Beyond classifications: Combining continuous and discrete approaches to better understand land-cover change within the lower Mekong River region

Abstract

Human activities operating at multiple spatial and temporal scales induce changes in economic, social and demographic structures, land cover and land use, and natural resource extraction. Reciprocally, human activities are themselves influenced by the manner in which climate variation, variability in the quantity and quality of natural resources, and other processes of environmental change interact with socio-cultural systems. In order to better capture such landscape dynamics, we combine continuous satellite data with discrete land-cover classifications Specifically, we use tasselled cap analysis (TCA), Normalized Difference Vegetation Index (NDVI) and thermal analyses of surface temperatures (Ts) to study land-cover change. In so doing, we highlight the importance, within land change science research, of moving beyond a simple classification scheme approach, and demonstrate how the integration of continuous data augments classification methodologies. In our study of a south-east Asian landscape, we find that though the dominant land-cover class in Sisaket, Thailand is rice, and in Ordar Mean Chey, Cambodia, it is forest, nevertheless the current changes in both locations follow a similar pattern. Tasselled cap (TC), surface temperature and NDVI analyses reveal these same patterns of cover: lower NDVI and TCA-Moist, and warmer temperatures in Sisaket, and higher NDVI and TCA-Moist values with cooler surface temperatures in Ordar Mean Chey. The analyses also highlight areas of substantial change, especially within-cover changes, showing how the use of means and variance surfaces for continuous satellite data can add value to a traditional land-cover analysis. Much of the significant change in these landscapes, as well as potential precursors to future changes, occurs within classes in these landscapes, suggesting that the true dynamism and variability of the social-ecological landscape is best captured by combining continuous and discrete analyses, given the importance of within-class variation.