Uncertainty from climate models can be reduced from "above" - how scientists report on it. There are three main sources of uncertainty:
1) Reflexive uncertainty from unknown future greenhouse gas emissions and climate-society feedbacks.
2) Epistemic uncertainty from incomplete representation in climate models and missing processes. Reducing this is a long term goal.
3) Aleatoric uncertainty from internal climate variability that is partially chaotic. For temperature, epistemic uncertainty dominates further in the future, while aleatoric uncertainty remains constant; thus constraining epistemic uncertainty is important. However, for rainfall over India, observational uncertainty is also significant, presenting a barrier to reducing model uncertainty.
Krishna AchutaRao - Uncertainty from above - can it be reduced?
1. Uncertainty from above: can it
be reduced?
Krishna AchutaRao
Indian Institute of Technology Delhi
2. Outline
• Uncertainty from “above”
– How have scientists reported it?
– Spatial & temporal scales
• Can we identify and quantify the various
sources of uncertainty
• Can we reduce or constrain uncertainty?
• Some inferences
Singh, R., and K. M. AchutaRao “Quantifying and Constraining Structural Uncertainty in
Future Climate Projections over India”, GC43C-1218, American Geophysical Union Fall
Meeting 2015. San Francisco, CA USA 14-18 December 2015.
3.
4. (IPCC AR5 Working Group – I, FAQ 12.1, Figure 1) Global mean temperature change averaged across all Coupled Model
Intercomparison Project Phase 5 (CMIP5) models (relative to 1986–2005) for the four Representative Concentration
Pathway (RCP) scenarios: RCP2.6 (dark blue), RCP4.5 (light blue), RCP6.0 (orange) and RCP8.5 (red); 32, 42, 25 and 39
models were used respectively for these 4 scenarios. Likely ranges for global temperature change by the end of the 21st
century are indicated by vertical bars. Note that these ranges apply to the difference between two 20-year means, 2081–
2100 relative to 1986–2005, which accounts for the bars being centred at a smaller value than the end point of the annual
trajectories. For the highest (RCP8.5) and lowest (RCP2.6) scenario, illustrative maps of surface temperature change at the
end of the 21st century (2081–2100 relative to 1986–2005) are shown for two CMIP5 models. These models are chosen to
show a rather broad range of response, but this particular set is not representative of any measure of model response
uncertainty.
5. Figure 7. CMIP5 model ensemble
mean precipitation change (%)
projected for 2030s (2021–2050),
2060s (2046–2075)
and 2080s (2070–2099) relative to
the pre-industrial period (1880s,
i.e. over 1861–1900).
Chaturvedi et al.,
Current Science 2012
6. Chaturvedi et al.,
Current Science 2012
Figure 3. CMIP5 model-based time series of temperature and
precipitation anomalies (historical and projections) from
1861 to 2099 relative to the 1961–1990 baseline for the RCP
scenarios. Shaded area represents the range of changes projected by
the 18 models for each year
7. IPCC, 2013: Annex I: Atlas of Global and Regional Climate Projections [van Oldenborgh, G.J., M. Collins, J.
Arblaster, J.H. Christensen, J. Marotzke, S.B. Power, M. Rummukainen and T. Zhou (eds.)]
8. Reflexive Uncertainty
“Reflexive: Scenario uncertainty” This is linked to
the fact that the 21st Century GHG emission
trajectory is unknown and involves unpredictable
climate–society feedbacks as well as non-climate
related (e.g., political and economic) factors (also
not predictable).
9. Epistemic Uncertainty
“Epistemic: of or relating to knowledge or
knowing” This is related to the lack of knowledge
in the model representation of physical and
dynamical processes including missing processes
and/or interactions within climate models. Also
called structural uncertainty or model
uncertainty/error. Reduction of this uncertainty is
a key long-term objective.
10. Aleatoric Uncertainty
“Aleatoric: characterized by chance or
indeterminate elements” This is related to the fact
that the climate (defined as a long-term mean) is
partially chaotic due to unpredictable internal
variability.
11. What do the uncertainties look like?
Global mean
temperature
East Asia
decadal mean
JJA rainfall
Figure 2 (Adapted from IPCC
AR5 Working Group – I, Figure
11.8) Sources of uncertainty in
climate projections as a
function of lead time based on
an analysis of CMIP5 results.
(Top). The fraction of variance
explained by each source of
uncertainty for East Asian (5°N
to 45°N, 67.5°E to 130°E)
decadal mean boreal summer
(June to August) precipitation
(Bottom).
Also see:
Hawkins & Sutton (2011)
12. Summer Temperature Change Over India
Aleatoric and
epistemic
uncertainty in
projected summer
(June-July-August)
temperature change
under RCP8.5. The
aleatoric (top row)
and epistemic
(bottom row)
uncertainties are
shown for 4 different
20-year periods into
the future from left
to right: 2020-2039
(left most column),
2040-2059, 2060-
2079 and 2080-
2099.
13. Focusing on temperature change in the
Central North East India region
Parthasarathy et al 1994: All-India
Monthly and seasonal rainfall series
1871-1993. Theor. Appl. Climatol, 49,
217-224
15. Winter rainfall change over Peninsular
India (RCP4.5 and RCP8.5 scenarios)
RCP 4.5 RCP 8.5
16. Where Aleatoric Uncertainty Exceeds
Epistemic Uncertainty
• We may not be able to reduce the overall
uncertainty any further
• Since aleatoric uncertainty does not grow over
time,
– we can use current information on variability to
inform future uncertainty.
– past knowledge of coping with variability may play
an important part in adaptation strategies.
17. Where Epistemic Uncertainty Exceeds Aleatoric,
can we Constrain it?
• Model democracy
• How to weight models?
– “Good” models get higher weights
– What does a “good model” mean?
– How does one compare models to observations?
• Our research indicates that model weighting
schemes do not constrain epistemic
uncertainty over India
• Observational uncertainty matters more
18. Collins, AchutaRao, Ashok, Mitra, Prakash,
Srivastava, Turner, Observational
challenges in evaluating climate models,
Nature Climate Change, 2013
Observational Uncertainty
• How good are our
observations?
• Fundamental barrier to
reducing model uncertainty