Numerical Terradynamic Simulation Group 8-2015 New satellite climate data records indicate strong coupling between recent frozen season changes and snow cover over high northern latitudes

Weexamined new satellite climate data records documenting frozen (FR) season and snow cover extent (SCE) changes from1979 to 2011 over all northern vegetated land areas (⩾45 °N). New insight on the spatial and temporal characteristics of seasonal FR ground and snowpackmelt changes were revealed by integrating the independent FR and SCEdata records. Similar decreasing trends in annual FR and SCEdurations coincidedwithwidespreadwarming (0.4 °C decade). Relatively strong declines in FR and SCEdurations in spring and summer are partially offset by increasing trends in fall andwinter. These contrasting seasonal trends result in relatively weak decreasing trends in annual FR and SCEdurations. A dominant SCE retreat response to FRduration decreases was observed, while the sign and strength of this relationshipwas spatially complex, varying by latitude and regional snow cover, and climate characteristics. The spatial extent of FR conditions exceeds SCE in early spring and is smaller during snowmelt in late spring and early summer, while FR ground in the absence of snow cover is widespread in the fall. The integrated satellite record, for the first time, reveals a general increasing trend in annual snowmelt duration from1.3 to 3.3 days decade (p< 0.01), occurring largely in the fall. Annual FR ground durations are declining from0.8 to 1.3 days decade. These changes imply extensive biophysical impacts to regional snow cover, soil and permafrost regimes, surfacewater and energy budgets, and climate feedbacks, while ongoing satellitemicrowavemissions provide an effectivemeans for regionalmonitoring.


Introduction
The seasonal freeze/thaw (FT) state transition to predominantly non-frozen (FR) conditions in the spring initiates processes that are nearly dormant during the winter FR season and is related to seasonal snowmelt and soil thawing in high northern latitude (HNL) ecosystems (Kimball et al 2001, Euskirchen et al 2007, Mortin et al 2012. The FT signal and associated FR season metric obtained from satellite microwave remote sensing characterizes the predominant FR or non-FR status of the land surface and the duration of FR conditions within the sensor footprint, without distinguishing among individual landscape elements, including vegetation, snow cover and surface soil conditions (Zhang et al 2011, Kim et al 2012. Snow cover extent (SCE) exerts a strong impact on surface climate and FT conditions by providing an effective thermal buffer reducing soil-atmosphere energy exchange and maintaining warmer winter soil temperatures than would otherwise occur under snow free conditions (Wang and Zender 2011, Mortin et al 2012. SCE and FT variations are also coupled with the lower atmosphere by regulating surface energy partitioning of net solar radiation into sensible and latent heat through associated changes in land surface albedo and evaporation (Betts et al 2014). Recent widespread decreasing trends in the FR season and SCE over HNL land areas (⩾45°N) have been reported (Dye 2002, Brown and Robinson 2011, Derksen and Brown 2012, Kim et al 2012 and may be associated with rising atmospheric heating (Dye 2002, Mioduszewski et al 2014, and accelerated by an amplified snow-albedo feedback (Chapin et al 2005, Dery and.
