Download presentation
Presentation is loading. Please wait.
Published byWinifred Rodgers Modified over 5 years ago
1
POLDER vs HadGEM3: Capturing the variability of aerosol direct radiative effect Ben Johnson, Nick Davies1, Fanny Peers1, Jim Haywood1 1 University of Exeter Hi I’m Ben Johnson, I work at the Met Office Hadley Centre, my main research is on aerosol radiation interactions and absorbing aerosol This is a piece of work that myself and Nick Davies worked on together. Many of the plots are from his PhD thsis. Nick just passed his PhD viva yesterday, so this work is obviously up to scratch. I’d also like to thank Fanny and Jim for their input on this work. © Crown copyright Met Office
2
BB plume over SE Atlantic has large direct impact on radiation budget
Direct Radiative Effect (DRE) of all aerosols at TOA from HadGEM3: Mean for Aug & Sept 2006 HadGEM3, Aug-Sept 2006 Models must capture: 1) Cloud deck 2) Aerosol plume 3) Elevation above cloud 4) Absorption properties The BBA plume over the SE Atlantic has a large direct impact on the radiation budget This the direct radiative of all aerosol at TOA from HadGEM3, showing the mean for Aug-Sept 2006 The BB plume over the SE Atlantic sticks out like a sore thumb with TOA effect of up to 20Wm-2 (during the dry season). Not all GCMs get positive forcing here, 4/16 AEROCOM 1 get –ve. HadGEM2-ES had very weak +ve effect, only few Wm-2 at most. Models only get +ve if they capture the cloud deck (HadGEM3 simulates low cloud fractions around 60-80% over this region), the aerosol plume (HadGEM predicts fine-mode AODs in the range 0.2 – 0.7 over the ocean), the elevation of the aerosol (black line) above the cloud (blue line), and have sufficiently absorbing aerosol (in HadGEM3 GA7.1 we upgraded the absorption by BC, which is an important factor in simulating the +ve forcing in this region). Here we see HadGEM3 captures low cloud fractions of 60-80% and fine-mode AODs ranging from 0.2 – 0.7 BB plume sticks out like a sore thumb! Not all GCMs get the positive forcing here (e.g. Zuidema et al., 2016). © Crown copyright Met Office © Crown copyright Met Office
3
New retrievals for aerosol above-cloud
New algorithm retrieves the properties of clouds and the absorbing aerosols above them from POLDER (Peers et al., 2015, ACP). Until recently there were very few observational constraints to evaluate modelled DRE in this region. However, there are now a few satellite algorithms, such as POLDER that have been developed to retrieve the properties of absorbing aerosol layers above cloud. Developed by Fanny Peers, the new algorithm retrieves the properties of clouds (COD) and the absorbing aerosols (AOD, SSA, Ang) above them from POLDER. It then also provides a estimate of the instantaneous aerosol radiative effect for the scenes with absorbing aerosol-above cloud. The above-cloud aerosol DRE has a very broad and highly skewed distribution with values ranging from ~-10 to 150Wm-2 Algorithm provides estimate of direct radiative effect (DRE) for the above-cloud aerosol © Crown copyright Met Office
4
Exploring the problem with SOCRATES offline radiative transfer
TOA aerosol DRE SOCRATES set up with absorbing aerosol* above stratocumulus Sign and magnitude of DRE is highly dependent on Cloud optical depth (COD) DRE switches to positive for COD >~3 The AOD becomes more important at higher COD (e.g. >~10) Before we start comparing HadGEM3 with POLDER, lets briefly explore the problem with SOCRATES offline rad trans model. Set up SOCRATES with an absorbing aerosol layer above StCu then you find that the DRE of the aerosol is highly dependent on the COD. For this set up DRE switched sign to positive for COD > 3. As you go to higher COD the AOD becomes increasingly important. However, from further analysis Nick showed that the variability of COD was more important in governing the variability of DRE in the SE Atlantic region. Finally, the SSA is also critical to the sign and magnitude of DRE. In these calculations the SSA was around 550nm. In earlier tests with less absorbing aerosol (original CLASSIC based on SAFARI) the DREs were much lower and remained negative until COD exceeded ~10. *Aged BB (CLASSIC GA8), SSA550~0.82 Cloud reff =10um SZA = 0.
