Long-Term Dynamics and Response to Climate Change of Different Vegetation Types Using GIMMS NDVI3g Data over Amathole District in South Africa

Abstract

Monitoring vegetation dynamics is essential for improving our understanding of how natural and managed agricultural landscapes respond to climate variability and change in the long term. Amathole District Municipality (ADM) in Eastern Cape Province of South Africa has been majorly threatened by climate variability and change during the last decades. This study explored long-term dynamics of vegetation and its response to climate variations using the satellite-derived normalized difference vegetation index from the third-generation Global Inventory Modeling and Mapping Studies (GIMMS NDVI3g) and the ERA5-Land global reanalysis product. A non-parametric trend and partial correlation analyses were used to evaluate the long-term vegetation changes and the role of climatic variables (temperature, precipitation, solar radiation and wind speed) during the period 1981–2015. The results of the ADM’s seasonal NDVI3g characteristics suggested that negative vegetation changes (browning trends) dominated most of the landscape from winter to summer while positive (greening) trends dominated in autumn during the study period. Much of these changes were reflected in forest landscapes with a higher coefficient of variation (CV ≈ 15) than other vegetation types (CV ≈ 10). In addition, the pixel-wise correlation analyses indicated a positive (negative) relationship between the NDVI3g and the ERA5-Land precipitation in spring–autumn (winter) seasons, while the reverse was the case with other climatic variables across vegetation types. However, the relationships between the NDVI3g and the climatic variables were relatively low (R < 0.5) across vegetation types and seasons, the results somewhat suggest the potential role of atmospheric variations in vegetation changes in ADM. The findings of this study provide invaluable insights into potential consequences of climate change and the need for well-informed decisions that underpin the evaluation and management of regional vegetation and forest resources.

1. Introduction

Vegetation is an important biophysical factor that controls the terrestrial exchanges of heat, moisture and carbon fluxes between the land and atmosphere, which consequently influences the microclimate through feedback processes at regional scales [1,2]. The vegetation-climate control is achieved through changes in surface albedo and surface roughness that determines the friction and turbulence, which may vary seasonally and spatially depending on the background climate and plant functional type [3,4]. For instance, grassland typically has a higher surface albedo than the forest ecosystem, suggesting a decrease in transpiration and ultimately leading to cooling or warming depending on which of these processes (e.g., transpiration, photosynthesis and heat exchange) dominates [5,6]. At the same time, vegetation response and growth also depend on subseasonal-to-seasonal changes in local surface and meteorological conditions such as soil moisture which is directly linked to precipitation, temperature, incident solar radiation, etc. For instance, under non-limiting water with the increasing solar radiation and temperature, vegetation responds by opening stomata, thereby enhancing water and carbon uptakes through the processes of transpiration and photosynthesis, respectively and consequently increasing plant productivity [7,8]. The vegetation response and performance also vary across ecosystem types/species [9,10], with changes in biogeochemical cycling across terrestrial ecosystems [11,12]. In the context of climate drivers, temperature, precipitation/soil moisture and incoming solar radiation are the main global climatological drivers of vegetation, whereas if one of these aspects deviates from optimal values, the vegetation becomes stressed and plant productivity declines [13,14]. The resultant deviations, therefore, affect ecosystem functioning and microclimate dynamics. In essence, vegetation plays a pivotal role in global climate, natural resources management, carbon sequestration and ecological biodiversity [15,16,17].

