Global Land Surface Albedo Parameterization with 19 Years of MODIS Data

A physics-informed machine learning parameterization for predicting direct and diffuse albedo across visible and near-infrared bands. The framework uses 19 years (2003–2021) of MODIS BRDF data and ten biogeophysical predictors, developed by Akarsh Ralhan. It includes models, Python code, and static input files for integration into Earth system models.

Use Cases

• Improving surface energy balance simulations in Earth system models based on the physics-informed ML parameterization.
• Analyzing spatiotemporal albedo variability across land-cover categories like grasslands and shrublands based on the described predictors.
• Investigating the role of subgrid heterogeneity using the included Surface Albedo Localization Factor (SALF) correction.
• Conducting feature-importance analysis for albedo drivers such as solar zenith angle and leaf area index.

Strengths

• High predictive accuracy reported with an overall R² of 0.88 and MAPE of 7.5% against MODIS albedo.
• Includes 19 years of MODIS BRDF data (2003–2021) for training and validation.
• Framework incorporates a static correction (SALF) that reportedly increased average R² by 0.16 and reduced MAPE by 5.1%.
• Performance is validated across diverse land-cover categories, including challenging barren regions.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The dataset size is listed as 0.0 B, indicating a very limited scope or potential metadata issue.

Provenance

Source

MODIS BRDF data, author Akarsh Ralhan.

Collection Method

Physics-informed machine learning using 19 years of satellite data and ten biogeophysical predictors.

Time Range

2003 to 2021

Freshness

Last updated 2026-04-22 06:55:15; freshness should be verified.

Geography

Global land surface