Dual Roles of Water Availability in Forest Vigor: A Multiperspective Analysis in China

Abstract

Water availability is one of the most important resources for forest growth. However, due to its complex spatio-temporal relationship with other climatic factors (e.g., temperature and solar radiation), it paradoxically shows both positive and negative correlations (i.e., dual roles) with forest vigor for unknown reasons. In this study, a multiperspective analysis that combined the deficit of the Normalized Difference Vegetation Index (dNDVI) and multitimescale Standardized Precipitation Evapotranspiration Index (SPEI) was conducted for the forests in China, from which their correlation strengths and directions (positive or negative) were linked with spatio-temporal patterns of environmental temperature (T) and water balance (WB) (i.e., precipitation minus potential evapotranspiration). In this way, the reasons for the inconsistent roles of water were revealed. The results showed that the roles of water availability greatly depended on T, WB, and seasonality (i.e., growing or pregrowing season) for both planted and natural forests. Specifically, a negative role of water availability mainly occurred in regions of T below its specific threshold (i.e., T ≤ T_threshold) during the pregrowing season. In contrast, a positive role was mainly observed in warm environments (T > T_threshold) during the pregrowing season and in dry environments where WB was below its specific threshold (i.e., WB ≤ WB_threshold) during the growing season. The values of T_threshold and WB_threshold were related to the vegetation type, with T_threshold ranging from 1.3 to 4.7 °C and WB_threshold ranging from 129.1 to 238.8 mm/month, respectively. Our study revealed that the values of T_threshold and WB_threshold for a specific forest were stable, and did not change with the SPEI time-scales. Our results reveal the dual roles of water availability in forest vigor and highlight the importance of environmental climate and seasonality, which jointly affect the roles of water availability in forest vigor. These should be considered when monitoring and/or predicting the impacts of drought on forests in the future.

1. Introduction

Forest ecosystems cover more than 4.1 billion hectares of the Earth’s surface [1] and offer products (e.g., food and timber) and services (e.g., sequestering atmospheric carbon dioxide and controlling water erosion) [2,3], making them crucial to sustainable development. Forest ecosystems are strongly impacted by drought, which is likely to intensify in the future [4,5]. Despite the importance of water to vegetation, researchers have found that water supply (e.g., precipitation) exerts two quite different impacts on forest vigor (e.g., Normalized Difference Vegetation Index (NDVI) and Gross Primary Production (GPP)), i.e., both positive and negative roles (i.e., positive and negative correlations) coexist and their roles depend on the spatial location and timing of precipitation [6,7]. To date, most studies have concentrated on the positive role of the water supply in vegetation vigor, ignoring the negative role and its underlying mechanism [8,9]. Revealing the dual roles of water availability in forest vigor is crucial, as it forms the basis for a broad comprehension of the mechanisms involved and will allow us to make better predictions regarding the impacts of climate change.

Given that the impacts of water on forest vigor have time-lag [10] and cumulative effects [11,12], the cumulative effect of water availability should be considered when assessing the forest vigor-water relationship. Previous studies have indicated that forest vigor is affected by water availability not only at the current time (month or season), but also in the preceding time, with the impacts of accumulative water in preceding months probably being greater than those of the current water conditions [13,14]. The Standardized Precipitation Evapotranspiration Index (SPEI) is widely used to reflect the cumulative effect of water availability in terms of the magnitude, accumulative time, and spatial extent [14,15]. Given that SPEI is a typical meteorological index that contains little information about vegetation, it is commonly combined with a forest vigor index generated from optical remote sensing data, such as the NDVI [11,16]. Green vegetation strongly absorbs radiation in the red wavelength region, but strongly reflects it in the near-infrared region. Additionally, satellite remote sensing data provide information for forest areas that are inaccessible, remote, or difficult to assess for a direct field study. Therefore, remote sensing may be used to monitor vegetation dynamics over large areas at various spatial and temporal scales. Considering this, the satellite-derived NDVI calculated by the reflectivity of red and near-infrared wavelengths is considered a robust indicator for vegetation greenness and vigor at regional and continental scales [17].

