Cumulative and Lagged Effects of Drought on the Phenology of Different Vegetation Types in East Asia, 2001–2020

Abstract

Drought disturbances are becoming more frequent with global warming. Accurately assessing the regulatory effect of drought on vegetation phenology is key to understanding terrestrial ecosystem response mechanisms in the context of climate change. Previous studies on cumulative and lagged effects of drought on vegetation growth have mostly focused on a single vegetation type or the overall vegetation NDVI, overlooking the possible influence of different adaptation strategies of different vegetation types and differences in drought effects on different phenological nodes. This study investigates the cumulative and lagged effects of drought on vegetation phenology across a region of East Asia from 2001 to 2020 using NDVI data and the Standardized Precipitation Evapotranspiration Index (SPEI). We analyzed the start of growing season (SOS) and end of growing season (EOS) responses to drought across four vegetation types: deciduous needleleaf forests (DNFs), deciduous broadleaf forests (DBFs), shrublands, and grasslands. Results reveal contrasting phenological responses: drought delayed SOS in grasslands through a “drought escape” strategy but advanced SOS in forests and shrublands. All vegetation types showed earlier EOS under drought stress. Cumulative drought effects were strongest on DNFs, SOS, and shrubland SOS, while lagged effects dominated DBFs and grassland SOS. Drought impacts varied with moisture conditions: they were stronger in dry regions for SOS but more pronounced in humid areas for EOS. By confirming that drought effects vary by vegetation type and phenology node, these findings enhance our understanding of vegetation adaptation strategies and ecosystem responses to climate stress.

1. Introduction

The impacts of global warming on ecosystems are far-reaching and complex. Changes in vegetation phenology (the timing of cyclical events in the life cycle of plants) are an important indicator of how vegetation responds to a changing environment. Vegetation phenology in turn plays a key role in terrestrial ecosystem dynamics and in guiding agricultural production [1]. Many studies have confirmed that vegetation phenology is influenced by meteorological factors such as temperature, radiation, and precipitation [2,3]. These influences vary among vegetation types, such as forests and grasslands [4,5,6,7,8].

Droughts—periods of extreme water scarcity—are expected to increase as the climate warms [9,10,11]. The cascading effects of drought can alter the structure and function of ecosystems. In recent decades, drought disturbances have been occurring more frequently; the area affected by drought has increased; and the impacts of drought on terrestrial ecosystems, biodiversity, and agricultural production are increasingly becoming a focus of global concern [12,13,14,15]. Drought directly modulates the physiological processes of plants by altering water availability, which in turn reshapes vegetation phenology, but these effects vary among the phenological nodes of different vegetation types [16,17]. Studies have implicated drought conditions in either accelerating [18] or delaying [19] the start of the vegetation growing season (start of season, or SOS) in different settings. Vegetation spring phenology is also constrained by background climatic conditions such as spring temperature and precipitation [20]. Winter moisture and temperature conditions at high latitudes influence the sensitivity of vegetation SOS to spring drought [21]. For shrubs, it was found that drought led to an earlier SOS, thereby extending the growing season [22]. However, in a study on Canadian grasslands, it was found that drought led to an earlier termination of the vegetation growing season (end of season, or EOS) [23]. It appears that mild drought can accelerate germination by triggering plants’ water stress response mechanisms, but extreme drought events can significantly delay both SOS and EOS [24,25].

Beyond the immediate effects of drought on vegetation phenology, cumulative or lagged drought effects can affect the vegetation performance in subsequent growing seasons. For example, drought may cause long-term damage to plant physiology, such as drought-induced root damage and xylem embolism, which remain after the drought is over and affect subsequent water uptake and transport [26,27,28]. Furthermore, vegetation consumes large amounts of non-structural carbohydrates (NSCs) during periods of drought, leading to a reduction in resource reserves and affecting subsequent growth and resilience [29,30]. Studies have increasingly pointed to the importance of both cumulative effects (potentially from multiple droughts over time) and lagged effects of previous droughts on vegetation phenology [31,32,33,34]. Specifically, cumulative effects involve the persistent impacts of drought that are gradually superimposed on the vegetation growth cycle during the year, while lagged effects manifest as the influence of previous periods of drought on the current characteristics of vegetation at a specific point in time [35]. These reflect two aspects of the legacy effects of water shortages on vegetation that can be highly variable across taxa in their lag time and manifestation [36]. However, previous studies on the cumulative and lagged effects of drought on vegetation growth have mostly focused on single vegetation types or on the overall vegetation NDVI, coverage, and productivity [35,37,38,39,40], overlooking possible differential responses to drought among vegetation phenological nodes and different vegetation types, such as forests composed of trees with different leaf forms (broadleaf and needleleaf), shrubs, and grasslands.

Different vegetation types have adopted different survival strategies (such as leaf form) and thus may display varying effects of drought on different phenological nodes (i.e., SOS and EOS). For example, the resource utilization strategies of trees are adapted to local thermal and moisture conditions [41,42]. Trees may be broadly categorized into four main types based on leaf form (needleleaf or broadleaf) and habit (evergreen or deciduous) [43]. These types represent different kinds of trade-offs between growth and survival [41]. In response to variations in water conditions, trees have developed different leaf forms and thus different growing season water use strategies: needle leaves often display a conservative adaptation strategy with cold- and drought-resistant structures as an adaptation to physiological drought, short growing seasons, and low temperatures [44], while broad leaves are less drought resistant, being distributed mainly in warmer and more humid climates [45]. Shrubs have longer leaf lifespan and higher biomass per unit leaf area for cold and drought resistance [41,46], and grasses generally exhibit a summer dormancy strategy that enhances their dehydration tolerance [47].

East Asia has unique climatic conditions and diverse vegetation types. In recent years, the severity of droughts in East Asia has shown a significant upward trend, and there is considerable unpredictability regarding changes in regional precipitation patterns accompanying global warming [48]. Frequent heatwaves and severe droughts occurring simultaneously in the region have exacerbated soil dryness, causing a detectable impact on vegetation growth [49]. Therefore, it is essential to conduct research on the relationship between drought and vegetation growth patterns, not only during the period of drought itself but also after the drought is alleviated.

