1. Introduction
The Weather Research and Forecasting (WRF) Model (Skamarock and Klemp 2008) has become one of the more commonly used models for dynamical downscaling (e.g., Caldwell et al. 2009; Salathé et al. 2010; Racherla et al. 2012; Mearns et al. 2012; Gao et al. 2012; Vautard et al. 2013; Harkey and Holloway 2013). The goal of dynamical downscaling is to use a physics-based regional climate model to project regional- and local-scale climate features at finer spatial and temporal resolutions than are available from global models (e.g., Dickinson et al. 1989; Giorgi 1990; Feser et al. 2011). The resultant regional climate fields can be used to examine the effects of climate change—particularly extremes—on air quality, human health, water quality and quantity, and ecosystem services (e.g., Nolte et al. 2008; Trail et al. 2013; Arnbjerg-Nielsen et al. 2013; Fann et al. 2015).
In WRF, when water bodies are unresolved by or missing from the driving data, the default method of setting the missing water temperatures extrapolates the temperature from the nearest water point in the driving data to the target grid cell. This approach is sufficient for historical simulations because high-resolution water temperature data are available and typically a limited search radius is required. However, the method degenerates for dynamical downscaling since high-resolution water temperatures do not exist for the future, so incongruous water temperatures are often used instead. For example, as shown in Fig. 2 in Mallard et al. (2015) and discussed in detail later in this paper, in our 36-km North American modeling domain, the default method assigns temperatures from off the New Jersey coast in the Atlantic Ocean (several hundred kilometers to the east) to Lakes Erie and Ontario, as well as southeastern Lake Huron and southern Lake Michigan. Meanwhile, temperatures for most of Lake Huron, northern Lake Michigan, and eastern and central Lake Superior are assigned from James Bay, several hundred kilometers to the north; and temperatures in western Lake Superior are assigned from southwestern Hudson Bay, also several hundred kilometers to the north. The temperatures for Great Salt Lake are set from the Pacific Ocean, several hundred kilometers and two mountain ranges to the west. The default method creates nonphysical discontinuities where the temperatures are extrapolated from different bodies of water. In winter, those discontinuities are on the order of 10–15 K in Lake Superior and 20–30 K in Lakes Huron and Michigan. Gao et al. (2012) (focusing on a 4-km domain), Bullock et al. (2014) (using a 12-km domain), and Mallard et al. (2014) (using a 12-km domain) noted similar discontinuities in Great Lakes water temperatures by using the default method for dynamical downscaling with WRF. Sun et al. (2015), who focused on a 4-km domain, found that using the default method with WRF adversely affected precipitation patterns and accumulations in simulations of Lake Victoria, Africa.
The Great Lakes strongly influence the climate of the surrounding region, moderating air temperatures in summer and enhancing precipitation in winter. Notaro et al. (2013) presented an idealized regional climate modeling experiment where the Great Lakes were replaced by vegetation and showed that the presence of the Great Lakes altered the hydrologic budget, the surface energy budget, and the development and passage of synoptic systems. While the example in Notaro et al. (2013) was deliberately extreme, it demonstrated that the Great Lakes are important for dynamical downscaling. Similarly, Great Salt Lake affects the climate of the Intermountain West (e.g., Wang et al. 2010). Considering the potential difficulties of prescribing lake temperatures in WRF for downscaling, one might ask how essential it is to set the temperatures of large unresolved lakes, especially if the focus of the research study is hundreds of kilometers from those lakes. This paper elucidates the effects of setting lake temperatures with unrealistic and unrepresentative values—as can be done by default in regional climate modeling with WRF if no inland water temperature data exist in the driving data—and compares the resulting regional climate to that generated by a method where lake temperatures are consistent with the driving data.
2. Regional climate modeling
Two dynamical downscaling simulations were conducted using WRF version 3.6.1 (WRFv3.6.1) where WRF was initialized at the time labeled 0000 UTC 1 October 1994 and run continuously for a 39-month period so that the analysis period corresponds to the period labeled 1995–97 from the global climate simulation. A domain with 36-km horizontal grid spacing (Fig. 1) covered North America using a 34-layer configuration that extended to a model top at 50 hPa. The physics options included the Rapid Radiative Transfer Model for global climate models (Iacono et al. 2008) for longwave and shortwave radiation, the WRF single-moment six-class microphysics scheme (Hong and Lim 2006), the Kain–Fritsch convective parameterization scheme (Kain 2004) with feedback between the convection and radiation scheme (Herwehe et al. 2014), the Yonsei University planetary boundary layer scheme (Hong et al. 2006), and the Noah land surface model (Chen and Dudhia 2001). Spectral nudging (Miguez-Macho et al. 2004) was applied to horizontal wind components, potential temperature, and geopotential at all levels above the planetary boundary layer using the nudging coefficients from Otte et al. (2012).
(top) WRF 36-km domain used for simulations indicating lake points (turquoise), nonlake water points (gray), and land points (white) following USGS land use classification. (bottom) Locations and station IDs of buoys in the Great Lakes with long climatological records.
