An Integrated Drought Index (Vapor Pressure Deficit–Soil Moisture–Sun-Induced Chlorophyll Fluorescence Dryness Index, VMFDI) Based on Multisource Data and Its Applications in Agricultural Drought Management

Abstract

To more precisely monitor drought, a new remote sensing-based drought index, the Vapor Pressure Deficit–Soil Moisture–Sun-Induced Chlorophyll fluorescence Dryness Index (VMFDI), with a spatial resolution of 1 km based on vapor pressure deficit (VPD), soil moisture (SM), and sun-induced chlorophyll fluorescence (SIF) data was constructed via a three-dimensional spatial distance model, and it was used to monitor dryness in the Yellow River Basin during 2003–2020. The spatiotemporal variations in and main factors of the VMFDI and agroecosystem responses were analyzed via the Theil–Sen median and Mann–Kendall tests and Liang–Kleeman information flow. The results revealed the following: (1) The VMFDI effectively monitors regional drought and is more sensitive than other indices like the standardized precipitation evapotranspiration index (SPEI) and GRACE drought severity index and single variables. (2) VMFDI values fluctuated seasonally in the Yellow River Basin, peaking in August and reaching their lowest in March. The basin becomes drier in winter but wetter in spring, summer, and autumn, with the middle and lower reaches, particularly Shaanxi and Gansu, being drought-prone. The VMFDI values in the agroecosystem were lower. (3) SM and VPD dominated drought at the watershed and agroecosystem scales, respectively. Key agroecosystem indicators, including greenness (NDVI), gross primary productivity (GPP), water use efficiency (WUE), and leaf area index (LAI), were negatively correlated with drought (p < 0.05). When VPD exceeded a threshold range of 7.11–7.17 ha, the relationships between these indicators and VPD shifted from positive to negative. The specific VPD thresholds in maize and wheat systems were 8.03–8.57 ha and 7.15 ha, respectively. Suggestions for drought risk management were also provided. This study provides a new method and high-resolution data for accurately monitoring drought, which can aid in mitigating agricultural drought risks and promoting high-quality agricultural development.

1. Introduction

Drought is one of the most serious natural disasters with wide distribution, long duration, and recurrent frequency, and it has become an urgent issue worldwide [1,2]. Drought severely affects food security, economic development, biological health, and even the course of civilizations [3,4,5]. Between 2000 and 2019, approximately 18% of the global population was affected by drought, leading to 16,000 deaths annually and economic losses of 6.2 billion dollars per year [6]. Owing to droughts, annual crop production losses also reached 3 billion dollars in the past decade [7]. In China, the average drought-related crop yield losses exceed 30 million tons annually [8]. Climate change and human activity are expected to intensify drought conditions, potentially displacing 700 million people by 2030 and 216 million by 2050 [6]. Given these far-reaching influences, effective drought monitoring is essential for early warning and mitigation strategies.

Drought manifests as an imbalance in the water cycle and is commonly categorized into meteorological, hydrological, agricultural, and socioeconomic types [9,10,11]. Precipitation (PRE), vapor pressure deficit (VPD), runoff, temperature, the vegetation spectrum, and other related factors have been employed to quantitatively monitor and assess droughts. Drought indices are vital for quantifying drought intensities, spatiotemporal characteristics, and evolutionary trends.

To date, numerous drought indices have been established to monitor meteorological, hydrological, agricultural, and socioeconomic droughts, but each has notable limitations. Meteorological indices such as the Palmer Drought Severity Index (PDSI), Standardized Precipitation Index (SPI), and standardized precipitation evapotranspiration index (SPEI) are widely used. However, they suffer from low spatial resolution due to reliance on station-based data, limiting their effectiveness in large or data-sparse regions [12,13,14,15]. While the SPEI integrates both precipitation and evapotranspiration, its spatial resolution remains inadequate for comprehensive drought monitoring. Hydrological drought indices like the standardized runoff index (SRI) and the standardized streamflow index (SSI) focus on runoff/streamflow anomalies, but their accuracy depends on the quality of runoff data [16,17]. Satellite-based indices, i.e., the GRACE drought severity index (GRACE-DSI), offer better spatial coverage but face challenges with temporal resolution and data gaps in certain regions [18]. Socioeconomic drought indices, including the Multivariate Standardized Reliability and Resilience Index (MSRRI), Socioeconomic Drought Index (SEDI) and standardized water supply and demand index (SWSDI), often struggle with socioeconomic data integration and are limited by data availability and complexity [19,20,21,22].

Agricultural drought, closely linked to food security, is defined as a deficit in soil moisture (SM) caused by abnormal meteorological conditions, such as insufficient precipitation, reduced evapotranspiration, and increased land surface temperature (LST) [23]. These SM shortages further impact crop photosynthesis, growth, and yields [24]. Thus, monitoring SM is critical and can be achieved through in situ measurements [25]; satellite-based models (e.g., SMAP) [26]; optical, thermal, and microwave sensors [27]; and global navigation satellite systems (GNSSs) [28,29]. Meteorological indicators, e.g., the PDSI, are also used to monitor agricultural drought, as they correlate with SM anomalies. Drought limits crop growth and yields and is characterized by low SM and a high VPD, which inhibits plant photosynthesis and growth by causing stomatal closure [30]. VPD, which represents the difference between saturated and actual vapor pressure, is influenced by both the atmospheric temperature and humidity and has gained popularity as an indicator for agricultural drought monitoring [31,32]. Additionally, vegetation indices, such as the normalized difference vegetation index (NDVI), the enhanced vegetation index (EVI), the vegetation conditions index (VCI), the temperature condition index (TCI), the normalized difference water index (NDWI), and the vegetation health index (VHI), are commonly used to assess crop health during drought [33]. While these indices provide valuable insights, Solar-Induced Chlorophyll Fluorescence (SIF) has proven more sensitive to SM deficits and responds more rapidly to drought stress, making it an important indicator for estimating drought-induced crop losses [34].

