Sea ice harbors diverse microbial communities, with phototrophic ice algae playing a crucial role in primary production and the food web. The distribution of these algae is influenced by gradients in temperature, salinity, porosity, and light transmittance, leading to significant microspatial variability. Current methods for measuring algal biomass lack the resolution to capture this variability, hindering mechanistic understanding and predictive modeling. This study addresses this gap by employing hyperspectral imaging (HI), a technique capable of quantifying biogeochemical properties at a per-pixel level. HI has revolutionized observations in terrestrial and marine ecosystems, and recent work has demonstrated its qualitative potential for capturing sub-mm scale biomass variability in sea ice. This research advances HI technology by developing quantitative methods for mapping the fine-scale distribution of sea-ice algae using chlorophyll *a* (Chl *a*) as a proxy. The goal is to provide spatially explicit quantitative estimates of Chl *a* concentration and demonstrate the critical role of HI in making scalable under-ice observations.

Existing sea-ice sampling methods, including ice coring, under-ice bio-optical sensing using L-shaped arms, and unmanned underwater vehicles (UUVs), have limitations in resolving the mm-scale spatial variability of algal biomass. While UUVs improve spatial coverage, their sensor footprints limit resolution. Traditional methods for assessing algal biomass include ice coring followed by laboratory analysis, but these methods can cause significant disruption and biomass loss. Moreover, the use of under-ice optical sensing from UUVs has broadened the spatial coverage, but resolution remains limited due to the large footprints of the underwater radiance sensors. Previous studies have used hyperspectral imaging (HI) to qualitatively assess biomass variability, but quantitative relationships applicable to specific sensor configurations and environments have been lacking. This study aimed to address these gaps by developing and testing novel spectral indices for quantitative biomass mapping.

The study was conducted at Cape Evans, Antarctica, during November-December 2018. Forty-two ice cores (14 cm diameter) were extracted from areas with varying snow cover. The ice-water interface microtopography was determined using structure-from-motion (SfM) digital photogrammetry. A custom optical setup was designed to measure transmitted radiation along the vertical and horizontal axes of ice cores at high spectral and spatial resolution. Both *in situ* (using an under-ice sliding platform) and *ex situ* (on extracted cores) hyperspectral images were acquired. The *ex situ* scans focused on the lower 9 cm of six cores (vertical scans) and the bottom 3 cm of all 42 cores (horizontal scans). Two LED light sources (solar-simulating and white) were used for illumination. Image pre-processing involved radiometric correction, masking, spectral subsetting (400-700 nm), and Savitzky-Golay filtering to reduce noise. Principal Component Analysis (PCA) was applied to explore spatial variability in the *ex situ* imagery. Chlorophyll *a* (Chl *a*) was quantified fluorometrically from melted ice-core samples. Several spectral indices (NDI, AUC_650-700, ANCB_650-700, ANMB_650-700, LAUC_650-700) were calculated from the transmittance spectra and regressed against Chl *a* values to develop predictive models. The best-performing model was applied to both *ex situ* and *in situ* images to create quantitative Chl *a* maps. Spatial autocorrelation and complexity of the *in situ* biomass map were analyzed using variograms and gradient magnitude calculations.

The study area at Cape Evans exhibited a relatively flat under-ice habitat with large cavities and brine channels. The mean Chl *a* concentration in the bottom 3 cm of the 42 cores was 18.74 ± 18.04 mg m⁻². SfM photogrammetry revealed detailed microtopography, including a skeletal layer, brine channels, and cavities. PCA of the *ex situ* images effectively separated light intensity variability (PC1) from Chl *a* concentration (PC2). Among the tested spectral indices, LAUC_650-700 (log-transformed area under the curve) showed the best performance in predicting Chl *a* (R² = 0.84), outperforming NDIs and AUC_650-700. The LAUC model was applied to both *ex situ* and *in situ* images to generate high-resolution quantitative maps of Chl *a* distribution. The *in situ* map showed highly heterogeneous Chl *a* patterns, with an autocorrelation length scale of ~12 cm and significant mm-scale gradients. The average pixel-based biomass estimates for the six core samples showed good consistency with their respective sampled values (variability of 25–35%).

The results demonstrate the effectiveness of hyperspectral imaging for mapping microspatial variability in ice algal biomass. The LAUC_650-700 index provided a robust and accurate method for quantitative Chl *a* estimation, overcoming limitations of traditional indices. The high-resolution maps revealed previously unseen details of biomass patchiness, revealing the influence of under-ice topography and microstructures on algal distribution. This has implications for understanding ecological processes within sea ice and improving biogeochemical models. The study highlights the critical need to improve sampling resolution to match the capabilities of high-resolution imaging techniques for accurate validation and model development.

This study demonstrates the successful application of hyperspectral imaging for high-resolution quantitative mapping of ice algal biomass. The developed LAUC_650-700 index offers a robust approach for predicting chlorophyll *a* concentrations. The high-resolution maps provide unprecedented insights into the microspatial heterogeneity of ice algal communities. Future research should focus on integrating this methodology with other techniques (e.g., time-lapse imaging, physical measurements) to provide a more holistic understanding of sea ice ecology and biogeochemistry. Improved validation techniques at relevant scales are also crucial.

The study primarily focused on a specific sea ice type (fast ice) with a relatively simple composition and minimal snow cover. The applicability of the developed models to other sea ice types and conditions with varying snow cover requires further investigation. Validation of the high-resolution *in situ* data is limited by the lack of comparable sampling methodologies at the mm-scale. Furthermore, the current approach focuses on the bottom 3 cm of ice cores; modification of the workflow might be necessary for sea ice exhibiting different vertical biomass distributions.