1 Satellite melt detection Melt detection from QSCAT followed the methods outlined by Trusel et al. [2012] with a minor change to one parameter. This detection process examines temporal changes in backscatter and considers a pixel as melting when its backscatter drops below a defined threshold when evaluated against the winter (July–September) mean backscatter of that pixel. For greater than 90% of the pixels examined (on average), a 2-dB threshold below the winter mean of each pixel was used. However, it is generally necessary to account for effects of snow accumulation in backscatter time series over ice sheets [Wang et al., 2007; Trusel et al., 2012]. As such, the melt detection process accounts for slow changes in backscatter owing to snow accumulation by examining the slope of the backscatter in each pixel over time. In Trusel et al. [2012], if the difference between pre-melt (winter) and post-melt (autumn: days 122-152; ~May) backscatter was greater than 1.0 dB, two melt thresholds were used: a smaller threshold for the beginning of the melt season, and a slightly larger threshold for the end of the season to prevent snowfall-induced backscatter reductions from being considered melt. In this study, we increased this pre-melt to post-melt threshold to 1.7 dB after finding that the original value (1.0 dB) resulted in unnecessarily using a dual threshold method at Neumayer Station during two years of the record. Effectively, this change increases the use of a single threshold and nearly all melt is still detected with a 2 dB threshold. As pixels requiring 2 thresholds were exceptional, previous results presented in Trusel et al. [2012] are not affected. 2 Surface energy balance modeling Fluxes of meltwater derived from surface energy balance (SEB) modeling of in situ observations are used for calibration and evaluation of remotely sensed data from QSCAT. In this study, we utilize SEB-based observations from five locations across Antarctica (Neumayer Station, Larsen C AWS, AWS 14 (on Larsen C), AWS 15 (on Larsen C), and Pine Island Glacier AWS A), all with slightly different methods for calculating the SEB owing to specifics of the experimental design or observational limitations. In all comparisons, SEB and QSCAT observations were temporally limited to coincident annual or sub-annual periods governed by the availability in in situ observations. Data from Neumayer Station are those previously published by van den Broeke et al. [2010]. These melt data are derived from observations of the radiative fluxes and modeled turbulent fluxes using the bulk method, which incorporates observations of wind speed, air temperature, and relative humidity to determine sensible and latent heat fluxes. The Neumayer SEB model was coupled to a snow model in order to calculate the subsurface heat flux and model meltwater percolation and refreeze. This dataset encompasses eight full summers (1999–2007) and one partial year (up to 31-December-2008) of surface meltwater fluxes and residual snowpack liquid water. The eight full and temporally continuous summers (1999–2000 to 2007–2008) were used to calibrate QSCAT MDD and SEB-based melt [Figure 1A], while the partial summer (up to 31-December-2008) was used in evaluating QSCAT-derived melt fluxes along with SEB-based data from elsewhere [Figure 1B]. We refer the reader to van den Broeke et al [2010] for further details on the Neumayer Station instrumentation and model. We emphasize that Neumayer is a Baseline Surface Radiation Network (BSRN) station [Konig-Langlo, 2013], making these SEB and melt calculations particularly reliable. SEB-derived melt flux observations are also incorporated from two automated weather stations on Larsen C ice shelf (AWS 14 and AWS 15) operated by the Institute for Marine and Atmospheric Research Utrecht and previously published in Kuipers Munneke et al. [2012]. These melt results are also based upon radiative flux observations and modeled turbulent and ground heat fluxes. The snow model used for these data uniquely incorporates subsurface absorption of solar radiation, and as such allows for melt initiation below the snow surface, which was found to be significant on Larsen C Ice Shelf [Kuipers Munneke et al., 2012]. For consistency with other SEB melt fluxes, our evaluation in Figure 1B only includes the surface melt component. However, QSCAT is potentially sensitive to both liquid water at the surface and shallow subsurface [Ashcraft and Long, 2006; Trusel et al., 2012], thus in section 3 of this Supplementary Material we also examined the total instantaneous melt flux incorporating both surface and subsurface melt at these two locations. AWS 14 and AWS 15 were installed in January 2009 and therefore only share several months of overlap with QSCAT. We refer the reader to Kuipers Munneke et al. [2012] for further details on the AWS 14 and AWS 15 dataset. Original SEB modeling was performed on observations from the PIG (Pine Island Glacier) AWS A (75.184°S, 101.729°W) operated by Environmental Fluid Dynamics Laboratory at New York University [http://efdl.cims.nyu.edu]. PIG AWS A operated from January 2008 to January 2011. This SEB model used observations of