Climate Model Simulations   

 

1.  Introduction

Investigations of historical and potential future climate variations often employ global climate model simulations to test hypotheses concerning the climate system and to examine the range of likely future scenarios.  In the context of Devils Lake, then, climate models may hold some promise for understanding the changes that have occurred and anticipating future trends.  If the models are capable of reproducing either the long-term variations in large-scale climate oscillations [see Observed Changes] that seem responsible for the Devils Lake changes, or the effects of natural and anthropogenic climate trends, then the climate simulations will provide useful predictability.  Thus, Prescient Weather has obtained and analyzed a range of climate model simulations, including many from the Fifth Assessment Report (AR5) of the Intergovernmental Panel on Climate Change (IPCC), and more specifically those from the Coupled Model Intercomparison Project Phase 5 (CMIP5, Taylor et al 2012), to assess whether decadal and longer-term simulations of climate might be useful in long-term planning and management at Devils Lake. 

 

2.  AR5 Model Simulations of 20th Century Devils Lake Precipitation

Before examining the AR5 future scenarios, Prescient Weather first analyzed CMIP5 historical (1850-2005) baseline model simulations from eight international agencies including the NASA Goddard Institute for Space Studies (GISS), the UK Met Office Hadley Centre, the Canadian Centre for Climate Modeling and Analysis, the National Center for Atmospheric Research, the Geophysical Fluid Dynamics Laboratory, the Japan Agency for Marine-Earth Science and Technology, the Meteorological Research Institute, and the Max Planck Institute.  Such a broad sample of international climate models incorporates diverse climate modeling expertise, and the multi-model consensus provides an assessment that is more reliable than the outcome of any single modeling framework. 

The CMIP5 historical simulations span the period 1850-2005 and are intended to facilitate comparison of model performance with present climate and observed climate change by  incorporating observed climate forcing during the 19th and 20th century.  An ensemble of simulations for each model is produced by using multiple physics options or by initializing the simulation from various branch times of the multicentury preindustrial control run after it has reached quasi equilibrium.  Consequently, the historical CMIP5 simulations do not begin with observed initial conditions, but rather with the range of initial conditions typical of a pre-industrial climate that is stable.  The historical simulations incorporate the observed climate forcing with the expectation that the effects of the forcing will become apparent in the trends in climate during the simulation period.  For this research, the CMIP5 data was obtained from the Earth System Grid (ESG), a multi-agency collaboration to make climate data accessible to the international research community (Williams et al. 2009). 

Twentieth-century monthly precipitation at the grid point closest to Devils Lake was extracted from ensemble realizations output from the eight models produced by eight agencies listed earlier.  For each model realization, an annual precipitation climatology was computed for the period 1900-1950 and this climatology was then subtracted during each year of the simulation to obtain the annual anomaly.  The period 1900-1950 was chosen as the base climatology in order to accentuate any possible changes in precipitation during the latter half of the century when anthropogenic climate forcing becomes more pronounced.  The anomalies from each year were then accumulated from 1900 to 2005 and plotted on the same graph with the observed Devils Lake basin-averaged accumulated precipitation anomaly from the Parameter-elevation Regressions on Independent Slopes Model (PRISM) data set (Daly et al. 1994, 2002).

Figure 1 shows the accumulated precipitation anomaly from the 36 climate simulations of the 20th century.  The bold blue line represents the elevation of Devils Lake, the bold black line the basin average PRISM precipitation anomaly, the bold gray line the ensemble mean of the 36 realizations, and the colored lines represent the individual realizations.  The inflection point in 1950 is a result of using the 1900-1950 climatology; the accumulation of the anomalies over that period is by definition zero.  Overall, the ensemble mean of all model simulations (bold gray line) shows minimal change in the precipitation climate of Devils Lake, except for a slight increase in precipitation from 1985 to 2005.  This increase late in the century may be an early sign of small influences of anthropogenic forcing on precipitation near Devils Lake, however this remains speculative.  Two models (GISS and GFDL) show more significant precipitation increases in the last 20-year period, but it is not known whether these increases are a result of anthropogenic climate forcing or part of natural variability from multi-decadal oscillations such as the Pacific Decadal Oscillation (PDO) or Atlantic Multidecadal Oscillation (AMO).  Without more consistency between the models, we can only speculate that anthropogenic climate change during the latter half of the 20th century had a small effect of increasing precipitation near Devils Lake.  The more significant observed increases in precipitation from 1990 onward are more likely a result of changes in circulation patterns associated with natural climate oscillations such as El Niño-Southern Oscillation (ENSO) or the PDO.