The evolution of satellite technology has enabled the development of important tools and ecosystem indicators used to monitor vegetation dynamics and trends from regional to global scales. Satellite-derived normalized difference vegetation index (NDVI) is one of such indicators, derived from green leaf scattering in the near-infrared wavelengths (0.72–1.0 µm) and chlorophyll absorption in the red wavelengths (0.58–0.68 µm). Increasing or decreasing the NDVI is considered a reflection of vegetation growth rate, used as an indicator for monitoring seasonal and inter-annual vegetation dynamics, phenological characteristics, land-use change, drought and fire activity, vegetation health and the associated response to climate variations [18,19,20,21,22]. Based on the third-generation Global Inventory Modeling and Mapping Studies (GIMMS; NDVI3g), a study demonstrates a stronger increase in cropland which is further linked to sufficient soil moisture, and a decline in the vegetation trend is linked to the temperature-induced moisture stress in India [23]. Depending on the season, other studies have also attributed increasing the NDVI or vegetation growth to increasing air temperature [24], rainfall variations [21], higher radiation associated with reduced cloud coverage [25] and CO₂ fertilization [26]. In South Africa, a study explored the rainfall–NDVI relationships over a semi-arid wetland in Limpopo Province and reported that vegetation productivity is largely influenced by rainfall variability in this area [27]. Weak to moderate correlations between the NDVI and precipitation and land surface temperature have also been found over Amathole District Municipality (ADM) in Eastern Cape Province [28]. While findings from previous studies provide insight on the potential effects of climate change on vegetation [29], the interactions among climatic variables and how they determine vegetation dynamics across land cover types are still poorly understood, particularly in ADM where climate-related events such as extreme drought often threaten vegetation performance and growth [29].

ADM in Eastern Cape Province of South Africa has witnessed a high record of climate and environment-related degradation such as vegetation stress, water scarcity and dry episodes in the last two decades [30,31]. For instance, ADM was reported to generally receive less than the region’s average rainfall amount of 400 mm/year with some few places recording more between 700 and 1000 mm/year occurring in spring and summer months [32,33]. These rainfall variations have effects on the hydrological cycle or water budgets and consequently on the rate of plant growth [34,35]. Hence, the interrelationship between vegetation and climate variables calls for constant monitoring to design appropriate targets to mitigate ecosystem collapse and food security. In essence, the knowledge of vegetation response to climate variations is fundamental to understanding the potential consequences of climate change in the future. To this end, the remote sensing-based NDVI3g and global reanalyses meteorological variables were explored to investigate the long-term (1981–2015) dynamics and responses to climate change of different vegetation types. The study attempts to answer two basic questions: (i) has the vegetation landscape in ADM experienced a greening or browning trend in the period 1981–2015?; (ii) what is/are the major climatic driver(s) of the response of vegetation types in ADM during this period? These contributions are essential to support the existing knowledge akin to vegetation dynamics and responses of different vegetation types to climate variations and ultimately to inform decision making on degraded areas of vegetation and forest resources.

2. Materials and Methods

2.1. Study Area

ADM in Eastern Cape Province of South Africa is located between 32°34′29.99″ S and 27°12′17.40″ E (Figure 1a), with an elevation of 1963 m. a.s.l. (6440 ft) and a total population of 892,637 [36]. Overall, the district covers a land area of 21,595 km² (8338 sq. mi) with approximately 200 km of coastline [37]. ADM comprises six local municipalities, which include Amahlathi, Ngqushwa, Great Kei, Mnquma, Mbhashe and Raymond Mhlaba. The district is ranked as the second richest biodiversity in South Africa [38]. This area is characterised by two bio-geographical regions, including the warm temperate south coast and the subtropical east coast [39]. In addition, the land comprises natural areas that are similarly diverse, which include moist mountainous, thornveld, well-watered coastal and semi-arid Karoo and succulent and thicket areas [32,40].

In general, a subtropical climate overrides ADM, with rainfall peaking in the summer [41]. The average rainfall is 400 mm/year but fluctuates between 700 and 1000 mm/year in some areas [30,42]. In addition, the average monthly temperature ranges between 1.5 and 2.5 °C and the winter temperature reaching an average of 21 °C, while the summer reaches an average temperature of 28 °C [43]. ADM comprises different eco-topographic undulating grassland and the Amatole mountain range along the Wild Coast from diverse landscape dynamics and climate, which results in various forms of habitats [44,45]. The formation of habitat in ADM varies as a result of the shifting climate and associated influencing factors including soil, topography and vegetation types across the landscape, which are crucial in vegetation distribution.

2.1.1. Satellite-Derived Products

The monthly global GIMMS NDVI3g product was used in this study. The data were obtained from the National Aeronautics and Space Administration (NASA) Ames Ecological Forecasting Lab (https://ecocast.arc.nasa.gov/data/pub/gimms/3g.v1/, accessed on 5 July 2021) for the period of 1981 to 2015. The data have a spatial resolution of 0.083° with a 15-day temporal resolution. The products were preprocessed and quality-controlled based on radiometric calibration, atmospheric attenuation, cloud screening, orbital drift, sensor degradation, view and illumination geometry and other effects that are not associated with vegetation change [46,47]. The monthly NDVI3g products are based on the maximum value composite (MVC) method. The MVC selects the highest value of each pixel from multi-temporal data to represent the current NDVI value [48]. For this study, the “Gimms v1.2.1” R package for downloading and processing global GIMMS NDVI3g data was used [49]. Finally, NDVI3g pixels with quality control flag 0 were used in the analysis.

