Melt Season Arctic Sea Ice Type Separability Using Fully and Compact Polarimetric C- and L-Band Synthetic Aperture Radar

Abstract

Sea ice mapping using Synthetic Aperture Radar (SAR) in the melt season poses challenges, due to wet snow and melt ponds complicating sea ice type separability. To address this, we analyzed fully polarimetric (FP) and simulated compact polarimetric (CP) C- (RADARSAT-2) and L- (ALOS-2 PALSAR-2) band SAR, in the 2018 melt season in the Canadian Arctic Archipelago, for stage-wise separation of first year ice (FYI) and multiyear ice (MYI). SAR scenes at both near- (19.1–28.3°) and far- (35.8–42.1°) range incidence angles and coincident high-resolution optical scenes were used to assess the impact of surface melt ponds on separability within a landfast ice zone of diverse ice thickness. C-band provided better separability between FYI and MYI during pond onset, while L-band was superior during pond drainage due to MYI volumetric scattering. CP parameters matched FP performance across the melt season. HH and HV, commonly offered in ScanSAR mode for both frequencies, presented good separability during pond onset and drainage. Using both C-band and L-band SAR along with constraining incidence angle ranges, enhances sea ice type identification and separability. Our results can support ice type classification and seasonal stage detection for climate studies and enhance existing frameworks for ice motion vector retrievals.

Résumé

La cartographie de la glace de mer à l’aide d’un radar à synthèse d’ouverture (SAR) pendant la saison de fonte pose des défis, en raison de la neige mouillée et des étangs de fonte qui compliquent la séparabilité des types de glace de mer. Pour résoudre ce problème, nous avons analysé les bandes complètement polarimétriques et des bandes compactes simulées (CP) de RADARSAT-2 en bande C et d’ALOS-2 PALSAR-2 en bande L au cours de la saison de fonte 2018 dans l’archipel arctique canadien, pour la séparation par étape de la glace de première année (FYI) et de la glace pluriannuelle (MYI). Des scènes SAR à des angles d’incidence proches (19,1-28,3o) et lointains (35,8-42,1o) et des scènes optiques à haute résolution coïncidentes ont été utilisées pour évaluer l’impact des étangs de fonte de surface sur la séparabilité dans une zone de glace côtière de diverses épaisseurs. La bande C offre une meilleure séparabilité entre FYI et MYI pendant l’apparition de l’étang, tandis que la bande L est plus efficace pendant le drainage de l’étang en raison de la diffusion volumétrique de la glace MYI. Les paramètres CP obtiennent les mêmes performances que les bandes FP tout au long de la saison de fonte. HH et HV, couramment offerts en mode ScanSAR pour les deux fréquences, présentaient une bonne séparabilité pendant l’apparition des étangs et le drainage. L’utilisation de SAR dans les deux fréquences ainsi que des plages d’angles d’incidence contraignantes améliore l’identification et la séparabilité des types de glace de mer. Nos résultats peuvent soutenir la classification des types de glace et la détection des stades saisonniers pour les études climatiques et améliorer les outils existants pour la récupération des vecteurs de mouvement de la glace.

Introduction

Observations of the last four decades indicate that Arctic sea ice is declining in age, thickness, and extent (Maslanik et al. 2007; Meier et al. 2014). This new ice regime represents a shift from thicker multiyear ice (MYI) to thinner, seasonally decaying, first year ice (FYI) (Kwok and Rothrock 2009; Laxon et al. 2013; Sumata et al. 2023). Particularly, Arctic sea ice extent decline has been most pronounced in the summer, with earlier and longer melt seasons accelerating sea ice loss (Markus, Stroeve, and Miller 2009; Stroeve and Notz 2015).

In the melt season, the sea ice conditions exhibit considerable heterogeneity, due to the melting of snow and the formation of melt ponds. Diverse patterns in melt pond formation and evolution manifest depending on the ice type. Generally, the smoother, undeformed FYI is characterized by higher melt pond fraction (MPF) compared to the rougher and undulated MYI (Eicken et al. 2004; Scharien and Yackel 2005; Polashenski, Perovich, and Courville 2012). The presence of melt ponds exerts a significant influence on ice dynamics, reduces sea ice albedo, intensifying heat absorption, and accelerating ice melting (Perovich et al. 2002). Hence, studying the surface conditions in the melt season, especially for distinguishing FYI from MYI, can improve sea ice forecasting and mapping.

Field observations during the melt season are often confined to specific regions, while optical remote sensing is limited by cloud coverage (Perovich et al. 2002; Webster et al. 2022). To address these challenges, active microwave sensors, particularly Synthetic Aperture Radar (SAR), offer a compelling alternative. SAR is unaffected by light and cloud cover and allows for high spatiotemporal coverage (Scharien et al. 2014; Howell et al. 2020). Additionally, the potential of using multi-frequency and multi-polarization SAR data for sea ice retrievals in challenging sea ice regimes, such as the advanced melt and freeze-up, has been demonstrated (Onstott and Gogineni 1985; Scheuchl, Hajnsek, and Cumming 2002; Dierking and Busche 2006; Casey et al. 2016; Singha et al. 2018).

Traditionally, higher frequencies like C-band, have been used for their adeptness in distinguishing between thicker, older ice and thinner ice types. For C-band SAR the measured radar backscatter from snow-covered sea ice is a function of snow grain size, brine volume, and roughness at the snow ice interface, hence MYI and FYI backscatter signatures are distinct in the winter. As temperatures rise during the melt season, liquid water in the snowpack and the upper ice layer reduces the microwave signal penetration and ice type differentiation. In contrast, lower SAR frequencies, such as L-band, allow for greater signal penetration depth through wet snow, thus enabling scattering responses from the upper ice layer (Dierking and Busche 2006). Despite previous research recognizing the value of multifrequency polarimetric parameters for sea ice mapping and classification (Scheuchl, Hajnsek, and Cumming 2002; Dierking and Dall 2007; Gill and Yackel 2012; Dabboor et al. 2017; Singha, Johansson, and Doulgeris 2021), the exploration of combined C- and L-band SAR sea ice signatures during the melt season is limited to dual-polarization acquisitions (Casey et al. 2016; Mahmud et al. 2020).

Backscatter for common single and dual-polarization SAR channels, is highly variable during melt conditions due to surface heterogeneity, leading to limited reliable ice observations (Livingstone et al. 1987; Yackel et al. 2007). The presence of liquid water alters dielectric permittivity, and fluctuations in wind-wave induced surface roughness from melt ponds on open water often result in overlapping HH and HV signatures from FYI and MYI (Barber, Hanesiak, and Yackel 2001). A promising alternative uses fully polarimetric (FP) SAR, which employs four linear backscatter channels (HH + HV + VH + HH) and relative phase measurements to comprehensively characterize a target’s polarimetric response. This mode offers a range of sea ice geophysical information, showcasing its potential for ice type separability and classification during the melt season (Fors et al. 2016; Singha, Johansson, and Doulgeris 2021; He et al. 2022). The wider application of FP mode data for both frequencies to understanding sea ice properties during melt has been largely limited by narrow (≤50km) achievable swath widths compared to single or dual-pol modes (Geldsetzer et al. 2015). Additionally, outside of open-access missions such as Sentinel-1 (dual-pol), data from both C- and L-band SAR missions are limited by access restraints.

