Warming diminishes the stability of primary productivity in global grass- and forb-dominated ecosystems

Global warming has induced increases in productivity in open, grass- and forb-dominated (OGFD) ecosystems. However, little attention has been given to the temporal stability of productivity responses to global warming. We show that the stability of productivity in OGFD ecosystems decreased significantly over the past 40 years. Using the satellite-derived normalized difference vegetation index (NDVI) data from the Global Inventory Modelling and Mapping Studies (GIMMS) group, we analyzed global patterns of the stability in productivity among OGFD ecosystems. We found that the global mean stability of NDVI-based productivity estimates significantly decreased from 1982 to 2015. Comparing different trends, we found that stability decreased by 36%, and increased by 27% of the total area of OGFD ecosystems. The stability of productivity in OGFD ecosystems decreased in the northern hemisphere, especially in the Mongolian plateau and Eurasian steppe. In contrast, stability increased significantly in the southern hemisphere. Increases in both mean annual temperature and annual temperature variability were correlated with decreases in the stability of productivity in the northern hemisphere. Although the productivity of OGFD ecosystems has generally increased with warming, the stability of production has decreased. OGFD ecosystems, particularly northern hemisphere systems with low baseline productivity may be vulnerable to the loss of grazing potential and grazing consistency in the warmer future. These observations highlight the need for adaptation strategies for animal husbandry to respond to variability in productivity and reduce the negative impact of climate change on grazed ecosystems.


Introduction
Trends in vegetation growth and net primary productivity (NPP) are important indicators to evaluate ecosystem function, climate change, and ecosystem services such as carbon storage, hydrological throughput, and food safety [1,2]. Based on satellite remote sensing data, many studies have confirmed increases in NPP among terrestrial ecosystems over the most recent half-century [2][3][4][5][6]. However, other studies have shown that this trend of increased NPP is weakening in northern latitudes [7,8], or that trend in changes of NPP differs across regions and among ecosystems [3,[9][10][11].
Open, grass-and forb-dominated (OGFD) ecosystems are both one of the most broadly distributed terrestrial ecosystems as well as one of the most important in terms of supporting animal husbandry and wild Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
herbivores across the world [11][12][13][14]. As with other systems, NPP is sensitive to climate change and human activities [11][12][13][14][15][16]. In OGFD ecosystems, precipitation and temperature are the main factors determining NPP. Previous research has indicated that precipitation and warming moderate the productivity of OGFD ecosystems, as the climate variability in each is associated with variation in productivity [16][17][18][19][20]. In addition to the mean value of NPP, the stability of productivity for seasonal, annual, and interannual time scales, related to the variability of NPP, is also an important indicator for ecosystem functioning in OGFD ecosystems, which are critical to both OGFD ecosystems management and livestock production [15,20]. Changes in average annual climate conditions, seasonal patterns, climate variations, and extreme climate events can have significant effects on the stability of ecosystem productivity. Recently, several field and experimental studies revealed the impacts of climate change on the stability of productivity in the OGFD ecosystems [16,20]. Furthermore, the stability of productivity is mainly related to species diversity, and complementary growth of different species or functional groups at the community level under climate change conditions [16,[20][21][22]. However, fewer studies [5,6,[23][24][25] have focused on broad-scale patterns in the variability of NPP among OGFD ecosystems and its responses to climate change at a global or regional scale.
We note that NPP is a proxy for grazing potential, an important ecosystem service in OGFD systems for both livestock and native grazers. Understanding the stability of NPP is important for two interrelated reasons. Firstly, OGFD systems bio-climatically reside at the cusp between forests and deserts and are vulnerable to environmental system change. Secondly, OGFD systems are vulnerable to degradation through over-grazing. With decreased stability comes the need for adaptive management of both native grazers and livestock, and the value of early season indicators of seasonal grazing potential. These two issues interact because overgrazing can drive desertification in stressed grasslands. We used global measures of remotely sensed vegetation that predicts NPP to understand over 30 years of change across OGFD systems around the globe. In OGFD ecosystems, annual precipitation is the main factor determining NPP [26][27][28]. In addition, the stability of productivity is closely related to the annual variability of climate [19,20,26,28]. We hypothesize that the stability of productivity is significantly correlated to both annual mean precipitation and annual variability of precipitation at the global scale.

Study area
We defined the OGFD ecosystems as those areas in which the vegetation is dominated by grasses and forbs [9], and are divided into three main types: tundra, grassland, and savanna. We based the definitions on land cover data and its classes as defined by the International Geosphere-Biosphere Programme (IGBP, http://www.igbp. net) [29]. On a climatic gradient, these ecosystems are situated between forest (wetter) and desert or ice-land (drier) [13,14]. OGFD ecosystems provide important ecosystem services, such as sand fixation, soil and water conservation, and provide food for both wild and domesticated livestock. In aggregate, these ecosystems contribute to the livelihoods of more than 800 million people [13]. OGFD ecosystems are among the largest ecosystems in the world, with an estimated total area of 3.73 billion ha (37 300 000 sq km), which represents about 19% of the world's land area [9] (supplementary figure 1).

