Abstract: The automatic ground-based observation of cloud is an exigent requirement for the current meteorology operation. Up to now, the detection algorithms for the groundbased cloud observation are mainly based on the threshold, but neither the fixed threshold nor the global threshold method can achieve satisfactory groundbased cloud detection effects. Using the properties of blue sky background versus white cloud in clear sky, an automatic ground based cloud detection method is presented based on the local threshold interpolation in which the original cloud image is resized to an appropriate size and then the normalized difference operation is performed on the blue band and the red band of the resampled image. After that, the normalized difference result is separated into a series of quadrilled subimages according to the spatial position of the image pixels automatically. Next, the improved maximum interclass variance adaptive threshold algorithm and some decisionmaking rules are used to compute the local threshold for each subimage, and,using the bilinear interpolated algorithm, the threshold array is interpolated to form a curved surface whose size is the same as the original cloud image. Finally, the curved threshold surface is used to finish groundbased cloud detection by comparing to the normalized difference result of the blue and red bands pixel by pixel. Compared with the fixed threshold and the global threshold algorithms, the proposed method obtains better detection effects for clouds in small, broken bits and weak contrast clouds. The quantitative assessment results show that the fixed threshold algorithm has a much lower correctness and accuracy than the global threshold method and the local threshold method. Furthermore, the proposed method acquired better results than the global threshold algorithm both in correctness and accuracy.