Conversely, compact polarimetry (CP) offers a good tradeoff between the scattering information provided by FP, and capability of acquiring data over wide swaths (Charbonneau et al. 2010; Geldsetzer et al. 2015; Espeseth, Brekke, and Johansson 2017). In CP, a circularly polarized wave is transmitted, and coherent linear polarized waves are received, providing relative phase information for polarimetric decompositions akin to FP (Raney 2006; Dubois-Fernandez et al. 2008; Cloude 2012). While several studies have examined the CP parameter potential for sea ice mapping and classification during winter and freeze-up periods (Li and Perrie 2016; Espeseth, Brekke, and Johansson 2017; Singha and Ressel 2017), a comprehensive analysis of CP parameter relationships across all melt stages is lacking (Dabboor and Geldsetzer 2014; Geldsetzer et al. 2015).

Geldsetzer et al. (2015) and Nasonova et al. (2018) used simulated CP data for C-band in the summer, showing that parameters sensitive to surface scattering and depolarization effectively separated FYI from MYI samples. Components of the mχ decomposition and the Stokes vectors exhibited good separation in both NR and FR incidence angles. NR acquisitions presented enhanced contrast owing to their sensitivity to Bragg scattering from wind-roughened melt ponds, accentuating differences in sea ice type-related melt pond fraction. Nonetheless, a detailed stage-wise analysis of ice type separability in the advanced melt remains unexplored. Furthermore, the assessment of simulated L-band CP data within the context sea ice variability in the melt season is limited. Future missions like Copernicus ROSE-L, NASA-ISRO SAR (NISAR), and ALOS-4 PALSAR-3 from JAXA, offer the potential for L-band CP data exploration, complementing existing C-band CP data with high spatial resolution, in both FP and CP modes from the Radarsat Constellation Mission (RCM).

For this work, we use FP and simulated CP polarimetric data from C- and L-band frequencies to study early and advanced melt stages and address a knowledge gap in understanding FYI and MYI separability in the summer. Our analysis covers the 2018 melt season in the Canadian Arctic Archipelago (CAA) and focuses on three main research questions:

1. Which frequency, C- or L-band, offers enhanced separability between FYI and MYI in advanced melt?

2. How do CP parameters compare to FP, for FYI and MYI type separability at each melt stage?

3. What is the role of melt ponds in the observed backscatter variability?

To address these questions, we investigate the relative scattering mechanisms of FYI and MYI during late winter, melt onset and advanced melt (defined below), and explore the role of surface melt ponds as a driving factor of observed changes in the advanced melt. Additionally, we study the impact of radar parameters, such as incidence angle and frequency, on ice type separability and scattering mechanisms during the melt season. Section “Seasonal stages, study site and data” describes the seasonally-evolving sea ice surface conditions, defining each seasonal stage and substage addressed in this study. We then provide details about the study site, C- and L-band SAR datasets, high-resolution satellite optical data of melt pond covered sea ice, and ancillary datasets. Data processing and the framework for analysis is provided in Section “Methods”, followed by results in Section “Results”. Summary and conclusions are given in Section “Summary and Conclusions”.

Seasonal stages, study site and data

Seasonal stages

In this study, the period pre-ceding melt is defined as late winter (LW) and two distinct stages comprising melt conditions are melt onset (MO) and advanced melt. MO on sea ice has been previously defined as the time when the surface temperature is above −1 °C for more than five consecutive days. During MO, the transition from the pendular and funicular regimes leads to an increasing snow wetness and snow metamorphism (Tiuri et al. 1984). The presence of liquid water within the snow and ice volume, alters the dielectric properties of snow and ice, and results in the convergence of previously contrasting single-polarization C-band backscatter from FYI and MYI. The advanced melt season, defined as the period after MO when the sea ice volume is isothermal and surface melt ponds form, is characterized by variable combinations of melting snow, bare ice, and melt pond covered ice. As mentioned in Section “Introduction”, this leads to C-band backscatter ambiguity (Barber, Hanesiak, and Yackel 2001).

Advanced melt is further divided into three substages, (i) pond onset (PO), defined as the initial period of melt pond formation, where the pond formation and evolution is driven by melting snow, and the relatively rapid melt water influx, combined with variations in surface topography between FYI and MYI can result in a large contrast in MPF (i.e., widespread ponding on FYI and topographically limited ponding on MYI); (ii) pond evolution (PE), after the meteoric snow cover has ablated and sea ice is composed of melt pond water and bare ice; melt pond fraction is driven by competing melt water input and drainage sources, particularly on FYI where a network of interconnected melt ponds is formed, and enhancing drainage occurs into cracks and flaws, and (iii) pond drainage (PD), where the FYI has become porous enough that melt water is vertically transported through the ice volume through drainage channels to the ocean (Weeks and Ackley 1986; Eicken et al. 2002; Polashenski, Perovich, and Courville 2012). During PD, the MPF on FYI decreases rapidly when the melt pond surfaces are above the ice freeboard line, and increases again when they reach the sea level. Importantly, PD is a pre-cursor to FYI break up as it occurs when the ice is isothermal and structurally weak (Eicken et al. 2002, 2004). FYI and MYI differ mainly in permeability, with the less permeable MYI retaining more melt ponds, in some cases into the subsequent freeze-up period (Eicken et al. 2002).

Overall, these substages provide an intuitive framework for evaluating key C- and L-band scattering differences between FYI and MYI, and backscatter changes in the context of seasonally evolving sea ice properties. It is acknowledged that these sub-stages are not discrete, e.g., a greater snow depth on MYI compared to FYI will result in a later transition from PO to PE for that ice type.

Study site

Data were acquired from M’Clintock Channel in the southwestern CAA in 2018 (Figure 1a), as part of an effort to characterize the seasonal evolution of sea ice physical and electromagnetic properties, including evaluating the utility of C- and L-band SAR for sea ice geophysical information retrievals (Scharien et al. 2018). This location comprises a combination of thermodynamically grown FYI, MYI and deformed ice, all of which become land-fast from winter until break up, normally in July (Canadian Ice Service 2011). Sea ice in this region is therefore characterized by a broad spectrum of roughness and thickness during the late winter and is ideal for establishing seasonal linkages with SAR backscatter since the same ice can be tracked from pre-melt through advanced melt without requiring ice motion tracking. Based on the histogram from an airborne electromagnetic (AEM) survey on 8 May 2018 (Figure 1c), the snow plus ice thickness ranges from approximately 1 to 8 m for the complete survey track, while the track portions overlapping with SAR datasets range from approximately 1.7 to 7 m. The AEM survey configuration is detailed in Section “Ancillary datasets”.

Figure 1. (a) Study area map showing the southwestern Canadian Arctic Archipelago. Areal extents of C- and L-band SAR images and optical scenes are shown along with an airborne electromagnetic (AEM) sea ice thickness track. (b) A late winter (LW) SAR image showing the ice thickness data and ROIs used for aggregating scattering data across all seasons. (c) Histograms of ice thickness derived from the complete survey track (orange) and the segment that coincides with SAR acquisitions (blue).

