Comparing Rainfall Prediction at Various Time Scales and Rainfall Interpolation at the Regional Scale Using Artificial Neural Networks

Abstract Precipitation prediction has important applications in a variety of industries, including agriculture, water conservation, and transportation. In previous study, analyses of rainfall prediction were rarely carried out separately from the spatial and temporal scales. The rainfall data of every observation point in Zhejiang Province in 2015 were compared on a spatial scale using Inverse Distance Weighted Interpolation (IDW), Kringing Interpolation, Moving Cube Interpolation (MC), Natural Neighbour Decomposition Interpolation (MSM), Nearest Neighbour Interpolation (NeaN), and Bilinear Interpolation (NN) in order to study the distribution trend of rainfall in the provincial area. The daily and monthly rainfall projections were simulated on a temporal scale using historical precipitation data from a Zhejiang Province observation site covering 30 years, from 1990 to 2020, as an example, using the RBFN and BP models. It is verified that the simulation and interpolation results generated by the two processes are accurate. In the realm of rainfall prediction, artificial neural networks (ANN) offer some advantages over the conventional classical interpolation method. Nevertheless, ANN is also constrained by nonlinear rainfall characteristics, training time, computational complexity, and the arrangement and quantity of meteorological stations, among other factors. The findings show that RBFN outperforms the BP model in rainfall simulation, particularly when it comes to predicting longer time horizons. The essay concludes with a discussion of the possible applications and future prospects of artificial neural networks (ANN) in the field of rainfall prediction at both geographical and temporal dimensions.