While existing indices based on ground observations can estimate drought events well, they suffer from poor spatial continuity, high data capture costs, and limited comparability across different spatiotemporal scales. Satellite-based drought indices, which are timely, extensive, continuous, and inexpensive, have been proposed and widely applied in global and regional agricultural drought monitoring [35]. However, satellite-based indices that rely on a single or limited set of factors are insufficient for capturing the complexity of drought conditions. To improve drought assessment accuracy, several comprehensive remote sensing (RS) drought indices have been developed, i.e., the temperature–vegetation–soil moisture drought index (TVMDI) [36], temperature–vegetation–precipitation dryness index (TVPDI) [37], and temperature–SIF–Water Balance Dryness Index (TSWDI) [38]. While these indices offer a more holistic view of drought, they are often complicated, require large amounts of data, and face challenges with regional applicability [39]. Additionally, indices with low spatiotemporal resolution are not suitable for agricultural drought assessments. Therefore, it is necessary to construct a novel integrated index that can accurately capture drought dynamics and meet the needs of agricultural management.

Consequently, in this study, a novel integrated RS drought index, the Vapor Pressure Deficit–Soil Moisture–Sun-Induced Chlorophyll fluorescence Dryness Index (VMFDI), based on high-resolution multisource data and key drought-causing factors (VPD, SM, and SIF) and drought principles was constructed via a three-dimensional spatial distance model. The objectives of this study are as follows: (1) to validate the VMFDI’s accuracy and reliability through multiple validation methods; (2) to apply the VMFDI to analyze the spatiotemporal dynamics of drought in the Yellow River Basin and the agroecosystem and assess the main factors influencing drought conditions; and (3) to explore the responses of crop growth, productivity, and yields in the agroecosystem to droughts, providing critical insights into drought-induced agricultural risks. Based on these findings, this study aims to propose evidence-based strategies for enhancing agricultural drought resilience, informing smart agriculture practices, and guiding high-quality agricultural development.

2. Materials and Methods

2.1. Study Area

The Yellow River Basin is not only a vital economic region and ecological barrier but also a major grain production area in China with a long history of agricultural cultivation. It is located at 32°10′N~41°50′N and 95°53′E~119°05′E, covers an area of 795,000 km², and spans the eastern, central, and western regions within nine provinces, namely, Qinghai, Sichuan, Gansu, Ningxia, Inner Mongolia, Shaanxi, Shanxi, Henan, and Shandong (Figure 1). The Yellow River Basin is generally divided into the upstream, midstream, and downstream parts [40]. The terrain of the Yellow River Basin is high in the west and low in the east, and the maximum difference in elevation is greater than 6200 m (Figure 1). There are many high mountains with abundant snow in the upper reaches; the Loess Plateau, which has severe soil erosion, is present in the middle reaches; and an alluvial plain is present in the lower reaches of the basin. The basin has diverse climate types, but arid and semiarid climates with low total precipitation, high evaporation, and low humidity are prevalent [41]. The land cover types in the basin are mainly grasslands and croplands, which account for approximately 65% and 21% of the total area of the basin, respectively (Figure 1). Water scarcity has become a major challenge for regional sustainable development. Specifically, the temperature has increased, and precipitation has decreased because of climate change [42]. The cropland in the middle and lower reaches of the basin continuously expanded by 2.3 times during the 1000 years before 2000 [43], and that in the whole basin increased by another 14% over the last two decades [44]. Notably, frequent agricultural drought; increased agricultural water demand; and prominent agricultural, industrial, domestic, and ecological water conflicts have severely constrained sustainable regional agricultural development. Indeed, the average use of water per unit of cultivated land is only 0.49 m³/m², which is 17% of that at the country level. Additionally, the complex climate, diverse landscape, and crop planting systems meet challenges for agricultural water management in the basin. Therefore, to achieve regional drought relief and food security, the Yellow River Basin should be studied as a typical area in urgent need of drought monitoring and agricultural water resource management.

2.2. Data Sources and Processing

Drought occurs with many factors and is a complex process. Indices that are based on one or two factors cannot fully capture drought information. Therefore, a comprehensive drought index that can better assess drought condition is needed. Compared with existing indices, synthesized drought indices with a one-month time scale have been verified to have better performance in capturing the impacts of drought on crop growth [39]. Therefore, a new integrated drought index with a high spatial resolution (1 km $\times$ 1 km) and a one-month time scale is proposed in this study. A flowchart of the data collection, processing, and new integrated drought index, namely, the VPD-SM-SIF Dryness Index (VMFDI), calculation, as well as its validation and application in this study, is shown in Figure 2. The details of the corresponding data sources and the main preprocessing steps and methods are presented in Section 2.2 and Section 2.3.

2.2.1. Data for Constructing the VMFDI

VPD, SM, and SIF have been verified for effective drought monitoring from different perspectives [24,38]. To establish the VMFDI, VPD, SM, and SIF data from 01/2003 to 12/2020 were collected (

Table 1. Description of related datasets used in this study.

Data Type	Production /Tile	Spatial Resolution	Temporal Resolution	Unit	Data Period	Source
VPD	HiMIC	1 km	Monthly	hPa	2003–2020	http://data.tpdc.ac.cn (accessed on 14 May 2024)
SM	SMCI1.0	30″	Daily	0.001 m3/m3	2000–2020	http://data.tpdc.ac.cn (accessed on 14 May 2024)
SIF	GOSIF	0.05°	Monthly	mW m−2 um−1 sr−1	2000–2023	https://data.globalecology.unh.edu/data/GOSIF_v2/ (accessed on 14 May 2024)
SPEI	SPEI base v2.8	0.5°	Monthly	/	1901–2021	https://spei.csic.es/spei_database (accessed on 15 May 2024)
TWS	GTWS-MLrec	0.25°	Monthly	mm	1940–2022	https://zenodo.org/records/10040927 (accessed on 15 May 2024)
PRE	Precipitation	1 km	Monthly	0.1 mm	1901–2022	https://www.tpdc.ac.cn/zh-hans/data/faae7605-a0f2-4d18-b28f-5cee413766a2/ (accessed on 17 May 2024)
ET	MOD16A2	500 m	8-day	mm	2001–	https://developers.google.cn/earth-engine/datasets (accessed on 17 May 2024)
NDVI	MOD13A1	500 m	16-day	/	02/2000–12/2023
GPP	MOD17A2H	500 m	8-day	gC m−2	02/2000–02/2023
LAI	MOD15A2H	500 m	8-day	Area fraction	02/2000–
Land	MCD12Q1 LC_Type1	500 m	Yearly	/	01/2001–01/2022
Maize Wheat	ChinaCropArea1km	1 km	Yearly	/	2003–2019	https://www.nesdc.org.cn (accessed on 04/07/2024 and 16 July 2024)
	ChnaWheat/Maize10	10 m	Yearly	/	2020

).