Citation: Journal of Climate 29, 2; 10.1175/JCLI-D-15-0233.1
The input data were 0.9° × 1.25° fields from phase 5 of the Coupled Model Intercomparison Project (CMIP5) using the “historical” run of the National Center for Atmospheric Research–Department of Energy Community Earth System Model (CESM) (Gent et al. 2011). The CESM fields used here are a representation of weather conditions experienced within the historical climate, but do not correspond to the actual weather conditions that occurred during any given year; therefore, comparisons cannot be made directly to observations for a specific year. The suite of variables in the input data included surface pressure and three-dimensional representations of temperature, specific humidity, wind components, pressure, and geopotential height extracted at 6-h intervals, as well as monthly sea surface temperature, ice fraction, soil moisture, and soil temperature. The only difference between the two WRF simulations was the method of setting lake temperatures: using the default method (“Default”), or using temperatures from the lake model within the Community Land Model (CLM) component of the CESM simulation (“CLM Lakes”). Selected CLM output fields are included in the CMIP5 archive, but they are not part of the typical suite of data set aside for regional climate downscaling. In CLM Lakes, lake surface temperatures were set in WRF using monthly averaged fields from CLM by adding several preprocessing steps (described below) to ensure that CLM data were used only at inland water points that are not represented by the atmospheric component of CESM [the Community Atmosphere Model (CAM)]. As part of the preprocessing, the monthly fields (including CLM lake surface temperatures) were temporally interpolated to 6-h intervals (where the monthly averaged value was assumed to be valid at 0000 UTC on the 15th day of each month) to avoid abrupt transitions in the downscaling simulations. The CESM fields provided initial, lateral boundary, and surface boundary conditions, and they served as the constraints for nudging.
In the absence of an ocean model or a lake model in WRF (e.g., Gu et al. 2015) for regional climate modeling, water temperatures (including those of inland water bodies) are not merely initialized. The temporal evolution of the water temperatures is prepared a priori from one or more external data sources, and those temperatures are typically used as lower boundary (i.e., surface) forcing and updated at the frequency of the lateral boundary forcing (which is every 6 h in the simulations shown here), but without temporal interpolation. In addition, the water temperatures in our WRF configuration do not respond to the overlying atmospheric conditions, although they are adjusted to reflect the presence of sea ice (which may be indicated from a different external data source). These water bodies exist only in the horizontal plane at the surface in WRF and therefore do not consider lake depth or varying temperatures and salinity below the surface. See Mallard et al. (2015) for a thorough discussion of alternative methods of setting lake temperatures in WRF.
To use the CLM Lakes method, the land use classification in WRF must distinguish between inland water bodies (i.e., lakes) and ocean points. The WRF modeling domain in this study (Fig. 1, top panel) used the 2-min, 24-category U.S. Geological Survey (USGS) classification available in WRF, which was augmented so the lakes are distinguished by an extra category using “usgs_lakes+2m” in the WRF Preprocessing System (WPS). As shown in Fig. 1 (top panel), the Great Lakes (Fig. 1, bottom panel) are well resolved in the 36-km domain. The only other U.S. lakes that are represented in the domain are Great Salt Lake in Utah, Lake Okeechobee in Florida, and Lake of the Woods in northern Minnesota at the border with Canada. Several lakes in Canada are also identified by this land use classification, including Lake Winnipeg, Lake Winnipegosis, Lake Manitoba, and Lake Nipigon in south-central Canada. The same land use classification was used for both the Default and CLM Lakes simulations. In our configuration of WRF (which does not include a lake model), the distinction between lake and ocean points in the land use classification is benign; both are treated generically as water during the WRF simulation.
In WRFv3.6.1, WPS typically processes both skin (ground) temperature and sea surface temperature simultaneously by reading a skin temperature array (named “SKINTEMP” in WPS) and masking that array using the land-water mask from the input model. The WPS default method interpolates SKINTEMP from the input model data by matching land cells in the input dataset to land cells on the WRF domain, and water to water. Interpolation schemes are successively attempted using 16-point and then 4-point zones surrounding each WRF grid cell. If there are no valid data within these zones on the input data grid to interpolate to the WRF target grid cell, a “search” option is then employed in which data from the closest available land/water point on the input data grid are extrapolated to the WRF target grid cell. The combinations of the nine interpolation methods in WPS (see http://www2.mmm.ucar.edu/wrf/users/docs/user_guide_V3/users_guide_chap3.htm) can be tailored by the user, and input variables for WRF do not need to conform to the same interpolation techniques. However, all of these interpolation methods are reliant upon the availability of adequate driving data to generate realistic values for WRF. Several permutations of those methods, as well as attempts to blend land and water masks and isolate processing of sea surface temperature, were explored unsuccessfully prior to developing the CLM Lakes method used in this study.
Default and CLM Lakes used the same preprocessing through WPS, where an independent software program was developed to translate CESM data into the WPS intermediate format. For Default, SKINTEMP in WPS is filled from the radiative surface temperature (TS) from CAM. For CLM Lakes, lake temperature (TLAKE) from CLM is also made available to WPS to be interpolated to the WRF grid. In CLM Lakes, another software program was added between WPS and WRF to modify SKINTEMP at inland water points identified by the augmented USGS classification (described above) and instead use the interpolated TLAKE field. The SKINTEMP field from WPS becomes the “SST” field in the WRF lower boundary forcing file, and it includes both ground temperature and sea surface temperature. Default and CLM Lakes differ only in the construction of the SST field; all other initial conditions, boundary conditions, lower boundary forcing, nudged fields, and model options are identical.
3. Analysis
a. Input datasets
Figure 2a shows radiative surface temperature (TS) from CAM for the time labeled 0130 UTC 10 July 1995 with data only displayed for CAM water points from the land/ocean mask (“landmask”). The pixel sizes in Fig. 2a are filled at the resolution of the data available from CESM. As shown in Fig. 2a, the Great Lakes (as well as other large lakes in Canada) are not considered as water points in the CAM land mask. At CESM’s resolution, each of the Great Lakes is at least partially represented. With no valid data in TS in the vicinity of the Great Lakes, the Default method extrapolates temperatures from the closest water points in the Atlantic Ocean (to the east) and from James Bay and Hudson Bay (to the north) to set the water temperatures for the Great Lakes, as discussed in section 1.
(a) CAM TS radiative surface temperature (K) masked by CAM’s “landmask” so that TS is only shown over water (i.e., ocean) points, valid at the time labeled 0130 UTC 10 July 1995. (b) Monthly average lake temperatures (K) from CLM’s TLAKE at lake level 1, valid for the period labeled July 1995. Data are cropped for North America.
