Cornell EAS seminar

Seminar talk given at Cornell EAS, February 14th 2025.

Transcript

1. David Lafferty (Cornell BEE) with Samantha Hartke (NCAR/USACE) Ryan Sriver (UIUC) Vivek Srikrishnan (Cornell) EAS Seminar February 14th 2025 What drives uncertainty in local climate risk estimates across the United States?
2. Demand for local climate information is increasing • Academia: impact assessments often rely on high resolution climate projections 2
3. Demand for local climate information is increasing • Public sector: local climate information is increasingly used to guide rules and regulations 3 Lopez-Cantu et al, Incorporating uncertainty from downscaled rainfall projections into climate resilience planning in U.S. cities, 2022 Environ. Res.: Infrastruct. Sustain. “most of the adaptation plans we assessed use future projections produced by statistical downscaling methods… whereas a few plans use dynamically downscaled projections”
4. Generating local climate information requires confronting many sources of uncertainty 4 1) Scenario uncertainty: how will greenhouse gas (GHG) emissions evolve in future?
5. Generating local climate information requires confronting many sources of uncertainty 5 1) Scenario uncertainty 2) Response/model uncertainty: how will the climate system respond?
6. Generating local climate information requires confronting many sources of uncertainty 6 1) Scenario uncertainty 2) Response/model uncertainty: how will the climate system respond?
7. Generating local climate information requires confronting many sources of uncertainty 7 1) Scenario uncertainty 2) Response/model uncertainty 3) Internal variability: what is the range of natural variability around the forced response?
8. Generating local climate information requires confronting many sources of uncertainty 8 1) Scenario uncertainty 2) Response/model uncertainty 3) Internal variability: what is the range of natural variability around the forced response? Lehner & Deser (2023) Origin, importance, and predictive limits of internal climate variability. Environ. Res.: Climate
9. Generating local climate information requires confronting many sources of uncertainty 9 1) Scenario uncertainty 2) Response/model uncertainty 3) Internal variability 4) Downscaling uncertainty: how do we infer local climate/weather information from coarse-scale model outputs?
10. Understanding the contribution from each source of uncertainty can help practitioners 10 What are the key uncertainties shaping long-term projections? • Given fixed computational and time constraints, where should we focus our energy? • Scenario, response, and downscaling uncertainties are reducible; internal variability is not Ithaca annual maximum temperature
11. Understanding the contribution from each source of uncertainty can help practitioners 11 Blanusa et al. (2023), Internal variability plays a dominant role in global climate projections of temperature and precipitation extremes. Climate Dynamics See also: Lehner et al. (2020), Partitioning climate projection uncertainty with multiple large ensembles and CMIP5/6. Earth System Dynamics Schwarzwald & Lenssen (2022), The importance of internal climate variability in climate impact projections. PNAS Recent work in this area: Ø Leverages large ensembles to highlight the importance of internal variability Ø Leverages multiple downscaled ensembles to highlight the importance of downscaling uncertainty
12. Understanding the contribution from each source of uncertainty can help practitioners 12 Recent work in this area: Ø Leverages large ensembles to highlight the importance of internal variability Ø Leverages multiple downscaled ensembles to highlight the importance of downscaling uncertainty Lafferty & Sriver (2023), Downscaling and bias-correction contribute considerable uncertainty to local climate projections in CMIP6. npj Clim. & Atm. Sc. See also: Wootten et al. (2017), Characterizing Sources of Uncertainty from Global Climate Models and Downscaling Techniques. J. Applied Met. & Clim.
13. Understanding the contribution from each source of uncertainty can help practitioners 13 Recent work in this area: Ø Leverages large ensembles to highlight the importance of internal variability Ø Leverages multiple downscaled ensembles to highlight the importance of downscaling uncertainty Outstanding questions: o How do internal variability and downscaling uncertainty combine to shape local risk estimates? Data limitations have prevented us analyzing internal variability and downscaling uncertainty in a single framework.
