Evaluating the temporal accuracy of grassland to cropland change detection using multitemporal image analysis

Abstract

Grasslands are valuable carbon sinks in the effort to mitigate climate change. However, they are not well protected and are consequently being replaced by agricultural systems worldwide. Current monitoring efforts using remote sensing and ground-based methods are insufficient, and accordingly the mapping of grassland to cropland conversions must be improved to better document these changes in the Canadian Prairies. The purpose of this study is to evaluate different structural break methods and remote sensing datasets for their temporal accuracy in detecting grassland conversions in two Alberta study areas from 2010 to 2018. Breaks For Additive Seasonal and Trend (BFAST), BFAST Seasonal and Bayesian Estimator of Abrupt change, Seasonality and Trend (BEAST) methods were applied to evaluate their sensitivity to rangeland and pasture conversions. The best model was BFAST Seasonal, correctly predicting the year of change for 76% of rangelands and 66% of pastures. This demonstrates that seasonal models are effective in detecting interannual changes in vegetation composition amidst background noise from climate and management induced phenological changes. MODIS data outperformed Landsat, outlining the importance of high temporal resolution remote sensing data to successful change detection, even at the expense of higher spatial resolution. Overall, this study demonstrates that structural break methods are effective in identifying grassland to agriculture transitions and may be useful for the operational monitoring of grassland inventories in the future.