Abstract

The authors analyze the ability of global climate models (GCMs) from phase 5 of the Coupled Model Intercomparison Project (CMIP5) multimodel ensemble to simulate very heavy daily precipitation and its supporting processes, comparing them with observations. Their analysis focuses on an upper Mississippi region for winter (December–February), when it is assumed that resolved synoptic circulation governs precipitation. CMIP5 GCMs generally reproduce well the precipitation versus intensity spectrum seen in observations to intensities as strong as 20 mm day−1. Most models do not produce the highest precipitation intensities seen in observations. Models show good agreement at the 95th percentile, while the coarsest resolution models generally show lower precipitation at high-intensity thresholds, such as the 99.5th percentile. There is no dominant month for simulated very heavy events to occur, although observed very heavy events occur most frequently in December. Further analysis focuses on precipitation events exceeding the 99.5th percentile that occur simultaneously at several points in the region, yielding so-called “widespread events.” Examination of additional fields during widespread very heavy events shows that the models produce these events under the same physical conditions seen in the observations. The coarsest models generally produce similar behavior, although features have smoother spatial distributions. However, the resolution in itself could not be identified as a major reason that separates one model from another. The capabilities of the CMIP5 GCMs examined here support using them to assess changes in very heavy precipitation under future climate scenarios.

1. Introduction

With enhancements in climate models' ability to simulate past and future climate, one topic that has gained attention is very heavy events accompanying global climate change. Increased variability in winds, temperature, and precipitation, among others, are of great interest to both the scientific community and the general public because of the social and economic impacts these events can cause. To validate these climate models, simulations need to be compared with observational data to determine if physical behaviors causing these events in models are similar to those in the real world. By using projections based on validated models, one can make analyses and decisions concerning future climate change with greater confidence.

Here we analyze very heavy daily precipitation events, as defined by Groisman et al. (2005), during the winter months in the upper Mississippi region. We use climate simulations produced by 21 global climate models (GCMs) for phase 5 of the Coupled Model Intercomparison Project (CMIP5) multimodel ensemble (Taylor et al. 2012). A major portion of this paper is motivated by previous work by Kawazoe and Gutowski (2013), who focused on very heavy winter precipitation in the same region, but by regional climate models (RCMs) from the North American Regional Climate Change Assessment Program (NARCCAP). The goals of this study are to assess the ability of the CMIP5 models collectively to reproduce very heavy daily precipitation in observations, to produce very heavy precipitation for the same physical conditions as in observations, and to provide a baseline for understanding how very heavy daily precipitation and its causal processes change under enhanced greenhouse warming scenarios. We note, however, that while this capability is a necessary condition for using the models to assess climate change for very heavy precipitation events, assessment of changes would have to assume that these models capture enhanced greenhouse gas scenarios appropriately.

2. Observations, simulations, and analysis methods

a. Observations

The analysis uses the University of Washington's (UW) gridded precipitation (Maurer et al. 2002) as the primary observational data. This dataset provides observation-based precipitation on a 0.125° grid that covers all of the contiguous United States. Interpolation for this gridded dataset used the scheme of Shepard (1984) as implemented in Widmann and Bretherton (2000). The dataset also uses corrections for systematic elevation effects given by the Parameter-Elevation Regressions on Independent Slopes Model (PRISM; Daly et al. 1994). The dataset in the Network Common Data Form (NetCDF) format covers the period 1950–99.

We use the UW data output as the basis for identifying days when very heavy precipitation occurs. For all other fields in the observational analysis, we used the North American Regional Reanalysis (NARR; Mesinger et al. 2006). The fields we use include 500-hPa geopotential heights, 2-m air temperature, 2-m specific humidity, and 10-m horizontal winds. These fields represent key environmental conditions during the development of very heavy precipitation events and are also common to the output archives for most of the models examined here.

b. Simulations

Model output comes from 21 global climate models that simulated the historical period 1850–2005 for CMIP5 (Table 1; Taylor et al. 2012). Analyses of all models are from the historical experiment using the r1i1p1 ensemble member. The GCMs used in this analysis are models available in archives on 1 June 2012, with an emphasis on models that provided daily precipitation, 500-hPa geopotential heights, 2-m air temperature, 2-m specific humidity, and 10-m horizontal winds.

Table 1.

CMIP5 GCMs analyzed in this paper.

