Accuracy Improvement of Land Cover Type Classification Using Nighttime AVHRR Data.

Abstract

The new method developed in this study was the use of nighttime AVHRR thermal data as an additional channel in classifying land cover types with respect to improve classification accuracy. The NDVI, ordinary brightness temperatures and land surface temperature, were examined using two-classification algorithms namely Maximum Likelihood and Decision Tree. Various band inputs were comparatively assessed for the classification accuracy. The results of classification in both cool and hot seasons showed that overall accuracy obtained by using combination of day and nighttime data was better than that obtained by using only daytime data. Using a combination of three bands including NDVI, daytime LST, and nighttime LST gave the best accuracy. Using three bands and only daytime data yielded a lower accuracy compared to when using only two bands, but with day and nighttime data. Although an overall accuracy of using only daytime and using both day and nighttime data was not remarkably different, the accuracies when adding a nighttime data were notably increased such as in a case of forest and built up classes. The nighttime data can well classify (1) forest from densely grown crops, (2) deciduous forest in hot season from sparsely growing crops and harvested agricultures, and (3) built up from harvested land. The results also indicated that the proposed approach using nighttime data was effective as a new classification method in vegetation environment associated with landscape. In addition, this study suggested an integrated approach involving day and nighttime data to monitor urbanization and heat island.