Albedo change from snow algae blooms can contribute substantially to snow melt in the North Cascades, USA

The Earth's energy balance depends heavily on the high reflectivity of snow surfaces. Clean snow reflects up to 99% of incoming solar radiation, mitigating atmospheric warming. However, even small reductions in snow cover or albedo significantly diminish this protective effect, contributing to climate change. Various microorganisms, including snow algae, inhabit snowpacks. Snow algae, flourishing in summer months with sufficient interstitial water, produce photoprotective pigments (primarily astaxanthin) that color the snow red, reducing albedo by approximately 20%. Precisely quantifying the impact of this reduced albedo across vast glacier and snowfield areas is challenging with in-situ studies alone. This study leverages the increasing capabilities of uncrewed aerial vehicles (UAVs) to assess the prevalence and albedo influence of snow algae across large, inaccessible areas. Previous cryosphere research using UAVs has focused on Structure-from-Motion (SfM) for high-resolution digital elevation models (DEMs) to evaluate snow and glacier ablation, glacier dynamics, and snow depth retrieval. While UAV studies of glacier algae exist (e.g., Greenland Ice Sheet), remote detection of snow algal blooms at mid-latitudes using UAVs is relatively unexplored. Previous research on snow algae detection primarily employed satellite imagery (e.g., AVIRIS, SPOT, Landsat, Sentinel-2A), but these often lack closely corresponding ground validation data and have limited spectral or spatial resolution. This study aims to adapt red/green indices used in satellite remote sensing to the UAV level, validating the approaches with coupled in situ snow algae data, potentially paving the way for broader application with more affordable hyperspectral sensors. The North Cascades' declining snowpack due to climate change exacerbates glacier melt, and snow algae's contribution to this melt is not currently factored into regional watershed models. This study addresses this gap by mapping snow algae extent and impact on a large scale, improving our understanding of North Cascades snowpack stability.

Existing literature extensively covers the impact of light-absorbing impurities, such as black carbon, on snow albedo and subsequent melt. However, the role of biological impurities, specifically snow algae, remains relatively less studied, especially in mid-latitude regions. Studies using satellite imagery have demonstrated the potential for remote detection of snow algae, using indices based on the ratio of red to green reflectance. While satellite-based approaches offer large-scale coverage, limitations exist in terms of spatial and spectral resolution, and the availability of coincident ground-truthing data. The use of UAVs in cryospheric research is growing, offering high-resolution spatial data and the potential for more frequent monitoring. However, dedicated studies focusing on snow algae detection using UAV-based multispectral imagery at mid-latitudes are limited. Previous work has explored the use of principal components analysis (PCA) in remote sensing applications, such as mapping snow cover and avalanche debris. However, to the authors' knowledge, the use of PCA for snow algae detection has not been previously explored. Radiative transfer models, such as SNICAR, have been used extensively to model the impact of light-absorbing impurities on snow albedo. Recently, SNICAR has been adapted to incorporate snow algae, allowing for more accurate estimations of their impact on snowmelt.

This study utilized a DJI M210 UAV equipped with a MicaSense Dual camera system (RedEdge MX and RedEdge MX Blue) to map snow algae in a 0.1 km² basin in the North Cascades. Two approaches were employed: principal components analysis (PCA) and spectral indexing. UAV flights and coincident ground sampling were conducted on July 2, 2021, and July 30, 2021, with additional ground sampling on other dates. Snow algae cell concentrations and pigment concentrations (chlorophyll a, chlorophyll b, lutein, astaxanthin, beta-carotene) were determined using Guava Flowcytometry and HPLC. For PCA, the 10-band multispectral imagery was reduced to three principal components, and thresholds were set based on training points to classify snow algae. Spectral indexing used a red/green band ratio, optimized based on training points, to map snow algae. A snow mask (R_R > 0.3) was applied to limit analysis to snow-covered areas. The optimized red/green index was defined as R_R − 0.015 > 1.02 × R_G. The SNICAR model was used to simulate spectral albedo based on field data, and instantaneous radiative forcing (IRF) was calculated. The IRF was calculated as the sum of the wavelength-specific downward flux, E_d, multiplied by the wavelength interval and the difference between clean snow reflectance, R_clean, and snow algae-covered snow reflectance, R_algae, over the wavelengths 350–850 nm. The study area was the Bagley Lakes basin in the Mount Baker-Snoqualmie National Forest. The accuracy of the orthomosaics generated was assessed using ground control points (GCPs) and ground validation points (GVPs). The study area was reasonably accessible, allowing for repeat surveys.