14. New downscaled datasets allow a better characterization of internal variability 14 Downscaled Ensemble No. of GCMs Spatial res. Initial condition sampling Emissions scenarios Training data Total outputs GARD-LENS 3 12.5 km EC-Earth3 (50 members) CanESM5 (50 members) CESM2-LENS (100 members) SSP3-7.0 GMET (1980-2014) 200 LOCA2 27 6 km Up to 10 members SSP2-4.5, SSP3-7.0, SSP5-8.5 Livneh (1950-2014) ~300 STAR-ESDM 24 ~5 km No SSP2-4.5, SSP5-8.5 NClimGrid (1951-2001) 48 Hartke et al. (2024), GARD-LENS: A downscaled large ensemble dataset for understanding future climate and its uncertainties. Sci. Data Pierce et al. (2023), Future Increases in North American Extreme Precipitation in CMIP6 Downscaled with LOCA. J. Hydrometeor Hayhoe et al. (2024), STAR-ESDM: A generalizable approach to generating high-resolution climate projections through signal decomposition. Earth's Future
15. We analyze uncertainty in extreme return levels 15 1) We fit a GEV distribution to annual maxima to estimate return levels 2) Scale the GEV calculation across all outputs 3) Estimate the uncertainty associated with each source
16. We analyze uncertainty in extreme return levels 16 - 0 - 1 , , ,1 , 1 ,- . 1) We fit a GEV distribution to annual maxima to estimate return levels 2) Scale the GEV calculation across all outputs 3) Estimate the uncertainty associated with each source
17. We analyze uncertainty in extreme return levels 17 1) We fit a GEV distribution to annual maxima to estimate return levels 2) Scale the GEV calculation across all outputs 3) Estimate the uncertainty associated with each source ( )
18. We analyze uncertainty in extreme return levels 18 1) We fit a GEV distribution to annual maxima to estimate return levels 2) Scale the GEV calculation across all outputs 3) Estimate the uncertainty associated with each source 5( 2 5 2 0) .. 0 23 ( 0 0 23 0 (4 12- 35 5( 2 5 2 0) ..
19. We analyze uncertainty in extreme return levels 19 1) We fit a GEV distribution to annual maxima to estimate return levels 2) Scale the GEV calculation across all outputs 3) Estimate the uncertainty associated with each source Internal variability: range across initial condition members
20. We analyze uncertainty in extreme return levels 20 1) We fit a GEV distribution to annual maxima to estimate return levels 2) Scale the GEV calculation across all outputs 3) Estimate the uncertainty associated with each source Downscaling uncertainty: range across downscaling methods -
21. We analyze uncertainty in extreme return levels 21 1) We fit a GEV distribution to annual maxima to estimate return levels 2) Scale the GEV calculation across all outputs 3) Estimate the uncertainty associated with each source ) ) ) ( ) ( ) ) )
22. Changes in risk metrics are less sensitive to downscaling uncertainty 22 - 0 - 1 , , ,1 , 1 ,- .
23. CONUS-wide analysis: extreme precipitation 23
24. CONUS-wide analysis: extreme heat 24
25. CONUS-wide analysis: extreme cold 25
26. Key takeaways and ongoing robustness checks 26 Key takeaways: Ø Internal variability drive large differences estimated return levels, particularly for extreme precipitation and extreme cold Ø Downscaling uncertainty can be important for precipitation, but generally diminishes when looking at long-term changes Ø Response uncertainty is generally more important for temperature metrics Robustness checks: o Does allowing for non-stationarity change these results? o How important is GEV fit uncertainty?
27. Implications for climate risk analysis? 27 Required resolution? Plausible change signals? Dynamical or statistical downscaling? Bias- correction? Large enough large ensemble? Cloud vs. HPC?
