We have documented in our paper
Pielke Sr., R.A., R. Wilby, D. Niyogi, F. Hossain, K. Dairuku, J. Adegoke, G. Kallos, T. Seastedt, and K. Suding, 2011: Dealing with complexity and extreme events using a bottom-up, resource-based vulnerability perspective. AGU Monograph on Complexity and Extreme Events in Geosciences, in press.
that regionally downscaled forecasts from global multi-decadal climate model predcitions have no skill beyond whatever is in the parent global model. I summarized the issues with dynamical regional downscaling in my post
The Failure Of Dynamic Downscaling As Adding Value to Multi-Decadal Regional Climate Prediction
1. As a necessary condition for an accurate prediction, the multi-decadal global climate model simulations must include all first-order climate forcings and feedbacks. However, they do not [see for example: NRC, 2005; Pielke Sr. et al., 2009].
2. These global multi-decadal predictions are unable to skillfully simulate major atmospheric circulation features such the Pacific Decadal Oscillation [PDO], the North Atlantic Oscillation [NAO], El Niño and La Niña, and the South Asian monsoon [Pielke Sr., 2010; Annamalai et al., 2007].
3. While dynamic regional downscaling yield higher spatial resolution, the regional climate models are strongly dependent on the lateral boundary conditions and interior nudging by their parent global models [e.g., see Rockel et al., 2008]. Large-scale climate errors in the global models are retained and could even be amplified by the higher spatial resolution regional models.
4. Since as reported, the global multi-decadal climate model predictions cannot accurately predict circulation features such as the PDO, NAO, El Niño, and La Niña [Compo et al., 2011] they cannot provide accurate lateral boundary conditions and interior nudging to the regional climate models.
5. The regional models themselves do not have the domain scale (or two-way interaction) to skillfully predict these larger-scale atmospheric features.
6. There is also only one-way interaction between regional and global models which is not physically consistent. If the regional model significantly alters the atmospheric and/or ocean circulations, there is no way for this information to alter the larger-scale circulation features which are being fed into the regional model through the lateral boundary conditions and nudging.
7. When higher spatial analyses of land use and other forcings are considered in the regional domain, the errors and uncertainty from the larger model still persists thus rendering the added complexity and details ineffective [Ray et al. 2010; Mishra et al. 2010].
8. The lateral boundary conditions for input to regional downscaling require regional-scale information from a global forecast model. However the global model does not have this regional-scale information due to its limited spatial resolution. This is, however, a logical paradox since the regional model needs something that can only be acquired by a regional model (or regional observations). Therefore, the acquisition of lateral boundary conditions with the needed spatial resolution becomes logically impossible.
Finally, There is sometimes an incorrect assumption that although global climate models cannot predict future climate change as an initial value problem, they can predict future climate statistics as a boundary value problem [Palmer et al., 2008]. With respect to weather patterns, for the downscaling regional (and global) models to add value over and beyond what is available from the historical, recent paleo-record, and worse case sequence of days, however, they must be able to skillfully predict the changes in the regional weather statistics.
There is only value for predicting climate change, however, if they could skillfully predict the changes in the statistics of the weather and other aspects of the climate system. There is no evidence, however, that the model can predict changes in these climate statistics even in hindcast. As highlighted in Dessai et al. [2009] the finer and time-space based downscaled information can be “misconstrued as accurate”, but the ability to get this finer-scale information does not necessarily translate into increased confidence in the downscaled scenario [Wilby, 2010].
There is an article in the July 2011 issue of the Intermountain West Climate Summary titled
Examining Regional Climate Model (RCM) projections: What do they add to our picture of future climate in the region? By Karen Cozzetto, Imtiaz Rangwala, and Jeff Lukas
which perpeturates the erroneous claim that skillful predictions of regional climate decades from now are possible.
The article starts with the text [highlight added]
“To prepare for future climate change, land and water resource managers want to know how key climate variables, such as temperature and precipitation, may change in the future relative to the present. The principal tools for investigating potential future climate changes on global-to-regional scales are global climate models (GCMs). Because of the relatively coarse spatial resolution of GCM output (100- 300 km), many user applications of GCM climate projections require processing of the GCM output to bring the effective scale of the data to a more local level. This process is called downscaling.
Excerpts exclude
“One of the main advantages of dynamical downscaling over statistical downscaling that the former represents the physical processes of climate—thus linking spatial scales of climate in a manner that can vary as the the future climate changes. By contrast, statistical downscaling is based on fixed historically-based assumptions regarding the spatial relationships of climate variables. In addition, a greater number of output climate variables from these RCMs relevant to resource managers, are being archived at subdaily timescales. For example, the RCMs simulate the individual terms in the water and energy budgets at the Earth’s surface, so that projected trends in evapotranspiration, solar radiation, and snowcover can be investigated at sub-GCM scales.”
“Mid-21st century changes in temperature and precipitation were determined as the average changes over the 2041-2070 time period relative to the average for the 1971-2000 time period. Changes in both temperature and precipitation were examined on a seasonal (3-month) basis, and precipitation changes were also analyzed on a monthly basis because some of the key precipitation features in southwestern Colorado occur at that timescale.”
“A major disadvantage of dynamical downscaling is that, as with GCMs, the process is computationally intensive and there are biases, or systematic errors, in the simulation of the present-day climate. And if modelers of water and ecosystem impacts require data at a yet finer spatial resolution than is provided by the RCMs, they would still need to make use of statistical methods to further downscale the data.”
“Researchers with the Western Water Assessment (authors Cozzetto and Rangwala, along with Jason Neff and Joe Barsugli) have examined in detail the dynamically downscaled temperature and precipitation projections available from the NARCCAP for southwestern Colorado for two 30-year periods: a historic period (1971-2000) and a future period (2041-2070). The area of investigation extends from 36.0° to 38.5° North latitude and 105.5° to 110° West longitude, and was centered on the San Juan Mountains….”