Citation: Journal of Climate 29, 2; 10.1175/JCLI-D-15-0233.1
Lake temperature (TLAKE) from CLM is shown in Fig. 2b for the period labeled July 1995 and only for CLM cells where there is a nonzero percentage of lake coverage. Some of the shaded cells shown in Fig. 2b also have partial land coverage in CLM. There is no overlap of CAM ocean cells (Fig. 2a) and CLM cells with lakes (Fig. 2b); they are mutually exclusive data fields. None of the lakes in the 36-km WRF domain (Fig. 1, top panel) has any valid water temperature data for interpolation in Fig. 2a, but all of the lakes in the 36-km domain are covered by data in Fig. 2b. Within CESM, the lake temperature data from CLM (Fig. 2b) are consistent with the evolution of the meteorological conditions simulated by CAM (Fig. 2a). As discussed in section 1, there can be very large differences between temperatures of the lakes and distant water cells from where data would be extrapolated by the WRF Default method. For example, there is more than a 15-K difference between the water temperatures in the Great Lakes (Fig. 2b) and those in the Atlantic Ocean off the coast of New Jersey (Fig. 2a) that are used in Default to set temperatures of the southern and eastern Great Lakes.
Figure 3 shows sea surface temperatures preprocessed for WRF using Default and using CLM Lakes for the times labeled 0000 UTC 15 July 1995 (Figs. 3a–c) and 0000 UTC 15 December 1995 (Figs. 3d–f). Using the Default method, the data in Fig. 3a are consistent with those shown in Fig. 2a for the ocean points during July. As discussed above, in Default, the lake temperatures in Fig. 3a can only be set using data shown in Fig. 2a. With no valid data in the vicinity of any of the inland water points on the WRF domain (see Fig. 1), the “search” option was used to extrapolate water temperatures from the closest water point with valid data. The lake temperatures for Lake Ontario, Lake Erie, southeastern Lake Huron, and southern Lake Michigan were identically set to the ocean temperature from the Atlantic Ocean off the New Jersey coast (to the east) at 297.6 K. The lake temperatures for most of Lake Huron, northern Lake Michigan, and most of Lake Superior are identical to the water temperature in southern James Bay (to the north) at 284.8 K, while western Lake Superior is set to 279.1 K from Hudson Bay, and most of the lakes in central Canada are also assigned temperatures from Hudson Bay. The temperature of Great Salt Lake was assigned a temperature from the Pacific Ocean off the California coast.
Input SST field (K) for WRF at 0000 UTC 15 July 1995 prepared using (a) Default and (b) CLM Lakes. (c) Difference between CLM Lakes and Default in instantaneous SST shown in (b) and (a). (d)–(f) As in (a)–(c), but for 0000 UTC 15 December 1995.
Citation: Journal of Climate 29, 2; 10.1175/JCLI-D-15-0233.1
In Default, using distant and often unrepresentative oceanic temperatures to set lake temperatures can result in nonphysical discontinuities within lakes. For example, in Fig. 3a, there are distinct temperature zones within Lakes Michigan and Huron that differ by nearly 13 K in July, reflecting the difference in water temperature between the mid-Atlantic Ocean and James Bay. A similar but less severe discontinuity of more than 5 K occurs in Lake Superior that delineates where water temperatures were assigned from James Bay and Hudson Bay. Using Default, the temperatures of inland water bodies are in persistent disequilibrium with the atmospheric conditions throughout the WRF simulation because this configuration of WRF (without a lake model) does not allow the water temperatures to respond to the atmospheric conditions. The water temperatures serve as lower boundary forcing throughout the simulation, so discontinuities in lake temperatures will not be overcome by using a spinup period of any length.
Figure 3b shows the SST field obtained by using the CLM Lakes method, which blends the data from Figs. 2a and 2b following the ocean and lake distinctions shown in Fig. 1. As in Fig. 3a, the oceanic temperatures in Fig. 3b are consistent with those shown in Fig. 2a. However, in Fig. 3b the temperatures of the lakes are set from CLM lake temperatures in Fig. 2b. The difference in SSTs set from CLM Lakes (Fig. 3b) and Default (Fig. 3a) is shown in Fig. 3c. Differences in SSTs between the two methods only occur at the inland water points. In the July case shown in Figs. 3a–c, the temperatures in the southern and eastern Great Lakes are more than 14 K colder in CLM Lakes than in Default, while western Lake Superior is nearly 5 K warmer in CLM Lakes than in Default. There is little difference in the water temperatures (0–2 K) between both runs at this time in the central Great Lakes, but CLM Lakes realistically allows for local variations of temperature that are not present in the Default, as can be seen in Fig. 3c in the differences between the two runs.
The differences between Default and CLM Lakes are accentuated in winter because of the sharper temperature gradient across the domain in winter, as shown for December (Figs. 3d–f). In Default, surface temperatures in the southern and eastern Great Lakes are set to 289.2 K, while the central Great Lakes are 270.8 K, creating within-lake discontinuities of 18.4 K in Lakes Michigan and Huron (Fig. 3d). Western Lake Superior is set to 265.6 K, creating a discontinuity of more than 5 K in that lake. By contrast, the temperatures of the Great Lakes in CLM Lakes (Fig. 3e) have gradual transitions spatially, with temperatures ranging between 260 and 270 K. As a result, the temperature differences at this wintertime (Fig. 3f) are as much as 23 K colder in CLM Lakes than in Default.
