Participatory mapping of forest plantations in the Southern Highlands of Tanzania with open source data and tools

Growing amount of the global demand on forest related services such as timber, wood fiber and fruits are produced in planted forests, especially in tropical regions where forest plantations have expanded during the last 25 years. As the plantations may form a substantial proportion of regional and local landscapes, spatially explicit knowledge of the dynamics of forest plantation cover is important to understand the environmental and socioeconomic impacts and to support sustainable forest management regimes. In this study, we used open image catalogues and cloud computing capacity of Google Earth Engine (GEE) and participatory reference data collection designed with Collect Earth tool to map the extent and species composition of forest plantations in Southern Highlands of Tanzania, a region experiencing rapid growth of planted forest area. A large training sample of forest plantations was collected in a Mapathon event where 22 local experts from different backgrounds interpreted plantation coverage, species and age information from high-resolution satellite images of Google Earth and Bing maps in 2 weeks’ time. The collected points were used to classify a stack of Landsat-8 OLI best pixel mosaic (2013-2015), Sentinel 2 median mosaic (2015-2016) and Sentinel-1A mean mosaic (2015), processed in GEE. Based on tentative classification results the plantation distribution in the Southern Highlands was mapped with high overall accuracy with coverage ranging between 200,000 and 220,000 hectares. Majority of the plantations were small and fragmented smallholder woodlots located outside the largest government and company-owned plantations. The main classification errors were caused by young plantations, heterogeneous structure of smallholder woodlots and human interpretation errors in sample collection. Nevertheless, the results show that integrating local knowledge on reference data collection provides good accuracy even in challenging interpretation task. Moreover, automated classification methods based on freely available and constantly updating geospatial data sets and tools allow repetition of the mapping, important for monitoring land cover dynamics especially in data scarce regions.