Figure 1. Devils Lake elevation from 1900-2005 (blue line), accumulated precipitation anomaly from PRISM observations (black line), ensemble mean accumulated precipitation anomaly from 36 AR5 climate realizations (gray line) and from each of the eight models (bold colored lines), and the accumulated precipitation anomalies from the 36 individual climate realizations (multicolored lines).

 

3. AR5 Model Simulations of 20th Century Large Scale Climate Oscillations

It would be easier to assess the historical simulation results from the AR5 model if we could differentiate the impacts of anthropogenic climate forcing from the impacts caused by natural climate oscillations.  Previous studies of the AR4 climate simulations show that the climate models are capable of simulating some modes of oceanic variability but that their temporal representations during the 20th century are not realistic (Furtado et al. 2011).  Thus to explore whether there have been improvements to the modeling for AR5, five realizations from the GISS-E2-H model and five realizations from the Canadian Centre for Climate Modeling and Analysis (CanESM2) models were examined to determine their ability to portray the PDO, which is a mode of Pacific variability that strongly influences North American climate (Mantua and Hare 2002).

Figure 2 shows the PDO temperature pattern, defined as the first eigenvector of the detrended North Pacific sea surface temperatures; we have used the Extended Reconstructed Sea Surface Temperature dataset (ERSSTv3, Smith et al 2008).  The PDO index was computed from the first principal component (PC1) as shown in Figure 3.  Several smoothing algorithms were applied to the data and plotted, but the 132-month Hamming moving average (green line) shows the decadal variations while still revealing periods of significant variability on the intra-decadal time scale.  Consequently, the 132-month Hamming average was used to smooth the PDO index reconstructions from the GISS and Canadian model realizations.

Figure 2.  Pacific Decadal Oscillation Index derived from the first principal component of detrended North Pacific ERSSTv3 SST anomalies.

The 1900-2005 monthly SST anomaly data from the five GISS-E2-H realizations and five CanESM2 realizations were projected onto the PDO eigenvector, thereby generating an ensemble of 20th century PDO index time series.  The ensemble mean PDO index from the 10 members was also calculated and then plotted with the time series from each ensemble member.  Figure 4 illustrates the observed and reconstructed PDO from 1905-2000 (note that 1900-1905 and 2000-2005 were excluded because of the 132-month smoothing).  The AR5 models are able to realistically represent intra-decadal and decadal PDO variability over the 20th century, but as expected the timing of the PDO phase varies because of the differences between the initial conditions in the historical CMIP5 realizations.  Thus, it seems plausible that the simulated climate oscillations from the CMIP5 models may be responsible for some of the variation in simulated precipitation near Devils Lake, but without further investigation it is difficult to distinguish between changes associated with anthropogenic forcing and changes caused by natural oscillations. 

Figure 3.  Eigenvector (PDO pattern) of the first principal component of detrended North Pacific ERSSTv3 SST anomalies.

Figure 4.  Observed 20th century PDO index (black line), reconstructed PDO index from the NASA GISS-E2-H model realizations (red lines) and the Environment Canada CanESM2 model realizations (green lines), and the ensemble mean PDO index (blue line).  All PDO index values were smoothed using a 132-month Hamming moving average.

 

It is interesting to note that the ensemble mean PDO index in Figure 4 shows a gradual increase beginning in 1930 and continuing to the present.  Such a trend is possibly a result of the way the PDO index is calculated (Zhang et al, 1997), i.e. the monthly mean global average SST anomalies are removed to separate the PDO pattern of variability from any global trends that may be present in the data.  Thus, if the simulated global SST anomaly trends are greater than the observed, then the model SST anomalies in the North Pacific could favor a more positive PDO regime in recent years.