Climate warming is changing global ecosystems by reshaping vegetation phenology rhythms [50]. Such changes not only threaten biodiversity but may also exacerbate the climate stress through carbon–water cycle feedbacks [51]. In a warming world, there is a need to better understand the cumulative and lagged effects of drought on the phenology of forests, shrubs, and grasslands, and among forests with different leaf forms. This study investigated the pattern of response to drought among different vegetation types in East Asia from 2001 to 2020. The objectives of this study were to determine (1) whether there were difference in response of phenology to cumulative and lagged effects of drought; (2) differences in effect of cumulative and lagged effects of drought on different phenological nodes; (3) differential effects of background moisture conditions on cumulative and lagged effects of drought; and (4) differences in dominant drought effects among shrublands, grasslands, and forests of different leaf forms in the study area. Understanding the mechanisms of drought-induced phenological changes among broadleaf and needleleaf forests, shrublands, and grasslands will be essential for better management of vegetation to maintain vital ecological functions.

2. Materials and Methods

2.1. Study Area

This study focuses on a region in the middle to high latitudes of northeast Asia (longitude 90–135°E, latitude 40–55°N), including portions of northern China and eastern Russia as well as most of Mongolia and a small part of the northern Korean peninsula [31]. The major geographical features of the study area include the Northeast China Plain, the Mongolian Plateau, the Greater and Lesser Khingan Mountains, and the Changbai Mountains (Figure 1a,b). The area’s topography descends from west to east. Precipitation varies seasonally and interannually with the summer and winter monsoons; moisture conditions range from semi-humid in the east to semi-arid and arid in the west, while thermal conditions range from the warm temperate zone in the south to the cold temperate zone in the north.

Within the region, the main vegetation types consist of deciduous needleleaf forests (DNFs), deciduous broadleaf forests (DBFs), shrublands, and grasslands. DNFs are located in the northern parts of the Greater Khingan Mountains and southern and southeastern Siberia; DBFs are mainly found in northeast China; shrublands occur in southern Siberia and the Greater Khingan Mountains; and grasslands cover most of the Inner Mongolian Plateau in China as well as northern Mongolia (Figure 1c). Each of the vegetation types has different water use strategies during the growing season, such as leaf stomatal conductance and hydraulic safety margins, which may influence their phenological response to drought. (Other types of land cover, such as deserts and wetlands, are excluded from the analysis).

2.2. Datasets

2.2.1. Vegetation Type Data and Processing

We use the International Geosphere-Biosphere Program (IGBP) MODIS land cover classification scheme. MODIS land cover data were obtained from the MCD12Q1 data product (https://ladsweb.modaps.eosdis.nasa.gov (accessed on 1 November 2024)) with a spatial resolution of 500 m. Based on the main leaf form and habit of trees in the study area, vegetation types in the study area were reclassified into four categories: DNFs, DBFs, shrubland, and grassland. In order to minimize the disturbance of vegetation type changes, we extracted the area where vegetation types did not change during the study period [52,53] to use as the basis for the vegetation types in our analysis (Figure 1c). For reference, we also obtained STRM15+ V2.5.5 global topography data at a 15 arc-second resolution from the General Bathymetric Chart of the Oceans (https://gebco.net (accessed on 15 April 2025)).

2.2.2. SPEI Data Source and Processing

The standardized precipitation evapotranspiration index (SPEI) constructed by Vicente-Serrano et al. [54] can reflect regional drought situations in a more comprehensive way by considering the distribution of both potential evapotranspiration and precipitation [55]. The global SPEI dataset, SPEIbase v.2.8 (https://spei.csic.es/database.html (accessed on 1 October 2024)), identifies drought conditions at a time scale from 1 to 48 months and with a spatial resolution of 0.05° (approximately 3.5–5.5 km). We selected SPEI data for 1- to 12-month time scales to quantify drought occurrence in terms of intensity and length. We resampled the SPEI grids using a bilinear interpolation method to match the 500 m spatial resolution of the MODIS data [31,56,57,58,59].

2.3. Methods

2.3.1. Detecting Vegetation Phenology

The Normalized Difference Vegetation Index (NDVI) quantifies vegetation by measuring the difference between the reflected near-infrared and red light detected by remote sensors [60]. This index can effectively reflect vegetation activity and is the most commonly used indicator of the robustness of vegetation cover [61]. As the NDVI is highly sensitive to moisture conditions, it can be used to assess the drought status of vegetation, enabling a comparison with the SPEI drought index to reveal the relationship between vegetation activity and drought conditions. We use NDVI data for the period between January 2001 and December 2020 obtained from NASA MOD13A1 (https://ladsweb.modaps.eosdis.nasa.gov (accessed on 1 November 2024)), with a spatial resolution of 500 m and temporal resolution of 16 days [62].

Using MATLAB R2024a software, we applied the Savitzky–Golay filtering method to smooth the NDVI data, setting the window length at 9 and the smoothing polynomial degree at 3. This method effectively filled in the data gaps between the 16-day intervals, reconstructing the time series and removing the effects of outliers caused by clouds, aerosols, sensor noise, or other background noise [63]. We then calculated the dates of the vegetation phenological nodes, including the start of the growing season (SOS) and the end of the growing season (EOS), using the dynamic thresholding method [64]: based on the NDVI time series characteristics of the study area and prior research [65,66,67,68], we set the date when the NDVI rises to 30% of the annual NDVI amplitude as the SOS and set the date when the NDVI falls to 50% of the annual NDVI amplitude as the EOS. The workflow of this study is shown in Figure 2.