The relationship between water availability and forest vigor is usually characterized using Pearson’s correlation coefficients [18,19]. Despite increasing recognition of the negative role (i.e., negative correlations) of water availability in forest vigor [6,20], the underlying mechanisms are unknown. A negative role means that forest vigor decreases with the increase of water availability, which is inconsistent with the ecology principle that water is a necessary resource for plants. Previous studies indicated that environmental climate, i.e., temperature (T) and water balance (WB) (i.e., precipitation minus potential evapotranspiration) and forest type, as well as seasonality, were related to water availability impacts on forest vigor. For example, the sensitivity of forests to water availability decreases with increases in environmental moisture [21], and varies according to the temperature zone and forest type [6,20]. Some studies have found a negative role of water availability in forest vigor in cold and wet regions [20]. Additionally, the main limiting factors for forest vigor change with the different seasons [8,9,19]. Forest vigor during spring and summer is more sensitive to temperature and water conditions [19,22]. However, previous studies did not use these influencing factors (i.e., environmental climate, season, and forest type) to explore the underlying mechanism of the dual roles of water availability in forest vigor.

China is a good region for exploring the underlying mechanism of water impacts on forest vigor, as drought is one of the most damaging and disastrous hazards there [23,24]. Due to the complex spatio-temporal relationship among climatic factors and vegetation, our study expected that water availability would exert a contrast effect on forests in China, driven by the environmental climate or season or forest type [6,8,9,19,20,21]. Therefore, this study focuses on forests in China, using satellite-derived NDVI and multiple time-scale SPEIs to quantify the impacts of environmental climate (i.e., the monthly mean temperature and water balance; the details of data can be found in Section 2.2.4) on the positive and/or negative role of water availability in forest vigor during different seasons. Through the multiperspective analysis, we try to reveal the underlying mechanisms causing a negative role and link them with the spatio-temporal patterns of T, WB, and forest types, in order to establish a baseline for monitoring and/or predicting the impacts of drought on the satellite-derived forest vigor (i.e., NDVI).

2. Materials and Methods

2.1. Research Area

Forests are widely distributed in China (Figure 1). The climatic conditions of China are complex, and forest ecosystems are affected by drought [23,24]. China is dominated by an Asian monsoon climate, with T and WB (detailed information about these datasets can be found in Section 2.2.4) both showing obvious seasonal variations (Figure 2a,b). Occurrences of T ≥ 0 °C and WB ≥ 0 mm/month are concentrated during May–September, indicating that the warm and wet seasons are concentrated in May–September. With these climatic conditions, forest vigor also displays obvious seasonal variation, being highest during the growing season of May–September (Figure 2c). To reduce the interference of leaf fall, which typically occurs from October–December, only changes in forest vigor during the pregrowing (January–April) and growing (May–September) seasons are considered in this research.

2.2. Materials

2.2.1. Satellite-Derived Vegetation Index

The NDVI derived from the Moderate Resolution Imaging Spectroradiometer (MODIS) of Terra was used to monitor forest vigor in our research. The MOD13A3 (Version 6) dataset is a global monthly composite product that spans 2001–2015 with a spatial resolution of 1 km (https://search.earthdata.nasa.gov/, accessed 9 December 2020). The NDVI data were resampled to a spatial resolution of 0.01° by using ArcGIS (Version 10.2) (Esri, Environmental Systems Research Institute, New York, NY, USA). Additionally, to reflect the relative changes in forest vigor and to facilitate comparisons of forests, as well as to reduce the complex impacts of both biotic and abiotic factors on the NDVI itself, the raw NDVI data for each pixel were transformed into the deficit of NDVI (dNDVI) based on the maximum value of NDVI during 2001–2015 (Equation (1)). Compared to the original value of NDVI, the dNDVI reflected the change in grid NDVI that would be more directly related to the changes in water availability.

$$dNDV{I}_{x,y,i}[\%]=\frac{(NDV{I}_{x,y,i}-NDV{I}_{x,y,i,\max })}{NDV{I}_{x,y,i,\max }}\times 100$$

(1)

where x and y represent the longitude and latitude of every forest pixel, respectively; i ranges from January to September; and max represents the maximum value of the monthly NDVI during 2001–2015.

