Abstract: Abstract: With the rapid development of remote sensing technology, large-scale and high timeliness satellite products provide digital, quantitative, and mechanistic support for agricultural production. To evaluate the accuracy and uncertainty of vegetation products retrieved by remote sensing better, the sampling design is very important in the process of ground measurement experiment for validation in heterogeneous vegetated areas. In this study, the remote sensing image was regarded as the prior knowledge, the initial sampling points were selected by the K-means algorithm, and the optimal sampling scheme was planned by Spatial Simulated Annealing (SSA) algorithm. Then, the research scheme was verified by the field data of the same period. Based on the prior knowledge and geostatistics theory, it provided a strong theory for the sampling scheme. The essence of spatial simulated annealing algorithm is to search randomly, transfer state, accept (or discard) new solutions before the cooling cut-off, to find the optimal combination. By constantly jittering the new sampling combination, it jumped out of the local optimal solution, avoided the randomness of sampling, and could find more satisfaction. It meant that the initial positions of sampling points determined by stratified sampling were constantly combined and changed. Finally, the optimal combination that minimizes Kriging variance was obtained. Compared with other sampling schemes, it could be concluded that the SSA had stable advantages on different sampling numbers, the sampling accuracy was less affected by the number of samples, and the sampling combination with lower prediction error could also be found when the sample numbers were small. Under the condition of ensuring the sampling accuracy, the sampling quantity was obviously less than the traditional sampling scheme, which effectively reduced the sampling cost. The representativeness and accuracy of sampling points were evaluated by the relationship between sampling points and population, the scale of the trend surface and the real surface sample site. From the aspect of geostatistics, the sampling points obtained by SSA had better simulation ability to the sample population; From the aspect of Kriging interpolation, the Kriging variance of the sampling points optimized by SSA was 3-4 orders of magnitude higher than that of the traditional sampling points. The root mean square error between the interpolation surface and the image surface of the two sample areas based on the SSA algorithm was 3.102 6 and 2.962 7, respectively, and the Pearson correlation coefficient was 0.45 and 0.73, respectively. Compared with the other three sampling methods, the result of SSA was the smallest root mean square error and the highest Pearson correlation coefficient. Compared with random sampling, systematic sampling, and threshold segmentation sampling, the correlation between interpolation surface and image surface based on SSA improved by 29%, 30%, and 6%, respectively; the Pearson correlation coefficients of the interpolation points based on SSA and the measured points were 0.601 and 0.757, respectively, which were higher than those of the other three sampling methods. Compared with random sampling, systematic sampling, and threshold segmentation sampling, the correlation coefficients of interpolation points and measured points based on SSA increased by 0.23, 0.14, and 0.07 on average. It was proved that SSA could provide a reliable and optimized sampling strategy for the ground experiment of validation.