2.3.2. Trend Analysis

This study uses the non-parametric Sen’s slope and Mann–Kendall methods to calculate trend [69]. We implemented this equation in the R 4.4.1 programming language to analyze the regional-scale and pixel-scale SOS and EOS trends:

$$\mathrm{S}\mathrm{l}\mathrm{o}\mathrm{p}\mathrm{e}=\mathrm{M}\mathrm{e}\mathrm{d}\mathrm{i}\mathrm{a}\mathrm{n}[{\displaystyle \frac{{\mathrm{X}}_{\mathrm{j}}-{\mathrm{X}}_{\mathrm{i}}}{\mathrm{j}-\mathrm{i}}}],\forall 1 \le \mathrm{i} \le \mathrm{j} \le \mathrm{n}$$

(1)

where x_i and x_j represent the values at times i and j, and n represents the length of the time series data. When the slope is positive, it indicates that SOS or EOS has been delayed; when the slope is negative, it indicates that SOS or EOS has advanced.

2.3.3. Detecting Cumulative and Lagged Effects of Drought

As the basis for comparing the cumulative and lagged effects of drought in terms of monthly time scales, we need to determine the starting months for SOS and EOS. If the average SOS date falls in the first half of the month, then we designate the previous month as the starting SOS month; if it falls in the second half of the month, then the current month is designated as the starting SOS month [70]. The same logic applies to the EOS month. For the vast majority of our study area from 2001 to 2020, the SOS falls in April and the EOS falls in September, so those are designated as the SOS and EOS starting months for comparison with the 1- to 12-month cumulative and lagged SPEI values [65].

Before performing a correlation analysis, we used a linearity test in SPSS 19.0 to verify that there is a linear relationship between phenology nodes and SPEI. Taking the phenology parameters and the cumulative SPEI (1 to 12 months) [32,71], we computed the Pearson correlation coefficient (r) to test the cumulative effect of drought on the phenology of different vegetation types in the study area [57,72,73]. Then, we used correlation analysis to calculate the correlation coefficients between vegetation phenology and the 1- to 12-month SPEI values, generating 12 correlation coefficients for both SOS and EOS from 2001 to 2020. We calculated the cumulative effect and flagged the cumulative month with the largest absolute value of the correlation coefficient to identify the time scale of the cumulative effect [53,73], as follows:

$${\mathrm{r}}_{\mathrm{i}}=\mathrm{c}\mathrm{o}\mathrm{r}\mathrm{r}(\mathrm{x},{\mathrm{y}}_{\mathrm{i}})1\le \mathrm{i}\le 12$$

(2)

$${\mathrm{R}}_{\mathrm{m}\mathrm{a}\mathrm{x}\_\mathrm{c}\mathrm{u}\mathrm{m}}=\mathrm{m}\mathrm{a}\mathrm{x}\left(\right|{\mathrm{r}}_{\mathrm{i}}\left|\right)1\le \mathrm{i}\le 12$$

(3)

where r_i is correlation coefficient between vegetation phenology and the SPEI for each cumulative month; x denotes the SOS or EOS; y_i denotes the SPEI for cumulative month i; R_{max_cum} is the maximum value of r_i; and the cumulative response time is determined to be months corresponding with R_{max_cum}.

The Pearson correlation coefficient (r) is used to test the lagged effect of drought on phenological parameters and 1-month SPEI₀₁. Twelve correlation coefficients were generated by combining the 2001–2020 SOS and EOS time series with previous j-month SPEI series and calculating correlation coefficients between vegetation phenology and 1-month SPEI at different lagged times (1 $\le \mathrm{j}\le$ 12). R_{max_lag} is the magnitude of the lagged effect and the month with the largest absolute value of the correlation coefficient is used as the time scale of the lagged effect. The drought lagged effect is calculated as follows:

$${\mathrm{r}}_{\mathrm{j}}=\mathrm{c}\mathrm{o}\mathrm{r}\mathrm{r}(\mathrm{x},{\mathrm{y}}_{\mathrm{j}})1\le \mathrm{j}\le 12$$

(4)

$${\mathrm{R}}_{\mathrm{m}\mathrm{a}\mathrm{x}\_\mathrm{l}\mathrm{a}\mathrm{g}}=\mathrm{m}\mathrm{a}\mathrm{x}\left(\right|{\mathrm{r}}_{\mathrm{j}}\left|\right)1\le \mathrm{j}\le 12$$

(5)

where r_j is the Pearson correlation coefficient between vegetation phenology and the SPEI for each lagged month; x denotes the SOS or EOS; y_j denotes the SPEI for lagged j months; R_{max_lag} is the maximum value of r_j; and the lagged response time is determined to be the time corresponding with R_{max_lag}.

In addition, we used one-way analysis of variance to test the significance of differences for R_max between different vegetation types. To assess drought’s cumulative and lagged effects on vegetation phenology across moisture conditions, this study selected SPEI₁₂ of December between 2001 and 2020 to represent annual average water conditions. The maximum correlation coefficient (R_{max_cum}, R_{max_lag}) and cumulative and lagged months were used to analyze changes in the effects of drought on vegetation phenology under different moisture conditions. The moisture gradient was determined by using the annual average SPEI at 0.1 intervals. To clarify the relative strengths of these correlations in terms of the areal extent affected by drought, we calculated the difference in magnitude between R_{max_cum} and R_{max_lag}:

∆R_max = |R_{max_cum}| − |R_{max_lag}|

(6)

where positive values indicate that the cumulative effect was more strongly correlated and negative values indicate that the lagged effect was more strongly correlated [31,53,74].

3. Results

3.1. Phenology Changes by Vegetation Type

The spatial distributions of multiyear mean values of vegetation phenology in the study area from 2001 to 2020 are shown in Figure 3. The SOS was mainly concentrated between the 90th and 130th day of the year (DOY), that is, from early April to late May, accounting for 84.44% of the study area. From east to west, the SOS appeared earlier in eastern China and later in the northwest and southwest of the study area. Among the vegetation types, DNF, DBF, and shrub SOS were concentrated at DOY 90–125, while the grassland SOS was concentrated at DOY 95–150 (Figure 3a). Vegetation SOS in the northeast China, including the Greater Khingan Mountains, Lesser Khingan Mountains, and Changbai Mountains, showed a trend of advancement, with the DBF and eastern grassland SOS significantly advancing. The Khangai and Khentii Mountains in the northern part of the Mongolian Plateau and part of southwestern Inner Mongolia showed a trend of postponement, with the SOS significantly delayed mainly in the grasslands there. As shown in Figure 3c,e, the SOS advanced in 89.43% of the study area (13.32% passed a significance test, p < 0.05) while the SOS was delayed in 10.57% of the study area (1.50% significantly, p < 0.05).