2.1.2. ERA5-Land Reanalysis Data

The ERA5-Land data climate data for ADM were obtained from the Copernicus climate database (https://cds.climate.copernicus.eu/, accessed on 7 July 2021) from 1981 to 2015. The climate variables include the 2 m temperature (°C), precipitation (mm), wind speed (m/s) and shortwave radiation downwards (W/m²). The data are available hourly and at temporal and spatial resolutions of 0.01° and 0.01°, respectively. The global reanalysis ERA5-Land is the fifth generation global reanalysis product from the European Center for Medium-Range Weather Forecast (ECMWF) and supersedes its predecessors [50,51].

2.1.3. Ancillary Data

The land cover data for ADM were obtained from Copernicus Climate Data Store (https://cds.climate.copernicus.eu/, accessed on 7 July 2021) for 1992 and 2015. The data for both years were compared in a logical approach to remove pixels or land cover that have been modified or changed due to urbanization, deforestation, agricultural practices, etc. over the 23 years [52]. The stable pixels over the years were subsequently retained for analysis in this study. The retained stable pixels were reclassified by grouping the major land covers in ADM into four types, namely grassland, cropland, forest and shrubland, which means that these four types at both dates were taken to establish a mask for different land cover types (Figure 1b). The land cover maps were generated by the European Space Agency Climate Change Initiative (ESA CCI) [53]. The global land cover maps were produced at a spatial resolution of 300 m. Finally, all the data were harmonised from their native resolutions into common spatial (0.083°) and temporal (monthly) resolutions before analysis.

2.2. Study Methods

First, spatial, seasonal and inter-annual characteristics of NDVI3g were analysed. For seasonal analysis, the entire period was characterised into four seasons, which included winter (June, July and August), spring (September, October and November), summer (December, January and February) and autumn (March, April and May). The long-time climatology, trend and coefficient of variation (CV) of NDVI3g and climate variables were subsequently calculated for each season and vegetation type. The NDVI values typically range from about −0.2 to 0.1 for inland water bodies, snow, deserts, bare soils and sparsely vegetated areas and from 0.1 to 1 for increasing amounts of vegetation.

2.2.1. Coefficient of Variation

The coefficient of variation (CV) is generally used in vegetation studies and is mainly used to reflect the discrete degree of the NDVI data and the inter-annual variability of vegetation [54,55]. The CV was estimated as following:

$$C{V}_{\mathrm{NDVI}}=\frac{{\mathsf{\mu }}_{\mathrm{NDVI}}}{{\mathsf{\delta }}_{\mathrm{NDVI}}},$$

(1)

where $C{V}_{\mathrm{NDVI}}$ denotes the CV of the seasonal NDVI values of each pixel from 1981 to 2015; ${\delta }_{\mathrm{NDVI}}$ is the standard deviation of the seasonal NDVI values for 1981–2015; ${\mu }_{\mathrm{NDVI}}$ is the long-term mean seasonal NDVI value. A higher $C{V}_{\mathrm{NDVI}}$ value indicates that the NDVI time series shows a higher data variation, while a lower $C{V}_{\mathrm{NDVI}}$ indicates a more stable NDVI time series.

