Hyperspectral Remote Sensing Inversion Models Of Crop Chlorophyll Content Based On Machine Learning And Radiative Transfer Models

An accurate quantitative estimation of crop chlorophyll content is of great importance for a wide range of monitoring crop grow health condition and estimating biomass,since radiative transfer model are complex caused by the nonlinear relationship between crop specral and chlorophyll content and the uncertainties in the land surface systems, traditional inversion techniques can not satisfied with the demand of accurate estimation of chlorophyll content.Alternatively, machine learning algorithms are able to cope with the strong nonlinearity of the functional dependence between the biophysical parameter and the observed reflected radiance, they may therefore be more suitable candidates for estimating crop biochemistry parameters from inversion of radiative transfer model. It is crucial to apply machine learning algorithms for inversion of radiative transfer model, so as to construct hyperspectral remote sensing estimation model for crop chlorophyll content.The thesis first linked crop leaf level optical properties and chlorophyll content throuth the inversion of radiative transfer model, PROSPECT. Next, crop chlorophyll content scaled-up to the crop canopy level was estimated using machine learning and PROSAIL. The main conclusions and creative points are as follows:(1) First derivative, continuum removal, wavelet transform denoising were applied to measured leaf reflectance spectral and crop canopy reflectance spectral from Hyperion images to enhance chlorophyll absorption features. Next, the responses of crop leaf reflectance spectral generated by PROSPECT and canopy reflectance spectral generated by PROSAIL were analyzed with different chlorophyll contents. The ill-posed inverse problem of remote sensing was solved using prior information. Then, leaf-leavel spectral datasets and canopy-level spectral datasets of crops were created.(2) Genetic algorithm (GA) and particle swarm optimization (PSO) based approaches for determination the penalty parameter C and the kernel function parameter y of support vector machines (SVM), term GA-SVM and PSO-SVM were proposed. GA-SVM and PSO-SVM were applied to invert PROSPECT for retrieval of crop leaf chlorophyll content. The results demonstrate that the estimation accuracy of PSO-SVM approach surpass GA-PSO.Therefore, the PSO-SVM approach is valuable for parameter determination of SVM.(3) This study is the first couple gradient boosting machines (GBM) with PROSPECT, a hyperspectral remote sensing model, term GBM-PROSPECT, was developed for estimating crop leaf chlorophyll content. The developed model was compared with SVM-PROSPECT and RF-PROSPECT. The results show that GBM-PROSPECT yield an R2 of 0.9714 and a mean square error (MSE) of 36.9652 using the demyCλspectral datasets. Compared with SVM-PROSEPCT and RF-PROSPECT, GBM-PROSPECT got the highest chlorophyll estimation accuracy, therefore, GBM-PROSPECT is more suitable for crop chlorophyll content estimation at leaf level.(4) This study demonstrated that couple random forests (RF) with PROSAIL, a hyperspectral remote sensing model, term RF-PROSAIL, was developed for estimating crop canopy chlorophyll content. The developed model was compared with GBM-PROSAIL. The results show that GBM-PROSPECT yielded an R2 of 0.9000 and a mean square error (MSE) of 1670.4 using the db9Cλspectral datasets. Compared with GBM-PROSAIL, RF-PROSAIL got the highest chlorophyll estimation accuracy, and the computation time of RF-PROSAIL less than that of GBM-PROSAIL, therefore, RF-PROSAIL is more suitable for crop chlorophyll content estimation at canopy level.(5) The limation of PROSPECT and SAIL were analyzed, the errors generated in developed hyperspectral remote sensing inversion model for estimating crop chlorophyll content at different scales were discussed. It is recommended that future research should explore a systematic upscaling framework which combines spatial pattern analysis, improved radiative transfer models, crop biophysical/biochemistry parameters database, RF to retrieve crop chlorophyll content from multi-source remote sensing data collaboratively at the canopy level.