The time range of the EOS was from early September to late October, occurring from DOY 250 to 280 in 92.52% of the study area. The earliest EOS mainly appeared in the northwest and the latest EOS was concentrated in the southeast. The EOS of DNFs and DBFs was concentrated at DOY 257–287, the EOS of shrublands was concentrated at DOY 250–290, and the EOS of grassland was concentrated at DOY 243–275 (Figure 3b). Most of the study area showed a trend toward a delay in the EOS (91.94%, with 14.72% significant), especially in DBFs and a small portion of grasslands. An earlier (advancing) EOS was detected in parts of Inner Mongolia Plateau and a small portion of northwestern Mongolia (8.07% of the study area, with 1.37% significant) (see Figure 3d,f). Across the study area, grassland areas displayed both delays and advancements in SOS and EOS, while the other vegetation types saw more consistent trends of an advancing SOS and delayed EOS. In terms of the magnitude of change, advancements in the SOS and delays in the EOS were both higher in the southern ranges of the grasslands and DBFs.

3.2. Cumulative Effects of Drought on Phenology by Vegetation Type

3.2.1. Spatial Characteristics of Cumulative Effects

Figure 4a,b shows the maximum correlation coefficient R_{max_cum} between the SOS and SPEI and the corresponding cumulative months, illustrating the spatial characteristics of the cumulative effect of drought on the vegetation SOS. Significant positive or negative R_{max_cum} values indicate where phenology was affected by drought. The magnitude of the correlation coefficient indicates the extent to which phenology was affected by drought. The corresponding time scale indicates the response time of the sensitivity of phenology to drought [32].

The effects on the SOS of cumulative drought varied considerably in different regions. Positive significant correlations, indicating where drought induced advances in the SOS, accounted for a substantial percentage of the study area, around 32.55%, mainly in the Greater Khingan Mountains, the eastern part of Northeast China, the Khangai Mountains, and parts of southern Siberia. These are mostly areas with DNFs, DBFs, and shrub vegetation types. Negative significant correlations, indicating that drought contributed to a delayed SOS, appeared in 12.14% of the study area, mainly grasslands, including the western part of the Inner Mongolia Plateau. Drought led to a delayed SOS in these grassland ecosystems, counter to the trend seen in DNFs, DBFs, and shrublands (Figure 4a). There are significant differences between different vegetation types in the extent of the SOS affected by drought (Figure 5a). The distribution of cumulative months, indicating the length of the response time, showed no obvious difference among the time scale of 1 to 12 months, with a peak occurring at a length of 2 months (12.78%), followed by a length of 9 months (11.66%) (Figure 4b).

For the EOS, R_{max_cum} showed positive significant correlations most obviously in the south—opposite to the spatial distribution of the SOS R_{max_cum} (Figure 4c). Significantly positively correlated areas, indicating where drought induced advances in the EOS, accounted for 32.46% of the vegetation area, mainly in the southern part of northeastern China and the east-central part of Inner Mongolia, including areas of DNFs, DBFs, grassland, and shrub vegetation types. The EOS R_{max_cum} displayed negative significant correlations, mainly in the south-central Greater Khingan Mountains and parts of southern Siberia, accounting for only 3.32% of the study area, predominantly DNFs. There were significant differences in the extent of the EOS affected by drought between DNFs and grasslands, as well as between DBFs and shrubs, but not between DNFs and shrubs or between DBFs and grasslands (Figure 5b). Time scales corresponding to the appearance of maximum correlation coefficients of the EOS vary across the study area, generally with longer response time scales in the east (Figure 4d). The peak cumulative months were mostly lower than for the SOS, with the 1-month scale having the largest proportion of affected area for the EOS (28.36%), followed by the 2-month scale (19.08%).

Overall, the area significantly affected by the R_{max_cum} cumulative drought accounted for 44.69% of the vegetation distribution area for the SOS and 35.69% for the EOS. Results showed that the drought’s cumulative effect on vegetation phenology showed different response times, extents, and sensitivities for different vegetation types and different phenology nodes.

3.2.2. Time-Scale Characteristics of Cumulative Effects

As seen in Figure 6, there were obvious differences in the drought effects as superimposed over multiple time scales on the phenology of different vegetation types. (Bar graph results were extracted from the analysis of correlation between phenology and the SPEI at 12 cumulative time scales before selecting the maximum R_{max_cum} [75,76]). The effect of drought on the DNF SOS was significant across most of the vegetation area (80.76% positive, 0.39% negative), with the peak occurring at 6 months of cumulative drought (69.52%) (Figure 6a). The drought effect on the DBF SOS was somewhat weaker, with 42.74% displaying significant correlations and the peak occurring at 2 months of cumulative drought (27.79%) (Figure 6b). The effect of drought on the shrub SOS was significant for 62.29% of its area, peaking at a cumulative 3 months (44.81%) and showing high sensitivity to drought at short time scales (Figure 6c). The effect of drought on grassland SOS achieved significance in 40.56% of that area, with a weak peak at a cumulative 7 months (14.87%) (Figure 6d). The positive effect of drought on the SOS in DNFs, DBFs, and shrublands was greater than the negative effect based on the proportions with significant correlations.

For EOS, the percentage of DNF area with significant correlations between cumulative drought and the EOS was small, only 13.76%, with little variation at different cumulative time scales (Figure 6e). By comparison, cumulative drought had a stronger effect on the DBF EOS (35.68%); the cumulative time scale was concentrated in the range of 6–12 months, with the peak occurring at 12 months (25.39%), indicating that prolonged drought had a stronger effect on the DBF EOS (Figure 6f). The effect of drought on the shrub EOS was relatively weak, with significant correlations in 21.42% of the distribution area and a peak occurring at 8 months (9.56%) (Figure 6g). The effect of drought on the grassland EOS is relatively stronger, with significant correlations in 39.86% of the distribution area and the peak occurring at 1 month (26.37%), indicating that grasslands were more susceptible to short-term drought stress (Figure 6h).