 

4.  AR5 Model Simulations of 21st Century Devils Lake Precipitation

While the CMIP5 historical model simulations do not conclusively indicate that observed climate forcing is responsible for the observed changes in 20th century precipitation near Devils Lake, the future scenarios from these models might still be useful if the expected changes in 21st century Devils Lake precipitation are generally consistent among the various models and ensembles.  Therefore, CMIP5 model simulations of precipitation spanning the period 2005 to 2100 were downloaded from the Earth System Grid (ESG).  Whenever possible, three experiments (RCP26, RCP45, RCP60) were obtained that represent the various 21st century forcing scenarios referred to as “Representative Concentration Pathways” or (RCPs, Moss et al 2010).  The number behind the RCP represents the radiative forcing of either 2.6, 4.5 or 6.0 W/m2 at the end of the 21st century.  The following table shows an inventory of the 21st century model data acquired.

Table 1.    Acquired AR5 model precipitation data for the 21st century RCP scenarios.

A preliminary analysis of the 21st century AR5 model simulations revealed a large range in annual precipitation near Devils Lake.  Thus, in order to compare the various model simulations properly, it was necessary to calibrate the 21st century precipitation realizations.  This was accomplished by obtaining each model’s historical (1900-2005) simulation, and then computing precipitation climatologies for designated periods (e.g. 1900-2005 and 1991-2005).  The climatologies were calculated for each model ensemble member at the model grid point closest to Devils Lake.  Finally, the accumulated precipitation anomaly for the 21st century was computed using a running sum of the difference between the simulated precipitation and historical climatology for each ensemble member and for each year during the 21st century.

Figure 5 shows the accumulated precipitation anomaly for 54 AR5 model simulations and the ensemble mean from each of the eight independent modeling agencies listed in Table 1 and colored as follows:

1.

NASA-Goddard Institute for Space Studies

cyan

2.

UK Met Office Hadley Centre

pink

3.

Canadian Centre for Climate Modeling and Analysis

red

4.

National Center for Atmospheric Research

green

5.

Geophysical Fluid Dynamics Laboratory

yellow

6.

Japan Agency for Marine-Earth Science and Technology

purple

7.

Meteorological Research Institute

blue

8.

Max Planck Institute

orange

In this case, the anomalies were computed using a model climatology spanning 1900-2005.  Nearly all of the AR5 model simulations show precipitation increasing in the 21st century compared to the 20th century model climatology.  The GFDL CM3 model ensemble mean (bold yellow line) shows the largest increases, while the NCAR CCSM 4 ensemble mean (bold green line) indicates only a modest increase.

Figure 5.  Accumulated 21st century precipitation anomaly at Devils Lake from 54 AR5 climate simulations.  The bold colored lines represent the ensemble mean accumulated anomaly from each model.

The simulations were then averaged by representative concentration pathways to get a better indication of the effects of increasing 21st century radiative forcing.  Figure 6 shows the same set of accumulated anomalies as Figure 5, but instead of averaging the ensembles by model, they are averaged by RCP (rcp26 in blue, rcp45 in green, and rcp60 in red).  Overall, there does not appear to be a significant difference in the rate of the accumulated precipitation anomaly among the various forcing scenarios, although near the end of the 21st century the rcp60 (red) scenario shows precipitation increasing more rapidly.  Therefore, while the models suggest that the 21st century will be wetter than the 20th century near Devils Lake, it does not appear that larger increases in greenhouse gas concentrations (e.g. rcp60 versus rcp26) cause significantly greater increases in accumulated precipitation. 

Figure 6.  Same as Figure 5 except that the bold line represent the mean from each RCP (rcp26 in blue, rcp45 in green, and rcp60 in red).

The largest increases in precipitation at Devils Lake have occurred in the past 20 years, and so it is worthwhile examining the 21st century accumulated precipitation anomaly relative to model climatologies spanning the period 1991-2005.  It has been shown that some CMIP5 models indicated a slight upward trend during the latter half of the 20th century.  Thus, it is possible that for these models the increase during the last 20 years will not accelerate further during the 21st century.  Figure 7 is the same as Figure 5 except that the model climatologies span 1991-2005.  Overall, there is still a consensus of models that foresee precipitation increasing during the 21st century when compared with the model climatology from 1991-2005.  However, there is a notable decrease in the magnitude of the accumulated anomalies, particularly with the GISS model, and the MIROC5 model now suggests a decrease in precipitation during the 21st century. 

