cuSTSG is a GPU-enabled spatial-temporal Savitzky-Golay (STSG) program based on the Compute Unified Device Architecture (CUDA). Firstly, the cosine similarity between time-series data of adjacent years is used to identify similar years without land cover type changes, hence to exclude the years with inaccurate quality flags from generating NDVI seasonal growth trajectory. Secondly, the computational performance is improved by reducing redundant computations, and parallelizing the computationally intensive procedures using the CUDA for GPUs. - View it on GitHub