b. Comparisons against climatology
To examine the different data sources at a location over time, time series were constructed (Fig. 4) for buoy location 45004 in eastern Lake Superior (see Fig. 1) for the CESM period labeled 10 July–9 August 1995. For reference, observations at 45004 from 2000 are plotted to provide context for the water temperatures and their day-to-day changes at that location. (Recall that data from CESM cannot be strictly matched to historical observations, as previously discussed.) Because there are no water data near Lake Superior in CAM’s TS field (see Fig. 2a), the Default approach sets the water temperature at the WRF cell that includes 45004 to a temperature from James Bay (Fig. 4). In general, the temperature extrapolated from James Bay is warmer than observed temperatures during midsummer in eastern Lake Superior. Alternatively, the approach used by Gao et al. (2012) could have been applied by simply using TS from CAM and ignoring land-to-land and water-to-water matching. Using that approach (see Fig. 4), the mean skin temperature during this period would be much warmer than typical water temperatures, and unrealistic diurnal oscillations of 15–25 K would have been used for water temperatures. With CLM Lakes, there is a gradual increase in water temperature over time (see comparison to 284 K over the time series) that is broadly consistent with the observed trend. The lack of diurnal variability in the CLM Lakes lake temperature is because data were only made available as monthly values and were interpolated to 6-h intervals in our preprocessing to ensure smooth transitions in water temperature over time. Although there is a modest cold bias in CLM Lakes compared to the observations at this location and time of the year (Fig. 4), the temperatures applied in the CLM Lakes approach are consistent with the meteorological conditions in the driving data in CESM, and they are preferable to the alternate approaches to set water temperature. We caution that a comprehensive evaluation of the quality and representativeness of CLM-derived lake temperatures for the historical period is beyond the scope of this research study.
Time series of surface temperature (K) that could be used as a reference for Lake Superior buoy 45004 for 10 July–9 August; see Fig. 1 for buoy location. Data shown include CAM TS surface (ground) temperature (masked from Fig. 2a) from the cell that includes 45004 (black); CAM TS surface (water) temperature from the water cell in James Bay that is closest to 45004 (purple); CLM TLAKE lake temperature (shown in Fig. 2b) from the cell that includes 45004 (green), and observed water temperatures at 45004 (orange). CAM and CLM data are from the year labeled 1995, and observations are from 2000. The horizontal dashed gray line denotes 284 K.
Citation: Journal of Climate 29, 2; 10.1175/JCLI-D-15-0233.1
Figure 5 shows a comparison of the lake temperature data generated from Default and CLM Lakes to monthly climatology of water temperature observations at the nine buoy locations shown in Fig. 1. Note that buoy observations are not recorded when the lakes are frozen, as indicated by the reduction in observations over the climatological period during the boreal winter months. At the southern and eastern buoys (45005, 45007, and 45012), using the Default method consistently results in lake temperatures that are warmer than the 75th percentile monthly climatological observation throughout the year. At those locations in the winter months (December–February), the monthly average lake temperature in Default is often more than 10 K higher than the highest instantaneous value recorded during the climatological period. At those locations, CLM Lakes generally matches the climatological record better than Default, although there is a notable cool bias in summer in CLM compared to climatology. At the northern and western buoy locations (Fig. 5), Default and CLM Lakes are comparable to each other and reasonably match climatology in the summer and fall months. However, Default is markedly colder than CLM Lakes in winter and spring, owing to its extrapolation of water temperature data from water bodies to the north of these lakes.
Climatological distribution of observed water temperature (°C) recorded at Great Lakes buoys shown with their numbers in Fig. 1. Box plots show monthly median, interquartile range, and extremes of recorded temperatures, as well as the number of observations for each month over the climatological period. Climatological period is 1979–2013 (buoys 45001 and 45002), 1980–2013 (buoys 45003, 45004, and 45005), 1981–2013 (buoys 45006, 45007, and 45008), and 2002–13 (buoy 45012). Monthly average input sea surface temperature at the WRF point closest to each buoy is shown for Default (purple) and CLM Lakes (green) for CESM year 1995.
Citation: Journal of Climate 29, 2; 10.1175/JCLI-D-15-0233.1
Figure 5 can also be used to examine lake temperatures obtained from both methods within the same lake. Buoys 45002 and 45007 are sited in Lake Michigan approximately 300 km apart (see Fig. 1). The climatology at 45002 and 45007 is very similar (Fig. 5), except during January and February as the lake water freezes. As suggested by Figs. 2 and 3, Default sets the temperature at 45002 from James Bay, and the water temperature at 45007 originates in the Atlantic Ocean. The Default approach results in unrealistic differences in the annual evolutions of monthly lake temperature at WRF cells near the buoy locations within the same lake. For example, the average January lake temperature at 45007 is nearly 35 K warmer than at 45002 using Default. By contrast, the lake temperatures derived from CLM Lakes at those buoys have similar temporal evolutions, and the temperatures at 45007 (to the south) are only slightly warmer than those at 45002 (to the north) throughout the year. In addition, comparing the climatology at buoys 45003 and 45008 in Lake Huron (Fig. 5) indicates that there is local variation in the observed temperatures within that lake, even though the buoys are approximately 150 km apart (see Fig. 1). The Default approach applied identical lake temperatures to the WRF grid cells that correspond to both buoy locations in Lake Huron, while CLM Lakes enabled local variations in lake temperature, as also suggested by Figs. 3c and 3f. Regardless, using lake temperatures that are in equilibrium with the prevailing atmospheric conditions (as in CLM Lakes) is preferable to and more representative than using the Default method’s ad hoc extrapolation of lake temperatures from distant oceanic points.
c. Comparing WRF runs with Default and CLM Lakes
Figure 6 shows differences between CLM Lakes and Default in monthly average sea level pressure and average daily maximum 2-m temperature (T2MAXDAVG) for each July of the analysis period. Differences in monthly average sea level pressure in July (Figs. 6a–c) are subtle, usually 0.25 hPa or lower, but they are as high as 0.40 hPa. These differences are entirely attributable to the differences in lake temperatures and their evolutions between the two WRF simulations. Recall that spectral nudging was used so the evolution of the large-scale meteorology is broadly consistent in both WRF simulations; without spectral nudging, the divergence between the simulations is more widespread and of much larger magnitude (not shown). There is little visual consistency in the spatial patterns of the differences in July sea level pressure from year to year, although larger, colder differences in sea surface temperature (refer to Fig. 3c) result in greater increases in monthly average sea level pressure. For example, CLM Lakes water temperatures for southern Lake Michigan, Lake Erie, and Lake Ontario are often more than 10 K colder than Default, which generally is reflected as the maximum increase in sea level pressure (Figs. 6a–c).
