Prediction of landscape pattern changes in a coastal river basin in south-eastern China

Abstract

Different landscapes in a river basin make different contributions to non-point source pollution according to the source–sink theory. Changes in landscape pattern have a considerable impact on river water quality. Precise predictions of the future landscape pattern provide strong support for manipulating land use patterns, which is beneficial for sustainable development in the watershed. The Jiulong River Basin in south-eastern China was selected as the study area. The Random Forest classifier, which combined textural characteristics with spectral information, was applied to interpret the landscapes from Landsat images acquired in 1995, 2005 and 2015 in this paper. The overall classification accuracy reached 86%, and the Kappa index was higher than 0.83, which was better than the accuracy of classifiers based on the exclusive use of spectral information. We used a hybrid cellular automaton–Markov (CA–Markov) model to simulate the landscape pattern of the Jiulong River Basin in 2015 and verified the accuracy according to the interpreted landscape classification map in 2015. The pixels that were predicted to the correct landscape account for 88.35% of total pixels, and the Kappa coefficient was 0.88, indicating that the CA–Markov model was credible for predicting the future landscape pattern in the Jiulong River Basin. The CA–Markov model was applied to forecast the landscape pattern of this watershed in 2025, and the transition area matrix from 2015 to 2025 was obtained. The predicted results demonstrated the imbalanced migration of source–sink landscapes in the future, which indicated the deterioration of water quality in the study area. Hence, government regulators need rational manipulation of the landscape pattern for sustainable development in the Jiulong River Basin.