The NDVI dataset used in this analysis is GIMMS NDVI version 3 (referred to as NDVI3g), derived from NOAA AVHRR data (National Oceanic and Atmospheric Administration Advanced Very High Resolution Radiometer) [30]. The dataset has a length of 31 years, covering July 1981 to December 2012, with a fortnightly temporal and 8 km spatial resolution. The NDVI3g archive has certain areas masked off over non-vegetated lands, or when measurements are ambiguous due to surface conditions [31]. These include oceans, major lakes, the Greenland icecap, and many parts of the Arctic and Antarctic. This dataset is considered to be the most accurate long-term AVHRR data record [32], which makes it appropriate for long-term studies of land surface trends in vegetation, seasonality, and coupling between climate variability and vegetation. We extracted NDVI values from January 1982 to December 2012 and applied the maximum-value-composite (MVC) technique to obtain monthly time-series to further remove the bias caused by atmospheric conditions [33]. The annual NDVI time-series were calculated by averaging the monthly data.

The climate data are obtained from the latest version of the high-resolution Climatic Research Unit time-series (CRU-TS) 3.22 dataset, provided by the University of East Anglia [34,35,36]. The gridded time-series dataset covers the period 1901–2013, at 0.5 degree resolution. The monthly data are calculated from daily or sub-daily data provided by more than 4000 weather stations distributed around the world and aggregated into annual time-series for trend estimation. Only temperature and precipitation data from 1982 to 2012 are used in the analysis in accordance with the NDVI data.

The Global Human Footprint Dataset (LWP-2)—a proxy for anthropogenic effects—is the normalized Human Influence Index (HII), created from nine global data layers including human population pressure (population density), human land use and infrastructure (built areas, nighttime lights, land use, land cover), and human access (coastlines, roads, railroads, navigable rivers). It is a comprehensive index representing human effects on the environment with a spatial resolution of 1-km. It ranges from 0 to 100, with higher values indicating a stronger anthropogenic impact. The dataset was produced by the Wildlife Conservation Society (WCS) and the Columbia University Center for International Earth Science Information Network (CIESIN) [37].

Ecoregions are large areas of similar climate where ecosystems recur in predictable patterns. Bailey’s Ecoregions Map characterizes the patterns of global climate and potential natural vegetation [38,39]. That enables us to explore the relationships between NDVI trends and climatic factors on a regional basis, covering different climate regimes. The map consists of four first-level categories or Domain regions—Dry, Humid Temperate, Humid Tropical, and Polar—each containing several sub-categories. The data can be obtained from the U.S. Forest Service, Inventory and Monitoring Institute, Ecoregions Center.

We estimated linear trends of NDVI and climate variables from their annual time-series for each pixel using the Theil–Sen (TS) median slope. The Theil–Sen procedure is a rank-based test that calculates the non-parametric slope and intercept of the time-series by determining the median of all estimates of the slopes derived from all pairs of observations [40]. The outputs of the TS procedure provide an annual rate of change of NDVI and a measure of significance. The TS procedure is commonly used in estimating NDVI trends [11,41] because it is resistant to outliers and is accurate even for skewed and heteroscedastic data [42].

In order to better capture temporal changes of NDVI, trends were estimated for three distinct periods: 1982–1994, 1995–2004 and 2005–2012. This division is not determined by the breakpoint detection (e.g., Rogier et al. [43]) because breakpoints and their frequency could vary greatly across the world [44]. Usually, breakpoint detection approach is applied to a certain region rather than the whole world [45,46]. In fact, the two breakpoints adopted here, 1995 and 2004, aimed to define a period during which both NDVI and Global Human Footprint data overlap. Therefore, they offer a unique time window for analyzing interconnections of human activity and NDVI trends. In addition, the breakpoints chosen here approximately match the breakpoints detected in the previous studies of Piao et al. [47] and Zhang et al. [45]. The TS procedure is also applied to the climate variables, including temperature and precipitation, for trend estimation. After estimating trends in NDVI, temperature, and precipitation by the TS procedure, we analyzed correlations between NDVI change and climate variables and also human footprint data across different ecoregions and geographical regions.

The estimated annual rate of change of NDVI from 1982 to 2012 is shown in Figure 1, with a positive average of 0.46 × 10^–3 per year. Most of the greening areas are in the Northern Hemisphere (NH), consistent with the previous studies [12,13,18,48]. The most evident greening trend can be found in Western Europe, followed by the Southeastern United States, Amazonia, Sahel, India, and the southeastern parts of China. Browning trends are mainly found in parts of the boreal region of North America, as well as Argentina and Southern Africa.