Overall, slightly more grassland area had significant negative correlations than significant positive correlations for the SOS, while significant positive correlations dominated in DNFs, DBFs, and shrublands. Cumulative drought had the greatest effect on the SOS in DNFs and the second in shrubs; cumulative drought had the greatest effect on the EOS in grasslands and the second in DBFs. The time scale of drought’s effect on grassland was shorter for the EOS, while the opposite was true for DBFs. These differences in the response to drought of the phenological nodes may relate to the adaptation strategies of different vegetation types, reflecting their physiological characteristics, water use efficiency, or preferred moisture and temperature conditions at different stages of growth and development.

3.2.3. Cumulative Effects Across Moisture Conditions

Figure 7 shows how peak drought’s cumulative months and R_{max_cum} means vary with mean annual moisture conditions for the SOS and EOS of different vegetation types. For the SOS, drought’s cumulative month values for DNFs, DBFs, and shrublands were lower in relatively dry areas and higher in relatively humid areas (Figure 7a–c). R⁺_{max_cum} (R_max > 0) was significantly negatively correlated with the annual mean SPEI for DBFs and shrublands (Figure 7f,g). The opposite was true for grasslands, where the drought’s cumulative months gradually increased with decreasing moisture (Figure 7d). R⁻_{max_cum} (R_max < 0) was significantly positively correlated with the annual mean SPEI for grasslands (R² = 0.602), with a greater effect in arid areas than in wet areas (Figure 7h).

For the EOS, drought’s cumulative months of DNFs and DBFs displayed nonlinear relationships with the annual mean SPEI. Drought cumulative months increase and then decrease with decreasing moisture (Figure 7i,j). Both shrub and grassland drought’s cumulative months were significantly positively correlated with the annual mean SPEI, suggesting that the shrub and grassland EOS respond faster to the cumulative effects of drought in dry regions and that increased water loss shortens the response time to drought. As shown in Figure 7m–p, for DNFs and DBFs, the annual average SPEI was significantly positively correlated with R⁺_{max_cum}, but only DNFs’ annual average SPEI was significantly correlated with R⁻_{max_cum}. Grasslands, by contrast, saw significantly negative correlations with R⁺_{max_cum}, meaning that the cumulative effects of drought had a much smaller effect in areas with greater moisture. These variations reflected the adaptive adjustment of different vegetation types to drought under different moisture conditions, reflecting the complex relationship between vegetation and moisture dynamics.

3.3. Lagged Effects of Drought on Phenology by Vegetation Type

3.3.1. Spatial Characteristics of Lagged Effects

Figure 8 illustrates the spatial variation in R_{max_lag}, the highest correlation coefficient between the SOS or EOS and monthly SPEI among the previous 12 months, and the corresponding lagged month. The distribution of the drought lagged effect on the SOS showed obvious spatial differences, with significant positive correlations accounting for 26.54% of the total area, mainly in eastern China and parts of southern Siberia, areas dominated by DNFs, DBFs, and shrublands. Significant negative correlations were mainly distributed in the Mongolian Plateau and western parts of Inner Mongolia, accounting for 23.19% of the total area, mostly grasslands (Figure 8a). There were no significant differences in the extent of SOS affected by drought between DNFs and shrubs in the study area, but there were significant differences between other vegetation types (Figure 9a). The largest percentage of drought impacts (14.32%) could be attributed to a lag of 1 month, followed by a lag of 10 months (10.70%) (Figure 8b).

The drought lagged effect on the EOS also showed spatial differences, with 28.86% of the total area displaying significant positive correlations, concentrated in northeast China and across all four vegetation types. Significant negative correlations were found in central and western parts of Mongolia, accounting for 15.83% of the total area, mainly grasslands (Figure 8c). There are significant differences in the extent of the EOS affected by drought between different vegetation types, but no significant differences between DNFs and shrubs (Figure 9b). The drought effect of the first previous month of the EOS was the most widespread, with 13.61%, while the previous 5 months were the next most widespread, accounting for 10.48% of the total area (Figure 8d). Considering both positive and negative lagged effects, correlations were significant for 44.70% of the total area for the EOS, compared with 49.73% for the SOS. The variation in the direction and time scale of the lagged effect of drought on the phenology of different vegetation types appears to reflect their differentiated ecological adaptation strategies.

3.3.2. Time-Scale Characteristics of Lagged Effects

Figure 10 displays obvious differences among the lagged effects of drought on different vegetation types and phenological nodes. (The bar graph results were extracted from correlation analysis between phenology and SPEI at 12 lagged time scales before selecting the maximum R_{max_lag}.) For SOS, the lagged effect of drought affected the largest area for DNFs (53.42% of the DNF area displayed a significant correlation), followed by grassland (50.82%), shrubland (50.03%), and then DBFs (47.93%). The lags affecting the DNF, DBF, and shrubland SOS were mainly concentrated in the time scales of 1–2 months, with a peak lagged effect for these three vegetation types occurring in the first month (23.34%, 16.49%, and 19.16% respectively). Thus, except for grassland, the SOS was mainly influenced by the lagged effect of drought in the previous month, and the closer the drought was to the phenological node, the greater the effect of drought. Only in grasslands did the proportion with negative significant correlations exceed that with positive correlations, and the lagged effects on grasslands did not vary much among the different time scales.

For the EOS, the drought lagged effects on all four vegetation types displayed significant correlations on similar proportions of their spatial distribution, ranging from 44.85% for grasslands and 44.64% for DBFs to 44.41% for shrublands and 43.54% for DNFs. The correlations for DNFs, DBFs, and shrubland peaked at lags of 3 months (6.13%), 8 months (8.94%), and 6 months (6.91%), respectively; drought had a more rapid effect on grasslands. peaking at a 1-month lag (11.28%).

