Measuring the impact of livestock grazing on grassland above-ground net primary production (ANPP) is essential for grass yield estimation and pasture management. However, since there is a lack of accurate and repeatable techniques to obtain the details of grazing locations and stocking rates at the regional scale, it is an extremely challenging task to study the influence of regional grazing on the grassland ANPP. Taking Zoige County as a case, this paper proposes an approach to quantify the

more »

... tial and temporal variation of grazing intensity and grazing period through time-series remote sensing data, simulated grassland ANPP through the denitrification and decomposition (DNDC) model, and then explores the impact of grazing on grassland ANPP. The result showed that the model-estimated ANPP while considering grazing had a significant relationship with the field-observed ANPP, with the coefficient of determination (R 2 ) of 0.75, root mean square error (RMSE) of 122.86 kgC/ha, and average relative error (RE) of 8.77%. On the contrary, if grazing activity was not considered in simulation, a large uncertainty was found when the model-estimated ANPP was compared with the field observation, showing R 2 of 0.4, RMSE of 211.51 kgC/ha, and average RE of 32.5%. For the whole area of Zoige County in 2012, the statistics of the estimation showed that the total regional ANPP was up to 3.815ˆ10 5 tC, while the total regional ANPP, without considering grazing, would be overestimated by 44.4%, up to 5.51ˆ10 5 tC. This indicates that the grazing parameters derived in this study could effectively improve the accuracy of ANPP simulation results. Therefore, it is feasible to combine time-series remote sensing data with the process model to simulate the grazing effects on grassland ANPP. However, some issues, such as selecting proper remote sensing data, improving the quality of model input parameters, collecting more field data, and exploring the data assimilation approaches, still should be considered in the future work. are subjected to serious progressive degradation [5] so that the dynamic balance of the carbon cycle in grassland ecosystems has been broken [6] . Dynamically monitoring the ecological environment changes of grassland ecosystems is critical to understand the role of grasslands in the global carbon cycle and is desirable for the local government to manage the grasslands resource. Primary production represents the major input of carbon and energy into ecosystems [7] and above-ground net primary production (ANPP) is considered as an integrative variable of the function of terrestrial ecosystems because of its relationships with animal biomass, secondary productivity, and nutrient cycling [8, 9] . Grassland ANPP is one of the most direct indicators of grassland ecological environment status [10] . Methodologically, grassland ANPP could be estimated through the ecosystem models, which have always been regarded as essential and indispensable tools in simulating net primary production (NPP), owing to their ability to describe the response of the grassland ecosystem to changing environmental conditions and human disturbances [11] . Since the 1980s, a variety of process-based ecological models have been developed to estimate grassland ANPP [12] . Among them, denitrification and decomposition (DNDC) is considered as one of the most widely used and successful biogeochemical models [13] , and it has been applied to simulate the amount and dynamics of carbon for almost all terrestrial ecosystems [14] . At the International Workshop on Global Change for Asia Pacific Region in 2000, the DNDC model was designated as one of the biogeochemical models applicable for the Asia Pacific region [15] . Although the DNDC model provides a powerful tool to estimate ANPP, significant uncertainties still exist when it is used to simulate the regional ANPP for grasslands because of the human disturbances, especially the grazing impact. Grazing could significantly influence primary production, vegetation composition, and root biomass [16] , and it would have a more direct and rapid impact on standing biomass than other factors, such as management practices, edaphic conditions, and climate in the short term [17] . Therefore, how to quantify the variation of the grazing effects on grassland ANPP is critical to simulate regional grassland ANPP. Despite a lot of debate about whether or not grazing can increase grassland ANPP at the individual site [18], due to the lack of sufficient historic grazing data, there have not been enough systematic and comprehensive investigations conducted on the regional grazing effects by far, especially in the nomadic or semi-nomadic pastures. Although the grazing data derived from the "Gridded Livestock of the World" (GLW) data could be used to model the grazing effect on dry grassland carbon cycling, it was too coarse to reflect the spatial variation of grazing intensity [19] . Additionally, because of the large area for livestock grazing, the relatively long duration of active interaction between animals and plants, and the difficulty encountered in measuring the forage consumed by free-ranging animals [20] , it is even hard to assess regional grazing effects in free grazing systems. A significant obstacle to quantifying the regional grazing effects is the difficulty in getting the data regarding the grazing locations and stocking rates through conducting grazing experiments at a large enough scale [21] . Moreover, there is a lack of accurate and repeatable techniques to obtain the details of grazing intensity and grazing period at regional scale [22] . Limited regional monitoring makes it difficult to pinpoint the regional impact of livestock on the grassland ANPP [23] . With the development of satellite and computer technologies, remote sensing techniques have become highly promising tools in monitoring spatial and temporal changes of the grassland ecosystem at regional scales with rapid data acquisition and at lower cost [24] . In view of the different vegetation status between grazing and non-grazing areas, and the good relationships between the vegetation indices and vegetation coverage, the vegetation indices are usually used to explore the ability of remote sensing data to quantify the grazing intensity or evaluate the grazing effects. For example, Kawamura et al. (2005) investigated the spatial distribution of grazing intensity based on the Normalized Difference of Vegetation Index (NDVI) and the tracking data recorded by global positioning system (GPS) [25]. Blanco et al. (2009) applied the NDVI to detect the spatial and temporal patterns of vegetation in two different grazing systems (i.e., continuous and two-paddocks rest-rotation) [22]. Yu et al. (2010) estimated the grazing capacity using the NDVI, above-ground biomass data, and theoretical livestock carrying capacity [26]. Li et al. (2014) identified the mowing and