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

The study addresses how biological impurities—specifically snow algae—reduce snow albedo and enhance snowmelt, a process less quantified than impacts from inorganic light-absorbing particles. Snow algae develop photoprotective pigments (notably astaxanthin) that redden snow and can reduce albedo by around 20%, potentially accelerating melt and affecting watershed hydrology. With declining snowpack in the North Cascades and projected further reductions, understanding biotic darkening is important for regional melt modeling. Remote sensing offers scalable observation of snow algae, but mid-latitude UAV-based detection has been limited. This work aims to (1) detect and map snow algae using UAV-mounted multispectral imaging via principal component analysis (PCA) and a red/green spectral index, (2) validate remote observations with coincident in situ pigment and cell measurements, (3) model spectral albedo impacts using SNICAR, and (4) quantify radiative forcing (RF) and associated snowmelt over a basin in the North Cascades.

Previous work has shown snow algae’s role in reducing snow albedo and accelerating melt, with early remote detection using AVIRIS and subsequent satellite studies (SPOT red/green ratio; applications to Landsat and Sentinel-2). Upcoming spaceborne imaging spectroscopy missions (PRISMA, EnMAP, NASA SBG) and UAV hyperspectral sensors promise improved detection of subtle spectral absorptions from algae. UAVs have been widely applied in cryospheric research for DEMs, ablation mapping, glacier dynamics, and snow depth retrievals, and for glacier algae on the Greenland Ice Sheet, but fewer studies target mid-latitude snow algae. PCA is established in remote sensing for feature extraction and band reduction across applications (snow cover, avalanche debris, wetlands), yet has not been broadly used for snow algae. Regional context indicates decreasing North Cascades snowpack, likely expanding algal habitat and influence, underscoring the need to integrate biological darkening into models. SNICAR provides a framework for modeling snow albedo with impurities and was recently extended to include algae contributions within a unified code base.

Study site: Bagley Lakes basin, Mount Baker–Snoqualmie National Forest (48.854°N, 121.692°W, 1277 m a.s.l.), ~0.1 km², with annual red snow algae blooms. Avalanche runout and lake areas were masked due to terrain and photogrammetry limitations. UAV surveys: DJI Matrice 210 with MicaSense Dual camera (RedEdge MX + RedEdge MX Blue) acquiring 10 bands: coastal blue (430–458 nm), blue (459–491), green-1 (524–538), green-2 (546–574), red-1 (642–658), red-2 (661–675), red edge-1 (700–710), red edge-2 (711–723), red edge-3 (731–749), near-infrared (813–871). RAW 12-bit images at 1 s interval. Flights on July 2, 2021 and July 30, 2021 (snow-on), and Sept 24, 2021 (snow-free). Flight altitude 90 m AGL, nadir view, 75% frontlap/sidelap, ~7 cm GSD, ~20 min per survey. Pre-flight reflectance calibration panel images acquired. Surveys near solar noon under direct sunlight. Positioning and control: EMLID Reach RS2 RTK GNSS base-rover setup. July 2: 12 targets (9 GCPs, 3 GVPs), total GVP accuracy 0.37 m; July 30: 10 targets (7 GCPs, 3 GVPs), GVP total error 20.69 cm; orthomosaics with 7 cm pixels; lateral XY error < pixel size. Ground sampling: Coincident with UAV flights, GPS-located snow algae samples collected from 0.1 m × 0.1 m × 2 cm depth using metal spatula into WhirlPak bags, kept dark, melted ~24 h. July 2: n=9; July 30: n=10. Additional dates: June 26 (n=5), July 9 (n=10), July 18 (n=10), July 22 (n=10). Analyses included pigments, cell counts, and ash-free dry mass (AFDM). Pigment analysis: Triplicate filters per homogenized sample; HPLC-QTOF (Agilent 1290 Infinity 2 + 6545XT LC-QTOF) with specified buffers, flow, gradient, and MS settings. Quantified beta carotene, lutein, astaxanthin, chlorophyll a, chlorophyll b; verification via mass fragments. Cell concentration: Samples fixed to 2% glutaraldehyde, stored at −20 °C. Guava easyCyte flow cytometer (blue laser); bead checks; four 200 µL replicates per sample; cells identified by high red fluorescence and forward scatter, validated by microscopy. AFDM: Filters dried at 104 °C for 1 h, combusted at 550 °C for 1 h; AFDM computed as pre- vs post-combustion mass difference normalized by volume. Image processing: Agisoft Metashape Pro 1.6.4 workflow with MicaSense reflectance calibration; altitude adjusted via Python script to meters a.s.l.; image quality filtering (threshold 0.5). Alignment high accuracy; removal of points with >0.5 pixel reprojection error; dense cloud generation (high quality, aggressive depth filtering); DEM and orthomosaic creation. GCPs/GVPs manually marked in ≥6 images each. PCA approach: ENVI 5.6 Forward PCA (rotate/new stats) on 10-band orthomosaics. Evaluation with 100 random training points (Snow vs Other) plus in situ algae points. Determined PC combinations that best separate algae vs clean snow; defined thresholds and applied basin-wide. Optimized red/green index: Developed from literature-based red/green (Takeuchi) ratio using MicaSense bands (red 642–658 nm; green 524–538 nm). Snow mask: pixels with red band reflectance R_R > 0.3. Optimized index threshold based on training data: classify algae if R_R − 0.015 > (1/1.02) × R_G (equivalently, Band 5 > 1.02 × Band 3 + 0.015). Applied within snow mask to estimate algae-covered area. Albedo and radiative forcing modeling: SNICAR-ADv3 used to simulate spectra for clean snow and algae-laden snow using measured cell and pigment data. Due to SNICAR input bounds (primarily tuned for glacier algae), pigment dry mass fractions were scaled by factor 0.05 to minimum allowable range (per communication with model developer). Parameters: average cell diameter 15 µm; solar zenith angle 40° (July 2) and 43–50° (July 30); snow depth from DEM differencing (snow-on vs snow-free); snow density 600 kg m−3; spheroidal grains, effective radius 100 µm; dust set to default/zero. Instantaneous radiative forcing (IRF) computed as spectral sum over 350–850 nm of downward flux (from pvlighthouse) times (R_clean − R_algae) × Δλ. Scaling to melt: Daily RF energy computed using mapped algae area; melt volume estimated assuming 0.334 MJ to melt 1 kg of snow and wet snow density 600 kg m−3. Basin-wide July melt integrated assuming linear scaling between survey dates; compared with DEM-derived total snowmelt.

