Climate AI Nordics

Climate AI Nordics

Nordic network for collaborations around tackling climate change using AI and machine learning.

Featured paper: Separating the Albedo-Reducing Effect of Snow Impurities Using Machine Learning

Lou-Anne Chevrollier from Aarhus University and Adrien Wehrlé from Zurich University collecting snow reflectance data at the Finse Research Station.

Authors: Lou-Anne Chevrollier, Adrien Wehrlé, Joseph M. Cook, Norbert Pirk, Liane G. Benning, Alexandre M. Anesio, and Martyn Tranter.

Paper: https:doi.org/10.5194/tc-19-1527-2025, The Cryosphere, 2025, 19, 1527–1538.

Snow albedo—the fraction of sunlight reflected by snow—plays a key role in Earth’s energy balance. When snow is contaminated by light-absorbing particles (LAPs) such as mineral dust, black carbon, or pigmented algae, its albedo decreases, leading to enhanced absorption of solar radiation. This process influences glacier mass balance and represents an important source of uncertainty in melt forecasts because it varies greatly in time and space, and remote detection methods able to identify the different types of LAPs are lacking.

To address this challenge, the study introduces an AI-based inversion framework using a deep-learning emulator of a radiative transfer model. Trained on over 5.8 million simulations, the emulator predicts spectral albedo from key surface parameters such as snow grain size, liquid water content, and LAP concentrations. Coupled with a gradient-based inversion algorithm, this approach simultaneously quantifies the role of each individual LAP type in surface darkening. The use of AI in this study enabled a speed up of the original model by 30 times while maintaining high accuracy, making it suitable for large-scale applications using drone or satellite imagery.

Figure 2 reproduced from the original study, showing the agreement between model and observations as well as the retrieved surface parameters for different surface types (from left to right): clean snow, algal-loaded snow, dust-loaded snow, black particles-loaded snow.

Applied to 180 observations collected at the Finse Research Station, the method revealed strong and variable impacts of LAPs on snow albedo. Red algal blooms reduced albedo by up to 0.13, mineral dust by 0.21, and dark particles by 0.25, corresponding to daily radiative forcing values as high as 44 W/m². Importantly, the presence of multiple LAPs attenuated individual effects by up to a factor of three, highlighting complex interactions that must be considered in predictive models. This open-source method enables high-resolution mapping of snow impurities and initiates progress toward improved climate projections and melt assessments.

You May Also Enjoy

Climate AI Nordics launches official YouTube channel

Published:

To celebrate our one-year anniversary, Climate AI Nordics is launching a new YouTube channel dedicated to sharing knowledge within the AI and climate community. The channel features a complete set of recordings from the inaugural Nordic Workshop on AI for Climate Change held in Gothenburg this past May. In addition to workshop content, viewers can access an extensive library of past webinars covering topics such as biodiversity monitoring, weather forecasting, and Earth system modelling. We invite researchers and practitioners to explore these valuable resources and join us as we look forward to another year of collaboration and innovation.

Featured member, November 2025: Joakim Bruslund Haurum

Published:

Joakim Bruslund Haurum, an Assistant Professor at the University of Southern Denmark’s new Vejle Campus and the Pioneer Centre for AI, is the featured member for November 2025. He is a Computer Vision researcher specializing in expert tasks, balancing pure research on algorithms with applied work in infrastructure inspection and biodiversity monitoring. Beyond his studies on hierarchical data, Joakim engages with the scientific community (and with Climate AI Nordics) by co-organizing events like the AI for Climate and Conservation workshop at EurIPS. He is currently seeking collaborations with others interested in solving challenges related to messy data in expert domains.

TESSERA: precomputed FAIR global pixel embeddings for earth representation and analysis

Event date:

Webinar with Zhengpeng (Frank) Feng, Cambridge. Petabytes of satellite Earth Observation (EO) data are freely available and can address critical global challenges. However, EO data quality is poor due to clouds and variable lighting conditions. To address this, practitioners typically use compositing, but this critically removes the temporal phenological signal. Moreover, supervised machine learning to map composited pixels to task-specific classes requires accurately labelled data that are rarely available. We present TESSERA, a pixel-oriented foundation model for EO time series that creates 128-dimensional latent embeddings requiring only a few labels for task-specific training to achieve state-of-the-art performance across diverse complex tasks. TESSERA uses two encoders that combine optical data with synthetic aperture radar backscatter coefficients at 10m resolution, creating embeddings fused with a multilayer perceptron to generate annual global embedding maps. TESSERA closely matches or outperforms state-of-the-art task-specific models and other foundation models across five diverse downstream tasks. It is unprecedented in ease of use, scale, and accuracy: no other open foundation model provides precomputed outputs with global, annual coverage at 10m resolution.