SAR and optical datasets

Details of the SAR and high-resolution optical image datasets are provided in Table 1. We analyzed 10 C-band (5.4 GHz) RADARSAT-2 (RS2), and 3 L-band (1.27 GHz) ALOS-2/PALSAR-2 (PS2) SAR scenes (CSA, 2011; JAXA, 2008). A Sentinel-1 mosaic was used only for visualization in Figure 1b, to represent LW sea ice conditions in the broader study area. The PS2 scenes are coincident to RS2 scenes (Figure 1a) at near-range (NR) (19.1-28.3°) and far-range (FR) (35.8-42.1°) incidence angles within ±19-hour time difference, enabling comprehensive substage analysis for advanced melt. Each scene was collected in single look complex (SLC) format, except for PS2-1 which was ground range detected (GRD).

Table 1. SAR and optical image characteristics.

ID	Sensor	Date (yyyy.mm.dd)	Time (UTC)	Stage	IA range (°)	IA class	Air temp. (^oC)	Wind speed (m/s)
Year 2018
PS2-1	ALOS-2 /PALSAR-2	2018.01.21	05:33	W	28 – 33	–	−34.7	6.3
RS2-1	RADARSAT-2	2018.05.20	14.10	LW	22.6 – 26.2	NR	−5.5	6.9
RS2-2	RADARSAT-2	2018.05.21	13.24	LW	38.7 – 41.3	FR	−5.1	4.4
S1	Sentinel 1-A	2018.05.28	23:59	LW	29.1 – 46	–	−4.1	6.1
RS2-3	RADARSAT-2	2018.06.10	23.25	MO	19.1 – 22.6	NR	0.8	4.7
GE1-1	GeoEye-1	2018.06.10	18:37	MO	16.1	ON	0.9	4.7
RS2-4	RADARSAT-2	2018.06.10	13.24	MO	35.8 – 38.6	FR	2.8	4.4
WV2-1	Worldview-2	2018.06.19	19:12	PO	2.3	ON	1.6	3.6
RS2-5	RADARSAT-2	2018.06.20	00.46	PO	36.9 – 42.1	FR	5.5	6.4
PS2-2	ALOS-2 /PALSAR-2	2018.06.20	09:34	PO	37 – 40	FR	5.2	8.9
RS2-6	RADARSAT-2	2018.06.20	14.01	PO	25 – 28.3	NR	6.6	8.9
WV3-1	Worldview-3	2018.06.26	19:12	PE	7.7	ON	8.7	2.5
RS2-7	RADARSAT-2	2018.06.27	23.49	PE	22.6 – 26	NR	3.5	1.4
RS2-8	RADARSAT-2	2018.06.27	00.28	PE	36.9 – 42.1	FR	5.4	4.4
GE1-2	GeoEye-1	2018.06.29	18:33	PE	16.2	ON	6.3	4.4
PS2-3	ALOS-2 /PALSAR-2	2018.07.14	11:14	PD	37 – 40	FR	7.8	8.9
RS2-9	RADARSAT-2	2018.07.14	23.43	PD	23.8–27	NR	3.4	3.3
RS2-10	RADARSAT-2	2018.07.14	00.12	PD	39 – 42	FR	2.5	5.3
WV3-2	Worldview-3	2018.07.15	18:45	PD	5.6	ON	17.6	2.8
WV3-3	Worldview-3	2018.07.15	19:12	PD	6.3	ON	15.7	6.7

IDs denote sensor type: PALSAR-2 (PS2); RADARSAT-2 (RS2); GeoEye-1 (GE-1); and Worldview-2/3 (WV2/3), with each acquisition from a sensor numbered sequentially. Seasonal stages are described in the text. Incidence angle (IA) ranges correspond to SAR swaths, with average off-nadir angles substituted for each optical scene. IA classes are near-range (NR), far-range (FR), or off-nadir (ON) for optical scenes. ERA5 hourly air temperature (at 2 m) and wind speed (at 10 m) corresponding to acquisition times are provided.

A total of 6 high-resolution, optical (VIS-NIR) images were acquired during advanced melt conditions from GeoEye-1 and WorldView-2 and −3 sensors. The optical imagery were at 2 m spatial resolution, enhanced to 0.5 m after panchromatic (pan)-sharpening. This enhancement enabled visual assessment of melting conditions, including surface flooding and drainage, determination of each advanced melt substage, and quantification of MPF.

Ancillary datasets

We collected ancillary datasets to assess the geophysical state of the sea ice and determine seasonal stages. A backscatter timeseries of FYI and MYI was extracted using the Advanced Scatterometer (ASCAT) Scatterometer Image Reconstruction (SIR) data product (Early and Long 2001). ASCAT operates at C-band (5.3 GHz) and employs vertical polarization (VV). The SIR data product consists of 2-day composites, with a pixel spacing of 4.5 km and normalization to a far-range incidence angle of 40°. Similar time series have previously been employed for detecting MO and attributing seasonal stages based on FYI and MYI backscatter variations (Mortin et al. 2014; Casey et al. 2016; Howell et al. 2018). Additionally, daily corrected reflectance at 250 m spatial resolution from the Moderate Resolution Imaging Spectroradiometer (MODIS) was used to support identifying advanced melt substages (Figure S1).

The AEM survey comprised spatially coincident airborne laser scanner (ALS) and electromagnetic (EM) ice thickness data collected along the flight line shown in Figure 1a. Only the EM ice thickness data are used in this study. The EM ice thickness instrument derives height above the ice-water interface by inducing an EM field that exploits the differences of electrical conductivity between sea ice and water (Haas et al. 2009). These instruments do not allow for differentiating between snow and sea ice thickness, thus the difference between those two measurements is referred as the total thickness, i.e., snow plus ice thickness (Haas and Howell 2015). After processing, thickness data are provided at approximately 6 m spacing along the survey track, with variations in spacing dependent on the speed of the aircraft. The AEM survey was conducted in the LW stage for retrieving the maximum sea ice thickness in the area while minimizing the atmospheric influence on surface conditions (Haas et al. 2009).

Hourly mean surface air temperature (2 m) and wind speed data (10 m) were derived from the 0.5° x 0.5° gridded, bias-corrected dataset from the fifth generation of the European Center for Medium-Range Weather Forecasts (ECMWF) atmospheric reanalysis (ERA5). The ERA5 dataset provides a reconstruction of near-surface meteorological variables, and it is used as a meteorological forcing dataset for surface modeling. Western Arctic regional ice analysis charts, published by the Canadian Ice Service (CIS), aided region of interest (ROI) selection. The charts are produced weekly from the integration of data from a variety of sources including surface observations, aerial, and satellite reconnaissance (Canadian Ice Service 2011).

Methods

Figure 2 illustrates the main data processing and analysis workflow, elaborated upon in the following sub-sections.

Figure 2. Workflow for SAR polarimetric parameter retrieval and optical scene processing to melt pond fraction. The two processing chains were executed concurrently to facilitate comparison between backscatter parameters and melt pond fraction.