The snow algae cell concentrations were significantly higher on July 2nd (peak bloom) compared to July 30th (post-peak). The ratio of astaxanthin to chlorophyll a increased on July 30th, suggesting a shift towards photoprotective pigment production. PCA effectively separated snow algae from clean snow on July 2nd, but less so on July 30th, possibly due to increased impurities in the snowpack as the season progressed. The optimized red/green band index also performed better on July 2nd. On July 2nd, approximately 1% of the surface snow was covered by snow algae using PCA, while the optimized red/green index estimated 1.16% coverage. On July 30th, PCA indicated 2.06% coverage, while the optimized index showed 1.37%. The optimized red/green index showed a significant correlation with the natural log of cell concentration for the July 30th data (r² = 0.86), but the correlation was not significant when both survey dates were combined (r² = 0.16). The SNICAR model showed a strong relationship between snow algae cell concentration and spectral albedo, with increasing cell concentration decreasing albedo in the visible spectrum. The average IRF was 236.56 ± 79.55 W m⁻² on July 2nd and 88.86 ± 40.96 W m⁻² on July 30th. The total estimated snowmelt due to snow algae in the basin was 1508 ± 536 m³. This accounted for approximately 0.02% of the total snowmelt observed in the basin between the two surveys. The maximum instantaneous radiative forcing reported for snow algae in this study was 360 Wm⁻², exceeding values found in previous studies at higher latitudes.

This study demonstrates the feasibility of using UAV-based multispectral imagery to detect and map snow algae blooms in mid-latitude regions. The results highlight the importance of considering bloom conditions when developing and applying algorithms for snow algae detection. The PCA approach provides a valuable method for exploring spectral features and identifying optimal band combinations, while the optimized red/green index offers a more readily applicable method for large-scale mapping. The strong correlation between the optimized red/green index and cell concentration on July 30th suggests the potential for accurate prediction of snow algae abundance, but the weaker correlation when combining data from both surveys underscores the influence of bloom stage and pigment ratios. The significant radiative forcing caused by snow algae highlights the need for their inclusion in regional watershed melt models and global climate models. The higher IRF values observed in this study compared to previous research at higher latitudes suggest that snow algae might play a more substantial role in snowmelt at mid-latitudes.

This research successfully demonstrated the ability to remotely map snow algae in mid-latitude regions using UAV-based multispectral imagery, employing both PCA and spectral indexing approaches. The findings underscore the significant contribution of snow algae to snowmelt via albedo reduction, particularly during peak bloom conditions. The development of optimized indices and the use of PCA for exploration suggest future directions for improving the accuracy and application of remote sensing methods for snow algae detection. The incorporation of snow algae in climate and watershed models is crucial for better understanding and prediction of snowmelt in a changing climate. Future studies should focus on the development of more robust algorithms that account for varying bloom intensities and the integration of these data into physically-based snowmelt models.

The study focused on a single basin and may not be fully generalizable to other locations. The limited number of sampling dates could not fully capture the dynamic nature of snow algae blooms. The SNICAR model used required scaling of pigment dry mass fractions to fit within the model's allowable range, which could affect the accuracy of albedo and IRF estimates. Additionally, the study did not account for other impurities in the snowpack, which may have influenced the albedo and melt rates. Further research with a larger dataset, more frequent sampling, and the inclusion of other factors is recommended.