CMIP5 GCMs analyzed in this paper.
CMIP5 GCMs analyzed in this paper.

c. Analysis

We analyzed the period 1980–99, consistent with available UW precipitation data. Our region of interest is the upper Mississippi region, designated here as the region bounded by (37°–47°N, 89°–99°W) and highlighted in Fig. 1. Resolution of each model within this region is listed on Table 2. This is the same region used in some of our previous precipitation analyses (Gutowski et al. 2008, 2010; Kawazoe and Gutowski 2013). Our analysis focuses on the winter season [December–February (DJF)], when synoptic dynamics are more important than in the warmer months, when smaller-scale convective events may be more important (e.g., Schumacher and Johnson 2005, 2006). This assumption here is that resolved circulation governs winter events, so that the other model fields we analyze are directly relevant to understanding the physical behavior of very heavy events (e.g., Gutowski et al. 2008).

Fig. 1.

Region covered by each CMIP5 model, UW, and NARR. The upper Mississippi analysis region is in the boxed area.

Fig. 1.

Region covered by each CMIP5 model, UW, and NARR. The upper Mississippi analysis region is in the boxed area.

Table 2.

Approximate resolution and nominal area for a model's grid box in the upper Mississippi domain in terms of a 0.5° × 0.5° grid box. UW/NARR resolution is for precipitation. All other fields use NARR's 0.3° × 0.3° resolution.

Approximate resolution and nominal area for a model's grid box in the upper Mississippi domain in terms of a 0.5° × 0.5° grid box. UW/NARR resolution is for precipitation. All other fields use NARR's 0.3° × 0.3° resolution.
Approximate resolution and nominal area for a model's grid box in the upper Mississippi domain in terms of a 0.5° × 0.5° grid box. UW/NARR resolution is for precipitation. All other fields use NARR's 0.3° × 0.3° resolution.

We converted the original UW output to a 0.5° grid by averaging all original grid points that fell in a 0.5° box centered on the new grid point. We did this to give the dataset the same nominal resolution as the NARR and the highest-resolution GCM.

CMIP5 models are in daily increments from 0000 to 0000 UTC (1800–1800 local standard time in the upper Mississippi region). The UW dataset is in daily increments from 0600 to 0600 UTC (0000–0000 local standard time in the upper Mississippi region), a factor that may affect some of our results. The analysis examining conditions other than precipitation during very heavy events used the 0000 UTC fields at the start of the day, which provided information on the initial state of the atmosphere.

We defined a precipitation event as precipitation above 0.25 mm day−1 recorded for 1 day at one observational or model grid point, which differs with Kawazoe and Gutowski (2013), who defined a precipitation event as all nonzero records. We extracted the top 0.5% of all precipitation events as very heavy daily events. From these events, we then found widespread very heavy events by searching for multiple very heavy events occurring on the same day. The GCMs have a range of different resolutions. To determine widespread very heavy events, we adjusted all GCM output to equivalent 0.5° grid boxes to match the grid we use for the UW data Thus, a 2.5° × 3.75° grid box in the HadCM3 covers roughly (2.5/0.5) × (3.75/0.5) = 37.5 times more area than a 0.5° grid box, so one HadCM3 grid box is the equivalent of 37.5 grid boxes at 0.5°. The nominal area equivalent to 0.5° for the other GCMs is listed on Table 2. For our analysis, we designated simultaneous very heavy events on 15 or more equivalent 0.5° grid boxes as widespread events. Note that for the coarsest models, a widespread event can occur with just one grid box having a very heavy event.

We examined several atmospheric fields, listed earlier, to understand conditions conducive to very heavy events. These fields give insight into the preferred conditions for very heavy precipitation events and become the basis for assessing simulated versus observed processes yielding very heavy precipitation. The 10-m winds were used as our primary indicator of moisture flux. Although it is not perfectly synonymous with moisture flux direction and convergence, it is a low-level circulation field available from all the models. For 500-hPa geopotential heights, 2-m air temperature, and 2-m specific humidity, we examined anomalies. These anomalies are composites of fields on the days of widespread very heavy events minus the 20-yr time average during the winter season. We compute time averages separately for each model and for the observations. To gauge the magnitude of the anomalies, we also computed the 2–5-day variability of the same fields throughout the analysis period, applying to daily time series a Lanczos filter with nine weights and a cutoff frequency of 5 days.

3. Widespread very heavy precipitation

Table 3 shows the average precipitation rate and frequency of daily precipitation events in the upper Mississippi region for the observations and for each model. The numbers in parentheses are the percentage of days with precipitation above 2.5 mm day−1. Other than NorESM1-M and MIROC-ESM-CHEM, the models produce too much precipitation. Other than GFDL CM3, BCC-CSM1.1, and BNU-ESM, the models also produce fewer days with precipitation than observed. Similar to Kawazoe and Gutowski (2013), days with precipitation above 2.5 mm day−1 agree well between observations and models. This shows that fewer precipitation events below 2.5 mm day−1 occurred in the models than observations, indicating that CMIP5 GCMs produce fewer “drizzle” events than observed, in contrast to the NARCCAP RCMs in Kawazoe and Gutowski (2013). Recall, however, that the definition of a precipitation event in Kawazoe and Gutowski (2013) is any nonzero precipitation, whereas here we count only events with precipitation exceeding 0.25 mm day−1. This difference may account for the different frequency of “drizzle” events between the two studies. The spreads across models in average precipitation rate and days with precipitation do not indicate that resolution in itself is an important factor for differences with observations.