C- and L-band SAR image processing

The SLC RS2 and PS2 scenes were calibrated to σ⁰, speckle filtered using a polarimetric 7 × 7 Boxcar filter, and map-projected to the WGS 1984/UTM projection using Bilinear resampling. Additionally, Faraday rotation correction was applied to the SLC PS2 scenes. Following pre-processing, the RS2 and PS2 images were simulated to right circular hybrid CP mode. The SAR scenes were simulated to CP mode, due to the unavailability of CP SAR data in the region prior to the launch of RCM.

From the FP data, 8 polarimetric parameters were retrieved, including polarimetric ratios. The total power (SPAN) is used to describe the backscatter variability for FYI and MYI per seasonal stage. From the simulated CP, we derived the Stokes vector components together with 4 child parameters describing the EM field. Additionally, the m_χ decomposition was selected for retrieving the relative scattering mechanisms for FYI and MYI at each seasonal stage. This choice was informed by its demonstrated ability to aid identification of changes in scattering processes. This method relies on two key parameters, m (degree of polarization) and χ (Poincaré ellipticity parameter) and delivers three channels corresponding to double bounce, volumetric, and surface scattering contributions per pixel (Raney et al. 2012). Importantly, the m_χ decomposition has demonstrated its efficacy in identifying ice types across winter and summer seasons (Geldsetzer et al. 2015; Nasonova et al. 2018). This comprehensive approach yielded a total of 24 polarimetric parameters from both RS2 and PS2 scenes, encompassing both FP and CP modes. Parameter description and definitions are listed in the Table S1 in the supplementary materials.

SAR polarimetric parameter analysis

Homogeneous ROIs of FP and CP parameters from FYI and MYI were used to evaluate ice type separability, relative scattering mechanisms, relationship to MPF during advanced melt, and parameter redundancy. In addition, the effect of incidence angle range was examined based on the NR and FR designations in Table 1. ROIs were selected using the HH and HV bands from W and LW RS2 and PS2 images, alongside reference ancillary data. We selected 30 FYI and 30 MYI ROIs, each measuring 20 by 20 pixels, with the deliberate intention of representing a spectrum of LW backscatter intensity levels, including varying levels of ice deformation (Figure 1b).

FYI and MYI sample separability was determined using the 2-sample Kolmogorov–Smirnov (KS) non-parametric test. This test measures the absolute maximum distance between the cumulative distribution functions of two samples, without assuming normally distributed data. The KS values range from 0 to 1, with 1 to indicating complete separability. Moreover, we calculated the p-value for each KS test to determine statistical significance and we report only significant relationships. Before conducting the KS test for each sample, we applied z-score criteria to exclude outliers, eliminating z-score values beyond three standard deviations from the mean.

For assessing FP and CP parameter redundancy, we employed the non-parametric Spearman’s correlation coefficient (ρ). Specifically, ρ was calculated based on the ROIs of each ice type and substage, and within a distinct incidence angle class (NR or FR). As a correlation criterion for parameter grouping, we defined ρ ≥ ±0.8. Beyond indicating redundancy, the correlation analysis provided insights into the grouping of parameters concerning ice type separability.

Melt pond fraction and SAR parameter comparison

High-resolution optical images from 2018 were classified to binary ice (or snow-covered ice) and melt pond pixels to: (i) retrieve areal MPF estimates for discrete regions of FYI and MYI; (ii) enhance our understanding of seasonal melt pond evolution; and (iii) assess the impact of MPF on C- and L-band backscatter. These images covered two regions in close north-south proximity and are named ROI-N, and ROI-S, respectively (Figure 1). Classification to MPF was done using a supervised Random Forest (RF) classifier applied to the pan-sharpened VIS-NIR bands with 0.5 m cell size obtained using the associated panchromatic band. The distinct spectral signatures of melt pond water compared to snow-covered or bare sea ice enabled overall classification accuracies >97%. Details about the image classification algorithm can be found in Scharien et al. (2014).

For analyzing MPF and SAR parameter relationships, the labeled FYI and MYI ROIs were used for aggregating SAR parameter statistics from images acquired within ± 20h of an optical scene in 2018. Additionally, a total of 30 ROIs representing pure melt pond and snow covered or bare ice over FYI and MYI types were selected from the PO stage and used to track the evolution of the pond area relative to bare ice as it subsequently evolved to become drained ice. This was done since it is expected that exposed ice that is flushed by a surface melt pond will have different physical and microwave scattering characteristics than the surrounding ice. The MPF for FYI and MYI was then calculated for each ROI, and date-wise summary MPF statistics for individual scenes and the ice types within, were derived. Pearson’s product moment correlation coefficient (r) between each polarimetric parameter and MPF was then examined to evaluate how melt ponds influence the ice type specific polarization properties of backscatter, to better understand separability potential.

Software usage

All the image pre- and post-processing explained in Section “C- and L-band SAR image processing” was done using scripts integrated into the Sentinel-1 toolbox (SNAP), version 8. The polarimetric parameters analysis described in Section “SAR polarimetric parameter analysis” was done using in house developed Python code utilizing Python version 3.8 (Python Software Foundation, https://www.python.org/). We performed melt pond fraction analysis stated in Section “Melt pond fraction and SAR parameter comparison” using QGIS (QGIS Geographic Information System, http://www.qgis.org) version 3.28.

Results

The RS-2 and PS-2 scenes are attributed to each seasonal stage and are then examined for backscatter variability. Using SPAN and the m_χ decomposition, we describe the observed backscatter differences between FYI and MYI and identify the dominated scattering mechanisms at each stage. Coincident MPF retrievals, enable a thorough comparison with polarimetric parameters, revealing the main drivers behind FYI-MYI backscatter changes. Each section findings, support and explain the observed FYI and MYI separability and parameter redundancy.

Melt season stage definition

In Figure 3, we present the ASCAT backscatter timeseries for each ice type. The data spans from mid-May, specifically Day of Year (DOY) 140, until DOY 200 (Figure 3b). MO occurs when the presence of liquid water in snow leads to a shift in MYI scattering mechanism from volumetric to surface, resulting in a large (∼6dB) decrease in backscatter (Casey et al. 2016). During this period, backscatter from FYI increases due to volumetric scattering from wet snow grains and high brine volume (Barber and Nghiem 1999). Consequently, the backscatter timeseries of FYI and MYI converge. In our dataset, MO was detected on DOY 160. After MO, daily mean surface air temperatures are closer to 0 °C (Figure 3c), and the snowpack transitions from pendular (low saturation) to funicular regime, where liquid water is present throughout the snow and the air is trapped in distinct bubbles in the pores (Tiuri et al. 1984). The advanced melt starts when melt ponds form on the surface. At this stage, the snow cover rapidly ablates, bare ice is exposed, and MPF evolves dynamically.

Figure 3. (a) ASCAT data and ROIs used for FYI and MYI time series backscatter coefficient σ⁰ extraction. (b) Backscatter time series and seasonal regimes denoted by green dashed lines (MO, PO, PD). (c) Time series ERA5 bias-corrected atmospheric reanalysis data of mean surface wind speed (at 10 m) and mean surface air temperature (at 2 m). Vertical, light blue lines represent the days of the RS2 and PS2 acquisitions.