VPD is an important indicator reflecting regional meteorological drought and inflowing vegetation growth, defined as the difference between saturated vapor pressure and actual vapor pressure at a specific temperature. In this study, monthly VPD data were extracted from the HiMIC-Monthly database, which provides 1 km high-resolution atmospheric moisture data for China in 2003–2020 [45]. The dataset has an accuracy of over 0.96, with a root mean square error and mean absolute error within acceptable ranges [46].

SM data were obtained from the SMCI1.0 (Soil Moisture of China by in situ data, version 1.0) of the National Tibetan Plateau/Third Pole Environment Data Center [25]. SMCI1.0 provides daily soil moisture at 10 cm depth intervals up to 100 cm, with a spatial resolution of ~1 km. The data are derived from machine learning based on the observed soil moisture from 1648 stations of the China Meteorological Administration, ERA5-land, leaf area index, land cover type, topography, and soil properties [38,47]. In this study, monthly SM was summed from daily data, averaged to the 0–40 cm soil depth, which effectively reflects agricultural drought. The data were downscaled to 1 km via the nearest neighbor sampling method to match VPD resolution.

SIF is less influenced by soil background and clouds and directly reflects plant photochemical activity. It has become a key parameter for monitoring vegetation growth and drought stress [38]. However, original Orbiting Carbon Observatory-2 (OCO-2) SIF data are spatially discontinuous and are not suitable for regional-scale studies. Therefore, global OCO-2-based SIF product (GOSIF) data were adopted [48]. GOSIF data were downscaled to 1 km to match the VPD and SM data for VMFDI calculation.

2.2.2. Data for Verifying the VMFDI

The SPEI has been widely applied to reflect regional meteorological drought [15,49]. The GRACE-DSI is a reasonable indicator for hydrological drought [18]. In this study, the SPEI, GRACE-DSI, VPD, SM, SIF, evapotranspiration (ET), PRE, observed drought events, and crop growth status indicators were used to validate the VMFDI (Table 1).

The SPEI product (SPEI base v2.8), with a 0.5° spatial resolution, was calculated via the FAO-56 Penman–Monteith model with monthly precipitation and potential evapotranspiration from the CRU TS 4.06 dataset. Data from three temporal scales—SPEI01, SPEI03, and SPEI12—were selected to evaluate the VMFDI’s drought monitoring ability.

The GRACE-DSI, based on standardized anomalies of GRACE terrestrial water storage (GRACE-TWS), was derived from the GTWS-MLrec dataset (0.25° × 0.25°) for 2003–2020 [50], reconstructed by a machine learning method.

ET data (2003–2020) were obtained from the MOD16A2 product (500 m resolution) based on MODIS Terra net evapotranspiration data processed with the Google Earth Engine.

Precipitation data (2003–2020) were obtained from the National Tibetan Plateau/Third Pole Environment Data Center [51], using a 1 km monthly dataset generated by downscaling global climate data [52].

Observed drought event information from the Bulletin of Flood and Drought Disasters in China (2003~2020) was used to test the monitoring capacity of the VMFDI and other drought indicators. During this period, the Yellow River Basin experienced droughts in 64 months.

2.2.3. Data of Agricultural Status in Response to Drought

Vegetation greenness reflects vegetation growth response to drought, decreasing as drought intensifies [53]. It is closely correlated with gross primary productivity (GPP), though this relationship weakens in less arid conditions [54]. GPP, representing carbon fixed by crop photosynthesis, is also sensitive to drought. Water use efficiency (WUE), which measures dry matter per unit water consumed through evapotranspiration [55], is an important index of regional agricultural water management and crop growth under drought. The risk of crop yield loss increases nonlinearly with drought severity [56]. In this study, vegetation greenness, GPP, WUE, and crop yield were used to assess crop responses to drought. The normalized difference vegetation index (NDVI) and leaf area index (LAI) were used to reflect crop greenness and crops yields, respectively.

Data on land cover, NDVI, GPP, and the LAI from 2003 to 2020 were analyzed with the Google Earth Engine based on MODIS products from NASA. Cropland was extracted from the MCD12Q1 LC_Type1 product (500 m resolution). Monthly GPP was calculated using the MOD17A2H product (8-day cumulative composite) [57]. The LAI was obtained from the MODIS MOD15A2H product (8-day, 500 m resolution). Finally, these data were upscaled to 1 km to match VMFDI resolution. Maize and wheat data from 2000 to 2019 were sourced from the ChinaCropArea1km, with 2020 data from ChinaWheat10 and 10 m maize cropland data, both resampled to 1 km.

2.3. Methodology

2.3.1. Construction of VMFDI

The VMFDI was established based on a modified three-dimensional spatial distance model. The spatial distance model, which uses Euclidean distance to measure the absolute distance between points in multidimensional space, has been widely applied in remote sensing for various environmental assessments, including drought monitoring and ecological studies [36,37,38]. The VMFDI was constructed using the following steps:

Data preprocessing: First, outliers in the variables VPD, SM, and GOSIF were removed using the cumulative probability dichotomy method [38]. Then, these variables were normalized to a range of 0–1 to address the differences in their value scales (Equation (1)).

$$\mathrm{N}\mathrm{V}\mathrm{a}\mathrm{r}(\mathrm{i},\mathrm{j})={\displaystyle \frac{\mathrm{V}\mathrm{a}\mathrm{r}(\mathrm{i},\mathrm{j})-{\mathrm{V}\mathrm{a}\mathrm{r}\left(\mathrm{i}\right)}_{\mathrm{m}\mathrm{i}\mathrm{n}}}{{\mathrm{V}\mathrm{a}\mathrm{r}\left(\mathrm{i}\right)}_{\mathrm{m}\mathrm{a}\mathrm{x}}-{\mathrm{V}\mathrm{a}\mathrm{r}\left(\mathrm{i}\right)}_{\mathrm{m}\mathrm{i}\mathrm{n}}}}$$

(1)

where $\mathrm{N}\mathrm{V}\mathrm{a}\mathrm{r}(\mathrm{i},\mathrm{j})$ represents the normalized value of the ${\mathrm{j}}^{\mathrm{t}\mathrm{h}}$ month for the ${\mathrm{i}}^{\mathrm{t}\mathrm{h}}$ variable (VPD, SM, or SIF). $\mathrm{V}\mathrm{a}\mathrm{r}(\mathrm{i},\mathrm{j})$ represents the original value, and ${\mathrm{V}\mathrm{a}\mathrm{r}\left(\mathrm{i}\right)}_{\mathrm{m}\mathrm{i}\mathrm{n}}$ and ${\mathrm{V}\mathrm{a}\mathrm{r}\left(\mathrm{i}\right)}_{\mathrm{m}\mathrm{a}\mathrm{x}}$ are the minimum and maximum values of the ${\mathrm{i}}^{\mathrm{t}\mathrm{h}}$ variable.