In a typical HNL seasonal snow cycle, snowmelt may last for several days to weeks in the spring until the snowpack is depleted and nighttime temperatures remain above freezing (Semmens et al 2013); surface air temperatures (SATs) may rise above freezing under daily solar radiation and thermal loading, while snow and underlying soil temperatures remain near 0.0°C or below freezing until the snow cover heat sink is gone. The seasonal transition from dry to wet snowpack conditions with spring snowmelt onset generally coincides with a rapid decline in land surface albedo (Ling andZhang 2003, Betts et al 2014), increased liquid water availability in the landscape (Kimball et al 2001, Mortin et al 2012 and a seasonal shift in the surface energy budget from predominantly sensible to latent energy, with commensurate increases in land surface evaporation (Chapin et al 2005, Ling and Zhang 2003, Zhang et al 2011. Prior to the beginning of the HNL snow cycle, FR ground conditions may occur in early to mid-fall before persistent snow cover (Kim et al 2014a). FR conditions without a buffering snow layer may promote colder soil temperature extremes, damaging vegetation and reducing subsequent winter soil decomposition and respiration processes (Daniels et al 2011, Kreyling et al 2012, Aanderud et al 2013, Du et al 2013. Differences in the seasonal timing and extent of FT and snow cover may also impact other ecosystem properties, including soil active layer development, permafrost temperature and soil organic carbon content (Haei et al 2013, Ling and Zhang 2003, Park et al 2014. Despite significant impacts of snow cover status on the terrestrial energy budget and associated carbon and climate feedbacks, few studies have examined regional changes in snow cover conditions over the HNL domain other than documenting changes in SCE. In this study, we document relationships between FR season and SCE changes over the HNL domain (i.e., vegetated land area poleward of 45°N) using new satellite climate data records (CDRs) for these parameters extending over more than 30 years . We characterize seasonal offsets between FR area and SCE, and regional trends and temporal correlations in these parameters. Finally, we quantify emergent regional patterns and trends in the snowmelt season and FR ground duration over snowfree land areas by integrating the satellite microwave FT and the combined visible-band and microwave sensor based SCE data records.
2. Data and methods 2.1. Satellite data Satellite passive microwave remote sensing is wellsuited for FT monitoring over the HNL domain due to strong contrast in land surface dielectric properties and brightness temperature (T b ) between predominantly FR and non-FR conditions, enhanced by the relative insensitivity of lower frequency (e.g. ⩽ 37 GHz) microwave retrievals to solar illumination and atmosphere cloud/aerosol contamination constraints , Frei et al 2012. We used a global landscape FT Earth system data record (FT-ESDR) derived from calibrated 37 GHz, vertically polarized and overlapping T b records from the scanning multi-channel microwave radiometer and special sensor microwave imager; the FT-ESDR distinguishes twice daily (AM and PM) FT conditions from ascending and descending orbit T b retrievals, posted to a 25 km resolution EASE-Grid (Brodzik and Knowles 2002) for the period 1979-2012 (Kim et al 2012(Kim et al , 2014b. Four categorical daily FT classifications are distinguished, including FR (AM and PM FR), non-FR (AM and PM thawed), transitional (AM FR and PM thawed) and inverse transitional (AM thawed and PM FR) conditions. The reported mean annual FT-ESDR spatial classification accuracy exceeds 84% over the HNL domain, but with approximately 7% lower accuracy during seasonal FT transition periods in spring and fall (Kim et al 2012). The FT retrieval represents predominant FR or non-FR conditions within the satellite footprint and does not distinguish individual landscape elements, including air, soil, vegetation and snow cover within each 25 km resolution grid cell (Zhang et al 2011, Kim et al 2012.
FR duration was defined from the daily FT record of the number of FR or transitional days for annual and seasonal periods (table S1). The four seasons are defined as Winter (December-February), Spring (March-May), Summer (June-August) and Fall (September-November), because seasonal snow melt and onset generally occur during spring and fall portions of the calendar year, respectively (Robinson and Frei 2000, Dye 2002. The HNL FR duration temporal trend patterns were compared against daily 2 m SAT estimates from the quarter-degree resolution ERA-Interim global reanalysis (Dee et al 2011). Satellite global CDRs of weekly SCE (Brodzik and Armstrong, 2013) projected in a consistent 25 km resolution EASE-Grid format (Brodzik et al 2012) were used to examine seasonal SCE trends and correlations with HNL FR duration changes for the 1979-2011 period. The SCE observations were regridded from the NOAA snow chart CDR Robinson 2011, Estilow et al 2013) which spans 1967-present (http://snowcover.org), and is manually derived by analysts from combined visibleband satellite imagery, including advanced very high resolution radiometer, geostationary operational environmental satellite and moderate resolution imaging spectroradiometer records (Helfrich et al 2007). Analysis of the updated SCE record indicates improved data quality during the Northern Hemisphere spring (Brown and Robinson 2011). For this study we selected the period 1979-2011, because the FT-ESDR was available from 1979 to 2012 and the SCE data records only extended up to 2011 at the time of this investigation. We examined seasonal and annual SCE duration, defined as the percentage of time in a given period that each grid cell was snow covered, ranging from 0 to 100 percent (table S1). SCE duration is preferred over the binary classification because it can more accurately account for SCE variability in patchy snow areas (Nolin 2010).
