Integration of Cloud Score+ with Sentinel-2 Harmonized for land use and land cover classification using machine learning algorithms

Abstract West Kalimantan, Indonesia, has exotic nature reserves, with Danau Sentarum National Park (Sentarum) as its icon. Considering this importance, observing the use of Land Use Land Cover (LULC) around the lake is very important, which can be done using a remote sensing approach. Being in the equatorial region often results in continuous cloud cover throughout the year, which results in inaccurate observations and analysis results. Almost all cloud cover filters are used, but the results are often less than satisfactory. This study shows how the Cloud Score+ (CS+) product with Sentinel-2 Harmonized (S2H) can monitor image clarity in Sentarum and process it using machine learning to produce reliable (LULC) maps. Finding reasonably eliminating clouds, and clear pixels, and cloud shadows from L2A (surface reflection) or L1C (top-of-atmosphere) photos are both possible with the help of CS+ output. Using the RF algorithm, we conducted LULC research in Sentarum and produced precise results for five different LULC classes, namely water bodies, agriculture, forests, barren land, and urban. Sentinel-2 data for 2022-2023 is the basis for mapping because of anomalies that occurred in both years. In 2022, there will be extreme rainfall in Sentarum, while in 2023, the start of El Nino will begin. To identify the study area, three satellite indices were used: Normalized Difference Vegetation Index, Modified Normalized Difference Water Index, and Normalized Difference Building Index. Overall, the final results of the research in 2022 and 2023 produced an overall accuracy of 94.52% and 93.97% and a kappa index of 92.22% and 91.79%.