Statistical relationship between surface PM10 concentration and aerosol optical depth over the Sahel as a function of weather type, using neural network methodology. Issue 23 (12th December 2013)

• Record Type: Journal Article
• Authors: Yahi, H.; Marticorena, B.; Thiria, S.; Chatenet, B.; Schmechtig, C.; Rajot, J. L.; Crepon, M.
• Is Part Of: Journal of geophysical research. Volume 118:Issue 23(2013)
• Page Start: 13, 265
• Page End: 13, 281
• Publication Date: 2013-12-12
• Subjects: Atmospheric physics -- Periodicals; Geophysics -- Periodicals; 551.5
• DOI: 10.1002/2013JD019465
• Languages: English
• ISSNs: 2169-897X

Abstract

<abstract abstract-type="main"> <title>Abstract</title> <p>[1] This work aims at assessing the capability of passive remote‐sensed measurements such as aerosol optical depth (AOD) to monitor the surface dust concentration during the dry season in the Sahel region (West Africa). We processed continuous measurements of AODs and surface concentrations for the period (2006–2010) in Banizoumbou (Niger) and Cinzana (Mali). In order to account for the influence of meteorological condition on the relationship between PM10 surface concentration and AOD, we decomposed the mesoscale meteorological fields surrounding the stations into five weather types having similar 3‐dimensional atmospheric characteristics. This classification was obtained by a clustering method based on nonlinear artificial neural networks, the so‐called self‐organizing map. The weather types were identified by processing tridimensional fields of meridional and zonal winds and air temperature obtained from European Centre for Medium‐Range Weather Forecasts (ECMWF) model output centered on each measurement station. Five similar weather types have been identified at the two stations. Three of them are associated with the Harmattan flux; the other two correspond to northward inflow of the monsoon flow at the beginning or the end of the dry season. An improved relationship has been found between the surface PM<sub>10</sub> concentrations and the AOD by using a dedicated statistical relationship for each weather type. The performances of the statistical inversion computed on the test data sets show satisfactory skills for most of the classes, much better than a linear regression. This should permit the inversion of the mineral dust concentration from AODs derived from satellite observations over the Sahel.</p> </abstract>