Dataset and pre-analysis 2.2.1. Climatic data
Monthly mean temperature and monthly total precipitation data were obtained from the Climatic Research Unit (CRU) for Time-Series Version 4.01 (TS 4.01) of High-Resolution Gridded Data. These data provide month-by-month climate variations with a resolution of 0.5 degrees (http://badc.nerc.ac.uk). These data were used to analyze the climatic impacts on the stabilities of OGFD ecosystems productivity at global and regional scales from 1982 to 2015.

Remote sensing data
The normalized difference vegetation index (NDVI) is an index of vegetation greenness and photosynthetic capacity, which is the normalized ratio of red and near-infrared (NIR) reflectance [28,31], widely used to detect the variability of productivity in the regional and global terrestrial ecosystem, most focused on Europe, the Sahel, Qinghai-Tibetan Plateau, and also the northern Hemisphere [2,26,[30][31][32][33]. To explore the response of the stability of productivity in OGFD ecosystems to climate change, we used the satellite-derived normalized difference vegetation index (NDVI) data from the Global Inventory Modelling and Mapping Studies (GIMMS) group. As an important indicator of productivity or vegetation growth, GIMMS NDVI data are widely used to study global vegetation productivity or other vegetation activities and their changes [3,36]. The annual maximum NDVI obtained from the monthly GIMMS NDVI in the period between 1982 and 2015, with a resolution of 0.083 degrees, was processed for analyzing the trend and the stability of OGFD ecosystem productivity, at both global and regional scales.

Land cover map
We used global land cover data from IGBP (http://www.igbp.net) [29], with a resolution of 8 km, to derive the global distribution map of OGFD ecosystems. The distribution map of grassland and savannas was derived directly from the 17 IGBP classification map [9]. We included grassland in Arctic Circle (66°33'N), defined as Tundra, to obtain three OGFD-dominated ecosystem types at a global scale [9]. OGFD ecosystems are mainly distributed in eight regions such as the Arctic Archipelago Region, the North American Plains and Cold Desert, mid-eastern South America, the seasonally arid African Plains, central Eurasia, the Mongolian Plateau, the Qinghai-Tibetan Plateau, and Oceania [9] (supplementary figure 1).
For analyzing the relationship between the climate measures and the stability and its changes in productivity of eight mainly distributed regions and global OGFD ecosystems, the NDVI and OGFD distribution maps were re-sampled with a 0.5-degree resolution to meet the resolution of climatic data.

Data analysis 2.3.1. Stability index (SI)
We defined the stability of OGFD ecosystems productivity as the ratio of mean annual maximum NDVI to its standard deviation. The SI was calculated in each pixel of OGFD ecosystems at a monthly scale across the period of 1982 to 2015.

Climatic measures
We calculated the annual total precipitation (ATP), the variation coefficient of precipitation (CVP), the annual mean temperature (AMT), and the standard deviation of annual mean temperature (SDT) based on the monthly mean temperature and monthly total precipitation of OGFD ecosystems. CVP was calculated as the standard deviation divided by the mean of annual precipitation for the full-time series [26].

Sliding-window analysis
Changes in stability index and climatic measures of OGFD ecosystems over time were compared using a slidingwindow analysis [26]. This technique was chosen to reduce the influence of outliers and highlight longer-term trends. We calculated 5-year windows by indexing one-time steps after each calculation and ending after the last complete 5-year window. We calculated sliding windows for the values of NDVI, the stability of NDVI values (SI), as well as the four climatic measures (ATP, CVP, AMT, and SDT). Consequently, the analysis of stability index and climatic measures using data from 1982 to 2015 begins in 1986, where the value for 1986 reflects the status over the previous 5 years.

Trend analysis
We detected a gradual change in the sliding-window stability index and the four climatic measures over all time steps at each pixel of OGFD ecosystems by a least-square linear regression model, as in many other studies [9-11, 34, 35]. We used the nonparametric Mann-Kendall (MK) test [32,[36][37][38][39] to estimate the abrupt points for all pixel values for sliding-window stability index of global OGFD ecosystems.

Regression analysis
Stepwise regression analysis was used to select the best model to explore the relationships between the stability index (SI) and the four climate measures in different eight regions and globally. All covariates were standardized and rescaled to a mean of 0 and unit variance. The Akaike information criterion [26,40] was used to select the best model by using the MASS package in the R 3.43 version.

