Application Of UAV Remote Sensing In Monitoring Damage Caused By Rice Leaf Folde

Rice leaf folder,Cnaphalocrocis medinalis(C.medinalis),is one of the important pests in the major rice-growing areas in China.Its outbreak could seriously affect China’s food security.This experiment conducted two consecutive years of unmanned aerial vehicle remote sensing observation and field observation experiments on the natural harm of C.medinalis in the field environment from 2021 to 2022.The canopy spectrum,leaf rolled rate,relative chlorophyll content and leaf area index data of damaged and healthy rice were acquired.Combined with the unmanned aerial vehicle(UAV)platform to obtain the multi-spectral data,canopy spectral curve,vegetation indexes and growth characteristic parameters of rice with the different pest levels were analyzed.The estimation model of chlorophyll content and leaf rolled rate were constructed by means of ordinary linear regression,polynomial fitting,stepwise multiple linear regression and partial least squares regression.This study provided important technical support for precise monitoring and early warning of rice diseases and pests in China.The main research conclusions were as follows:(1)In 2021,the chlorophyll content and leaf area index of rice in non-control fields changed in the same growth cycle.However,in 2022,the chlorophyll content mainly increased first and then decreased,while the leaf area index continued to increase.When the rice was exposed to insect pests,the canopy spectra showed significant differences began from the green band.The multispectral reflectance values of rice in the three bands of green light,near infrared and red edge were significantly reduced in the larval gnawing rice.At jointing stage,the partial least square method and the stepwise multiple linear regression were used to construct multiple equations to estimate SPAD.The estimation effect of SMLR model at heading stage is not ideal.(2)There was no significant correlation between the multispectral bands and the hyperspectral bands at tillering stage in 2021.At the jointing stage and the booting stage,the correlation between the hyperspectral bands and the multispectral bands was good.There was a good correlation between the hyperspectral and the multispectral bands on the tillering stage,jointing stage and booting stage in 2022.In general,the correlation levels between vegetation indices of different data sources in 2021 were significant,even extremely significant,with a correlation coefficient as high as 0.7.The correlation coefficients between the vegetation indices at the tillering stage and the booting stage were the high in 2022.There were significant negative correlations between the same vegetation indices at the jointing stage and the heading stage.(3)The linear fitting results of hyperspectral NDVI and multispectral NDVI at the four rice growth stages(tillering,jointing,booting and heading)in 2021 reached at least P<0.05significance level,and the best fitting effect was at the booting stage,while the weak fitting effect was at the heading stage.It showed the best fitting effect with a determination coefficient of 0.6 at the tillering stage in 2022.At jointing stage,the fitting effect was not good,and it was consistent with that in 2021.The determination coefficient of the fitting curve at the booting stage decreased compared with that in 2021.There was a good conversion relationship between the ground hyperspectral and UAV multispectral vegetation indexes at tillering stage and booting stage respectively.(4)Most of the vegetation indices in the four main growth stages of rice were significantly correlated with leaf rolled rate.The correlation between vegetation indexes and leaf rolled rate fluctuated greatly in different growth stages.The fitting effect of the multi-factor estimation models was better than that of the single factor estimation models in each growth period.Among them,the leaf rolled rate inversion model at jointing stage was the best,and the R~2 of its verification set was 0.87,and the RPD was up to 2.6.Tillering stage was the next.The results of model validation at heading stage were slightly worse,with R~2 of 0.74 and RPD of1.9.The multi-factor models at booting and heading stages were closer to the 1:1 line at lower infestation levels(leaf rolled rate<10%),and underestimated infestation levels at more severe infestations.In addition,the optimal leaf rolled rate estimation model was suitable for tillering stage and booting stage.The coefficient of determination of the fitting curve between the estimated leaf rolled rate and the real value of the control field reached 0.75.According to the spatial distribution of leaf rolled rate,the damage degree of C.medinalis increased gradually with the progress of growth period.The results of model interpolation at tillering,jointing and booting stages of untreated fields were consistent with actual conditions.In conclusion,it is feasible to use UAV multi-spectral image to monitor the damage of C.medinalis in field.