2.2.2. Forest Inventory Map

A forest distribution map derived from the Eighth Forest Inventory (2009–2013) of China was used in our research (Figure 1), which includes planted coniferous (PC), planted mixed (PM), natural coniferous (NC), and natural mixed (NM) forests [25]. The forest inventory map was obtained by remote sensing and a massive forest field survey, and can well-distinguish forests and nonforests [11]. To relate the distribution to the remote sensing NDVI, the map was first digitized and then converted into a raster file with the same spatial resolution of 0.01° through spatial adjustment and resampling using ArcGIS. Given that many researchers have discarded regions with NDVI values below a threshold to minimize soil background influences and enhance green forest signals [26,27,28], and the NDVI value of forests during the pregrowing season (January–April) is commonly greater than 0 [29], the regions with the minimum NDVI below 0 during January-December for the entire study period were excluded from analysis in this study. As a result, the area analyzed in this study occupied 90.93% of the total forest cover (Figure 1).

2.2.3. Meteorological Drought Index

Meteorological drought indexes, such as the Palmer Drought Severity Index (PDSI) [30], Standardized Precipitation Index (SPI) [31], and SPEI [14], are commonly used indicators for drought monitoring and assessment because they simplify complex interrelationships between many climate and climate-related parameters. The SPEI is a widely used and multiscalar drought index that can be effective for characterizing the cumulative effect of water availability on a pixel scale provided that a suitable time-scale is selected [11,12,21]. The SPEI is a standardized drought index that reflects the relative change of the water condition. The main steps employed to calculate SPEI were as follows: (1) The water balance represented as the difference (D) between precipitation (P) and potential evapotranspiration (PET obtained by Thornthwaite or Penman—Monteith methods) was calculated; (2) the D values were aggregated at various time-scales Equations (2) and (3) the series of D was fitted by a three-parameter log-logistic probability density function and converted by the standard normal distribution [14,18].

$$D_j^i={\displaystyle {\sum }_{n=0}^{j-1}(P_{i-n}}-PE{T}_{i-n}),i\ge j$$

(2)

where i represents the calendar month, and j (months) is the timescale of the water balance aggregation.

Given that the Climatic Research Unit (CRU, University of East Anglia, UK) SPEI was calculated based on global monthly observed meteorological datasets that were interpolated to 0.5° and the PET was estimated using the well-known formula of FAO-56 Penman—Monteith [32,33], the CRU SPEI could characterize the actual water conditions more precisely on a regional scale [34,35]. In addition, the selected values of time-scales employed for capturing the cumulative effect were diverse in different research studies, with the longest time-scales of SPEI reaching 48 months [20]. Therefore, in this study, the monthly SPEI values with a 1–48 month time-scale derived from the latest CRU dataset (Version 2.5) were used (http://digital.csic.es/handle/10261/153475, accessed 10 March 2020). To match with other data, such as remote sensing NDVI and the forest map, only the SPEI datasets spanning 2001–2015 were selected and resampled to a spatial resolution of 0.01° using the nearest neighbor method of ArcGIS.

2.2.4. Environmental Temperature and Water Balance

To characterize environmental climatic conditions, the monthly observed meteorology datasets of CRU TS3.24.01 with a 0.5 degree spatial resolution (http://data.ceda.ac.uk/badc/cru/data/cru_ts/cru_ts_3.24.01/data/, accessed 9 December 2020) were obtained from the CEDA catalogue (the Natural Environment Research Council’s Data Repository for Atmospheric Science and Earth Observation) and used in this study, representing the basic data of CRU SPEI (Version 2.5). The datasets included the monthly temperature (K), monthly precipitation (mm/month), and daily potential evapotranspiration (mm/day). The environmental temperature and water balance were calculated according to the following steps. First, the total potential evapotranspiration was calculated for every month based on the daily potential evapotranspiration and the number of days. Second, the monthly WB was calculated using the monthly potential evapotranspiration and monthly precipitation (Equation (3)). Finally, the monthly averaged T and WB during 2001–2015 were calculated and resampled to a spatial resolution of 0.01°.

$$W{B}_{x,y,i}=P_{x,y,i}-PE{T}_{x,y,i}$$

(3)

where WB is the monthly water balance; x and y represent the longitude and latitude of every forest pixel, respectively; i represents months ranging within January–September; and P and PET represent the monthly precipitation and monthly potential evapotranspiration, respectively.

