
31b551556f6d05a6050d475134f5099d.ppt
- Количество слайдов: 17
GEWEX Aerosol Panel: A critical review of the efficacy of commonly used aerosol optical depth retrieval GEWEX Panel: Sundar Christopher, Richard Ferrare, Paul Ginoux, Stefan Kinne, Gregory G. Leptoukh, Jeffrey Reid, Paul Stackhouse Programmatic support: Hal Maring, Charles Ichoku, Bill Rossow Comments on this presentation: jeffrey. reid@nrlmry. navy. mil
bottom line up front • Since our Last Briefing (Aug 2010), the GEWEX Aerosol Assessment Panel (GAAP) has nearly completed the first report. • By late October we expect to send sections to the instrument science teams for “fact checking. ” • Report will layout the nature of the aerosol problem, with a synopsis of the literature and commentary on verification methods and findings. • Phase 2 -detailed independent evaluation, will not start until MODIS collection 6 and MISR v 23 is officially released.
the aerosol problem • The aerosol field has recently grown exponentially, with literally dozens of both products and applications. • But, most products are in the twilight zones of “research, ” “development” and “production. ” • This is reinforced with the funding situation where money for product development, maintenance, and verification is limited. Developers spend more time “using” than “supporting” their products. • By the time the wider community figures out how a product is doing, a new version is released. • Situation: Confusion and some rancor in the community as to the actual efficacy and appropriate application of these data sets • Response: Reform the GEWEX Aerosol Panel (GAP)
the GAP mandate • NASA HQ wished the development of a comprehensive evaluation of the current state-of-the science performed within the GEWEX framework. • Team is to be small and well rounded. Team size is to be no larger than what is necessary (e. g. , no large committees). • Members are to be ‘jurors drawn from accomplished peers’ to examine the 7 most common global products: AVHRR (GACP and NOAA), MISR, MODIS (Standard & Deep Blue), OMI, and POLDER.
the GAP mandate • Phase 1: Perform comprehensive literature review and evaluation. Deliverable will be a report on the state-of the science, the application of satellite aerosol data, and the identification of shortcomings, and broad recommendations to the field for future development and verification needs. • Phase 2: Based on Phase 1 examine in detail specific issues in the generation of retrieval and gridded products. • Peer review of the peer review: Findings given to teams for comment before release. After an iteration, team rebuttals can be made public record. • Where are we now? Nearing the end of phase 1. Report is 100+ pages and growing…
customers and issues • Often aerosol products are thought of as climate products. • However, all of the world’s major Numerical Weather Prediction (NWP) have aerosol assimilation programs and aerosol data has worked its way into numerous applications. • There aerosol observability issues – reliable and timely delivery – bias (contextual, sampling) understanding / removal – error characterization (essential for assimilations)
opinion data product evaluation, validation, and verification. Reliable and timely delivery of satellite aerosol and fire products is only half the challenge. If products are to be integrated, then biases need to be removed through careful product evaluation and verification. Contextual & sampling biases need to be understood. Because of possible degradation in model performance through data assimilation, aerosol product error characterization has been emphasized more in the operations than climate communities. Indeed, despite popular misconceptions, operational data characterization requirements are often more strict than what is commonly used in the climate r e s e a r c h c o m m u n i t y. Reid, J. S. , Benedetti, A. , Colarco, P. R. , Hansen, J. A. , 2011. International operational aerosol observability workshop, Bulletin of the American Meteorological Society, 92, Issue 6, pp. ES 21 -ES 24 doi: 10. 1175/2010 BAMS 3183. 1
panel members HQ: Hal Maring and Charles Ichoku • Sundar Christopher (UAH): chair, algorithm development, multi sensor products • Richard Ferrare (NASA La. RC): lidar, field work, multisensor products • Paul Ginoux (NOAA GFDL): Global modeling and aerosol sources • Stefan Kinne (Max Plank): GEWEX Cloud, AEROCOM • Gregory Leptoukh (NASA GSFC): Level 3 product development and distribution • Jeffrey Reid (NRL): co-chair, aerosol observability, field work, verification, operational development • Paul Stackhouse: GEWEX radiation, atmospheric radiation and energetics.
report outline 1. Introduction 2. Nature of the Problem 3. Overview of Assessed Satellite Products 4. Evaluation of Verification and Intercomparison Studies 5. Phase 1 Synopsis and Recommendations
relative levels of efficacy required (Approximate and not meant to offend…) Studies Imagery/ Contextual Seasonal Climatology Model Aps, V&V, Inventory Data Assimilation “Advantage of Human Eye” Basically want to know were stuff is. Can do one-up corrections Have stronger time constraints and need spatial bias elimination. Quantify bias & uncertainty everywhere and correct where you can. Parametric Modeling and Lower Order Process Studies Correlations de-emphasize bias Trend Climatology Need to de-trend biases in retrieval and in sampling Higher Order Process Study Push multi-product and satellite data V&V statistics must speak to these applications! Hence, there is no “one size fits all” error parameter. Sorry….
bias examples (1) global average, time-series over water differences are a mix of radiometric bias cloud bias microphysical bias sampling differences contextual bias. satellite retrievals tend to overestimate AOD (at low AOD) over oceans especially MISR global AOD difference between sensors Mischenko et al. , 2007 Zhang and Reid, 2010
bias examples (2) more • consideration of “what the satellite actually sees” is often overlooked ASO clear sky bias, Zhang and Reid 2009 • basic matchup between sensors is not trivial. • core retrieval biases related to clouds, lower boundary condition and microphysics are nonrandom, and spatially / temporally correlated MODIS vs. AERONET - slope -
If we knew AOD then we would not need MODIS. All we have is MODIS’s own estimate of AOD…. AERONET AOD RMSE (MODIS, AERONET) diagnostic versus prognostic error models MODIS over ocean example worse here better here From Shi et al. , 2010, ACP MODIS AOD
verification (1) • Since you can measure AOD, all aerosol science projects drive to validate to AOD, whether it is appropriate or not. • There is no shortage of validation studies. But, they tend to be direct regression based, have important details missing, and are conducted over limited periods of times and/or locations. Hence, they tend to be of limited utility. • While there are many cases of satellite cal-val components from field missions, analyses are usually not repeated for new product versions. • Even well designed third party studies are generally not utilized or cited by the production teams.
verification (2) • Over ocean, there tends to be remarkable consistency both in AOD and in correlated bias across sensors. Cloud masking is still a problem. • Over land, there is strong regional and temporally correlated biases across both algorithms and sensors, largely due to the lower boundary condition. • Radiance calibration is a significant problem and pops up in indirect ways. NASA is working on it. • Demonstration of diversity in aerosol products has little barring on relative product efficacy. • bottom line: The difference between “face value statistics” and an error bar for an individual retrieval is vast. How does this effect you? Depends on what you do with the data.
key recommendations (1) • Algorithms need better documentation. The ATBDs are a good start, but they need to be kept current and perhaps even expanded. • Better strategies for “level 3 products” need to be devised and supported. One size fits none…. . • One size fits all verification does not work either. But, there is a total lack of agreement on key verification metrics. The USER community needs to agree on what they think is important. • It should be a programmatic requirement of the science teams to develop prognostic error models as part of any mass produced and distributed product. Program offices need to fund this.
key recommendations (2) • AERONET and MPL-net are clearly backbone networks for verification and we strongly endorse their financial support as a critical community resource. Similarly, targeted aircraft observations should be encouraged. • Developers and outside entities to work more together in verification studies. • Field work needs to be better utilized. After a first round of verification studies, next generation algorithms do not typically make use of older studies. Field work should focus more on verifying higher level products.