Final Report: Statistical Approaches to Detection and Downscaling of Climate Variability and Change

EPA Grant Number: R829402C006
Subproject: this is subproject number 006 , established and managed by the Center Director under grant R829402
(EPA does not fund or establish subprojects; EPA awards and manages the overall grant for this center).

Center: Center for Integrating Statistical and Environmental Science
Center Director: Stein, Michael
Title: Statistical Approaches to Detection and Downscaling of Climate Variability and Change
Investigators: Wuebbles, Donald J. , Cai, Airong , Hayhoe, Katharine , Hertel, Anne , Stein, Michael , Tiao, George , Vrac, Mathieu
Institution: University of Illinois at Urbana-Champaign , Texas Tech University , University of Chicago
EPA Project Officer: Packard, Benjamin H
Project Period: March 12, 2002 through March 11, 2007
RFA: Environmental Statistics Center (2001) RFA Text |  Recipients Lists
Research Category: Ecological Indicators/Assessment/Restoration , Environmental Statistics , Health , Ecosystems


This project applied statistical techniques to analyze global climate model simulations of present-day and future atmospheric circulation patterns and their relationship to surface climate, in order to determine likely impacts from future climate change. Assessing the potential impacts of climate change over the coming century requires an understanding of multi-scale connections between large-scale atmospheric circulation patterns and local surface conditions. Development of such an understanding is a challenging task, but one in which significant advances can be made through application of statistical techniques to a combination of observed and model-simulated atmospheric conditions. In this project, we explored multiple avenues by which statistical analyses can improve our understanding of atmospheric processes and enhance future projections of climate change and its impacts at the regional to local scale.

Summary/Accomplishments (Outputs/Outcomes):

There are four key questions that were examined as part of this project. Here, we summarize the key research questions addressed by this project, briefly describe our findings in each area, and list the presentations and journal articles (submitted, accepted, and published) and other reports (e.g., theses) that have resulted from the various components of this project to date.

Research Question 1. To what extent are global models able to reproduce the dominant circulation patterns of the atmosphere and their relationship to surface climate across the United States and around the world?

Atmospheric circulation patterns — including cyclical phenomenon such as El Ni˜no as well as everyday high and low pressure systems that pass across the continent each week — are key to determining surface climate conditions. To this end, we examined the extent to which the latest global climate model simulations prepared for the Intergovernmental Panel on Climate Change Fourth Assessment Report are able to simulate these large-scale atmospheric patterns. Their performance in this area is highly indicative of their overall skill in capturing the behavior of the real-world climate system.

We first examined six large-scale oscillatory patterns known to be linked (or teleconnected) to surface climate: El Ni˜no Southern Oscillation, the North Atlantic and Pacific Decadal Oscillations, the Arctic and Antarctic Oscillations, and the Pacific North American Pattern. Initial simulations using eight of the latest global climate models (PCM, CCSM3, HadCM3, GFDL CM2.1, ECHAM5, CSIROMk-2, Miroc-med, and CGCM3) showed that, in general, the models are able to simulatee the spatial and temporal characteristics of these patterns. This work was then expanded to all models included in the 4th IPCC assessment. Detailed statistical analysis revealed, however, that models have a tendency to overestimate the spatial strength of the pattern and produce time series that oscillate too quickly relative to the real world. Moreover, some models produce patterns that are clearly closer to “observed” patterns based on reanalysis fields than other models, allowing us to identify those best able to reproduce specific features of atmospheric circulation. We then examined global climate model ability to simulate primary day-to-day atmospheric circulation patterns over North America. Using both a hierarchical and an EM algorithm to cluster reanalysis geopotential height fields, we identified five primary patterns that showed a strong seasonal dependence. Two patterns tended to appear most frequently during winter months; two were typical of “transitional” spring and autumn months, and one tended to dominate summer months. Using the PCM and CCSM models, we found that these models were able to reproduce the same five seasonal patterns, although with a bias towards an early emergence of the summer pattern in spring months and longer duration into the fall as compared with reanalysis-based pattern frequencies.

This work generated the following conference presentations and journal publications: Hayhoe, Wuebbles and Hertel (2005), Hayhoe, et al. (in press), Hertel, Hayhoe and Wuebbles (2006), Vrac, Hayhoe and Stein (2007) and Stoner, Hayhoe and Wuebbles (2008a,b) as well as the M.S. thesis of Hertel (2006).

Research Question 2. When can anthropogenic climate change be expected to produce a statistically significant impact on regional patterns of seasonal temperature and precipitation?

Human-driven climate change is already manifesting itself in many ways around the globe, including rising temperatures and sea level, melting ice sheets, and advances in the timing of spring phenology. However, many of these changes are not yet large enough to affect the consciousness of the general public or regional planners, who often make decisions that will affect a city or a region’s ability to cope with climate and weather events over future years and decades.

Based on historical simulations from the same eight global climate models listed above that take into account observed natural variability and human emissions over the past century, we divided the globe into 256 grid cells and fit a linear statistical model to the simulated seasonal temperature trends in each grid cell. Using a historical database of monthly observed surface temperatures from 1960 through 2000, we then calculated inter-annual variability in seasonal temperatures. From the observationally-based variance and model-based trend estimates, we then estimated the year in which a seasonal temperature trend will become significant at the 95th percentile level or higher for each grid cell.