The satellite FT and SCE records were integrated to estimate the extent and duration of snowmelt and FR ground conditions. Snowmelt was defined for grid cells where the SCE record indicated snow cover presence, but the FT-ESDR indicated transitional (AM FR, PM non-FR) FT conditions. FR ground cells were defined for coincident FT-ESDR defined FR and snow-free SCE conditions. Durations of snowmelt and FR ground were defined from the weekly data record of the number of grid cells for annual and seasonal periods (table S1). Weekly FT status was determined for each grid cell from the daily FT-ESDR using a 50% temporal threshold within the corresponding coarser 7-day SCE time step. The resulting snowmelt and FR ground metrics were then aggregated on a similar seasonal and annual basis as the FR and SCE metrics described above. The combination of SCE and FT transitional conditions was used as a more conservative indicator of snowmelt relative to the potential addition of FT-ESDR defined non-FR (AM and PM non-FR) conditions. This was done to minimize potential impacts from assumed greater SCE retrieval uncertainty under patchy and transient snow cover conditions later in the spring melt season when non-FR (AM and PM) conditions are more prevalent , Kim et al 2014a. Here, FT-ESDR classified transitional conditions coinciding with SCE defined snow cover are assumed to represent snowmelt and associated wet snowpack conditions. Likewise, the use of only FT-ESDR defined FR (AM and PM FR) status to estimate FR ground conditions represents a more conservative indicator than the potential use of both FR and transitional (AM FR PM non-FR) FT states.

Statistical methods
Annual trends were calculated using pre-whitened Kendall's tau statistics following removal of temporal autocorrelation using the ZYP package in R statistics (Yue and Pilon 2004). When trends were analyzed, outliers were screened as a non-systematic variation identified on a grid cell-wise basis as quantities surpassing ±2 times the standard deviation (SD) of the long-term record means to minimize any remaining sensor discontinuity occurring from inter-sensor calibration and resolution difference of various satellite sensors (Jeganathan et al 2014, Kim et al 2014b. The Spearman's correlation coefficient (r-value) was used to assess the sign and strength of relationships between FR and SCE duration anomalies on an annual and seasonal basis. Temporal anomalies of the parameter series were initially computed as annual differences from average conditions characterized from the period of record; where a significant (p ⩽ 0.1) trend was identified, the temporal anomalies were determined as differences from the long-term detrended mean (Barichivich et al 2014, Kim et al 2014c). The significance of these relationships was stratified according to relative strong (p ⩽ 0.05), moderate (0.05 < p ⩽ 0.1) and weak (p > 0.1) categories. The correlations between annual FR and SCE duration anomalies were determined for each grid cell within the HNL domain and analyzed within different segmentation levels determined from independent ancillary geospatial data, including latitudinal zones, characteristic seasonal snow types determined from a 0.45-degree resolution snow physical properties database (Sturm et al 1995), and 0.1-degree resolution Köppen climate zone map (Peel et al 2007). For consistency, all data records used in this investigation were resampled to the same 25 km resolution EASE-grid format using nearest-neighbor resampling of coarser resolution grid cells or drop-in-bucket averaging of finer resolution pixels. The HNL domain in this study includes all vegetated land areas where seasonal FR temperatures are a major constraint to land surface water mobility and ecological processes over the area poleward of 45°N, while approximately 63% of the HNL domain is also underlain by permafrost (Brown et al 2014).

