|
As one of the important indexes of the climate characteristics, the daily mean temperature in climate change research, agricultural meteorological disaster monitoring and other fields plays an important role; compared with the traditional way of monitoring and estimating the average daily temperature, the remote sensing technology has comprehensive, macroscopic, dynamic and other incomparable absolute advantages, and can accurately describe the spatial heterogeneity of the daily mean temperature. In order to improve the quality of agricultural meteorological service and increase the monitoring accuracy on agricultural disasters, we obtain the optimal inversion model of the daily mean temperature in soybean growing area. In this paper, based on the FY-3D-MERSIⅡ remote sensing data, a random forest model and multiple regression model were constructed, respectively, to inversion of spatially continuous daily mean temperature in Liaoning Province. Results are as below: (1) On the whole, the random forest model has good applicability in the daily mean temperature retrieval, the root mean square error (RMSE) and mean absolute error (MAE) of the random forest method are 0.95 °C and 1.75 °C; but the multiple regression model inversion accuracy is relatively low, RMSE is 1.24, MAE is 1.15°C. (2) Combined with the soybean growing area, data found that although the inversion results of random forest model and multiple regression model in the eastern mountains of the study area have great deviation, the proportion of soybean planting in this area is relatively low; therefore, both models have good applicability in retrieving daily average temperature in soybean growing area, and the random forest model is relatively more stable. (3) Based on the spatial interpolation, results show that the random forest model and multiple regression model in describing the spatial distribution of the daily mean temperature is more exquisite and accurate, especially in the coastal areas, and the inversion results are more consistent with the reality, which proves the feasibility of daily mean temperature inversion based on remote sensing.
|
