D-Markov: A Sparse Sample-based Model for Interannual Precipitation Prediction during the Rainy Season

A timely and accurate precipitation prediction method is necessary to reduce the huge loss from heavy precipitation during the rainy season. In recent years, data-driven machine learning model has been developed rapidly because of its high efficiency and low-cost. The greater the amount of data, the higher the prediction accuracy of machine learning model. However, emerging machine learning models cannot be directly used because the sparse data of interannual precipitation cannot offer enough data for training. In this study, we propose a sparse sample-based, efficient, and low-cost model, i.e., D-Markov, to forecast rainy season precipitation accurately. First, to resolve the problem of sparse samples, the Markov model is used to capture the state transition of samples. However, the prediction accuracy of the Markov model cannot fully satisfy the requirements of forecast, especially for abnormal precipitation. Therefore, in accordance with the hidden oscillation of rainy season precipitation, a wavelet-guided ensemble empirical mode decomposition algorithm is introduced into the Markov model. In this manner, prediction accuracy is dramatically improved because the ensemble empirical mode decomposition algorithm can successfully extract some hidden oscillation modes i.e., intrinsic mode functions(IMFs), which are usually ignored by the traditional Markov model. The effectiveness and efficiency of the D-Markov model are demonstrated through comparison with the BCC_CSM1.1(m) system provided by authorities from the China Climate Center.