USE OF GIS GEOGRAPHICALLY WEIGHTED REGRESSION TO DETERMINE NATURAL RUBBER PRODUCTIVITY AND THEIR DRIVING FORCES: A CASE STUDY FROM THE KALUTARA DISTRICT OF SRI LANKA

Abstract

The goal of this study was to analyze the productivity variation in smallholder rubber lands in the Kalutara district which is located in the Wet Zone of Sri Lanka and spatial relationship of key drivers to the productivity variation. Low productivity has been a major challenge in rubber plantations in the country in recent years. In this study we used spatial modelling tools available in geographic information science to explore the spatial variability of the rubber productivity and explore the key drivers of it in spatial context. Geostatistical kriging analysis is the simple type of prediction method which include the cross validation of prediction and error terms in the forecasting techniques. The productivity of smallholder rubber lands in Kalutara District varied from 777 to 1463 kg/ha/year, while the highest average productivity was recorded in the Divisional Secretariat (DS) divisions; Palindanuwara, Beruwala and Kalutara. Low productivity was recorded in the Matugama and in a few areas in Ingiriya and Bandaragama divisions. Local variation of driving forces behind the average productivity was explored using Geographically Weighted Regression (GWR) method. GWR explored the spatial variability of the relationship between the productivity and fertilizer usage, weeding, soil conservation, number of tappable trees and age of the trees under tapping. All the variables were found to present significant spatial variability. Apart from generating global significant value model resulted the local variation of each parameter estimates with respect to the projected coordinates of the area. Emerge of sign change of local parameters observed in some areas that cannot be observed in globally. It is necessary to understand the significance level of local coefficient subject to the multicolinearity and spatial auto correlation. 

Keywords: geographic information science, geographically weighted regression, kriging geostatistical analyst, multicolinearity, productivity, spatial auto correlation