The CASA model is widely used in estimating vegetation net primary productivity (NPP), but its accuracy needs to be improved. Based on the geographical factor regression method (AMMRR) and land surface water index (LSWI), this study aimed to correct the temperature stress coefficient and water stress coefficient, two key parameters of CASA model. Next, we continued to estimate NPP and analyze the effect of the correction on vegetation and on the relationship between NPP and other factors. Results showed that:(1) The correction could effectively improve the estimation accuracy of the CASA model. The amount of corrected NPP was 34.29 TgC/a, but the original NPP was 34.52 TgC/a; therefore, the NPP of the original model was overestimated by 0.23 TgC/a. (2) This method can not only correct the influence of terrain on NPP, but also corrects the impact of human activities areas on NPP under flat terrain. In areas with high altitude, large topographic relief, as well as in human activities areas, the correction had a large impact on NPP estimation, and the original model of the oasis area was overvalued. (3) The effect of correction on growth season was greater than that during the non-growth season. Slope had a substantial influence on NPP, and the higher the slope was, the greater the overestimation of the original model. Before correction, sunny slope NPP values were overestimated, whereas, the shady slope NPP values were underestimated.