About AEROMAP

AEROMAP logo
  
 
  Gateways into AEROMAP
1

Go to: [Contributors]  [Objectives]  [Methods]  [Results]  [Milestones]  [Phases]


Abstract

The International Panel on Climate Change (IPCC) has identified that the largest uncertainty in current estimates of planetary radiation forcing is due to atmospheric aerosols and has called for an urgent expansion of global studies to monitor and characterise them. Aerosols are characterised by their optical and microphysical properties and while the aerosol robotic network (AERONET) of remote sensing instruments provides accurate values for them, AERONET's coverage of the Earth surface is sparse (mostly city-based) and patchy. The AEROMAP project is designed to provide a solution to overcome this lack of information on global aerosols without the need to invest in hundreds of new AERONET sites. To achieve this, AEROMAP will harness and capitalise on the high resolution full-Earth measurements of aerosol optical depth provided daily by satellite remote sensing instruments. Furthermore, by using satellite-driven data, AEROMAP will be able to gain access to aerosol characteristics over the oceans which are of paramount importance to the estimate of the overall radiation budget. As a result, the retrieval of aerosol optical and microphysical properties by AEROMAP will significantly extend the efforts to globally monitor and characterise atmospheric aerosols.

AEROMAP has developed and validated data mining tools based on neural networks to convert satellite measurements of aerosol optical depth and columnar water vapour into aerosol optical and microphysical properties like the single scattering albedo, particle asymmetry factor, complex refractive index and the aerosol size distribution. This is a very challenging problem as satellite instruments, unlike their ground-based counterparts in AERONET, do not yet have the capability to provide polarisation information obtained from the diffuse radiation field. In essence, AEROMAP is attempting to produce a robust inversion algorithm for the retrieval of aerosol parameters analogous to that used by AERONET but without this source of information. AEROMAP's inversion algorithm will then open the door to global aerosol monitoring and characterisation. The inputs to AEROMAP are a small but specific set of satellite measurements. Specifically, daily, full-Earth measurements of the aerosol optical depth (AOD) provided by the MODerate resolution Imaging Spectrometer (MODIS) satellite instrument in 3 wavelength bands (438-448nm, 673-683nm, 862-877nm) spanning the visible spectrum and the near-infrared taken together with a near-infrared measurement of the columnar water vapour, are used as well as the best estimate of the absorption aerosol optical depth (AAOD) at 500nm provided by the Ozone Monitoring Instrument (OMI) satellite. The outputs from AEROMAP are retrievals of over 40 optical and microphysical parameters needed to characterise aerosols. This is accomplished by feeding the satellite data into neural networks that have been trained on AERONET "ground-truth" data to learn the relation between the inputs and output parameters. The result will be global maps of aerosol optical and microphysical parameters at a resolution of 1o x 1o (50km x 50km) which will be used to monitor and classify aerosols as they move daily across the Earth's surface.

The main objectives of AEROMAP are to code neural networks to learn the relation between existing satellite inputs and aerosol optical and microphysical parameters for different types of aerosol. AEROMAP will use co-located and synchronous ground-truth data to test that the trained networks can extrapolate the aerosol parameters retrieved from the satellite inputs at the training data location to new and distant geo-locations. AEROMAP will then generate daily-updated global maps of aerosol optical and microphysical parameters and classify aerosol in each pixel by type. AEROMAP will apply the networks to satellite measurements of important extreme aerosol events such as desert dust storms, forest fire outbreaks, urban brown cloud episodes, volcanic eruptions and radiation clouds to assess whether or not it is possible to monitor and track their spatio-temporal characteristics. By updating the global maps with daily satellite over-pass data, AEROMAP will function as a near-real-time monitor of aerosols whose maps and data files will be made publicly-available at the project website/portal. The global aerosol maps will be assessed to develop an aerosol impact scale that will be used for the issuing of alerts and early-warning information about aerosol-related hazards.

This project is funded by Marie-Curie Actions at APCG.


