Objective The value of solar-induced chlorophyll fluorescence (SIF) is of great significance for characterizing the growth status of crops and early diagnosis of vegetation stress. At present, a large range of SIF values are retrieved by remote sensing technology. However, due to the limitations of low spatial resolution and sparse sampling, it is difficult to meet the application of spatial scales below the provincial level. To solve the above problems, this paper proposes a SIF downscaling model construction method based on random forest (RF) algorithm.
Methods First, multi-source remote sensing data are fused with orbiting carbon observation-2 (OCO-2) SIF, and the nonlinear relationship with characteristic variables is established under the same scale of SIF. Based on the assumption that the spatial scale relationship is constant, 500 m×500 m spatial resolution SIF data are inversely performed by RF algorithm.
Results The coefficients of determination, mean absolute error and root mean squared error of the prediction model reached 0.72, 0.24 mW·m-2·nm-1·sr-1 and 0.33 mW·m-2·nm-1·sr-1, respectively, indicating the effectiveness of this method. Then, based on the SIF data after the model downscaling, the spatial correlation analysis is carried out with the enhanced vegetation index, normalized difference vegetation index, and the new OCO-2 SIF data, and verified with the SIF of the Chinese carbon dioxide observation satellite mission and the ground SIF station data. The results show good correlation. In addition, the time series analysis of the downscaled SIF in Henan Province, shows that the downscaled products can better capture vegetation change information.
Conclusions The RF algorithm can perform SIF products with 500 m×500 m spatial resolution well by integrating multi-source data.