NDVI trends for three separate periods are presented in Figure 2. The period 1982–1994 is characterized by widespread greening trends in the NH, especially over Europe, parts of the United States, India, and China. In contrast, the majority of the Southern Hemisphere (SH) shows insignificant NDVI trends, with some noticeable browning. This is in agreement with the study of Xiao and Moody [48]. During the period 1995–2004, areas with significant NDVI change were dominated by browning trends all over the world, while greening only took place in the high latitudes of North America and Asia. The increase in vegetation growth that took place prior to 1995 over broad areas in the NH became insignificant or even reversed during this period. This is also supported by other studies [47,49,50,51]. For example, Piao et al. [47] found that the growing season NDVI significantly increased from 1982 to 1997 and then decreased from 1997 to 2006 in the temperate and boreal Eurasia. In the latest period, 2005–2012, while NDVI changes in most parts of the world remained insignificant, significant greening trends appeared in Northern Russia and Eastern Australia. Moreover, browning trends appeared mainly in the Western United States, northern parts of South America and Central Africa. These results clearly indicate that spatial patterns of NDVI change vary greatly over different periods. We observe an overall greening, browning, and greening for the three periods 1982–1994, 1995–2004, and 2005–2012, respectively, at annual rate of change of 0.65 × 10⁻³, −0.55 × 10⁻³ and 0.97 × 10⁻³.

Table 1 shows the spatio-temporal patterns of NDVI trends aggregated by Bailey’s Ecoregions Map. Letter G represents a greening trend (positive NDVI change rate), while letter B represents a browning trend (negative NDVI change rate). Figure 3 shows the spatial distribution of Bailey’s Ecoregions and the corresponding temporal patterns of NDVI change according to Table 1. It is observed that the G-B-G trend is the dominant pattern in the world, covering large parts of Asia and Europe, middle-eastern North America, North Africa, and much of Australia. These regions experienced a greening trend interrupted by browning in the period 1995–2004, consistent with the conclusions of previous studies [20,43]. The B-B-B trend only exists in the Rainforest Division of both South America and Africa, indicating that the rainforest regions underwent continuous browning throughout the 31 years. The G-G-G trend, indicating continuous greening, is observed in high latitude regions of the NH, namely, Tundra Regime Mountains and Tundra Division. The B-B-G trend is observed in the Savanna Division and Subtropical Division regions, indicating that a browning trend have turned into greening over the latest period of 2005–2012. The G-G-B trend, showing a greening trend that turned into browning after 2005, is only observed in the Temperate Desert Division region. The G-B-B trend is observed in regions such as the Marine Division and Marine Regime Mountains.

Climate variability is one of the main drivers of vegetation dynamics [13,43,52]. The spatial relationships between NDVI change and the changes in climate variables (i.e., temperature and precipitation) are analyzed for each domain of the Bailey’s Ecoregions (Table 2). The results show that the strengths of relationships vary among ecoregions over different periods. We only include pixels with significant NDVI trend in the analysis to better reflect vegetation change, whereas those with insignificant trend are excluded. Significant NDVI trends account for about 10% to 18% of all samples in each period among four ecoregions and are able to well represent the regional trends compared to all samples (Table S3). In most cases, the influence of temperature and precipitation changes on NDVI trends are positive, suggesting that the increase in both temperature and precipitation could promote vegetation growth. However, some exceptions exist in the Dry Domain Region and the Polar Domain Region, where the partial correlation is negative over certain periods. The relationship between the NDVI trend and temperature trend becomes weaker as the partial correlation decreases over time, especially in the Humid Temperate and Humid Tropical Domain Regions. On the other hand, the relationship between NDVI trends and precipitation trends becomes stronger after 2005, especially in the Humid Temperate Domain Region. The results of multiple regression analysis (Table S1) show similar spatial patterns for the relationship between NDVI trend and climate variables (i.e., temperature and precipitation).

In the Dry Domain Region, both mean temperature and precipitation increase over each period. The positive partial correlation between NDVI and temperature trends decreases over time and becomes even negative (−0.415) in the last period. In contrast, the correlation between NDVI and precipitation trends stays positive throughout all three periods. These results suggest that in the Dry Domain Region, the effect of temperature change on NDVI trends is more variable. The relative strength of the impact of temperature and precipitation on NDVI changes also varies with time. For example, temperature trends had stronger correlation with NDVI trends than precipitation during the period 1982–1994. Nevertheless, precipitation played a more significant role over the last two periods, as can be seen from the high correlation coefficient of 0.452 for the period 2005–2012.

In the Humid Temperate Domain Region, effects of both temperature change and precipitation change on NDVI trends are consistently positive during each period, although their strengths vary. While the partial correlation of NDVI trends and precipitation change increased, the partial correlation of NDVI trends and temperature change decreased. This weakened spatial relationship between NDVI trends and temperature change is partly supported by Piao et al. [53]. They show that the relationship between the interannual temperature variability and NDVI has weakened substantially in the northern mid and high latitudes over the past three decades, possibly due to extreme climate events.

