Abstract: The variations of leaf area index (LAI) can be used to monitor growth status and estimate yields of winter wheat. However, there are several limitations in available studies, including scale mismatch between ground observation data and satellite data leading to scale effect and the weak sensitivity of commonly used multispectral data relative to hyperspectral data. To effectively use the hyperspectral information and select the best bands, LAI estimation models were proposed with the aim of improving LAI estimation accuracy. In this study, we used unmanned aerial vehicle (UAV) hyperspectral data, serving as a bridge between ground observations and satellite data. We acquired ground-measured LAI data, UAV hyperspectral data, and GF-1 data during the greening period of winter wheat. Different forms of transformation and calculation of characteristic variables were conducted on the hyperspectral data, followed by establishing the area of interest. UAV image pixels with the same scale as ground observation were calculated. We further conducted correlation analyses between LAI and various spectral transformation forms as well as vegetation indices at the same scale. LAI sensitive bands or indices were filtered. LAI inversion at different scales based on UAVs and GF-1 satellites was developed. The results showed that the sensitive bands or indices of winter wheat LAI were 635, 655, 693, 704, 714, 721, 724, 763, 806, 813, 900, 936 nm first derivative, 714, 717, 763, 767, 784, 806, 813, 900, 903, 936 nm  second derivative, as well as the sensitive spectral indices SD_y, DVI, MSAVI2, NLI, and SAVI. An inversion model for estimating LAI from UAV hyperspectral images was developed utilizing various techniques, such as stepwise regression, partial least squares, and ridge regression. According to accuracy comparisons, ridge regression model was considered the optimal. Based on the upscaling method, a GF-1 winter wheat LAI estimation model was constructed. The simulation results were used as relative true values to validate the LAI remote sensing inversion products. The correlation coefficient between FY3_1KM_LAI products and GF_1KM_LAI products reached 0.787, indicating a strong correlation between FY3_LAI products and relative truth. This suggests that the results could be applied in daily operational services and research. In this study, we addressed the accuracy of inversion models under different data sources by upscaling, and discussed the ability of different remote sensing information sources in estimating LAI of winter wheat. Our results provided scientific guidance for crop management and theoretical basis for precision agriculture research.

Key words: leaf area index, GF-1, UAV hyperspectral data, validation, ridge regression