Rainfall and runoff time-series trend analysis using LSTM recurrent neural network and wavelet neural network with satellite-based meteorological data: case study of Nzoia hydrologic basin

Abstract This study compares LSTM neural network and wavelet (WNN) for spatio-temporal prediction of rainfall runoff time-series trends in scarcely gauged hydrologic basins. Using long-term situ observed data 30 years (1980–2009) from ten rain gauge stations three discharge measurement stations, the Nzoia River basin are predicted through satellite-based meteorological comprising of: precipitation, mean temperature, relative humidity, wind speed solar radiation. The modelling was carried out sub-basins corresponding to stations. WNN were implemented with same deep learning topological structure consisting 4 hidden layers, each neurons. In five parameters using WNN, both models performed well respective R 2 values 0.8967 0.8820. MAE RMSE measures predictions ranged between 11–13 m 3 /s monthly prediction. With data, within = 0.8610 as compared 0.7825 WNN. trend 9 11 mm, while varied 15 21 mm. performance improved increase number input parameters, which corresponded size sub-basin. terms computational time, converged at lowest nearly epochs, taking slightly longer attain minimum RMSE. shows that basins scarce hydrological monitoring networks, use suitable spatial temporal analysis trends.