🔭 astrophysics

HAMSTER: Hyperspectral Albedo Maps dataset with high Spatial and TEmporal Resolution

This paper introduces HAMSTER, a dataset of hyperspectral albedo maps with high spatial, spectral, and temporal resolution, generated via PCA regression on MODIS data and laboratory measurements to enable more accurate Earth energy budget calculations and climate simulations across the 400–2500 nm wavelength range.

Original authors: Giulia Roccetti, Luca Bugliaro, Felix Gödde, Claudia Emde, Ulrich Hamann, Mihail Manev, Michael Sterzik, Cedric Wehrum

Published 2026-05-27

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Giulia Roccetti, Luca Bugliaro, Felix Gödde, Claudia Emde, Ulrich Hamann, Mihail Manev, Michael Sterzik, Cedric Wehrum

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: Painting the Earth in Full Color

Imagine the Earth is a giant, complex painting. To understand how much sunlight the planet absorbs versus how much it bounces back into space (which is crucial for understanding our climate), scientists need to know the exact "color" of every patch of land, ocean, and ice.

For a long time, scientists have had a map of this painting, but it was like looking at a low-resolution photo that only had seven specific colors (spectral bands). It was like trying to describe a rainbow using only seven crayons. While useful, this limited view made it hard to calculate the Earth's energy budget perfectly because the "paint" (sunlight) interacts differently with surfaces at every single wavelength, not just those seven spots.

This paper introduces HAMSTER (Hyperspectral Albedo Maps with high Spatial and TEmporal Resolution). Think of HAMSTER as a magic paintbrush that takes those seven limited crayons and uses them to reconstruct the entire rainbow of colors for the Earth's surface, from violet (400 nm) to deep infrared (2500 nm).

How They Did It: The "Musical Chord" Analogy

The scientists didn't just guess the missing colors. They used a clever mathematical trick called Principal Component Analysis (PCA).

Here is an analogy: Imagine you have a library of thousands of musical recordings of different instruments (violins, drums, flutes, etc.). You want to recreate a specific song, but you only have a recording of the song played on a very basic, seven-key piano.

The Library: The researchers gathered a massive "library" of real-world surface "sounds" (spectra) from three different databases. These included dry soils, rocks, vegetation, snow, ice, and even man-made materials from 82 different countries.
The Piano: They took the existing "seven-key piano" data from NASA's MODIS satellites (which measures the Earth's reflectivity in seven specific bands).
The Reconstruction: Using PCA, they figured out the "musical chords" (principal components) that make up all those different surfaces. They then used the seven notes from the satellite data to figure out which "chords" were playing at any given spot on Earth.
The Result: Once they knew the chords, they could "play" the full song. This allowed them to fill in the gaps between the seven satellite bands and create a smooth, continuous spectrum for every pixel on the map.

What HAMSTER Looks Like

The final product is a dataset that is incredibly detailed in three ways:

Spatially (Where): It covers the Earth in high detail (0.05 degrees, roughly 5-6 km per pixel).
Temporally (When): It provides a map for every single day of the year, capturing how the Earth changes with the seasons.
Spectrally (Color): It doesn't just have seven colors; it has a smooth gradient of colors every 10 nanometers across the visible and near-infrared spectrum.

What They Found: The Earth's Seasonal Dance

By using these new, full-color maps, the authors were able to see things the old seven-color maps missed:

The "Vegetation Red Edge": They could clearly see the sharp jump in reflectivity that happens when plants start photosynthesizing (around 700 nm). It's like seeing a forest suddenly "light up" in a specific color as spring arrives, something the old maps missed because their "crayons" were too far apart to catch that specific transition.
Snow vs. Ice: They tracked how snow and ice melt. In winter, high-latitude areas (like Canada and Russia) look very bright (high albedo) because of snow. In summer, as the snow melts, the reflectivity drops. The maps show this dance clearly.
Deserts are Stable: Deserts like the Sahara and the Australian Outback barely change their "color" throughout the year. They are the steady, unchanging background of the Earth's painting.
Cities are Complex: Urban areas like Beijing and Mexico City showed a mix of colors. They aren't just one color; they are a blend of concrete, asphalt, vegetation, and soil. The maps showed that cities generally reflect less light (are darker) than forests or deserts.

Checking the Work: The "Double-Check"

To make sure their magic paintbrush worked, the scientists compared their new HAMSTER maps against other satellite instruments (SEVIRI and TROPOMI) that look at the Earth differently.

The Result: The new maps matched the other satellites very well. The differences were small (about the same size as the differences you'd see if you compared two different standard maps). This proves that their method of "filling in the gaps" is accurate and reliable.

Why It Matters

The paper concludes that having these full-spectrum maps is essential for climate models. Just as a painter needs the full spectrum of light to capture a realistic sunset, climate scientists need the full spectrum of surface reflectivity to accurately calculate the Earth's energy budget. Without this detail, models might miss subtle but important changes in how the planet heats up or cools down.

In short: The authors took a limited, seven-color view of the Earth and used a massive library of ground-truth data to mathematically "color in" the rest of the spectrum, creating a high-definition, full-color movie of the Earth's reflectivity that changes every day of the year.

Technical Summary: HAMSTER: Hyperspectral Albedo Maps dataset with high Spatial and TEmporal Resolution

Problem Statement
Surface albedo is a fundamental parameter for calculating the Earth's energy budget and regulating climate dynamics, particularly through feedback mechanisms involving snow and ice. While satellite instruments like the Moderate Resolution Imaging Spectroradiometer (MODIS) provide continuous monitoring of surface reflection, they are limited to seven spectral bands in the visible and near-infrared (VIS–NIR) range. Most Earth System Models (ESMs) rely on simplified broadband assumptions (e.g., two-stream approaches with fixed values for photosynthetically active radiation and near-infrared bands) or lack the spectral resolution to capture the distinct spectral signatures of different land surfaces. This limitation introduces uncertainties in radiative-transfer simulations, cloud property retrievals (e.g., using the O2 A band), and the modeling of regional climate variability. A complete set of albedo maps covering the entire wavelength range is required to accurately simulate radiance spectra and retrieve atmospheric properties, yet such hyperspectral data are not directly available from current satellite missions for every wavelength.

