Prediction of Daily Precipitation Based on Deep Learning and Broad Learning Techniques

Precipitation forecasting, which is closely related to economic development and human life, is a hot topic in meteorological research. It is well known that the weather system is extremely complex and the precipitation intensity is affected by many meteorological variables. Hence, traditional techniques can only provide a short term prediction. In the study, the National Oceanic and Atmospheric Administration (NOAA) dataset, which contains the daily weather information that is recorded by over 100,000 observation sites including the whole word from 1800 to 2017, is employed to examine the prediction of daily precipitation. We utilize the State-of-the-art machine learning techniques, that is Deep Learning (BPNN, CNN, LSTM) and Broad Learning System (BLS) techniques, to predict daily precipitation intensity. Then, the performance of these techniques is evaluated by four criterions (Accuracy, Precision, Recall, F1-score). Experiments show that the completion and cleaning reconstruction of real-world data is conducive to improving the prediction results. Compared to Deep Learning, the Broad Learning System has a better performance in precipitation prediction.