I found that for scenes over the areas where SRTM30 is absent, TC step still runs, but produces scenes that are not geolocated properly. Which makes sense - Luis earlier wrote: “For pixels without elevation it will still add the gravitational model. Using Ellipsoid Correction RD is the same thing but it doesn’t use a DEM and instead uses the average elevation given in most products.”
However, the resulting error is quite large. I assume this procedure will always be applied for the purely ocean scenes, but it’ll be sad if they will always contain errors. What is the estimate of the expected error over such no-DEM areas?
It won’t help you with your question, but could you write here your projection parameters: EPSG (from… to…), type of interpolation, product type, pixel sampling.
Consider using polar stereographic projections if your study is somewhat in a polar region. As well as for purely ocean or iceberg regions without land I would suggest using Ellipsoid Correction only.
Thank you, I will try ellipsoid correction only for comparison.
We (Earth Engine) are processing all GRD scenes. We project them into EPSG:4326 using SRTM 30, with the rest of TC parameters at default values.
Out of curiosity, what would be the advantage of using a polar projection? We reproject everything into Web Mercator for browsing and analysis anyway (though this makes working with polar areas difficult, and we at some point want to add an alternative view that uses a polar projection).
@simonf
What error magnitude are you observing? What is your reference to say that ocean-only images are not properly geolocated?
I don’t see why Terrain Correction over the ocean should be much worse than over land, provided that the appropriate input data (i.e. heights) is used.
DEM heights are used over land. Geoid height (i.e. difference between the geoid and the reference ellipsoid) can be used as a substitute of the DEM in ocean images. If I understand Luis’s sentence above correctly, this is what S1TBX does (gravitational model = geoid height). The geoid closely matches the sea surface. There can be a small difference between the two (see ‘sea surface topography’, tides, waves…) which will translate in a shift (error) in range, but I’d expect this shift to be in the order of a few metres only.
Depending on the location, the geoid height goes from approx. -100 to +80m wrt WGS84 reference ellipsoid.
I’d expect Ellipsoid Correction of ocean images to be less accurate than Terrain Correction. All such images I have looked at are annotated with very low heights (e.g. 0m, 10^-4m…) probably because there is no DEM in that location. Since Ellipsoid Correction seems to be using those annotated heights, then we could be in a worse case of 300m shift in ground range (worse case: when vertical shift is 100m, and incidence angle is 20deg). Terrain Correction should not give such a large error if a gravitational model is used.
@css: To clarify, my original complaint was that TC over land for an area where no SRTM 30 is available showed a very large error (hundreds of meters), but this is not a situation I expect to occur over land once I switch to the complete SRTM dataset. But that made me wonder about what to do for ocean areas with no SRTM, though I have not observed any errors over ocean yet - this was a preemptive question.
Following the recommendations above, I have tried ellipsoid correction using both Range-Doppler and Geolocation-Grid without TC on a scene that covers Miami S1A_IW_GRDH_1SSV_20151221T231936_20151221T232001_009147_00D292_B9FC), which is flat enough for elevation averaging to be valid. The result is quite well geolocated in both cases - it’s shifted a couple of pixels east, but it’s certainly good enough for pure oceans scenes.
I’ll modify my processing chain to apply Ellipsoid Correction for areas where no DEM is available. Does this sound reasonable?
Ultimately, geolocation based on Range-Doppler will be as accurate as the exactness of the height data you provide. S1TBX Terrain Correction and Ellipsoid Correction use two different sets of height data or models. In no-DEM areas, TC uses a gravitational model (geoid); EC uses the annotated heights (which for S1 are referenced to sea level) and, by design or by omission, no gravitational model. Which operation (TC or EC) is more accurate for a given no-DEM image will depend on which of the height models better represents the true height of the scene. For ocean-only images, TC will be more accurate, since the geoid is a good representation of the true height of the sea surface (EC will be as accurate as TC in the locations where geoid height = 0). I don’t think it is possible to predict which method will be more accurate in no-DEM land areas: it will depend on true height, geoid height and annotated height (is it 0 for no-DEM areas?).
The geolocation error you are seeing when you EC the Miami scene is because in that part of the world the geoid height is around -30m. EC does not take this height into account, and the result is a shift in the range direction. I’d expect S1TBX TC to be more accurate in that image even if no DEM were available.
You are right - when I choose TC with ASTER DEM without using a local copy, the result is geolocated better than EC output. This is great - then I don’t even need to switch between EC and TC for no-DEM sea areas. I’ll use ASTER for land areas where SRTM is missing. ASTER has no data only for very top of Greenland and large areas of Antarctica (where elevation will definitely be a problem), so might revisit this issue after processing everything else.