Difference between CLM Lakes and Default (computed as CLM Lakes minus Default) for monthly average sea level pressure (hPa) for (a) July 1995, (b) July 1996, and (c) July 1997. (d)–(f) As in (a)–(c), but for monthly average daily maximum 2-m temperature (K). Color shading is applied for areas with significance at the 0.95 level from a paired Mann–Whitney (Wilcoxon) test.
Citation: Journal of Climate 29, 2; 10.1175/JCLI-D-15-0233.1
Differences in T2MAXDAVG are shown in Figs. 6d–f. During July CLM Lakes is 7–10 K colder than Default over the southern and eastern Great Lakes, and 4–6 K warmer over western Lake Superior. The largest differences in T2MAXDAVG are over the inland water bodies where the monthly difference is often greater than 1 K. There are also notable differences in T2MAXDAVG across the North American continent, and those differences occasionally exceed 1 K at grid cells that are distant from the lakes, such as along the Missouri–Iowa border and in central Nebraska in 1995; in southern Louisiana, along the Alabama–Georgia border, and in central Minnesota in 1996; and in northern Nebraska, central Missouri, and northern Florida in 1997. Differences in T2MAXDAVG between CLM Lakes and Default over oceanic points rarely exceed ±0.25 K (the smallest increment plotted in Figs. 6 and 7) during any month of the 3-yr analysis period; see isolated examples off the Newfoundland coast in Fig. 6d.
The differences in monthly average sea level pressure and T2MAXDAVG may seem random in Fig. 6, other than where they are local to the changes in lake temperatures. However, comparing Figs. 6a–c to Figs. 6d–f suggests that the change in lake temperatures between CLM Lakes and Default has affected the dynamics of the regional climate model throughout much of the WRF domain. Often areas in Figs. 6a–c that incurred larger monthly average sea level pressure differences between the runs align with areas that also incurred larger differences in T2MAXDAVG. Over the Great Lakes, where the signal is strongest because of the change to the underlying forcing from the lake temperatures (e.g., Figs. 3a–c), colder (warmer) lake temperatures in CLM Lakes are aligned with differences of the same sign in T2MAXDAVG (Figs. 6d–f), which typically resulted in local increases (decreases) of sea level pressure (Figs. 6a–c). The sign of the change in sea level pressure throughout the rest of the domain is often, but not always, consistent with the behavior over the lakes, possibly due to the more zonal weather patterns in July in this domain. There is spatial consistency between areas with maximum differences in sea level pressure and T2MAXDAVG beyond the lakes; compare, for example, Figs. 6c and 6f over northern Nebraska, central Missouri, and northern Florida, which were mentioned above as having differences in T2MAXDAVG of more than 1 K in July 1997.
Unlike in the summer, in the winter a consistent pattern of differences emerges between Default and CLM Lakes that repeats each year and extends to regions hundreds of kilometers from the lakes, as shown in Fig. 7 for each December of the analysis period. The Default lake temperatures in WRF create thermally induced perturbations in the simulated sea level pressure that are most prominent in winter over the lakes (Figs. 7a–c). These perturbations are most intense where the differences in lake temperature are greatest, such that lower (higher) sea level pressures occur where there are increases (decreases) in lake temperature (cf. Figs. 7a and 8a over the Great Lakes), as in summer. In Default, the monthly average lake temperatures in the southern Great Lakes are substantially warmer in winter (more than 15 K; see Fig. 8a) than are supported by the evolution of the meteorology, and they artificially create localized low pressure. Because of the size of the Great Lakes and because the lake temperatures cannot respond to the atmospheric conditions in the WRF default configuration, the disturbances in sea level pressure manifest as a wave that emanates from the Great Lakes and extends toward southern Florida (Figs. 7a–c). The amplitude of the wave is maximized where the differences in water temperature are maximized, such that the peak difference in monthly average sea level pressure is more than 2.0 hPa in December 1995, and more than 1.5 hPa and 1.7 hPa in each of the following two Decembers, respectively. The wave amplitude decreases as it moves outward from the Great Lakes, but it occurs annually with geographic consistency based on distance from the Great Lakes. The difference in lake temperature for Great Salt Lake between the two WRF simulations creates a similar wave in the western United States, which is more subtle because Great Salt Lake is smaller. Similar thermally induced waves in sea level pressure from snow cover were simulated by Sobolowski et al. (2007).
Difference between CLM Lakes and Default (computed as CLM Lakes minus Default) for December 1995 for (a) average sea surface temperature, (b) mean 2-m temperature, (c) 2nd percentile 2-m temperature, (d) 98th percentile 2-m temperature (all K), as well as number of (e) freeze days, (f) precipitation days, (g) days with 0.5 in. of precipitation, and (h) average 850-hPa wind speed (m s−1).
