Rice Growth Monitoring Based On A Fixed-Wing Unmanned Aerial Vehicle Platform

Monitoring crop growth status timely and accurately is of great interest in precision agriculture that seeks to match agronomic inputs to crop demand,both spatially and temporally.Unmanned aerial vehicle(UAV)-based remote sensing has attracted extensive interest for high flexibility and easiness of operation.In this study,field experiments were conducted over two consecutive years(2017-2018),involving different N rates(0-405 kg N ha-1),planting patterns(pot-seeding mechanical transplanting,carpet-seeding mechanical transplanting,and direct-sowing),and rice cultivars(Nanjing-9108 and Yongyou-2640).Images were collected by a compact multispectral camera mounted on a fixed-wing UAV during key rice growth stages.The reflectance of rice canopy was obtained after image pre-processing.Simultaneously,leaf nitrogen accumulation(LNA),Leaf area index(LAI),and aboveground biomass(AGB)were obtained through destructive field samplings.The objectives of this study were:i)to establish rice LNA estimation model using vegetation index method;ii)to establish rice LAI and AGB estimation models using vegetation index method(simple regression),artificial neural networks(ANN)regression model,random forests(RF)regression model,and linear mixed-effect(LME)model.Our work might provide effective technical support for real-time monitoring of rice growth in large-scale farmland.Firstly,we explored the potential of building rice LNA estimation model based on multispectral images from the fixed-wing UAV.The selected retrieval method was vegetation indices(VIs)based method and the VIs were calculated from the fixed-wing UAV-born MS image.In order to find the best VI for model building,the correlation analysis was conducted between LNA and VIs within different growth stages.The results showed that compared with the subsequent growth stages of rice,the vegetation index of each growth period before heading had a good relationship with LNA,and the booting stage showed the best performance.The R2 of CIRE NDRE and LNA reached 0.87.During the whole growth period modeling process,several vegetation indices including CIRE NDRE,mSR,CCCI and LNA with red and near infrared bands were more strongly correlated,and R2 of CIRE and NDRE both reached 0.76.The best vegetation index in the model validation process was CIRE,which verified R2 was 0.57 and the RMSE was 3.69 g m’2.The results show that the fixed-wing UAV platform can better monitor the nitrogen nutrition status of rice leaves.Secondly,we explored the potential for a linear mixed-effect(LME)model and multispectral images from a fixed-wing UAV to estimate AGB and LAI for rice.Images were collected by a compact multispectral camera mounted on a fixed-wing UAV during key rice growth stages.LME,simple regression(SR),artificial neural networks(ANN),and random forests(RF)models were developed for growth parameters(AGB and LAI)and spectral information.Furthermore,all regression models were evaluated by cross-validation using pooled data,based on the coefficient of determination(R2)and root mean square error(RMSE).The results showed that several NIR-based and red edge-based vegetation indices(VIs)had optimum relationships with growth parameters for the whole season(AGB:R2>0.60,LAI:R2>0.70)and pre-heading stages(AGB:R2>0.70,LAI:R2>0.70),but all the selected VIs were weakly related to growth parameters for the post-heading stages(AGB:R2<0.30,LAI:R2<0.60).Compared with other models(SR,ANN,and RF),the LME model had the highest AGB estimation accuracy for all stage groups(whole season:R2=0.89,RMSE=1.75 t ha-1;pre-heading stages:R2=0.86,RMSE=1.49 t ha-1;post-heading stages:R2=0.75,RMSE=1.97 t ha-1).Satisfactory results were also obtained for LAI estimation while the superiority of LME model was not as significant as it was for AGB estimation.This study demonstrates that LME model could accurately estimate rice AGB and LAI,and fixed-wing UAVs are promising for monitoring crop growth status over large-scale farmland.