Vegetation leaf chlorophyll is an important parameter for agricultural remote sensing inversion. The change of Chlorophyll content is intensely associated with the stress degree and physiological changes of vegetation growth environment. Therefore, it is vital for agricultural safety production to conduct real-time and dynamic detect in vegetation chlorophyll. Nevertheless, it is difficult to accurately measure space chlorophyll based on traditionally empirical models. In this study, using high-resolution Sentinel-2A data, three modeling patterns were driven by spectral information, optimal spectral index and biological covariates based on PROSAIL radiation transmission model within the framework of machine learning (random forest) (Scheme 1: the consolidation of spectral information and optimally spectral index combination, Scheme 2: the combination of spectral information and physical model biological covariate, Scheme 3: the merger among spectral information, optimally spectral index and physical model biological covariate). The chlorophyll content of cotton leaves was mapped based on the optimized modeling scheme. The results show that: (1) the correlation between the optimally spectral index Ratio Vegetation Index (RVI) with red edge band and the SPAD value of cotton leaves is the highest r=0.767, P^{* *}=0.195. (2) The importance analysis of the 17 constructed variables shows that the construction between the optimally spectral index Ratio Vegetation Index (RVI) and the physical model biological covariate LAI-Cab imposes a great contribution on the precise of the estimated model. (3) The red-edge band is determined as the optimal band when the vegetation index is constructed by the modeling scheme, which presented main role in constructing vegetation index. Analyzing the three schemes through model evaluation criteria, the order of prediction accuracy was model scheme 3 > model scheme 1 > model scheme 2. The decision coefficient R2 of scheme 3 was the highest at 0.826, which showed model scheme 3 has greater capability in predicting the SPAD value of the cotton leaves. It can provide advanced theory for the inversion of physiological parameters of crops in arid areas. Moreover, it also supplied scientific data support in detecting agricultural safety and allocating reasonable water and fertilizer.