Correcting for bias in satellite-retrieved cloud droplet effective radius and its impact on estimates of aerosol-cloud interactions

Fu, Dongwei

Permalink

https://hdl.handle.net/2142/117530 Copy

Description

Title

Correcting for bias in satellite-retrieved cloud droplet effective radius and its impact on estimates of aerosol-cloud interactions

Author(s)

Fu, Dongwei

Issue Date

2022-09-14

Director of Research (if dissertation) or Advisor (if thesis)

Di Girolamo, Larry

Doctoral Committee Chair(s)

Di Girolamo, Larry

Committee Member(s)

• Rauber, Robert M
• Nesbitt, Stephen W
• McFarquhar, Greg M

Department of Study

Atmospheric Sciences

Discipline

Atmospheric Sciences

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Satellite Remote Sensing
• Cloud Microphysics
• Cloud droplet effective radius
• Cloud droplet number concentration
• Field validation
• Aerosol-cloud interactions
• 3-D Radiative Transfer

Abstract

Our longest record (spanning nearly four decades) of observations for cloud optical and microphysical properties from space is derived from the bi-spectral approach. It retrieves cloud optical depth and cloud effective radius (Re) from a pair of measured cloud reflectances, typically one in the visible/near-infrared and the other in the shortwave infrared spectral range. This approach makes several critical assumptions, including 1-D radiative transfer, single-mode droplet size distribution, and cloud horizontal and vertical homogeneity. Deviations from these assumptions lead to systematic errors in the cloud retrievals that co-vary with the underlying scene heterogeneity and sun-view geometry. Given the wide use of bi-spectral derived cloud properties in climate research, it is crucial to quantify the errors and bias in satellite cloud retrievals. Recently, an effort to characterize the bias in bi-spectral Re retrieved from the Moderate Resolution Imaging Spectroradiometer (MODIS) through data fusion between MODIS and the Multi-angle Imaging SpectroRadiometer (MISR) revealed monthly zonal-mean values of Re biases from 2 to 11 μm, depending on latitude (Liang et al., 2015). Here, in a push towards bias-correction of MODIS-retrieved Re, we further examine the MISR-MODIS fusion estimate of Re bias as it relates to other observed cloud properties, such as cloud horizontal heterogeneity, cloud optical depth, and sun-view geometry. By stratifying the Re bias by observed MODIS cloud properties, latitude, and month, we introduce a bias-correction approach for MODIS-retrieved Re at regional scales. Our estimates reveal global distribution of MODIS-retrieved Re monthly mean bias ~1 to 10 μm (15 to 60%) depending on scene heterogeneity, cloud optical depth, and solar zenith angle. Bias-correction was separately applied to the 1.6, 2.1 and 3.7 μm MODIS spectral channels, the bias-adjusted monthly-mean Re from the 2.1 and 3.7 μm channels show difference of ~ +0.6 μm in the coastal marine stratocumulus regions and difference of ~ -2 μm in the cumuliform cloud regions, compared to the uncorrected values of ~ -1 to -6 μm from the original MODIS Re products. Bias-adjusted Re values compare favorably to other independent data sources, including field observations, global model simulations, and interpretations of satellite retrievals that do not use retrieval techniques similar to MODIS. The robustness of our bias-correction is evaluated in one of the first field validations of bi-spectral Re in heterogeneous cloud regions. Using data collected during the Cloud, Aerosol and Monsoon Processes Philippines Experiment (CAMP2Ex) from August-October 2019, cross-comparisons between Terra-MODIS bi-spectral Re, airborne remotely sensed Re from the Research Scanning Polarimeter (RSP) onboard the NASA P-3, and in-situ measurements from both the P-3 and the SPEC Learjet are applied to matching cumulus cloud fields. Re from the polarimetric technique is used for validating the bi-spectral approach because it is less sensitive to the assumptions of plane parallel and homogeneous clouds. RSP’s ability to derive collocated bi-spectral and polarimetric Re allows for comparing Re retrievals from the two techniques with the same sampling. When compared to RSP polarimetric Re, on average, RSP bi-spectral Re overestimates by 6 μm, and the MODIS Re