To identify the advanced melt substages, we utilized optical images from high-resolution optical sensors and MODIS (Figure S1). The rapid and extensive flooding of relatively shallow, low-albedo ponds over smooth FYI, compared to deformed FYI and MYI, enables PO identification. Similarly, the rapid drainage of melt ponds from a FYI surface with freeboard above the sea-level, which occurs when a permeability threshold is crossed and ice-ocean connectivity is achieved, enables PD identification. PE thus represents the intervening stage. In the study area, PO is identified at DOY 166, and PD at DOY 186. The ice in the extended area started breaking around DOY 196. MPF for the FYI samples was high during PO, then moderate in PE and decreased to low in PD, reflecting surface drainage. MYI samples had moderate MPF during PO (∼50%), dropping in PE, and stabilizing at around 20% in PD. Reference surface air temperature and wind speed data are also shown in Figure 3c, and values coincident to SAR image acquisitions are provided in Table 1.

FYI and MYI scattering properties across the melt season

The HH backscatter variability for SAR C- and L-band subsets in the advanced melt is shown in Figure 4. Additionally, Figure 5 presents mean SPAN for FYI and MYI, as well as initially pure pond and bare ice ROIs from PO, for all stages. Here, the same samples are examined across the season and are compared to the scattering mechanisms per type, frequency, and incidence angle. Interestingly, in each substage SPAN in NR exceeds that in FR, with the most notable difference occurring in PO. This is due to the anticipated effect of surface flooding reducing the backscatter intensity at larger incidence angles.

Figure 4. Image subsets of optical RGB composites and C- and L-band HH backscatter variability during advanced melt season substages. The top panel shows the PO stage with: a) WV2-1; a_i) RS2-6; a_ii) RS2-5; and a_iii) PS2-2. The middle panel shows the PE stage with: b) WV3-1; b_i) RS2-7; b_ii) RS2-8 (there is no L-band scene). The bottom panel shows the PD stage with: c) WV3-2; c_i) RS2-9; c_ii) RS2-10; and c_iii) PS2-3.

Figure 5. Mean total power (span) for C- and L-band ROIs in late winter (LW), melt onset (MO), pond onset (PO), pond evolution (PE), and pond drainage (PD). The analysis encompasses NR and FR for C-band during all stages, and L-band FR during PO and PD. Additional melt pond water and bare ice samples in the advanced melt (see text for description). L-band scenes are indicated by an "L" suffix.

In LW, the C-band SAR scenes showcase the anticipated distinct contrast between FYI and MYI. Progressing to MO, this contrast weakens in C-band, with FYI SPAN increasing and MYI decreasing compared to LW, aligning with the ASCAT timeseries. The relative scattering mechanisms in Figure 6b, show reduced volumetric scattering for MYI in both NR and FR compared to LW, indicating wet snowpack.

Figure 6. Bar plots of relative scattering mechanisms from the m_χ decomposition, from C-band acquisitions in NR (left panel) and FR (right panel). Groups (a) to (e) correspond to late winter (LW), melt onset (MO), pond onset (PO), pond evolution (PE), and pond drainage (PD). Bare ice and melt ponds samples are shown in addition to FYI during stages PO to PD.

In the advanced melt, the C- and L- band SAR are 5 days after the detected PO in the MODIS timeseries, and the surface conditions are highly variable. In this stage, we notice that the SPAN of melt ponds is quite high (−4dB), similar to FYI SPAN in the NR. From Table 1, we observe that the C- and L-band scenes in PO, are acquired during moderate to high wind speeds affecting backscatter and SPAN. Specifically, a high wind speed (8.9 m/s) is observed for C-band in NR, and is particularly affecting the FYI backscatter, which in the NR is higher than MYI (Figure 4ai). At these wind speeds a strong positive relationship is expected between C-band backscatter and MPF (Yackel and Barber 2000). Wind roughened melt ponds result in higher Bragg scattering and SPAN for both ice types. FYI is characterized by higher MPF compared to MYI during PO, and the relatively smooth surface of FYI, compared to morphologically complex and fetch-limited MYI, promotes capillary wave development. These factors lead to a greater density of capillary waves in the SAR footprint for FYI and higher SPAN. In the FR scene from the same day, FYI SPAN is lower than MYI, and the melt pond SPAN is lower than bare ice. Despite the wind speed during this acquisition being sufficiently high to produce wind-waves (6.4 m/s), the large reduction in span is consistent with observations of C-band backscatter from wind-roughened ocean surfaces at the same wind speed and incidence angle (Snoeij et al. 1991). Thus, in the C-band and at NR, during PO, FYI SPAN is stronger than MYI due to rough surface scattering from melt ponds, whereas in FR the SPAN from FYI is weaker than MYI due to those same melt ponds. This is expected to affect the FYI-MYI separability especially in NR, allowing for greater difference between the two ice types.

In the L-band PO scene, greater SPAN contrast between FYI and MYI is observed than in C-band in the FR and under strong wind (8.9 m/s). Nevertheless, SPAN remains low for melt ponds and FYI, and moderate for bare ice and MYI, suggesting minimal impact from the wind. This is because L-band in FR incidence angles is less sensitive to moderate wind speeds (5 to 10 m/s) compared to C-band, thus the effect is less pronounced (Isoguchi and Shimada 2009). Moreover, Figure 4aiii depicts the L-band HH backscatter. Consistent with Arkett et al. (2008) and Casey et al. (2016), we find that regions of deformed ice are clearly distinguishable, and MYI floe boundaries are better delineated suggesting good FYI-MYI separability potential at this substage.

The m_χ decomposition provides useful insights into the relative contributions of individual scattering mechanisms to total backscattered energy: surface, volumetric and double bounce. Figures 6c and 7a present the relative scattering mechanisms during PO for C- and L-band, respectively, highlighting differences between NR and FR acquisitions. The additional bare ice and melt pond samples reveal their relative roles in the observed scattering mechanisms for FYI and MYI. For C-band, NR, we observe a clear dominance of surface scattering for all ice types. In the FR, all types present a shift where the surface scattering contribution is lower (∼60%) and the volumetric scattering contribution increases (∼40%). FYI and MYI scattering variability indicate that both bare ice and melt ponds contribute to their signatures. However, in secondary contributions, FYI is on par with melt ponds, both showing higher volumetric scattering compared to bare ice and MYI, in NR and FR. For the L-band scene, the FYI shows similar responses to both bare ice and melt pond signatures, while MYI shows equal contributions from surface and volumetric scattering.

Figure 7. Bar plots of relative scattering mechanisms from the m_χ decomposition, from L-band acquisitions in FR during (a) pond onset (PO) and (b) pond drainage (PD). Bare ice and melt ponds samples are shown in addition to FYI during stages PO to PD.

