TL;DR
Quantifying how cloud droplet numbers change with aerosol pollution (a key climate forcing) is tricky from satellites. This study uses multiple satellite datasets and reanalysis to show that updraft, precipitation, retrieval errors, and vertical alignment all significantly impact these estimates, and suggests ways to improve them.
What Problem Does This Paper Address?
Aerosol-cloud interactions are a major source of uncertainty in climate models, particularly the "Twomey effect," which describes how more aerosols lead to more cloud droplets, making clouds brighter. The sensitivity of cloud droplet number concentration ($N_d$) to aerosol concentration ($S$) is a crucial parameter for estimating this effect. However, satellite observations, which are essential for global estimates, face significant challenges. It's difficult to isolate the true aerosol impact on $N_d$ from other factors like cloud dynamics (updrafts, precipitation), retrieval errors in both aerosol and cloud properties, and how aerosols and clouds are vertically aligned. This paper aims to disentangle these confounding factors to provide a more accurate assessment of $S$ from satellite data.
Methods
The authors employ a multi-dataset approach, combining active and passive satellite observations with reanalysis data. They investigate the impact of four key confounders:
- Updraft: Proxied by cloud-base height (CBH) and cloud geometrical thickness (CGT) from MISR data.
- Precipitation: Identified using CloudSat radar data.
- Retrieval Biases: Quantified using metrics related to the distance of clear pixels to cloudy pixels (AL) for aerosol retrieval and the difference between all cloudy pixels and quality-assured cloudy pixels for cloud retrieval (Δ$N_d$).
- Vertical Co-location: Assessed by comparing column-integrated aerosol proxies (SO4C) with near-cloud-base proxies (SO4B) derived from MERRA-2 reanalysis.
They use a "pre-binned" regression approach to calculate $S$, where data is grouped into bins based on the confounding factor, and then a linear regression is performed on the binned medians. This is compared to a standard "all-data" regression.
Datasets
The study utilizes a comprehensive suite of datasets:
- Terra and Aqua MODIS:
- MYD08/MOD08 (Level 3): Aerosol Optical Depth (AOD) at various wavelengths.
- MYD06/MOD06 (Level 2): Cloud optical properties (Cloud Effective Radius - CER, Cloud Optical Thickness - COT), Cloud Mask, Cloud-top temperature, Cloud-top height, Cloud phase, and Cloud multi-layer flag.
- CloudSat:
- 2B-CLDCLASS product: Precipitation flag.
- 5x5 km² resolution data: Used for cloud properties.
- MISR (Multi-angle Imaging SpectroRadiometer):
- MIL2TCSP (Level 2): Cloud-base height (CBH) and Cloud-top height (CTH).
- Reanalysis Data:
- MERRA-2 (Modern-Era Retrospective analysis for Research and Applications version 2): Sulfate mass mixing ratio profiles, air density, and temperature.
- ERA5 (ECMWF Reanalysis v5): Temperatures at 700 and 1000 hPa (used for calculating Lower Tropospheric Stability).
Key Insights and Ideas
- Updraft Matters: Stronger updrafts (indicated by higher CBH and CGT) are associated with a significantly higher sensitivity of cloud droplet number to aerosols. This makes sense, as stronger updrafts can lead to higher supersaturation and activate more CCN.
- Precipitation is a Major Confounder: Precipitating clouds artificially inflate the estimated aerosol-cloud sensitivity ($S$) by about 21%. This is because precipitation acts as a sink for cloud droplets, masking the true aerosol effect, and aerosols can also suppress precipitation, creating a complex feedback loop.
- Retrieval Biases Undermine Estimates: Both aerosol and cloud retrieval errors tend to underestimate $S$. This is particularly challenging because cloud fraction plays a dual role: it can exacerbate aerosol retrieval biases (e.g., near clouds) but improve cloud retrieval accuracy (by reducing heterogeneity). Balancing these is difficult.
- Vertical Alignment is Crucial: Using column-integrated aerosol proxies (like SO4C) instead of near-cloud-base proxies (like SO4B) can nearly double the estimated sensitivity ($S$). This is because column-integrated values don't capture the crucial variability of aerosols near the cloud base, which is what directly influences CCN activation.
Key Results
- $S$ increases remarkably with cloud-base height and cloud geometric thickness, indicating stronger aerosol-cloud interactions at higher updrafts.
- Including precipitating clouds artificially amplifies $S$ by an estimated 21% for AI (AOD). The SAI for all clouds is 0.68, compared to 0.56 for non-raining clouds.
- Retrieval biases in aerosol and cloud properties tend to underestimate $S$. For example, aerosol retrieval biases (first AL bin) lead to an underestimation of SAI by ~29% and SAOD by ~50%, though the overall impact on $S$ is smaller due to limited data in that bin. Cloud retrieval biases (comparing $N_d$ from all pixels vs. quality-assured pixels) lead to an underestimation of $S$ by ~8% for SAOD and ~17% for SAI.
- Using column-integrated sulfate (SO4C) as a CCN proxy results in an approximately twofold enhancement of $S$ compared to using sulfate near cloud base (SO4B). The correlation between $N_d$ and SO4C is also degraded compared to SO4B.
Limitations and Future Work
- Nd Retrieval Assumptions: The study acknowledges that satellite-based $N_d$ retrievals rely on several assumptions, which can introduce systematic biases. Future work could explore more advanced sampling strategies or methods to better constrain cloud adiabaticity.
- CCN Proxies: While AOD/AI and sulfate are used as CCN proxies, they don't fully capture aerosol properties like size or hygroscopicity, and organic aerosols are also important. Direct CCN retrievals from polarimetric satellites are suggested for future work.
- Linear Regression: The use of linear regression to quantify $S$ might not fully capture non-linear relationships, especially in polluted conditions. Joint histogram methods are suggested as an alternative.
- Entrainment Mixing: The impact of entrainment mixing, another crucial droplet sink process, was not explicitly considered due to the difficulty in quantifying it from satellite observations. However, it's noted that updraft constraints might implicitly incorporate some of its effects.
- Spatial Scale: The study acknowledges that $S$ can vary with spatial resolution and scale of analysis.
- Future Directions: The authors suggest using combined active and passive satellite observations (e.g., CALIOP/CloudSat) that provide simultaneous retrievals of aerosol profiles, precipitation, and cloud-base height to perform a more comprehensive analysis accounting for all potential biases. They also highlight the importance of directly retrieved CCN for accurate radiative forcing estimates.
Summary updated