Downscaling model of SMAP L4 soil moisture product in Inner Mongolia region based on machine learning
BIAN Chaoyang, HUANG Fang, HE Weibing, ZHANG Qiaofeng, LU Tongtong, GUAN Hao
2025, 0(9):
19-25.
doi:10.13474/j.cnki.11-2246.2025.0904
Asbtract
(
)
PDF (1429KB)
(
)
References |
Related Articles |
Metrics
With the intensification of global warming,drought disasters are increasingly frequent in Inner Mongolia region,posing a serious threat to agricultural and animal husbandry production,ecological environment and regional sustainable development.Soil moisture,as the most direct indicator reflecting drought disasters,has a significant impact on agricultural and animal husbandry production and ecological environment in Inner Mongolia region.However,there are difficulties in obtaining soil moisture data with high temporal and spatial resolutions at present,and traditional monitoring methods are difficult to meet the demand.This study utilizes the powerful cloud storage and computing capabilities of Google Earth Engine to obtain long-time series remote sensing data including Sentinel-1 SAR,SMAP L4,Landsat 8,MODIS LST,DEM and other products of Inner Mongolia region,and performs preprocessing,unifying time scales and spatial resolutions.Through correlation analysis,downscaling factors with the greatest correlation with SMAP L4 soil moisture are selected.Random forest,support vector machine and classification and regression tree algorithms are respectively used,combined with the selected downscaling factors,to carry out soil moisture downscaling experiments and obtain soil moisture data with 1 km high spatial resolution and high accuracy in the study area.Finally,the downscaling results are compared and verified with SMAP L4 resampling data and public soil moisture datasets.The results show that the downscaling model based on random forest achieves an average R value of up to 0.84,an average MAE of 0.049 m3/m3,and both RMSE and ubRMSD are significantly smaller than the other two models,demonstrating the best downscaling performance in the study area.Based on innovative data processing methods,fine downscaling factor screening mechanisms,and comparative application of multiple machine learning algorithms,this paper provides an effective solution for obtaining long-term series,high-resolution,and high-precision soil moisture data in Inner Mongolia region,which is of great significance for local agriculture and animal husbandry,drought monitoring,and sustainable development.