Methodology
The authors introduce a novel methodology to generate hyperspectral albedo maps by combining satellite observations with laboratory spectral measurements using a Principal Component Analysis (PCA) regression algorithm.

MODIS Climatology Construction: The process begins with the MODIS Surface Reflectance Black-Sky Albedo dataset (MCD43D version 6.1). The authors compile a 10-year climatology (2013–2022) for each Day of Year (DOY) across seven MODIS bands. To address missing data caused by cloud cover or solar geometry constraints, a multi-step filling procedure is employed:
- Temporal interpolation using neighboring DOYs (within a 40-day window).
- Linear interpolation to balance seasonal transitions near solstices.
- Spatial averaging over surrounding grid boxes (3x3 to 9x9).
- Longitudinal averaging within latitude bands.
- Global averaging over the 10-year period for persistent gaps.
- Ocean pixels are filled using a "deep-ocean" spectrum from the USGS ECOSTRESS library.
Spectral Training Data: To reconstruct the full spectrum, the authors aggregate hyperspectral reflectance measurements from three distinct datasets:
- ECOSTRESS Library: 1,023 spectra from the US, including vegetation, rocks, soils, man-made materials, and water/ice.
- ICRAF–ISRIC: A global dataset of 4,440 soil spectra from 58 countries.
- LUCAS: 21,782 soil spectra from 28 EU countries (corrected for negative values in the 400–500 nm range using a regression model trained on ICRAF–ISRIC data).
- In total, 26,635 spectra covering 400–2500 nm are used. To balance the dataset and prevent soil spectra from overwhelming vegetation and water signatures, a k-means clustering algorithm is used to select representative spectra (100 for ICRAF–ISRIC, 100 for LUCAS, and 128 for ECOSTRESS).
PCA Regression: The authors apply PCA to the matrix of laboratory spectra ( $U$ ) to extract principal components (eigenvectors). Six principal components are derived from the soil and vegetation data, and a seventh constant eigenvector is added, following previous studies (Vidot and Borbás, 2014; Jiang and Fang, 2019).
- The MODIS albedo vector ( $R$ ) for the seven bands is inverted to solve for the coefficient vector ( $c$ ) using the equation $c = R \tilde{U}^{-1}$ , where $\tilde{U}$ is the convolved principal component matrix matching the MODIS bandpasses.
- The full hyperspectral albedo spectrum ( $R_\lambda$ ) is then reconstructed for every pixel and DOY using $R_\lambda = c \tilde{U}_\lambda$ .

Key Contributions and Results
The primary output is the HAMSTER (Hyperspectral Albedo Maps dataset with high Spatial and TEmporal Resolution) dataset, characterized by:

Spectral Resolution: 10 nm steps from 400 to 2500 nm.
Spatial Resolution: 0.05° (180 arcsec) in latitude and longitude.
Temporal Resolution: 1 day, averaged over the 2013–2022 period.

Validation:

MODIS Reconstruction: The hyperspectral maps successfully reconstruct the original MODIS climatology with a Root-Mean-Square Error (RMSE) of less than 0.0003 for all seven bands.
SEVIRI Comparison: Validation against the Spinning Enhanced Visible and Infrared Imager (SEVIRI) on the Meteosat satellite shows discrepancies of up to 0.06, consistent with known differences between various satellite land surface products. The RMSE is slightly higher for the SEVIRI channel furthest from MODIS bands (810 nm).
TROPOMI Comparison: Comparison with the TROPOMI Lambertian-equivalent reflectivity (LER) product over Africa and the Middle East (excluding snow/ice regions) yields RMSE values between 0.02 and 0.07, confirming the model's performance even in spectral regions far from the training MODIS bands (e.g., the vegetation red edge).

Spectral Analysis of Land Surfaces:
Using HAMSTER, the authors analyze the spectral profiles of various land types:

Forests: Tropical rainforests show minimal seasonal variation, while boreal forests exhibit significant reflectivity changes due to snow cover. All forests display the characteristic Vegetation Red Edge (VRE) steep increase around 700 nm.
Deserts: Generally show low seasonal variability, though mineralogical composition causes significant spectral differences between deserts (e.g., Sahara vs. Gobi).
Ice Sheets: Greenland and Antarctica show high VIS reflectivity (~1.0) decreasing in the NIR. Seasonal melting leads to lower albedo in summer.
Urban Areas: Cities like Beijing and Mexico City show lower overall albedo (<20%) and complex spectral shapes resulting from linear combinations of man-made materials, soil, and vegetation.

Significance
The paper claims that HAMSTER enables the refinement of Earth's energy budget calculations and the improvement of climate simulations by providing a complete, high-resolution spectral view of surface albedo. The dataset allows for the study of seasonal variability across the entire VIS–NIR spectrum, which is impossible with standard broadband or limited-band products. The authors note that the dataset has already been implemented in the radiative-transfer software package libRadtran. They highlight that previous studies (e.g., Braghiere et al., 2023) have shown that using simplified broadband albedo assumptions can lead to significant deviations in radiative forcing, cloudiness, and surface temperature predictions; HAMSTER offers a pathway to mitigate these errors by providing the necessary hyperspectral detail for both global and regional climate models.