In the Humid Tropical Domain Region, both NDVI and precipitation decrease in all three periods. The decline in vegetation activity might be due to the drying climate, as indicated by the previous studies [49,50,51]. Still, both temperature change and precipitation change have a positive effect on NDVI change in each period, although the partial correlation with temperature change decreases as in the Humid Temperature Domain Region. The close values of correlations of NDVI trends with temperature and precipitation in each period indicate that both climate variables are of similar significance in determining vegetation growth change.

In the Polar Domain Region, temperature and precipitation increased significantly over all three periods. Nevertheless, there is no clear correlation between NDVI change and climate variables because the sign of the correlation coefficient, R, changes over time. Temperature has a higher partial correlation with NDVI than precipitation, indicating that heat might play a more important role in vegetation growth over this region. In addition, the very low vegetation cover and unfavorable atmospheric conditions [43] in this region could influence the quality of the data.

The changes in vegetation growth are dependent upon the collective changes of temperature and precipitation. The positive relationship found between NDVI and precipitation trends in most cases demonstrates that an adequate increase in precipitation is a prerequisite for realizing the positive effect from the rising temperature; otherwise, warming can even reduce the vegetation growth as observed recently [49]. This can happen because moisture deficit due to increased evapotranspiration in a warmer world may offset the positive effect of temperature on vegetation growth (extended length of the growing season) [53]. For example, in the Dry Domain Region, although both temperature and precipitation increased, temperature increase had a negative effect on NDVI trend while the effect of increasing precipitation was positive.

The greening/browning trends in many areas are only partially explained by temperature and precipitation. In large regions, other factors—such as CO₂ fertilization, reforestation, forest regrowth, woody plant proliferation, and agricultural management practices—may drive these trends. The increase in atmospheric CO₂ concentration is likely to cause an increase in vegetation productivity, supported by modeling studies [54], experimental studies [55], and studies based on remote sensing data [56].

Changes in vegetation activity can also be influenced by human activities (e.g., land use). Anthropogenic effects on vegetation change could either be positive or negative, depending on the specific human actions (e.g., positive effects from fertilization and irrigation on crop lands and negative effects from deforestation or urbanization). Global Human Footprint (GHF) is used as a proxy to represent the extent of anthropogenic effects as a whole, since it cannot distinguish between actions with positive or negative effects on vegetation. Therefore, here we test the hypothesis that more intensive human activities could lead to greater changes in vegetation activity as reflected by NDVI regardless of the sign of the change. For this purpose, the relationship between the magnitude of the NDVI trend (trend in absolute value) and human footprint is examined at the country level across different continents. We conduct the analysis at the country level because human influence is closely linked to its political, economic, and social status. Since the GHF dataset only covers the time period 1995–2004, NDVI trends for the same period (1995–2004) as well as for the whole time span of 1982–2012 are analyzed. The results for the two time periods are generally consistent and only differ in their statistical significance (Table S2), suggesting that this analysis is not sensitive to the choice of period.

Figure 4 shows the relationship between NDVI trends and human footprint for the period 1982–2012. Clearly, there are strong positive correlations in Africa, Asia, and Europe (p < 0.01), weak positive correlations in Oceania (p > 0.1), and weak negative correlations in both North and South America (p > 0.1). The positive correlation shows that intensive human footprint is indeed associated with a larger NDVI change in terms of the magnitude. On the other hand, human influence can enhance the vegetation growth. This positive effect could result from development activities such as land transformation, intensive agriculture (irrigation, fertilization), plantations, and nitrogen deposition [27]. For example, evidence from China suggests that greening in croplands and forested lands are mainly caused by agricultural practices as well as afforestation and reforestation [18]. Fertilization and irrigation probably resulted in greening in India from 1982 to 2003 [57]. Greening in the Sahelian region of Africa is partially explained by the land use changes [58]. On the other hand, human influence can also negatively impact the NDVI trends and cause browning (e.g., deforestation). For instance, it is suggested that the majority of the browning from 1982 to 2003 in Central Asia was caused by the abandonment of agricultural lands [59]. However, the weak relationships observed in Oceania, South America, and North America demonstrate that the relationship between human influence and magnitude of NDVI trends is not straightforward, probably because the negative (positive) effects of human activity could be counteracted by the positive (negative) effects of climate change, leading to a weak correlation between NDVI change and GHF.

Preliminary results of the statistical analysis cannot specify which human factors may be responsible for NDVI increase or decrease, because human activities are diverse and have substantial regional differences. More importantly, climate variables also contribute greatly to NDVI change, which can modify the impact of human activities to some extent. In fact, multiple mechanisms drive vegetation dynamic changes, including climate factors (temperature, precipitation), climate change induced disturbance (fire, plant, diseases, and insect pests), and human activity induced environmental changes (land use/cover change, forest regrowth, nitrogen deposition, CO₂ fertilization, aerosols, etc.). Therefore, further research is needed to better understand the mechanisms of human activities on vegetation changes, especially studies with process-based ecosystem models capable of separating climate and human factors (e.g., Mao et al. [60]).