Correspondence between FR season and snow cover variability
The satellite records show similar trends  in FR and SCE durations over the HNL domain (table 1), with strong declines in FR and SCE durations for the spring and summer. Strong decreasing regional trends in FR and SCE durations occur in summer over both Eurasia (EA) and North America (NA), but with relatively stronger decline over NA. The decreasing trend in mean annual FR duration over the HNL is largely driven by strong decreasing spring and summer FR trends, offset by moderately increasing fall and strongly increasing winter FR trends. A stronger decreasing mean annual FR season trend over EA results from a strong FR duration decrease in spring relative to NA. The mechanisms for fall and winter cooling are uncertain, but coincide with reported snow cover increase (Bulygina et al 2009) and fall/winter cooling attributed to summer warming and sea ice decline (Cohen et al 2012a(Cohen et al , 2012b. Relatively weak or inconsistent seasonal SCE trends may result from potential SCE overestimation during spring and fall over some northern land areas , Brown and Derksen 2013, Mudryk et al 2014, likely due to remaining cloud contamination in the satellite visible band sensor data (Tang et al 2013). The mean annual FR season changes also correspond with ERA-Interim SAT annual anomalies over the HNL (correlation (r) = −0.297; p < 0.1), NA (r = −0.503; p < 0.05) and EA (r = −0.563; p < 0.05). Mean annual and seasonal correlations between FR and SCE duration anomalies over the HNL and NA, and EA portions of the domain are summarized in table 2. FR season variability generally coincides with similar annual and seasonal changes in SCE duration. A relatively weak HNL correlation in spring contrasts with stronger correspondence for continental sub-regions and may reflect opposing oscillations in spring climate conditions between EA and NA (Zhang et al 2007). The HNL correlations between annual FR season and SCE variations (figure 1) are predominantly positive, indicating a general SCE retreat coincident to FR duration decreases. The sign and strength of the mean correlations vary according to latitude, with generally stronger, positive correspondence at higher latitudes (figure 2(a)); the correlations are also lower or negative where the FR and SCE seasons occur over a smaller portion of the annual cycle, and consistent with the generally shorter cold season at lower latitudes. The correlations of FR and SCE durations also vary according to regional climate and characteristic snow cover conditions. Lower correlations occur in maritime, prairie and alpine snow conditions, whereas relatively higher correlations occur in tundra and taiga snow zones ( figure 2(b)). Alpine and maritime snow areas are characterized by complex terrain and microclimate heterogeneity that may not be effectively resolved by the coarse (25 km) resolution satellite retrievals (Du et al 2014). Maritime and prairie snow areas are also characterized by ephemeral or shallow, patchy snow conditions and lower SCE duration (Painter et al 2009, Betts et al 2014 that may vary independent of FR season changes. Negative correlation areas also coincide with lower latitude temperate and dry climate zones, including Central NA and Western Europe (figure 1). In these areas, longer FR durations coincide with colder, drier atmosphere conditions that promote less precipitation and snow cover. In lower latitude dry climate areas, significant snow cover may also be lost to the atmosphere through wind redistribution and sublimation even under

Snowmelt and FR ground characteristics
The climatology of HNL FR area and SCE variations established from the long-term satellite records reveals a large dynamic seasonal range and strong interannual variability in the temporal progressions of these parameters (figure 3). The HNL FR area ranges from a summer minimum of approximately 0.2 million km 2 (temporal SD ±0.97 million km 2 mo −1 ) to a winter maximum of 32.5 million km 2 (SD ± 0.60 million km 2 mo −1 ). The mean differences between FR area and SCE are approximately 1.0 and 2.7 million km 2 (3.0% and 8.1% of the domain) for respective seasonal transition periods in spring and fall. FR area generally exceeds SCE in the fall and spring, confirming that FR conditions are a prerequisite for persistent fall and winter snow cover, whereas SCE exceeds FR conditions during wet snow conditions in mid-spring and early summer.
Prevailing FR or SCE grid cells are defined in figure S1, where classified FR or snow covered conditions exceed 50% of respective spring and fall periods over the entire record . FR areas generally exceed SCE in the early spring, while SCE exceeds FR area where snowmelt is occurring in the late spring and early summer, extending from generally coastal and southern areas in early spring to inland areas and northern latitudes as the melt season progresses. In the fall, FR areas occur in the absence of snow cover over large areas including portions of Asia and the Pacific Northwest. Table 1. Kendall's tau trends for mean annual and seasonal frozen (FR) duration (days decade −1 ) and seasonal SCE duration (% decade −1 ) enclosed in parentheses for HNL and continental North America (NA) and Eurasia (EA) sub-regions, and 1979-2011 record; trend significance levels are denoted by asterisks as: *moderate (0.05 < p ⩽ 0.1) and **strong (p ⩽ 0.05).