shortwave and longwave radiative fluxes, as well as wind speed, air temperature, and relative humidity to model the turbulent fluxes using the bulk method. The SEB model was not coupled to a snowpack model, rather the conductive heat flux to the subsurface was set to a fixed 10 W m-2. The resulting small melt fluxes from January 2008 (1.29 mm w.e.) and the full 2008–2009 austral summer (5.38 mm w.e.) were compared with QSCAT. Additional melt fluxes generated from a SEB model using the Larsen C AWS and presented by van den Broeke [2005] are included for comparisons with QSCAT. This model is unique in that no radiative flux terms were monitored in situ. Rather, radiative fluxes were derived from top of the atmosphere fluxes and assumptions about cloudiness and atmospheric transmissivity based upon measured air temperature and wind speed. Thus, the absolute accuracy of this dataset is expected to be much smaller than at the other locations, although we find an overall consistency with QSCAT and RACMO melt fluxes. Here, melt fluxes from 1999–2000, 2001–2002, and 2002–2003 calculated by van den Broeke [2005] are used in evaluating QSCAT, while the 2000–2001 melt season was rejected owing to insufficient air temperature observations during this austral summer. For more information on the Larsen C AWS melt fluxes we refer the reader to van den Broeke [2005]. 3 Additional evaluations of satellite-based melt fluxes This section presents additional analyses comparing ground and satellite-based results in order to more fully assess melt fluxes estimated from QSCAT MDD. Figure 1B shows the relationship between QSCAT- and SEB-based melt fluxes using all available in situ observations over annual and sub-annually summed periods as listed above in Section 2 of this Supplementary Material. However, if the calibration points from Neumayer Station (data from 1999-2000 to 2007–2008; n = 8) are removed from Figure 1B, the relationship remains significant (R^2 = 0.973, p < 0.001, n = 8) and linear regression coefficients (slope = 1.054, intercept = 20.54) are similar to those in the calibration equation, particularly for the slope coefficient which is nearly identical. The relationship using only data from AWS locations on Larsen C Ice Shelf (Larsen C AWS, AWS 14, and AWS 15) is also strong (R^2 = 0.841, p = 0.028, n = 5) although the slope coefficient becomes shallower (slope = 0.674) and the intercept is large (intercept = 146.78) owing to a lack of low melt data to constrain this relationship. A regression based on data only from AWS 14 and AWS 15 (n = 2) or from only PIG AWS (n = 2) is infeasible given insufficient data points. Subsurface melt fluxes resulting from shortwave penetration into the snowpack were generated alongside surface melt fluxes by the SEB model at AWS 14 and AWS 15 on LCIS [Kuipers Munneke et al., 2012]. Because QSCAT is potentially sensitive to melt in the subsurface, we examined the effect of including these melt fluxes in the evaluation of QSCAT melt results. Statistics generated from this analysis are shown in Table S1. Including the subsurface melt component produces weaker correlations as well as larger root-mean-square errors (RMSE) with all combinations of ground stations examined [Table S1]. Bias (mean error) is lower with subsurface melt included [Table S1], but this is only because QSCAT then greatly underestimates AWS 14 melt, compensating for overestimation at Larsen C AWS. These factors suggest QSCAT is most sensitive to melt at the surface. As such, the potential exists for underestimation of total meltwater fluxes in places such as AWS 14, where the subsurface melt component is significant [e.g., Kuipers Munneke et al., 2012]. 4 References Ashcraft, I. S., and D. G. Long (2006), Comparison of methods for melt detection over Greenland using active and passive microwave measurements, International Journal of Remote Sensing, 27(12), 2469, doi:10.1080/01431160500534465. König-Langlo, G. (2013), Basic measurements of radiation from Neumayer Station in the year 1999-2007. Alfred Wegener Institute, Helmholtz Center for Polar and Marine Research, Bremerhaven, doi:10.1594/PANGAEA.819774. Kuipers Munneke, P., M. R. van den Broeke, J. C. King, T. Gray, and C. H. Reijmer (2012), Near-surface climate and surface energy budget of Larsen C ice shelf, Antarctic Peninsula, The Cryosphere, 6(2), 353–363, doi:10.5194/tc-6-353-2012. Trusel, L. D., K. E. Frey, and S. B. Das (2012), Antarctic surface melting dynamics: Enhanced perspectives from radar scatterometer data, J. Geophys. Res., 117, F02023, doi:201210.1029/2011JF002126. Van den Broeke, M. (2005), Strong surface melting preceded collapse of Antarctic Peninsula ice shelf, Geophys. Res. Lett., 32(12), doi:10.1029/2005GL023247. Van den Broeke, M., G. Koenig-Langlo, G. Picard, P. Kuipers Munneke, and J. Lenaerts (2010), Surface energy balance, melt and sublimation at Neumayer Station, East Antarctica, Antarct. Sci., 22(1), 87–96, doi:10.1017/S0954102009990538. Wang, L., M. Sharp, B. Rivard, and K. Steffen (2007), Melt season duration and ice layer formation on the Greenland ice sheet, 2000-2004, J. Geophys. Res.-Earth Surf., 112(F4), doi:10.1029/2007JF000760.