2.3. Methods

2.3.1. Definition of Positive and Negative Roles of Water Availability in Forest Vigor

Our research defined the positive and negative roles of water availability in forest vigor based on the strongest correlation between dNDVI and SPEI. First, we determined Pearson’s correlation (r) between monthly dNDVI and SPEI for time-scales of 1–48 months during the period of 2001–2015 (Equation (4)). Secondly, for each pixel, the absolute maximum correlation coefficient (|r|_max) was obtained by ENVI/IDL (Version 5.0) (Exelis Visual Information Solutions, Boulder, CO, USA) (Equation (5)). r was used to further characterize the strongest positive (r_p) or negative (r_n) correlations of forest vigor with water availability. Finally, the spatial distribution of r_p and r_n were determined by ArcGIS to show spatial patterns. In addition, their histograms for various forest types (PC, PM, NC, and NM) and different periods (pregrowing and growing seasons) were determined using MATLAB (R2012b) (MathWorks, Natick, MA, USA), respectively.

$$r_{x,y,i,j}=corr(dNDV{I}_{x,y,i},SPE{I}_{x,y,i,j})$$

(4)

$$|r_{x,y,i}{|}_{\max }=\max ({\mathrm{r}}_{x,y,i,all=1,\dots ,j})$$

(5)

where x and y represent the longitude and latitude, respectively; i ranges within January-September; and j ranges within 1–48 months. Because r was calculated using values of dNDVI and SPEI for 15 years (N = 15, where “N” represents the number of data samples), and we selected the value of 0.05 as the significance level of the correlation coefficient, corresponding to the threshold of 0.514, the value of |r| > 0.514 (i.e., critical value) means significant at a 0.05 level [36].

2.3.2. Quantifying the Impacts of Environmental Climate, Seasonality, and Forest Types

Based on the environmental climate data (i.e., T and WB) and the strongest correlations (i.e., r_p and r_n) of forests with water availability, our research aimed to quantify the impacts of environmental climate on the dual roles (positive or negative correlations) of water availability in forest vigor among seasons (i.e., pregrowing season in January–April and growing season in May–September) and among forest types (i.e., planted and natural forests). First, we counted the ratio of the area (the number of pixels) of significantly positive (i.e., r_p) and negative correlations (i.e., r_n) to the whole calculation area (the number of pixels) for every 5 °C and every 20 mm/month for pregrowing and growing seasons, respectively, which we also named the area percentage. Second, the statistical relationships between the average value of T and WB and the area percentage of significant r_p and r_n in the same climatic condition were constructed using a linear regression model (i.e., y = ax + b) and an exponential regression model (i.e., y = a[exp(bx)] + c[exp(dx)]), respectively. Additionally, to further eliminate the influence of the heteroscedasticity of residuals, we also combined the fitting line of the weighted regression model with that of the linear regression model. To determine whether planted and natural forests exhibited response differences, the independent-samples t-test with a bootstrapped sampling method (1000 samples) was used to compare the differences in the area percentage of both significant r_p and r_n between planted and natural forests. Finally, to confirm whether any climatic threshold existed, which can indicate changes of the difference in the area percentage between significant r_p and r_n, two methods were used in the research: (1) The intersection points of two linear fitting lines were obtained as the environmental temperature thresholds, and (2) a series of the difference between the area percentage of significant r_p and r_n was calculated for various ordered groups of environmental water balance that ranged from large to small (e.g., WB > 220 mm/month, WB > 200 mm/month, WB > 180 mm/month, etc.) by using the paired t-test method. We found that the difference between significant r_p and r_n appears to be significant at a 0.05 statistical level when WB is greater than one, and the value is the threshold of WB. Additionally, the actual data of T or WB around each threshold were added in the research, respectively.

2.3.3. Sensitivity Analysis for the Time-Scales of SPEI

Recently, a number of studies have found that the cumulative effect of water availability on forest vigor is important, but selecting a suitable time-scale of the cumulative time is often arbitrary, which can range within 1–48 months [6,18,20]. Therefore, to determine whether the dual roles (i.e., positive and negative correlations) of water availability in forest vigor are sensitive to SPEI time-scales, we compared the difference in climatic thresholds among time-spans of 1–12, 1–24, 1–36, and 1–48 months. A detailed flow diagram of the analysis is shown in Figure 3.

