Development towards a global operational aerosol consensus: basic climatological characteristics of the International Cooperative for Aerosol Prediction Multi-Model Ensemble (ICAP-MME)

Abstract

Abstract. Over the past several years, there has been a rapid development in the number and quality of global aerosol models intended for operational forecasting use. Indeed, most centers with global numerical weather prediction (NWP) capabilities have some program for aerosol prediction. These aerosol models typically have differences in their underlying meteorology as well as aerosol sources, sinks, microphysics and transformations. However, like similar diversity in aerosol climate models, the aerosol forecast models have fairly similar overall bulk error statistics for aerosol optical thickness (AOT)-one of the few aerosol metrics that is globally available. Experience in climate and weather prediction has shown that in situations such as this where there are several independent models, a multi-model ensemble or consensus will be top performing in many key error metrics. Further, multi-model ensembles provide a highly valuable tool for forecasters attempting to predict severe aerosol events. Here we present the first steps in developing a global multi-model aerosol forecasting ensemble intended for eventual operational and basic research use. Drawing from members of the International Cooperative for Aerosol Prediction (ICAP) latest generation of quasi-operational aerosol models, five day AOT forecasts are analyzed for December 2011 through November 2012 from four institutions: ECMWF, JMA, NASA GSFC, and NRL/FNMOC. For dust, we also include the NOAA NGAC product in our analysis. The Barcelona Supercomputing Centre (NMMC) and UK Met office dust product have also recent become available with ICAP, but have insufficient data to be included in this analysis period. A simple consensus ensemble of member and mean AOT fields for modal species (e.g., fine and coarse mode, and a separate dust ensemble) is used to create the ICAP Multi-Model Ensemble (ICAP-MME). The ICAP-MME is run daily at 0Z for 6 hourly forecasts out to 120 h. Basing metrics on comparisons to 21 regionally representative Aerosol Robotic Network (AERONET) sites, all models generally captured the basic aerosol features of the globe. However, there is an overall AOT low bias among models, particularly for high AOT events. Biomass burning regions have the most diversity in seasonal average AOT. The southern oceans, though low in AOT, nevertheless also have high diversity. In regard to root mean square error, as expected the ICAP-MME placed first over all models worldwide, and was typically first or second in ranking against all models at individual sites. These results are encouraging; as more global operational aerosol models come on line, we expect their inclusion in a robust operational multi-model ensemble will provide valuable aerosol forecasting guidance.

Ground‐based High Spectral Resolution Lidar observation of aerosol vertical distribution in the summertime Southeast United States

“…Those cases with significant coarse mode contribution may have some cirrus contamination. The International Cooperative for Aerosol Prediction multimodel aerosol ensemble [ Sessions et al ., ] suggested periodic injections of Saharan dust into the SEUS, but not with appreciable influence at the site.…”

“…Contextual information on regional biomass burning smoke activity is provided in Figure , with aerosol backscatter and total depolarization information at Huntsville showing smoke to >13 km altitude being presented in Figure . The roots of this event occurred the night of 21 June, with global multispecies aerosol models [e.g., Sessions et al ., ] and the Navy Aerosol Analysis and Prediction System reanalysis [ Lynch et al ., ] forecasted smoke from multiple fires in Colorado being transported up into an MCS forming that night over Nebraska. At that point, the forecast global models failed to forecast smoke properly as they are not equipped to cope with a deep convective transport.…”

Assimilation of AERONET and MODIS AOT observations using variational and ensemble data assimilation methods and its impact on aerosol forecasting skill

“…The temporal variability in the ensemble was better represented in regions like India, which are impacted by large AOT events (>1), as is the case for Beijing in East Asia and Kanpur in India (Figures c and d). Beijing and Kanpur are examples of locations where aerosol prediction is a challenge, especially for large events [ Sessions et al , ]. In this regard, assimilating AERONET in the ensemble NAAPS system is promising for extreme aerosol events, especially in places like Beijing where the AOT frequently exceeds 1.…”

“…One consistency across all of these systems is the assimilation of derivatives of aerosol optical thickness (AOT) retrievals from the Moderate Resolution Imaging Spectroradiometer (MODIS) with its daily global spatial coverage. While the assimilation of MODIS AOT has helped the various forecasting systems to successfully capture the basic aerosol features of the globe [ Sessions et al , ], it is important to consider how other aerosol‐related data streams can be used in aerosol forecasting. A multitude of satellite‐based products have been assimilated, including passive Multi‐angle Imaging SpectroRadiometer [ Lynch et al , ] and active Cloud‐Aerosol Lidar with Orthogonal Polarization [ Sekiyama et al , , Zhang et al , ].…”

“…This was shown in analysis time series in India and Asia as well as in the aforementioned case studies. There have been demonstrated challenges in the global aerosol forecasting community, particularly when it comes to predicting high AOT aerosol events [ Sessions et al , ]. This is a problem as a central motivation for operational aerosol forecasting is predicting significant aerosol events [ Sessions et al , ].…”