Code and derived data forTraining Sample Location Matters: Accuracy Impacts in LULC Classification

This repository contains the analysis code and derived outputs for the study “Training Sample Location Matters: Accuracy Impacts in LULC Classification”. The workflow was implemented in Google Earth Engine (JavaScript API) and replicated in Python notebooks (Jupyter/Kaggle) for reproducibility.The code executes a reproducible Land Use and Land Cover (LULC) classification workflow over the Ferizaj region (Kosovo), using both PlanetScope (3 m) and SkySat (0.5–0.8 m) imagery (licensed data).ContentsGEE scripts (.js): segmentation, training/validation sample handling, Random Forest (RF) and CART classifiers, accuracy assessment, misclassification flows, and buffer-based validation.Python/Kaggle notebooks (.ipynb): reproducibility pipeline for accuracy metrics and statistical analysis.Derived data products (.csv, .tif): accuracy tables, confusion matrices, and misclassification flow datasets generated by the experiments.README file: instructions on how to replace placeholders with private Planet assets and run the code in Google Earth Engine.Data AccessRaw PlanetScope and SkySat imagery cannot be shared publicly due to Planet Labs licensing restrictions.Derived LULC maps, metrics tables, and flows are provided here to support reproducibility of the results.