Climate envelopes and uncertainty

• 1 Main points from Bruce Hewitson's presentation: [where? when?]

We start with some assumptions:

• there is need for people to be able to access and appropriately interpret credible climate information so as to avoid people assuming either extreme views of climate change - crisis or unfounded rumour

• there are a number of plausible, credible and reasonable climate futures that could evolve but only one will become a reality, however we need to work with a number of possibilities

• it is true that not everything is certain BUT not everything is uncertain either and we need to be able distinguish between these and communicate them appropriately (some of these uncertainties are scale dependent)

• the question becomes how does one effectively communicate the nuances and sometimes conflicting but important information

• it is important that stakeholders gain a good understanding of the climate baseline, that they are able to interpret projected trends and that there is a sustained flow of information between climate scientists and users - this process is time intensive (need resources to enable that) and usually needs to be communicated in person (the wiki will have limitations in facilitating this)

Of late we are having an active discussion on this topic, mostly in emails, a few meetings and a lot of project work. The main issues are:

• Can we a priori restrict the number or range of climate scenarios that an analyst should look at? Myles Allen suggested three views (last year in a meeting in Oxford, but I'm paraphrasing from memory):

--Rationalist: There is no rational (verifiable, proven, transparent) way to decide if a climate scenario, whether global or local, is good or bad, better or worse than another. This is the strict view of many climatologists: it will take some decades before the evolution of climate change will constrain model projections of the future.

--Subjectivist: The Bayesian practitioners point to scenarios as decision tools and accept that a reasonable panel of climate experts should be expected to assign probabilities to each projection (or sometimes to the outcomes). There is no real empirical way to test whether such probabilistic distributions are 'right', 'acceptable' or indeed 'misleading'. The IPCC maintains that all scenarios are equally likely (or more accurately, that the scenarios do not come with a priori probabilities). However, the IPCC also uses a rating of confidence and uncertainty in its final statements, so there is a logical disconnect there! This approach does assume that decision making needs probabilities to handle uncertainty (which is indeed misleading).

--Realist: Decision makers need some guidance. Pick a good scenario doesn't work. Take three (high, low and best guess) is not credible any longer. But equally, look at all of them (that could be thousands if you include the large ensemble runs and millions if you go beyond GCMs!) might not be helpful. (The weADAPT approach examines this sequence of dealing with uncertainty elsewhere.) So, is there an ad-hoc way to reduce the range of results? This mostly involves dropping some of the upper end outcomes: can this decision context deal with a global change of 12 degC? Or, sometimes we find climatologists scratching their heads: 'that GCM is not very good', meaning it doesn't capture the present climate very well or has some model flaws that are undesirable. However, reducing this 'head scratching' to a quantitative rule is not fully sorted out.

The real questions for an adaptation science are:

• Do we have a way to capture vulnerability and the decision context (quite different domains!) that inform the selection of scenarios?

• Can we map the adaptation landscape (the outcome space) in a way that deciphers the value of information implied in the range of scenarios?

• When the decisions do require robust decision making, do we have a way to a priori screen the range of climate futures we might expect to include?

Just trying to get some collective thinking on this topic, as we plan the redesign of weADAPT 3.0...

By the way:

• Our metaphor of climate envelopes needs updating. In most cases, the chart of future climate changes is more akin to the unraveling of a braided rope. Or if we take the Adaptation Landscape as our metaphor, then the climate scenarios are more like a cloud (maybe smog) that obscures some features and focuses attention on others. Anyone with a good metaphor?

Further notes are from Bruce Hewitson's presentation from a workshop in 2007:

--Tom Downing 21:10, 14 July 2009 (CEST)

Main points from Bruce Hewitson's presentation: [where? when?]

Background basics

• climate defines the range of weather events experienced in a region
• climate is a complex interconnection of many global and local processes - the fundamental physics and dynamics are well characterised; the 'self organising criticality' of the climate system has major gaps in sensitivity, non-linearity, teleconnections, thresholds and hysteresis, etc.
• climate change can be expressed though changes in seasonality, intensity, frequency, the average, mode, etc.
• (un)certainty has spatial, temporal and parameter dependence e.g. globally simple, regionally complex

Historical change

Detection and assessment subject to data limitations a) use quality controlled station observations b) interpolate carefully if possible c) rigorously assess trends for direction, magnitude and significance There is an urgent need to extend, coordinate and collate data records and observing systems

Future projections

There are four sources of information within which one looks for commonality on which to base ones understanding of future climate

1) historical trends (which are not a guarantee of future trend)

2) understanding of physical processes governing regional climate responses

3) information from GCM simulations of the future (coarse resolution)

4) downscaling of GCM data to regional scales

Continental warming

Observed patterns only simulated by models that include anthropogenic forcing

• decades
• coverage
• sampling of models
• observation uncertainty
• Antarctica

Africa

• a mixed picture of uncertainty and agreement
• regions of low consensus related in part to spatial positioning of boundaries of the climate processes

Approach taken at UCT

a) characterise baseline observational climate as best as possible

• global station data
• Africa gridded data
• measured parameters + derivatives (e.g. dry spell duration)

b) characterise process change to inform understanding

• consider circulation change as a means to gain confidence in location-specific climate

c) use as many models as possible

• ideally focus on AR4 suite of model simulations

d) downscale where possible

• RCM downscaling still problematic, empirical more robust

Current first order priority products

• emphasis on precipitation, T max and min and derivatives (dry spell, raindays > threshold, T > or < threshold, frequency and seasonality of events, etc.)
• analysis of past trend and variability (where possible)
• analysis of circulation - somewhat technical but still accessible
• maps of seasonal climatology: past; GCM control; future
• raw date of mean anomaly (future - present)
• 12 GCMs from AR4 archive
• 2 RCMs - limited work on this
• downscaling to grid and station scale from GCMs
• focus on median and percentiles rather than labels of models
• dissemination from web portal
• support documentation on caveats, limitations, assumptions, guidelines of use, etc.