Spring Snow Depth Over Pan-Arctic Sea Ice Derived from AMSR2 Using Machine Learning Methods (2013–2023)

Description This data set provides daily estimates of spring snow depth over pan-Arctic sea ice from 1 March to 30 April for the period 2013–2023. The retrievals are produced at a spatial resolution of 25 km × 25 km using brightness temperature observations from the Advanced Microwave Scanning Radiometer-2 (AMSR2) onboard the GCOM-W satellite. Snow depth retrievals are generated using a multi-model machine learning framework that integrates three complementary models: K-Nearest Neighbors (KNN), Extremely Randomized Trees (Extra Trees), and TabNet. The merged product, AMSR2_Merged, is derived through a performance-weighted fusion scheme constrained by OIB-MEDIAN airborne snow depth observations from 2013–2015. Snow depth values are provided only for sea ice-covered regions with sea ice concentration greater than 50%. The dataset includes daily snow depth estimates from the merged model and the three individual ML models, as well as a total uncertainty estimate for AMSR2_Merged. The uncertainty field, AMSR2_Merged_Unc, is derived through Gaussian error propagation by combining multiple uncertainty sources, including AMSR2 brightness-temperature noise, IceBird training-reference uncertainty, spatial scale-dependent uncertainty, inter-model spread uncertainty, and weighting uncertainty in the multi-model fusion framework. In addition to the NetCDF data files, spatial maps and animation videos of the snow depth estimates are also provided to facilitate visualization and interpretation of the daily and seasonal evolution of Arctic snow depth. An early melt-onset flag is also included to support cautious use of the snow depth retrievals under melt-affected conditions. This flag is based on the passive microwave melt-onset record of Markus et al. (2009), which identifies the first day of early melt onset from satellite passive microwave observations. A value of 999 indicates that early melt onset has already occurred for the corresponding grid cell and date, while NaN indicates that early melt onset has not occurred or that no valid melt-onset information is available. For any inquiries, please contact: [e-mail] . Data Files Filename format: YYYYMMDD.ncwhere YYYYMMDD represents the date of the daily snow depth retrieval. Each NetCDF file contains the following variables: lonLongitude on the NSIDC Sea Ice Polar Stereographic North grid.Unit: degrees_east latLatitude on the NSIDC Sea Ice Polar Stereographic North grid.Unit: degrees_north AMSR2_MergedMerged snow depth estimate from the performance-weighted fusion of KNN, Extra Trees, and TabNet models.Unit: centimeter AMSR2_Merged_UncTotal uncertainty of AMSR2_Merged derived through Gaussian error propagation from multiple uncertainty sources, including AMSR2 brightness-temperature noise, IceBird training-reference uncertainty, spatial scale-dependent uncertainty, inter-model spread uncertainty, and weighting uncertainty in multi-model fusion.Unit: centimeter AMSR2_KNNSnow depth estimate derived from the KNN model.Unit: centimeter AMSR2_ExtraTreesSnow depth estimate derived from the Extra Trees model.Unit: centimeter AMSR2_TabNetSnow depth estimate derived from the TabNet model.Unit: centimeter Early_Melt_Onset_FlagEarly melt-onset occurrence flag based on the passive microwave melt-onset record of Markus et al. (2009).A value of 999 indicates that early melt onset has already occurred; NaN indicates that early melt onset has not occurred or that no valid melt-onset information is available.Unit: 1 Platform(s): GCOM-W satellite Sensor(s): Advanced Microwave Scanning Radiometer 2 (AMSR2) Data Format(s): NetCDF Temporal Coverage: March 1, 2013 to April 30, 2023 Temporal Resolution: Daily Spatial Resolution: 25 km × 25 km Spatial Reference System(s): NSIDC Sea Ice Polar Stereographic North (EPSG:3411) Spatial Coverage: North: 90°N, South: 0°N, East: 180°E, West: 180°W Example of Data Reading Using MATLAB >> ncdisp('20130301.nc') Source: 20130301.nc Format: netcdf4_classic Global Attributes: Title = 'Daily spring snow depth estimates over Arctic sea ice from AMSR2 using a multi-model machine learning framework' Description = 'This data set provides daily