My Comment: If there are “biases, or systematic errors, in the simulation of the present-day climate”, why would the regional forecasts decades from now be accepted by the impacts community? This is a flawed scientific approach.
Also, to state that “dynamical downscaling …..represents the physical processes of climate” ignores peer reviewed papers and national assessments that conclude i) the global models do not have all of the important human climate forcings (e.g. NRC, 2005), and ii) they fail to properly simulate the physics of the climate system (e.g. see). These models have also failed so far to skillfully simulate large scale atmospheric circulations (e. g. see and see). The mult-decadal global climate models DO NOT accurately simulate the climate system and its response to human climate forcings.
The article has a section headlined
“Limitations and uncertainties in the NARCCAP results”
which reads
The NARCCAP data are useful in that they facilitate comparison among the results of multiple RCMs and GCMs and allow examination of the additional information that dynamical downscaling can provide about future climates at smaller spatial scales. However, the four GCMs used by NARCCAP to provide boundary conditions for the RCMs represents only one-sixth of the GCMs available in the CMIP3 archive. Thus, the analysis for the San Juans presented here does not capture the full range of available GCM climate projections.
Also, while the 50-km resolution of the NARCCAP data is a large improvement over the resolution of the GCMs, resulting in 20 to 40 times more grid cells, it is still inadequate to fully resolve both the horizontal and vertical scales of local topographic features. For
example, the RCM representations of the San Juan Mountains only reach 3300 m (11,000’) instead of the actual 4300 m (14,000’), thus constraining the climatic influence of topography (e.g., on the terrain-induced lifting of air masses above the condensation level). And as noted above, none of the RCM simulations captured the monthly precipitation climatology of the region. In particular, all of the models had trouble reproducing various features of the North American Monsoon from July through September. In most cases, no monsoon was simulated, and in the remaining ones, the monsoon was not maintained for a long enough period. Additionally, a majority of the RCM simulations had problems reproducing the observed trend of increasing precipitation with elevation during the fall, winter, and spring months, the period during which the snowpack is accumulated. This analysis indicates that climate model projections of changes in precipitation have much greater uncertainty than temperature and, therefore, should be treated with greater caution. That said, there is no clear evidence for a future trend toward greater annual precipitation in this region that would be large enough to counterbalance the drying effect of the projected increase in temperature. Thus, it is likely that water availability will decline in the future.”
My Comment: The authors confess that “….none of the RCM simulations captured the monthly precipitation climatology of the region…” This is a remarkable statement! The models do not accurately simulate the precipitation yet they are provided to the water resource community as robust results in which they should base decisions.
The title of the paper by Greame Stephens
Stephens, G. L., T. L’Ecuyer, R. Forbes, A. Gettlemen, J.‐C. Golaz, A. Bodas‐Salcedo, K. Suzuki, P. Gabriel, and J. Haynes (2010), Dreary state of precipitation in global models, J. Geophys. Res., 115, D24211, doi:10.1029/2010JD014532
by itself shows how far we have yet to go to accuratelt predict precipitation. There is no way the regional models can correct for the deficiencies in the physics and dynamics of the global model results since its temperature, wind and moisture fields feed into the regional models. There is also no way the regional multi-decadal temperature forecasts can be correct if the precipitation and clouds are not properly modeled.
They write about the next step in the section
“What’s coming: High-resolution dynamical downscaling”
which reads
“To properly simulate precipitation in mountainous regions such as the Colorado Rocky Mountains, where terrain-induced lifting of air masses is dominant, a realistic depiction of the topography is essential. Such orographic processes are almost non-existent in the current GCMs and remain weak in NARCCAP-type RCMs. High-resolution RCMs at spatial scales as low as 1-2 km are now being used, in a limited research mode, to simulate climate over mountainous regions. To properly simulate snowfall over the Colorado Rockies,
researchers at NCAR have found that the RCMs need to be run at spatial scales of 6 km or less (Rasmussen et al. 2011). In one climate change experiment, they found the high-resolution RCM simulated a much greater increase in winter precipitation (+26%) for the Colorado Rockies than that projected by the driving GCM (+4%).Even if we accept that high-resolution RCMs are more realistically simulating winter precipitation in Intermountain West, as appears likely, we would still need several such RCM runs driven by a suite of different GCM boundary forcings to confidently project future precipitation trends from them. These runs would be extremely computationally intensive and it would not be economically viable to perform them over large areas in the near future. And this approach would not necessarily address the poor
representation of the North American Monsoon and summer precipitation in the region. Nonetheless, we can expect to see more results from fine-scale RCMs in near future, and it is expected that they will increase our understanding of the complex atmospheric processes over mountains.”
My Comment: They write ‘[e]ven if we accept that high-resolution RCMs are more realistically simulating winter precipitation in Intermountain West, as appears likely….” The idea of “accepting” the model results without observational validation is absurd. Moreover, it is an illusion to assume that model predicted finer scale structure due to more detailed terrain structure is actually model skill. It is asking too much of the RCMs to correct systematic biases from the parent global model by the finer grid resolution of the terrain and other landscape features as the RCMs are slaves to the parent model.
My Bottom Line:
There is absolutely no value in regional downscaling from multi-decadal global climate models. such work, as exemplfied by
Examining Regional Climate Model (RCM) projections: What do they add to our picture of future climate in the region? By Karen Cozzetto, Imtiaz Rangwala, and Jeff Lukas
Such are studies are misleading policymakers.
Climate Science: Roger Pielke Sr. 