5
New retrievals for aerosol above-cloud
New algorithm retrieves the properties of clouds and the absorbing aerosols above them from POLDER (Peers et al., 2015, ACP). Algorithm provides estimate of direct radiative effect (DRE) for the above-cloud aerosol
6
Now for some climate simulations
Three simulations with HadGEM3-GA7.1: N96 (140km), N216 (60km), N512 (25km) Each run nudged with ERA-40 reanalyses for July-Sept 2006 Emissions from CMIP6 (BB = GFED4s) Hourly diagnostics 12-15Z for comparison with POLDER overpasses Now for some climate simulations. We ran 3 simulations with HadGEM3-GA7. at 3 resolutions: 140km, 60km, 25km. Each runs was nudged with ERA-40 reanalyses for July-Sept 2006 to take advantage of some POLDER retrievals that had already been processed for this season in Peers et al. (2016) Emission were all from CMIP6… I output hourly diagnostics (various cloud, aerosol and radiation diagnostics) from 12-15Z to facilitate comparisons with POLDER retrievals collocated with the overpasses. © Crown copyright Met Office
7
DRE estimated from POLDER and HadGEM3
Model sampled for POLDER overpass times and COD > 3 Modelled DRE lower than POLDER but PDF almost as broad Model resolution makes little difference to mean or variability of DRE!? POLDER HadGEM3 N96 HadGEM3 N216 HadGEM3 N512 So here are the DRE estimates from POLDER and HadGEM3 for Aug-Sept 2006 and a sample region 20S-0, 10W -10E) As we saw in the introduction POLDER gives a very broad skewed PDF when looking at all space-time points. When averaged over time the DRE is positive everywhere in this domain. HadGEM3 output was sampled for POLDER overpass times and with equivalent sampling criteria including a COD threshold of 3. The model gives a similar distribution but is shifted to slightly lower values on average and with less extreme high values. The model also predicts positive DRE everywhere and with a similar spatial distribution but lower mean values (POLDER mean value is 40Wm-2, HadGEM3 around 25). The biggest surprise is that model resolution makes little difference to the mean or variability of modelled DRE. (:-o * Data for Aug-Sept 2006 and the region (20˚S – 0˚, 10˚W – 10˚E) © Crown copyright Met Office Figure 3. Variability of aerosol DRE calculated by the POLDER retrieval and HadGEM3, averaged over the period Aug-Sept 2006 and for the region (20˚S – 0˚, 10˚W – 10˚E). Plots include (a) PDFs, (b) DRE from POLDER, (c-e) DRE from HadGEM for the low (N96), medium (N216) and high (N512) resolution simulations, respectively.
