Assessing Random Forest and SARIMA Models for Wet Period Rainfall Prediction in Northern Territory, Australia

The present study investigates the simultaneous effect of climate indices on wet rainfall in the Northern Territory (NT). To achieve this, a comprehensive set of input variables was employed, including lagged indices of the Indian Ocean Dipole (IOD), El Niño-Southern Oscillation (ENSO), and MaddenJulian Oscillation (MJO). These climatic predictors were utilized to calibrate and validate the Random Forest (RF) and Seasonal Auto-Regressive Integrated Moving Average (SARIMA) models, ensuring robust model performance and accurate rainfall forecasting. To evaluate the model’s performance, three statistical measures were used: mean absolute error (MAE), correlation coefficient (r), and root mean square error (RMSE). Among the large-scale climate factors considered, MJO, SOI, and NINO 3.4 were identified as having a significant impact on the Northern Territory’s wet period rainfall forecast. For NT forecasts at one selected station, the RF model demonstrated the most favorable outcomes. Similarly, the SARIMA model displayed a root-mean-square error (RMSE) that ranged from 8.64 mm to $\mathbf{6 0. 6 5 \mathrm {mm}}$ across the selected site. In contrast, the RF model demonstrated better performance, exhibiting an RMSE range of $\mathbf{5. 2 4} \mathbf{~ m m}$ to $\mathbf{4 9. 3 3} \mathbf{~ m m}$.