Geospatial Analysis of NDVI-Rainfall Dynamics under High ENSO Influence in Peninsular Malaysia
DOI:
https://doi.org/10.11113/jagst.v5n1.103Keywords:
CMORPH, ENSO, Geographically Weighted Regression, NDVI, Peninsular Malaysia, RainfallAbstract
Malaysia (PM) rainfall varies significantly due to El Niño-Southern Oscillation (ENSO), making it an important region to study the relationship between NDVI and rainfall. These connections are complex, spatially non-linear, non-stationary, and scale-dependent, challenging conventional regression models. To address this, a local modelling approach, Geographically Weighted Regression (GWR) was employed, accommodating spatial variability in relationships. This study utilizes the CMORPH gridded dataset to explore the NDVI-rainfall relationship in PM during very strong El Niño events of 2015/2016, as well as strong La Niña events of 2010/2011. During the El Niño 2015/16 event, the Gaussian weighting function achieved an AIC of 398.48 and a quasi-global R² of 0.63, outperforming the Bisquare function's AIC of 434.12 and quasi-global R² of 0.48. In the La Niña 2010/11 event, the Bisquare function excelled with an AIC of 442.01 and a quasi-global R² of 0.52, while the Gaussian recorded an AIC of 505.69 and a quasi-global R² of 0.10. The median local R² for El Niño (0.6 to 0.8) was higher than that for La Niña (some areas dropping below 0.4), highlighting the GWR model's superior performance in capturing spatial variation. In terms of predictive power, the metrics demonstrate superior model performance during El Niño, with a Mean Absolute Error (MAE) of 0.5, a Root Mean Square Error (RMSE) of 0.66, an R² value of 0.63 indicating significant variance explained, and Nash-Sutcliffe Efficiency (NSE) and Kling-Gupta Efficiency (KGE) both at 0.63 and 0.6, respectively.