CONTRIBUTORS

Dr Stelios Kazandzis  (Scientist in Charge)  APCG-IERSD-NOA
Dr Michael Taylor  (Research Fellow)  APCG-IERSD-NOA

For individual contributors to publications during the course of the AEROMAP project, please see AEROMAP Co-author Index


OBJECTIVES (O1-9)

✔  O1: to train MIMO-ANNs to identify the functional relationship between AERONET direct sun AOD data and AERONET AMP retrieval data.
✔  O2: to investigate the ability of ANNs trained and validated for one site to extrapolate to qualitatively-similar sites worldwide (e.g. desert dust, biomass burning, urban pollution).
✔  O3: to train MIMO-ANNs to identify the functional relationship between MODIS satellite AOD data and AERONET direct sun AOD data.
✔  O4: to investigate the ability of ANNs trained and validated on co-located sites to extrapolate to regions where no sites exist using MODIS AOD and MIMO-ANN derived functions.
✔  O5: to classify and characterise the global ANN-derived AMP by aerosol type using cluster analysis.
✔  O6: to study the time-dependent dispersal of aerosol clusters worldwide by tracking spatial variations in characterized ANN-derived AMP on global maps for a number of climatologically and/or socio-economically important cases.
✔  O7: to produce accurate, daily-updated, global maps of AMP characterised by aerosol type.
✔  O8: to test the accuracy and feasibility of creating a global near real-time monitor of aerosols and an alerting service for the assessment of climatological risks and the issuing of early-warnings.
✔  O9: to engage in science communication to inform the public of the project, its results and the potential impact of global aerosol characterisation on Euroepan Research Area (ERA) environmental policy.


METHODS (M1-5)

✔  M1: Machine learning, function approximation and neural network generalisation (O1-3)
✔  M2: Independent assessment and statistical uncertainty analysis of AMP extrapolations (O4)
✔  M3: Cluster analysis and aerosol type classification (O5)
✔  M4: Dispersion analysis (O6) and real-time monitoring/alerting feasibility assessment (O7-8)
✔  M5: Website design and science communication (O9)


RESULTS (R1-9)

✔  R1: Identification of the mathematical mapping between AERONET direct sun AOD data and AERONET AMP retrieval data.
✔  R2: Verification and uncertainty analysis of the ability of ANNs to extrapolate AMP retrievals to regions of similar aerosol type.
✔  R3: Identification of the mathematical correlation mapping between MODIS satellite AOD data and AERONET direct sun AOD data.
✔  R4: Verification and uncertainty analysis of the ability of ANNs to extrapolate AMP retrievals to regions where no sites exist.
✔  R5: Rendering of global ANN-derived AMP maps with aerosol typing.
✔  R6: Pilot studies of aerosol temporal variation (tracking) for a number of climatologically and/or socio-economically important cases.
✔  R7: Production of accurate, daily-updated, global maps of AMP on a global grid of resolution 1 degree CMG.
✔  R8: Feasibility testing of a global real-time monitor of aerosols and production of microphysics-based air quality indices for the assessment of climatological risks and the issuing of early-warnings.
✔  R9: Creation of a project website/portal, science communication & dissemination of results.


MILESTONES (M1-4)

✔  M1: Assessment report on the function approximation ability and extrapolation power of the MIMO-ANNs (paper in preparation)
✔  M2: Mid-term report (month 12)
✔  M3: Assessment report on the classification of aerosols and the results of assessing the feasibility of developing microphysics-based air quality indices and a real-time monitor and online alerting service (month 18)
✔  M4: Final report (month 24)


PHASES (A-E)

pdf

✔  Phase A (Months 1-6): Machine learning, function approximation and generalisation:
✔  A1 (months 0-3): Data acquisition and pre-processing
✔  A2 (months 0-3): Function approximation of the relation between MODIS AOD and AERONET AOD data
✔  A3 (months 3-6): Function approximation of the relation between AERONET AOD and AERONET AMP inversion products at selected sites

✔  Phase B (months 6-12): Independent assessment of results using AMP inversion algorithms:
✔  B1 (months 6-9): Validation of extrapolation power by training with capped data
✔  B2 (months 6-9): Validation of extrapolation power by training with one aerosol type and validating with another

✔  Phase C (months 9-15): Cluster analysis, aerosol typing and case studies:
✔  C1 (months 9-12): Cluster analysis of global AERONET sites
✔  C2 (months 9-15): Pilot studies
✔  C3 (months 12-15): Production of global maps:

✔  Phase D (months 15-24): 3D spatio-temporal mapping and real-time monitoring/alerting feasibility study
✔  D1 (months 15-18): Assessment of the feasibility of real-time monitoring
✔  D2 (months 18-24): Development of microphysics-based air quality indices for issuing alerts

✔  Phase E (Months 1-24): Website design, maintenance and public engagement:
✔  E1 (months 0-9): Design and initiation of the project website
✔  E2 (month 12): Public engagement
✔  E3 (months 12-21): Outreach science communication & dissemination of results activities
✔  E4 (month 23): Development of the project website with educational resources