Response to reviewers

Edward GryspeerdtOpenalex Unknown 2022 · DOI · OpenAlex

Abstract Cloud droplet number concentration ($N_d$) is of central importance to observation-based estimates of aerosol indirect effects, being used to quantify both the cloud sensitivity to aerosol and the base state of the cloud. However, the derivation of $N_d$ from satellite data depends on a number of assumptions about the cloud and the accuracy of the retrievals of the cloud properties from which it is derived, making it prone to systematic biases. A number of sampling strategies have been proposed to address these biases by selecting the most accurate $N_d$ retrievals in the satellite data. This work compares the impact of these strategies on the accuracy of the satellite retrieved $N_d$, using a selection of in-situ measurements. In stratocumulus regions, the MODIS $N_d$ retrieval is able to achieve a high precision ($r^2$ of 0.5–0.8). This is lower in other cloud regimes, but can be increased by appropriate sampling choices. Although the $N_d$ sampling can have significant effects on the $N_d$ climatology, it produces only a 20% variation in the implied radiative forcing from aerosol-cloud interactions, with the choice of aerosol proxy driving the overall uncertainty. The results are summarised into recommendations for using MODIS $N_d$ products and appropriate sampling.

Introduction

Predicting future climate requires a precise understanding of aerosol-cloud interactions (ACI). When aerosols—such as smoke or industrial pollution—interact with clouds, they typically increase the number of droplets while decreasing their size. This phenomenon increases cloud albedo and longevity, exerting a cooling effect that partially offsets greenhouse gas warming. Cloud droplet number concentration ($N_d$) serves as a critical metric for quantifying this effect.

However, satellites do not measure $N_d$ directly. Instead, instruments like MODIS retrieve cloud optical depth ($\tau_c$) and droplet effective radius ($r_e$), which are then used to calculate $N_d$ under the assumption of an "adiabatic" cloud (one that is well-mixed and lacks significant precipitation or entrainment). Because these assumptions frequently fail in real-world conditions, various "sampling strategies" or data filters have been proposed to isolate high-quality pixels. This study provides a rigorous evaluation of these strategies by comparing satellite estimates against an extensive library of aircraft measurements.

Methodology and Data

The study evaluates five distinct sampling strategies for the standard MODIS Collection 6.1 $N_d$ product:

  1. All: A baseline with minimal filtering.
  2. Q06: Excludes thin clouds or very small droplets.
  3. G18: Incorporates filters for high solar/viewing angles and sub-pixel inhomogeneity (SPI).
  4. BR17 ($r_e$-stacking): A stringent method comparing $r_e$ at different wavelengths to identify adiabatic pixels.
  5. Z18 (Cloud Cores): Selects only the thickest 10% of cloud pixels within a region.

Comparisons were conducted at both the pixel level (1 km) and the gridbox level ($1^\circ \times 1^\circ$), the latter being the standard scale for global climate modeling. The "ground truth" was provided by 11 aircraft campaigns conducted between 2010 and 2020, covering diverse regimes from stable stratocumulus decks (e.g., VOCALS-REx) to complex convective environments (e.g., COPE) and the Southern Ocean (SOCRATES).

Key Findings

Accuracy Across Cloud Regimes

The performance of MODIS $N_d$ retrievals varies significantly by cloud type. In stable stratocumulus regions, the retrievals are remarkably robust, achieving $r^2$ values between 0.5 and 0.8 even with minimal filtering. In contrast, broken or convective clouds present a significant challenge due to the breakdown of the adiabatic assumption. In these complex regimes, the BR17 ($r_e$-stacking) method proved most accurate, achieving an overall pixel-level $r^2$ of 0.71, though it requires discarding approximately 63% of the available data.

Impact on Climatology vs. Radiative Forcing

While sampling strategies dramatically alter the global climatology of $N_d$—with stringent filters like BR17 and Z18 yielding much higher mean values, particularly over land—their impact on calculated aerosol-cloud radiative forcing is surprisingly modest. The variation in implied radiative forcing due to $N_d$ sampling was approximately 20%.

The Dominance of Aerosol Uncertainty

A critical insight from this work is that the choice of aerosol proxy (the measurement used to represent pollution levels) introduces far more uncertainty into radiative forcing estimates than the $N_d$ sampling strategy itself. This suggests that while refining $N_d$ retrievals is valuable, improving aerosol data is the more urgent priority for reducing uncertainty in ACI estimates.

Conclusions and Future Directions

The study demonstrates that current satellite $N_d$ products are "good enough" for global forcing estimates, provided appropriate sampling is used. However, convective clouds remain a significant hurdle that current bispectral retrieval strategies cannot fully overcome.

Future advancements will likely involve moving beyond current limitations through the use of multi-angle polarimetry or lidar. Additionally, expanding in-situ validation to under-sampled regions, such as the remote ocean and polluted terrestrial environments, remains essential for refining global $N_d$ recommendations.

This summary is based on research published in Atmospheric Measurement Techniques.

Summary updated

Abstract

<strong class="journal-contentHeaderColor">Abstract.</strong> Cloud droplet number concentration (N<sub><em>d</em></sub>) is of central importance to observation-based estimates of aerosol indirect effects, being used to quantify both the cloud sensitivity to aerosol and the base state of the cloud. However, the derivation of N<sub><em>d</em></sub> from satellite data depends on a number of assumptions about the cloud and the accuracy of the retrievals of the cloud properties from which it is derived, making it prone to systematic biases. A number of sampling strategies have been proposed to address these biases by selecting the most accurate N<sub><em>d</em></sub> retrievals in the satellite data. This work compares the impact of these strategies on the accuracy of the satellite retrieved N<sub><em>d</em></sub>, using a selection of insitu measurements. In stratocumulus regions, the MODIS N<sub><em>d</em></sub> retrieval is able to achieve a high precision (r<sup>2</sup> of 0.5&ndash;0.8). This is lower in other cloud regimes, but can be increased by appropriate sampling choices. Although the N<sub><em>d</em></sub> sampling can have significant effects on the N<sub><em>d</em></sub> climatology, it produces only a 20 % variation in the implied radiative forcing from aerosol-cloud interactions, with the choice of aerosol proxy driving the overall uncertainty. The results are summarised into recommendations for using MODIS N<sub><em>d</em></sub> products and appropriate sampling.
BibTeX
@inproceedings{gryspeerdt2022response,
  author = {Edward Gryspeerdt},
  title = {{Response to reviewers}},
  booktitle = {Openalex Unknown},
  year = {2022},
  doi = {10.5194/amt-2021-371-ac1},
}
Figure