3. Results

3.1. Dual Roles of Water Availability in Forest Vigor

Based on the strongest correlations of forests with water availability, as characterized by the r corresponding to |r|_max, our research showed that water availability played dual roles (i.e., positive or negative correlations) in forest vigor in China (Figure 4 and Figure S1). The positive (r_p) and negative (r_n) correlations both displayed great spatial and temporal heterogeneity (Figure 4 and Figure S1). The role of water availability was related to the season (i.e., pregrowing season and growing season), but was not related to the forest type (i.e., planted and natural). During the pregrowing season, the values of r_p and r_n exhibited a double peak curve. The double peak was concentrated at 0.514 and −0.514, respectively. This phenomenon can be found in all forests (i.e., PC, PM, NC, and NM) and suggested that nearly 50% of forests showed no significant change when the water availability increased or decreased (Figure 5a).

Additionally, for all forests, there was no difference in the area percentage of significant r_p (27.37%) and significant r_n (25.61%) in the pregrowing season; during the growing season, however, the significantly negative role of water decreased (13.35%) and the significantly positive role of water increased (28.92%) (Figure 5b). This demonstrates that the forest region with a significant positive correlation was mostly observed during the growing season. Furthermore, the phenomenon can also be found in all forest types, as the insets of Figure 5 show. To eliminate interference of the normal variations for forest vigor, the regions with the significant positive and negative correlations (i.e., |r| > 0.514) were selected and analyzed further.

3.2. Factors Impacting the Roles of Water Availability

3.2.1. Temperature Dependency

The roles of water availability in forest vigor were highly dependent on the temperature and the time (i.e., season of concern), but independent of the forest type (i.e., planted and natural forests). During the pregrowing season, the area percentages of significant positive/negative correlations for all forests were linear increases/decreases with increased T (Figure 6a,c). The differences in the area percentage between positive and negative correlations also changed with increases in T (Figure 6a,c); that is, the role of water availability in forest vigor was related to T. Specifically, the forest regions with a negative correlation were mostly located in cooler environments (T ≤ T_threshold), while those with a positive correlation were mainly situated in warmer environments (T > T_threshold) (Figure 7). The value of T_threshold varied according to the forest type, with 1.3 (−0.85~2.60), 2.6 (2.42~2.87), 4.7 (4.55~6.65), and 3.4 (3.30~6.90) °C for PC, NC, PM, and NM, respectively (Figure 6a,c and Figure 7). These results indicated that T during the pregrowing season should be considered to better explore the impacts of drought. During the growing season, however, both the positive and negative correlations of all forests exhibited no obvious difference (Figure 6b,d); that is, the T does not show any effect on the differences in positive and negative roles of water availability during the growing season.

Despite the values of T_threshold being somewhat different for planted and natural forests, their relationship of temperature dependency is similar. Moreover, there were no significant differences in the area percentage of positive/negative correlations between planted and natural forests (i.e., PC vs. NC and PM vs. NM) during the pregrowing season and growing season, because all of the p-values of the independent-samples t-test with a bootstrapped sampling method (1000 samples) were greater than 0.05 (Table S1). This indicates that the role of water in vegetation vigor was independent of the forest type.

3.2.2. Water Balance Dependency

The roles of water availability in forest vigor were also highly dependent on environmental WB, and differed among the pregrowing season and growing season. Similar to the temperature dependency, the water balance dependency is not related to the forest type, as planted and natural forests show a similar relationship. During the pregrowing season, the area percentages of significant positive and negative correlations for all forests both linearly decreased with increases in WB (Figure 8a,c). There was no difference in the area percentages between significant positive and negative correlations, indicating that the environmental WB did not explain the differences between positive and negative roles of water availability in forest vigor during the pregrowing season. During the growing season, however, the area percentages of significant positive/negative correlations for most of the forests exponentially decreased/increased with increases in environmental WB (Figure 8b,d). At the same time, the differences in the area percentages between significant positive and negative correlations decreased gradually (Figure 8b,d). In drier environments (i.e., WB ≤ WB_threshold), most forests had a positive relationship with water availability (Figure 9). In contrast, in wetter environments (i.e., WB > WB_threshold), there was no difference in area percentages between significant positive and negative correlations (Figure 9). These results indicated that the role of water in vegetation vigor was related to environmental WB, which played a positive role in relatively drier environments. The values of WB_threshold were also related to the forest type, with values of 238.8 (241.4~238.8), 186.0 (189.5~186.0), 129.1 (129.7~129.1), and 186.3 (186.5~186.3) mm/month for PC, NC, PM, and NM, respectively (Figure 8b,d and Figure 9). These results indicated that WB during the growing season should be taken into consideration to better explore the effects of drought.

