Satellite-Based Vegetation Biomass Predictions for Colorado Rangelands, 2014-2023

Nearly 10,000 ground samples collected from 2014 to 2023 at the Central Plains Experimental Range in Colorado support research on machine learning model transferability. The dataset, aggregated to plot and transect levels, includes in-situ biomass estimates, geographic coordinates, and satellite-derived indices from Harmonized Landsat Sentinel (HLS) data. It was created by the USDA Agricultural Research Service to evaluate the performance of geospatial MLAs for predicting herbaceous biomass.

Use Cases

• Training and comparing machine learning algorithms for biomass prediction based on satellite-derived indices.
• Evaluating model transferability to unseen conditions based on geographic and temporal blocking covariates.
• Assessing model performance under extreme conditions using in-situ biomass and visual obstruction readings.
• Analyzing the relationship between satellite spectral bands and ground-measured herbaceous standing biomass.

Strengths

• Contains nearly 10,000 ground samples collected over a 9-year period (2014-2023).
• Data is aggregated at two spatial levels: 2,322 plots and 9,647 transects.
• Includes multiple data types: unique identifiers, geographic polygons, in-situ biomass, and satellite-derived indices.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown for the provided CSV/XML files, which may limit suitability assessment.

Provenance

Source

USDA Agricultural Research Service, SCINet project/AI Center of Excellence.

Collection Method

Ground samples collected using visual obstruction methods, combined with satellite data from the Harmonized Landsat Sentinel (HLS) product.

Time Range

2014 to 2023

Freshness

Last updated 2026-03 13 22:48:08.661013; freshness should be verified.

Geography

Central Plains Experimental Range (CPER), near Nunn, Colorado, USA.