(2011), History of climate modeling. WIREs Clim Change, 2: 128-139 • Climate change presents risks to human and natural systems o These risks manifest at local scales • Coupled general circulation models (GCMs) can help quantify these risks
Scenario uncertainty Model uncertainty Internal variability Sources of uncertainty in climate projections • Future greenhouse gas emissions trajectories https://www.dkrz.de/en/communication/climate-simulations/cmip6-en/the-ssp-scenarios
Sources of uncertainty in climate projections • Different GCM structures and parameters • Chaotic nature of the Earth system Scenario uncertainty Model uncertainty Internal variability • Future greenhouse gas emissions trajectories
temperatures in Ithaca Scenario uncertainty: variance across scenarios of the multi-model mean Model uncertainty: variance across models, averaged over scenarios
scenarios Sources of uncertainty in climate projections Variance decomposition: average temperatures in Ithaca Model uncertainty: variance across models, averaged over scenarios Scenario uncertainty: variance across scenarios of the multi-model mean
need for localized climate information • Many end-users require higher-resolution information • Coarse resolution can contribute to biases related to: o Subgrid-scale processes o Heterogeneity in surface properties
information o Dynamical: high-resolution regional model forced via GCM boundary conditions o Empirical/statistical: spatial disaggregation, regression or analog methods, machine learning
information o Dynamical: high-resolution regional model forced via GCM boundary conditions o Empirical/statistical: spatial disaggregation, regression or analog methods, machine learning Bias-correction aims to correct systematic biases o Often a quantile mapping approach Pierce, D. W., et al., Improved Bias Correction Techniques for Hydrological Simulations of Climate Change. J. Hydrometeor (2015)
information o Dynamical: high-resolution regional model forced via GCM boundary conditions o Empirical/statistical: spatial disaggregation, regression or analog methods, machine learning Ø Both depend on training (observational) data Bias-correction aims to correct systematic biases o Often a quantile mapping approach Ø Many impact or risk assessments rely on one downscaled & bias-corrected ensemble
projections into climate resilience planning in U.S. cities, 2022 Environ. Res.: Infrastruct. Sustain. 2 045006 Ø 16 of 21 (76%) climate adaptation plans rely on downscaled climate projections Ø None rely on raw GCM outputs Downscaled ensembles are widely used • Impact assessments • Public sector decision-making
mean Model uncertainty: variance across models, averaged over scenarios and ensembles Internal variability: magnitude of residuals, averaged over models, scenarios, and ensembles Ø Downscaling uncertainty: variance across ensembles, averaged over models and scenarios Variance decomposition: Ithaca average temperatures Characterizing uncertainty associated with downscaling & bias-correction
analysis at global scale ü Suite of climate metrics measuring climate averages and extremes ü Downscaled & bias-corrected ensembles from CMIP6 • Downscaling and bias-correction uncertainty is relatively more important: o over the near-term o in projections of climate extremes o in projections of precipitation Ø Lafferty, D. & Sriver, R., Downscaling and bias-correction contribute considerable uncertainty to local climate projections in CMIP6, npj Clim Atmos (2023)
bias-correction uncertainty can represent considerable sources of uncertainty: o over the near-term o in projections of climate extremes o in projections of precipitation o in regions where observations disagree
bias-correction uncertainty can represent considerable sources of uncertainty: o over the near-term o in projections of climate extremes o in projections of precipitation o in regions where observations disagree • Additional insights: o Threshold metrics: hot days, wet days, dry days, hot & dry days o Temporally compounding extremes o Large observational disagreements over coastal areas
downscaled ensembles may be necessary to characterize the full range of possible climate futures o Doing so may significantly increase data and computational requirements!
& precipitation drivers LOCA ü Initial condition sampling ü Several scenarios and models OakRidge ü Dynamical & statistical downscaling ü Trained on multiple observational datasets + others... UNH Water Balance Model Ø Parametric uncertainty Statistical yield model Ø Structural uncertainty Ø Parametric uncertainty • Sensitivity analysis • Generalizable insights Hydrology Soil moisture simulation Agriculture US maize outcomes • Lafferty et al. (in prep.) Combined sensitivity analysis of climate, hydrologic, and agricultural modeling uncertainties for US maize yield projections • Alipour et al. (in prep.) Identifying the key parameters driving uncertainty in crop yield prediction
david0811
morganepeng
pedronauck
mza
bermonpainter
stephaniewalter
sergeychernyshev
akosma
geeforr
tammyeverts
samlambert
caitiem20
cromwellryan
