I was alerted to another paper that documents the limitations of multi-decadal regional climate predictions [h/t Robert Pollock] .
The paper is
Mishra, V., F. Dominguez, and D. P. Lettenmaier (2012), Urban precipitation extremes: How reliable are regional climate models?, Geophys. Res. Lett., 39, L03407, doi:10.1029/2011GL050658.
The abstract reads [highlight added]
We evaluate the ability of regional climate models (RCMs) that participated in the North American Regional Climate Change Assessment Program (NARCCAP) to reproduce the historical season of occurrence, mean, and variability of 3 and 24-hour precipitation extremes for 100 urban areas across the United States. We show that RCMs with both reanalysis and global climate model (GCM) boundary conditions behave similarly and underestimate 3-hour precipitation maxima across almost the entire U.S. RCMs with both boundary conditions broadly capture the season of occurrence of precipitation maxima except in the interior of the western U.S. and the southeastern U.S. On the other hand, the RCMs do much better in identifying the season of 24-hour precipitation maxima. For mean annual precipitation maxima, regardless of the boundary condition, RCMs consistently show high (low) bias for locations in the western (eastern) U.S. Our results indicate that RCM-simulated 3-hour precipitation maxima at 100-year return period could be considered acceptable for stormwater infrastructure design at less than 12% of the 100 urban areas (regardless of boundary conditions). RCM performance for 24-hour precipitation maxima was slightly better, with performance acceptable for stormwater infrastructure design judged adequate at about 25% of the urban areas.
Their experimental design is explained as
We used RCM-simulated precipitation output from participating models in the North American Regional Climate Change Assessment Program (NARCCAP) [Mearns et al., 2009]. For most of the NARCCAP RCMs, two distinct simulations were made: the first simulation forced the RCMs with output from the National Center for Environmental Prediction/Department of Energy (NCEP/DOE) reanalysis [Kanamitsu et al., 2002] at the boundaries for the 1979–2000 period (RCM-reanalysis henceforth). For the second simulation, output from selected GCMs was used to provide the RCM boundary conditions both in the historical (1968–2000) and future (2038–2080) periods (RCM-GCM henceforth). In this study, we focus only on the RCM reanalysis and RCM-GCM for the historical period, because our objective is to evaluate model skill when compared to observations.
Thus, the downscaling runs using the Reanalysis is a Type 2 downscaling as defined in
Castro, C.L., R.A. Pielke Sr., and G. Leoncini, 2005: Dynamical downscaling: Assessment of value retained and added using the Regional Atmospheric Modeling System (RAMS). J. Geophys. Res. – Atmospheres, 110, No. D5, D05108, doi:10.1029/2004JD004721.
The runs with the GCMs for the period 1968-2000 appears to be a Type 3 downscaling (i.e. the SSTs are prescribed over this time period, but their paper is not clear on this). If SSTs, and all other aspects of the GCM runs were predicted, not prescribed, this would be a Type 4 downscaling simulation run in a hindcast mode.
Their conclusions include the summary
1. RCM performance is satisfactory in simulating the seasonality of 24-hour precipitation extremes across most of the U.S. However, for most urban areas in the western and southeastern U.S., the seasonality of 3-hour precipitation extremes was not successfully reproduced by the RCMs with either reanalysis or GCM boundary conditions. Specifically, the RCMs tended to predict 3-hour precipitation maxima in winter, whereas the observations indicated summer.
2. RCMs largely underestimated 3-hour precipitation maxima means and 100-year return period magnitudes at most locations across the United States for both reanalysis and GCM boundary conditions. However, performance was better for 24-hour precipitation maxima (means and 100-year events), although there were generally overestimates in the west, and underestimates in the east.
3. For both 3 and 24-hour annual precipitation maxima, RCMs with reanalysis boundary conditions underestimated interannual variability and overestimated interannual variability with GCM boundary conditions.
4. At only a very small number of locations was the bias in RCM-simulated 3 and 24-hour 100 year return period precipitation maxima within +/-10% of the observed estimates, which might be deemed acceptable for stormwater infrastructure design purposes.
This is an informative study. Using reanalyses, where real-world observations are used to constrain the regional climate model predictions (through later boundary conditions and nudging), provides the benchmark upon which the multi-decadal climate forecasts must improve on.
Papers that we have completed on extreme rainfall events in urban areas; e. g.
Lei, M., D. Niyogi, C. Kishtawal, R. Pielke Sr., A. Beltrán-Przekurat, T. Nobis, and S. Vaidya, 2008: Effect of explicit urban land surface representation on the simulation of the 26 July 2005 heavy rain event over Mumbai, India. Atmos. Chem. Phys. Discussions, 8, 8773–8816.
show that landscape effects must also be considered in planning for extreme rainfall events.
See also for Atlanta, research on this subject by Marshall Shepherd and by Dev Niyogi
The Mishra et al 2012 paper shows that participating models in the North American Regional Climate Change Assessment Program (NARCCAP) have not provided evidence that their predictions would have the required skill for the future time period (2038–2080).
They have biases for the recent climate, and have not even been tested in this paper with respect to their ability to skillfully predict changes in urban climate statistics over the period 1968 to 2000. If they are being provided to urban planners as being robust estimates of the envelope of what could occur during 2038-2080, they are misleading those policymakers.