Table 3.

Properties of CMIP models and UW: overall average precipitation rate and percentage of days reporting precipitation (numbers in parentheses are the percentage of days with precipitation exceeding 2.5 mm day−1).

Properties of CMIP models and UW: overall average precipitation rate and percentage of days reporting precipitation (numbers in parentheses are the percentage of days with precipitation exceeding 2.5 mm day−1).
Properties of CMIP models and UW: overall average precipitation rate and percentage of days reporting precipitation (numbers in parentheses are the percentage of days with precipitation exceeding 2.5 mm day−1).

Figures 2 and 3 show histograms of normalized frequency versus intensity in the upper Mississippi region using 2.5 mm day−1 bins. Figure 2 contains models with all supporting environmental fields, and Fig. 3 has additional models that did not have all supporting fields, so they are used for the precipitation analysis only. Observations and the models are in relatively good agreement up to around 20 mm day−1. Other than the BNU-ESM, NorESM1-M, and MIROC-ESM-CHEM, the models show a higher frequency of precipitation in bins greater than 20 mm day−1 compared to observations, while not producing events at the highest intensity spectrum. MIROC4h agrees well with observations over the whole intensity spectrum, perhaps because it has a resolution similar to the grid for observations.

Fig. 2.

Normalized frequency of precipitation as a function of daily intensity for 1980–99 in models and observations that provided all analyzed supporting fields. Arrows mark the 99.5th percentile; black is UW and blue are GCMs.

Fig. 2.

Normalized frequency of precipitation as a function of daily intensity for 1980–99 in models and observations that provided all analyzed supporting fields. Arrows mark the 99.5th percentile; black is UW and blue are GCMs.

Fig. 3.

As in Fig. 2, but for observations and for additional models that did not provide all analyzed supporting fields.

Fig. 3.

As in Fig. 2, but for observations and for additional models that did not provide all analyzed supporting fields.

Table 4 shows precipitation for each model and for the observations at the 95th, 99th, and 99.5th percentiles. The models and observations show fairly good agreement at the 95th percentile. At higher percentiles, finer resolution models have very heavy events that tend to be greater than observations, with MPI-ESM-LR being a slight outlier. Excluding the CanESM2, the coarsest-resolution GCMs show precipitation lower than observations at all percentiles. This suggests that the coarsest-resolution models do not replicate intense, small-scale circulation features that are necessary for producing very heavy events.

Table 4.

Precipitation intensity for models and observations at the 95th, 99th, and 99.5th percentiles for all nonzero precipitation.

Precipitation intensity for models and observations at the 95th, 99th, and 99.5th percentiles for all nonzero precipitation.
Precipitation intensity for models and observations at the 95th, 99th, and 99.5th percentiles for all nonzero precipitation.

Figures 4 and 5 show the distribution of days with simultaneous very heavy events on a given number of grid boxes. Figure 4 contains models with all supporting environmental fields, and Fig. 5 has additional models that did not have all supporting fields, so they are used for the precipitation analysis only. The x axis indicates the minimum area of a multigrid-point event, thus suggesting its spatial scale. The models tend to produce very heavy events covering a larger area than the observations. MIROC4h has approximately the same resolution as the observational dataset and has a similar pattern in Fig. 5.

Fig. 4.

Days with simultaneous very heavy events on at least the given number of grid points for observations and models that provided all analyzed supporting fields.

Fig. 4.

Days with simultaneous very heavy events on at least the given number of grid points for observations and models that provided all analyzed supporting fields.

Fig. 5.

Days with simultaneous very heavy events on at least the given number of grid points for observations and models that did not provide all analyzed supporting fields.

Fig. 5.

Days with simultaneous very heavy events on at least the given number of grid points for observations and models that did not provide all analyzed supporting fields.

Further analysis focuses on very heavy events occurring on at least 15 equivalent 0.5° grid boxes on the same day. We denote these as widespread very heavy events. For at least the higher-resolution models, these events are more likely to be the outcome of resolved behavior.

