Monthly geographically weighted regression between climate and vegetation in the Eastern Cape Province of South Africa: Clustering pattern shifts and biome-dependent accuracies

Abstract

Remote sensing provides cost-effective and unbiased data and thus is ideal for assessing climate–vegetation relationships. Such relationships can be quantified using geographically weighted regression (GWR) approach to account for variations of the relationships across space. This approach was applied in the Eastern Cape province of South Africa that is rich in biodiversity hosting 10 of the country's 11 biomes. The study aimed to determine if the GWR accuracy for relating Enhanced Vegetation Index (EVI) with rainfall and Land Surface Temperature (LST) shows an optimal pattern with time and space. and to explore if the correlation of EVI with rainfall and LST varies with biome type. Monthly data covering February 2000 to December 2017 were used for the three variables. The coefficient of determination (R2) was greater than 0.5 for 75% of the locations, with month-to-month change of R2 exceeding 25% for many locations. Optimized Hot Spot Analysis returned well-defined broad clusters of high and low R2 values separated by clusters of randomly distributed R2 values. These clusters shifted with month, further stressing the benefit of modelling at the monthly scale. Assessment of R2 by biome showed the importance of biomes in characterizing GWR of climate and vegetation, with better correlations found in low biodiversity (Succulent Karoo and Nama-Karoo biomes) than in higher biodiversity (Forest and Indian Ocean Coastal Belt biomes) zones. Further, the estimation residuals of the Forest Biome varied significantly from 3 to 5 other biomes across the year indicating the complex interaction of this biome with rainfall and LST. The study encourages further research by using high temporal resolution data for detailed monitoring within the GWR framework.