2.2.2. Trend Analysis

The trend in NDVI3g and climate variables were analysed using the non-parametric Mann-Kendall test [56,57]. The method has been widely used to identify trends and climate and vegetation series [58,59,60]. For the trend analysis, the confidence intervals were expressed using the Mann-Kendall static (M):

$$M={{\displaystyle \sum }}_{i=1}^{n-1}{{\displaystyle \sum }}_{j=i+1}^{n}sgn\left(y\right),$$

(2)

where n is the length of time series data and y = $P_j-P_i$; $sgn\left(y\right)$ is estimated as:

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

(3)

The negative and positive differences in data samples are represented by statistics. For the null hypothesis, the mean of $M$ is zero, i.e., $E\left(M\right)=0$, and the variance, σ, was expressed as:

$$\sigma =\frac{n\left(n-1\right)\left(2n+5\right)-{{\displaystyle \sum }}_{j=1}^n\left(t_j-1\right)\left(2t_j+5\right)}{18},$$

(4)

where $t_j$ is the size of any given jth tie. The $M$ test statistic distribution is subject to Z-transformation, which was described as:

$$Z=\{\begin{array}{c}\frac{M-1}{\sqrt{\sigma }}, M>0\\ 0, M=0\\ \frac{M+1}{\sqrt{\sigma }}, M<0\end{array}.$$

(5)

The “Sen” slope’s magnitude [61] was estimated as the median overall values of the entire data and expressed as:

$$S_k=Median\left(\frac{P_j-P_i }{j-i}\right), for \left(1 \le i<j\le n\right),$$

(6)

where P_j and P_i represent data values at times j and i (j > i), respectively, while n is the number of data. A positive (negative) slope indicates an upward (downward) trend.

2.2.3. Partial Correlation Analysis

To understand the relationship between vegetation and climate variables across ADM, pixel-wise partial correlation analysis was carried out between the NDVI3g and each climate variable for each season and the land cover type, for the period 1981–2015 [62,63,64]. The partial correlation coefficients between variables can be expressed as:

$$R_{ab,c,d,e}=\frac{r_{ab}-r_{ac}{r}_{ad}{r}_{ae}{r}_{bc}{r}_{bd}{r}_{be}{r}_{cd}{r}_{ce}{r}_{de}}{\sqrt{1-r_{ac }^2}\sqrt{1-r_{ad}^2}\sqrt{1-r_{ae }^2}\sqrt{1-r_{bc}^2}\sqrt{1-r_{bd }^2}\sqrt{1-r_{be}^2}\sqrt{1-r_{cd }^2}\sqrt{1-r_{ce}^2}\sqrt{1-r_{de }^2}},$$

(7)

where $R_{xy,z,i}$ is the partial correlation coefficients between variables $a$ (NDVI3g product) and $b$ (e.g., precipitation) while controlling the remaining variables $c, d, e$; $r$ is the Pearson correlation coefficient between variables $x$ and $y$, between $x$ and $z$, or between $y$ and $z$ for product $i$.

3. Results

3.1. Spatial and Temporal Variations in Vegetation across ADM

The study assessed spatiotemporal variations in the NDVI3g across Amathole District during the period from 1981 to 2015. The analysis of the NDVI3g climatology indicated that there was no distinct difference in vegetation characteristics across the seasons (Figure 2). However, for spatial characteristics, the values of NDVI3g increased from approximately 0.3 in the northern and western fringes to 0.8 in the southeastern part of ADM. The areas with low NDVI3g values were majorly dominated by grass and/or shrubs, while areas with high NDVI3g values were characterised by forest cover and/or crop landscape (Figure 1b).

The spatial trend in the NDVI3g for each season is shown in Figure 3. For the analysis period, the results indicated decreasing (browning) trends of NDVI3g across a wider part of ADM in spring (September, October and November), summer (December, January and February) and winter (June, July and August), except over grass landscapes in which the NDVI3g showed significant (p-value < 0.05) increasing (greening) trends in spring and winter. In autumn, the NDVI3g also showed significant increasing trends largely across the entire landscape. Overall, the trends in the NDVI3g were weak across the landscapes and seasons.

The boxplots (Figure 4) showed the long-term CV of the NDVI3g ($C{V}_{\mathrm{NDVI}}$) across seasons and vegetation types. Again, the variations of the NDVI3g were uniform (mean $C{V}_{\mathrm{NDVI}}$ ≈ 10) across the seasons and vegetation types, except for forest landscapes, which showed higher changes (mean $C{V}_{\mathrm{NDVI}}$ ≈ 15) in the NDVI3g across seasons, for the analysis period.

