the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 23 Oct 2024
Strong inter-model differences and biases in CMIP6 simulations of PM2.5, aerosol optical depth, and precipitation over Africa
Abstract. Poor air quality and precipitation change are strong, rapidly changing, and possibly linked, drivers of physical hazards in sub-Saharan Africa. Future projections of sub-Saharan air quality and precipitation remain uncertain due to differences in model representations of aerosol, aerosol-precipitation interactions, and unclear future aerosol emission pathways. In this study, we evaluate the performance of CMIP6 models in simulating PM2.5, aerosol optical depth (AOD), and precipitation over Africa, relative to a range of observational and reanalysis products, including novel observational datasets, over the 1981–2023 period. While models accurately capture the seasonal cycle of PM2.5 concentrations over most regions, the concentration magnitudes show strong inter-model diversity. Dust AOD shows generally accurate seasonal spatial distribution, with multi-model mean (MMM) pattern correlation coefficients within 0.77–0.94, despite strong inter-model diversity in magnitude. Seasonal spatial patterns of non-dust AOD are poorly represented, with MMM pattern correlation coefficients of 0.25–0.58, and poorest performance during SON. Emission inventory inaccuracies may explain systematic biases for non-dust AOD fields, with differences in circulation and precipitation patterns, and aerosol treatment causing inter-model diversity. Both monsoon regions are generally well captured, though there is poorer performance in simulating the east African monsoon. Biases found relate to the intertropical convergence zone, more apparent over east Africa, and rainfall magnitude, more apparent over west Africa. This evaluation highlights strong inter-model diversity in the representation of African air quality and climate, and identifies model performance over sub-Saharan Africa, and the reasons behind the biases, as critical gaps to address for improving confidence in climate projections.
- Preprint
(14563 KB) - Metadata XML
-
Supplement
(2401 KB) - BibTeX
- EndNote
Status: open (until 04 Jan 2025)
-
RC1: 'Comment on egusphere-2024-3057', Anonymous Referee #1, 25 Nov 2024
reply
Review of “Strong inter-model differences and biases in CMIP6 simulations of PM2.5, aerosol optical depth, and precipitation over Africa”
Toolan et al. evaluate the performance of CMIP6 models in simulating PM2.5, aerosol optical depth (AOD), and precipitation over Africa. The authors provide a thorough analysis using observational and reanalysis datasets, identifying regions and variables where models perform well and where they require improvement. This study has potential to contribute significantly to understanding the reliability and limitations of CMIP6 models for regional applications in Africa. I recommend the manuscript for publication in Atmospheric Chemistry and Physics, subject to the following major and minor comments.
Major Comments
The analysis highlights model biases and performance differences across multiple variables and regions in CMIP6 models, but its scientific contribution and novelty could be improved by addressing certain gaps in linking these variables. Currently, the explanations regarding PM2.5, AOD, and precipitation are treated somewhat independently, and the connections between them are not explicitly discussed. For instance, how systematic discrepancies in the ITCZ influence aerosol transport / air quality and precipitation variability. This exploration could provide essential insights into understanding and identifying (or rule out) the key physical processes to drive the inter-model uncertainties.
The discussion on AOD performance across models is detailed and robust, especially in separating dust and non-dust components. However, it could benefit more from some discussions on following climate impacts, such as aerosol-radiative and aerosol-cloud interactions in the CMIP6 climate models (which have been previously shown to largely affect African climate but with large uncertainties), to further strengthen the impacts of this work.
I have some questions on the motivation of the choice of CMIP6 models for evaluating air pollution, due to their coarse resolution and relatively simplified chemical schemes. While these models are suitable for large-scale climate studies, their limitations in capturing small-scale and chemical processes critical for air quality should be explicitly acknowledged. A discussion on how model resolution and the simplified chemical schemes affects the results, and whether higher-resolution regional models like WRF-Chem might be more appropriate for certain aspects of this analysis, would further strengthen the manuscript.
Minor Comments
Line 12: “well captured” is vague. Specify whether this refers to precipitation magnitude, variability, monsoon onset time, or other variables.
Line 18: “Africa is a region of large heterogeneity in both air quality and precipitation”, 'heterogeneity' refers to spatial heterogeneity?
Line 139: Specify the time period and resolution of CAMS data.
Section 2.1.3: Specify the time period, and resolution. Also, I think the section should be 'precipitation observation and reanalysis datasets' as ERA5 is not an observational data set.
Line 202: provide the definition of the acronym “ITCZ” when it is first mentioned.
Line 210: Define MMR (mass mixing ratio) as it is first time mentioned here.
Line 223: How is pattern correlation calculated? For example, convert the time-averaged 2d date into 1d and then calculate the Pearson correlation coefficient?
