Taskforce IV: Treatment, quantification and integration of uncertainties in CarboEurope-IP Component uncertainties (Inventory, Eddy fluxes, Atmosphere measurements,…)

European carbon balance uncertainties

Bottom-up modelling

Top-down modelling

Propagation, CCDAS

Objectives• The overall goal of this task force is to develop a coherent strategy

of how uncertainties in CarboEurope have to be treated in order to achieve a scientifically defensible estimate of the European carbon balance and the associated uncertainties at different temporal and spatial scales”.

• 1. Sectoral component: – Common definitions– Guidelines for quantification– Importance ranking of uncertainties Recommendations for strategies to reduce uncertainties

• 2. Integrative component:– Multiple constraint approach make use of the complementary

information in the different data streams– Analysis of data flow between components– Define UA/UQ in bottom-up modelling

General considerations IDefinitions

• Uncertainty: the state of being unsure of something

• In field science (ISO 1995 - the GUM ): “Uncertainty: parameter, associated with the result of a measurement, that characterizes the dispersion of the values that could reasonably be attributed to the measure”

• The uncertainty in the result of a measurement arises from the remaining variance in the random component and the uncertainties connected to the correction for systematic effects (ISO 1995).

General considerations II(Importance ranking)

• When we are considering a ranking of uncertainties within the different sectors reduced the following general equations should be considered:

• Importance of Uncertainty = Magnitude * Sensitivity of goal value

• Efficiency of reduction = Magnitude * Sensitivity * ‘Cost’ per Reduction of Magnitude

General considerations IIICharaterization of uncertainties

• Spatial characteristic (scale, domain, absolute values versus gradients)

• Temporal characteristic to considered (mean fluxes, trend, interannual variability, seasonal, synoptic, temporal domain)

• Type of uncertainty (random, systematic, autocorrelation, scaling/aggregation, total)

• Quantity of interest (GPP, NPP; NEP, NBP….)

General considerations IVCombination of uncertainties

‘Truth’

Method A

Method B

Method A: good a variability (also different scales!)Method B: good at mean

General considerations IVCombination of uncertainties

High precision,Spatial coverage

High precision, high temporal resolution

Provide understanding

Extrapolation cap., incl. of history

Large-scale constraint

Spatial/temporally consistent data, stochastic events

Session planTuesday, 11:15-13:00

1. Overview about the taskforce objectives (Reichstein/Smith/Wattenbach/Gerbig)

2. Uncertainties in inventories (Luyssaert)3. Uncertaintes in flux estimates (Aubinet)´4. Uncertainties in carbon balances inferred from atmospheric

measurements (Rödenbeck/Peylin/Schumacher)5. Integration of uncertainties in bottom-up modeling (Wattenbach)6. Bottom-up modelling: Model input&structure uncertainties

(Jung)7. Bottom-up modelling: Parameter uncertainties (Zaehle)8. Bottom-up modelling: Scaling-aggregation-representation

uncertainties (Tenhunen) 9. Integrating and propagating uncertainties via CCDAS (Rayner)10. Uncertainty quantification and analysis (UQ/UA) in NitroEurope

(NEU) (van Oijen/Smith)

Superficially: need of CE-IP to provice uncertaintes (contract)

Also clear: we need distributions instead of point estimates

Show CE-IP integration (multiple-constraint) slide

There is a new view emerging: there no ‘validation’ of models or

methods, but only via a new combination of methods with their uncertainties we can reduce those