Citation: Journal of Climate 29, 2; 10.1175/JCLI-D-15-0233.1
The T2MAXDAVG values are impacted over and in the vicinity of the Great Lakes (Figs. 7d–f) during December because the Default lake surface temperatures in the southern Great Lakes are much warmer than supported by meteorology. By instead using lake temperatures that are in equilibrium with the meteorology in CLM Lakes, there are widespread decreases of more than 1.0 K in T2MAXDAVG in upstate New York, Michigan, and southern Ontario. In the colder two simulation years (1995 and 1997), decreases of more than 0.5 K extend through Pennsylvania and the southern New England states. More surprisingly, a band of increases in T2MAXDAVG occurs each December to the west and south of the Great Lakes, including Wisconsin, Illinois, Indiana, Ohio, West Virginia, and Virginia. These increases in T2MAXDAVG in CLM Lakes relative to Default are typically 0.25–0.50 K but are locally greater than 1.0 K. In addition, another area of decreases in T2MAXDAVG of comparable magnitude is focused over Georgia and South Carolina. The differences in T2MAXDAVG (Figs. 7d–f) align with the perturbation in sea level pressure (Figs. 7a–c). The undulations in sea level pressure and T2MAXDAVG (Fig. 7) are spatially aligned and reappear each winter, and are larger in magnitude during colder winters when there is a greater divergence between lake and atmospheric temperatures, particularly in the southern Great Lakes. The spatial pattern of differences in sea level pressure and T2MAXDAVG during December (Fig. 7) begins in November or December and persists through February or March (not shown), and each winter’s weather events and the timing of the lake freezes in CLM affect the onset, dissipation, and magnitude of the wave shown by the differences.
Differences in monthly mean 2-m temperature (T2MEAN) (Fig. 8b) are also influenced by the difference in lake temperatures (Fig. 8a) and are spatially aligned with differences in sea level pressure (Fig. 7a) and T2MAXDAVG (Fig. 7d), and those differences extend into the southern United States. Using CLM Lakes rather than Default results in decreases in T2MEAN of more than 2.5 K in the lee of Lakes Michigan, Erie, and Ontario, and more than 0.5 K decreases in T2MEAN are evident throughout the Great Lakes region and into the northeastern United States. Differences in T2MEAN are also seen in the shadows of Great Salt Lake in the western United States and of Lakes Nipigon, Winnipeg, Winnipegosis, and Manitoba in south-central Canada, and those changes are consistent with the differences in lake temperatures.
Incongruous lake temperatures also affect the distribution of the hourly 2-m temperatures, particularly in the vicinity of the Great Lakes. Using CLM Lakes rather than Default results in widespread increases of more than 5.0 K in monthly second-percentile temperature near Lake Superior, northern Lake Michigan, and Lake Huron, as well as near the south-central Canadian lakes (Fig. 8c). By contrast, CLM Lakes results in widespread decreases of the same magnitude in monthly 98th percentile temperature (Fig. 8d) and an overall compression of the distribution and reduction in variability of the 2-m air temperature. Comparing Figs. 8c and 8d indicates that the improper lake temperatures specified in Default artificially enhance the range of winter 2-m air temperatures (and, thus, the extremes) near the large lakes by more than 10 K. Default lake temperatures also squelch the number of freeze days (i.e., days with 2-m air temperatures less than 273.15 K) relative to CLM Lakes (Fig. 8e) in the lee of the southern Great Lakes because the lake temperatures are much too warm (Fig. 8a). CLM Lakes results in an increase in freeze days that is most pronounced near the Great Lakes, but also extends through much of the United States, except along the band where an increase in temperature (Figs. 7d and 8b) aligns with the sea level pressure wave (Fig. 7a) such that there are fewer freeze days.
Using Default lake temperatures artificially enhances the number of days of precipitation (Fig. 8f) and the number of days with precipitation exceeding 0.5 in. (DP05; Fig. 8g) downwind of the southern Great Lakes in winter. The lake temperatures in the southern Great Lakes are too warm in Default, so the lakes artificially create lower sea level pressure over the lakes and too readily generate lake-effect precipitation when cold winter air masses traverse the lakes. Under the prevailing meteorological conditions which are captured in CLM Lakes, parts of the southern Great Lakes would begin to freeze, thereby eliminating one physical mechanism in Default that adds precipitation in the lee of the Great Lakes. With CLM Lakes, a decrease of 3–7 precipitation days over the month relative to Default occurs not just in the lee of the Great Lakes, but also extends through the eastern Ohio Valley. A decrease of more than 4 days in DP05 occurs immediately downwind of the southern Great Lakes and is a direct result of reducing lake-effect precipitation.
The incongruous specification of the lake temperature is evident aloft, as there is a notable influence on the mesoscale circulation. The effect on monthly average 850-hPa wind speed (Fig. 8h) is concentrated where the differences between lake temperatures are greatest (Fig. 8a). By using CLM Lakes in WRF, lake temperatures are decreased, and 850-hPa wind speeds are increased by more than 1.0 m s−1 over southern portions of the Great Lakes. Decreases in 850-hPa wind speed of similar magnitude occur in land areas that are adjacent to and downwind from the lakes (i.e., Michigan, southern Ontario, upstate New York, and northern Vermont). These changes to the low-level convergence contribute to the artificial enhancement in precipitation in Default. In addition, subtle but coherent decreases in 850-hPa wind speed are simulated in Indiana and Virginia, which are spatially consistent with the differences in sea level pressure (Fig. 7a) and average temperatures (Figs. 7d and 8b).
4. Discussion
Improperly setting lake temperatures in WRF for downscaling adversely affects the resulting regional climate not just in the region surrounding the lakes, but also in areas far removed from the lakes. The effects are strongest when there is a greater difference between the default (incongruous) lake temperature setting and the improved method of defining lake temperatures–which occurs in winter—and they reappear each simulation year. As of WRFv3.6.1, several options are available to set lake temperatures such as employing a lake model within WRF (e.g., Gula and Peltier 2012; Mallard et al. 2014; Gu et al. 2015) or using a hysteresis based on 2-m air temperature (see Mallard et al. 2015). Using CLM-simulated lake temperatures instead of the default approach enabled the regional climate in WRF to evolve with physical forcing that is consistent with the atmospheric data from CAM.