- Bloom intensity: Average cell concentration on July 2 was 431,000 ± 398,000 cells/mL versus 55,000 ± 60,000 cells/mL on July 30 (nearly 8× reduction). Chlorophyll a concentrations were 29× higher on July 2 than July 30. Pigment ratios showed increased astaxanthin:chlorophyll a and decreased lutein:chlorophyll a by July 30, indicating a shift toward photoprotective pigments and post-peak bloom conditions. - PCA ordination: July 2—PC1 explained 99.67%, PC2 0.30%, PC3 0.01% of variance; July 30—PC1 99.78%, PC2 0.19%, PC3 0.01%. July 2 classification used PC1 > −90,000 and PC2 < −2,500, mapping ~1% algae coverage. July 30 used PC1 < 50,000 and PC3 < −500, mapping 2.06% algae coverage of remaining snow. - Optimized red/green index: Clear separation on July 2; less effective on July 30. Snow mask R_R > 0.3. Optimized threshold: Band5 > 1.02 × Band3 + 0.015. Mapped algae coverage: July 2—1.16% of snow area (1,352.18 m² of algae over 116,226.95 m² snow); July 30—1.37% (556.07 m² of algae over 40,593.57 m² snow). Snow area decreased markedly between dates. - Index–biology correlations: Across both dates, optimized index vs ln(cell concentration) was nonsignificant (F1,17=3.232, P=0.09, r²=0.16). July 2 alone: r²=0.30 (F1,7=3.038, P=0.13). July 30 alone: significant strong relationship r²=0.86 (F1,8=50.23, P<0.001), ln(y)=5.0575x+4.4519. Index vs astaxanthin:chlorophyll a ratio significant but modest (F1,17=5.30, P=0.034, r²=0.24). - Spectral behavior: SNICAR simulations showed decreasing visible albedo (350–750 nm) with increasing cell concentration; average spectral albedo (205–750 nm) increased from 0.56 ± 0.17 (July 2) to 0.82 ± 0.09 (July 30) as blooms waned. UAV reflectance showed algae absorbing more at 400–600 nm and reflecting more at 600–900 nm relative to clean snow; July 2 separation between classes was larger than July 30. - Radiative forcing and melt: Average IRF 236.56 ± 79.55 W m−2 (max 359.95) on July 2; 88.86 ± 40.96 W m−2 (max 156.9) on July 30. Estimated daily RF energy: 18,290 ± 6,150 MJ (July 2) and 2,671 ± 1,231 MJ (July 30), corresponding to 91 ± 31 m³ day−1 and 13 ± 6 m³ day−1 snowmelt due to algae. Integrated over July (linear scaling), total algae-driven melt ~1,508 ± 536 m³ in the 0.1 km² basin. DEM differencing indicated ~8,000,000 m³ total snowmelt; algae accounted for ~0.02% despite covering ~1% of snow. - Comparative context: July 2 IRF exceeded prior reports for red algae in Antarctica (~88 W m−2 max 186) and Alaska (~22 W m−2 max 88), comparable to or higher than dust IRF in other mountain regions.

