Leaf Area Index Inversion In Kiwifruit Orchard Based On Multi-source Remote Sensing Data

The northern foot of the Qinling Mountains and the two sides of the Weihe River in Shaanxi Province is the advantage of kiwifruit production areas,kiwifruit has become a characteristic industry to drive local economic development and increase farmers’income.However,due to the deterioration of ecological conditions and the neglect of resource utilization,some varieties of kiwifruit have been endangered.Therefore,remote sensing monitoring of major kiwifruit growing areas to clarify the spatial distribution of kiwifruit orchard is of great significance for the protection of kiwifruit germplasm and the sustainable development of genetic resources.Based on the spectral reflection characteristic of ground feature information of remote sensing technology to make large and fast growing conditions to be real-time monitoring.LAI(leaf area index)as an ideal parameter estimation of vegetation growth,reflecting the crop physiological process and ecological system function has an irreplaceable role.At present,LAI estimation is mainly applied to winter wheat,corn,soybean and other food crops,but there are few studies on the application of kiwifruit and other orchards.In view of the large number of available remote sensing information sources and the small number of LAI inversion studies for cash crops in arid areas,it is particularly important to make comparative analysis of multi-source remote sensing inversion of LAI and select the optimal scale suitable for LAI inversion.LAI estimation based on accurate and detailed ground hyperspectral platform and fast and convenient high resolution UAV digital camera platform is of great significance for obtaining three-dimensional information of kiwifruit orchard,as well as its growth,yield estimation and orchard management.In this study,the growth parameters,canopy hyperspectral data and high-resolution RGB images of kiwifruit at four growth stages,including initial flowering stage(IF),young fruit stage(YF),fruit enlargement stage(FE),and mature stage(MS),were obtained from May to August 2021.Based on the measured data,the growth parameters of kiwifruit like LAI were discussed,besides,the original spectrum of kiwifruit canopy and its ten mathematical transformation spectra were analyzed.And a series of hyperspectral processing methods,such as spectral mathematical transformation,vegetation indices and spectral indices,were carried out on the spectrum of kiwifruit canopy.The correlation between LAI and measured values in corresponding periods was calculated,and the corresponding estimation model was built.Spectral and textural features of UAV RGB images at different kiwifruit growth stages were extracted in this study,and LAI estimation models of corresponding periods were constructed through a series of new variable sets,so as to estimate spatial distribution characteristics of LAI in the kiwifruit orchard in the study area.And The scale transformation of LAI in kiwifruit orchard was studied by combining sentinel-2 images.The study showed that:1)Canopy structure parameters of kiwifruit have been changing,with the advance of growth period,LAI showed a trend of gradual increase,and LAI reached the maximum at mature stage.The absorption valley of kiwifruit canopy spectral reflectance in the visible band of blue and red spectrum decreased with time,while the reflection peak in the green spectrum increased gradually.In the near infrared long wave region,the absorption intensity of water decreased with the decrease of wavelength.After different forms of spectral transformation,the spectral characteristics of kiwifruit canopy were obvious,and the characteristic bands were easier to highlight.Reciprocal transform reversed the"peak and valley"features of the original spectrum;The logarithmic reciprocal transform elongated the extreme change;After the transformation of the first derivative of the reciprocal,the curve fluctuated greatly,and the visible blue,green,yellow and red edges were obvious.The first derivative transformation of the reciprocal of logarithm enhanced the characteristic bands of visible light as well as the absorption bands of water and carbon dioxide in the short-wave and long-wave regions of near infrared,which could add more characteristic bands to the differential transformation.2)LAI measured values were significantly correlated with spectra processed by various methods,and had the highest correlation(-0.684)with the reflectance at 474 nm at maturity stage of original spectrum,and the highest correlation(0.760)with the reflectance at 2056nm at young fruit stage of square root transform spectrum.It had the highest correlation with MSR(0.686)of the vegetation index and DSI(1382,414)of the optimized spectral index(0.786).Among the four hyperspectral processing methods,mathematical transformation spectral model(modeling R~2=0.700,RMSE=0.542,n RMSE=17.25%)was superior to spectral index model(modeling R~2=0.635,RMSE=0.214,n RMSE=20.11%)and superior to original spectral model(modeling R~2=0.579,RMSE=0.230,n RMSE=30.17%),and the traditional vegetation index model(modeling R~2=0.564,RMSE=0.234,n RMSE=23.70%).3)The UAV univariate spectral index quadratic polynomial models could basically meet the prediction of LAI at YF and FE,but the accuracy of univariate model is very low for IF and MS.The new index sets combining spectral and textural features could estimate LAI with higher accuracy.The accuracy(validation R~2=0.789)of SWR model combined with texture feature was higher than that of pure spectral feature model(validation R~2=0.690).Compared with pure spectral feature model,the estimation ability of SWR model combined with texture feature was stronger.Therefore,the combination of spectral and texture features was more suitable for growth monitoring of kiwifruit orchard.In addition,the RFR model significantly improved the predictability and accuracy of the model.The results showed that the prediction accuracy of RFR model in different growth stages was higher,and the validation accuracy of the model at FE(R~2=0.829,RMSE=0.069,n RMSE=13.49%)was better in four stages.4)Through the study of"point to surface"and"surface to surface"up-scaling conversion on LAI inversion of kiwifruit orchard,it was found that the ordinary kriging interpolation method based on UAV high-resolution scale is suitable for the study on LAI inversion scale of kiwifruit orchard,which the validation R~2 reached 0.656.And the LAI spatial distribution characteristics showed a similar trend before and after the"point-plane"ascending scale transformation.LAI spatial distribution characteristics were basically similar before and after the"surface to surface"up-scaling transformation of UAV images,but different from the LAI distribution estimated by Sentinel-2.Besides,the predicted values of LAI were underestimated,and when the UAV scale increased to 10 m satellite scale,its fitting effect with Sentinel-2 satellite at 10m scale was general.