In conclusion, according to the AR5 climate models it appears likely that 21st century precipitation will increase near Devils Lake relative to the 20th century because of anthropogenic effects.  However, the changes in precipitation are not likely to be as significant as the recent observed increases, which appear to be caused primarily by natural climate oscillations and not by anthropogenic changes.

Figure 7.  Same as Figure 5 except the anomalies were calculated using a model climatology from 1991-2005.

 

5. CFSv2 Decadal Forecasts

The CMIP5 historical and future scenario simulations focus primarily on studying the effects of anthropogenic forcing, but the CMIP5 experiment also includes decadal simulations that were designed to more accurately simulate observed and future decadal climate variations.  The CMIP5 decadal experiments use the observed ocean state at the start of each decadal simulation, which is necessary so that the model is initialized with the observed state of slowly evolving climate oscillations such as the PDO and AMO.  By then creating an ensemble of observed states, the models are expected to more accurately predict the trajectory of future climate, including influences from both natural climate oscillations and from anthropogenic climate change.

In parallel with the CMIP5 decadal experiment, the new Climate Forecast System version 2 (CFSv2) developed at NCEP (Saha et al., 2013) includes decadal-scale reforecasts to allow calibration and skill estimation for predictions of future multi-decadal climate variability.  Prescient Weather examined the CFSv2 reforecasts of precipitation at Devils Lake and across North America to assess whether the model showed skill in anticipating the changes that have occurred.

First, the monthly precipitation amounts were extracted for the grid point closest to Devils Lake from the forecasts initialized on November 1 of 1960, 1980, and 2005.  For each initialization date, four ensemble members were available, each with a forecast length of 30 years (i.e. extending to December of 1990, 2010, and 2035).   The trailing decadal precipitation totals from each ensemble member and the ensemble means are shown in Figure 8, along with the observed Devils Lake basin-average precipitation from the Climate Prediction Center unified precipitation analysis (Chen et al. 2008).  There are very large differences between the ensemble members, with some members deviating strongly from observations in the 1970-2010 period.  However, the ensemble mean forecasts from the 1980-2010 simulations show some resemblance to the observations, with a large increase in precipitation after 1990.  The forecasts for 2005-2035 show generally decreasing precipitation until about 2025, followed by another rapid increase.

Figure 8.  Trailing decadal precipitation totals from the CPC unified precipitation analysis (green line, right scale) and from ensemble member (red, purple, blue, and orange lines) and ensemble mean (black line) CFSv2 decadal forecasts (left scale).  No bias adjustment or calibration is applied.

 

The spatial patterns of observed and predicted decadal precipitation were also examined from the same set of forecasts.  The model successfully anticipated the large-scale shift to wetter conditions in the northern Plains in about 1990 (e.g. Figure 9), but in some other areas the forecast changes were of the wrong sign.  Thus the anomaly correlations between the forecasts and observations (Table 2) are small and positive for a regional domain surrounding Devils Lake, but are close to zero for the entire domain shown in Figure 9.

Given that only a few decades are sampled in this limited set of retrospective forecasts, it is difficult to draw definitive conclusions about the skill of the CFSv2 decadal precipitation forecasts.  However, the model did show some success in anticipating the wetter conditions after 1990 in the northern Plains and near Devils Lake.  For the years ahead, the CFSv2 simulations initialized in 2005 expect a notable decrease in precipitation until about 2025, followed by a sharp increase by 2035 (Figure 10).


Figure 9.  Decadal precipitation anomalies from the CFSv2 reforecasts for 1981-1990 and 1991-2000 (top panels) and from the CPC unified precipitation analysis (bottom panels).  The anomalies are computed relative to the modeled and observed 1981-2010 climatologies respectively.  The light blue polygon indicates the Devils Lake drainage basin.

