by Siti H. Latifah (Indonesia)
Slash and burn agriculture has been commonly practiced in the tropics for generations. Besides the fact that this system is important for local people, it has been changing from the traditional pattern to a more intensive practice that could degrade ecosystems. There is a need to develop an automated method able to efficiently quantify land-use intensity within slash and burn system, i.e., the number of agricultural cycles and the fallow period lengths in order to define a sustainable land-use management which accommodates both food supply and forest conservation. In this research, we have proposed and evaluated the methods using breakpoints with the Breaks for Additive Season and Trend (BFAST) framework on Landsat time-series from 1984 to 2014 to detect changes within slash and burn system. The method was implemented in the tropical evergreen forest in the municipality of Tefe and Alvares, Brazil where the slash and burn field patches were located along the Amazonian river. We tested eight different vegetation indices and found that NDMI was successfully discriminated forests (old-growth and secondary forests) from agricultural fields, where recognizing the change between forests and agricultural fields is important to characterize changes in the slash and burn system. Here, we explored the use of change detection algorithm at the object level to derive the spatial context of changes. Objects, which refer to land cover types, were defined using multi-temporal Object Based Analysis (OBIA). Validation was done by comparing the predicted land-use history (number of agricultural cycles and length of fallow period) derived from remote sensing results with observed data from farmer interviews. Our approach could predict the observed data with the RMSE of 1.24 (range 4) and 1.35 years (range 10 years) for agricultural cycles and fallow period durations, respectively. The sources of error were most likely induced by a lack of Landsat data and irregularity of observations within the Landsat time series, something that is inevitable in the tropics. Another source of error was the time differences between the monitoring periods and the farmers’ approximation time. Overall, our remote sensing based approach is relevant to quantify land-use intensity within slash and burn agriculture systems in the tropical forest. This approach offers an easy and cost-efficient way to map land-use intensity within slash and burn agriculture at any scale (local, regional, etc.).
Keywords: Slash and burn agriculture; Time-series; Remote Sensing; Tropics; Normalized Difference Moisture Index (NDMI); Multi-temporal Object Based Image Analysis (OBIA); Breakpoints; Break for Additive Season and Trend (BFAST)