The stability of OGFD productivity
From 1982 to 2015, the stability index of OGFD ecosystems productivity was relatively low across central Eurasia, the Mongolian plateau, and the Arctic Archipelago Region (figure 1). In contrast, the stability index of OGFD ecosystems productivity was relatively high across much of mid-eastern South America and New Zealand ( figure 1, supplementary figure 2). In general, the stability index of OGFD ecosystems productivity is positively correlated with NDVI (R 2 = 0.539, supplementary figure 3(1)) and negatively correlated with the standard deviation of NDVI (R 2 = 0.396, supplementary figure 3(2)) according to the summarized data from eight major OGFD ecosystem distribution regions.

Impacts of climate change on the NPP stability in OGFD ecosystems
We selected the best-fit linear model to indicate the correlation of four key climate measures, including annual total precipitation, the variation coefficient of precipitation, annual mean temperature, and standard deviation of annual mean temperature with the stability index of OGFD ecosystems from 1982 to 2015 (see Methods). The results showed that the stability of productivity was most strongly correlated with temperature (table 1).
The impacts of climate change on the stability of NPP were analyzed by stepwise regression in each region with different trends and globally (table 1 and supplementary table 1). In locations with significant increasing trends of stability index from 1982 to 2015, annual mean temperature and variation coefficient of precipitation were the dominant drivers of stability index change (p < 0.001). Stability index was positively correlated with annual mean temperature while negatively correlated with the variation coefficient of precipitation, but the specific correlations varied regionally. Specifically, we observed three types of responses (supplementary table 1). First, in those locations where temperatures were warming, the stability index increased. Second, in locations where precipitation became less variable, the stability of NPP increased. Third, the stability index was has  negatively correlated with annual mean temperature in locations with significantly decreasing trends of stability index. In total, these patterns show that where the temperature had warmed the most had decreasing stability in NPP. Moreover, the standard deviation of annual mean temperature had a negative impact harmed on the stability index in locations (Asia, Europe) that experienced a significant decrease of stability index (p = 0.002) (supplementary table 1). Finally, in locations that experienced insignificant trends of stability index from 1982 to 2015, the stability index was negatively correlated with both the variation coefficient of precipitation and standard deviation of annual mean temperature (p < 0.001).
For different types of OGFD ecosystems globally, the stability index was negatively correlated with both annual mean temperature and standard deviation of annual mean temperature (p < 0.001) in Grassland, and negatively correlated with both annual total precipitation and annual mean temperature (p < 0.001) in Tundra (supplementary table 2). In the Arctic Archipelago Region, mid-eastern South America, central Eurasia, the Mongolian Plateau, and the Tibetan Plateau, we observed that the increases in annual mean temperature are correlated with lower stability of productivity in OGFD ecosystems. In mid-eastern South America, central Eurasia, and Oceania, we observed that increasing standard deviation of annual mean temperature results in low stability of productivity in OGFD ecosystems (supplementary table 1). In the North American Plains and Cold Desert, Mongolian Plateau, and Oceania, we observed that the increases in annual total precipitation lead to higher stability of productivity. In the North American Plains and Cold Desert and Mid-eastern South America, we observe increasing variation coefficient of precipitation results in low stability of productivity in the OGFD ecosystem (supplementary table 1).

Regional change of the NPP stability
Many studies have confirmed the enhanced global vegetation activity and terrestrial ecosystem productivity in recent years [5-8, 23, 24], especially in the Northern Hemisphere [7][8][9]. Due to the inconsistency of periods, the increasing trends of ecosystem productivity in different regions may not be completely consistent in recent years [1][2][3][4]32]. It is worth noting that studies analyzing trends of global increasing vegetation productivity have shown a slowing of this trend between 1998 and 2012 [41]. It could be a saturation of ecosystem capacity to increase productivity. This shift in ecosystem productivity may be caused by the changes in climatic conditions (temperature and precipitation) and human activities (e.g., grazing) [7,9,16,42]. Whatever the cause, the year 2007 was a turning point when trends in NPP significantly changed in the Eurasian steppe region [11]. However, little attention has been given to the stability of productivity, and the stability of productivity has generally decreased (P < 0.05) across the world. Nevertheless, both regions with a negative trend in stability (36% of OGFD ecosystems area) and regions with a positive trend in stability (27% of OGFD ecosystems) trends exist from 1982 to 2015 ( figure 3(b)). The geographic distribution of producivity reveals a spatial pattern of both increasing and decreasing trends of NPP stability index across the world ( figure 3). Declines in the NPP stability index of OGFD ecosystems were observed in most areas of the northern hemisphere, especially in the Mongolian plateau (over 65% of the area). The grassland productivity on the Mongolian Plateau also showed a decreasing trend [9], and the changes in climate factors and overgrazing activities may lead to reduced stability and increased fluctuations in grassland productivity on the Mongolian Plateau [9,20,32]. In contrast, patterns of increasing productivity dominated the seasonally arid African Plains and Oceania ( figure 3(a), supplementary  figure 5), although there have both positive and negative trends in stability of OGFD ecosystems productivity in the southern hemisphere ( figure 3(a)). It may be that the relatively stable and beneficial change of climate are resulting in the higher and more stable ecosystems productivity in the Southern Hemisphere [14,19,28,36,43], where mainly distributed the grasslands and Savannah ecosystems [12,13]. Table 1. Model selection results for linear regression models of stability index (SI) for productivity in locations with differing trends. SI, stability index of OGFD ecosystems productivity; ATP, annual total precipitation; CVP, the variation coefficient of precipitation; AMT, annual mean temperature; SDT, the standard deviation of annual mean temperature.