We found that the opposing factors of variance and trends often tend to cancel each other out, with relatively long detection times being found for both equatorial regions with small variance and small trends as well as for high latitude areas with relatively high trends but also high variance. Instead, we found that continental locations sufficiently poleward to have a significant trend but close enough to the coasts so as to enjoy the moderating effects of the ocean on interannual variability are the most sensitive “indicator” regions where temperature trends either have already or are likely to become significant within the next decade.

This work generated the following conference presentations and journal publications: Cai, Hayhoe and Tiao (2006) and Cai, Hayhoe, Kao, Tiao and Wuebbles (in preparation).

Research Question 3. What impact is anthropogenic climate change expected to have on atmospheric circulation? How will this affect climate averages and extremes at the Earth’s surface?

Changes in the heat content of the atmosphere due to increasing concentrations of greenhouse gases are projected to drive substantial changes in atmospheric circulation and surface climate over the coming century. Here, we used future simulations driven by a range of emission scenarios, in order to capture the uncertainty in future projections due to the human choices and activities that determine our emissions, to examine projected changes in key circulation features and surface climate.

Using the seasonal circulation patterns identified in research question 2, we first examined projected shifts in monthly frequencies of these patterns using simulations of climate over the coming century. In general, we saw a trend towards earlier emergence of transition patterns in winter and of summer patterns in spring and fall, relative to present-day simulations and reanalysisbased pattern frequencies. This is consistent with a picture of a warming planet, and suggests that surface temperature trends over North America may be at least partially driven by earlier and more prolonged summer-like circulation patterns.

We next used the global model simulations to calculate projected changes in 10 indices of climate extremes, five related to temperature and five to precipitation. These extremes measure projected changes in heat wave days, warm nights, annual temperature ranges, dry days, extreme rainfall events, and precipitation intensity. We first found that the historical trends simulated by the models over the past century are remarkably consistent with observed trends over that same time period.

Turning to the future, model simulations presented a consistent picture of a warmer and wetter world. Dramatic increases in heat-related extreme events were seen around the world, as well as increases in both heavy rainfall and drought events across much of the Northern Hemisphere.

Finally, the same model simulations were used to examine changes in the strength and frequency of the six primary teleconnection patterns examined earlier in research question 1. In general, we found that the frequency of El Ni˜no events increases, and many patterns appear to strengthen under a higher emissions scenario in the future. Under a lower emissions scenario, little change is seen. These results seem to suggest a possible threshold of change. If we remain below a certain level of change, atmospheric teleconnection patterns may not be affected; beyond that, however, atmospheric circulation patterns begin to alter as do their impacts on surface climate and weather patterns around the world.

This work generated the following conference presentations and journal publications: Tebaldi, Hayhoe, Arblaster and Meehl (2006), Vrac, Hayhoe and Stein (in preparation), Hertel, Hayhoe, Cai and Wuebbles (in preparation).

Research Question 4. To what degree are statistical methods able to downscale global climate model simulations to reproduce climate means and extremes at the local to regional scale?

Present-day computing resources currently limit the spatial resolution of global climate model simulations to one degree or larger. However, most of the impacts of climate change are expected to be significantly modified by local and regional-scale features. For this reason, we used statistical downscaling methods to develop relationships between large-scale and local-scale climate fields in order to determine the likely future impact of global climate change at the regional level. To this end, we developed a categorization and transition modeling method to provide accurate and rapid simulations of local-scale precipitation features based on statistically-defined weather states, at low computational cost. This statistical method was based on a stochastic weather typing approach. Two different kinds of weather states were defined: “circulation” patterns, developed by a mixture model applied to large-scale NCEP reanalysis data, and “precipitation” patterns, developed by a hierarchical ascending clustering method applied directly to the observed rainfall amounts in Illinois. By modeling the transition probabilities from one pattern to another by a nonhomogeneous Markov model, we saw that the precipitation states allowed us to model conditional distributions and local simulated intensities of precipitation more accurately than with the traditional approach based on upper-air patterns alone.

A fundamental challenge with statistical downscaling approaches to predicting climate change is evaluating their effectiveness without having to wait decades to see how the climate actually changes. We developed a general strategy for evaluating the effectiveness of statistical downscaling approaches by using the results from regional climate models as a surrogate for the future climate on a local scale. We applied our approach to predicting changes in precipitation in Illinois. We found that our downscaling approach worked fairly well at reproducing precipitation patterns from the regional climate model, but had trouble with more extreme precipitation levels and with changes in precipitation induced by the higher emission scenarios.

This work generated the following conference presentations and journal publications: Vrac, Stein and Hayhoe (2007), Hayhoe, Vrac and Stein (2006) and Vrac, Stein, Hayhoe, Liang (2007).