Although CESM outputs were used here for demonstration, the absence of lake temperature data for downscaling is endemic to the CMIP5 archive. The CMIP5 standard output guidelines (Taylor 2013) indicate that daily mean sea surface temperatures are reported on the global model’s ocean grid. Because lakes, such as the Great Lakes, are not included on the ocean grid (Fig. 2a) and are typically modeled on a distinct lake grid in global models (Fig. 2b), water temperatures for those lakes are not reported with the high-frequency data provided for downscaling in the CMIP5 archive. Therefore, additional care is required for defining the temperatures of lakes that are unresolved on the global model’s ocean grid when using the CMIP5 archive for downscaling, particularly with the WRF Model, to ensure that lake temperatures are consistent with the driving atmospheric conditions. Omitting temperatures from the lake grid of the global models is an important oversight in the construction of the CMIP5 archive for downscaling—one that should be addressed if there are further coordinated modeling efforts and as long as lake models are not commonly using within regional climate models for downscaling. In particular, we suggest that lake temperature data (and other land surface fields) be made available for downscaling at the same frequency as the atmospheric data.
One of the advantages of dynamical downscaling is better representing mesoscale atmospheric processes at land–water interfaces. Properly setting the lake temperatures in WRF is essential for achieving high-quality results when downscaling, and using incongruous lake temperatures undermines this advantage. This paper clearly demonstrates that the default method of prescribing lake temperatures for downscaling with WRF over North America introduces systematic biases, especially in winter. These biases impair the simulation and interpretation of regional climate means and extremes, even at locations distant from the lakes.
Acknowledgments
CESM data were obtained from the Earth System Grid (http://www.earthsystemgrid.org). Buoy observations were obtained from the NOAA National Data Buoy Center (http://www.ndbc.noaa.gov). The WRF Model was obtained from the National Center for Atmospheric Research (http://www.wrf-model.org). Russ Bullock and Rohit Mathur from the United States Environmental Protection Agency (U.S. EPA) provided technical feedback on this paper. The critique of three anonymous reviewers served to strengthen this manuscript. The U.S. EPA through its Office of Research and Development funded and managed the research described here. It has been subjected to the Agency’s administrative review and approved for publication.
REFERENCES
Arnbjerg-Nielsen, K., P. Williams, J. Olsson, S. Beecham, A. Pathirana, I. Bülow Gregersen, H. Madsen, and V.-T.-V. Nguyen, 2013: Impacts of climate change on rainfall extremes and urban drainage systems: A review. Water Sci. Technol., 68, 16–28, doi:10.2166/wst.2013.251.
Bullock, O. R., Jr., K. Alapaty, J. A. Herwehe, M. S. Mallard, T. L. Otte, R. C. Gilliam, and C. G. Nolte, 2014: An observation-based investigation of nudging in WRF for downscaling surface climate information to 12-km grid spacing. J. Appl. Meteor. Climatol, 53, 20–33, doi:10.1175/JAMC-D-13-030.1.
Caldwell, P., H.-N. S. Chin, D. C. Bader, and G. Bala, 2009: Evaluation of a WRF dynamical downscaling simulation over California. Climatic Change, 95, 499–521, doi:10.1007/s10584-009-9583-5.
Chen, F., and J. Dudhia, 2001: Coupling an advanced land surface–hydrology model with the Penn State–NCAR MM5 modeling system. Part I: Model implementation and sensitivity. Mon. Wea. Rev., 129, 569–585, doi:10.1175/1520-0493(2001)129<0569:CAALSH>2.0.CO;2.
Dickinson, R. E., R. M. Errico, F. Giorgi, and G. T. Bates, 1989: A regional climate model for the western United States. Climatic Change, 15, 383–422, doi:10.1007/BF00240465.
Fann, N., C. G. Nolte, P. Dolwick, T. L. Spero, A. Curry Brown, S. Phillips, and S. Anenberg, 2015: The geographic distribution and economic value of climate change-related ozone health impacts in the United States at 2030. J. Air Waste Manage. Assoc., 65, 570–580, doi:10.1080/10962247.2014.996270.
Feser, F., B. Rockel, H. von Storch, J. Winterfeldt, and M. Zahn, 2011: Regional climate models add value to global model data. Bull. Amer. Meteor. Soc., 92, 1181–1192, doi:10.1175/2011BAMS3061.1.
Gao, Y., J. S. Fu, J. B. Drake, Y. Liu, and J.-F. Lamarque, 2012: Projected changes of extreme weather events in the eastern United States based on a high resolution climate modeling system. Environ. Res. Lett., 7, 044025, doi:10.1088/1748-9326/7/4/044025.
Gent, P. R., and Coauthors, 2011: The Community Climate System Model version 4. J. Climate, 24, 4973–4991, doi:10.1175/2011JCLI4083.1.
Giorgi, F., 1990: Simulation of regional climate using a limited area model nested in a general circulation model. J. Climate, 3, 941–963, doi:10.1175/1520-0442(1990)003<0941:SORCUA>2.0.CO;2.
Gu, H., J. Jin, Y. Wu, M. Ek, and Z. Subin, 2015: Calibration and validation of lake surface temperature simulations with the coupled WRF-lake model. Climatic Change, 129, 471–483, doi:10.1007/s10584-013-0978-y.
Gula, J., and W. R. Peltier, 2012: Dynamical downscaling over the Great Lakes basin of North America using the WRF regional climate model: The impact of the Great Lakes system on regional greenhouse warming. J. Climate, 25, 7723–7742, doi:10.1175/JCLI-D-11-00388.1.
