Data granularity is one of the most important issues of parallel computing based on large volume of spatial data. After a comparison with the different types of terrain analysis algorithms, a Geo-Data Granularity Model (GDGM in short) was proposed, which can be used for the quantitative description of data partition granularity in parallel computing process of massive spatial data. In this algorithm, according to the parallel evaluation method, the parallel data granularity was adaptively adjusted and finally an optimized data grain was obtained. The execution time of each parallel computing was recorded for the comparison of computing efficiency values from different data granularities. Furthermore, by means of the comparison, a dynamic algorithm was designed for the dynamic scheduling of different data granularity so that the optimal performance of specific algorithm was achieved. The preliminary experiments show that the algorithm has much better efficiency and portability than the traditional ones so far.