estimates of spring snow depth over pan-Arctic sea ice from 1 March to 30 April for 2013-2023 at a spatial resolution of 25 km x 25 km. Snow depth retrievals are derived from Advanced Microwave Scanning Radiometer-2 (AMSR2) brightness temperature observations using a multi-model machine learning framework that integrates K-Nearest Neighbors (KNN), Extremely Randomized Trees (Extra Trees), and TabNet. The merged product, AMSR2_Merged, is generated using a performance-weighted fusion scheme constrained by OIB-MEDIAN airborne snow depth observations from 2013-2015. The data set also includes snow depth estimates from the three individual ML models, a total uncertainty estimate for AMSR2_Merged derived through Gaussian error propagation from multiple uncertainty sources, and an early melt-onset flag based on the passive microwave melt-onset record of Markus et al. (2009).' Spatial_resolution = '25 km x 25 km' Temporal_resolution = 'Daily' Temporal_coverage = '1 March to 30 April, 2013-2023' Projection = 'NSIDC Sea Ice Polar Stereographic North, EPSG:3411' References = 'Markus, T., Stroeve, J. C., and Miller, J. (2009), Recent changes in Arctic sea ice melt onset, freezeup, and melt season length, Journal of Geophysical Research, 114, C12024, https://doi.org/10.1029/2009JC005436.' Creation_date = '2026-04-28 11:46:09' Dimensions: x = 304 y = 448 Variables: lon Size: 304x448 Dimensions: x,y Datatype: single Attributes: Long_name = 'Longitude on the NSIDC Sea Ice Polar Stereographic North grid' Units = 'degrees_east' lat Size: 304x448 Dimensions: x,y Datatype: single Attributes: Long_name = 'Latitude on the NSIDC Sea Ice Polar Stereographic North grid' Units = 'degrees_north' AMSR2_Merged Size: 304x448 Dimensions: x,y Datatype: single Attributes: _FillValue = NaN Long_name = 'Merged snow depth estimate from the performance-weighted fusion of KNN, Extra Trees, and TabNet models' Units = 'centimeter' Description = 'Daily merged snow depth estimate over Arctic sea ice derived from AMSR2 brightness temperatures using a performance-weighted multi-model ML fusion framework.' AMSR2_Merged_Unc Size: 304x448 Dimensions: x,y Datatype: single Attributes: _FillValue = NaN Long_name = 'Total uncertainty of the merged snow depth estimate derived through Gaussian error propagation' Units = 'centimeter' Description = 'Total uncertainty of AMSR2_Merged estimated by Gaussian error propagation, including contributions from AMSR2 brightness-temperature noise, IceBird training-reference uncertainty, spatial scale-dependent uncertainty, inter-model spread uncertainty, and weighting uncertainty in multi-model fusion.' AMSR2_KNN Size: 304x448 Dimensions: x,y Datatype: single Attributes: _FillValue = NaN Long_name = 'Snow depth estimate derived from the KNN model' Units = 'centimeter' Description = 'Daily snow depth estimate over Arctic sea ice derived from AMSR2 brightness temperatures using the KNN model.' AMSR2_ExtraTrees Size: 304x448 Dimensions: x,y Datatype: single Attributes: _FillValue = NaN Long_name = 'Snow depth estimate derived from the Extra Trees model' Units = 'centimeter' Description = 'Daily snow depth estimate over Arctic sea ice derived from AMSR2 brightness temperatures using the Extra Trees model.' AMSR2_TabNet Size: 304x448 Dimensions: x,y Datatype: single Attributes: _FillValue = NaN Long_name = 'Snow depth estimate derived from the TabNet model' Units = 'centimeter' Description = 'Daily snow depth estimate over Arctic sea ice derived from AMSR2 brightness temperatures using the TabNet model.' Early_Melt_Onset_Flag Size: 304x448 Dimensions: x,y Datatype: single Attributes: _FillValue = NaN Long_name = 'Early melt-onset occurrence flag' Units = '1' Flag_values = '999, NaN' Flag_meanings = 'early_melt_onset_occurred no_early_melt_onset_or_missing' Description = 'Flag indicating whether early melt onset has occurred at each grid cell for the corresponding date, based on the passive microwave melt-onset record of Markus et al. (2009). A value of 999 indicates that early melt onset has already occurred; NaN indicates that early melt onset has not occurred or no valid melt-onset information is available. This flag is intended to support cautious use of snow depth retrievals under melt-affected conditions.'>>