Harkey, M., and T. Holloway, 2013: Constrained dynamical downscaling for assessment of climate impacts. J. Geophys. Res. Atmos., 118, 2136–2148, doi:10.1002/jgrd.50223.
Herwehe, J. A., K. Alapaty, T. L. Spero, and C. G. Nolte, 2014: Increasing the credibility of regional climate simulations by introducing subgrid-scale cloud–radiation interactions. J. Geophys. Res. Atmos., 119, 5317–5330, doi:10.1002/2014JD021504.
Hong, S.-Y., and J.-O. J. Lim, 2006: The WRF single-moment 6-class microphysics scheme (WSM6). J. Korean Meteor. Soc, 42, 129–151.
Hong, S.-Y., Y. Noh, and J. Dudhia, 2006: A new vertical diffusion package with an explicit treatment of entrainment processes. Mon. Wea. Rev., 134, 2318–2341, doi:10.1175/MWR3199.1.
Iacono, M. J., J. S. Delamere, E. J. Mlawer, M. W. Shephard, S. A. Clough, and W. D. Collins, 2008: Radiative forcing by long-lived greenhouse gases: Calculations with the AER radiative transfer models. J. Geophys. Res., 113, D13103, doi:10.1029/2008JD009944.
Kain, J. S., 2004: The Kain–Fritsch convective parameterization: An update. J. Appl. Meteor., 43, 170–181, doi:10.1175/1520-0450(2004)043<0170:TKCPAU>2.0.CO;2.
Mallard, M. S., C. G. Nolte, O. R. Bullock, T. L. Spero, and J. Gula, 2014: Using a coupled lake model with WRF for dynamical downscaling. J. Geophys. Res. Atmos., 119, 7193–7208, doi:10.1002/2014JD021785.
Mallard, M. S., C. G. Nolte, T. L. Spero, O. R. Bullock, K. Alapaty, J. A. Herwehe, J. Gula, and J. H. Bowden, 2015: Technical challenges and solutions representing lakes when using WRF in downscaling applications. Geosci. Model Dev., 8, 1085–1096, doi:10.5194/gmd-8-1085-2015.
Mearns, L. O., and Coauthors, 2012: The North American Regional Climate Change Assessment Program: Overview of phase I results. Bull. Amer. Meteor. Soc., 93, 1337–1362, doi:10.1175/BAMS-D-11-00223.1.
Miguez-Macho, G., G. L. Stenchikov, and A. Robock, 2004: Spectral nudging to eliminate the effects of domain position and geometry in regional climate model simulations. J. Geophys. Res., 109, D13104, doi:10.1029/2003JD004495.
Nolte, C. G., A. B. Gilliland, C. Hogrefe, and L. J. Mickley, 2008: Linking global to regional models to assess future climate impacts on surface ozone levels in the United States. J. Geophys. Res., 113, D14307, doi:10.1029/2007JD008497.
Notaro, M., K. Holman, A. Zarrin, E. Fluck, S. Vavrus, and V. Bennington, 2013: Influence of the Laurentian Great Lakes on regional climate. J. Climate, 26, 789–804, doi:10.1175/JCLI-D-12-00140.1.
Otte, T. L., C. G. Nolte, M. J. Otte, and J. H. Bowden, 2012: Does nudging squelch the extremes in regional climate modeling? J. Climate, 25, 7046–7066, doi:10.1175/JCLI-D-12-00048.1.
Racherla, P. N., D. T. Shindell, and G. S. Faluvegi, 2012: The added value to global model projections of climate change by dynamical downscaling: A case study over the continental U.S. using the GISS-ModelE2 and WRF models. J. Geophys. Res., 117, D20118, doi:10.1029/2012JD018091.
Salathé, E. P., Jr., L. R. Leung, Y. Qian, and Y. Zhang, 2010: Regional climate model projections for the state of Washington. Climatic Change, 102, 51–75, doi:10.1007/s10584-010-9849-y.
Skamarock, W. C., and J. B. Klemp, 2008: A time-split nonhydrostatic atmospheric model for weather research and forecasting applications. J. Comput. Phys., 227, 3465–3485, doi:10.1016/j.jcp.2007.01.037.
Sobolowski, S., G. Gong, and M. Ting, 2007: Northern Hemisphere winter climate variability: Response to North American snow cover anomalies and orography. Geophys. Res. Lett., 34, L16825, doi:10.1029/2007GL030573.
Sun, X., L. Xie, F. Semazzi, and B. Liu, 2015: Effect of lake surface temperature on the spatial distribution and intensity of the precipitation over the Lake Victoria basin. Mon. Wea. Rev., 143, 1179–1192, doi:10.1175/MWR-D-14-00049.1.
Taylor, K., 2013: CMIP5 standard output. Accessed 2 Feb 2015. [Available online at http://cmip-pcmdi.llnl.gov/cmip5/docs/standard_output.pdf.]
Trail, M., A. P. Tsimpidi, P. Liu, K. Tsigaridis, Y. Hu, A. Nenes, and A. G. Russell, 2013: Downscaling a global climate model to simulate climate change over the US and the implication on regional and urban air quality. Geosci. Model Dev., 6, 1429–1445, doi:10.5194/gmd-6-1429-2013.
Vautard, R., and Coauthors, 2013: The simulation of European heat waves from an ensemble of regional climate models within the EURO-CORDEX project. Climate Dyn., 41, 2555–2575, doi:10.1007/s00382-013-1714-z.
Wang, S.-Y., R. R. Gillies, J. Jin, and L. E. Hipps, 2010: Coherence between the Great Salt Lake level and the Pacific quasi-decadal oscillation. J. Climate, 23, 2161–2177, doi:10.1175/2009JCLI2979.1.
