venkateshkolluru95/cross-scale-grassland-cc-agb-data

Cross-Scale Grassland Canopy Cover and Aboveground Biomass Dataset

Overview

This dataset contains wall-to-wall canopy cover (CC, %) and aboveground biomass (AGB, g/m²) predictions at 10 m spatial resolution for grasslands across Mongolia and two provinces of Kazakhstan (Almaty and Aqmola). The maps were derived using a cross-scale mapping approach integrating in situ measurements, small unoccupied aerial systems (sUAS) imagery, and Sentinel-2 satellite data.

Repository Structure

Code

cross_scale_grassland_CC_AGB_code/
├── 01_sUAS_vegetation_indices_GLCM.js
├── 02_RF_model_quadrat_to_site.R
├── 03_raster_to_points_conversion.R
├── 04_spatial_thinning_moransI.R
├── 05_Sentinel2_predictor_extraction.js
├── 06_RF_model_site_to_regional.R
├── 07_uncertainty_mapping.R
└── README.md

Script	Description
01_sUAS_vegetation_indices_GLCM.js	GEE script for calculating vegetation indices and GLCM texture metrics from sUAS imagery
02_RF_model_quadrat_to_site.R	R script for Random Forest model to upscale quadrat-based estimates to site scale
03_raster_to_points_conversion.R	R script for converting predicted raster maps to point datasets
04_spatial_thinning_moransI.R	R script for k-means clustering-based spatial thinning using Moran's I optimization
05_Sentinel2_predictor_extraction.js	GEE script for extracting Sentinel-2 predictor variables
06_RF_model_site_to_regional.R	R script for Random Forest model to upscale site-scale estimates to regional scale
07_uncertainty_mapping.R	R script for generating per-pixel uncertainty estimates using 30-fold cross-validation

Data

Canopy Cover Maps (CC, %)

• CC_MN_10m.tif — Mongolia
• CC_Almaty_10m.tif — Kazakhstan, Almaty Province
• CC_Aqmola_10m.tif — Kazakhstan, Aqmola Province

Aboveground Biomass Maps (AGB, g/m²)

• AGB_MN_10m.tif — Mongolia
• AGB_Almaty_10m.tif — Kazakhstan, Almaty Province
• AGB_Aqmola_10m.tif — Kazakhstan, Aqmola Province

Data Specifications

Attribute	Value
Spatial resolution	10 m
Coordinate reference system	Albers Conical Equal Area (WGS84)
File format	GeoTIFF with LZW compression
Temporal coverage	Kazakhstan (2022), Mongolia (2023)

Methods

Random Forest regression models were developed using a two-step cross-scale approach:

1. Quadrat-to-site upscaling using sUAS-derived spectral, textural, and structural metrics
2. Site-to-regional upscaling using Sentinel-2 imagery and environmental covariates

Uncertainty was quantified using 30-fold cross-validation.

Requirements

• Google Earth Engine account (for .js scripts)
• R version 4.0+ with packages: caret, randomForest, raster, terra, dplyr, doParallel

Citation

If you use this dataset, please cite:

@article{Kolluru2026_sUAS_Sentinel2,
  author = {Kolluru, Venkatesh and John, Ranjeet and Chen, Jiquan and Henebry, Geoffrey M. and Xiao, Jingfeng and Shinde, Rajat and Kussainova, Maira and Ganzorig, Ulgiichimeg},
  title = {Leveraging sUAS-Sentinel-2 synergy for cross-scale mapping of canopy cover and aboveground biomass across Mongolia and Kazakhstan},
  journal = {Remote Sensing of Environment},
  volume = {336},
  pages = {115302},
  year = {2026},
  issn = {0034-4257},
  doi = {10.1016/j.rse.2026.115302},
  url = {https://doi.org/10.1016/j.rse.2026.115302}
}

Contact

For questions regarding this dataset, please contact the corresponding author:

Venkatesh Kolluru
📧 Venkatesh.Kolluru@coyotes.usd.edu
University of South Dakota