Background With destroying the ecological environment and affecting the regional environmental carrying capacity, soil erosion has become one of the important issues restricting the sustainable development of society. Pengyang county is located in the second sub-region of the Loess Hilly and Gully Region, and is a key control area for soil erosion in the country. After years of vigorous management, the ecological environment has been effectively improved. In order to consolidate the management results, accurate soil erosion prediction results are the scientific basis for reasonable prevention of soil erosion.

Methods Three phases of remote sensing images, DEM, and daily rainfall were used as basic datain 2000, 2008, and 2015, meanwhile soil erosion was calculated based on GIS and RUSLE models and the analysis area was quantified and characteristics of annual spatiotemporal distributionin 2000-2015. An ANN-CA model was constructed using an artificial neural network coupled with a cellular automaton and the six soil erosion factors R, K, C, P, L, and S were invoked as model input variables. The soil erosion intensity level was the initial state of the cell which made a prediction of soil erosion in Pengyang country in 2025.

Results 1) Overall situation of soil erosion in Pengyang country from 2000 to 2015 indicated a positive trend. The area of erosion above the grade strong decreased by 652.81 km² and high-grade erosion gradually shifted to low-grade. 2) The accuracy of soil erosion intensity simulation results in 2015 was ranked as follows: slight > light > medium > ultra strong > severe > strong, the overall accuracy was 7.9%, the Kappa coefficient was 0.82 and the prediction accuracy was high. 3) Soil erosion in Pengyang county in 2025 would be mainly slight and light with areas of 1 366.67 km² and 7 486.61 km², accounting for 84.69% of the total area. The area above the strong is only 2.69 km², accounting for 1.1% of the total area which has improved further with compared with 2015. Erosion above the strong mainly occurred in Mengyuan and Xinjitown so that the construction of soil erosion conservation measures could be enhanced in this area to prevent soil erosion.Conculsions The research results illuminate that the ANN-CA model has significantly self-learning ability and spatial dynamic simulation function which are universal in regional soil erosion prediction.