Investigating covariances in and for carbon dioxide surface flux inversions
Open Access
- Author:
- Wesloh, Daniel Robert
- Graduate Program:
- Meteorology and Atmospheric Science
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- December 03, 2021
- Committee Members:
- Klaus Keller, Outside Unit & Field Member
Sha Feng, Major Field Member
Kenneth Davis, Chair & Dissertation Advisor
David Stensrud, Program Head/Chair
Steven Greybush, Major Field Member
Thomas Lauvaux, Special Member - Keywords:
- carbon dioxide
atmospheric trace gas flux inversions
data assimilation
covariance assumptions
temporal biogenic CO2 flux error correlations - Abstract:
- Current CO2 flux inversion systems use matrix representations of the spatial correlations and do not include correlations between the parts of the daily cycle. We set up a framework to test these assumptions and compare stochastic and deterministic representations of the posterior CO2 flux uncertainty. Producing deterministic posterior uncertainty matrices at reduced resolution with the same transport error as the full inversion produced uncertainty estimates that were lower than expected, and became lower still with coarsening resolution. Stochastic estimates of the posterior CO2 flux uncertainty provide similar information to the deterministic estimates at full resolution in the ideal case, and become more variable as the number of ensemble members used to construct the stochastic estimate decreases. We then investigate the temporal biogenic CO2 flux error correlations using the difference between eddy covariance and terrestrial carbon cycle estimates of the CO2 flux, construct a family of temporal correlation functions to describe these data, and recommend a member of that family for inversions. The new temporal correlation function performs as well as two correlation functions used in previous regional inversions at matching previously-published estimates of the uncertainties in the hourly CO2 fluxes at the site level and in the annual fluxes at the continental-average level at the same time. However, neither the new nor the existing correlations were able to match previously-published estimates of the uncertainty in the monthly flux at the continental-average scale. The investigation indicates that much of the room for improving the prior mean flux estimates from terrestrial carbon cycle models lies in improving the daily cycle, either within an inversion or in the terrestrial carbon cycle models. We integrate the new correlations into a pseudo-data experiment to see whether the new correlations perform better than previously-used correlations from the literature in a best case scenario. The new correlations recover the ``true'' continental-average CO2 flux better than the correlations from previous inversions in the ideal case. The new CO2 flux error temporal correlation functions merit further investigation for suitability for real-data inversions.