Steve Platnick 1, Gala Wind 2,1, Zhibo Zhang 3, Hyoun-Myoung Cho 3, G. T. Arnold 2,1, Michael D. King 4, Steve Ackerman 5, Brent Maddux NASA Goddard Space Flight Center, 23 2 SSAI, 3 U. of Maryland Baltimore County, 45 4 U. Colorado/LASP, 5 U. Wisconsin, Madison AGU Fall Meeting A44B 6 Dec 2012 San Francisco, CA Sensitivity of Marine Warm Cloud Retrieval Statistics to Algorithm Choices: Examples from MODIS Collection 6 Development Code
What is a Cloud: The Pixel-Level Choices Algorithm Developer’s Make -Explicit (partly cloudy pixel filtering by the developer) -Implicit (filtering invoked by retrieval failures) Sensitivity of Cloud Optical Property Retrievals to Choices Sampling fraction, τ, r e Outline
Cloud Clear What Do We Mean by a Cloud Mask? Ideal pixel Clear
What Do We Mean by a Cloud Mask? Cloud Clear Overcast Clear Sky Partly Cloudy
Cloud Clear Satellite Cloud Mask (likelihood of “Not Clear”) What Do We Mean by a Cloud Mask?
MODIS Cloud Pixel Filtering Choices: Explicit & Implicit Masked as Clear & Not Clear Total Number of Pixels (1 km) = Developer Choices Retrieve edge/250m partly cloudy pixels? Provide a τ- only retrieval when multispectral retrievals fail? Not Clear Categories: Overcast (?) Cloud Edge 250m “hole” Possibly heavy smoke/dust, glint? Explicit filtering Retrieval Outcomes: Successful τ & r e No τ or r e possible τ only (ignore r e spectral information)? Implicit filtering
Cloud Pixel Filtering/QA Choices: C5 Granule Example 1 April 2005, MODIS Aqua MODIS 250/500 m composite
Cloud Pixel Filtering/QA Choices: C5 Granule Example 1 April 2005, MODIS Aqua Clear Sky Restoral Flags cloud edges 250m partly cloudy pixels spatial/spectral tests (glint, dust, smoke)
MODIS 250m Heterogeneity global analysis, low maritime water clouds Pixel Counts km cloud edges 250m partly cloudy 1km cloud edge & 250m partly cloud removed 3D artifacts more likely
Pixel Filtering: Retrieval Outcome Terra MODIS April 2005, maritime water clouds CTP ≥ 680mb, ±30° latitude Successful COT & r e COT r e (2.1 µm)
Successful COT & r e COT r e 2.1 – r e 3.7 Pixel Filtering: Retrieval Outcome Terra MODIS April 2005, maritime water clouds CTP ≥ 680mb, ±30° latitude Retrievals consistent w/breakdown of 1D forward model
44% of cloudy pixels are associated w/edges or designated as partly cloudy by the 250m cloud mask 40% of edge/partly cloudy pixel retrievals fail (simultaneous COT and r e solution fall outside LUT space) Successful COT & r e Failure (minor) Failure (major) Pixel Filtering: Sampling Statistics Terra MODIS April 2005, maritime water clouds CTP ≥ 680mb, ±30° latitude
Pixel Filtering: Retrieval Outcome SEVIRI, 15 min imagery, 11 August 2009, maritime water clouds CTP ≥ 680mb, ±30° latitude, ±55° VZA Successful COT & r e COT r e (1.6 µm) Successful COT & r e Failure Fraction of Population (%) 20% of cloudy pixels are associated w/edges, 68% of those retrievals fail
Pixel Filtering/QA Choices: Global Mean Sensitivity Cloud Retrieval Difference: with edge/250m filtering – w/out τ r e,2.1 ∆ τ =±4 ∆ r e,2.1 =±2 µm April 2005, MODIS Terra
Summary (1) Tropical/subtropical marine warm cloud partly-cloudy retrievals (edge pixels and those identified by 250m observations) are biased w.r.t. the filtered pixel population. Biases are consistent w/breakdown of 1D cloud model. - Retrievals will not correctly describe interaction of the cloud with the radiation field, microphysics, or derived water path. Frequency of these pixels depends on the spatial scales of the satellite observations and the clouds. MODIS Cloud Product Collection 5: These pixels were removed/filtered (“Clear Sky Restoral” algorithm). Collection 6: Will attempt retrievals on these pixels. Allow users to explore the consequences of the partly cloudy categories. Regardless, a significant fraction of such retrievals “fail” for the latitude zone studied.
All algorithms do consider the suitability of a pixel/FOV for use with the forward model – either explicitly or implicitly. Spatial heterogeneity and related sampling issues ARE NOT unique to the MODIS product. Other satellite sensors have similar issues and consequently inherent sampling biases for low marine clouds, e.g., CloudSat [Zhang et al., A33G], microwave imagers, etc. - How to communicate to this to the variety of users is a challenge. Summary (2)