The disclosure provides a method for predictive soil mapping based on solar radiation in large flat area, comprising: Step 1, capturing the response process of the earth's surface heat: after an observation day or an observation period is selected, a curve graph depicting a dynamic response of the surface heat to solar radiation captured by using a remote sensor of a moderate-resolution imaging spectrometer; Step 2, constructing environmental covariates: quantitative analysis is performed for the dynamic response curve graph obtained in Step 1 by a mathematical method, and characteristic parameters are extracted and taken as environmental covariates; and Step 3: a machine learning model is established for the predictive soil attribute mapping. This method solves the challenge of effective soil mapping in flat areas, significantly improves the accuracy and efficiency of predictive soil mapping in flat areas, reduces the cost in time and economy of soil survey mapping.