The findings demonstrate that snow algae at mid-latitudes can impose substantial radiative forcing and measurable melt enhancement, particularly during peak bloom conditions, addressing the study’s objective to quantify biological contributions to snow darkening. UAV-based multispectral mapping, validated by coincident cell and pigment measurements, effectively detected algae presence, with PCA leveraging all bands and an optimized red/green index exploiting known pigment absorption features. The strong dependence of detection performance and index–abundance relationships on bloom state and seasonal progression underscores the need to consider temporal bloom dynamics and snowpack age in remote sensing algorithms. The large IRF observed relative to higher-latitude studies suggests that mid-latitude conditions (higher insolation, longer melt seasons) may enhance algae impacts on melt. Modeled and measured spectra were broadly consistent in shape, supporting use of SNICAR for biological darkening scenarios, though absolute reflectance differences indicate parameterization refinements are needed for snow algae pigments. The mapping-to-modeling workflow provides a pathway to scale with satellite platforms (e.g., Sentinel-2, Landsat, upcoming imaging spectroscopy missions) and to integrate biological darkening into melt and watershed models, potentially improving predictions of snowmelt timing and volume.

This study demonstrates that UAV-mounted multispectral sensors can reliably detect and map mid-latitude snow algae, and that the optimal remote sensing approach varies with bloom state. A PCA-based method and an optimized red/green index both distinguished algae from clean snow during peak bloom, with reduced separability later in the season as snowpack thinned and impurities accumulated. Ground-validated mapping combined with SNICAR modeling quantified high instantaneous radiative forcing (up to ~360 W m−2) and estimated that algae contributed ~1,508 ± 536 m³ of melt over July in a 0.1 km² basin. These results indicate that snow algae can be a non-negligible driver of melt in the North Cascades and likely other mid-latitude mountain ranges. Incorporating snow algae into water resource and climate models is warranted. Future work should: expand temporal sampling to capture bloom evolution; apply hyperspectral sensors for quantitative concentration retrievals; co-develop indices with numerical bloom models; and scale validated approaches to satellite platforms to assess watershed-scale impacts.

- Temporal coverage: Only two UAV surveys during the bloom (July 2 and July 30) limit characterization of temporal dynamics; linear interpolation was assumed to integrate July melt. - Algorithm dependence on bloom state: Both PCA and red/green indices were less effective post-peak; detection and concentration prediction varied with season. - Spectral confusion: Debris-covered snow and other impurities (dust, black carbon) were not quantified and can confound indices; “Other” class aggregated non-algal impurities. - Spatial masking: Avalanche runout and lake areas were excluded; photogrammetry performs poorly over water; steep, debris-laden terrain complicates detection. - Training data requirement: PCA thresholds relied on coincident ground truth; generalization without training is challenging. - Sensor constraints: MAPIR NIR camera lacked spectral resolution to discriminate algae; even with MicaSense, late-season separability was reduced; hyperspectral data may be needed for quantitative retrievals. - Modeling assumptions: SNICAR pigment inputs were scaled to fit allowable ranges (tuned for glacier algae), potentially biasing absolute albedo; fixed parameters (grain size/shape, density 600 kg m−3, dust set to zero) and external downward flux spectrum were assumed; melt conversion used a single latent heat and density value. - Biological detail: Species composition and its influence on pigment spectra were not assessed; potential species shifts could affect optical properties. - Vegetation influence: Later survey imagery had more vegetation, potentially influencing PCA ordination dominated by non-snow variance. - Positional/photometric uncertainties: Although GVP errors were within pixel size, residual geolocation and reflectance calibration uncertainties persist.