8
Above-cloud aerosol optical depth (ACAOD) from POLDER and HadGEM3
POLDER retrieval Model AOD is lower, especially further out over Atlantic Mean value is 40% lower, enough to explain the weaker DRE HadGEM3 N96 HadGEM3 N216 HadGEM3 N512 So naturally we then evaluated the aerosol and cloud properties to see how these compare to POLDER. Here is the evaluation of above-cloud AOD by integrating modelled extinction from the TOA down to cloud-top. From the PDFs we see that the model AOD is systematically lower than that retrieved by POLDER, especially with a higher frequency of very low AODs. As shown in the maps these very low values are generally on the Southern or Western edge of the sample region. The mean value is 40% lower in the model, enough to explain the weaker DRE. The modelled distribution of AC-AOD doesn’t vary that much with model resolution. © Crown copyright Met Office
9
Single-scattering albedo of above-cloud aerosol layer (ACSSA)
CLARIFY aged BB EXSCALABAR POLDER retrieval HadGEM3 N96 HadGEM3 N216 HadGEM3 N512 Modelled SSA ~0.05 lower than POLDER (30-40% more absorption per AOD) POLDER retrieves the imaginary part of the refractive index, hence enabling an estimate of SSA in the above-cloud aerosol layer. This leads to a very broad distribution with a mean of around 0.86 for the sample region. The POLDER estimates are in pretty good agreement with in-situ aircraft data from CLARIFY. This is the SSA from the EXSCALABAR instruments for runs sampling aged BB. The modelled SSA is around 0.05 lower than POLDER (indicating 30-40% more absorption per unit of AOD). There is a slight increase in SSA in the higher resolution runs. I don’t know what causes this. It could be a change in distribution of relative humidities at the altitude of the elevated aerosol layers. © Crown copyright Met Office
10
Cloud optical depth (COD) from POLDER and HadGEM
POLDER retrieval Plots show distributions of COD for cloudy scenes (COD >3) Model underestimates mean COD by ~25% Model fails to capture higher COD over stratocumulus decks around 10oS. Little change with model resolution! HadGEM3 N96 HadGEM3 N216 HadGEM3 N512 These plots show the distributions of cloud optical depth for cloudy scenes (where COD > 3). POLDER shows that COD has a fairly broad skewed distribution with a mean of around 11. Model underestimates the mean COD by ~25%, and underestimates the breadth of the distribution. Model also fails to capture the higher COD over the stratocumulus tongue around 10S. The underestimation of COD will strongly influence DRE and is probably the main reason why the modelled DRE is lower than POLDER. Surprisingly, again there is little change in COD with model resolution, except for finer scale structure. © Crown copyright Met Office
11
Conclusions Positive DRE in SE Atlantic but large spread. Requires good simulation of cloud properties and elevated plume of absorbing aerosol HadGEM3 captures variability of cloud, aerosol and DRE quite well, even at low resolution (140km) No apparent advantage from changing resolution from 140km to 60km or 25km! Model mean DRE is lower than POLDER, related to lower COD and AC-AOD. SSA was actually too low. Aerosol biases can be fixed (various trials completed) but can not reconcile the DRE with POLDER. Clouds would need brightening to do so. Just read the slide
12
Planned future work Widen evaluation, more observations: vertical distributions, AC-AOD… Improve aerosol properties in UKCA / RADAER Tweaks to thicken / brighten clouds? Daily variability of emissions, important? (vs monthly means) © Crown copyright Met Office
13
Thank you for your attention Questions?
© Crown copyright Met Office
14
HadGEM3 puts most of the BB plume above the cloud
Snapshot from 1st Aug 2006 For BB plume 60 – 80% of the AOD is above cloud © Crown copyright Met Office
15
Processes and parameters that could be improved / tuned
Parameter / process Current value Best value / range Impact BB emission scaling 2.0 1 – 3.5 Scale up/down to match AOD observations OC:POM 1.4 1.8 – 2.6 Increase OC mass & scattering 30-80% Hygroscopic growth parameter: FHYG_AOM 0.65 Reduce scattering 20-40% Density OC 1.5 1.2 Increase OC volume & scattering 25% Density BC 1.9 Reduce absorption 20% OC refractive index imaginary part 0.005 – 0.015i Reduce SSA in OC-rich regions BC mixing rule Volume-weighted Maxwell-Garnett? Reduce absorption 20-25%
16
Trial 1: Increase aerosol scattering
Default model (GA7.1) OC:POM increase from 1.4 to 2.0 AOD up by 30% SSA up from 0.80 to 0.85 (now match POLDER) But same AAOD! DRE more –ve in clear sky but not much different over cloud OC:POM increased to 2.0
17
SOCRATES estimates PDF of DRE from realistic AOD, COD distributions
SOCRATES now fed with AOD, COD and solar zenith angle (SZA) distributions from POLDER retrievals* COD variability alone creates most of the variability in DRE AOD variability also contributes to variability of DRE ---SZA variability less important in this case --- * Data covers the period Aug-Sept 2006 and region (20˚S – 0˚, 10˚W – 10˚E) © Crown copyright Met Office
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.