Abstract: 【Objective】 Based on GF-5 hyperspectral images, a model for inverting soil organic carbon（SOC） content in the irrigation area of Central Yunnan Plateau was constructed, which could provide reference basis for subsequent research on SOC content inversion in the irrigation area of Central Yunnan Plateau. 【Method】 Yao’an County, Chuxiong Prefecture, Yunnan Province was selected as the research area, and GF-5 hyperspectral image was used as the basic data source to screen out preprocessing methods with high correlation with SOC content and spectral index. The feature band combination was screened based on continuous projection algorithm（SPA） and competitive adaptive reweighting algorithm（CARS）. The selected feature band, spectral index, topographic factor and Sentinel-1 backscattering coefficient were combined as auxiliary variables, combined with the SOC content data collected in the field, XGBoost model was used to invert SOC content. 【Result】 Among the 21 data preprocessing methods, AM-Normalize had the best preprocessing effect, with a correlation coefficient of 0.7544 with the measured SOC content; followed by SG-FD, SD and FD,with correlation coefficients with the measured SOC content of 0.6791, 0.6671 and 0.6202 respectively. The band inversion effect of SPA screening was the best, with coefficient of determination（R²） increasing by 0.0739 and 0.1524 compared to CARS and full-band data respectively, while root mean square error（RMSE） decreased by 0.9279 and 1.2793 respectively. The variable model G2, which introduced topographic factors, had an R² increase of 0.0398 compared to the variable model G1（characteristic bands + spectral indexes）, and RMSE decreased by 0.1685; further adding the Sentinel-1backscatter coefficient, the R² of the variable model G3 increased by 0.0255 compared to the variable model G2, and RMSE decreased by 0.1385. The SOC content inversion results based on GF-5 hyperspectral images showed that the SOC content range in the Yao’an irrigation district of the Central Yunnan Plateau was 9.8443-29.2514 g/kg, with an average of19.4447 g/kg, which was relatively close to the SOC content measured value range of soil samples（10.47-30.11 g/kg）and the average value（20.6307 g/kg）.【Conclusion】 The XGBoost model has been built on the basis of GF-5 hyperspectral images, after AM-Normalize preprocessing effectively reduces noise interference, SPA screens feature bands, and introduces spectral index, terrain factor and Sentinel-1 backscatter coefficient, the accuracy and applicability of SOC content inversion can be effectively improved, which can provide technical support for SOC content prediction in the Central Yunnan Plateau.