Spatial distance model: The VMFDI was calculated using a three-dimensional spatial distance model, where the variables VPD, SM, and SIF were set as axes X, Y, and Z, respectively (Figure 3). The driest point D (1, 0, 0), representing the highest VPD and the lowest SM and SIF, served as the reference point, while the wettest point W (0, 1, 1) represents the lowest VPD and the highest SM and SIF. The dry–wet line connecting points D and W defined the range of drought conditions.

VMFDI calculation: The Euclidean distance from any point in the hyperspace to point D (1, 0, 0) was calculated, with the distance representing the relative drought value. The formula for the VMFDI is as follows:

$$\mathrm{V}\mathrm{M}\mathrm{F}\mathrm{D}\mathrm{I}=\sqrt{{({\mathrm{N}\mathrm{V}\mathrm{P}\mathrm{D}}_{\mathrm{M}\mathrm{a}\mathrm{x}}-\mathrm{N}\mathrm{V}\mathrm{P}\mathrm{D})}^2+{(\mathrm{N}\mathrm{S}\mathrm{M}-{\mathrm{N}\mathrm{S}\mathrm{M}}_{\mathrm{M}\mathrm{i}\mathrm{n}})}^2+{(\mathrm{N}\mathrm{S}\mathrm{I}\mathrm{F}-{\mathrm{N}\mathrm{S}\mathrm{I}\mathrm{F}}_{\mathrm{M}\mathrm{i}\mathrm{n}})}^2}$$

(2)

where NVPD, NSM, and NSIF represent the normalized values of the VPD, SM, and SIF, respectively. ${\mathrm{N}\mathrm{V}\mathrm{P}\mathrm{D}}_{\mathrm{m}\mathrm{a}\mathrm{x}}$ refers to the maximum normalized values of VPD, while ${\mathrm{N}\mathrm{S}\mathrm{M}}_{\mathrm{m}\mathrm{i}\mathrm{n}}$ and ${\mathrm{N}\mathrm{S}\mathrm{I}\mathrm{F}}_{\mathrm{m}\mathrm{i}\mathrm{n}}$ refer to the minimum normalized values of SM, and SIF, respectively. The VMFDI ranges from 0 to $\sqrt{3}$, with higher values indicating wetter conditions and lower values indicating drier conditions.

2.3.2. Validation of VMFDI

To assess the performance of the VMFDI, it was compared with established drought indices such as the SPEI and GRACE-SDI, using observed drought events from 2003 to 2020. Drought events were identified using the SPEI values (SPEI01, SPEI03, and SPEI12) and GRACE-SDI values, with a threshold of −0.5 for drought detection [18,44,58]. For the VMFDI, drought events were identified based on the multiyear average VMFDI values (${\mathrm{V}\mathrm{M}\mathrm{F}\mathrm{D}\mathrm{I}}_{\mathrm{m}\mathrm{e}\mathrm{a}\mathrm{n}}$) for the corresponding month. The details of the drought classification are shown in

Table 2. Drought classification based on the VMFDI, SPEI, and GRACE-DSI.

Degree of Drought	VMFDI	SPEI	GRACE-DSI
Drought	(−∞, ${\mathrm{V}\mathrm{M}\mathrm{F}\mathrm{D}\mathrm{I}}_{\_\mathrm{m}\mathrm{e}\mathrm{a}\mathrm{n}}$)	(−∞, −0.5]	(−∞, −0.5]
No drought	[${\mathrm{V}\mathrm{M}\mathrm{F}\mathrm{D}\mathrm{I}}_{\_\mathrm{m}\mathrm{e}\mathrm{a}\mathrm{n}}$, +∞)	(−0.5, +∞)	(−0.5, +∞)

.

A correlation analysis was conducted to examine the relationship between the VMFDI and other variables (VPD, SM, ET, PRE, SIF, NDVI, WUE, GPP, and LAI) to evaluate the sensitivity of the VMFDI compared to other drought indices. The correlation coefficient (Equation (3)) was calculated to quantify the strength of the relationships, with higher correlations suggesting greater sensitivity.

$$\mathrm{R}={\displaystyle \frac{{\displaystyle \sum _{\mathrm{i}=1}^{\mathrm{n}}\left({\mathrm{X}}_{\mathrm{i}}-\overline{\mathrm{X}}\right)\left({\mathrm{Y}}_{\mathrm{i}}-\overline{\mathrm{Y}}\right)}}{\sqrt{{\displaystyle \sum _{\mathrm{i}=1}^{\mathrm{n}}{\left({\mathrm{X}}_{\mathrm{i}}-\overline{\mathrm{X}}\right)}^2}}\sqrt{{\displaystyle \sum _{\mathrm{i}=1}^{\mathrm{n}}{\left({\mathrm{Y}}_{\mathrm{i}}-\overline{\mathrm{Y}}\right)}^2}}}}$$

(3)

where ${\mathrm{X}}_{\mathrm{i}}$ and ${\mathrm{Y}}_{\mathrm{i}}$ are the values of the corresponding variables at time $\mathrm{i}$; $\overline{\mathrm{X}}$ and $\overline{\mathrm{Y}}$ are their mean values.

