干旱区生态环境知识资源中心

There is a need for the up-to-date assessment of desertification/land degradation maps that are dynamic in nature at different scales for comprehensive planning and preparation of action plans. This paper aims to develop the desertification vulnerability index (DVI) and predict the different desertification processes operating in Anantapur District, India, based on machine language techniques. Climate, land use, soil, and socioeconomic parameters were used to prepare DVI by a multivariate index model. The computed DVI along with climate, terrain, and soil properties was used as explanatory variable to predict the desertification processes by using a random forest model. About 14.2% of the area was created as a training dataset in 9 places for modeling and remaining area was tested for prediction of desertification processes. We used desertification status map (DSM) of Anantapur District prepared under Desertification status mapping of India-2nd cycle as a reference dataset for calculation of accuracy indices. Kappa and classification accuracy index were calculated for training and validation datasets. We recorded overall accuracy rate and kappa index of 85.5% and 75.8% for training datasets and 71.0% and 51.8% for testing datasets. The results of variable importance analysis of random forest model showed that DVI was the most important predictor followed by potential evapotranspiration and Normalized Difference Vegetation Index for prediction of desertification processes. The results from this work given new insight into using the existing knowledge on prediction of desertification in unvisited areas and also quick update of DSM maps.