ÃÛÌÒ´«Ã½Æƽâ°æÏÂÔØ

Snow • Remote sensing • Software • Computing

I'm a detective of satellite images, using remote sensing algorithms to uncover secrets hidden within pixels. My focus is on clouds and snow data—how much there is and how reflective it is—which then helps communities build resilience to snowmelt water patterns. It's about using tech to crack environmental puzzles.

Current and Upcoming Work

Ìý

Starting in September 2025, I will develop a prototype to distinguish snow and clouds in Alaska using VIIRS satellite imagery. This deep-learning-based solution, trained with CALIOP data, addresses a persistent challenge in polar regions. The goal is to demonstrate the feasibility of a new product with improved accuracy, providing crucial data for a diverse group of stakeholders, including industries, energy, transportation, and insurance companies, as well as landowners. Ultimately, this research will help build stronger, more resilient communities. I will present this work at the AGU Fall Meeting 2025 in the session "A049. Decision-Relevant Understanding of Impactful Weather and Extremes."

Ìý

Ìý

Recent Research at INSTAAR (2022-2025)

Ìý

At INSTAAR, my research focused on improving the accuracy and accessibility of snow and albedo products. I refactored spectral unmixing algorithms (STC and SPIReS), updating SPIReS to use VIIRS data.

I designed and built a near-real-time pipeline that now provides critical snow and albedo data to hydrologists, water resource managers, and other stakeholders in the Western US, where snowmelt is a vital resource. This pipeline is run by NSIDC and also feeds their public-facing Snow Today web application, allowing anyone to track snow patterns.

Ìý

Ìý

Prior Work and Expertise (Before 2022)

Ìý

My PhD research in Earth Sciences (2019) demonstrated that long-term erosion patterns in the Himalayas were not significantly impacted by past climate change. I achieved this by combining cosmogenic nuclide measurements and numerical modeling.

From 2018 to 2021, I taught Earth and environmental sciences, GIS, geostatistics, and software development as an Assistant Professor in France. My work also includes developments in landscape modeling and algorithms for seismic landslide modeling in collaboration with researchers at INSTAAR/CIRES and the Université de Rennes.

Publications

Ìý

Education

• PhD (Geosciences):ÌýUniversite de Lorraine, 2019
• MS (Earth & universe sciences): Paris-Sud University (Paris XI), 2015
• MS (Computer science): Ecole des Mines d'Alès, 2005
• BA (Logistics & transport business management):ÌýUniversité d'Évry, 1999

Deep Learning-Based Discrimination of Polar Cloud and Snow Using VIIRS Imagery.

Accurate cloud detection is essential for Earth observation; however, reliably distinguishing clouds from snow and ice remains a persistent challenge, especially in dynamic polar regions and under low-light conditions. This impacts the accuracy of cryospheric products and hinders robust atmospheric and climate analyses. This study presents a novel deep learning prototype centered onÌýVIIRSÌýobservations to improve cloud-snow discrimination, addressing a long-standing challenge in optical remote sensing.

The proposed prototype is applied to complexÌýAlaskan scenes, representing a testbed for polar remote sensing. It leverages VIIRS multispectral data, including the unique Day/Night Band (DNB) for enhanced nighttime detection, the 1.61 µm (I3) band for its pronounced snow-cloud reflectance contrast, and thermal infrared bands (I4, I5, M14, M15) to incorporate surface temperature and cloud phase information. The prototype employs aÌýU-NetÌýdeep learning architecture to perform pixel-level semantic segmentation, a proven method for handling complex spatial patterns in remote sensing imagery. This architecture effectively fuses multi-scale features, enabling precise cloud delineation even for subtle structures.

We are constructing a high-quality, regionally-focused dataset derived from challenging VIIRS granules over Alaska to support model training and evaluation. This involves rigorous manual refinement of initial masks, supplemented by collocatedÌýCALIOPÌýlidar data for robust, independent ground truth. This prototype is especially designed to enhance the detection of thin clouds over snow and ice, particularly under polar night conditions, representing a substantial advancement over conventional threshold-based methods.

This research demonstrates the transformative potential of advanced computer vision techniques in optimizing the use of VIIRS observations for both cryospheric and atmospheric applications. Improved cloud-snow discrimination enables the generation of more accurate cloud masks, which directly enhances the reliability of snow and ice products. In turn, this contributes to a more precise retrievals of cloud property in climatically sensitive polar regions, providing invaluable input for climate modeling efforts and operational forecasting systems.

Ìý

Collaborations

My work is supervised by Albert Kettner and funded until August 2025 on NASA grants for projects of Karl Rittger. I collaborated with Ross Palomaki, and the Members of the Mountain Hydrology Group, along with the Members of the CSDMS Group, Eric Hutton, Mark Piper, and Greg Tucker.

Mountain Hydrology GroupÌý