Why we need a new way to describe spatial resolution in SAR imagery

We’d like to promote the use of a new index of spatial resolution for radar images, specifically designed for distributed targets (such as forests, agricultural fields, glaciers, etc).  The influence of speckle on radar imagery means that, for terrestrial application at least, the proper definition of resolution for Synthetic Aperture Radar is inadequate. At best, it’s unhelpful – at worst, it’s misleading.

The increasing availability over the last decade of both space-borne and airborne high-resolution radar systems has given the user community an ever wider selection of observational tools. But this abundance of choice has created a new problem – users must be able to easily assess the data quality and compare performance across sensors. Spatial resolution is one key parameter that is used to make such a comparison and in my experience spatial resolution of radar imagery is neither presented consistently across data providers, nor is it widely understood by many users of the data. Inconsistency and confusion often leads to inappropriate application and eventual disappointment on behalf of many users.  I’ve talked to many users, more used to optical imagery, who get frustrated when data that is labelled “10m resolution” turns out in practice to be nearer a 30m Landsat than a 10m Spot 5.

We’ve tried to tackle this in a recent publication (Woodhouse et al 2011) by describing a more generic index of spatial resolution – the ”equivalent spatial resolution” (ESR)  – that quotes the resolution of a square pixel at 12 looks. Our approach considers the separability of distributed targets that differ in intensity by a known contrast ratio. Using a 3 dB difference as an arbitrary, but reasonable, threshold value, we show how the resolution of a 12 look image is closer to what might reasonably be expected by someone more used to working with optical imagery.

Spatial resolution is a key performance indicator of remotely sensed imagery and is defined as the separability of two idealised point targets. The point target is characterized by a Dirac delta function and so does not indicate visibility, which is a property of the contrast of any individual target with its background. The limiting factor of any imaging system is therefore the width of the end-to-end point spread function (PSF), or more exactly the ability to separate two overlapping PSFs. The problem in radar imagery is that speckle, the noise-like modulation of the signal due to wave interference, does not effect point targets.  This strict definition of spatial resolution does not allow direct comparison between differently processed SAR data (quoted at different Looks), nor between SAR data and comparable optical data.

We hope our paper offers a neat little summary of the issues of spatial resolution in radar imagery, and offers a more consistent way to describe the spatial resolution in these images. Given its clear meaning, if this index was widely used it would allow the user community to better assess the performance of different SAR systems and how they compare to their optical counterparts.

If you agree with this post, please SHARE it using the buttons below. 

If you disagree, or want to offer an alternative, please comment below.  

If you don’t have access to the IJRS (required to download the full paper, as published, from link below), the pre-published (unformatted) submission version is available here

I. H. Woodhouse, A. Marino & I. Cameron (2011), A standard index of spatial resolution for distributed targets in synthetic aperture radar imagery. International Journal of Remote Sensing, 32(23):7929-7938.   DOI:10.1080/01431161.2010.502546

4 thoughts on “Why we need a new way to describe spatial resolution in SAR imagery

  • This is a helpful discussion of the issues with measuring resolution of SAR images. For polarimetric images, one needs even more looks to obtain a usable product. The multilook polarimetric images released by the NASA/JPL UAVSAR airborne radar (http://uavsar.jpl.nasa.gov) have a total of 36 looks (3 in range and 12 in azimuth) for an approximate pixel spacing (and effective resolution) of 7 m.

  • Thanks, Eric. Good point. We didn’t really tackle the issue of polarisation as it will depend on the application and method of analysis too much. You could imagine one user would just use the span image, while another wants the fully polarimetric information. Perhaps we should be saying that we should quote a square pixel size for an equivalent number of looks of AT LEAST 12. A user wouldn’t complain if the data quality is better! 🙂

  • I appreciate your point but 12 looks is an unusual degree of multilooking for most applications. You’d also need to explain multi-looking to new users to allow them to judge the trade off between speckle suppression and the degradation in effective spatial resolution.

  • Blamannen, Your are right that 12 looks might seem unusual. Any “standard” would be arbitrary, but the important thing (in my opinion) is that there IS a standard, and everyone uses it. That way there is actually less need to explain to new users the speckle-resolution trade off. If all radar data is provided to users at different looks, and at a number of looks which isn’t actually fit for their purpose, then they DO need to understand it before they can use it properly.
    There is an argument that adaptive averaging keeps resolution-integrity while reducing speckle, and it is true that a standard 12-look image format would not allow that. However, our suggestion is that the spatial resolution is always QUOTED as an equivalent 12 look image, not necessarily supplied as 12 look.

Leave a Reply

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