Multi-faceted land cover and land use change analyses in the Yellow River Basin based on dense Landsat time series: Exemplary analysis in mining,agriculture, forest,and urban areas

The Yellow River Basin is one of China's most dynamic regions, where past and recent anthropogenic land use activities and polices have had a remarkably impact on the basin's surface. Over the past decades, the rapid socio-economic development has increased the pressure on the prevailing water and land resources with various repercussions on the environment and society. Counteracting ecological degradation in the basin, large-scale conservation and restoration plans have been initiated to expand vegetation coverage on deteriorated land, simultaneously fostering rural sustainable agriculture production. In this context, we derived precise spatial thematic products from long-term satellite time-series about high-frequency temporal dynamics. This information, available in a consistent and repeatable fashion is rare and relevant for many regional and local stakeholders and must be monitored annually to capture the rapid rate of change. Such information serves as a valuable base for decision-making processes. In this study, we used all the archived Landsat images between 2000 and 2015 (4520 scenes) to computed annually the spatially continuous spectral-temporal and textual metrics based on dense Landsat time-series to derive annual maps showing the most prominent land-cover change types related to mining, agriculture, forestry, and urbanization in four sub-regions spread over the Yellow River Basin. These novel land cover/use products provide new insights into recent regional and local dynamics. For final classification, we employed random forest classifiers for each thematic focus-region, trained and tested based on a stable-pixels data set. The resulting maps achieved high accuracies and show afforestation on the Loess Plateau and urbanization as the most prominent drivers of land use/cover dynamics. Agricultural land remained stable, showing local small-scale dynamics. Our study highlights the great potential of using consistent spectral-temporal metrics derived from dense Landsat time-series data together with a stable pixels reference set, allowing for local and regional land surface dynamics mapping at high spatial resolution and the prediction of implications of future change for effective and sustainable basin management.