2.3.3. Spatiotemporal Analysis of VMFDI

The spatiotemporal distribution of the VMFDI at grid cells was analyzed using univariate linear regression to identify interannual trends. It can be expressed as Equation (4):

$${\mathrm{VMFDI}}_{\mathrm{slope}}={\displaystyle \frac{(\mathrm{n}{\displaystyle \sum _{\mathrm{j}=1}^{\mathrm{n}}\mathrm{j}\times \mathrm{V}\mathrm{M}\mathrm{F}\mathrm{D}{\mathrm{I}}_{\mathrm{i},\mathrm{j}}}-{\displaystyle \sum _{\mathrm{j}=1}^{\mathrm{n}}\mathrm{j}{\displaystyle \sum _{\mathrm{j}=1}^{\mathrm{n}}\mathrm{V}\mathrm{M}\mathrm{F}\mathrm{D}{\mathrm{I}}_{\mathrm{i},\mathrm{j}}}})}{\mathrm{n}{\displaystyle \sum _{\mathrm{j}=1}^{\mathrm{n}}{\mathrm{i}}^2-({\displaystyle \sum _{\mathrm{j}=1}^{\mathrm{n}}\mathrm{i}{)}^2}}}}$$

(4)

where ${\mathrm{VMFDI}}_{\mathrm{slope}}$ is the trend in the change, n is the length of the research period, $\mathrm{i}$ is the month ($1\le \mathrm{i}\le 12$), $\mathrm{j}$ is the year ($1\le \mathrm{j}\le \mathrm{n}$), and ${\mathrm{VMFDI}}_{\mathsf{i},\mathsf{j}}$ is the value of the VMFDI in the $i^{th}$ month of the $j^{th}$ year. ${\mathrm{VMFDI}}_{\mathrm{slope}}$ > 0 indicates a decreasing trend in drought; ${\mathrm{VMFDI}}_{\mathrm{slope}}$ < 0 suggests an increasing trend.

To assess the significance of the trends, the Theil–Sen median test and Mann–Kendall (M-K) test were conducted [59,60]. These tests are widely used to analyze trends in climate and hydrological data, and they provide a nonparametric approach to determining whether the observed trends are statistically significant at a 95% confidence level.

The center of gravity model and standard deviation ellipse were used to identify changes in the drought center in this study, which have been widely used in meteorological disasters [61,62]. The standard deviational ellipse, which can reflect the direction of the overall dominant distribution of spatial elements and the dispersion of each direction, is generally used to analyze the directional distribution of spatial elements [61,62]. In this study, the spatial aggregation characteristics and trends in changes in drought in the Yellow River Basin were analyzed by plotting the standard deviational ellipse of drought in different periods.

2.3.4. Identifying the Causal Relationship Method

To identify the causal relationships among the variables influencing drought, the Liang–Kleeman information flow method was employed. This method calculates the rate of information flow between two time series, offering insights into the direction and strength of causal relationships [63,64]. When given two time series ${\mathrm{X}}_1$ and ${\mathrm{X}}_2$ in a linear system, the rate of the information flow from ${\mathrm{X}}_2$ to ${\mathrm{X}}_1$ can be calculated as follows:

$${\mathrm{T}}_{2\to 1}={\displaystyle \frac{{\mathrm{C}}_{11}{{\mathrm{C}}_{12}\mathrm{C}}_{2,\mathrm{d}1}-{\mathrm{C}}_{12}^2{\mathrm{C}}_{1,{\mathrm{d}}_1}}{{\mathrm{C}}_{11}^2{\mathrm{C}}_{22}-{\mathrm{C}}_{11}{\mathrm{C}}_{22}^2}}$$

(5)

$${\mathrm{C}}_{\mathrm{i}\mathrm{j}}=\sqrt{({\mathrm{X}}_{\mathrm{i}}-{\overline{\mathrm{X}}}_{\mathrm{i}})({\mathrm{X}}_{\mathrm{j}}-{\overline{\mathrm{X}}}_{\mathrm{j}})}$$

(6)

where ${\mathrm{C}}_{\mathrm{i}\mathrm{j}}$ is the sample covariance between ${\mathrm{X}}_{\mathrm{i}}$ to ${\mathrm{X}}_{\mathrm{j}}$. ${\mathrm{C}}_{\mathrm{i},{\mathrm{d}}_{\mathrm{j}}}$ is the covariance between ${\mathrm{X}}_{\mathrm{i}}$ and ${\dot{\mathrm{X}}}_{\mathrm{j}}$. ${\dot{\mathrm{X}}}_{\mathrm{j}}={\displaystyle \frac{{\mathrm{X}}_{\mathrm{j},\mathrm{n}+1}-{\mathrm{X}}_{\mathrm{j},\mathrm{n}}}{∆\mathrm{t}}}$, where $∆\mathrm{t}$ is the time step. If ${\mathrm{T}}_{2\to 1}=0$, there is no causality from ${\mathrm{X}}_2$ to ${\mathrm{X}}_1$; if ${\mathrm{T}}_{2\to 1}\ne 0$, there is a causal relationship between them. If ${|\mathrm{T}}_{2\to 1}|>{|\mathrm{T}}_{1\to 2}|>0$, ${\mathrm{X}}_1$ and ${\mathrm{X}}_2$ are mutually causal, and the causality from the latter to the former is larger than its counterpart.

3. Results

3.1. Verification of VMFDI

3.1.1. VMFDI vs. Other Dryness Indicators

The VMFDI was compared with other drought indices, such as the GRACE-DSI and SPEI, as well as the individual variables of PRE, VPD, SM, and SIF, to assess its ability in monitoring regional dryness. The GRACE-DSI is a single drought index based on TWS anomalies, while the SPEI is a two-dimensional drought index derived from the PRE and potential evapotranspiration (PET). In contrast, the VMFDI is a three-dimensional drought index built from VPD, SM, and SIF, offering a more comprehensive representation of drought conditions. The VMFDI peaked in August and was the lowest in March, aligning with seasonal variations in VPD, SM, and SIF (Figure 4f–h and Figure 5). Over 82% of the region had positive correlations between the VMFDI and PRE, SM, and SIF, while nearly 60% showed a negative correlation with VPD (Figure 4i). This confirms that the VMFDI can effectively monitor regional drought conditions. To evaluate the performance of the VMFDI, it was compared with the GRACE-DSI, SPEI01, SPEI03, and SPEI12 using both temporal and spatial correlation analyses (Figure 4a–d, Figure 5, and Figure 6). The VMFDI showed high spatial correlation with these indices, with 65.17%, 97.47%, 94.42%, and 97.99% of the basin positively correlated with the GRACE-DSI, SPEI01, SPEI03, and SPEI12, respectively (Figure 4i). Temporal trends also aligned particularly with SPEI01 and SPEI03 (Figure 5). Correlation analysis revealed significant positive correlations with the GRACE-DSI (R = 0.19, p < 0.05), SPEI01 (R = 0.39, p < 0.05), SPEI03 (R = 0.40, p < 0.05), and SPEI12 (R = 0.17, p < 0.05) (Figure 6). These results indicate that the VMFDI is highly consistent with other drought indices, and it can be used to evaluate drought. Additionally, according to the records of the Bulletin of Flood and Drought Disasters in China, 64 months of drought occurred in the basin from 2003 to 2020. The VMFDI identified 37 months of drought from 2003 to 2020, outperforming the GRACE-DSI (10 months), SPEI01 (26 months), SPEI03 (31 months), and SPEI12 (31 months) (Figure 6). This suggests that the VMFDI is more sensitive in detecting drought events. When compared to individual variables, VPD, SM, and SIF identified only 19, 22, and 22 drought months, respectively, underscoring the superiority of the VMFDI in capturing drought conditions.

