Melt onset (MO) indicates a transition from winter to spring and is an important component of the Arctic sea ice energy balance [1
]. During melt, the sea-ice or snow surface becomes wet, causing a reduction of surface albedo. The MO date is therefore representative of an initiation of a positive feedback loop. The reduced albedo allows for greater absorption of shortwave solar radiation, which can lead to increased surface melt [3
]. During melt, as the ice begins to break-up, the appearance of open water leads to increased absorption of solar radiation into the ocean, which can both accelerate bottom melt and delay freeze-up [4
]. Melt onset dates are therefore correlated with melt season length [6
], and the September minimum sea ice extent, making long term MO information of value for climate studies [7
The Canadian Arctic Archipelago (CAA) is a region of islands in the Canadian Arctic between the Arctic Ocean and the Hudson Bay System. The ice cover in the CAA is a mixture of first-year ice, and thicker multi-year ice. In the CAA, the duration of the melt season is linked with the quantity of first-year ice that remains at the end of the ice season, and is therefore promoted to multi-year ice [8
]. Due to the thickness of multi-year ice, it is less likely to break-up during melt than first-year ice. However, during melt the thickness of the multi-year ice may be decreased, impacting its ability to survive future melt seasons [8
]. Melt onset and the duration of the melt season in the CAA are therefore particularly important toward understanding the fraction and characteristics of multi-year ice in the CAA. Due to the narrow waterways of the CAA, high-resolution satellite data are often used to determine MO in this region.
Data from synthetic aperture radars (SARs), scatterometers and passive microwave radiometers have been used in previous studies to determine MO dates [8
]. While data from both SARs and scatterometers can yield MO dates at reasonably high spatial resolution (100 m–5 km), the most commonly used MO retrieval methods use passive microwave data from the scanning multi-channel microwave radiometer (SMMR), and/or the special sensor microwave/imager (SSM/I), or the special sensor microwave imager/sounder (SSMIS). These passive microwave data are available each day over the entire Arctic from 1979 to the present, making these data ideal for climate studies. Current passive microwave MO estimation methods include the advanced horizontal range algorithm (AHRA), developed by Drobot and Anderson in 2001 [13
], and the passive microwave algorithm (PMW) developed by Markus in 2009 [7
]. Both methods use daily averaged 19 GHz and 37 GHz brightness temperature data from SMMR and SSM/I. However, the instrument field of view (IFOV) of the SSM/I sensor is 69 × 43 km for the 19 GHz channel, with similar IFOV for SMMR. These IFOVs are too large to resolve MO dates for large portions of the CAA without significant signal contamination due to land. Since 2002, advanced microwave scanning radiometers (AMSR-E and AMSR-2) have been collecting data over the arctic region at a much higher spatial resolution. For example, for the 19 GHz channel on AMSR2 the IFOV is 14 × 22 km, while for the 36.5 GHz channel (hereafter 37 GHz) it is 12 × 7 km. The IFOVs are similar for the AMSR-E instrument. Given these significantly smaller IFOVs, it is logical for MO in the CAA to be determined using the higher resolution AMSR data rather than the lower resolution SMMR and SSM/I data, at the expense of a shorter time series. In addition, it would be advantageous to utilize only the 37 GHz channel if possible. In this regard, the method developed in [14
] for MO over the Greenland ice sheet is of interest. This method uses the brightness temperature difference between ascending and descending passes at 37 GHz. When this difference exceeds a certain threshold, MO is said to have occurred. A drawback of this approach (and others) is that the threshold used to determine MO is fixed, and needs to be re-calibrated for specific geographic regions [14
]. Ideally, a method that does not rely on a fixed threshold could be used.
In this paper, we present a new MO determination method, named the dynamic threshold variability method (DTVM). Similar to [14
], the method uses swath data to pick up on diurnal variability patterns associated with MO [15
] that would otherwise be missed using daily averaged data, which are used in the AHRA and PWM algorithms. In this sense, our method is designed to detect early melt onset, which corresponds to the first appearance of water in the snowpack and is often followed by diurnal cycles of melt and refreeze [7
]. An early melt onset date is similarly defined in the PMW algorithm, while AHRA has a single MO date. Details pertaining to the MO detection of these two methods can be found in [7
]. The main difference between DTVM and previous approaches is that instead of using a fixed threshold, a range of threshold values are evaluated. To present this new approach, we first review the two leading MO methods for passive microwave data; the AHRA and PMW methods. We then describe the methodology used in the DTVM for MO estimation. This is followed by a comparison of the passive microwave methods with each other, in addition to MO estimated using 2 m air temperature from reanalysis as well as data from temperature profilers installed in sea ice. Finally, we discuss benefits and possible shortcomings of the DTVM that can be improved upon in future studies.
Comparisons between MO dates estimated using three passive microwave methods, PMW, AHRA and DTVM, show that the modes of the differences are centred around 0. This indicates that, generally, the three passive microwave methods are in agreement. MO dates estimated using the DTVM and PMW methods agree more closely to each other than to the AHRA MO dates. This is likely a result of both the DTVM and PMW methods relying partly or entirely on the variability of TB37V. Both histograms in Figure 5
, in which the AHRA is compared to the PMW and the DTVM, indicate bi-modal distributions centred around −1:1 and 8:10. This is strong evidence to suggest that, in the CAA, the 20-day window test, which is used by the AHRA, is estimating MO 9 days earlier than DTVM and PMW. The AHRA determined MO before a significant increase in surface temperature and before any significant change in the HR parameter. As the AHRA uses different channels it may be sensitive to different features of the surface signal. For example, horizontally polarized channels are known to be more sensitive to surface roughness than vertically polarized channels [31
]. The HR parameter used in AHRA also has substantial overlapping signatures with wind-slab (due to blowing snow) and emission from a brine-rich snow layer [32
], in particular during the range (−10 °C to 4 °C) that could trigger the 20 day window test.
To investigate the possibility that weather effects could be impacting MO dates, we examined the values of the polarization and gradient ratios that are used as weather filters in passive microwave ice concentration retrieval algorithms [19
]. These ratios were generally found to stay below the thresholds that would indicate atmospheric effects, such as contamination of the surface signal by cloud liquid water or water vapour, hence, similar to previous studies, we believe 37 GHz can be used without correcting the brightness temperatures for atmospheric effects (at least in the landfast region considered here). However, this does not take into account the fact that increasing water vapour (and downwelling longwave radiation) can indirectly lead to changes in the snow cover and hence emissivity of the snowpack, which would then lead to changes in brightness temperature. This cannot be fully taken into account without a more detailed investigation. For example, by carrying out emission modelling (e.g., [33
]). This is an exploratory area for current microwave emission models, and is outside the scope of the present study.