Mapping cropland abandonment and distinguishing from intentional afforestation with Landsat time series

Abstract

Detecting cropland abandonment in a timely manner is essential to unlock the potential of such abandoned lands, for instance, to alleviate world hunger and steer environmental restoration. However, the challenge remains how to separate cropland abandonment from spectrally similar land-cover change trajectories, such as intentional afforestation (i.e., tree plantations) on former agricultural lands. Taking the South Sichuan province of China as a study area, this study developed a new approach by integrating land-cover change trajectories mapped using the random forest classifier and LandTrendr, as well as NDVI change based on Landsat time series to reveal abandoned cropland from 2003 to 2018. Results showed that the developed methodology could help to distinguish cropland abandonment with 76% producer’s and 80% user’s accuracies. The study showed that, by 2018, 0.37 million ha (approximately 15.54%) of previously cultivated land became truly abandoned, and 0.53 million ha (approximately 22.27%) of previously cultivated land became intentionally afforested. Annual abandonment rates were high at the beginning of the study period and low by 2018. Overall, our study highlights how the magnitude and pace of NDVI change helped to distinguish abandoned cropland from other land uses, such as intentional afforestation. The method can be adapted to map cropland abandonment accurately elsewhere; thus, our results can assist in evaluating land-use policies which aimed at guiding the cropland abandonment process.

Keywords

• Cropland abandonment
• Remote sensing
• Random forest
• Land-cover probability
• Biomass change