The 10-year average change in the underlying global warming signals in sea surface temperatures (DSST) for the three global warming cases A–C, compared with present- day conditions (1999–2008). (Source: Lauer et al., 2010)
Current climate-change projections from Global Circulation Models (GCMs) have limited application to Hawaii and the Pacific Islands. Most very long climate forecasts have been performed at large horizontal spatial grid resolutions of 200 to 300 km. Without further downscaling analyses, predictions from GCM’s are not useful as the basis for planning adaptation measures at the regional and island scale because there is often a mismatch between the timescales at which managers make decisions and the timescales at which climate models make projections. For example, while a freshwater manager might plan a city’s freshwater infrastructure 10 to 20 years in the future, a climate model provides projections of rainfall that are for the end of the century. While models of climate change at the end of the century are useful for long-term planning and education, it is also helpful for resource managers to have shorter-term climate projections at the seasonal to interannual scale. By downscaling a GCM’s predictions, researchers get more regionally applicable predictions of variables. This project supported two types of climate downscaling efforts to produce the best guidance possible on long-term climate changes expected in Hawaii and other islands in the region.
Global Models
Regional projections for Hawaii and the Pacific Islands require input from global coupled ocean-atmosphere model projections. A large suite of 21st century climate projections for different emission scenarios and different global climate models was made available as part of recent IPCC Assessment Reports. At the time of this research, a similar effort (Coupled Model Intercomparison Project 5 or CMIP5) had more advanced models and projections on 10 to 30 year time frames. Appropriate AR5 and CMIP5 models were then used to represent climate processes in the tropical and subtropical Pacific region.
Downscaling allows researchers to take coarse data from a GCM and create regional predictions on a finer spatial scale. These figures show how both coarse and downscaled information might look on the mainland United States and the main Hawaiian Islands. (UCAR) (left), and Axel Lauer (right). (Source: The University Corporation for Atmospheric Research)
Statistical Downscaling
One approach to producing downscaled data for the Pacific Islands is to use statistical relationships between local variables such as rainfall and the regional-scale weather patterns. If the relationships determined from present-day climate observations can be assumed to hold for future climate conditions, then these relationships can be used to downscale coarse climate projections.
Current work at IPRC using linear statistical downscaling of seasonal rainfall for the main Hawaiian Islands will be extended to consider nonlinear relationships. Downscaled climate data will be extended to consider the distribution of rainfall intensities associated with extreme events throughout the Pacific islands in collaboration with efforts underway in the PRICIP project.
Dynamical Downscaling
A team led by the University of Hawaii conducted dynamical downscaled predictions of precipitation at seasonal to interannual timescales (e.g., during different phases of ENSO) at high spatial resolution, starting on the Hawaiian Islands. Specifically, NOAA operational seasonal prediction model (CFSv2) variables are routinely made available, and they were downscaled to a resolution that was relevant to island spatial scale (~1 km). Pacific RISA has previously supported the development of theHawaiʻi Regional Climate Model (HRCM)by the UH International Pacific Research Center (IPRC). The HRCM is a dynamically downscaled regional model for the Hawaiian Islands at 15 km, 3 km, and 1 km horizontal grid scales. The HRCM has been used to perform a 20-year (1990-2009) present-day climate and projected 20-year (2080-2099) simulations for the late 21st century conditions. One salient result is that both the improved model physics and high model resolutions are needed for realistic simulation of current climate and future projections (Fig. 1).
Fig. 1 compares annual precipitation amounts over the island of Maui simulated with 1 km and 3 km resolutions. Observations show uneven distribution (Fig. 1a) with a peak (> 5000 mm) along the windward side of east Maui and the West Maui mountains, while < 500 mm in the central isthmus. Clearly, the 3 km resolution simulation does not capture this distribution well (Fig. 1c) while the 1 km resolution simulation (Fig. 1b) appears realistic.
