Evaluating the performance of CHIRPS and CPC precipitation data for streamflow forecasting using multiple linear regression and Long Short-Term Memory Neural Network model

Accurate streamflow forecasting is essential for effective water resources management and planning. Traditionally, streamflow prediction has relied heavily on a large volume of precipitation data from ground-based weather stations, which are often expensive to build and maintain and provide rainfall data at a coarse spatial resolution. However, machine learning techniques have introduced cost-effective tools for streamflow forecasting that require minimal input data. This study evaluates two prediction models utilizing gauge-based (CPC) and satellite-based (CHIRPS) rainfall data for the Wolf River watershed, comparing their effectiveness in streamflow forecasting. Employing Multiple Linear Regression (MLR) and Long Short-Term Memory Neural Network (LSTM-NN) models, daily precipitation data from Climate Hazards Group InfraRed Precipitation with Station (CHIRPS) and NOAA Climate Prediction Center (CPC) spanning the period from 1991 to 2021, were used for streamflow prediction. The models were developed using daily stream flow data from the Wolf River Watershed stream gauge USGS 07031650 during the specified timeframe. Results indicate that CHIRPS data outperforms CPC data when used with the LSTM-NN model, yielding lower Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) values at 15.02 and 21.53, respectively. Therefore, CHIRPS data emerges as a viable alternative rainfall data source for this study area in scenarios where gauge-based data is unavailable. The main contribution of this study includes:•Demonstrating CHIRPS as a viable alternative to gauge-based CPC data for streamflow forecasting.•Showing that LSTM-NN outperforms MLR in streamflow prediction, achieving lower RMSE and MAE values.