INFIQC - CONICET / CLCM / Departamento de Fisicoquímica, Facultad de Ciencias Químicas, Universidad Nacional de Córdoba, Ciudad Universitaria, 5000 Córdoba, Argentina
Cite this article: Lanzaco, B.L., Olcese, L.E., Palancar, G.G. and Toselli, B.M. (2017). An Improved Aerosol Optical Depth Map Based on Machine-Learning and MODIS Data: Development and Application in South America.
Aerosol Air Qual. Res.
17: 1623-1636. https://doi.org/10.4209/aaqr.2016.11.0484
HIGHLIGHTS
A method to obtain improved MODIS values in large areas has been developed.
The improvement zone covered to 62% of South America area.
Validated against AERONET, the data fraction inside the MODIS error increased 86%.
Monthly averages differences of up to ±0.6 AOD units have been found.
ABSTRACT
In zones where aerosol properties have been poorly characterized, satellite-based (MODIS) and ground-based (AERONET) aerosol optical depth (AOD) values typically differ. In this work, we use machine-learning based methods (artificial neural networks and support vector machines) to obtain corrected AOD values taken from MODIS in regions that are positioned far from AERONET stations. The method has been validated using several approaches.
The area suitable for improvement covers 62% of the South American continent, and the degree of improvement compared to MODIS values, expressed in terms of the fraction of data within the MODIS error, was found to be 38% and 86% for the Terra and Aqua satellites, respectively. The results show absolute monthly average differences between the MODIS and the proposed method of up to ± 0.6 AOD units. The MODIS AOD distribution for the analyzed period shows a mode of –0.04, while that for the method presented here is 0.08.
