Inversion Of Leaf Area Index Of Different Growth Stages Of Maize/Rice Based On Gaofen-1 Satellite

Leaf area index(LAI)is an important vegetation structure parameter in the biogeochemical cycle.As an important physiological and biochemical parameter of vegetation,LAI is the main input parameter describing the model of land surface related process.It can describe the geometric characteristics of vegetation canopy and provide relevant structural quantitative information for describing the initial energy exchange of vegetation canopy surface.This paper uses Luancheng County of Hebei Province and Zhaojue County of Sichuan Province as experimental bases to carry out LAI inversion of GF-1 image based on empirical model,PROSAIL model and BP neural network model,and explores a BP neural network based on improved genetic algorithm.These models are compared with the inversion results of the four models in corn,rice jointing and heading stage.The inversion accuracy is analyzed and the optimal model of corn and rice in different growth stages is determined.The main research results are as follow:(1)Empirical model,the correlation coefficient between the 6 planting index and LAI is above 0.67.The SAVI model has the highest accuracy in corn jointing stage,and the EVI model has the highest accuracy in rice heading stage.The NDVI model has the highest inversion accuracy in the corn heading stage and rice jointing stage.Indicating that the appropriate vegetation index should be used for inversion in different growth stages instead of using the NDVI model alone.(2)The PROSAIL model has better inversion results than the empirical model.The accuracy of corn jointing inversion is R~2=0.6630,RMSE=0.6108,and the accuracy of heading stage inversion is R~2=0.7730,RMSE=0.5037,inversion of rice jointing stage The accuracy is R~2=0.6982,RMSE=0.5027,and the accuracy of heading stage inversion is R~2=0.7035,RMSE=0.6523.Compared with the empirical model,the inversion accuracy of corn growth period is increased by 3.6%and 16.0%respectively.The inversion accuracy of the growth period increased by 3.8%and 3.2%respectively.(3)BP neural network model,the LAI estimation results are in good agreement with the measured values.The accuracy of corn jointing inversion is R~2=0.7016,RMSE=0.4051,and the accuracy of heading stage inversion is R~2=0.7025,RMSE=0.6308,The inversion accuracy of rice jointing stage was R~2=0.6699,RMSE=0.5254,and the inversion accuracy of heading stage was R~2=0.7290,RMSE=0.4264.(4)Based on the BP neural network model improved by genetic algorithm,the inversion results have higher prediction accuracy than the BP neural network model,and the inversion precision of corn growth period is increased by 4.3%and 2.9%respectively.The inversion precision of rice growth period was increased by 9.5%and 2.6%respectively.The inversion results of corn growth period were both high precision,and the inversion results of rice growth period were the highest precision.In summary,the GA-BP neural network model inversion is more in line with the crop growth status,and the applicability is good in both growth stages of corn and rice.When the parameter conditions are not met,the empirical model is a good method,and the different vegetation index inversion should be better according to different growth periods.The above conclusions can provide theoretical basis and method support for the application of GF-1 satellite in crop leaf area index inversion.