Inversion of photosynthetically active radiation based on GF-1 image by dark object method

Abstract: Photosynthetically active radiation (PAR) is an important influential factor of vegetation photosynthesis, and the study on its size and spatial distribution has important significance in many fields such as ecosystem study, agriculture monitoring, and ground-air energy exchange. To meet the demand of GF-1 satellite PAR remote sensing inversion, the paper establishes an atmospheric parameter look-up table (LUT), utilizes major atmospheric parameters, such as linear relation of reflectivity between blue band and red band of dark dense vegetation as well as inversed aerosol optical depth (AOD), and finally successfully inverses the ground PAR based on atmospheric radiation transfer model. The paper also conducts accuracy verification on the inversion result by using actually measured ground PAR. Dark dense vegetation algorithm uses the fixed linear relation of reflectivity between blue band and red band of dark dense vegetation in the GF-1 images to inverse the AOD. Identification of dark dense vegetation is mainly based on NDVI (normal difference vegetation index) values. Using typical vegetation spectral library provided by the USGS (United States Geological Survey), and based on GF-1 spectral response function convolution, the surface reflectivity of the vegetation in various wavebands were obtained, the proportionality coefficient of ground surface reflectivity between red band and blue band was analyzed, and finally obtains the proportionality coefficient of 1.7977 with the intercept of 0.0034, and the correlation coefficient of 0.9826. By taking linear relation of vegetation reflectivity between red band and blue band as constraint condition, and combined with radiation transfer equation, the aerosol and atmospheric parameter LUT is established and the AOD is inversed based on the LUT. Meanwhile, the spatial interpolation is made by using the property of continuity of AOD to acquire the AOD of the overall study area. After the inversion of AOD, atmospheric parameters such as atmospheric transmittance and hemisphere albedo are calculated. Finally ground solar radiation intensities in red, green and blue wave band of GF-1 satellite image are calculated. By studying the relation between ground surface solar radiation intensities of blue, green and red waveband in GF-1 satellite image and overall 400-700 nm PAR value, the paper has worked out the 3 conversion coefficients of 0.09156, 0.09951, and 0.1007 respectively, and thus realized the inversion from the ground solar radiation intensity of 3 discrete wavebands to PAR. The study selects 12 pieces of GF-1 effective data from April 2014 to December 2014 in the study area in Yucheng City, Shandong Province to inverse AOD and PAR, and verifies the results by comparing them with actually measured data in Yucheng experimental station of Chinese Ecosystem Research Network (CERN). The result shows that the overall accuracy of PAR has reached 95.77% with the average absolute error of 11.36 W/m2 and the average error less than 5%, indicating the correctness and precision of the method proposed by this paper, and also indicating the feasibility of PAR inversion by using GF-1 satellite images. The study also shows that the spatial distribution of PAR has significant correlation with AOD, and the higher the AOD is, the lower the PAR will be. By changing proportionality coefficient of vegetation reflectivity between red band and blue band, the study has found that when the coefficient value is approximately 1.5-2.1, its impact on PAR inversion accuracy is small. It is suggested to set the value to approximately 1.8 in order to achieve high accuracy. The method proposed by this paper can accurately inverse clear sky PAR with only original GF-1 satellite images without additional support data. This method is featured with simple production process and easy to be applied in the PAR operation inversion. It can provide important early-stage product data support for crop production estimation based on PAR in agricultural remote sensing monitoring practice.