3.1.2. Drought Monitoring Capability of VMFDI for Ecosystems

To compare drought monitoring accuracy, the correlations between the VMFDI, SPEI (SPEI01, SPEI03, and SPEI12), GRACE-DSI, and various indicators of water, heat, and growth conditions (PRE, SM, VPD, ET, LST, NDVI, SIF, GPP, WUE, and LAI) were analyzed (Figure 6). The VMFDI showed strong positive correlations with these drought indices, as well as with PRE (R = 0.71, p < 0.05), SM (R = 0.63, p < 0.05), ET (R = 0.62, p < 0.05), NDVI (R = 0.65, p < 0.05), SIF (R = 0.65, p < 0.05), GPP (R = 0.56, p < 0.05), and the LAI (R = 0.61, p < 0.05) (Figure 6a). These results indicate that the VMFDI effectively monitors meteorological and hydrological drought, as well as ecosystem drought, especially in agroecosystems. Compared to the SPEI and GRACE-DSI, the VMFDI exhibited stronger correlations with the NDVI, GPP, and LAI, indicating its superior sensitivity for monitoring ecosystem drought. Although the correlation of the VMFDI with VPD, LST, and WUE was weak across the full 2003–2020 period, significant negative correlations emerged in the monthly data (Figure 6a,b1–b12). The atmosphere can hold only a certain amount of water vapor at a certain temperature. As the LST increases, the upper limit for atmospheric water vapor rises, which in turn increases VPD, lowering the VMFDI and indicating drier conditions. A high VPD limits vegetation transpiration and restricts vegetation photosynthesis by closing vegetation’s stomata, thereby affecting carbon sequestration and evapotranspiration [65]. The relationship between WUE and the VMFDI depends on the rate of change between carbon sequestration and water loss. The VMFDI was positively correlated with growth status indices (GPP and LAI) from April to September but negatively correlated with these indices from October to March (Figure 6b1–b12). During low-VPD months, VPD was positively correlated with GPP and the LAI, but during high-VPD months, this relationship became negative, which is consistent with previous studies [65]. This shift underscores the importance of VPD thresholds in determining VPD’s effect on vegetation productivity, thus explaining the observed seasonal variation in the VMFDI’s correlation with GPP and the LAI.

3.2. Spatiotemporal Changes in the VMFDI

3.2.1. The Spatiotemporal Heterogeneity of the VMFDI in the Yellow River Basin

Figure 5 and Figure 7 show the monthly spatiotemporal patterns and variations in the VMFDI of the Yellow River Basin from January 2003 to December 2020. The results in Figure 5 show that the mean VMFDI value was $0.861\pm 0.07$ (mean $\pm$ standard deviation) and fluctuated annually, with the largest value occurring in August and the smallest value in March. Spatially, the multiyear average monthly VMFDI in the southern region of the middle and lower reaches of the Yellow River Basin is always lower than that in other regions, and the low-value areas shifted northward with seasonal changes (Figure 7a). This indicates that the middle and lower reaches of the Yellow River Basin were relatively dry and drought-prone. From 2003 to 2020, VMFDI values from November to March showed a decreasing trend over time, while those from April to October showed an increasing trend (Figure 7a). This suggests that the Yellow River Basin becomes drier in winter and wetter in spring, summer, and autumn. Specifically, the areas where the VMFDI decreased over time accounted for more than 80% of the total area in winter, with areas showing significant decreases (p < 0.05), accounting for 23.72~40.65% of the total area (Figure 7b,c). The regions with significant decreases were mainly in Inner Mongolia and Shanxi, Shaanxi, and Henan provinces (Figure 7c). From March to June and September, the areas where the VMFDI increased significantly (p < 0.05) over time covered more than 15.6% of the total area, primarily in parts of the regions of Qinghai, Gansu, and Shaanxi provinces. Figure 7 also shows that the VMFDI decreased over time in regions with initially lower VMFDI values, while it increased in regions with higher VMFDI values. This indicates that drier areas are becoming drier, while wetter areas are becoming wetter in the Yellow River Basin.

3.2.2. The Migration of the Drought Center in the Yellow River Basin

Changes in the 12-month and 18-year drought centers in the Yellow River Basin from 2003 to 2020, based on monthly and annual VMFDI anomalies and the gravity model, are shown in Figure 8. The monthly drought centers were concentrated in Shaanxi and Gansu, with a relatively small range of variation (Figure 8b). The direction of the standard deviation ellipse is 33.93° north by east. The monthly drought centers were clearly discrete in the west–east direction, with a maximum distance of 47.46 km movement of from March to April. The trajectory of the monthly drought center was shaped like an “8”. The drought centers during November and January were distributed in the lower part of the ‘8’ (Figure 8b). The annual drought centers were generally concentrated in the central areas of the basin, except in 2003. These centers were distributed mainly in the northeast to southwest belt (Figure 8c). The direction of the standard deviation ellipse of the annual drought center is 78.70° northeastward. The long axis is oriented in the southwest–northeast direction and has a length of 271,610.23 km. The elliptical flattening rate is 0.93. The drought centers were distributed mainly from southwest to northeast, with a high degree of dispersion. Overall, the annual drought center in the Yellow River basin moved southwest, with a total movement distance of more than 3837.90 km from 2003 to 2020 (Figure 8c).