28. 28 Thank you! Key takeaways: Ø Internal variability can drive large differences in decision-relevant risk metrics Ø How should we design and analyze climate model ensembles to efficiently sample relevant uncertainties? david0811.github.io dcl257@cornell.edu
29. Back up slides 29
30. Demand for local climate information is increasing • Private sector: emergence of climate services companies to provide bespoke analyses 30 Source: https://www.jupiterintel.com/solutions/climate-risk-disclosure
31. We analyze uncertainty in extreme return levels 31 1) We fit a GEV distribution to annual maxima to estimate return levels 2) Scale the GEV calculation across all outputs 3) Estimate the uncertainty associated with each source Scenario uncertainty: range across scenarios of the forced responses
32. We analyze uncertainty in extreme return levels 32 1) We fit a GEV distribution to annual maxima to estimate return levels 2) Scale the GEV calculation across all outputs 3) Estimate the uncertainty associated with each source Response uncertainty: range across forced responses within each scenario
33. Long-term changes are less sensitive to downscaling uncertainty 33 • Training datasets used for downscaling often exhibit systematic differences • Many downscaling methods preserve the long-term GCM-simulated trends
34. Extreme temperatures are more sensitive to response uncertainty 34 5 0 4 2 1 0 120 - 01) ( - - 01) - 3 .0 15 5 0 4 2 1 0 120 ) ) ) ) ) ) (
35. Extreme temperatures are more sensitive to response uncertainty 35 5 5 5 5 5 5 4 0 - 2 1 0 120 - 01) ( - - 01) - 3 .0 14 4 0 - 2 1 0 120 ) ) ) ( ) ( ) )
36. Robustness checks: NYC extreme rainfall 36 4( 1 4 1 ) -- .. 13 ( 13 3 3 0 14 4( 1 4 1 ) -- .. 13 ( 13 0 23 3 0 14 4( 1 4 1 ) -- .. 13 ( 13 0 23 3 0 14 4(2        
37. Robustness checks: NYC extreme rainfall 37 ) ) ) ( ) ) ) ) ) ( ) ) ) ) ) ( ) ) ) )        
38. Robustness checks: NYC extreme heat 38 4 3 2 1 . 12 1) ( 1) 1 1 - 4 4 3 2 1 . 12 1) ( 1) - 01 1 - 4 4 3 2 1 . 12 1) ( 1) - 01 1 - 4 4 0      
39. Robustness checks: NYC extreme heat 39 ( ( ) ( ( ( ) ( ( ( ) ( (      
40. Robustness checks: NYC extreme cold 40 4 4 4 4 4 3 2 1 . 12 1) ( 1) 1 1 - 3 4 4 4 4 4 3 2 1 . 12 1) ( 1) - 01 1 - 3 4 4 4 4 4 3 2 1 . 12 1) ( 1) - 01 1 - 3 3 0      
41. Robustness checks: NYC extreme cold 41 ) ( ( ) ( ( ( ( ) ( ( ( ( ) ( ( ) ( ( ( ( ) ( ( ( ( ) ( ( ) ( ( ( ( ) ( ( ( ) ( ( ( (      
42. Example: San Francisco 42 4 ) 4 3) 1 0 . 01 0( 30 ( 0 5 5 5 4 ) 4 ) ) 1 0 . 01 30 - 4 4 ) 2 0 30 )      
43. Example: San Francisco changes 43 4 ) 4 3) 1 0 . 01 0( 30 ( 0 4 ) 4 ) ) 1 0 . 01 30 - 5 4 4 ) 2 0 30 )      
44. Example: Chicago 44 4 ) 4 3) 1 0 . 01 0( 30 ( 0 5 5 5 5 4 ) 4 ) ) 1 0 . 01 30 - 4 4 ) 2 0 30 )     
45. Example: Chicago changes 45 4 ) 4 3) 1 0 . 01 0( 30 ( 0 4 ) 4 ) ) 1 0 . 01 30 - 5 4 4 ) 2 0 30 )