Probability Evaluation Study of Soil Erosion Types in the Songhua River Basin Based on Deep Learning and Gaussian Kernel Density Estimation

• Published in: Shui tu bao chi xue bao Vol. 38; no. 5; pp. 116 - 128
• Main Authors: XING Zhenxiang, WANG Jiaqi, ZHANG Hongxue, SONG Jian, WANG Yinan, DUAN Weiyi, GONG Ming, HUANG Changli
• Format: Journal Article
• Language: Chinese
• Published: Editorial Department of Journal of Soil and Water Conservation 01.10.2024
• Subjects: deep learning; gaussian kernel density estimation method; probability evaluation; soil erosion; songhua river basin
• ISSN: 1009-2242
• DOI: 10.13870/j.cnki.stbcxb.2024.05.012

[Objective] To scientifically identify the types of soil erosion at the watershed scale and give the corresponding probability of occurrence. [Methods] This study constructs a deep learning (DL)-based model for calculating soil erosion modulus in the Songhua River Basin and calculates different types of soil erosion modulus. Using three erosion modulus influencing factors, namely rainfall, air temperature and wind speed, as random variables, numerical simulation and Gaussian Kernel Density Estimation (GKDE) were used to construct the EM probability evaluation method, which gives the probability of occurrence of different combinations of soil erosion intensities. [Results] The R2 of the validation period of the EM computational models were all >0.86; 74.47% of the average annual occurrence of slight water erosion and slight wind erosion in the watershed; 12.86% of the area of slight and above water erosion and slight wind erosion; 12.56% of the area of slight and above wind erosion and slight water erosion; 0.