Mosaic landscapes under shifting cultivation, with their dynamic mix of managed and natural land covers, often fall through the cracks in remote sensing–based land cover and land use classifications, as these are unable to adequately capture such landscapes’ dynamic nature and complex spectral and spatial signatures. But information about such landscapes is urgently needed to improve the outcomes of global earth system modelling and large-scale carbon and greenhouse gas accounting. This study combines existing global Landsat-based deforestation data covering the years 2000 to 2014 with very high-resolution satellite imagery to visually detect the specific spatio-temporal pattern of shifting cultivation at a one-degree cell resolution worldwide. The accuracy levels of our classification were high with an overall accuracy above 87%. We estimate the current global extent of shifting cultivation and compare it to other current global mapping endeavors as well as results of literature searches. Based on an expert survey, we make a first attempt at estimating past trends as well as possible future trends in the global distribution of shifting cultivation until the end of the 21st century. With 62% of the investigated one-degree cells in the humid and sub-humid tropics currently showing signs of shifting cultivation—the majority in the Americas (41%) and Africa (37%)—this form of cultivation remains widespread, and it would be wrong to speak of its general global demise in the last decades. We estimate that shifting cultivation landscapes currently cover roughly 280 million hectares worldwide, including both cultivated fields and fallows. While only an approximation, this estimate is clearly smaller than the areas mentioned in the literature which range up to 1,000 million hectares. Based on our expert survey and historical trends we estimate a possible strong decrease in shifting cultivation over the next decades, raising issues of livelihood security and resilience among people currently depending on shifting cultivation.

Over the past half century, countries of Mainland Southeast Asia (MSEA) -Cambodia, Laos, Myanmar, Thailand, and Vietnamhave witnessed increases in commercialized agriculture with rapid expansions of boom-crop plantations. We used MODIS EVI and SWIR time-series from 2001-2014 to classify tree-cover changes across MSEA and performed a supervised change detection using an upscaling approach by deriving samples from existing Landsat classifications. We used the random forest classifier and distinguished 24 classes (16 representing boom-crops) with an accuracy of 82.2%. Boom-crops occupy about 18% of the landscape (8% of which is rubber). Since 2003 74,960 km 2 of rubber have been planted; 70% of rubber is planted on former forest land, and 30% on low vegetation area (mainly former croplands). Timing, patterns of change, and deforestation rates, however, differ among the MSEA countries and the high spatial and temporal detail of our classification allowed us to quantify dynamics and discuss political and socio-economic drivers of change.

No abstract

No abstract

Abstract:We performed a multi-date composite change detection technique using a dense-time stack of Landsat data to map land-use and land-cover change (LCLUC) in Mainland Southeast Asia (MSEA) with a focus on the expansion of boom crops, primarily tree crops. The supervised classification was performed using Support Vector Machines (SVM), which are supervised non-parametric statistical learning techniques. To select the most suitable SMV classifier and the related parameter settings, we used the training data and performed a two-dimensional grid search with a three-fold internal cross-validation. We worked in seven Landsat footprints and found the linear kernel to be the most suitable for all footprints, but the most suitable regularization parameter C varied across the footprints. We distinguished a total of 41 LCLUCs (13 to 31 classes per footprint) in very dynamic and heterogeneous landscapes. The approach proved useful for distinguishing subtle changes over time and to map a variety of land covers, tree crops, and transformations as long as sufficient training points could be collected for each class. While to date, this approach has only been applied to mapping urban extent and expansion, this study shows that it is also useful for mapping change in rural settings, especially when images from phenologically relevant acquisition dates are included.