Land use and land cover dynamics in dryland ecosystem of Northwestern Ethiopia: taking into account uncertainties and correcting bias in satellite-based maps.

Information on Land use and land cover (LULC) dynamics is crucial for environmental change studies. Remote sensing offers option for monitoring multitemporal LULC dynamics. However, remote sensing-based LULC maps include classification errors that may have effect on area estimates. A variety of area estimators have been developed to correct the errors despite they have not been fully utilized. The study investigated spatiotemporal LULC dynamics that account bias and uncertainty analysis in dry woodlands of Northwestern Ethiopia. Reference data were collected from Google Earth for classification and accuracy assessment. Then, a random forest model was used to produce LULC maps of 1986‒2019 with the overall accuracies ranged from 92.0 to 97.8%. A stratified estimator was used to estimate a bias-adjusted area with its uncertainty using reference data. The result indicated the bias-adjusted area estimates of LULC class had minor uncertainty ranged 0‒13.0%. Cropland substantially increased by 116.8%, while grassland and woodland decreased by 59.1 and 24.7%, respectively between 1986 and 2019. However, woodland and forest gained 21.4 and 24.0% between 2000 and 2010 and 2010 and 2019, respectively. There was insignificant difference between the mapped area and the corresponding bias-adjusted area estimates, with a relative error ranged from ‒6.7 to 4.3%. Despite these, errors in maps were somehow critical, accounting for significant variation in the estimations of some classes. The study highlighted the importance of reference data to produce robust estimates along with spatially explicit maps. Landscapes of the study area have experienced widespread LULC changes requiring strategy that properly manage woodland ecosystems.