 

Initialization Year

Forecast Period

North America domain

Devils Lake domain

1960

1981-1990

0.13

0.19

1980

1981-1990

-0.12

0.33

1980

1991-2000

0.06

0.24

1980

2001-2010

-0.08

0.03


Table 2.    Anomaly correlation between CFSv2 reforecast decadal precipitation totals and the corresponding observed totals in the CPC unified precipitation analysis, for the domain shown in Figure 9 (“North America domain”) and for a regional domain (“Devils Lake domain”) of approximately 800x800 km centered on Devils Lake.  The anomalies are computed relative to the modeled and observed 1981-2010 climatology respectively.

 


Figure 10.  Ensemble mean CFSv2 forecast precipitation anomaly for 2006-2015 (left), 2016-2025 (center) and 2026-2035 (right), from the simulations initialized on November 1, 2005.  The anomalies are computed relative to the 1981-2010 model climatology.

 

6.  Conclusions

The IPCC AR5 climate simulations of the 21st century suggest an increase in Devils Lake precipitation relative to the 20th century and perhaps even greater than the recent wet regime.  However, the climate models have been generally incapable of simulating the increases in precipitation during the 20th century, suggesting that recent observed trends may be more closely related to natural climate variability than anthropogenic climate change.  The inability of the AR5 simulations to predict the timing of natural climate oscillations creates difficulty in distinguishing the two potential causes of the observed change.

A preliminary examination of CMIP5 decadal simulations reveals that the CFSv2 model was able to anticipate the late 20th century shift to wet conditions and is therefore slightly more credible in its portrayal of future changes.  According to the CFSv2 decadal forecast for 2006-2035, precipitation amounts will diminish until 2025 but then rapidly increase again after that, reaching a level comparable to the wettest recent decade by 2035.  Future research should examine a larger ensemble of CMIP5 decadal simulations to reveal the effects of both natural and anthropogenic climate change on future precipitation at Devils Lake.

 

References

Chen, M., W. Shi, P. Xie, V. B. S. Silva, V. E. Kousky, R. W. Higgins, and J. E. Janowiak, 2008: Assessing Objective Techniques for Gauge-Based Analyses of Global Daily Precipitation.  J. Geophys. Res., 113.

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.

Daly, C., W. P. Gibson, G. H. Taylor, G. L. Johnson, and P. Pasteris, 2002: A Knowledge-Based Approach to the Statistical Mapping of Climate.  Climate Research, 22, 99-113.

Furtado, Jason C., Emanuele Di Lorenzo, Niklas Schneider, Nicholas A. Bond, 2011: North Pacific Decadal Variability and Climate Change in the IPCC AR4 Models. J. Climate, 24, 3049–3067.

Mantua, N. J, and S. R. Hare, 2002: The Pacific Decadal Oscillation.  J. Oceanography, 58, 35-44.

Richard H. Moss, Jae A. Edmonds, Kathy A. Hibbard, Martin R. Manning, Steven K. Rose, Detlef P. van Vuuren, Timothy R. Carter, Seita Emori, Mikiko Kainuma, Tom Kram, Gerald A. Meehl, John F. B. Mitchell, Nebojsa Nakicenovic, Keywan Riahi, Steven J. Smith, Ronald J. Stouffer, Allison M. Thomson, John P. Weyant1 and Thomas J. Wilbanks, 2010. The next generation of scenarios for climate change research and assessment. Nature 463: 747-756.

Smith, T.M., R.W. Reynolds, Thomas C. Peterson, and Jay Lawrimore, 2008: Improvements to NOAA's Historical Merged Land-Ocean Surface Temperature Analysis (1880-2006). Journal of Climate, 21, 2283-2296.

Saha, Suranjana et al., 2013,  The NCEP Climate Forecast System Version 2., submitted to Journal of Climate, manuscript available at http://cfs.ncep.noaa.gov/cfsv2.info/
CFSv2_paper.pdf.

Taylor, Karl E., Ronald J. Stouffer, Gerald A. Meehl, 2012: An overview of cmip5 and the experiment design. Bull. Amer. Meteor. Soc., 93, 485–498.

Williams, D. N., and Coauthors, 2009: The earth system grid: enabling access to multimodel climate simulation data. Bull. Amer. Meteor. Soc., 90, 195–205.

Zhang, Y., J.M. Wallace, D.S. Battisti, 1997: ENSO-like interdecadal variability: 1900-93. J. Climate, 10, 1004-1020.