3.2. Relationship between Vegetation and Climatic Variables

3.2.1. NDVI3g and Air Temperature

The pixel-wise partial correlation coefficients between the NDVI3g and the ERA5-Land temperature across the region are presented in Figure 5. The results showed negative correlations between the NDVI3g and the air temperature, majorly in the northeast in winter and spring. In summer, the negative correlations covered a wider part of ADM, and it occupied a much larger part, in autumn. The boxplot shows the upper and lower quartiles ranges with different classes of vegetation types in ADM. Specifically, the negative correlations (R ≈ 0.2) were dominated over croplands and forests in winter and spring. All vegetation types show a negative relationship with the air temperatures in summer and autumn, but stronger (R > 0.2) over grasslands and shrublands, relative to forests and croplands during these periods. These results are counter-intuitive, because vegetation growth is expected to increase with the rising temperature normally in the course of a growing season (spring–autumn). However, an extremely high temperature above the optimum for vegetation growth may induce vegetation stress in association with the rising soil moisture deficit [23]. Interestingly, the trends in ERA5Land air temperature show somewhat an increase over the analysis period, particularly from winter to summer (Figure S1). This suggests that air temperature could potentially drive vegetation changes during this period. Generally, the correlations between the NDVI3g and the ERA5-Land air temperature were low (R < 0.4) and largely insignificant across the seasons, particularly in spring and summer, and vegetation types.

3.2.2. NDVI3g and Precipitation

The pixel-wise partial correlation coefficients between the NDVI3g and the ERA5-Land precipitation across the region are presented in Figure 6. The NDVI3g showed a positive correlations (R > 0.25) with precipitation during spring–autumn across the landscapes, except for a few isolated areas dominated by forests where the correlations were negative. In winter, the correlations (R > 0.3) were negative, largely over the entire domain. In the context of vegetation types, the positive correlations were more dominated over forests and shrubs in spring and over all vegetation landscapes except forests in summer and autumn. In winter, negative relationships occurred over all the vegetation types, except croplands. Concurrent with the NDVI3g, the ERA5-Land precipitation trends showed that precipitation decreased insignificantly in many areas from autumn to spring over the analysis period (Figure S2). However, the correlation results, which were also statistically insignificant, suggested that precipitation is less an important driver of long-term vegetation dynamics in ADM.

3.2.3. NDVI3g and Incoming Shortwave Radiation

The pixel-wise partial correlation coefficients between the NDVI3g and the ERA5-Land shortwave radiation across the region are presented in Figure 7. The correlation results were nearly similar to the NDVI3g vs. the air temperature (see Section 3.2.1 and Figure 5). This is anticipated, since the incoming shortwave radiation was linearly related to air temperature. In essence, a rise in the incident radiation would imply an increase in the air temperature and vegetation growth in an ideal situation. Figure 7 shows that the negative correlations (R > 0.2) between the NDVI3g and the radiation were mostly dominated over shrubs and forests in spring while shrubs and grasslands retain the negative correlations in summer. In autumn, all the vegetation types showed a negative correlation with the radiation, with shrubs indicating the strongest (R > 0.3). In contrast, the NDVI3g were positively related to the radiation (R ≈ 0.1–0.2) over all the vegetation types in winter. These results reflected the insignificant positive trends in shortwave radiation (Figure S3) in spring and, to some extent, in summer, in contrast to negative trends in the NDVI3g during these seasons (see Figure 3). Generally, the correlation values were also weak (R < 0.4) and largely insignificant across ADM during the analysis period.

3.2.4. NDVI3g and Wind Speed

The pixel-wise partial correlation coefficients between the NDVI3g and the ERA5-Land wind speed across the region are presented in Figure 8. The correlations between the NDVI3g and the wind speed were negative, majorly over forests and shrubs in the south of ADM in spring. In summer and autumn, the entire domain was dominated by negative correlations (R > 0.2), while in winter, all the vegetation landscapes showed positive correlations with the wind speed. Similar to other climatic variables, the relationship between the NDVI3g and the ERA5-Land wind speed was generally weak and largely insignificant. Nonetheless, the negative correlation results are consistent with the results of trends in the wind speed (Figure S4) and the NDVI3g (Figure 3).