Table 5 shows the distribution of widespread very heavy events by winter months. There does not seem to be a dominant month for widespread very heavy events. This contrasts with the NARCCAP RCMs, driven by the National Centers for Environmental Prediction–Department of Energy (NCEP–DOE) reanalysis (Kanamitsu et al. 2002), for which five of the six had maximum frequency in December (Kawazoe and Gutowski 2013), in agreement with observations. Here only nine of the 21 models have the highest frequency of very heavy events occurring in December. The speculation for the NARCCAP models was that warmer Gulf of Mexico sea surface temperatures (SSTs) in December promote higher humidity over the Gulf and thus more atmospheric moisture for transport in the upper Mississippi region (Kunkel et al. 2002; Kawazoe and Gutowski 2013). Such a climatological control does not appear to operate here. We also examined monthly changes in large-scale baroclinicity, using temperature differences between the Gulf of Mexico and the upper Mississippi region, and found no systematic relationship with the occurrence of our very heavy events. We do find, however, that for each model, the average Gulf of Mexico SST during our widespread very heavy events is warmer than the model's climatological SST for each DJF month, usually by more than 1.5°C (not shown). Thus, warmer Gulf temperatures do promote very heavy events in these GCMs, but not for a particular month.

Table 5.

Percentage of widespread very heavy events by month for observations and for each model. Highest values during the season are in bold. GCM averages are 37.7% for December, 27.8% for January, and 34.5% for February.

Percentage of widespread very heavy events by month for observations and for each model. Highest values during the season are in bold. GCM averages are 37.7% for December, 27.8% for January, and 34.5% for February.
Percentage of widespread very heavy events by month for observations and for each model. Highest values during the season are in bold. GCM averages are 37.7% for December, 27.8% for January, and 34.5% for February.

Figure 6 shows composite precipitation during widespread very heavy events. Composite fields are from models that provided all supporting environmental fields for this analysis, that is, models used for Figs. 2 and 4. Models and observations show similar locations of very heavy precipitation, centered near the southeastern corner of our analysis region. Our analysis region in winter is warmest to the south. The warmer air can have more precipitable water, so the composite very heavy precipitation occurs where there will generally be more moisture in the atmosphere. Also, the southern end of the analysis region is closest to the primary source of the region's precipitable water, the Gulf of Mexico. This behavior is consistent with NARCCAP RCM analysis in Kawazoe and Gutowski (2013). Precipitation intensity by the CMIP5 GCMs agrees well with the NARCCAP RCMs in Kawazoe and Gutowski (2013), though the events in the CMIP5 GCMs typically cover a broader area and thus show a smoother composite.

Fig. 6.

Composite daily precipitation (mm day−1) during widespread very heavy events: (a) UW, (b) MRI-CGCM3, (c) MIROC5, (d) CNRM-CM5, (e) HadGEM2-CC, (f) IPSL-CM5A-MR, (g) FGOALS-s2, (h) GFDL CM3, (i) GFDL-ESM2M, (j) IPSL-CM5A-LR, (k) BCC-CSM1.1, (l) CanESM2, (m) MIROC-ESM-CHEM, and (n) BNU-ESM. Contour scale for all plots is in the upper right.

Fig. 6.

Composite daily precipitation (mm day−1) during widespread very heavy events: (a) UW, (b) MRI-CGCM3, (c) MIROC5, (d) CNRM-CM5, (e) HadGEM2-CC, (f) IPSL-CM5A-MR, (g) FGOALS-s2, (h) GFDL CM3, (i) GFDL-ESM2M, (j) IPSL-CM5A-LR, (k) BCC-CSM1.1, (l) CanESM2, (m) MIROC-ESM-CHEM, and (n) BNU-ESM. Contour scale for all plots is in the upper right.

4. Supporting environmental conditions

Figures 710 show composite fields produced by averaging over the widespread event days from each data source. Again, the anomaly fields for a given source come from subtracting the 20-yr DJF average from the composite. The NARR provided the observational results, with the days to composite determined from analysis of the UW precipitation. Composite fields are from models that provided all supporting environmental fields for this analysis. Inspection of individual events shows that the composites for each field are representative of the behavior of individual events.

Fig. 7.

Composite 500-hPa heights (m) during widespread very heavy events: (a) NARR, (b) MRI-CGCM3, (c) MIROC5, (d) CNRM-CM5, (e) HadGEM2-CC, (f) IPSL-CM5A-MR, (g) FGOALS-s2, (h) GFDL CM3, (i) GFDL-ESM2M, (j) IPSL-CM5A-LR, (k) BCC-CSM1.1, (l) CanESM2, (m) MIROC-ESM-CHEM, and (n) BNU-ESM. Contour scale for all plots is in the upper right.

Fig. 7.

