Robin's Blog

Data sources for parameterising Radiative Transfer Models & atmospheric correction algorithms

There was a question recently on the Py6S mailing list about what data sources are best to use to provide atmospheric parameters (such as AOT, water vapour and ozone) for use with 6S, other atmospheric Radiative Transfer Models (such as MODTRAN) or other atmospheric correction algorithms (such as ATCOR). In the spirit of ‘reply to public’ I decided I’d post the response here, as it’d certainly be easier for other people to find in future!

The first thing to say is if you’re doing atmospheric correction, then by far the best solution is to derive as many of the parameters as possible from the satellite data that you are trying to correct. Sometimes this is relatively easy (for example, when using MODIS data, as there are standard MODIS products for AOT, water vapour and ozone) and sometimes harder (there are no standard atmospheric products for SPOT, and algorithms for estimating atmospheric parameters from Landsat are relatively limited). The main reasons for this are:

  1. The atmospheric data will be from the exact time of the satellite image acquisition
  2. Depending on the satellite, the atmospheric data may be provided for every pixel in the image, and if this isn’t the case then there might be more than one value across the image (for example, the LEDAPS algorithm estimates AOT over all pixels containing Dense Dark Vegetation). This is key as it allows the atmospheric correction algorithm to take into account changes in atmospheric conditions across the image. I have published research (Wilson et al., 2014) showing that failing to take into account this variability when doing atmospheric correction can lead to significant errors.

However, if you’ve found this post then you’ve probably either already tried using data from the satellite you’re working with, and found it isn’t possible – or you’re doing radiative transfer modelling for some other purpose than atmospheric correction, so the explanation above had no relevance to you. So, on with the other data sources – and in Part 1 we’re going to cover aerosol-related data.

Aerosol Optical Thickness

(Also known as Aerosol Optical Depth, just to confuse you!)

This is a measure of the haziness of the atmosphere, and is set in Py6S as: s.aot550 = x Please note that this means AOT at 550nm (yes, AOT is wavelength dependent!), so you may have to interpolate the AOT data from different wavelengths. I’ll post a separate post soon about how best to do this.

When choosing an AOT data source, you usually have a trade-off between accuracy, spatial resolution and temporal resolution. If you’re lucky, you can manage to get two of those, but not three! So, going in order of accuracy:

AERONET

The Aerosol Robotic Network (AERONET) is the gold-standard in terms of accurate measurement of AOT. The measurements are acquired from ground-based sun photometers, which have the major benefit of being able to produce very accurate measurements at a high frequency – as they don’t depend on satellite orbits. However, there are two major disadvantages: they have a very low spatial resolution, and they only produce measurements when the sun is shining.

Specifically, although there are over 1,000 measurement sites listed on the AERONET website, many of these are for short-term measurement campaigns and so the number of sites regularly recording measurements over a long period of time is much lower. For example, excluding short-term sites, there are only three measurement sites in the UK (Hampshire, London and Edinburgh), and there are many parts of the world with very few sites (such as central Africa).

Also, as mentioned above, measurements are only acquired when there is a direct view of the sun from the instrument. This obviously means that measurements are not acquired during the night (which can stretch for many months near the poles), but also that cloud cover has a significant impact on the frequency of measurements. The AERONET cloud-screening algorithms have to be very strict, as even a tiny amount of cloud will cause a large error in the AOT measurement.

Good for: Very accurate measurements. High temporal resolution.

Bad for: Spatially-dense measurements. Cloudy areas.

Data available from: NASA AERONET site

Other sun photometer data

AERONET run a large network of sun photometers, but there are many other sun photometers run by other people/groups, and these can be good sources of data. Indeed, you may even have acquired your own measurements using your own sun photometer (such as the Microtops sun photometer pictured below). The advantages and disadvantages above still apply, although if you’ve taken measurements yourself then I assume they are in the right place and at the right time!

Data available from: There is a (slightly old now) list of other sun photometer data sources online. If you’ve got Microtops data that you want to process then have a look at my PyMicrotops module that makes the process a lot easier.

MODIS AOT

Data from the MODIS sensor on the Aqua and Terra satellites is operationally processed to produce AOT data at 10km and 3km resolution, the MOD04 and MOD04_3K products. AOT estimates are produced over all land surfaces, but measurements over bright surfaces (such as deserts and snow) tend to have a higher error.

