Monitoring Oriental Migratory Locust Damage Based On Multi-platform Remote Sensing Techniques

Locust damage is among the major worldwide agricultural pests,which has caused severe loss through around 100 countries and regions.The outbreak of Oriental Migratory Locust is the most frequent and the most serious among all kinds of locusts in China.Traditional methods on locust disaster monitoring and controlling mainly depend on field investigations.But such methods usually consume too much time and labor.In contrast,with the rather rich information that can be provided from the remotely sensed datasets,remote sensing technology makes it possible to implement large-scale monitoring of pests and diseases in a large-scale area effectively and rapidly,and has been an important research sub-branch in the field of agricultural informatization technology.Therefore,this paper conducted researches on monitoring and loss estimation of oriental migratory based on remote sensing platforms of different spatial scales.The primary objectives and conclusions are as follows:(1)Extraction of locust preferred vegetation areas based on satellite datasetsIn this study,we employed multi-temporary Landsat8 imagers to extract locust preferred vegetation areas using a decision tree classification method in Dongying city in 2015.The classification results manifested that around 12%of the entire study area is identified as oriental migratory locust preferred vegetation areas.Such habitats were mainly distributed in Hekou district and Kenli district of the city,and were especially concentrated near the estuary of Yellow River.The overall classification accuracy was 87.16%,with a Kappa coefficient of 0.849.The producer accuracy of oriental migratory locust preferred vegetation areas was 81.25%and the user accuracy was 92.86%.The results revealed that this method successfully extracted the potential occurrence areas of oriental migratory locust in Dongying.(2)Hyperspectral characteristic analysis of reed damaged by oriental migratory locustBased on the reed canopy spectral data collected during our field experiment of locust damage simulation,we conducted a series of analyses which focused on both the raw and the first-order derivative spectra differences between healthy and damaged reed caused by oriental migratory locust.We also compared the spectral characteristic variations of the 'red edge' parameters on different locust damage extents,including non-damage level,moderate-damage level and severe-damage level.The results showed after the occurrence of a locust damage,the characteristics of "green peak" and"red valley" faded and the overall spectral curve became much more flat.In the spectral range of 750nm and 900nm,the canopy spectra of damaged reed was featured with lower reflectance as opposed to those of healthy reed.And the worse the damage,the more significant the decrease went.The "red edge position" for the damaged reed moved to shorter wavelengths whereas this trend was not obvious.On the other hand,the locust damage did make the amplitude of the red edge peak decrease,mainly with the increase of the damage duration.(3)Modeling of reed loss estimation based on both ground level sensors and low altitude UAV platformsBased on the locust damage simulation field experiment,we employed canopy spectra derived on both ground level and low altitude UAV to respectively calculate the loss of vegetation indices and the loss of red edge parameters.Such loss indicators were then respectively used as model inputs to estimate green leaf loss of reed.The results showed that:1)The ASD ground spectral instrument and the low attitude UAV-hyperspectral platform were both reliable in obtaining high quality spectra data.The loss of NDVI derived from UAV performed best in estimating dry weight loss of reed green leaf,with a determination coefficient(R2)of 0.93 during the construction process of the estimation model as well as an RMSE of 8.8 g/m2(a relative RMSE of 7.6%)during the validation process.2)No matter for the ASD ground spectral instrument or for the low attitude UAV,loss estimation models based on vegetation indices performed better than those based on red edge parameters in estimating dry weight loss of reed green leaf.3)For both sensors,the parameter of "red edge position" was found impractical for estimating the reed green leaf loss since the R2 of the estimation model constructed using this parameter was only 0.02.In conclusion,this study employed satellite images to extract locust preferred vegetation areas and constructed reed loss estimation models based on the spectral characteristic variations in response to different locust damage levels.The locust damages were simulated in a field experiment and the spectra information were collected on two different platforms,i.e.the ground-level ASD instrument and the hyperspectral instrument onboard the low attitude U AV.The method for estimating reed loss has significant implications on further development of remote sensing techniques so as to realize the monitoring of Oriental Migratory Locust infestation and the post-disaster loss assessment in a more rapid and more effective manner.