During PE, 9 days after PO, the C-band HH backscatter is reversed with FYI showing higher backscatter than MYI in both NR and FR (Figure 4bi and 4bii). At this stage, FYI is about 45% drained, compared to 20% drainage over MYI. With the relatively low wind speeds of 3.1 m/s for NR and 1.9 m/s for FR acquisitions, this data suggests an ice-dominated backscatter response in place of the pond-dominated response from PO. Higher SPAN from FYI compared to MYI, in both NR and FR, is likely due to stronger relative backscatter due to surface drainage from FYI (exposure of fresh, drained ice instead of melt pond) and the presence of an absorptive wet snow surface remaining on MYI. Figure 6d shows dominant surface scattering for both types in NR and FR, with higher contribution for FYI.

Progressing to PD, drainage is greater for FYI (∼70%), while MYI reaches a plateau at 25% (Figure 8a). C-band SPAN for initially melt ponds and bare ice in Figure 5 is fairly similar, indicating that they both represent drained ice. FYI is considerably drained compared to earlier substages, leaving drained melt pond areas that have lower C-band SPAN than bare ice in the NR, and higher SPAN than bare ice in the FR. In the NR, the FYI SPAN is lower than MYI, while in the FR, SPAN of FYI, MYI, and bare ice are very similar. For the L-band scene in FR, there is a larger disparity in SPAN between FYI and MYI, with the latter much greater. SPAN of drained melt ponds at L-band, expected to be a greater proportion of FYI compared to MYI, is lower than bare ice.

Figure 8. (a) MPF time series of FYI and MYI, retrieved from optical imagery. (b) RGB color composite image subsets demonstrating surface conditions in the corresponding dates during PO, (c) and (d) in PE, and (e) in PD substages of the advanced melt.

In Figure 6e, the contributions from volumetric scattering for FYI and drained melt ponds present a 10% decrease compared to PE, in both NR and FR. In Figure 7b, the L-band FR samples present a clear dominance of volumetric scattering for all types, as well as a small contribution of double bounce. This is probably due to the greater signal penetration depth at L-band compared to C-band, combined with the relative drainage of surface melt ponds compared to the PO stage.

Based on this stage-wise analysis, in NR, C-band surface scattering is dominant across all stages with 10% higher contribution for MYI compared to FYI in MO and PO. Volumetric scattering presented higher variability between the ice types with higher contribution for FYI (∼10 to 15% more than MYI) in MO and PO. When surface drainage occurs in PE and PD, these relationships reverse for both ice types. In the FR, C-band and L-band presented a 60–40% divided contribution of surface and volumetric scattering across all stages.

Melt pond fraction retrievals

In this section, the ice type specific seasonal MPF evolution is briefly characterized. Following that, relationships between ice type specific MPF and C-band and L-band co-pol and cross-pol ratios for PO and PD stages are examined. Co- and cross-pol ratios of landfast FYI previously presented strong relationships with MPF (Scharien et al. 2012; Scharien et al. 2014).

MPF was highly variable between FYI and MYI (Figure 8). Specifically, during PO, FYI presented higher MPF (95% ±0.05) than MYI (50% ±0.2). In PE, FYI MPF decreased to 70%, with notable variance (±0.26) on June 26th, and further reduced to 52% ±0.12 on June 29th. At the end of the season in PD, it dropped to 30% ± 0.16. Conversely, MYI MPF decreased to 23% ± 0.09 on June 26th, remained relatively similar on June 29th, and during PD on July 15th, it slightly reduced to 20% ± 0.13. It is likely that the surface conditions were highly dynamic, with MPF variability during the PE stage rapidly changing over short time intervals. In particular, this is expected for FYI, which is smooth and covered by relatively shallow melt ponds, resulting in a MPF that can rapidly change diurnally due to competing meltwater input and drainage processes.

In Figure 9a, the co-pol ratio shows significant positive linear relationships with MPF for both ice types and frequencies, in NR and FR in PO. In PD, these relationships are not significant expect for FYI at L-band which is slightly negative (p < 0.05). During PO, the strongest positive relationship with MPF (r = 0.83) is observed for MYI in C-band FR. Interestingly, there are differences in the FR C- and L-band co-pol ratios between FYI and MYI across the observed MPF ranges (∼1.25–3). We expect that melt ponds are influencing the FR co-pol ratios, as increasing liquid water in the SAR footprint, with increasing MPF, leads to more preferential backscatter of vertically compared to horizontally polarized energy. However, the separation between the FYI and MYI co-pol ratios at the FR during PO suggests that, in addition to the melt pond influence on the co-pol trend, the background ice property differences influence the polarization properties of backscattered energy at both C- and L-band as well. In the PD, the limited co-pol ratio values (∼1 to 1.30) and non-significant or weak linear relationships can be explained by the dampening effect of volumetric scattering on the polarization properties of backscatter caused by the drained ice in this substage.

Figure 9. (a) Scatter plots of the co-polarized ratio (VV/HH) and (b) the cross-polarized (HV/HH) ratio against MPF from PO and PD substages. Pearson’s product moment correlation coefficient (r) is given as a measure of linear association between the polarimetric ratio and MPF along with the significance level. The shaded area represents the size of the confidence interval for the regression estimate and here is set to 95%.

Relationships between the C- and L-band cross-pol ratios and MPF are shown for NR and FR scattering at C-band, and FR scattering at L-band in Figure 9b. During PO, we observe significant negative correlation between MYI and MPF at C-band in the NR (r= −0.68), while in the FR at both C-and L-band there are strong positive relationships for MYI (r = 0.72 and r = 0.70, respectively). The negative trend for MYI in the NR can be explained by a stronger relative increase in HH backscatter, compared to HV, due to sensitivity to increasing MPF, which in this case is associated with strong wind forcing and likely Bragg scattering during this acquisition. This effect is not seen for FYI at C-band in the NR, however. Our assumption here is that the effect of wind-wave roughening on pond surfaces is dominant over the narrow range of observed MPF on the smooth FYI surface regardless of the specific MPF. The opposite effect occurs in the FR during PO, at both C- and L-band, where the effect of melt pond wind-waves on backscatter is negligible, and HH decreases faster than HV with increasing MPF. This supports that melt ponds dominate scattering of both FYI and MYI during PO for both C- and L-band frequencies and NR and FR incidence angles. In PD, we see negative relationships for FYI at C-band (both NR and FR), while MYI presents no significant relationships. The strong cross-pol ratio associated with FYI at C-band likely indicates the emergence of volume scattering effects from the large portion of freshened ice that emerges from areas where melt ponds have drained. No significant relationships are found between MPF and the cross-pol ratio at L-band during this stage.

Overall, the results in Figure 9 highlight sensitivities to MPF, and potential separability, on the basis of polarization diversity at C- and L-band frequency. In terms of separability potential, it is noteworthy that L-band ratios for FYI and MYI do not overlap during PO, despite the influence of MPF.

C- and L-band ice type separability for FP and CP parameters

The polarimetric parameter values are calculated for each sea ice type for all scene pairs, and an equal number of samples within the 30 ROIs representing FYI and MYI are used as input to calculate the KS distance. The results are presented in Figures 10 and 11. KS separability values from LW, when high values are expected, are used as a reference for the melt season evaluation. Heatmaps for both FP and CP modes present the FYI-MYI KS values, in NR and FR for C- and L-band and have been color-coded based on three ranges: from 0 to 0.3 we define low KS, from 0.4 to 0.6 medium KS, and from 0.7 to 1 high KS separability.