Despite the values of WB_threshold being somewhat different for planted and natural forests, their general relationship of water balance dependency is similar. The independent-samples t-test with a bootstrapped sampling method (1000 samples) indicated that nonsignificant differences exist in the area percentage of significant positive/negative correlations between planted and natural forests (i.e., PC vs. NC and PM vs. NM) during the pregrowing season and growing season at a 0.05 level (Table S2).

3.3. Sensitivity of the Roles of Water Availability in Forest Vigor to the SPEI Time-Scale

Comparisons of three time-spans of SPEI (i.e., SPEI 1–12, SPEI 1–24, SPEI 1–36, and SPEI 1–48) showed that the T dependency and WB dependency of the water availability impact on forest vigor were independent of the SPEI time-scale, with similar T_threshold and WB_threshold values for a certain forest type (Figure 10). Specifically, all results demonstrated that environmental climate and time (pregrowing season or growing season) jointly affected the dual roles of water availability in forest vigor and there were similar climatic thresholds for the different SPEI time-spans.

4. Discussion

4.1. Uncertainties and Prospects

The CRU-TS grid dataset (e.g., P, PET, and T) and their integrated index (i.e., CRU SPEI) were used in our research. To assess the suitability of these CRU data, we also compared them with site-based values (Figure 11). Specifically, we calculated PET and SPEI based on 593 stations of observed meteorological variables in China and compared them with those in CRU datasets. We found that all of the CRU data were consistent with those derived from observation stations, demonstrating a high consistency (Figure 11a–d and Figure S2a–d). Therefore, the CRU data were suitable for this study.

Because the statistical significance level of 0.05 (i.e., p = 0.05) is commonly used in related research, our research focused on the area with |r| < 0.514 (i.e., p < 0.05). However, the correlation is an estimate with associated uncertainty and statistical significance itself is a rather artificial criterion. Therefore, we compared the results of the statistical significance level of 0.05 with that of 0.1 (i.e., |r| < 0.441). We found that there was no difference in the thresholds between two significant levels (Figure 12), because the p values of the t-test (with a bootstrapped sampling method, 1000 samples) for temperature and water balance thresholds were 0.899 and 0.191, respectively.

Despite these efforts, some uncertainties still remain, such as the occurrence of possible errors during the matching process between the forest distribution map and remote sensing images, as well as in the spatial resolution resampling process. Therefore, more matchable data on time and space are needed in the future to continue to decrease the uncertainties. In addition, changes in forest vigor may be affected by other factors (e.g., insects, diseases, and fire) [23,37], as well as interactions among them and climate change/drought [38,39,40], which adds uncertainties to our results. However, the effect of these factors on forests are always represented by relatively low percentages [37], for example, <0.17% of the forest area in Yunnan was damaged by fire during 2001–2014 [12], and such uncertainties would not change our main conclusions. In addition, the background of global warming and the changes in drought frequency, duration, and intensity [16,41] will require some consideration in future study [23].

The synergic use of multiple remote sensing data (e.g., optical and microwave remote sensing, as well as light detection and ranging (LiDAR) data), could provide better results, because current data still present some uncertainties. For instance, optical remote sensing data (e.g., the vegetation index based on reflectivity) have been widely used to investigate the responses of forests to climate change due to them providing continuous information on forest dynamics, both temporally and spatially [42,43]. However, optical remote sensing only acquires canopy surface information [44]. In contrast, microwave remote sensing (e.g., synthetic aperture radar (SAR)) is superior to optical remote sensing in reducing high-resolution imagery of the Earth’s surface in all weather conditions [45,46], and LiDAR techniques have been proven to accurately capture the three-dimensional structure of a vegetation ecosystem [47,48]. However, SAR and LiDAR data also have limitations, such as limitations in temporal and spatial coverage, uncertainties in estimation, expensive datasets, and difficulties in data processing [49,50,51]. Therefore, a growing number of research focused on the synergic use of multisensor optical, SAR, and LiDAR, could provide better results as all data types complement each other, overcoming their respective limitations [49].

