Evaluating the feasibility of GF-1 remote sensing comparison with hyperspectral data for soil organic carbon prediction and mapping

Abstract

High-resolution remote sensing data play a very important role in agriculture. However, the major sources of high-resolution images are not owned by China. The Chinese "High Resolution Earth Observation Systems" was deployed in 2010, and several major projects have been implemented. The present study focused on assessing the feasibility of Gaofen (GF) multi-spectral data for monitoring bare soil organic carbon (SOC) at field and regional scales. The data sources are hyperspectra measured under laboratory conditions and simulated multi-spectral data from GF-1 remote sensing images. Partial least squares regression (PLSR) was used to estimate SOC. At the field scale, the SOC hyperspectral prediction model produced better R2=0.9688, RMSE=0.3818, and RPD=5.6393 than the simulated multi-spectral SOC prediction model (R2=0.8179, RMSE=0.9913, RPD=2.3401). At a regional scale, the SOC hyperspectral prediction model also produced a better R2=0.9319, RMSE=1.097, and RPD=3.8758 than the simulated multi-spectral SOC predicted model (R2=0.8445, RMSE=1.6574, RPD=2.4228). For the simulated GF-1 multi-spectra model, regional scale predications had advantages over field scale predictions. The spatial distribution characteristics of SOC measurements and predictions from hyperspectral data and simulated GF-1 multi-spectral data were similar. Thus, satisfactory performance of the predictive and calibrated models validates the feasibility of these methods for rapid large-scale SOC monitoring.