Composite 500-hPa heights (m) during widespread very heavy events: (a) NARR, (b) MRI-CGCM3, (c) MIROC5, (d) CNRM-CM5, (e) HadGEM2-CC, (f) IPSL-CM5A-MR, (g) FGOALS-s2, (h) GFDL CM3, (i) GFDL-ESM2M, (j) IPSL-CM5A-LR, (k) BCC-CSM1.1, (l) CanESM2, (m) MIROC-ESM-CHEM, and (n) BNU-ESM. Contour scale for all plots is in the upper right.

Fig. 8.

Composite 10-m horizontal winds (m s−1) during widespread very heavy events: (a) NARR, (b) MRI-CGCM3, (c) MIROC5, (d) CNRM-CM5, (e) HadGEM2-CC, (f) IPSL-CM5A-MR, (g) FGOALS-s2, (h) GFDL CM3, (i) GFDL-ESM2M, (j) IPSL-CM5A-LR, (k) BCC-CSM1.1, (l) CanESM2, (m) MIROC-ESM-CHEM, and (n) BNU-ESM. Wind vector for all plots is in the upper right.

Fig. 8.

Composite 10-m horizontal winds (m s−1) during widespread very heavy events: (a) NARR, (b) MRI-CGCM3, (c) MIROC5, (d) CNRM-CM5, (e) HadGEM2-CC, (f) IPSL-CM5A-MR, (g) FGOALS-s2, (h) GFDL CM3, (i) GFDL-ESM2M, (j) IPSL-CM5A-LR, (k) BCC-CSM1.1, (l) CanESM2, (m) MIROC-ESM-CHEM, and (n) BNU-ESM. Wind vector for all plots is in the upper right.

Fig. 9.

Composite 2-m temperature anomalies (K) during widespread very heavy events: (a) NARR, (b) MRI-CGCM3, (c) MIROC5, (d) CNRM-CM5, (e) HadGEM2-CC, (f) IPSL-CM5A-MR, (g) FGOALS-s2, (h) GFDL CM3, (i) GFDL-ESM2M, (j) IPSL-CM5A-LR, (k) BCC-CSM1.1, (l) CanESM2, (m) MIROC-ESM-CHEM, and (n) BNU-ESM. Contour scale for all plots is in the upper right.

Fig. 9.

Composite 2-m temperature anomalies (K) during widespread very heavy events: (a) NARR, (b) MRI-CGCM3, (c) MIROC5, (d) CNRM-CM5, (e) HadGEM2-CC, (f) IPSL-CM5A-MR, (g) FGOALS-s2, (h) GFDL CM3, (i) GFDL-ESM2M, (j) IPSL-CM5A-LR, (k) BCC-CSM1.1, (l) CanESM2, (m) MIROC-ESM-CHEM, and (n) BNU-ESM. Contour scale for all plots is in the upper right.

Fig. 10.

Composite 2-m specific humidity anomalies (kg kg−1) during widespread very heavy events: (a) NARR, (b) MRI-CGCM3, (c) MIROC5, (d) CNRM-CM5, (e) HadGEM2-CC, (f) IPSL-CM5A-MR, (g) FGOALS-s2, (h) GFDL CM3, (i) GFDL-ESM2M, (j) IPSL-CM5A-LR, (k) BCC-CSM1.1, (l) CanESM2, (m) MIROC-ESM-CHEM, and (n) BNU-ESM. Contour scale for all plots is in the upper right.

Fig. 10.

Composite 2-m specific humidity anomalies (kg kg−1) during widespread very heavy events: (a) NARR, (b) MRI-CGCM3, (c) MIROC5, (d) CNRM-CM5, (e) HadGEM2-CC, (f) IPSL-CM5A-MR, (g) FGOALS-s2, (h) GFDL CM3, (i) GFDL-ESM2M, (j) IPSL-CM5A-LR, (k) BCC-CSM1.1, (l) CanESM2, (m) MIROC-ESM-CHEM, and (n) BNU-ESM. Contour scale for all plots is in the upper right.

a. 500-hPa geopotential heights

As suggested by Fig. 7, a key ingredient for very heavy precipitation in the upper Mississippi region is the transport of warm, moist air from the Gulf of Mexico. Composite 500-hPa heights show very heavy events occurring when a deep trough develops around the southern Rockies, promoting a more pronounced southerly flow into the region when compared with the seasonal climatology. The presence of lower heights to the west and higher heights to the east of the analysis region highlighted in Kawazoe and Gutowski (2013) is also evident in the composite plots shown in Fig. 7. The magnitudes of the largest anomalies were roughly 5 times greater than the 2–5-day variability in 500-hPa heights for the same locations (not shown). Resolution does not seem to affect the depth or location of the trough.

b. 10-m horizontal wind

Figure 8 shows the composite 10-m winds for widespread very heavy events. As with 500-hPa heights, the composites are representative of the behavior of individual events. As discussed earlier, the winds indicate the direction of moisture transport and also the location of surface pressure centers, although these winds are not perfectly synonymous with the moisture flux direction and convergence, as discussed earlier.

