Derived data for Polat Lake (Eastern Anatolia): DSM, MLP land-cover classification, TWI, solar radiation

Description

This dataset contains the derived geospatial products, trained machine-learning model and source code accompanying the article "Geomorphological and Radiative Controls on a Hypersaline Karst Doline Lake: Insights from Polat Lake (Eastern Anatolia)" submitted to CATENA (Elsevier, 2026). The study characterises Polat Lake — a small (~1.1 ha) hypersaline doline lake in the Erzincan-Divrigi Basin (Eastern Anatolia, Turkiye) — through UAV-based photogrammetry, deep-learning point-cloud classification, geomorphometric analysis and solar-radiation modelling. All raster products are georeferenced to WGS 84 / UTM zone 37N (EPSG:32637); the original DSM was produced at 5.68 cm/pixel ground sampling distance. Folder structure: 01_DSM_and_Orthomosaic -- UAV DSM (5.68 cm + 1 m) and multidirectional hillshade 02_Classification_outputs -- MLP-classified 341-million-point cloud + figure source PNGs 03_MLP_model -- PyTorch weights, scaler params, feature names, architecture 04_TWI_and_SolarRadiation -- TWI, annual solar radiation, aspect classes 05_Vector_layers -- Faults, geology (MTA-derived), water, peaks, contours (shapefiles) 06_Scripts_and_Code -- Training, inference, plotting Python scripts 07_Training_data -- Per-class training-sample CSVs (Bedrock, Evaporite, Soil, Wetland) 08_Geomorphometric_outputs -- Cross-section profile CSV and 2.5D bathymetric raster 09_UAV_metadata -- Agisoft Metashape processing report (camera positions, residuals) 10_Field_photos -- 18 representative UAV / handheld field photographs See README.md and LICENSE.txt at the root of the dataset for full documentation and citation guidance.

Steps to reproduce

1) DSM-based analyses: use DSM_full_5.68cm.tif with ArcGIS Pro Spatial/3D Analyst (Fill -> FlowDir -> FlowAcc -> Slope -> TWI; Area Solar Radiation). 2) Train MLP: run train_MLP.py on training CSVs in 07_Training_data/ (weighted cross-entropy + Adam, lr=0.001, early stopping patience=15). 3) Inference: load MLP_land_classifier_weights.pth + scaler_mean.npy + scaler_scale.npy, run predict_MLP.py on feature-enriched point cloud. 4) Geomorphometric indices and bathymetric model derived in CloudCompare (Cross-Section + 2.5D Volume modules).

Institutions

• Erzincan Binali Yıldırım University

  Erzincan, Erzincan

Categories

Licence