Mapping of Oriental Migratory Locust Habitat Using Landsat OLI Images in Dongying City, China
Xiaomei Zheng,Jingfeng Huang,Hongmei Li,Lamin R. Mansaray,Peilin Song,Yujie Dou
DOI: https://doi.org/10.1109/agro-geoinformatics.2018.8476141
2018-01-01
Abstract:Oriental migratory locust seriously threatens agricultural production in croplands and can lead to substantial yield losses. Located in the northern part of Shandong Province and in proximity to the Yellow River Delta, Dongying City is one of areas most prone to the migratory locust plague in East of China. Traditional methods on monitoring and preventing outbreaks of locust are mainly based on field activities. It costs a lot of labors and time actually. Therefore, to generate data that would guide policies aimed at timely prevention and control of such a plague efficiently, we have monitored locust habitat distribution in this area using multi-temporal Landsat OLI images acquired from February to October in 2015. There are six Landsat OLI images were used all. A decision tree classification approach was employed to discriminate locust habitat from the surrounding land-cover based on the seasonal variation of different land-cover types. There are two commonly used spectral vegetation indices in decision tree classification algorithms for this study including the Normalized Difference Vegetation Index (NDVI) and the Land Surface Water Index (LSWI). Based on the production rules from the decision tree algorithms, seven land-cover categories were identified in this study as follows: locust habitat, dry land, water, paddy rice, urban, forest, and others. The classification results were evaluated by 327 random verification points from yield investigated (forest and locust habitat) and Google Earth images (dry land, water, paddy rice, urban, and others) respectively. According to the classification mapping result, approximately 8.5% of the total area was extracted as “oriental migratory locust habitat”. Also, locust habitat mainly distributed in Hekou district and Kenli district near the Yellow river especially. The accuracy assessment of the resultant classification map indicated an overall accuracy of 87.16% with a Kappa coefficient of 0.849 from the error matrix, which is typically higher than traditional supervised classification methods such as maximum likelihood. Also, water produced the highest accuracy as expected. For locust habitat land-cover types, the classification result produced the producer accuracy and the user accuracy with 81.255 and 92.86% respectively. These results have demonstrated that multi-temporal-data-based decision tree mapping methods can well exploit the tremendous potential of different vegetation phenological characteristics for extracting oriental migratory locust habitat and improving overall land-cover classification accuracy to some degree. In general, the classification result from decision tree method can be used for early locust damage prevention and produce a scientific guidance in Dongying city further.