (hydrological drought) [63,64,65]. In this study, the clear signal of spring drying across SPEI1, SPEI3, and SPEI6 suggests that the propagation from meteorological drought (precipitation deficit) to agricultural drought (soil moisture deficit) is highly efficient during this season. In the southern and southeastern regions, the significant drying trend for SPEI12 can be interpreted as the cumulative impact of repeated short- and medium-term drought events over years, ultimately affecting long-term regional water availability. This reflects the progression of drought from the meteorological and agricultural spheres to the hydrological domains. These findings highlight the necessity of adopting an integrated, multi-timescale drought assessment framework.

4.3. Limitations and Future Perspectives

In addition to trend analysis, this study also assessed drought prevalence through the occurrence rate of drought months, as detailed in Section 3.2.2. It is essential to clarify that this metric captures the cumulative duration of drought conditions rather than the frequency of independent drought events. Except for SPEI1, the time series of longer-scale SPEIs exhibit significant serial correlation, which means that a single prolonged drought event can result in multiple consecutive months of low SPEI values. Despite this, the approach remains valuable for identifying regions susceptible to droughts, providing a reliable basis for drought risk zoning.

Beyond this specific methodological consideration, the study has several other limitations. First, the selection of reanalysis and remote sensing products, while representative, is not exhaustive. In recent years, high-resolution global reanalysis and remote sensing products for precipitation and evapotranspiration (e.g., MSWEP, CHIRPS, ERA5, SiTHv2) have continued to emerge [31,66,67,68]. Incorporating a broader range of datasets in future studies would enable a more comprehensive comparison of SPEI estimation capabilities, improving the robustness and applicability of the findings.

Second, while the SPEI is a widely applied drought index, it does not account for all dimensions of drought. This study did not incorporate direct ecological response indicators such as soil moisture or vegetation indices, which circumscribes the ability to comprehensively identify ecological and compound droughts [69,70].

Third, the assessment of drought drivers relied on multiple linear regression to quantify the relative contributions of precipitation and PET. Although effective in identifying dominant climatic drivers, this method assumes linear relationships and thus simplifies the complex, non-linear processes governing drought dynamics.

Moreover, anthropogenic influences, such as aerosol pollution and land-use change, were not considered. Aerosols can influence evapotranspiration by altering radiation fluxes and surface energy balance, ultimately influencing soil moisture availability [71,72]. Land-use changes, such as urbanization and deforestation, affect surface albedo, runoff patterns, and hydrological cycling, thereby altering regional drought regimes [73]. Given the historically high levels of aerosol emissions and rapid land-use changes in eastern China, accounting for these human-induced drivers is essential for a more comprehensive understanding of drought evolution.

In summary, while this study provides valuable insights into the spatiotemporal variations and climatic drivers of drought in Zhejiang, further research is needed. Future work integrating additional data sources, ecological indicators, and anthropogenic drivers (including land-use changes and aerosols) into process-based models will deepen the understanding of drought mechanisms and enhance predictive capabilities.

5. Conclusions

This study conducts a systematic evaluation of nine reanalysis and remote sensing data combinations for multi-timescale SPEI-based drought assessment in Zhejiang Province. The results demonstrate the dominant role of precipitation data in determining the accuracy of SPEI estimation. Among the data combinations tested, the integration of CMFD V2.0 precipitation and GLEAM v4.2a evapotranspiration (PIEII) consistently yields the most accurate and robust performance across multiple timescales.

Utilizing the optimal PIEII-based SPEI dataset, the analysis of drought conditions in Zhejiang during 1980–2020 reveals significant spatiotemporal heterogeneity. Temporally, the trends in the SPEI exhibit a marked seasonal asymmetry, characterized by wetter winters and, more critically, drier springs. The intensified spring drying poses a substantial threat to regional agricultural water security during the crucial planting season. Spatially, a distinct spatial divergence emerges, with central and northeastern Zhejiang exhibiting wetting trends while the southern and southeastern coastal areas show a significant drying tendency. This drying is particularly pronounced for long-term hydrological drought (SPEI12), underscoring a growing risk of cumulative water deficits in these vulnerable regions. Additionally, a prevalence of light droughts is observed across the province, suggesting an elevated baseline of water stress that could impact long-term agricultural and hydrological stability. Furthermore, attribution analysis indicates that precipitation is the dominant driver across all timescales, while PET co-regulates short- to medium-term droughts.

These findings collectively indicate an escalating and multifaceted drought risk in Zhejiang, a traditionally humid region. This study not only provides a validated, high-resolution data combination for regional drought assessment applications but also offers valuable scientific insights for developing targeted, season-specific, and spatially differentiated water resource management and drought adaptation strategies to enhance regional resilience in a changing climate.