Abstract: Net primary productivity (NPP) plays an important role in global carbon cycle and is important to understanding drivers of climate changes. Precise and rapid estimation of vegetation NPP is important for evaluating ecological carrying capacity at a regional scale and managing natural resources reasonably. In this study, field-measured grassland above ground biomass (AGB) from 2011 to 2014, MODIS remote sensing data and meteorological data in Gannan prefecture were used. In combination with the ratio of belowground biomass to AGB, we calculate the grassland NPP, and evaluate the accuracy of the MOD17A3 product and Carnegie-Ames-Stanford approach (CASA) model, and analyze the dynamic changes of grassland NPP from 2000 to 2016 using the better method. The results show that the accuracy of grassland NPP predictions from the CASA model (root mean square error (RMSE) = 9.94 g C ·m^-2·yr^-1) is higher than that of MOD17A3 product. The average annual grassland NPP determined by the CASA model shows a decreasing trend from the southwest to northeast between 2000 and 2016 in our study area. Comparing different vegetation types, the annual NPP for marsh grassland (469.07 g C ·m^-2·yr^-1) was the highest, while that of temperate steppe grassland was the lowest (324.18 g C ·m^-2·yr^-1). In addition, the annual NPP of alpine meadow and alpine shrub meadow grasslands (which have relatively large area in Gannan prefecture) was, respectively, 370 and 430 g C ·m^-2·yr^-1. Over the past 17 years, the annual grassland NPP was generally stable in most regions, (75.31% of the total grassland area). Meanwhile, an increasing NPP trend was seen in 22.63%, and a decreasing trend in just 2.06% of the Gannan prefecture area. These results suggest that the CASA model has an important role in grassland NPP estimation and will assist in the sustainable management of grassland resources in alpine areas.

Key words: Gannan prefecture, grassland net primary productivity (NPP), MOD17A3 product, CASA model, dynamic change