Operational Dust Prediction - Mineral Dust: A Key Player in the Earth System - page 227

the same figure. The variability in the dust surface concentrations at the various
forecast ranges can be very large, especially compared to variations closer to the
source at Djougou. Variations in dust can be of the same order of magnitude as the
maximum concentrations of aerosols regulated by air quality policies.
Compounding predictability issues are challenges in dust observability. Both
satellite- and ground-based observations are needed for nowcasting, data assimi-
lation and evaluation tools. From satellite, a host of dust enhancement products
is available to identify major dust features (see Chap. 7 ) . However, many of
these are qualitative in nature and as such cannot be readily used for assimilation
in models. More quantitatively, aerosol optical depth (AOD) retrievals can be
assimilated with corrections (e.g. Zhang and Reid 2006 ) but are commonly available
only over water or dark surfaces. Over bright desert surfaces, where the aerosol
signal cannot be so easily distinguished from the surface reflectance, “dark target”
retrieval techniques (e.g. Kaufman et al. 1997 ) fail, and retrievals must exploit either
different wavelengths like the Deep Blue algorithm (Hsu et al. 2004 ), multi-angle
viewing such as with the MISR instrument (Martonchik et al. 2004 ) or polarimetric
observations (e.g. Deuzé et al. 2001 ). But even in these circumstances, large errors
exist, which can prohibit assimilation, and for the largest events, AODs are so high
that the retrievals fail. This leaves models without reliable data for assimilation near
source regions. Furthermore, while AOD is a common model benchmark, models
carry mass, and there is virtually no reliable or representative data sets for mass
evaluation in major dust source regions, where data is most needed. The little mass
data that is available tends to come from short, episodic field missions. Lidar data
of aerosol extinction and backscatter show promising potential towards constraining
vertical profiles of aerosol fields and the height of the aerosol layers (Winker et al.
2007 ). Moreover, lidar depolarisation observations can be used to discriminate dust
from other aerosol species.
The chapter is structured as follows. In Sect. 10.2 , several operational and quasi-
operational dust prediction models are described. Section 10.3 describes examples
of regional and global multi-model ensembles for dust prediction that have been
established in recent years to offer better information and products to the users. Key
aspects of data assimilation for dust prediction are discussed in Sect. 10.4 , while
related technical details are given in Appendix A . Section 10.5 offers an overview
of the type of verification and evaluation procedures these systems are subject to.
Finally, Sect. 10.6 presents a summary and a future outlook on dust prediction
activities.

10.2 Dust Prediction Models

This section summarises the characteristics of some of the current aerosol prediction
models that are run in an operational or quasi-operational manner at various centres
around the world. This compilation is not intended to be exhaustive. In an effort
to be as inclusive as possible, both global and regional systems are described.