Skip to main content
. Author manuscript; available in PMC: 2023 Aug 30.
Published in final edited form as: Atmos Chem Phys. 2021 May 27;21(10):8127–8167. doi: 10.5194/acp-21-8127-2021

Figure 3. Assessment of the effectiveness of the inverse model in reducing errors against observationally-informed constraints on regional dust aerosol optical depth (DAOD).

Figure 3.

Figure 3.

(a-d) Comparisons of the 15 observational constraints on regional DAOD (purple squares) against the inverse model results (blue circles) and the models in our ensemble (brown numbers; 1 = CESM, 2 = IMPACT, 3 = GISS ModelE2.1, 4 = GEOS/GOCART, 5 = MONARCH, 6 = INCA) for each of the four seasons. Results are grouped by the major source region nearest to each of the observed regions. Also listed are the root-mean-squared errors for each regional group for both the inverse model and model ensemble results, and the reduced chi squared metric (χv) for the comparisons of the inverse model results against all 15 DAOD constraints. Error bars denote one standard error. (e) Taylor diagram summarizing the statistics of the comparison against the seasonally averaged regional DAOD constraints for the different models (Taylor, 2001). The different symbols represent the measurements (purple triangle), the 13 AeroCom models (black letters; A = CAM, B = GISS ModelE, C = GOCART, D = SPRINTARS, E = MATCH, F = MOZGN, G = UMI, H = LOA, I = UIO_CTM, J = LSCE, K = ECHAM5, L = MIRAGE, M = TM5), the MERRA-2 dust product (red “R”), the six models in the model ensemble (brown numbers, as for panels a-d), the six improved model results (green numbers with a prime), and the inverse model results (blue star). The horizontal axis shows the standard deviation of the data set or model prediction, the curved axis shows the correlation, and the grey half-circles denote the centered root-mean-squared difference between the observations and the model predictions. As such, the distance between a model and the observations is a measure of the model’s ability to reproduce the spatiotemporal variability in the observations; Taylor diagrams do not capture biases between model predictions and observations. (f) Same as panel (e), except showing a comparison against the annually averaged regional DAOD constraints.