To validate the model’s predictive ability, a commonly adopted approach is to ask the model to retrospectively forecast past events (called “hindcasting”), such as the 1982–1983 and 1997–1998 strong El Niño events. Assessing how well different variables in the model hindcast against observed data gave researchers an idea of the model’s ability to predict real-time or future climate. Based on NOAA-recommended metrics, the consensus was that CFSv2 is able to capture modest skill when assessed over the Hawaiian archipelago. Encouraging enough, during strong El Niño events (i.e., 1997-98) the large swings in rainfall and persistence of dryness (from fall through winter and following spring) are skillfully forecast at longer leads by all ensemble members, primarily attributed to realistic representation of physical processes. Due to coarse resolutions employed in CFSv2 model (~125 km), however, the island-scale spatial distribution of precipitation climatology (e.g., Fig. 1a) is not resolved. Keeping in mind the increasing local demands for reliable and future forecast information at island scales, the team employed HRCM downscaling for seasonal prediction of precipitation.
Comparing Statistical and Dynamical Modeling
Both of the statistical and dynamical downscaling approaches used results from global climate models to make spatially detailed projections of future changes. However, the approaches are very different and can produce conflicting results, which make communicating the results and understanding the confidence of the projections complex. Stakeholders in policy and management positions need to know both how and why the projections differ. While both downscaling methods have inherent uncertainties, each has advantages and disadvantages. Therefore, it is important that continuing efforts to improve regional climate projections should be done using both approaches. The Pacific RISA team compared the downscaling results for Hawaii to determine sensitivities to technical assumptions and to ascertain the reasons for apparent discrepancies. Overall, the wet season (November-April) shows an enhanced drying trend in regions with climatological low rainfall amounts, and slightly enhanced precipitation in the wet windward regions.
Interpolated maps of the statistically downscaled wet-season rainfall for scenarios RCP4.5 and RCP 8.5 for the period 2041–2071 (31-yr time-mean). Shown is the ensemble median result from 32-members from CMIP5. Units are in percent.
The general pattern of the projected rainfall changes appears consistent with recent dynamical downscaling results. However, the drying trend was less severe in the dynamical results, and a larger increase in the wet region rainfall was projected in the dynamical downscaling scenario. This project examined how robust the statistical model results were and explored ways to improve the reliability of this approach. The goals were to test: (1) the sensitivity to changes in large-scale climate information; (2) the use of new large-scale climate variables as predictors; and (3) the robustness against changes in the statistical downscaling method. Statistically and dynamically downscaled historical rainfall anomalies were also compared to identify the weaknesses in each approach to provide guidelines for improving projections from both methods.
Our Vision
Pacific Island communities that are resilient to climate impacts and using climate information to manage risks.
Climate Scenarios
Current climate-change projections from Global Circulation Models (GCMs) have limited application to Hawaii and the Pacific Islands. Most very long climate forecasts have been performed at large horizontal spatial grid resolutions of 200 to 300 km. Without further downscaling analyses, predictions from GCM’s are not useful as the basis for planning adaptation measures at the regional and island scale because there is often a mismatch between the timescales at which managers make decisions and the timescales at which climate models make projections. For example, while a freshwater manager might plan a city’s freshwater infrastructure 10 to 20 years in the future, a climate model provides projections of rainfall that are for the end of the century. While models of climate change at the end of the century are useful for long-term planning and education, it is also helpful for resource managers to have shorter-term climate projections at the seasonal to interannual scale. By downscaling a GCM’s predictions, researchers get more regionally applicable predictions of variables. This project supported two types of climate downscaling efforts to produce the best guidance possible on long-term climate changes expected in Hawaii and other islands in the region.
Global Models
Regional projections for Hawaii and the Pacific Islands require input from global coupled ocean-atmosphere model projections. A large suite of 21st century climate projections for different emission scenarios and different global climate models was made available as part of recent IPCC Assessment Reports. At the time of this research, a similar effort (Coupled Model Intercomparison Project 5 or CMIP5) had more advanced models and projections on 10 to 30 year time frames. Appropriate AR5 and CMIP5 models were then used to represent climate processes in the tropical and subtropical Pacific region.
