UNCERTAINTY IN ESTIMATION OF BIOENERGY INDUCED LULC CHANGE: DEVELOPMENT OF A NEW CHANGE DETECTION TECHNIQUE.

Abstract:

Recent estimates of bioenergy induced land use land cover change (LULCC) have large uncertainty due to misclassification errors in the LULC datasets used for analysis. These uncertainties are further compounded when data is modified by merging classes, aggregating pixels and change in classification methods over time. Hence the LULCC computed using these derived datasets is more a reflection of change in classification methods, change in input data and data manipulation rather than reflecting actual changes ion ground. Furthermore results are constrained by geographic extent, update frequency and resolution of the dataset. To overcome this limitation we have developed a change detection system to identify yearly as well as seasonal changes in LULC patterns. Our method uses hierarchical clustering which works by grouping objects into a hierarchy based on phenological similarity of different vegetation types. The algorithm explicitly models vegetation phenology to reduce spurious changes. We apply our technique on globally available Moderate Resolution Imaging Spectroradiometer (MODIS) NDVI data at 250-meter resolution. We analyze 10 years of bi-weekly data to predict changes in the mid-western US as a case study. The results of our analysis are presented and its advantages over existing techniques are discussed.