Grasslands are one of the most widely distributed land cover vegetation types across the globe. They play a significant role in developing animal husbandry, protecting biodiversity, maintaining soil and water, and keeping ecological balance. Estimating grassland production, a fundamental variable in grassland resource management, is helpful to measure grassland productivity and diagnose its health status. In recent years, the combination of remote sensing and ground measurements into models has become an important method of estimating grassland production. Normally, large number of measurements are required for remote sensing modeling. Mongolia is an example of a traditional grassland animal husbandry country with the largest per capita grassland area in the world and is also part of the China-Mongolia-Russia Economic Corridor under the “Belt and Road” initiative. Constrained by multiple factors of overseas sampling, it is usually difficult to obtain sufficient, accurate, and evenly distributed production samples. Thus, the accuracy of estimation models will be affected. Until now, there is still no effective solutions to get more samples. In this study, a 200-kilometer buffer zone along the China-Mongolia Railway (Mongolia) was taken as the study area. Given the inhomogeneity of grassland distribution and the correlation between the samples, the Point estimation model of BSHADE (P-BSHADE) was introduced. We derived the grassland production dataset in the study area from 2000 to 2019 based on the sample measurements and interpolated samples, and a combination of Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), and Net Photosynthesis (PsnNet) for remote sensing modeling. Our method extrapolated sparse and unevenly distributed sampling points to supplement ground information by spatial interpolation, and used both the measured sample points and interpolated sample points for modeling. Six types of linear models and exponential models were established using above three vegetation indices. Our results show that the accuracy of the optimal model was 80%, higher than that from previous studies. The spatial pattern and interannual variation of grassland production estimated in our study were consistent with previous studies, which further confirmed the accuracy of our results and the feasibility of the interpolation method. Using interpolation method to optimize the data source is an entirely new attempt that improve the accuracy of the model estimation, which could be potentially applied to other overseas regions to monitor grassland resources.