06. Accuracy assessment of landcover classification based on various machine learning algorithms and remotely sensed data through Google Earth Engine: Application in Dak Lak province

Abstract

Machine learning algorithms allow for increasing the accuracy of land cover classification models from earth observation satellite images. Combined with cloud computing platforms, the procedure is performed automatically to ensure efficient processing of a big dataset containing a plethora of valuable spatial information extracted from remotely sensed images. This study evaluates the performance of various classification models from Landsat 8 (LS8) and Sentinel 2 (S2) optical satellite images through Google Earth Engine (GEE). Spectral surface reflectance values are used as input to the Classification And Regression Tree (CART) and Random Forest (RF) machine learning models to classify 7 land cover classes in Dak Lak province in 2021. Statistical error revealed that, with an area of about 13,000 km² on the scale of Dak Lak province, the LS8 image combined with the CART algorithm achieved the highest accuracy with Kappa of 0.85. The all-inclusive procedure provides an automatic solution of accurate and reliable land cover mapping, as well as supporting natural resources management and the environmental monitoring.