Polat Lake high-resolution environmental baseline: UAV-SfM, spatially validated classification, geomorphometry, wetness and solar-radiation data
Description
This dataset accompanies the article "High-resolution environmental baseline assessment of a hypersaline karst doline lake using uncrewed aerial vehicle photogrammetry and spatially validated deep learning", prepared for Environmental Monitoring and Assessment. It supersedes version 2 and provides the author-generated data, model artefacts, validation outputs and provenance code used for the audited single-campaign environmental baseline of Polat Lake, a hypersaline gypsum-karst doline lake in Eastern Anatolia, Türkiye. Five archives contain: (1) the 4.83 cm UAV-SfM DSM; (2) the 0.09996 m ground-filtered DTM, georeferenced RGB orthomosaic, October 2025 visible-water footprint and author-derived doline boundary; (3) the final 341,462,425-point classified LAZ and 1 m dominant-class raster; (4) training tables, trained model/scalers, nested five-fold blocked spatial cross-validation and RGB/geometry ablation results; and (5) TWI, whole-year potential solar-radiation layers, cross-section and sill-relative geomorphometry, reproducibility scripts, processing report and 18 field photographs. The mapped doline boundary is 25.987109 ha and the selected visible-water footprint is 0.892998 ha. Spatial data use WGS 84 / UTM zone 37N (EPSG:32637). This is a single-site, single-campaign baseline, not a time series. Blocked cross-validation estimates internal spatial interpolation and is not external validation at another lake. The sill-relative raster is not bathymetry, water depth or measured storage. No water-temperature time series, surveyed ground-control/checkpoint network, hydrological budget or causal environmental inference is included. Raw UAV frames, very large intermediate point clouds, the Metashape project and third-party MTA/Esri/OpenStreetMap content are excluded as documented in README.md.
Files
Steps to reproduce
1. Download the five Polat_Lake_V3_*.zip archives together with README.md, CITATION.cff, LICENSE.txt, MANIFEST_SHA256.csv, UPLOAD_SHA256.csv and dataset_metadata.json. 2. Verify the downloaded archive hashes against UPLOAD_SHA256.csv. 3. Extract all five archives into the same directory. Each preserves the common top-level folder Polat_Lake_Data_V3, so the contents merge without overwriting unrelated files. 4. Run the integrity check: python Polat_Lake_Data_V3/08_Reproducibility_code/validate_dataset.py Polat_Lake_Data_V3 5. Read 08_Reproducibility_code/README_reproduction.md and requirements.txt before rerunning analyses. Update the preserved workspace paths to the local extraction directory. 6. Classification reproduction uses the deposited finite-filtered training tables, model/scalers and spatial block/fold assignments. The deployment and validation scripts document the audited 13-predictor MLP workflow. 7. D8 TWI and whole-year potential solar-radiation provenance is supplied; ArcGIS Pro with Spatial Analyst is required to repeat the original GIS processing. Descriptive coupling statistics use the deposited aligned 1 m environmental grids. 8. Treat DSM/DTM elevation as within-scene relative geometry and the sill-relative raster as geometric screening only, not bathymetry or measured lake storage.
Institutions
- Erzincan Binali Yıldırım UniversityErzincan, Erzincan