Mapping the timing of cropland abandonment and recultivation in northern Kazakhstan using annual Landsat time series


Much of the world’s temperate grasslands have been converted to croplands, yet these trends can reverse in some regions. This is the case for the steppes of northern Kazakhstan, where the breakdown of the Soviet Union led to widespread cropland abandonment, creating restoration opportunities. Understanding when abandonment happened and whether it persists is important for making use of these opportunities. We developed a trajectory-based change detection approach to identify cropland abandonment between 1988 and 2013 and recultivation between 1991 and 2013. Our approach is based on annual time series of cropland probabilities derived from Landsat imagery and resulted in reliable maps (89% overall accuracy), with abandonment being detected more accurately (user’s accuracy of 93%) than recultivation (73%). Most of the remaining uncertainty in our maps was due to low image availability during the mid-1990s, leading to abandonment in the 1990s sometimes only being detected in the 2000s. Our results suggest that of the ~4.7 million ha of cropland in our study area in 1985, roughly 40% had been abandoned by 2013. Knowing the timing of abandonment allowed for deeper insights into what drives these dynamics: recultivation after 2007 happened preferentially on those lands that had been abandoned most recently, suggesting that the most productive croplands were abandoned last and recultivated first. Likewise, knowing the timing of abandonment allowed for more precise estimates of the environmental impacts of abandonment (e.g., soil organic carbon sequestration estimated at 16.3 Mt. C compared to 24.0 Mt. C when assuming all abandonment happened right after the breakdown of the Soviet Union, with the uncertainty around emission estimates decreasing by 63%). Overall, our study emphasizes the value of the Landsat archive for understanding agricultural land-use dynamics, and the opportunities of trajectory-based approaches for mapping these dynamics.

Remote Sensing of Environment, 213 49-60
Andrey Dara
PhD student
Tobias Kuemmerle
Tobias Kuemmerle
Professor & Head of the Conservation Biogeography Lab