Figure 10. Kolmogorov–Smirnov (KS) separability heatmaps for FYI and MYI at C-band. (a) C-band FP parameters. (b) CP parameters. Seasonal stages and NR and FR incidence angles are shown at the bottom.

Figure 11. Kolmogorov–Smirnov (KS) separability heatmaps for FYI and MYI at L-band. (a) L-band FP parameters. (b) CP parameters. Seasonal stages and FR incidence angles are shown at the bottom.

In LW, we observe high separability for the majority of FP and CP parameters related to surface and volumetric scattering. However, parameters such as co- and cross-pol ratios, RV/RH, RR/RL, m, S₁ and μ show lower separability, especially in the FR, in agreement with findings from Geldsetzer et al. (2015) and Nasonova et al. (2018).

In the MO, all KS values drop as expected relative to LW, though interestingly, FP and CP parameters sensitive to volume scattering HV, and mχ_v in the NR, show good KS separability of 0.6. This can be attributed FYI having higher volumetric scattering compared to MYI at this stage, due to brine wetted snow grains as reported by Barber and Nghiem (1999). In the FR, parameters sensitive to surface, single bounce scattering such as SPAN, RR, S₀, and mχ_s have good scores (KS = 0.6).

Progressing to the PO substage, the KS separability is higher compared to MO, for both NR and FR. Especially in the NR, parameters sensitive to surface scattering including SPAN, HH, VV for FP and RH, RV, S₀ and mχ_s in CP, denote high separability between FYI and MYI (KS = 0.7). Parameters sensitive to double bounce such as RL, show also high KS in NR incidence angles. In the FR parameters sensitive to Bragg scattering due to wind roughened melt ponds such as the co-pol ratio present the highest separability for this substage (KS = 0.8). The cross-pol ratio shows high separability in NR (KS = 0.7). These findings are supported by the SPAN and MPF analysis, confirming enhanced separability due to wind roughened melt ponds. Moreover, similar findings from Scharien et al. (2014) demonstrate the responsiveness of polarization ratios to surface melt ponds. The polarization ratios in each case, NR or FR, are sensitive to MPF and facilitate discrimination of ice types based on their MPF difference (FYI MPF > MYI MPF).

In PE, the separability between FYI and MYI in the NR is lower compared to PO. Nonetheless, in FR, high KS scores (KS = 0.7) are observed for parameters sensitive to surface scattering: SPAN and HH in FP mode, and RV, RH, S₀ and S₃ in CP mode. Highlighted in Figure 5, SPAN of FYI is higher than MYI, due to surface drainage, enabling a difference between the two ice types. Additionally, parameters sensitive to depolarization due to volumetric scattering (HV and mχ_v) present good separability results in FR, supporting the observed increase in the volumetric scattering component at this stage (Figure 6d).

During PD, parameters associated to depolarization due to volumetric scattering such as HV and mχ_v in the FR, show moderately good separability. This is consistent with PE, and the observed scattering mechanism differences for FYI and MYI occurring in PD (Section “FYI and MYI scattering properties across the melt season”), attributed to surface drainage. However, most of the C-band FP and CP parameters present low separability at this stage.

Comparing C-band to L-band, in the PO stage, slightly lower KS scores (KS ≤0.7) are found at L-band, whereas during PD, L-band KS scores are substantially higher. During PO, separability at L-band is optimized (KS = 0.7) for the FP parameter HH, and several CP parameters, including RH and mχ_v. Similar to C-band FR, the co-pol ratio shows high separability supporting the MPF analysis, while in CP the parameter RR associated to strong Bragg scattering shows high separability, reinforcing the contribution of melt ponds at this stage. During PD, parameters sensitive to surface scattering (SPAN, RH and S₀) and volumetric scattering (HV and mχ_v) enable very good separation (KS = 0.8), reinforcing that surface drainage allows for stronger volumetric scattering contributions.

It follows from the above analysis that C-band backscatter is more sensitive than L-band backscatter to surface melt pond coverage during PO. This enables greater ice type separability based on the polarization response of backscattered energy, which is optimized by using HV/HH in NR and VV/HH in FR. On the other hand, during PD the surface is considerably drained of melt ponds and the combined effect of enhanced volumetric scattering, and lower surface scattering, of MYI compared to FYI, from L-band backscatter leads to better separability compared to C-band.

FP and CP parameter redundancy

The correlation analysis revealed consistent patterns for both ice types, with akin polarimetric parameters being correlated. Overall positive correlations are observed for surface scattering related FP and CP parameters at both C- and L-band frequencies (ρ ≥ 0.8) (Figures S2 and S3). In the FR, C- and L-band scenes presented nearly identical relationships, while more decorrelated parameters were found for FR for both FP and CP parameters across the season. The correlation relationships were weakened during PE for both FP and CP, especially for MYI. During PD, the relationships were consistent, with the same correlation ranking between NR and FR, C- and L-band for both CP and FP with scores comparable to LW.

Following Dabboor, Montpetit, and Howell (2018) and Geldsetzer et al. (2015), to minimize redundancy, the FP and CP correlated parameters can be further grouped (Table 2). We have identified 3 groups of parameters that presented high correlation to each other for both FYI and MYI and a group of independent. The parameters in Group 1 were found strongly positively correlated to each other. Interestingly, these parameters are sensitive to changes in surface scattering, and performed similarly in the KS separability analysis, especially during PO and PE stages, when surface scattering dominated. Group 2 parameters were positively correlated and provided higher FYI and MYI separability during MO and PD, where we observed changes in volumetric scattering. Group 3 presented variable positive and negative relationships across the season, and variable KS scores. Based on Geldsetzer et al. (2015) this group shows more promising results in the medium range of incidence angles (30–39°). Finally, the independent group (Group 4) included the co- and cross-pol ratios, which are related to Bragg scattering and can be utilized in the cases of wind roughened melt ponds. Additionally, S₂ and RR (only for C-band) did not show any strong correlation (ρ≥ ±0.8) with other parameters across the season.

Table 2. Grouping of FP and CP polarimetric parameters, using Spearman’s correlation coefficient ≥ ± 0.80 for FYI and MYI types (n = 6000).

Group 1		Group 2	Group 3	Group4
FP	SPAN	HV		VV/HH
	HH			HV/HH
	VV HV*HH VV*HH
CP	RL RH	mχv	m S1	S2 RR
	RV		RR/RL*
	S0		RV/RH *
	S3		mχd
	mχs		αs μ

With * we denote the negative correlations.

Consequently, it is possible to utilize a representative parameter from each uncorrelated group, representing different scattering mechanisms, for the purpose of classification. Nonetheless, it is important to consider the effects of incidence angle and seasonality trends in parameter relationships as our findings emphasize, to aid FP and CP data classification in the advanced melt.

