ViitGEE: an open-source tool for visual image interpretation and reference data collection in google earth engine

Abstract

Cloud-based GIS platforms have altered how geospatial data is accessed, processed, and interpreted, yet efficient visual image interpretation and reference data collection tools remain limited. This study introduces ViitGEE, an open-source tool developed within the Google Earth Engine (GEE) environment, designed to facilitate the visual interpretation of very high-resolution satellite imagery and generate reference datasets for land cover classification studies. ViitGEE allows users to define custom spatial resolutions, interpret imagery using Google Maps base maps, and incorporate NDVI, NDWI, and NDBI time-series data to support decision-making in cases of visual ambiguity. A case study was conducted over a heterogeneous region in the central United States to evaluate the tool’s performance. Using 4,000 visually interpreted pixels via ViitGEE, a land cover map was produced with a Random Forest model, achieving an overall accuracy of 92.88% and a Kappa coefficient of 0.903. An independent validation with 1,000 stratified pixels produced with ViitGEE yielded similar accuracies. Comparative analysis with the NLCD 2023 map also demonstrated strong agreement. ViitGEE’s customizable classification scheme, open-access code, and export features enhance its adaptability across research needs. As a community-driven prototype, ViitGEE represents a scalable, flexible, and reproducible solution for land cover studies and is expected to evolve with ongoing user feedback and development.