overestimates by 7 μm. By contrast, in-situ derived Re and our bias-adjusted MODIS Re are both within 2 μm. Sensitivity analysis using RSP data reveals that small cumuli frequently exist in the sampled region (50% of the clouds sampled by RSP have transect lengths less than 0.6 km). The overestimates of Re from RSP bi-spectral technique compared to polarimetric technique increased as cloud size and cloud optical depth decreased. Drizzle, cloud top bumpiness and sampled solar-zenith angle, however, are not closely correlated with the overestimate of bi-spectral Re. We show that for shallow clouds that dominate the liquid cloud cover for the CAMP2Ex region and period, 3-D radiative transfer and cloud heterogeneity, particularly for the optically thin and small clouds, appear to be the leading cause for the large positive biases in bi-spectral retrievals. Finally, we investigate the impact of the observed overestimate in bi-spectral Re retrievals on aerosol-cloud interaction estimates in the CAMP2Ex domain. Using Re retrievals from various techniques, cloud droplet number concentration (Nd) and the aerosol first indirect effect (IE) are estimated from both satellite and airborne perspectives. Our results show that removing the ~7 μm overestimate in MODIS Re results in a ~300% increase in derived Nd values, on average, when compared to the original MODIS derived Nd. The bias-adjusted MODIS Nd (i.e., using our bias-corrected MODIS Re values) distribution has a median value of ~109 cm-3, which is similar to the in situ measured Nd median value of ~114 cm-3. The RSP polarimetric derived Nd distribution also has a similar median Nd value of ~117 cm-3. The original MODIS Nd and RSP bi-spectral derived Nd distributions, however, shows much smaller median Nd values of ~34 cm-3 and ~43 cm-3, respectively. IE is estimated as ∂ln(Nd)/ ∂ln(AOD), where AOD is aerosol optical depth. Our estimates of IE are ~0.16 for both RSP-polarimetric Nd (r = 0.44 with HSRL-2 AOD) and bias-adjusted MODIS Nd (r = 0.2 with MISR AOD), ~0.19 (r = 0.2) for the original MODIS Nd, and ~0.02 (r = 0.2 with MISR AOD) for RSP bi-spectral Nd (r = 0.03 with HSRL-2 AOD). The very poor correlation coefficient for IE derived from the RSP bi-spectral Nd values appears to arise from the very high variability in RSP bi-spectral Re, which are not observed in either in situ or RSP-polarimetric Re values. The very high variability in RSP bi-spectral Re exists because of the large uncertainty in the retrievals caused by cloud heterogeneity. Sensitivity of these results to the choice of scale and sampling strategy are acknowledged as an additional source of uncertainty, but not explored. Because these uncertainties vary with the underlying structure of the cloud field, caution continues to be warranted in studies of aerosol-cloud interaction that use bi-spectral Nd retrievals in cumulus cloud fields. However, the good agreement between MODIS bias-adjusted Nd, RSP polarimetric derived Nd, and in situ measured Nd increases our confidence in their use in studies of cumulus cloud fields. This is the first work to provide bias estimates from MODIS Re products in a global perspective, and it further confirms the large bias of bi-spectral Re in cumuliform cloud fields where deviations from the 1-D assumption is the largest. The bias-adjusted MODIS Re shows good agreement with field observations, model simulations, and other satellite retrievals that do not use the bi-spectral technique. By correcting for the MODIS Re bias using our newly developed bias-correction procedure, the resulting Nd also compares favorably to in situ measurements in cumulus regions. However, the bias in MODIS Re does not greatly impact estimates of the aerosol first indirect effect, as the bias-correction was obtained from stratification of solar-zenith angle, latitude, cloud optical depth and cloud heterogeneity, but not aerosol related properties. Therefore, when the estimated Re is corrected for bias, the derivative of the estimated Re bias is zero. Still, this work shows potential in using passive derived Re and Nd to evaluate model parameterization, particularly in cumuliform cloud regions.

Graduation Semester

2022-12

Type of Resource

Thesis

Copyright and License Information

Copyright 2022 Dongwei Fu

Owning Collections