3.3.3. Lagged Effects Across Moisture Conditions

Both R_{max_lag} and the corresponding number of lagged months were linked to moisture conditions (Figure 11). We find linear relationships for both the SOS and EOS of DBFs, shrubland, and grassland, but nonlinear relationships for DNFs. The number of drought effect lagged months for the DNF SOS was nonlinearly correlated with the annual mean SPEI (R² = 0.621), first decreasing and then increasing with decreasing moisture (Figure 9a). The lagged months for DBFs, shrubland, and grassland SOS showed clear increases with increasing annual average SPEI, with shrubland having the largest response coefficient (Figure 11b–d). The annual mean SPEI of the four vegetation types was significantly negatively correlated with R⁺_{max_lag} (R_max > 0), and shrubland and grassland were significantly positively correlated with R⁻_{max_lag} (R_max < 0); the correlations were stronger in relatively dry areas than in relatively humid areas (Figure 11e–h).

For the EOS, the number of drought effect lagged months for DNFs was nonlinearly correlated with the annual mean SPEI (R² = 0.314). The number of lagged months increased and then decreased with decreasing moisture conditions (Figure 11i). For DBFs, lagged months did not change obviously with changing moisture conditions. Both shrublands and grasslands displayed obvious increases in the number of lagged months with the increasing annual mean SPEI, meaning that decreases in water conditions shortened the response time to drought (Figure 11k,l). The correlation between the annual average SPEI and R⁺_{max_lag} for all four vegetation types was positive, indicating that drought had a greater impact on the vegetation EOS in wetter areas. The correlation between the annual mean SPEI and R⁻_{max_lag} for DNFs was negative (R² = 0.804) (Figure 11m–p). In areas with more severe drought, moisture conditions show a weakened effect on the EOS.

3.4. Comparison of Lagged and Cumulative Effects

The difference between R_{max_cum} and R_{max_lag} (ΔR_max) indicates whether cumulative or lagged effects had a stronger statistical association with observed phenology. While this measure does not inherently indicate causal dominance, it may suggest avenues for further research. The ΔR_max values for both the SOS and EOS were largely concentrated in a narrow range between −0.15 and +0.15, indicating that the lagged and cumulative effects were close in strength (Figure 12a,b).

For the SOS, the lagged effect dominates 55.67% of the study area, mainly concentrated in eastern China and northern Mongolia; the cumulative effect prevailed in 44.33% of the area, mainly concentrated in the far north, including northern parts of the Greater Khingan Mountains, southern Siberia, and portions of southeast Inner Mongolia. For the EOS, the study area was also dominated by the lagged effect (61.55%), mainly concentrated in the north, including northern Mongolia, the Greater Khingan Mountains, and parts of southern Siberia; the cumulative effect prevailed in 38.45% of the total area, concentrated in the southeast part of northeast China and Inner Mongolia.

For different vegetation types, we found that the DNF SOS was mainly affected by the drought’s cumulative effect in 83.17% of the distribution area, with most of them having ΔR_max values between 0 to 0.3. In contrast, the DBF SOS was more strongly affected by the lagged effect of drought, accounting for 56.47% of the distribution area, with ΔR_max values mostly between −0.15 and 0. Shrublands, like DNFs, were mainly affected by drought’s cumulative effect, with a share of 62.71% of the distribution area and ΔR_max values mostly between 0 and 0.3. Grasslands, similar to DBFs, were strongly affected by drought’s lagged effect, with a share of 60.94% of the distribution area and ΔR_max values mostly between −0.15 and 0 (Figure 12a and Figure 13a). For the EOS, all four vegetation types were dominated by the lagged effect, accounting for 84.08%, 61.10%, 76.03%, and 57.63% of the distribution area of DNFs, DBFs, shrublands, and grasslands, respectively; among them, DNFs and shrublands had greater proportions of ΔR_max values between −0.3 and 0, while most of DBFs and grasslands had ΔR_max values between −0.15 and 0 (Figure 12b and Figure 13b).

4. Discussion

4.1. Drought’s Cumulative and Lagged Effects on the Phenology of Different Vegetation Types

We found that drought has cumulative and lagged effects on vegetation phenology and that there were significant differences in these effects between vegetation types with different trade-off strategies for growth and survival. Drought led to the delayed SOS in grasslands and advanced SOS in DNFs, DBFs, and shrublands in this study. This is in line with previous findings, where pre-seasonal drought events were found to cause the grassland SOS to be delayed and the forest SOS to be advanced in northeast China [77].

For perennial trees and shrubs, the SOS is a key physiological turning point in the transition from dormancy to active growth. Pre-SOS drought can increase temperatures by reducing evaporation energy consumption, thus promoting vegetation’s new leaf germination in spring [18,78]. In addition, deep-rooted forest trees are better able to utilize water in the deep soil to resist the effects of previous droughts on current vegetation growth, further reducing the negative effect of drought on the SOS [79]. Furthermore, precipitation moderates the effect of temperature on phenology in dry regions [80], when temperature is the main driver [6,81,82].

By comparison, grass root systems are shallow, so when drought depletes soil moisture quickly, grasses do not have enough available water for germination [83]. After drought conditions wane, grassland vegetation can respond rapidly to light rainfall events (<10 mm). This fast-growing, resource-accessing vegetation has the advantage of enhancing stability by engaging in drought escape or drought avoidance strategies [17,84,85], rather than the more conservative strategy of drought tolerance [86,87,88]. As a result, the germination of vegetation in grassland ecosystems under drought conditions occurs much later than under normal conditions, resulting in the delayed grassland SOS.

Prior studies proposed that the cumulative effects of drought affect a greater proportion of grassland areas compared with forests [32,89]. To the contrary, in this study, the results for the SOS showed that the percentage of grasslands affected by cumulative drought was lower than the percentage of DNFs affected. An important reason for this is that grassland ecosystems in the study area span arid, semiarid, and semihumid regions [77]—including areas of northern Mongolia and northeastern Inner Mongolia where precipitation is relatively high—and thus moisture is not a major constraint on grassland rejuvenation. In our study area, DNFs are mainly found in eastern Siberia and the Greater Khingan Mountains. DNFs that extend into permafrost regions tend to have shallow root systems and mainly utilize shallow soil moisture from winter snowmelt. With global warming, earlier snowmelt replenishes soil moisture in early spring and reduces the possible negative effects of drought [90]. However, drought can also promote increased temperatures [91], facilitating the earlier development of new leaves. As a result, prolonged cumulative drought has a greater impact on the advancement of the SOS in DNFs.

