Mapping Rubber Plantations and Natural Forests in Xishuangbanna (Southwest China) Using Multi-Spectral Phenological Metrics from MODIS Time Series

Cornelius Senf, Dirk Pflugmacher, Sebastian van der Linden, Patrick Hostert
2013 Remote Sensing  
We developed and evaluated a new approach for mapping rubber plantations and natural forests in one of Southeast Asia's biodiversity hot spots, Xishuangbanna in China. We used a one-year annual time series of Moderate Resolution Imaging Spectroradiometer (MODIS), Enhanced Vegetation Index (EVI) and short-wave infrared (SWIR) reflectance data to develop phenological metrics. These phenological metrics were used to classify rubber plantations and forests with the Random Forest classification
more » ... classification algorithm. We evaluated which key phenological characteristics were important to discriminate rubber plantations and natural forests by estimating the influence of each metric on the classification accuracy. As a benchmark, we compared the best classification with a classification based on the full, fitted time series data. Overall classification accuracies derived from EVI and SWIR time series alone were 64.4% and 67.9%, respectively. Combining the phenological metrics from EVI and SWIR time series improved the accuracy to 73.5%. Using the full, smoothed time series data instead of metrics derived from the time series improved the overall accuracy only slightly (1.3%), indicating that the phenological metrics were sufficient to explain the seasonal changes captured by the MODIS time series. The results demonstrate a promising utility of phenological metrics for mapping and monitoring rubber expansion with MODIS. OPEN ACCESS Remote Sens. 2013, 5 2796
doi:10.3390/rs5062795 fatcat:fkbgdekcy5dxvpddpjddk73xoe