4.2. Adjustment Mechanisms of Environmental Climate and Seasonality

Our study found that the dual roles of water availability in forest vigor were jointly controlled by the environmental T and the seasonality (i.e., pregrowing season or growing season). In this study, we found that the roles of water availability in forest vigor in China had both positive and negative roles (Figure 4 and Figure 5 and Figure S1), which is consistent with other research [6,20]. Specifically, the negative role of water availability in forest vigor was mainly found in regions of T ≤ T_threshold during the pregrowing season. The T_threshold values ranged within 1.3–4.7 °C and varied according to the forest type, consistent with other phenological research which found that a temperature lower than 5 °C is too cold for forest growth during the pregrowing season [52,53,54]. As a result, forests located in cold regions (T ≤ T_threshold) are not limited by water availability, but by temperature [22]. As a result, there was no difference in the frequency of r_p and r_n in the pregrowing season, which means that nearly 50% of forests showed no significant change when the water availability increased or decreased at a 0.05 significant level (i.e., |r| > 0.514) (Figure 5). Additionally, given that temperature is usually negatively correlated with water availability [55], the negative correlation is probably an indirectly response of those forests located in cold regions (T ≤ T_threshold) to water availability during the pregrowing season (Figure 13a). With increases in environmental T, the importance of water availability for forest vigor increased, which caused the area percentage of the positive correlations to increase accordingly (Figure 13a,b). As T further increased and reached a threshold (i.e., T > T_threshold), the main limiting factor for forests became water availability and thus the positive role of water availability appeared (Figure 6, Figure 7 and Figure 13a,b).

Additionally, the dual roles of water availability in forest vigor were driven by the environmental WB and the seasonality. Previous research indicated that forest vigor during the growing season is commonly controlled by water availability [19,22], especially for drier regions of WB ≤ WB_threshold (Figure 13b). Therefore, they mainly show a positive response to water availability (Figure 8 and Figure 9). With increases in environmental WB, the importance of water availability for forests decreases, while the impacts caused by other limited factors (e.g., nutrients, diseases, and insects) emerge [23,37,56] (Figure 13b,c). As a result, forests located in wet regions (i.e., WB > WB_threshold) showed not only positive correlations, but also negative correlations, with water availability, which caused nonsignificant differences in the area percentages between the positive and negative correlations (Figure 8b,d). The results are consistent with other relevant research showing a negative correlation in wet regions [6,20] (Figure 13c). The WB_threshold is a key parameter for determining whether or not the forest is limited by water availability. In this study, we found WB_threshold of 129.1–238.8 mm/month, which varied according to the forest type (Figure 8 and Figure 9). In addition, values of WB_threshold were not sensitive to the selected SPEI time-scale (Figure 10), implying that these thresholds were invariant and useful for evaluating the effects of drought.

However, the roles of water availability in forest vigor were vegetation type-independent. The planted and natural forests showed similar rules due to the fact that temperature and water were the necessary resources for all forests [22,54]; that is, both environmental climate (i.e., T and WB) and seasonality (i.e., pregrowing season or growing season) adjusted the sensitivity of forests to temperature and water. Therefore, planted and natural forests exhibited consistent phenomena of temperature and water balance dependency (Figure 6 and Figure 8). However, the forest age, canopy height, stock volume, and other factors may lead to the differences in the sensitivity of forests to temperature and water availability [11,15,57], which means that the values of T_threshold and WB_threshold varied to a certain degree with forest type (Figure 7 and Figure 9).

5. Conclusions

Water is an important climatic resource for vegetation. However, due to the complex spatio-temporal relationship among climatic factors and vegetation, water shows both positive and negative correlations with forest vigor, and the reasons for the negative impact are not well-known. In this study, we conducted a multiperspective analysis for the forests in China. Specifically, to reveal the underlying mechanism causing the positive and negative role of water availability, we considered the influences of multiple factors (i.e., temperature and water balance, forest types, season differences, and multiple time scales of water availability) by combining multiple data (e.g., satellite-derived NDVI and the multiple time-scale SPEIs). The results indicated that the negative role of water availability in forest vigor mainly occurred in regions of T ≤ T_threshold during the pregrowing season. In comparison, positive roles mainly occurred in warmer regions (T > T_threshold) during the pregrowing season and in wetter regions (WB > WB_threshold) during the growing season. The T_threshold and WB_threshold were vegetation type-dependent, with values ranging from 1.3 to 4.7 °C for T_threshold and 129.1 to 238.8 mm/month for WB_threshold. These results revealed that both environmental climate and seasonality should be considered when monitoring or predicting the impacts of drought on the satellite-derived forest vigor (i.e., NDVI). Therefore, to explore the direct effects of drought on forests in the future, relevant research should focus on forests limited by water availability.