At the other end of the growing season, this study found that drought leads to an earlier EOS in DNFs, DBFs, shrublands, and grasslands. It has been found that summer and fall droughts significantly advance the vegetation EOS [92], which is consistent with our findings. Drought-induced premature senescence is another mitigation strategy that vegetation can adopt to cope with drought stress. When water stress coincides with natural senescence processes, plants optimize their survival strategies by accelerating senescence [93]. Continued summer warming, lower total precipitation, and higher evapotranspiration all promote vegetation leaf senescence [94]. Fall drought prompts leaves to close stomata, reducing transpiration and photosynthesis rates, but respiration is maintained at higher rates that accelerate carbon assimilation degradation, leading to earlier vegetation senescence [92]. Furthermore, the differences among vegetation types of the cumulative effects of drought are more obvious than among their lagged effects, which may be related to the fact that the cumulative effect builds on and deepens the lagged effect.

4.2. Effects on Phenology of Trees with Different Leaf Forms

This study found significant differences in the effects of drought on trees with different leaf forms (needleleaf or broadleaf). Drought had a greater effect on the SOS in DNFs, and on both the SOS and EOS in DBFs, and the character is more obvious in the cumulative aspect. This may stem from ecological adaptation strategies that tree species evolved to adapt to differential climates with different leaf forms [43,95,96]. These strategies are the main determinants of plants’ biomass allocation patterns [97]. Plant physiological traits and environmental adaptation strategies together shape phenological drought response patterns [5]. For example, conifer species typically exhibit conservative water use strategies [98], with lower stomatal conductance and lower photosynthetic rates [99]. Consequently, the SOS of DNFs reflects its low moisture requirement to initiate new shoot germination and photosynthesis. DBFs generally develop deep root systems, which reduce the negative effect of drought; their evapotranspiration requirements are low in the early stages of leaf development. Thus, drought may benefit the DBF and DNF SOS through the effect on thermal conditions.

At the other end of the growing season, there are obvious differences for trees with different leaf forms. This might be related to how the leaves of DNFs are highly adapted to drought and consume less water for photosynthesis [100], so drought only weakly regulates the processes of maturation to defoliation. In needleleaf forests, the needle-like leaves and conservative water-use strategy of DNFs make the EOS less sensitive to both short-term and long-term drought. By contrast, DBFs have a high leaf area, a high transpiration rate, and a greater water demand during the growing season [43], thus requiring large amounts of water to maintain transpiration and photosynthesis. The “opportunistic strategy” of DBFs tends to prioritize the allocation of limited water and nutrients to the root system or storage tissues rather than maintaining leaf function under drought stress [101,102]. When subjected to drought stress, these trees inhibit transpiration by decreasing stomatal conductance while accelerating leaf senescence signals (e.g., abscisic acid ABA accumulation), leading to chlorophyll degradation [103,104]. This resource redistribution drives early leaf abscission to reduce overall water consumption.

Consequently, this study found that the SOS of DNFs and DBFs are both highly affected by drought, mainly limited by tree water-use strategies and water availability, while the EOS of DBFs is also sensitive to drought, mainly stemming from hormonal regulation and resource redistribution to accelerate leaf senescence. These differences reflect the differentiation in phenological adaptation strategies among different leaf forms and functional types of plants. Future studies should incorporate multiscale observations (e.g., near-surface remote sensing, gene expression) to further validate the mechanisms behind these responses to drought.

4.3. Effects on Phenology Under Different Moisture Conditions

This study showed that the cumulative and lagged effects of drought on vegetation phenology also varied across different moisture conditions. The effect of drought on the vegetation SOS was stronger in relatively dry areas, intensifying with decreasing moisture. The effect on the vegetation EOS was stronger in relatively humid areas, which was consistent with a prior study [72]. This is because the vegetation SOS has a low resistance to moisture shortages, which tend to be more acute in dry regions, so drought’s influence on the SOS is stronger in dry regions. Vegetation that grows in wetter areas grows more foliage in the summer, accumulates more biomass, and consumes more water, and thus defoliation is more affected by fall drought [105]. Vegetation growing in arid areas, on the other hand, accumulates less biomass in summer and consumes less water, so leaf litter is relatively less affected by fall drought [18].

This study found that the time scales of drought’s cumulative effect on the DNF, DBF, and shrub SOS were shorter in response to drought in dry areas [106]. Conversely, the time scale of grassland’s response to drought is longer in arid regions [31]. This is mainly due to the fact that grasslands are more resilient and adaptive to shorter and more intense droughts in relatively arid regions, and are better able to self-regulate growth, morphology, and physiology to withstand drought stress on shorter time scales [107].

The drought’s cumulative effect response time scales showed a pattern of increasing and then decreasing with decreasing humidity for the DNF and DBF EOS, with different vegetation types exhibiting differentiated critical response thresholds. These patterns may be related to the differences in sensitivity to drought across vegetation types and need to be further explored in the future. The response time scales of drought’s cumulative effect were gradually shortened with decreasing humidity in the grassland and shrub EOS. We attribute this to the fact that vegetation growing in arid regions is frequently subjected to water stress, and in order to adapt quickly to water scarcity, plants in arid regions have evolved physiological, anatomical, and functional strategies to reduce their water loss, respiratory costs, photosynthetic activity, and growth rates [108].

For drought’s lagged effect, the lag time scales for the SOS in DNFs show a pattern of decrease and then increase with the intensification of drought, while for the EOS, the reverse pattern (increase and then decrease) was observed. The response time for the DBF SOS gradually shortened with the intensification of drought. The SOS and EOS of shrubs and grasslands responded faster to the lagged effect in arid regions: increased water loss shortened the response time to drought, and arid regions responded more rapidly to drought [106]. Arid regions have less soil moisture and generally poorer water-holding capacity, so vegetation growing in arid zones is usually sensitive to fluctuations in surface water resources [109]. In contrast, soil moisture content is more abundant in humid areas, so stable water reserves can continue to supply plant growth requirements even when precipitation is deficient [110].