4. Discussion

The results from this study revealed spatial and seasonal characteristics of the NDVI3g climatology and trend for different climate variables across ADM from 1981 to 2015. The climate variables exhibited a seasonal variation from one season to another during the study period. The results of negative correlations between NDVI3g and climate variables in the region might have significant implications on the hydrological cycle, human and livestock health and nutrition as well as ecosystem functioning. For instance, the variations in rainfall distribution in ADM might pose negative threats to food security, water availability and plant growth which may lead to recurrent droughts that are exacerbated by climate change. In the Northern Hemisphere, NDVI is largely influenced by temperature which connotes that a rising temperature/radiation implies an in-crease in vegetation activities [6,26]. Moreover, in a semi-arid environment, particularly in ADM where declining precipitation was observed in many areas, suggesting that precipitation is less an important driver of vegetation dynamics in the region. The declining precipitation and rising temperature are likely to have contributed to the drying trends and the resultant observed vegetation activity in the region. Therefore, it is important to recognise heterogeneous vegetation landscape characteristics and climate in ADM associated with the environment-related effect on vegetation cover. In essence, the major climatic drivers of the response of vegetation types can help to improve our understanding of vegetation landscapes and their impacts on a local scale as well as regional climatic conditions. Depending on land heterogeneity, the response of vegetation change has an impact on the regional circulation pattern, resulting in the various responses of climatic variables [65,66]. The seasonal characteristics of the NDVI3g climatology in ADM revealed a high spatial heterogeneity in vegetation growth trends. Although the overall vegetation-covered areas in ADM have been growing, the increase of the spatial heterogeneity in vegetation change trend requires management and development planning.

The CV of the seasonal NDVI3g vs. temperature, precipitation, solar radiation and wind speed yielded time-dependent NDVI characteristics for each vegetation type. This quantifies the NDVI3g trends in relation to the dynamic effects of individual variables contributing to the spatial variability of forest, grassland, shrubland and cropland across ADM. The results showed evidence of the areas with high NDVI3g values characterised by forest cover and crop landscape and areas with low NDVI3g, where grassland and shrubland showed the highest productivity for different seasons. Based on the findings of this study, positive and negative correlations of vegetation with climatic variables were observed across landscapes and seasons, which suggested negative vegetation changes (browning trends) dominated most of the landscape from winter to summer while positive (greening) trends dominated in autumn during the study period. Overall, the trends in the NDVI3g were weak across the landscapes and seasons, which might be attributed to the semiarid microclimate dynamics of the region. This connotes that climate characteristics in large areas are more complex and changeable than those in microscale areas, because the climates vary significantly at the microscale level than those in large-scale areas with varying influences on vegetation dynamics [10,29,67]. The non-parametric trend analysis revealed a changing trend of the NDVI3g from decreasing (browning) trends across a wider part of ADM, except over grass landscapes in which the NDVI3g showed significant increasing (greening) trends largely across the entire landscapes. Both the increased and decreased NDVI3g showed a statistically insignificant biogeophysical relationship with the climatic variables under investigation. The results revealed different patterns and fluctuations between the NDVI3g and the temperature, precipitation, solar radiation and wind speed (Figure S5). As a result, the increased (decreased) NDVI3g was linked to a decreasing (increasing) insignificant positive trend in temperature/radiation and subsequently increased (decreased) the precipitation as the length of the period increased. A recent study further corroborates the findings of these results based on NDVI trends and precipitation variations over the study area [28]. The results of vegetation dynamics with different climatic factors identified in this study differed seasonally and are broadly consistent with results of previous studies [28,29,67,68,69,70,71,72]. The changes in the climatic conditions further induced changes in the lower and upper at-mospheric conditions. The decreasing (increasing) vegetation cover in ADM can alter the surface energy balance and hydrological cycles, thus influencing the feedback mechanism of vegetation on global climate change.

