Restricted (Penn State Only)
Diaz Isaac, Liza Ivelisse
Graduate Program:
Doctor of Philosophy
Document Type:
Date of Defense:
March 16, 2017
Committee Members:
  • Kenneth J. Davis, Dissertation Advisor
  • Kenneth J. Davis, Committee Chair
  • Fuqing Zhang, Committee Member
  • Chris E. Forest , Committee Member
  • Andrew Carleton, Outside Member
  • Thomas Lauvaux, Dissertation Advisor
  • Natasha Miles, Committee Member
  • Carbon Dioxide
  • Atmospheric Transport Errors
  • Ensemble
  • WRF
  • Multi-Physics
  • Genetic Algorithm
  • Simulated Annealing
  • Calibration
  • Down-Selection
  • Atmospheric Inversion
Atmospheric inversions are used to assess biosphere-atmosphere CO2 surface exchanges at various scales. Recently, higher resolution inversions were performed using mesoscale models to improve the spatial and temporal resolution of these inversions, but variability among inverse flux estimates remains significant. One of the main contributors to these uncertainties is the atmospheric transport model errors. Past studies have used ensembles to understand these transport model uncertainties, but have some limitations including the small number of measurements, coarse resolution of the models, small number of members, centered only in the variation of Planetary Boundary Layer (PBL) schemes and no assessment was performed to test whether the spread of the ensemble represents the true uncertainty. For this work, I evaluate and quantify the transport model errors with a large set of simulations generated with the Weather Research and Forecasting (WRF) mesoscale model. The large ensemble of 45-members was constructed using different physics parameterizations (i.e., land surface models (LSMs), planetary boundary layer (PBL) schemes, cumulus parameterizations and microphysics parameterizations) and initial/boundary conditions. All the different models were coupled to CO2 fluxes and lateral boundary conditions from CarbonTracker to simulate CO2 mole fractions. I evaluate the atmospheric transport errors over a highly instrumented area, the Mid-Continental Intensive (MCI) region, for 2008 summer period. Both modeled meteorological variables (i.e., wind speed, wind direction and PBL height) and CO2 mixing ratios are compared to observations to evaluate the performance of the different models and the ensemble. In Chapter 2, I performed statistical analyses to evaluate the impact of both physics parameterizations and the meteorological dataset on CO2 mixing ratios and meteorological variables. The different model configurations show varying performances across the region that impede the selection of an optimal solution or least biased simulation for all the meteorological variables except for PBL height (PBLH). In general, physical parameterizations contribute equally to the model-to-model variability in atmospheric CO2 and meteorological variables, with the microphysics parameterization being the exception. It was also found that daily variations in CO2 mole fractions across the region are correlated primarily with errors in the PBLH. In Chapter 3, I introduce two calibrations (or down-selection) methodologies using Simulated Annealing (SA) and Genetic Algorithm (GA) over 2008 summer. I applied the calibration process to the multi-physics/multi-analysis ensemble of 45-members to select the optimal ensemble using the flatness of the rank histogram as the main criteria. The calibrated ensemble representing the model errors is based on all three meteorological variables. Using multiple model configurations (i.e. 45 configurations of varying physics), I show that a reduced number of simulations (less than 10 members) is sufficient to characterize the transport errors, reproducing the statistics of the model-data differences while minimizing the size of the ensemble. The CO2 error correlations of the calibrated ensembles were compared to the large ensemble to identify any impact of the calibration. Compared to the initial error structures, the calibrated ensembles revealed sampling noise across the region which indicates that additional filtering or modeling of the errors would be required to construct the error covariance matrix for regional CO2 inversion. Using the multi-physics and multi-analysis ensemble, I showed the importance that other physics parameterization besides the PBL schemes has on the atmospheric CO2 mixing ratio errors. In addition, the challenges that future atmospheric inversions still need to confront including correction of systematic biases and representation of errors, to avoid the propagation of these errors into inverse fluxes.