Interpreting the importance of the observed trends
In recent years, large-scale vegetation growth and its response to climate change have received extensive attention from scientists and policymakers [5,25]. Drought reduces vegetation productivity at global and regional scales [2,18,44]. Precipitation is the most important climatic factor affecting annual productivity and the stability of the grassland ecosystem on the Mongolian Plateau [20]. Variation in precipitation is a robust explanatory climate variable for predicting NDVI and acts as a key limiting factor on grazing livestock density by constraining pasture productivity [26]. We found that annual total precipitation was positively correlated with the NPP stability, while variation in annual total precipitation was negatively correlated with that of OGFD ecosystems in most regions in this study (supplementary table 1). It is worth noting that studies have shown that the global average surface temperature warming rate between 1998 and 2012 was significantly slower than years prior [41,45]. Over the same period (1998-2012), net increases in the rate of global vegetation productivity have decreased [41]. We also found a trend shift in the stability of productivity in global OGFD ecosystems with a lag of 2 to 3 years from the start of the slow warming period ( figure 2, supplementary figure 4). Lags in response times may come about because of ecosystem carryover effects, such as underground nutrient storage, or be expressed by lagged changes in herbivore populations.
In-situ warming experiments showed that warming reduces the stability of the plant community biomass production on the Qinghai-Tibetan Plateau [15]. In contrast, warming has been shown to increase the productivity of the grassland ecosystem in the USA [21]. Our results showed that in 27% of global OGFD ecosystems, climate warming induced an increase in the stability of that productivity ( figure 3; table 1). However, in other regions, warming reduces the stability of productivity in OGFD ecosystems ( figure 3; table 1). In these regions, the standard deviation of annual mean temperatures is negatively correlated with stability of net primary production within global OGFD ecosystems (figure 3; table 1; supplementary tables 1 and 2). These results suggest that the threat of climate change may both impact the productivity of critical ecosystems as well as change the predictability of that productivity with projections of a warmer future [42,46]. This is particularly important in ecosystems where that productivity is used for grass-based animal husbandry.

Uncertainties in this study
We have provided measures of the temporal stability of OGFD ecosystems productivity responses to global climate change based on the GIMMS3g NDVI and CRU TS4.01 data. There may be having certainly uncertainty due to a single remote sensing dataset, and also there are the main limitations that we mainly reported on the pattern observed and only speculated on the mechanisms driving these changes in this study. The integrated analysis of multiple long-term time series data should be carried out to reduce the uncertainty caused by a single data source in the future. More detailed divisions of mainly distributed regions in OGFD ecosystems should be also considered in the future study for reducing the uncertainty.
In addition, this study focus on the relationship between stability indexes with four climate measures derived from monthly and annual precipitation and temperature, more climate metric covering various time scales main have different impact on the stability of OGFD ecosystem productivity. The frequency and magnitude of extreme climate events increase with global warming [42], the compound disasters such as droughts and heat waves may cause severe risks on the structure, function, and stability of OGFD systems. Moreover, the mechanisms underlying the changes in the productivity stability and its responses to climate change in OGFD ecosystems should be needed to verify and deeply analyze by long-term controlled experiments and multimodel ensembles in the future study.

Conclusion
In summary, we show trends in the stability of productivity in OGFD ecosystems from 1982-2015 decreased while global vegetation productivity has increased over that period. We also found that warming, specifically, is correlated with the decline in the stability of productivity in global OGFD ecosystems. Some of these results are inconsistent with some in situ experimental results. These inconsistencies may emerge from challenges with a spatial and temporal resolution of global data sources or from variations in system responses regionally, and the specific locations in which these experiments have been conducted. For example, changes in patterns of precipitation do not significantly predict the stability of biomass temporal stability in alpine grassland. Further, increased warming is related, experimentally, to increases in the stability of OGFD ecosystems production in the USA. Thus, additional data integration across scales and among geographies, from field experiments to remote sensing studies is needed to better understand climatic impacts on important attributes of the OGFD ecosystems productivity under global change.