Annual
Winter  The estimated mean annual number of weeks of respective snowmelt and FR ground durations derived from the integrated FT and SCE records are 3.8 ± 1.7 (spatial-SD) weeks and 0.6 ± 1.1 (spatial-SD) weeks for the 33-year record and the HNL domain (figure 4). The highest mean snowmelt durations are located along the western NA coastal mountain zone, including Southern Alaska, coastal British Columbia, CN and the Pacific Northwest, USA ( figure 4(a)). These regions are characterized by cool to moderate climate conditions with extensive orographic driven precipitation and seasonal snow cover, increasing in depth and duration at higher elevations (Daly et al 2008). Moderate winter temperatures, humid atmosphere conditions, and frequent rain-on-snow events in this region promote frequent thawing and re-freezing of the snowpack, and transient snow cover conditions at lower elevations that contribute to the extended snowmelt season . The snowmelt duration is generally shorter in colder and drier climate areas, including boreal and Arctic zones characterized by stable winter FR temperatures and relatively rapid seasonal snowmelt onset, and snow cover depletion (Sturm et al 1995, Pomeroy et al 2006. In addition to climate constraints, snowmelt duration is also related to snow depth and vegetation stature (Sturm et al 2005, Marsh et al 2010. Longer FR ground durations coincide with lower latitude temperate and dry climate zones, including central NA, large portions of Europe and Southeastern EA ( figure 4(b)). FR ground conditions in the absence of insulating snow cover is less frequent for colder and higher latitude boreal and Arctic domains, where the FR season and SCE records are more tightly coupled (e.g., figure 2(a)).
The Kendall's tau regional trends for snowmelt and FR ground duration determined from the integrated FT and SCE records are presented in figure 5. These results show significant HNL regional trends of decreasing FR ground duration (p < 0.01) and increasing snowmelt duration (p < 0.01) over the 33-year record. The HNL mean annual snowmelt duration is increasing by approximately 0.19 weeks decade −1 while the FR ground duration is decreasing by −0.18 weeks decade −1 ( figure 5(c)). These trends also coincide with a 0.44°C decade −1 (p < 0.05) HNL warming trend derived from the ERA-interim SAT record. The relatively strong regional trends in snowmelt and FR ground durations from the integrated satellite record contrast with relatively weak or inconsistent SCE trends (table 1), and are largely driven by regional warming and associated increasing trends in the frequency of transitional FT events, and longer non-FR seasons (Kim et al 2014a). However, the increase in annual wet snowpack duration predominantly reflects increasing snowmelt duration in the fall (0.17 weeks Figure 1. Pixel-wise Spearman correlations (r) between annual FR season (days) and snow cover extent (SCE) duration (%) anomalies for the 1979-2011 record; areas outside the HNL domain are denoted in grey (land) and white (100% open water). Black areas on adjacent inset map represent significant (p < 0.1) correlations. Spatial convolution of a Gaussian weighting filter was applied to these maps. decade −1 ; p < 0.05) rather than the other seasons (table S2); this implies more frequent occurrence of transient snow cover and snowmelt prior to the onset of a more persistent seasonal snowpack in the fall, while the decrease in FR ground conditions is more uniformly distributed throughout the annual cycle (table S2).