3.3. The Responses of the Agroecosystem to the VMFDI

3.3.1. Changes in the VMFDI of the Agroecosystem of the Yellow River Basin

The monthly VMFDI value in the agroecosystem of the Yellow River Basin was $0.82\pm 0.007$ (mean $\pm$ SE) and fluctuated annually, with the largest value in August and the smallest value in March (range: 0.64~1.09) from 2003 to 2020 (Figure 9a). Compared to the total basin (Figure 5), VMFDI values in the agroecosystem were significantly lower. Correlation analysis revealed that VMFDI values in the agroecosystem significantly increased with increasing SM (r = 0.70, p < 0.05) and SIF (r = 0.63, p < 0.05) values. However, no significant correlation was observed between the VMFDI and VPD in agroecosystems, likely due to spatial heterogeneity in the VMFDI–VPD relationship. The results of the VMFDI anomalies in the agroecosystem of the Yellow River Basin revealed that 96 months experienced drought from 2003 to 2020, during which the VMFDI anomalies were lower than zero (Figure 9b).

3.3.2. The Response of the Crop Growth Status to the VMFDI

Correlations between the monthly VMFDI and the crop growth status indicators were analyzed (Figure 10). In the agroecosystem, crop greenness (NDVI) was significantly positively correlated with the VMFDI (r = 0.66, p < 0.05), indicating that reduced crop greenness occurred in response to regional dryness. The GPP, WUE, and LAI of crops were also significantly positively correlated with the VMFDI, with the correlation coefficients of 0.54 (p < 0.05), 0.20 (p < 0.05), and 0.59 (p < 0.05), respectively (Figure 10a). These findings suggest that crop productivity, water use efficiency, and yield decline as regional dryness increases. In maize and wheat planting systems, the NDVI, GPP, and LAI had strong positive correlations with the VMFDI (correlation coefficients > 0.62, p < 0.05, and >0.70, p < 0.05, respectively) (Figure 10b,c). This indicates that increasing drought intensity leads to decreased crop greenness, GPP, and yields. Notably, WUE was negatively correlated with the VMFDI in maize and wheat planting systems, primarily due to the inconsistent rates of change in GPP and ET, as well as crop resistance to drought. Although both GPP and ET decreased with increasing drought, crop drought resistance helped mitigate the reduction in carbon sequestration, particularly in maize (C₄ plants), which is more effective than C₃ plants in fixing CO₂ under the conditions of high light levels, temperature, and drought. As a result, the WUE of crops, especially maize, increases as drought intensifies.

4. Discussion

4.1. Driving Forces of Changes in Drought in Yellow River Basin

Droughts in the Yellow River Basin are influenced by a complex interplay of natural and anthropogenic factors, including geographical location, topography, atmospheric circulation patterns, vegetation cover, global climate change, and human activities [11,66]. In this study, the regions with low VMFDI values in the Yellow River Basin showed a clear seasonal migration from south to north and then south again. And the basin became drier in winter and wetter in the other three seasons from 2003 to 2020 (Figure 7 and Figure 8). This seasonal variation was strongly driven by changes in key climatic and environmental factors such as VPD, SM, and SIF, all of which were significantly correlated with the VMFDI and demonstrated strong causal relationships (Figure 6a and Figure 11a). Among these factors, SM played the most significant role in driving temporal and spatial variations in drought intensity within the basin (Figure 11a). As demonstrated in the Results Section, a reduction in SM directly influenced vegetation health, as evidenced by decreases in key ecosystem indicators such as the NDVI, GPP, WUE, and LAI (Figure 6a). This supports the hypothesis set forth in the Introduction regarding the impacts of soil moisture deficits on ecosystem productivity. Notably, the Liang–Kleeman information flow analysis revealed that changes in the VMFDI had a stronger causal influence on these vegetation indices than the reverse, indicating that drought conditions, driven by decreasing SM, exert a significant negative impact on crop and vegetation productivity (Figure 11a). This finding aligns with the conclusions of Liu et al. [24]. Evidence for vegetation growth was limited when SM was lower than the threshold, which varied among different vegetation types [67]. Furthermore, the results indicate that the decreasing VMFDI resulted in a higher VPD and LST but a lower SIF, PRE, and ET (Figure 6a and Figure 11a). These relationships suggest that as drought conditions intensify, atmospheric conditions such as VPD and LST increase, which further exacerbate the drying effects on vegetation. Conversely, the reduction in SIF, PRE, and ET under drought conditions highlights the limitations on photosynthetic activity and water availability in the ecosystem. In terms of the underlying driving forces of drought in the Yellow River Basin, atmospheric circulation is also one nuclear factor causing drought seasonality and annual movement, the anomalies of which generally lead to spatiotemporal variations in precipitation. The Yellow River Basin is located in a transitional climate zone between arid, semiarid, and sub-humid regions, which makes it particularly susceptible to variations in atmospheric circulation. The basin is influenced by both the East Asian monsoon and the mid-latitude westerly belt, making precipitation patterns highly sensitive to shifts in monsoonal activity [68]. Additionally, circulation factors—the Arctic Oscillation (AO), El Niño–Southern Oscillation (ENSO), and Pacific Decadal Oscillation (PDO)—significantly impact drought patterns in the Yellow River Basin [66,69,70]. For instance, the AO affects the transport of warm and wet water vapor through activities such as the Siberian high, westerlies, and Rossby waves, thereby affecting winter precipitation in the basin [69]. A positive phase of the winter AO has been shown to be associated with a decrease in winter precipitation in northern East Asia [71]. The variation in land–sea thermal differences during ENSO led to the East Asian monsoon transition and further affected precipitation in the basin. A strong correlation between rainy season precipitation in the Yellow River Basin and sea surface temperature in Niño regions has also been reported [72]. The PDO, by weakening the East Asian monsoon during certain phases, also has implications for regional drought dynamics [73]. Above all, spatiotemporal changes in precipitation and temperature induced regional SM changes; therefore, drought occurred, resulting in the migration of the basin. Natural factors play a central role in shaping the intensity and distribution of drought conditions, while land use practices and human interventions further influence the region’s vulnerability to drought. Understanding these driving forces is crucial for improving drought forecasting and management strategies, particularly in vulnerable agroecosystems, which are sensitive to fluctuations in soil moisture and VPD.

