The growth forecasting model for apple tree based on ground-based remote sensing

In order to monitor the growth statues of apple tree non-destructively and effectively, the field experiments were conducted at five different stages of apple tree annual growth season. The spectral reflectance of apple leaves was collected and the nutrient parameters of leaf (chlorophyll content (LCC) and moisture content (LMC)) were measured in the lab. The relationship between the apple tree leaf spectral reflectance and the apple growth parameters was analyzed. In order to select optimal spectral bands, the transformation forms of spectra were calculated including first derivative, second derivative, reciprocal, logarithm, the logarithm of reciprocal and the first derivative of logarithm. The sensitive detecting wavelengths were selected based on the correlation between the apple tree leaf spectra (original spectra and its transformation forms) and the apple tree growing parameters (LCC and LMC). The result showed that the original spectrum was most correlated with LCC from 511nm to 590nm and 688nm to 718nm; the correlation coefficients of September were the highest and the maximum value was 0.6. Three apple tree growth models were built using Multiple Linear Regression Analysis (MLRA), Principal Component Analysis (PCA) and Artificial Neural Network (ANN) respectively. The result showed that the forecasting model based on PCA was the optimal model to predict the apple leaves chlorophyll, and its calibration R2 was 0.851 and validation R2 was 0.8289. The apple leaves moisture content forecasting model based on ANN was optimal, and its calibration R2 was 0.8561 and validation R2 was 0.8375.