During the widespread very heavy events, winds turn counterclockwise to the west of the area of very heavy precipitation. The behavior corresponds to a surface low in the vicinity of Oklahoma accompanying the 500-hPa trough. This was seen in Kawazoe and Gutowski (2013) and Wendland et al. (1983), who focused on higher than average precipitation during the 1982/83 winter. In addition, the behavior shows low-level convergence. Because relatively strong winds blow from the Gulf of Mexico, the momentum convergence likely coincides with the moisture convergence, especially in the vicinity of the very heavy precipitation. Momentum convergence of 10-m winds at the 99th percentile during widespread very heavy events is shown in Table 6. There is no evident correlation between precipitation intensities and momentum convergence, nor are these values substantially different from corresponding quantities for all days in the analysis period (not shown). Table 6 also does not show momentum convergence varying with resolution.

Table 6.

The 99th percentile values of surface air temperature and specific humidity gradients and horizontal 10-m wind convergence on very heavy event days for observations and for each model.

The 99th percentile values of surface air temperature and specific humidity gradients and horizontal 10-m wind convergence on very heavy event days for observations and for each model.
The 99th percentile values of surface air temperature and specific humidity gradients and horizontal 10-m wind convergence on very heavy event days for observations and for each model.

Winds in the Gulf of Mexico highlight the importance of surface high pressure to the east of the analysis region. Strong winds in the composites tend to start as southwesterly flow around the southern tip of Florida. Over the Gulf, the winds turn clockwise toward the northern coast. This pattern provides substantial fetch for moistening air before it enters the southern U.S. Similar results were found in Brubaker et al. (2001), which emphasized the presence of anticyclonic flow around the Bermuda high, promoting moisture transport not only from the Gulf of Mexico, but also from the Caribbean and tropical Atlantic. Figure 8 does show flow possibly originating south of the Gulf. Although Brubaker et al. (2001) focused on the warm half of the year, Fig. 8 highlights the importance of the moisture fetch during the winter season when, climatologically, Gulf of Mexico moisture does not often penetrate our upper Mississippi region and existing terrestrial moisture supply within the region is low (Brubaker et al. 2001; Kunkel and Liang 2005). The BCC-CSM1.1, MIROC-ESM-CHEM, and BNU-ESM show lower-intensity winds in the Gulf of Mexico compared to the models. This may explain the lower-intensity precipitation events at higher percentiles, since it lowers the moisture from the Gulf of Mexico. NARCCAP RCMs from Kawazoe and Gutowski (2013) show stronger 10-m winds in the Gulf than CMIP5 GCMs.

c. 2-m air temperature and specific humidity

We also analyzed 2-m air temperature and specific humidity from most of the models and the NARR. Figures 9 and 10 show these two fields as composite anomalies. Regions of very heavy precipitation tend to occur in regions of positive temperature and specific humidity anomalies. Like the 500-hPa height anomalies, the maximum temperature and humidity anomalies are roughly 5 times greater than their corresponding 2–5-day variability in the same locations (not shown). Thus, by this measure, all three anomaly fields examined have large, comparable departures from typical daily variability.

The composite temperatures (not shown) in areas of very heavy precipitation are above 275 K, which increases the likelihood that the precipitation type during these events is rain, not snow. Comparisons with Kawazoe and Gutowski (2013) show warm temperature anomalies are stronger for most NARCCAP RCMs than for the CMIP5 GCMs. The BCC-CSM1.1, MIROC-ESM-CHEM, and BNU-ESM mentioned in the 10-m wind analysis also show weaker specific humidity anomalies compared to most of the other models. This supports the results in Table 4, which shows these models having weaker very heavy events, and the lower moisture fetch in these models discussed above. Comparisons with Kawazoe and Gutowski (2013) show specific humidity anomalies agree well between NARCCAP RCMs and CMIP5 GCMs. Finally, Table 6 shows 99th percentile temperature and specific humidity gradients during widespread very heavy events. Temperature gradients show a slight decrease in values as model resolution becomes coarser, while specific humidity does not. As with momentum convergence, there is no evidence of correlation between temperature and specific humidity gradient with respect to precipitation intensities, nor are these values substantially different from corresponding quantities for all days in the analysis period (not shown). This suggests that the variety of modeling differences such as cloud microphysics and boundary layer parameterizations may have a larger impact than model resolutions.

