Spectral Features And Inversion Models Of Water And Nitrogen Status In An Apple Orchard With Dwarf Tocks And Dense Planting In Southern Xinjiang

Southern Xinjiang is an important production base of feature fruits in China.To improve the utilization efficiency of water and nitrogen,precise monitoring and scientific management of water and nitrogen should be carried out during the production process.However,traditional methods of water and nitrogen measurements are inefficient and difficult to implement for large-scale accurate detection of water and nitrogen conditions in orchards.The rapid development of spectral remote sensing technology provides a feasible approach for real-time and accurate water and nitrogen status monitoring in orchards.This study focused on an apple orchard with dwarf stocks and dense planting and used a handheld hyperspectral spectrometer and an unmanned aerial vehicle multispectral camera to obtain reflection spectra data of soil and apple tree canopies.Spectroscopic characteristics of soil and canopy were analyzed,and the partial least squares regression(PLSR),support vector machine regression(SVR),BP neural network(BPNN),and random forest regression(RF)models were adopted to develop appropriate inversion models for orchard water and nitrogen status.The main conclusions of this study were as follows:(1)The spectral characteristics of water,nitrogen,and salt in the surface soil were screened and determined,and the spectral inversion models were developed.The feature spectra of soil moisture were around 1900 nm,nitrogen was around 1 490-1 506 nm,1 540-2 006 nm,and 2 011-2 500 nm,and salt was around 1 880-1 883 nm and 1 890-1 942 nm.The Savitzky-Golay(SG)smoothing,principal component analysis(PCA),and partial least squares regression was the appropriate procedure for inverting water,nitrogen,and salt content of sandy soil,while the Savitzky-Golay,logarithmic,successive projections algorithm(SPA)and PLSR model was suitable for loam soil moisture inversion,and the Multiplicative Scatter Correction(MSC),logarithm,competitive adaptive reweighted sampling reduction,and SVR model was feasible for predicting loam soil salt content.(2)The spectral characteristics of water status in the apple tree canopy were analyzed,and suitable spectral inversion models were established for different growing stages of the apple tree.The feature vegetation indices of NDVI,RVI,OSAVI,GNDVI,GRVI,and GOSAVI were highly correlated with the leaf water content of the apple tree canopy and could well characterize the water status.For canopy water status inversion,the SVR model was the best one during the flowering and fruit set period,the BPNN model was the best one during the pre-expansion period,the SVR model was the best model during the mid-expansion period of fruit,and the PLSR model was the best one during the expansion-late period of fruit.The SVR model had the best inversion performance on the canopy water status of the apple tree during the whole growth period.(3)The spectral characteristics of nitrogen status in the apple tree canopy were determined,and suitable nitrogen inversion models were developed for different growing stages.The stable nitrogen feature vegetation indices in the apple canopy were NDVI,RVI,and OSAVI,which increased when the nitrogen content in the leaves was high.The BPNN model was best for inverting canopy nitrogen status during the flowering and fruit set period,the RF model was best for the early-expansion period of fruit,the PLSR model was feasible for the mid-expansion period of fruit,and the BPNN model was appropriate for the lateexpansion period of fruit.Overall,the BPNN model had the best inversion performance during the whole growth period of the apple tree.