Figure 2 caption: NRMSE is not defined until section 3.3.1
Line 293-294: The explanation is a bit simplified here. Different model schemes can produce varying dust emissions, but the situation is more complex because the dust emission is interactively calculated. For instance, even when two models employ the same dust scheme, variations in simulated meteorological conditions (e.g., wind speed, soil moisture, and atmospheric stability) can result in very different dust emissions. It would be helpful to clarify (or comment on) whether the discrepancies arise primarily from differences in the dust schemes themselves, or whether they are predominantly driven by meteorological variability across the models.
Line 306: overestimated in all models?
Line 567-570: This sentence is long and hard to read. Suggest to break down or rephrase.
Line 582:The phrase "Difficulties in replicating the circulation over these regions are also responsible for some issues" could be more precise, e.g., "Challenges in accurately modeling regional atmospheric circulation contribute to these discrepancies"
Generally I would suggest avoiding vague language like "not well captured”, which repeatedly appears in the manuscript several times. Instead, specify the what is the discrepancy, for instance "models fail to replicate the observed seasonal peak in non-dust AOD over central Africa during SON"
Citation: https://doi.org/10.5194/egusphere-2024-3057-RC1 -
RC2: 'Comment on egusphere-2024-3057', Anonymous Referee #2, 21 Dec 2024
reply
Toolan et al. investigate how CMIP6 models perform in simulating PM2.5 levels, aerosol optical depth (AOD), and precipitation over Africa, focusing on east and west Africa. Overall, 56 CMIP6 models are evaluated relative to observational and reanalysis datasets on annual and seasonal scale during the 1981-2023 period. PM2.5 is evaluated relative to the AirNow database, dust and non-dust AOD is evaluated separately relative to CAMS, while precipitation is examined on monthly and daily basis to assess whether CMIP6 models accurately capture precipitation patterns throughout the year over east and west Africa, with focus on the east and west African monsoons. The authors use a variety of metrics to evaluate the performance of CMIP6 models and focus on interpreting their results throughout the manuscript. The results are comprehensive and advance our understanding of the weaknesses of climate models in accurately reproducing the climate state over Africa.
Major Comments
- Lines 44-45: You may also consider a number of recent studies that quantify the effective radiative forcing of anthropogenic aerosols and its decomposition in aerosol-cloud interactions and aerosol-radiation interactions using CMIP6 Earth system models (e.g., Thornhill et al., 2021; Zelinka et al., 2023; Kalisoras et al., 2024).
- In Section 3.2 PM2.5 is evaluated relative to AirNow, a novel yet limited database. Have you considered any other dataset that may extend further back in time or over a larger geographic area in Africa?
- In Section 3.3 the evaluation of seasonal dust and non-dust AOD distributions is presented in Fig. 6. The information obtained from Fig. 6 could be put on a separate table (maybe in the supplement) so the reader would know how well each CMIP6 model performed. The same could be applied to the rainfall results in Fig. 11.
- In Section 3.3 non-dust AOD is evaluated altogether. Maybe an evaluation of black carbon, organic aerosols and sulphates AODs separately could shed light on the cause of poorer model performance for non-dust AOD over Africa.
Minor Comments
Figure 1: As AOD is unitless, the brackets “[1]” in the titles of the four colorbars can be omitted. Same goes for the colorbar titles in Figs. 4 and 5, and the Y-axis labels in Figs. 7 and 8.
Line 77: “Precipitation biases occur spatially” can be changed to “Precipitation biases occur both spatially”.
Line 83: “between the Sahara and Gulf of Guinea discussed in Cook (1999)” could be changed to “between the Sahara and the Gulf of Guinea (Cook, 1999)”.
Line 85: “with a focus on both regions” can be changed to “focusing on both regions”.
Line 99: “demonstrating” can be replaced by “presenting” or “showing”.
Line 101: “regional annual cycles in AOD” could be changed to “regional annual AOD cycles”.
Line 102: “performance of CMIP6 models over Africa”.
Line 102: “model evaluation is performed for CMIP6 models’ seasonal spatial”. Please rephrase.
Table 1: It would greatly benefit this table if a column containing the measurement period for each station were added.
Table 1 caption: “The numbers in the final column”. Also, a dot should be added at the end of this sentence.
Lines 143-144: “MODIS face, where surface brightness over deserts causes a lack of contrast between aerosol signal and the underlying surface brightness (Wagner et al., 2010), or where systematic biases are present when clouds” could be altered to “MODIS face, such as surface brightness over deserts causing a lack of contrast between aerosol signal and the underlying surface brightness (Wagner et al., 2010), or systematic biases being present when clouds”.
Line 148: “(non-dust AOD = total AOD − dust AOD)” could be changed to “(non-dust AOD, i.e., the total AOD minus dust AOD)”.
Line 175: “These datasets were concatenated”.