Based on the results from the pixel-wise partial correlation analysis, the spatial and temporal trends between the seasonal NDVI3g and climatic variables showed that the wind speed was the most influential factor among other climatic variables influencing seasonal vegetation variation, especially for winter and spring in ADM. During winter, the spatial characteristics of the seasonal NDVI–wind speed trend appeared to follow a particular trend that might be described by some large-scale atmospheric systems. This might have negatively influenced vegetation production and trend as a result of the damage caused by strong winds [73,74]. Although, the relationship between the NDVI3g and the wind speed was generally weak and largely insignificant. A strong south easterly summer wind that blows along the west coast, causes the offshore flow of the ocean, which might have a large impact on vegetation dynamics among other environmental factors [75,76]. In contrast, the reduction in the wind strength over the study area might exert a positive effect on vegetation growth [77,78]. The findings of this study suggest that the positive (negative) processes of vegetation change with the wind speed could be a dominant factor in winter, which is a transition period from winter to spring in ADM. Moreover, the positive processes of vegetation change with the wind speed in winter proposed by this study could further influence soil moisture and water availability as well as ecosystem functioning. The identified wind speed characteristics help to better understand the contribution of each season to the changes in spatial and seasonal characteristics of NDVI3g climatology and the relationship between vegetation types and climate interactions in ADM. Several studies have revealed the importance of surface wind conditions in global climate system [77,78,79]. The identified positive correlation between the NDVI3g and the wind speed in all vegetation landscapes particularly in winter was based on remotely sensed vegetation data and other climatic variables from the ERA5-Land reanalysis data using various statistical methods. Nevertheless, a detailed study of feedback mechanisms based on an idealised simulation of a coupled vegetation-climate model is needed to identify more robust relationships between the vegetation activity and the climate in ADM. In addition, future investigations using observational data rather than reanalysis data are needed to fully understand the role of vegetation change on simulated vegetation-climate interactions and feedbacks in ADM. The newest GIMMS NDVI3g spatial resolution used for this study was relatively coarse, thus revealing overall trends for the entire ADM, but may not have enough resolution to accurately separate ecological processes. Therefore, future studies may use a higher spatial resolution dataset to better characterise vegetation trends and disentangle the impacts of external factors on vegetation change such as drought, warming, urbanization and ecological restorations.

5. Conclusions

In this study, pixel-wise partial correlation coefficient and non-parametric trend analyses were used to explore the NDVI3g for different seasons and different vegetation types and associated driving factors using long time-series remote sensing data from GIMMS (NDVI3g). The main conclusions of the study were summarised as follows. Firstly, during the 35 years from 1981 to 2015, the relationships between the NDVI3g and the climatic variables were relatively low (R < 0.5) and largely insignificant across vegetation types and seasons. The results somewhat suggested the potential role of atmospheric variations in vegetation changes in ADM. The large strength in inter-annual variation in the seasonal NDVI3g climatology and trends was observed in autumn and winter seasons with the highest productivity in these areas. The spatial and temporal trends in the NDVI3g peaked in the northwestern area majorly dominated by grassland/or shrubland and decreased towards the northeastern area of the region, where forest cover and/or crop landscape is predominant. Secondly, the correlation coefficient between the NDVI3g and the wind speed showed a stronger influence as a major driver of vegetation variation compared to other climate variables but was insignificant across the landscapes and seasons. Therefore, the wind as an essential climate component can affect carbon fluxes by changing carbon uptake and emission rates and also transport moisture and temperature from one area to another. Nevertheless, the impact of the wind speed has not received full attention in previous studies, thus assessing the characteristics of the variation of seasonal vegetation responses to the wind speed in the study area are crucial for sustainable forest resource utilization and ecological restoration. In general, the spatial and seasonal characteristics of the NDVI3g climatology in ADM indicated positive vegetation changes (increase) and vegetation types for the analysis period. Thirdly, the correlation coefficient between the NDVI and the climatic variables revealed an overall fluctuation across the seasons for different vegetation types with the largest decline in shrubs and/or crop landscape in these areas. This might be linked to the intended improvements on vegetation cover by the government which has not been fully undertaken in ADM, because the issues with the land use and land cover change and over-cultivation are yet to be fully resolved at the national level.

The deductions from this study are fundamental in designing strategies for targeted effective adaptation and mitigation to achieve sustainable vegetation and forest resource management. Uncertainties in terrestrial vegetation responses to future climate change require appropriate strategies to respond to climate issues and mechanistic approaches using remote sensing and GIS in light of the current result to address global change. This study suggests more studies on the assessment of ENSO impacts on spatiotemporal vegetation dynamics during the austral summer season to determine the degree and extent of vegetation response to ENSO warm events. In addition, there is a need to use actual biophysical variables such as LAI to assess vegetation changes to validate remotely sensed trends. This will reveal useful information on the use of different data such as remote sensing indices to assess El Niño impact on vegetation cover for effective vegetation monitoring and assessment.