Go with the floe: Spatio-temporal evolution of Arctic sea ice albedo using satellite imagery in a Lagrangian framework

Arctic sea ice plays a key role in the polar shortwave (SW) surface energy/radiative budget. At the start of the polar day in March, the central Arctic ice pack is relatively homogeneous and slow-moving. However, by the August melt peak, the ice transitions into a highly dynamic regime characterized by rapid drift and the development of extensive melt ponds, leads/breakups, and deformations, which drastically change the albedo within a few days. This complex nature of sea ice is a key reason why climate models struggle to accurately characterize surface albedo in the Arctic. Observationally, polar-orbiting satellites encounter several unique challenges in the Arctic. First, the observations need to be clear of clouds and atmospherically corrected; however, low-level clouds are ubiquitous in the Arctic and are difficult to detect using existing cloud detection algorithms. Second, it is assumed that the surface does not move while these observations are acquired over several days, an assumption that is invalid for Arctic sea ice due to the drift. Third, traditional land Bidirectional Reflectance Distribution Function (BRDF) models are insufficient to capture the anisotropic property of snow and sea ice to accurately estimate its albedo. Finally, the sparsity of in-situ and field observations in the Arctic has limited any development of a satellite data product for sea ice albedo. 

 

Despite these challenges, the frequent overpasses of polar-orbiting satellites over the polar regions provide valuable opportunities for Arctic surface remote sensing through the abundance of observations. We present a Lagrangian framework for tracking sea ice using a multi-overpass, multi-angular approach. Instead of observing sea ice at geographically fixed locations, we use a moving reference frame that “goes with the floe”. Using a suite of existing satellite (MODIS) data products in a scalable, modular pipeline, we employ machine learning to identify sea ice floes in a given scene. Once a floe is identified, we utilize a composite, kernel-driven snow BRDF model to populate the angular space, integrating these samples to derive daily spectral and SW broadband albedo. We then track each identified floe in a scene across multiple days (when possible), enabling us to build a spatio-temporal evolution of the albedo. Crucially, we use data from the NASA ARCSIX aircraft campaign, which took place during late spring and summer of 2024, to validate the satellite-derived albedo. Measurements from two instruments—All-Sky Camera (nadir-looking; 400–650 nm) and Solar Spectral Flux Radiometer (SSFR; 400–2000 nm)—are used to evaluate the accuracy and quantify the uncertainty in the satellite albedo product. By leveraging the multi-angular sampling from multiple MODIS instruments within this Lagrangian framework, we capture the change in albedo associated with the onset of melt and the subsequent increase in surface anisotropy and heterogeneity. Our work is a step towards developing an operational sea ice BRDF/albedo product for passive imagers like MODIS.