Abstract
The purpose of the present study was to develop two modeling frameworks to predict suitable areas for sugarcane (Saccarum officinarum) cultivation in Sri Lanka using Geographical Information Systems (GIS). The first approach consisted of neural network-based GIS modeling and the second approach consisted of GIS-based cartographic modeling. Soil properties, meteorological data, current vegetation and slope accessibility were considered to be major factors to identify potential lands for sugarcane cultivation. The Levenberg-Marquardt (LM) algorithm was used to develop the Artificial Neural Network (ANN) model and the Normalized Weighting Method was used to obtain the weight values for the cartographic model. The results showed that the highly suitable lands obtained from the two models were differed by 7% in the current study area. According to the final suitability map obtained from the ANN model, 17.24%, 29.74% and 23.71% of the lands were classified into highly, moderately and marginally suitable categories, respectively. The results of the weighted overlay model showed that 10.34%, 32.33% and 28.02% of the lands corresponded to highly, moderately and marginally suitable categories, respectively. Neither model enabled to identify ‘unsuitable’ land parcels. Cartographic modeling did not enable to handle noisy and missing data. It was concluded that both approaches have their advantages and drawbacks for different purposes. However, these results revealed that neural network-based GIS modeling could become a powerful alternative approach towards automated spatial decision-making.