Line 176: “The list of models used is shown in Table 2, with the nominal atmosphere and ocean resolutions also shown” could be altered to “The models used in the study, along with their nominal atmospheric and ocean resolutions are shown in Table 2”.
Table 2: Please check if the model reference for CAMS-CSM1-0 is correct in the following URL: http://www.climatechange.cn/EN/10.12006/j.issn.1673-1719.2019.186
Figure 2 caption: The acronym NRMSE is first defined in Line 340. It should also be explained here.
Line 276: “analysis in shown later in Sect. 3.3” should be changed to “analysis shown in Sect. 3.3”.
Line 278: Acronyms EAM and WAM are not defined until Line 490. They should be defined here instead.
Line 292: The word “generation” can be omitted.
Figure 4 caption: “(d) intermodel standard deviation for dust AOD (shading) and wind speed”.
Line 381: “during JJA, with westerlies bringing in”.
Lines 395-396: “over east Africa is less well represented than of that over west Africa, with the MMM failing the capture the AOD climatology” could be changed to “over east Africa is not represented as well as over west Africa, with the MMM failing to capture the AOD climatology”.
Line 421: “the areas with the largest intermodel spread”.
Line 431: “models analysed are found”. For consistency with the rest of the manuscript.
Line 432: “biases in the location of rainfall, and magnitude of rainfall in regions” could be changed to “biases in the location and magnitude of rainfall over some regions”.
Line 434: “small areas of localised biases over”. For the same reason as in Line 431.
Line 444: “a ranking of the pattern correlations of each of the models evaluated for each season” could be altered to “a ranking of the pattern correlations of individual models evaluated for each season”.
Line 494: “individual models have pattern”.
Line 519: “daily rainfall characterisation”. For the same reason as in Line 431.
Lines 534-535: “from this section of the analysis”.
Line 541: “a much higher range” could be changed to “a wider range”.
Line 570: “<” should be replaced with “less than”.
Line 574: “intermodel spread caused by”.
Line 583: “inability to correctly model transport of aerosols from biomass burning over central Africa” can be changed to “inability to correctly simulate the transport of biomass burning aerosols over central Africa”.
Lines 599-602: “Therefore, these findings… region of interest.”. This sentence is quite lengthy and can be confusing. I suggest rephrasing it or breaking it into two smaller ones.
References
Kalisoras, A., Georgoulias, A. K., Akritidis, D., Allen, R. J., Naik, V., Kuo, C., Szopa, S., Nabat, P., Olivié, D., van Noije, T., Le Sager, P., Neubauer, D., Oshima, N., Mulcahy, J., Horowitz, L. W., and Zanis, P.: Decomposing the effective radiative forcing of anthropogenic aerosols based on CMIP6 Earth system models, Atmospheric Chemistry and Physics, 24, 7837–7872, https://doi.org/10.5194/acp-24-7837-2024, 2024.
Thornhill, G. D., Collins, W. J., Kramer, R. J., Olivié, D., Skeie, R. B., O’Connor, F. M., Abraham, N. L., Checa-Garcia, R., Bauer, S. E., Deushi, M., Emmons, L. K., Forster, P. M., Horowitz, L. W., Johnson, B., Keeble, J., Lamarque, J.-F., Michou, M., Mills, M. J., Mulcahy, J. P., Myhre, G., Nabat, P., Naik, V., Oshima, N., Schulz, M., Smith, C. J., Takemura, T., Tilmes, S., Wu, T., Zeng, G., and Zhang, J.: Effective radiative forcing from emissions of reactive gases and aerosols – a multi-model comparison, Atmos. Chem. Phys., 21, 853–874, https://doi.org/10.5194/acp-21-853-2021, 2021.
Zelinka, M. D., Smith, C. J., Qin, Y., and Taylor, K. E.: Comparison of methods to estimate aerosol effective radiative forcings in climate models, Atmospheric Chemistry and Physics, 23, 8879–8898, https://doi.org/10.5194/acp-23-8879-2023, 2023.
Citation: https://doi.org/10.5194/egusphere-2024-3057-RC2
Data sets
CHIRPS: Rainfall Estimates from Rain Gauge and Satellite Observations C. C. Funk et al. https://doi.org/10.1038/sdata.2015.66
The Earth System Grid Federation: An open infrastructure for access to distributed geospatial data L. Cinquini et al. https://doi.org/10.1016/j.future.2013.07.002
ECMWF Reanalysis v5 (ERA5) H. Hersbach et al. https://www.ecmwf.int/en/forecasts/dataset/ecmwf-reanalysis-v5
CAMS global reanalysis (EAC4) A. Inness et al. https://doi.org/10.5194/acp-19-3515-2019
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 180 | 56 | 6 | 242 | 14 | 3 | 3 |
- HTML: 180
- PDF: 56
- XML: 6
- Total: 242
- Supplement: 14
- BibTeX: 3
- EndNote: 3
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1