Evaluating the efficiency of future crop pattern modelling using the CLUE-S approach in an agricultural plain

Abstract

Land Use Land Cover (LULC) change detection is an essential source of information for understanding the magnitude of environmental change to implement future development strategies. Sophisticated techniques (i.e. modelling) have been applied in the last decades worldwide for accurate LULC classification and future pro-jections. However, using these techniques in heterogeneous agricultural regions to extract crop-related infor-mation is still challenging. This study aimed to evaluate the efficiency and applicability of crop pattern prediction for the year 2050 with the CLUE-S model in an agricultural plain. The model was calibrated and validated based on the LULC changes to model future changes of the crop pattern by 2050. Twelve driving factors were utilised to quantify the relationship of LULC classes. The statistical relationship among the factors was examined with a Binomial Logistic Regression approach. Additionally, the magnitude of change in agricultural crop patterns between 2015 and 2050 was calculated according to local/regional policies and incorporated to the model as scenario layer. Future model results indicated that the cotton would increase by % 45 whereas maize would decrease by % 10 compared to 2015. The model performance was evaluated using the ground truth from the field observations considering the agricultural policies through the ROC (Receiver Operating Characteristic) indicators. The mean ROC value for the agricultural crop patterns was calculated as 0.71, while ROC values for other LULC classes were over 0.90. Overall a 0.79 ROC value was achieved as the model accuracy.