5. Conclusions

Twenty-one GCMs from the CMIP5 project were compared with observational data (University of Washington precipitation and the North American Regional Reanalysis) to determine the ability of models to reproduce very heavy daily precipitation events during winter (December–February) between 1980 and 1999 in an upper Mississippi region. Our very heavy daily precipitation was the top 0.5% of all daily values exceeding 0.25 mm day−1. Widespread very heavy precipitation was defined as very heavy daily precipitation occurring on at least 15 equivalent 0.5° grid boxes simultaneously. For these events, we analyzed daily 500-hPa heights, 2-m air temperature and specific humidity, and 10-m surface winds from a subset of 13 models that archived all these variables to diagnose the environment favorable for the production of very heavy precipitation.

The models, for the most part, tend to produce too much precipitation compared to observations. Also, the models tend to produce fewer precipitation days than observed, which differs from NARCCAP RCM results from Kawazoe and Gutowski (2013). The frequency of days with precipitation above 2.5 mm day−1 agrees well between models and observations, indicating that the GCMs produce too few light precipitation, or drizzle, events, again in contrast with results from the NARCCAP RCMs. The CMIP5 models and observations are in good agreement for frequency versus intensity of precipitation up to about 20 mm day−1 compared to observations. Above this value, most models produce a higher frequency of events than observed, but fail to produce the very intense events seen in observations. The finer-resolution models tend to show more intense precipitation at the 99.5th percentile, with MPI-ESM-LR having the largest value. With the exception of CanESM2, the coarsest-resolution GCMs have lower precipitation intensities at the 99.5th percentile than observations, which may indicate that coarser-resolution GCMs are unable to capture small-scale events that produce very heavy events.

The models do not have a dominant winter month when very heavy precipitation events occur. In the NARCCAP RCMs analyzed by Kawazoe and Gutowski (2013), December showed the highest frequency of very heavy events, likely due to warmer SSTs in the Gulf of Mexico in December. Such a climatological control did not appear in this analysis. However, warmer SSTs were seen during widespread very heavy events, supporting the assumption that warmer SSTs do allow more moisture to enter the atmosphere for transport into the central United States.

For environmental features, the observations and models show similar characteristics. Composite 500-hPa heights show a predominant southwesterly flow into the upper Mississippi region, caused by a deep trough or cutoff low near the Rockies. This allows increased moisture transport into the central United States from the Gulf of Mexico, which aids the development of very heavy precipitation. The 500-hPa heights in CMIP5 GCMs are similar in location and depth of the trough compared to the NARRCAP RCMs studied in Kawazoe and Gutowski (2013). Anomaly plots show that areas experiencing very heavy precipitation tend to occur in areas of positive anomalies of surface air temperatures, which provide an environment capable of holding more moisture compared to climatology. Areas experiencing very heavy precipitation also tend to occur in areas of positive moisture anomalies, showing that the warmer air does indeed have greater moisture. Temperature anomalies tend to be stronger in the NARCCAP RCMs analyzed by Kawazoe and Gutowski (2013) than in the CMIP5 GCMs, while specific humidity anomalies are similar between NARCCAP RCMs and CMIP5 GCMs. Surface wind analysis suggests a strong transport of Gulf of Mexico moisture into the upper Mississippi region. Features of a surface low exist slightly to the west of the area of very heavy precipitation. Low-level momentum convergence of 10-m winds near very heavy events is also present, indicating moisture convergence. Very heavy events tend to occur near the southern portion of the analysis region, centered over central Missouri. This is likely due to the warmer air in the southern part of the analysis region and transport of moisture into the part of the domain that is closest to the moisture source, the Gulf of Mexico. Aside from CanESM2, the coarsest models show slower 10-m winds in the Gulf compared to both higher-resolution GCMs and the NARCCAP RCMs analyzed by Kawazoe and Gutowski (2013). This is consistent with their smaller specific humidity anomalies and 99.5th percentile precipitation, possibly indicating that these models do not have the adequate transport of moisture into the region.

Resolution in itself could not account for differences in precipitation values or composite fields between models. However, these models appear to be capable of producing very heavy precipitation in the analysis region with the correct physical behavior compared to observations. Further diagnosis would be possible if other variables were available for all CMIP5 models studied here. These would include vertically integrated moisture transport, vertical velocity, and horizontal winds at multiple atmospheric levels. However, based on the fields we could analyze, the capability of both the NARCCAP RCMs and CMIP5 GCMs should support using them to assess changes in very heavy precipitation under future climate scenarios.

Acknowledgments

This work was supported by National Science Foundation Grants AGS-1125971 and BCS-1114978. Daily gridded observed precipitation data were obtained from the Surface Water Modeling group at the University of Washington. We acknowledge the World Climate Research Programme's Working Group on Coupled Modelling, which is responsible for CMIP, and we thank the climate modeling groups (listed in Table 1 of this paper) for producing and making available their model output. For CMIP the U.S. Department of Energy's Program for Climate Model Diagnosis and Intercomparison provided coordinating support and led development of software infrastructure in partnership with the Global Organization for Earth System Science Portals. We thank the reviewers for their helpful comments.