Summary and conclusions

This study provided an assessment of FP and CP polarimetric data, from C- and L-band SAR, for sea ice type separability in advanced melt conditions. We analyzed coincident RS2 and PS2 images, as well as high-resolution GeoEye-1 and Worldview optical imagery for understanding surface melt pond evolution and fractional coverage, in an area of the CAA containing a wide range of ice thickness, in 2018. Simulated CP polarimetric parameters were assessed in consideration of current, for example RCM, and future mission. In addition to statistical separability analysis, the parameter behaviors related to incidence angle, melt pond fraction and, by extension, ice fraction, were examined. Dominant scattering mechanisms for each ice type at each stage were considered using the m_χ decomposition.

For this work we addressed the following research questions: (1) Which frequency, C- or L-band, offers enhanced separability between FYI and MYI in advanced melt? (2) How do CP parameters compare to FP for FYI and MYI separability at each melt stage? (3) What is the role of melt ponds in the observed backscatter variability?

For question (1) we found the following: (i) for FP C-band, HV and cross-pol ratio in the NR, and SPAN in the FR presented consistent separability during MO, aligning with reports from Barber and Nghiem (1999) and Casey et al. (2016). An increase in volumetric scattering for FYI due to the presence of brine-wetted snow, and a decrease for MYI due to the masking effect of the wet snowpack, allows for better separability. (ii) separability during PO at C-band, when melt ponds dominate the returned signal, is best achieved using parameters sensitive to surface scattering in the NR. These include SPAN, HH, RH, RV and S₀ as well as decomposition mχ_s. This agrees with Geldsetzer et al. (2015) and Nasonova et al. (2018), who found high separability for the same parameters in the NR. In the FR, VV/HH and RV/RH presented the highest separability (KS = 0.8/0.7). At L-band during PO, both surface, Bragg and volume scattering associated parameters provide separability at the FR. These include HH, VV/HH, RH, RR and mχ_v. In PO, C-band performed better than L-band in terms of separability as evidence by higher KS values. (iii) separability during PE at C-band frequency is higher for FP surface scattering related parameters SPAN and HH in FR, while in CP parameters in both incidence angles show moderate separability. (iv) separability during PD using C-band frequency is possible using FR parameters HV and mχ_v, though, L-band in the FR provides higher separability (KS = 0.8) for surface (SPAN, RH, RV), volumetric (HV, mχ_v,) and single bounce (RR) scattering related parameters in FP and CP. The increased penetration depth from L-band into wet bare ice provides a higher contrast between FYI and MYI, due to volumetric scattering that dominates this substage. In addition, L-band data improved the identification of ice floe boundaries and deformation features (e.g., ridges) across the advanced melt season. These findings agree with previous works and strengthen our understanding when co-examining these datasets in such dynamic conditions.

For question (2), our analysis found that CP parameters were as effective as FP. The best CP parameters for both C- and L- band during the advanced melt were the first Stokes component S₀ together with the m_χ polarimetric decomposition and RR. These parameters highlighted the differences in surface, volumetric and Bragg scattering between the two ice types. The m_χv parameter, although not always the highest in KS separability value, ensures FYI-MYI discrimination across all advanced melt substages, depending on incidence angle utilized (here, NR or FR). Comparing to FP, at C-band HH enables separability during PO and PE substages, as HV does during MO and PD. At L-band, separability is achieved using HH during PO and HV during PD. These findings are potentially advantageous since SAR missions commonly operate in large swath ScanSAR dual-polarization HH + HV modes over sea ice, and CP data can also be collected over larger swaths. Future work should test real, not simulated, CP data and the various modes of missions such as RCM.

Another important contribution of this work comes from question (3). By examining SPAN and polarimetric ratios, together with MPF, and initially pure pond and pure ice samples, we were able to identify the influence of melt ponds on backscatter and polarimetric parameter variability. On the basis of the comparison between polarimetric ratios and MPF, we were therefore able to better understand the role of melt ponds on backscatter and ice type separability. We observed that, in the C-band frequency, NR and FR backscatter is strongly coupled to MPF during both PO and PD stages, though the effect of MPF is minimized when using the co-pol ratio in the NR and the cross-pol ratio FR. In the L-band frequency, the influence of MPF is much stronger during PO than PD, when surface flooding is widespread, affecting both ice types. These results highlight the influence of melt ponds on separability and the potential utilization of these parameters as indicators for melt pond variability, an important consideration in research in atmosphere-ice-ocean exchange processes during advanced melt. In addition, we found that the cross-pol ratio in NR and FR was sensitive to differences in FYI and MYI due likely to surface roughness differences during the PD stage. On the other hand, the co-pol ratio is a consistent indicator of MPF for FYI despite environmental changes such as wind roughened melt ponds. In the PD stage, the C-band co-pol ratio does not correlate well with MPF and a general reversal in trend compared to FYI is the subject of future study.

Our findings demonstrated the utility of both C- and L-band for improved sea ice type separability retrievals in the advanced melt season. A detailed stage-wise analysis of polarimetric parameters and relative scattering mechanisms at each stage and incidence angle range, supported a deeper understanding of the surface properties and mechanisms driving FYI and MYI backscatter in the advanced melt. The comparison of FP and simulated CP parameter highlighted the potential of missions such as RCM and the parameters presented here, to be utilized from current data at different CP modes. Importantly, convectional polarimetric channels such as HH and HV presented promising FYI and MYI separability in the advanced melt, together with the co-and cross pol ratios, indicating that missions such as Sentinel-1 can be utilized for studies in the advanced melt.

For expanding this work, the FP and CP parameters proposed, should be evaluated using image classification algorithms to determine a classification scheme for the melt season. In addition, regions of drifting pack ice should be examined since there will be different melt pond and sea ice characteristics and dynamic processes compared to land fast ice conditions examined here. Similarly, more years should be examined during the advanced melt season following this analysis to enhance the robustness of this work. In-situ observations of relevant sea ice properties such as surface roughness, at radar and geometric scales (ridges, hummocks, etc.), as well as dielectric properties, ice properties microstructure and texture/stratigraphy, and snow properties including snow depth, snow wetness, and grain dimension, should be obtained directly coincident to SAR acquisitions. The concurrent use of data from airborne platforms, either for the purpose of scaling from local to regional scales, or for direct SAR measurement, would be optimal. With the imaging capability of RCM, combined with pending missions such as NISAR, a new era for CP data requires expanded study of the use of CP SAR for overcoming challenges associated with highly dynamic melt season conditions, building on this, and concurrent work.

Acknowledgments

Special thanks to Malin Johansson for providing two of the ALOS-2 PALSAR-2 scenes from JAXA under the 3rd Research Announcement on Earth Observations (PI: Malin Johansson PER3A2N093). Thank you to Environment and Climate Change Canada’s Climate Research Division for providing high-resolution GeoEye-1 and Worldview-2 images. We also thank Christian Haas for reviewing the paper and providing EM bird ice thickness data.

Disclosure Statement

No conflict of interest was reported by the author (s).

Additional information

Funding

Research was funded by Marine Environmental Observations and Predictions and Response (MEOPAR) Network [1-02-02-004.5], Natural Sciences and Engineering Research Council of Canada Discovery [RGPIN-2022-05217], and Polar Knowledge Canada (POLAR) Science and Technology [NST-1718-0024] grants to Scharien. ALOS-2/PALSAR-2 scenes were provided by JAXA under the 6th ALOS Research Announcement (Project 3348; PI: Scharien).