Monitoring Locusta migratoria manilensis damage using ground level hyperspectral data

Locusta migratoria manilensis is one of the major migratory locusts in China which prefers phragmites australis (Cav.) Trin.ex Steudel (here after called reed). Locust damage is one of the major agricultural pests in the world which has a serious impact on agricultural production. With the development of optical remote sensing techniques, detection of plant diseases and pests by measurements of canopy spectra has been implemented on wheat, barely leaves, cotton, etc. However, rare studies have been focused on reed, especially on estimation of loss component caused by locust until now. Therefore, the objective of this study was to investigate hyperspectral characteristics of reed from ground level canopy spectral data by ASD FieldSpec® 3 Spectroradiometer and to establish loss estimation models based on a field simulated L. m. manilensis damage experiment. Up to now, Kenli District of Dongying City is an important region of locust monitoring and prevention in China. Therefore, we carried out the simulated damage experiment during July 2017 in Kenli district, Dongying city, Shangdong province of China. The simulated locust damage experiment was based on six simulated locust density levels and three different damage durations. According to the experiment schedule, hyperspectral field data were obtained in four times and corresponding aboveground biomass (AGB) were cut immediately after each of the three damage durations. Loss estimation models were based on 40 sample points between loss component of selected vegetation indices (including RVI, NDVI, GNDVI SAVI) and dry weight loss of green leaf of reed. The results indicated that: 1) After L. m. manilensis damage, reed canopy reflectance decreased in near infrared region whereas the gap between visible light and near infrared region was narrowed. Also, the more serious the damage, the more serious the decline of near infrared region. The near infrared region was more sensitive to locust damage extent than visible light region. 2) Models based on four selected loss component of vegetation indices ($\Delta, \Delta, \Delta, \Delta$) all had good correlations with dry weight loss of reed green leaf with their R$^{2\,}$ ranging from 0.60 to 0.74. Among these models, the model based on $\Delta$ and $\Delta$ performed better with being 0.74 and 0.72 respectively. Assessment on the loss estimation models were conducted by additional 20 sample points. The assessment results also indicated that $\Delta$ and $\Delta$ produced a higher estimation accuracy with the RMSE being 14.3 g/m2 and 14.2 g/m2 respectively on dry weight loss of green leaf. Therefore, the result concluded that loss component of NDVI and GNDVI can further improve the results and be the optimal choice for loss estimation after locust damage.