Statistical Downscaling
One approach to producing downscaled data for the Pacific Islands is to use statistical relationships between local variables such as rainfall and the regional-scale weather patterns. If the relationships determined from present-day climate observations can be assumed to hold for future climate conditions, then these relationships can be used to downscale coarse climate projections.
Current work at IPRC using linear statistical downscaling of seasonal rainfall for the main Hawaiian Islands will be extended to consider nonlinear relationships. Downscaled climate data will be extended to consider the distribution of rainfall intensities associated with extreme events throughout the Pacific islands in collaboration with efforts underway in the PRICIP project.
Dynamical Downscaling
A team led by the University of Hawaii conducted dynamical downscaled predictions of precipitation at seasonal to interannual timescales (e.g., during different phases of ENSO) at high spatial resolution, starting on the Hawaiian Islands. Specifically, NOAA operational seasonal prediction model (CFSv2) variables are routinely made available, and they were downscaled to a resolution that was relevant to island spatial scale (~1 km). Pacific RISA has previously supported the development of the Hawaiʻi Regional Climate Model (HRCM) by the UH International Pacific Research Center (IPRC). The HRCM is a dynamically downscaled regional model for the Hawaiian Islands at 15 km, 3 km, and 1 km horizontal grid scales. The HRCM has been used to perform a 20-year (1990-2009) present-day climate and projected 20-year (2080-2099) simulations for the late 21st century conditions. One salient result is that both the improved model physics and high model resolutions are needed for realistic simulation of current climate and future projections (Fig. 1).
To validate the model’s predictive ability, a commonly adopted approach is to ask the model to retrospectively forecast past events (called “hindcasting”), such as the 1982–1983 and 1997–1998 strong El Niño events. Assessing how well different variables in the model hindcast against observed data gave researchers an idea of the model’s ability to predict real-time or future climate. Based on NOAA-recommended metrics, the consensus was that CFSv2 is able to capture modest skill when assessed over the Hawaiian archipelago. Encouraging enough, during strong El Niño events (i.e., 1997-98) the large swings in rainfall and persistence of dryness (from fall through winter and following spring) are skillfully forecast at longer leads by all ensemble members, primarily attributed to realistic representation of physical processes. Due to coarse resolutions employed in CFSv2 model (~125 km), however, the island-scale spatial distribution of precipitation climatology (e.g., Fig. 1a) is not resolved. Keeping in mind the increasing local demands for reliable and future forecast information at island scales, the team employed HRCM downscaling for seasonal prediction of precipitation.
Comparing Statistical and Dynamical Modeling
Both of the statistical and dynamical downscaling approaches used results from global climate models to make spatially detailed projections of future changes. However, the approaches are very different and can produce conflicting results, which make communicating the results and understanding the confidence of the projections complex. Stakeholders in policy and management positions need to know both how and why the projections differ. While both downscaling methods have inherent uncertainties, each has advantages and disadvantages. Therefore, it is important that continuing efforts to improve regional climate projections should be done using both approaches. The Pacific RISA team compared the downscaling results for Hawaii to determine sensitivities to technical assumptions and to ascertain the reasons for apparent discrepancies. Overall, the wet season (November-April) shows an enhanced drying trend in regions with climatological low rainfall amounts, and slightly enhanced precipitation in the wet windward regions.
The general pattern of the projected rainfall changes appears consistent with recent dynamical downscaling results. However, the drying trend was less severe in the dynamical results, and a larger increase in the wet region rainfall was projected in the dynamical downscaling scenario. This project examined how robust the statistical model results were and explored ways to improve the reliability of this approach. The goals were to test: (1) the sensitivity to changes in large-scale climate information; (2) the use of new large-scale climate variables as predictors; and (3) the robustness against changes in the statistical downscaling method. Statistically and dynamically downscaled historical rainfall anomalies were also compared to identify the weaknesses in each approach to provide guidelines for improving projections from both methods.
Our Vision
Pacific Island communities that are resilient to climate impacts and using climate information to manage risks.
Follow Pacific RISA
Archives