4.2. VPD Thresholds Affecting Crop Growth

VPD is a critical driver of atmospheric water demand for crops, and elevated VPD levels are known to adversely affect crop growth and yields [30,74]. In the agroecosystem of the Yellow River Basin, a significant correlation was found between the VMFDI and both SIF and SM (Figure 9a). Furthermore, causal analysis revealed a stronger influence of the VMFDI on SIF and SM than the reverse (Figure 11b), indicating that reductions in the VMFDI are associated with declines in both SIF and SM. Additionally, the NDVI, GPP, and LAI were significantly affected by the VMFDI (Figure 10 and Figure 11b). The VMFDI, in turn, was primarily driven by variations in VPD, with a significant causal relationship observed between VPD and the VMFDI (${|\mathrm{T}}_{\mathrm{V}\mathrm{P}\mathrm{D}\to \mathrm{V}\mathrm{M}\mathrm{F}\mathrm{D}\mathrm{I}}|>{|\mathrm{T}}_{\mathrm{V}\mathrm{M}\mathrm{F}\mathrm{D}\mathrm{I}\to \mathrm{V}\mathrm{P}\mathrm{D}}|$) (Figure 9a and Figure 11b). In this study, VPD-induced drought significantly affected the NDVI, GPP, and LAI of the agroecosystem. Importantly, a critical threshold was identified in VPD, beyond which the NDVI, GPP, and LAI shifted from positive to negative responses (Figure 12a). The VPD threshold varied across crops, ranging from 7.11 to 7.17 ha in the Yellow River Basin agroecosystem. While, in maize and wheat planting systems, the thresholds were 8.03 to 8.57 ha and 7.15 ha, respectively (Figure 12a). These findings suggest that a modest increase in VPD may enhance photosynthesis in vegetation by increasing transpiration and CO₂ uptake, but excessive VPD could weaken stomatal conductance and photosynthesis, even leading to leaf rolling and further limiting plant growth [75]. Notably, maize, as a C₄ plant, exhibits higher tolerance to heat and drought compared to wheat, which is reflected in the higher VPD thresholds for maize.

4.3. Suggestions for Drought Risk Management in Agroecosystems: Insights from VMFDI and VPD Thresholds

Based on the findings of this study, several strategies are recommended for enhancing drought risk management in agroecosystems. First, integrating the VMFDI into agricultural drought monitoring and assessment systems is crucial. The automated joint monitoring of VPD, SM, and SIF throughout the growing season could provide real-time insights into drought conditions. The VMFDI could serve as a key indicator in visual drought monitoring platforms for smart agriculture. This could improve the early detection of drought conditions and inform decision-making in real time. Second, sustainable irrigation schemes should be designed and implemented, considering the critical thresholds for SM and VPD. Tailoring irrigation schedules to these thresholds is crucial for optimizing water use, reducing water wastage, and improving crop resilience to water scarcity. For instance, irrigation schedules could be adjusted dynamically based on real-time VPD and SM data, allowing farmers to respond proactively to changing environmental conditions and minimize the adverse effects of drought. Third, crop planting layouts should be optimized by considering the VPD and SM thresholds for different crops, alongside regional climate and geographical conditions. Optimizing crop selection and planting dates based on these thresholds can improve drought adaptation and resource utilization. For instance, drought-tolerant crops, such as maize, may be more suitable in regions with higher VPD thresholds, enhancing overall productivity and sustainability. Finally, integrating climate projections into long-term agricultural planning is essential. Forecasting changes in VMFDI patterns and future drought risks would enable the development of adaptive management practices, ensuring the long-term resilience of agroecosystems under climate change.

4.4. Limitations of VMFDI

The VMFDI, with a 1 km resolution, effectively monitors regional drought, particularly in agroecosystems, and is more sensitive than the SPEI, GRACE-DSI, and individual indicators such as VPD, SM, and SIF. However, it has some limitations. The GOSIF downscaling product used in this study lacks high spatial resolution and long-term coverage, which may introduce uncertainties. Future work will refine downscaling techniques, such as the use of machine learning, to minimize these effects. Additionally, while the VMFDI was applied at the monthly and annual scales in this study, future research could explore multiscale assessments to enhance decision support for agriculture and water management. A further validation of the VMFDI across diverse regions and agroecosystems is also needed. Ultimately, these limitations can be addressed by improving data quality and optimizing the model. As remote sensing technologies advance, the VMFDI has the potential for more accurate drought monitoring across larger regions and over extended time periods.

5. Conclusions

This study presents the development and application of the VMFDI, a novel drought index designed to provide a high-spatial-resolution (1 km) and monthly temporal-scale monitoring of drought conditions in the Yellow River Basin. By integrating a three-dimensional spatial distance model with key drought-related factors (VPD, SM, and SIF), this study offers a comprehensive tool for assessing regional drought dynamics, particularly in agroecosystems. The key findings of this study are as follows: (1) The VMFDI demonstrates superior sensitivity and accuracy in detecting regional drought conditions compared to existing drought indices such as the SPEI, GRACE-DSI, and individual drought indicators (VPD, SM, and SIF). Its 1 km spatial resolution and monthly scale allow for a more detailed and timely monitoring of ecosystem drought, making it a valuable tool for both research and practical applications in drought risk management. (2) An analysis of the VMFDI series reveals significant spatiotemporal variability in drought severity across the Yellow River Basin from 2003 to 2020. Seasonal trends indicate that the basin experienced drier conditions in winter and wetter conditions in spring, summer, and autumn. The areas most affected by drought are in the middle and lower reaches of the river, with seasonal shifts in the drought center observed in Shaanxi and Gansu. These findings highlight the dynamic nature of drought in the region and the importance of localized monitoring. (3) In the agroecosystem, the VMFDI effectively captures the negative impacts of drought on plant productivity, including significant reductions in the NDVI, GPP, WUE, and LAI (p < 0.05). The Liang–Kleeman information flow analysis reveals that the basin and agroecosystem experienced SM-induced drought and VPD-induced drought, respectively. This study also identifies critical VPD thresholds that trigger changes in crop responses, particularly in maize and wheat systems. This information is crucial for developing targeted drought mitigation strategies to minimize crop yield losses and realize water-saving in agricultural systems. Future research could focus on refining data downscaling methods and further validating the VMFDI across diverse regions and agroecosystems. Overall, the VMFDI represents a significant advancement in drought monitoring, offering a robust and high-resolution approach for understanding regional drought dynamics. And this study offers a framework for more precise drought risk assessment by identifying key drought drivers and response thresholds, which can inform the development of early warning systems, drought mitigation strategies, and water-saving agricultural practices.