REFERENCES

REFERENCES
Brubaker
,
K. L.
,
P. A.
Dirmeyer
,
A.
Sudradjat
,
B. S.
Levy
, and
F.
Bernal
,
2001
:
A 36-yr climatological description of the evaporative sources of warm-season precipitation in the Mississippi River basin
.
J. Hydrometeor.
,
2
,
537
557
.
Daly
,
C.
,
R. P.
Neilson
, and
D. L.
Phillips
,
1994
:
A statistical-topographic model for mapping climatological precipitation over mountainous terrain
.
J. Appl. Meteor.
,
33
,
140
158
.
Groisman
,
P. Y.
,
R. W.
Knight
,
D. R.
Easterling
,
T. R.
Karl
,
G. C.
Hegerl
, and
V. N.
Razuvaev
,
2005
:
Trends in intense precipitation in the climate record
.
J. Climate
,
18
,
1326
1350
.
Gutowski
,
W. J.
,
S. S.
Willis
,
J. C.
Patton
,
B. R. J.
Schwedler
,
R. W.
Arritt
, and
E. S.
Takle
,
2008
:
Changes in extreme, cold-season synoptic precipitation events under global warming
.
Geophys. Res. Lett.
,
35
,
L20710
,
doi:10.1029/2008GL035516
.
Gutowski
,
W. J.
, and
Coauthors
,
2010
:
Regional extreme monthly precipitation simulated by NARCCAP RCMs
.
J. Hydrometeor.
,
11
,
1373
1379
.
Kanamitsu
,
M.
,
W.
Ebisuzaki
,
J.
Woollen
,
S. K.
Yang
,
J. J.
Hnilo
,
M.
Fiorino
, and
G. L.
Potter
,
2002
:
NCEP–DOE AMIP-II reanalysis (R-2)
.
Bull. Amer. Meteor. Soc.
,
83
,
1631
1643
.
Kawazoe
,
S.
, and
W. J.
Gutowski
,
2013
:
Regional, very heavy daily precipitation in NARCCAP simulations
.
J. Hydrometeor.
,
14
,
1212
1227
.
Kunkel
,
K. E.
, and
X.-Z.
Liang
,
2005
:
GCM simulations of the climate in the central United States
.
J. Climate
,
18
,
1016
1031
.
Kunkel
,
K. E.
,
K.
Andsager
,
X.-Z.
Liang
,
R. W.
Arritt
,
E. S.
Takle
,
W. J.
Gutowski
, and
Z.
Pan
,
2002
:
Observations and regional climate model simulations of heavy precipitation events and seasonal anomalies: A comparison
.
J. Hydrometeor.
,
3
,
322
334
.
Maurer
,
E. P.
,
A. W.
Wood
,
J. C.
Adam
,
D. P.
Lettenmaier
, and
B.
Nijssen
,
2002
:
A long-term hydrologically based dataset of land surface fluxes and states for the conterminous United States
.
J. Climate
,
15
,
3237
3251
.
Mesinger
,
F.
, and
Coauthors
,
2006
:
North American Regional Reanalysis
.
Bull. Amer. Meteor. Soc.
,
87
,
343
360
.
Schumacher
,
R. S.
, and
R. H.
Johnson
,
2005
:
Organization and environmental properties of extreme-rain-producing mesoscale convective systems
.
Mon. Wea. Rev.
,
133
,
961
976
.
Schumacher
,
R. S.
, and
R. H.
Johnson
,
2006
:
Characteristics of U.S. Extreme Rain Events during 1999–2003
.
Wea. Forecasting
,
21
,
69
85
.
Shepard
,
D. S.
,
1984
: Computer mapping: The SYMAP interpolation algorithm. Spatial Statistics and Models, G. L Gaile and C. J. Willmott, Eds., D. Reidel, 133–145.
Taylor
,
K. E.
,
R. J.
Stouffer
, and
G. A.
Meehl
,
2012
:
An overview of CMIP5 and the experiment design
.
Bull. Amer. Meteor. Soc.
,
93
,
485
498
.
Wendland
,
W. M.
, and
Coauthors
,
1983
: Review of the unusual winter of 1982–83 in the upper Midwest. Bull. Amer. Meteor. Soc.,64, 1346–1350.
Widmann
,
M.
, and
C. S.
Bretherton
,
2000
:
Validation of mesoscale precipitation in the NCEP reanalysis using a new gridcell dataset for the northwestern United States
.
J. Climate
,
13
,
1936
1950
.