The Aqua and Terra satellites are in a sun-synchronous orbit, so passes over each location at approximately 10:30 and 13:30 local solar time, giving a total of two measurements a day. Again, cloud cover is an issue, and the MODIS algorithm masks out any pixel containing cloud.

The MODIS data is provided as HDF files with loads of Scientific Data Sets (SDS, basically bands) – the one you want to use is called Optical_Depth_Land_And_Ocean, which is high quality AOT data (where the QAC, the measure of retrieval quality, is 3) from both land and ocean, using all algorithms.

The MOD08 dataset provides the same data at a lower resolution (1 degree), if spatial resolution is not important for you.

Good for: Moderate spatial resolution. Operational data availability. Reasonable accuracy.

Bad for: Temporal resolution. Cloudy areas.

Data available from: LAADSWeb (Unfortunately it comes in an annoying projection, but you can reproject it using the MODIS Reprojection Tool or HEG. MOD08 doesn’t have this issue, and can also be accessed through Google Earth Engine)

MAIAC

The Multi-Angle Implementation of Atmospheric Correction (MAIAC) algorithm (Lyapustin et al., 2011) produces AOT data at 1km resolution, through a novel method which combines a range of MODIS observations with differing view geometries.

Good for: High spatial resolution.

Bad for: Not currently operationally available.

Data available from: Data is not currently available online, but I think it is due to be released as an official MODIS product within the next six months or so.

Aerosol Type

There are lots of ways of parameterising aerosol types in radiative transfer models (see here for a list of the ways you can do this in Py6S) – and it is often difficult to work out what the best option to use is. I’ve listed some options below, but this is unlikely to be a comprehensive list – so please leave a comment if you know of any other data sources.

AERONET

As well as providing AOT data, AERONET sun photometers also acquire other measurements which can be used to infer the aerosol type. This is almost certainly the most accurate way of setting the aerosol type – and there is a built-in function in Py6S that will help you do it. See here for full details on how to use it: basically you just download the AERONET data and give Py6S the filename and date/time and it does the rest.

Data available from: NASA AERONET site (find AERONET Inversions on the left-hand side)

MODIS

The MOD04 product mentioned above also provides an estimation of aerosol type, in five categories: dust, sulfate, smoke, heavy absorbing smoke and mixed. It’s not entirely clear how best to match these to the predefined 6S aerosol types (which are biomass burning, continental, desert, maritime, stratospheric and urban) – although there is an option in 6S to set a user-defined aerosol profile, which allows you to specify the relative fraction of each of dust, water, oceanic and soot aerosols (see AeroProfile.User)

Data available from: LAADSWeb (Unfortunately it comes in an annoying projection, but you can reproject it using the MODIS Reprojection Tool or HEG.)

Aerosol type climatologies

A number of aerosol type climatologies have been produced, giving an indication of the typical aerosol type observed at each location, over time. These climatologies vary in the level of detail – both in terms of spatial resolution (which is often coarse) and temporal resolution (often monthly or seasonally). The best way to find these is to search for papers about aerosol climatologies, and then either look for a URL in the paper or contact the authors to ask for the data. I’ve only included a couple of examples here as new climatologies are being published frequently:

So, that’s it for this part. Please let me know of any datasets that I’ve missed – and stay tuned for Part 2 which will cover other parameters (water vapour, ozone and so on).


If you found this post useful, please consider buying me a coffee.
This post originally appeared on Robin's Blog.


Categorised as: Academic, Py6S, Remote Sensing


4 Comments

  1. Adhithya says:

    WRT MODIS AOT section: Isn’t MODIS spatial resolution low (3km)? With high temporal resolution?

  2. Robin Wilson says:

    Thanks for the comment. Unfortunately, MODIS AOT is considered reasonably high-resolution as far as AOT data goes: there isn’t much proper high-resolution data available. However, to make it a bit clearer I’ve updated it to say ‘Moderate spatial resolution’.

  3. Andrea Nesdoly says:

    Did you happen to publish a part two outlining the other parameters (water vapour, ozone, etc)? Where might I find it? 🙂

  4. Robin Wilson says:

    I don’t think I ever got round to part two – sorry! I’m no longer working directly in atmospheric correction, so I’m not sure I’d know exactly what the best datasets currently available are, so I’m probably not the right person to write Part 2 now. Sorry I can’t be more help.

Leave a Reply

Your email address will not be published. Required fields are marked *