Uncertainty in eddy covariance datasets Dario Papale, Markus Reichstein, Antje Moffat, Ankur Desai, many others..

Uncertainty in eddy covariance datasets

Dario Papale, Markus Reichstein, Antje Moffat, Ankur Desai, many others..

Raw data (20 Hz)

Half hourly data

Half hourly, daily, monthly, annual data

Filtering (u*, spike, qc) and corrections (storage)

Errors or uncertainties??

Gap filling

Partitioning

A

BC

Footprint problems

Advection

Tower setup

ADVEX

Goeckede et al. in prep

IMECC

Corrections, filtering, etc.

IMECC

Random errors

Richardson et al.

This is not uncertainty,this is an error

This is uncertainty

Sites and years used

Papale et al. 2006

A – Uncertainty due to quality check and filtering

Difference between minimum and maximum value obtained at different time resolution using different correction setting (u* thresholds, spike thresholds, storage measurement)

NEE

Papale et al. 2006

A – Uncertainty due to quality check and filtering

Uncertainty due to different corrections settings (u* thresholds, spike thresholds, storage measurement) at annual scale.

< 100 gC m-2 (50 gC m-2)

Papale et al. 2006

A – Uncertainty due to quality check and filtering

GPP

Uncertainty due to different corrections settings (u* thresholds, spike thresholds, storage measurement) at annual scale.

~/< 100 gC m-2 (5-10%)

Papale et al. 2006

A – Uncertainty due to quality check and filtering

TER

Uncertainty due to different corrections settings (u* thresholds, spike thresholds, storage measurement) at annual scale.

~/< 100 gC m-2 (5-10%)

Moffat et al. 2007

B – Uncertainty due to gapfilling (15 methods, 50 artificial gaps scenarios)

RMSE for different sites and different methods (50 scenarios)

Richardson et al. 2006

B – Uncertainty due to gapfilling & random errors in the half hourly data

Richardson et al. estimated the random errors in eddy covariance measurements comparing data acquired by two systems in the same footprint and also comparing half hourly data acquired at the same site, under the same meteorological conditions but at different time.

Random error frequency distribution for three different US sites (double-exponential distribution)

Moffat et al. 2007

B – Uncertainty due to gapfilling

MAE boxplot of the different techniques and random uncertainty estimation using Richardson et al. method

B – Uncertainty due to gapfilling

Moffat et al. 2007

RMSE in function of different methods and gaps length

At annual bases the average uncertainty introduced by the “good” methods has been estimated to be +/- 25 gC m-2 year-1

Desai et al. 2007, in press

C – Uncertainty due to partitioning (23 methods, 10 artificial gaps scenario)

GPP and RE boxplot for each dataset using all the methods. Large part of the methods are in about 100 gC m.2 yr-1

Desai et al. 2007, in press

C – Uncertainty due to partitioning

Annual sun bias due to artificial gaps. Each boxplot is based on 10 gaps scenarios

Desai et al. 2007, in press

C – Uncertainty due to partitioning

GPP and RE monthly boxplot

RE GPP

Conclusions

Big effort is ongoing to assess uncertainty in the eddy covariance measurements in CarboeuropeIP and other projects.

There are uncertain assumptions in all steps of data acquisition and processing

Standardization of data processing helps to reduce uncertainty particularly in multi sites synthesis analysis

There is still a lot to do in the uncertainty definition due for example to advection and footprint

We need to discuss with the modeling community about how to incorporate the uncertainty in the measurement in the model parameterization