Author(s): Rashid Farooq; Monzur Alam Imteaz

Linked Author(s):

Keywords: Climate indices; CatBoost; Machine learning; Rainfall; SARIM

Abstract: Accurately predicting seasonal rainfall variations in the Northern Territory (NT) is critical for water resource management. These variations are complex, influenced by several influential climatic anomalies. In this study, a supervised machine learning (ML) and statistical model were employed to delineate seasonal rainfall in two distinct stations within the Northern Territory, Australia, utilizing monthly dataset from 1900 to 2023. This research attempts to find a nonlinear relationship between the NT wet-period rainfall and the lagged climate indices, using seasonal autoregressive integrated moving average (SARIMA) and CatBoost model, incorporating input datasets featuring various climate indices. Comparative assessments, utilizing metrics such as root mean square error (RMSE), Mean Absolute Error (MAE) and correlation coefficient (R), indicated a substantial enhancement in model performance with the statistical approach. It was observed that SARIMA model can provide higher correlations using the lagged indices for forecasting wet-period period rainfall in compared to ML method. Using these indices in SARIMA, the model correlation in the testing phase increased up to 83%, and 77% for the two case study stations of Alexandria and Anthony Lagoon in NT, respectively.

DOI: https://doi.org/10.3850/iahr-hic2483430201-117

Year: 2024