<p>This research is designed to improve bi-monthly and seasonal rainfall predictions in Ethiopia during the Kiremt season (June–September), which is critical for agriculture. The multiple linear regression (MLR) and nonlinear autoregressive network with exogenous inputs (NARX) forecasting models are used in this study. Monthly rainfall data or Enhancing National Climate Services (ENACTS) and four predictors SST3.4, ERA5 total precipitation, and zonal wind data at the 850 and 200 mb levels from 1990 to 2020 are used to improve rainfall forecasting. The model performance was evaluated using root mean square error (RMSE), mean absolute error (MAE), bias percentage calculation (PBIAS), Nash–Sutcliffe efficiency coefficient (NSE), and the coefficient of determination (<i>R</i>²). The MLR and NARX models explained 87 and 86% of the variance in rainfall data, respectively. The spatial averages of RMSE and MAE are 17.1 and 14.1 units, respectively. The models performed variably across stations and time periods, with the NARX model proving more accurate at select sites and during wet seasons, while MLR produced more consistent findings in other regions. Both models performed well in predicting rainfall on a seasonal and bi-monthly basis. The findings help to improve Ethiopian agricultural planning, hydrological management, and early warning systems.</p>