The statistical strength of the association between drought effects and phenological nodes of different vegetation types was explored at the pixel scale. This study found that the lagged effect exceeded the cumulative effect in a larger proportion of the study area, which is consistent with the results of previous studies [31]. However, there were obvious differences in the strength of the correlations between drought and phenology nodes for different vegetation types: the DNF and shrub SOS was dominated by cumulative effects, while the DBF and grassland SOS was dominated by lagged effects. The EOS for all four vegetation types was dominated by lagged effects. While correlation strength is not sufficient to show a causal effect, these results suggest avenues for further research into the different growth strategies of different vegetation types. DNFs and shrubs focus on conservative growth strategies: their maximum photosynthetic rate is low and their photosynthetic and respiratory capacities are weak, making them more drought tolerant and resistant to prolonged cumulative drought [41]. For DBFs and grasslands, where resource-acquisitive strategies dominate, plants tend to close stomata earlier to show greater resistance to pre-existing droughts through drought escape and drought avoidance [88,102,111].

In addition, there are differences in water and nutrient requirements for vegetation growth at different phenological stages. Temperature and precipitation affect plant growth and ecosystem carbon cycling, which in turn affects the time of cyclical events in the plant life cycle [112]. As the climate warms and drought intensifies, the effects of drought on the phenological nodes of different vegetation types can be studied to “decode” the response of terrestrial ecosystems to climate stress. Its scientific value lies not only in revealing the micro-mechanisms of plant survival strategies, but also in providing decision support for macro-scale ecological risk management, the realization of carbon neutrality, and the sustainable development of human society.

5. Limitations and Prospects

This study used only the NDVI and SPEI datasets to investigate the effects of drought on the phenology of various types of vegetation. The relatively coarse spatial and temporal resolutions of these datasets and the lack of ground-based observational data for validation necessarily limit the applicability of the findings presented here. We rely on interpolation methods to fill gaps in the time series and to align data sources of different resolutions, but these methods cannot compensate for phenomena that are manifested at finer spatial or temporal resolutions. Furthermore, the study area extends into regions of permafrost, where the SPEI indicator may not capture the special hydrological processes dominated by snowmelt rather than incoming precipitation.

The interaction mechanism between climate change and vegetation phenology is relatively complex, and other potential factors such as land surface temperature or human activities will inevitably have a direct or indirect impact on vegetation phenology. Future studies should consider changes in the cumulative and lagged effects of intensified droughts in the context of climate change and incorporate multiscale observations (e.g., near-surface remote sensing, in situ validation, and gene expression) as well as more meteorological elements to further validate the mechanisms behind these responses to drought.

6. Conclusions

In this study, by analyzing vegetation phenology data and the multiscale SPEI index, we documented the differential phenological responses of shrubs, grasslands, and forests with different leaf forms to the cumulative and lagged effects of drought at multiple time scales from 2001 to 2020. The effect of drought on the vegetation SOS was generally more obvious than that on the EOS. There were differences in the cumulative and lagged effects of drought on grass, shrubs, and trees with different leaf forms. Our major findings include the following:

(1)

Drought led to a delayed SOS for grassland ecosystems, but an advanced SOS for DNFs, DBFs, and shrublands. Drought caused the advancement of the EOS in all four vegetation types. Drought had a large effect on the SOS in DNFs, and similar effects on the SOS and EOS in DBFs. The time scale for the grassland SOS response to drought was longer than that for the EOS, while the time scales for the DBF SOS response were shorter than for the EOS. Except for grasslands, the drought’s lagged effect on the SOS was mainly from the previous month, and the closer the drought to the phenological node, the greater the effect. The grassland EOS responded very rapidly to drought, mainly concentrated in the time scale of 1–2 months.

(2)

The effect of drought on the vegetation SOS was more obvious in relatively dry areas, while the effect of drought on the vegetation EOS was higher in relatively wet areas. There were differences in the change in the main drought response time scale to changes in average annual water availability. The cumulative time scale of the grassland SOS response to drought gradually increased with decreasing average annual water availability, while the cumulative time scale of the DNF, DBF, and shrub SOS response to drought gradually decreased with decreasing average annual water availability. The cumulative time scale of the grassland and shrub EOS response to drought gradually decreased, showing a pattern in DNFs and DBFs of first increasing and then decreasing with decreasing average annual water availability. In DNFs, with decreasing average annual water availability, the SOS drought lag time scales showed a decreasing and then increasing pattern, but an increasing and then decreasing pattern for the EOS. The lag gradually shortened in response to drought for the SOS in DBFs as well as for both the SOS and EOS in arid shrublands and grasslands.

(3)

The drought lagged effect on vegetation phenology exceeded that of the drought’s cumulative effect in 55.67% of the study area for the SOS and 61.55% for the EOS. The dominant drought effects varied among vegetation types and phenological nodes: For the SOS, drought’s cumulative effects dominated in DNFs (83.17%) and shrublands (62.71%), and lagged effects dominated in DBFs (56.47%) and grasslands (60.94%). For the EOS, the lagged effect was dominant in all four vegetation types, ranging from 84.08% of DNFs to 57.63% of grasslands.

Previous studies have focused more on cumulative and lagged effects of single vegetation types or the overall vegetation NDVI. The difference in cumulative and lagged effects of drought on different phenological nodes has not been fully studied and remains largely unknown, especially for trees with different leaf forms. This study found differences in the responses to drought of shrubs, grasses, and trees with different leaf forms, most significantly between grasslands and other vegetation types, especially for the SOS and less for the EOS. Differences in dominance effects across vegetation types reflect the differences between conservative growth strategies and resource-acquisitive strategies.

In summary, differences in vegetation survival strategies are the main reason for observed differences in the cumulative and lagged effects of drought. Therefore, a clear understanding of how different vegetation types and leaf forms respond to drought is essential for modeling the sensitivity of vegetation to future climate and moisture conditions. As drought intensifies, future empirical studies and models will need to focus on the varying effects of different growth strategies of vegetation as they pertain to different phenological nodes.