The trend pattern of snowmelt and FR ground durations is spatially complex (figure 5). The snowmelt duration trend map shows a general lengthening of the snowmelt season over western NA and eastern EA. Significant (p < 0.1) increasing and decreasing trends in snowmelt duration represent approximately 22.0% and 4.4% of the HNL domain, respectively. The significant positive snowmelt duration trend areas average 0.76 ± 0.29 (spatial SD) weeks decade −1 (figure 5(a)) and are four times larger than the HNL regional trend ( figure 5(c)). Widespread increases in snowmelt duration may reflect increasing winter snow accumulations (table 1) and more frequent transitional FT conditions associated with regional warming trends (Bulygina et al 2009, Kim et al 2012. Significant increasing and decreasing trends in annual FR ground duration represent approximately 0.7% and 4.9% of the domain, respectively ( figure 5(b)); the significant negative trend areas show a mean FR ground duration decrease of −0.75 ± 0.38 (spatial SD) weeks decade −1 , which is 4.2 times larger than the HNL regional trend (figure 5(c)). Widespread decreasing FR ground duration trends are predominantly located along the southern HNL domain, while northern boreal and Arctic areas show generally weak, increasing trends.

Discussion
In this study, we utilized recent global satellite CDRs to analyze FR season and SCE changes over the HNL domain from 1979-2011, and provided new insight on seasonal snowmelt and FR ground changes revealed by a novel data fusion approach integrating the two independent and relatively well-calibrated satellite records. The satellite observations show similar changes in FR and SCE durations, with contrasting decreasing and increasing seasonal trends between spring/summer and fall/winter conditions, respectively. We find generally strong correspondence between mean annual and seasonal FR and SCE duration changes, but with regional variations in these relationships for different climate zones and snow cover regimes, and generally closer coupling at higher latitudes where the FR and SCE seasons are a larger portion of the annual cycle. The geographic extent of FR conditions is generally larger than SCE in fall and winter, resulting in large areas of freezing conditions without an insulating snow layer. In contrast, FR area is smaller than SCE in the spring, while regional . Spatial convolution of a Gaussian weighting filter was applied to these maps. differences in these two parameters delineate the zone and duration of snowmelt. Our results document an increasing HNL trend in the snowmelt duration, which predominantly occurs from an extension of snow cover and transitional FT conditions in the fall, and is coincident with regional warming. We also find a significant HNL decreasing trend in FR ground duration year-round, but which largely occurs over the southern HNL domain rather than the northern boreal-arctic permafrost zone. The timing and duration of FR ground in the absence of an insulating snowpack can influence the soil thermal regime and ecosystem processes, including soil litter decomposition and respiration, and potential freezing injury to plants (Beier et al 2008, Saito et al 2013. Our results show a general decreasing trend in FR ground duration over the HNL domain coincident with regional warming trends and consistent with a general relaxation of FR season constraints to land surface water mobility and ecosystem processes (Kim et al 2014a). However, the effect of these changes on regional soil temperature trends is uncertain because the soil thermal regime is strongly coupled to changes in timing and duration of SCE, Figure 5. Regional Kendall's tau trend patterns (weeks decade −1 ) of (a) snowmelt, (b) frozen ground durations, and (c) annual variation and significant trends (p < 0.01) of HNL mean snowmelt and frozen ground durations for the 1979-2011 record; mapped areas outside the HNL study domain are denoted in grey (land) and white (100% open water), while spatial convolution of a Gaussian weighting filter was also applied to the maps. which shows a more variable response pattern, as well as changes in snow depth and structure from which we have a general lack of regional information.
Significant snowmelt duration increases occur over 24.3% of the domain characterized by underlying permafrost. These changes may have considerable impacts on the terrestrial energy budget and landatmosphere climate feedbacks associated with lower surface albedos and enhanced net solar radiation loading under wet snowpack conditions , and associated non-FR season increases in surface latent energy exchange and evaporation (Zhang et al 2011). An extended snowmelt trend may also promote warmer soil temperatures, deepening active layer development and permafrost warming (Zhang et al 2005, Anisimov 2007). Snow structural changes from increasing melt events may also promote the development of thick ice layers, adversely affecting animal habitats, migration and foraging success (Bartsch et al 2010, Vincent et al 2011). Further research is needed to clarify the impacts of these changes on HNL ecosystems and regional climate feedbacks. Continuity of the FT-ESDR and SCE records enabled from ongoing satellite observations will support future studies of continuing changes and longerterm trends in these parameters.