Contributions to Understanding of Environmental Problems

Statements about climate change are often phrased in terms of an overall change in the Earth’s surface temperature. However, the degree of climate change will vary substantially with location and the effects on people will vary even more so. A major tool for studying climate change is general circulation models, or GCMs. However, the spatial resolution of GCMs is limited, so even if they provided accurate assessments of climate change on large scales, there would still be a need to downscale these assessments to more local scales. The observational record is another route to assessing climate change on a local scale, but it has the obvious disadvantage of providing no direct measure of future changes. This project investigated various ways of combining GCM results and observations to assess climate change at local scales.

Our results showed that GCMs do a good job of capturing many of the large-scale oscillatory patterns in the surface climate, thus increasing confidence in their ability to predict changes in climate. However, no matter how good these models get, direct observational evidence for climate change will always be important. We showed that the best locations to look for statistically significant changes in temperature are at coastal sites with fairly high latitudes.

One reason that climate change may vary with location is that a warming atmosphere may change large-scale circulation patterns in the atmosphere. By running GCMs under a number of scenarios, we found that for more optimistic assumptions about future greenhouse gas emissions, these large-cirulation patterns are not predicted to change much by the GCMs, but for more pessimistic emission scenarios, many of these patterns, such as El Ni˜no events, may increase substantially in strength and frequency. These results suggest a possible tipping-point in the climate system in which if greenhouse gas concentrations pass some threshold, substantial changes in large-scale circulation may ensue with profound consequences for the Earth’s climate.

Finally, we investigated new approaches to what is called statistical downscaling, which combines observational records and GCM outputs to predict climate changes on a local level. We developed a new approach to downscaling precipitation that does a better job of reproducing observed precipitation levels in Illinois than do standard approaches. We also developed a general strategy for evaluating the effectiveness of statistical approaches to downscaling projections of future climate changes obtained from GCMs. Applying this strategy to our downscaling methods, we showed that our approach to downscaling precipitation appears to capture most aspects of local changes of precipitation in Illinois for the less extreme emission scenarios.

Journal Articles on this Report : 5 Displayed | Download in RIS Format

Other subproject views: All 12 publications 7 publications in selected types All 5 journal articles
Other center views: All 115 publications 69 publications in selected types All 47 journal articles
Type Citation Sub Project Document Sources
Journal Article Stoner AMK, Hayhoe K, Wuebbles DJ. Assessing general circulation model simulations of atmospheric teleconnection patterns. Journal of Climate 2009;22(16):4348-4372. R829402 (Final)
R829402C006 (Final)
  • Full-text: Journal of Climate - full text
  • Abstract: Journal of Climate-Abstract
  • Other: Journal of Climate - PDF
  • Journal Article Tebaldi C, Hayhoe K, Arblaster JM, Meehl GA. Going to the extremes: an intercomparison of model-simulated historical and future changes in extreme events. Climatic Change 2006;79(3-4):185-211. R829402 (Final)
    R829402C006 (2006)
    R829402C006 (Final)
  • Abstract: SpringerLink-Abstract
  • Other: UCAR-PDF
  • Journal Article Vrac M, Stein ML, Hayhoe K, Liang X-Z. A general method for validating statistical downscaling methods under future climate change. Geophysical Research Letters 2007;34(18):L18701 (5 pp.). R829402 (Final)
    R829402C006 (Final)
    R830963 (Final)
  • Full-text: AGU-Full Text PDF
  • Abstract: AGU-Abstract and Full Text HTML
  • Journal Article Vrac M, Hayhoe K, Stein M. Identification and intermodel comparison of seasonal circulation patterns over North America. International Journal of Climatology 2007;27(5):603-620. R829402 (Final)
    R829402C006 (Final)
  • Abstract: Wiley Online- Abstract
  • Other: Wiley Online-PDF
  • Journal Article Vrac M, Stein M, Hayhoe K. Statistical downscaling of precipitation through nonhomogeneous stochastic weather typing. Climate Research 2007;34(3):169-184. R829402 (Final)
    R829402C006 (Final)
  • Abstract: Inter-Research-Abstract
  • Other: PDF
  • Supplemental Keywords:

    RFA, Scientific Discipline, Economic, Social, & Behavioral Science Research Program, Air, Health Risk Assessment, climate change, Air Pollution Effects, Environmental Statistics, Ecological Risk Assessment, biostatistics, environmental monitoring, particulate matter, risk assessment, health risk analysis, environmental risks, air pollution, climate models, data analysis, environmental indicators, infant mortality, climate variability, statistical methods

    Progress and Final Reports:

    Original Abstract
  • 2002
  • 2003
  • 2004 Progress Report
  • 2005

  • Main Center Abstract and Reports:

    R829402    Center for Integrating Statistical and Environmental Science

    Subprojects under this Center: (EPA does not fund or establish subprojects; EPA awards and manages the overall grant for this center).
    R829402C001 Detection of a Recovery in Stratospheric and Total Ozone
    R829402C002 Integrating Numerical Models and Monitoring Data
    R829402C003 Air Quality and Reported Asthma Incidence in Illinois
    R829402C004 Quasi-Experimental Evidence on How Airborne Particulates Affect Human Health
    R829402C005 Model Choice Stochasticity, and Ecological Complexity
    R829402C006 Statistical Approaches to Detection and Downscaling of Climate Variability and Change