Publishing satellite observations to GBIF

1

This is related to this topic but maybe slightly different (also comments closed there). Has anyone published satellite species observations to GBIF? We’re curious what your approach is. I am working with the Institute for Marine Remote Sensing team and they have been assessing distribution of seagrass using satellite imagery as well as red mangrove. These are rasters. Given that data shared to GBIF is tabular, should it be converted to polygons and a footprintWKT created for each polygon? Do we need a centroid for each polygon so we have a decimalLatitude and decimalLongitude? Or would it make more sense for each pixel in the raster (centroid of pixel = decimalLatitude and decimalLongitude) to be a present or absent for each species? Are GBIF users using footprintWKT for analyses? Is it easy for users to work with?

@daiesco I am working with the NOAA Passive Acoustic Reporting System on a crosswalk of their template to Darwin Core and sharing to GBIF and OBIS. We are waiting for them to finalize their database and then hopefully publish a dataset. Happy to share our process with you once we have that published.

3 Likes
2

Hi Abby,

GBIF currently does not have guidelines for publishing satellite occurrence data, but I will do my best to answer your questions below.

Has anyone published satellite species observations to GBIF?

Yes! Some publishers have published satellite based occurrence datasets: Search (The link referred to migratory datasets using GPS location tracking. Not relevant for satellite image-based occurrences.)

Given that data shared to GBIF is tabular, should it be converted to polygons and a footprintWKT created for each polygon?

As I mentioned, there is not really any standard for how such data should be shared with GBIF, but perhaps you can find inspiration here and other parts of the guide may be relevant to you as well.

One option could be, for each occurrence record, to provide both a footprintWKT for the entire polygon, a centroid of the polygon as decimalLatitude and decimalLongitude and coordinate uncertainty. How large are the pixels you refer to? I guess it is satellite system dependent? I still need to mull over the idea of using each pixel as a basis of the occurrence and would like to know how many pixels on average are in a polygon for example. But it may be the most detailed way to share the data (and use the data as well).

Are GBIF users using footprintWKT for analyses? Is it easy for users to work with?

I will answer these two in one go. We do not know if users use footprintWKTs for analysis. This might be possible to find out through a literature search, but @dnoesgaard probably knows better than me whether it is possible.

Working with footprints require some technical spatial/GIS skills and my understanding is that most people rather deal with coordinates and the uncertainties associated with the coordinates. But for the technically proficient users, I think footprints could potentially be very useful and perhaps we will be able to visualise them in the portal with the new data model.

We are waiting for them to finalize their database and then hopefully publish a dataset. Happy to share our process with you once we have that published.

Thank you - we would love to hear more of your process! It is some very relevant questions you pose since we see more and more of these satellite-based monitoring efforts across the world. I would also be very interested to hear how certain the identification of the different species is - it is my understanding that it can be challenging to differentiate between e.g. a rock bed and seagrass, especially with high turbidity etc.

I hope this answers your questions but let us know if you if you have any follow up question.

1 Like
3

@abbybenson it would be wonderful to know the process. With @camisilver we will be looking forward to it!

1 Like
4

Yes! Some publishers have published satellite based occurrence datasets: Search

From a quick browse I see only satellite-tag data. The data Abby and I are working with is species occurrence derived from satellite imagery. These are more akin to aerial survey data, but on a larger scale. The seagrass and mangrove maps we are generating have a millions of pixels where seagrass or mangrove were detected. The two leading representations are 1) many occurrence points spread (~2m) apart or 2) chunks of occurrences with many individuals and a footprintWKT encircling each field/forest. Neither seems ideal.

Are GBIF users using footprintWKT for analyses? Is it easy for users to work with?

I consider myself a relatively advanced GBIF/OBIS consumer, and I don’t want to deal with footprintWKT. Its usage seems rare, and there seem to be multiple valid interpretations of its use (spatial area covered over time by one individual, spatial area covered by one big individual, spatial area covered by multiple individuals). For something like a forest, I would prefer a grid of points where trees were detected rather than a single occurrence row for the entire forest. For the latter I would likely end up dropping it instead of creating my own sampling grid.

If footprintWKT becomes more frequently used, then I suggest the API will need to include methods for “raster”-izing these into sets of points in a grid. Otherwise these large areas will be undersampled in analyses, and the centroids will confound models unknowingly using vector data.

5

Yes, you are right - the datasets I linked to refer to migration datasets and not what you were looking for. @trobertson I think this discussion is relevant for your work with the new data model - it would be interesting to have a remote sensing use case with occurrences derived from satellite imagery.

6

This sounds like

would be most fit for use in that case?

7

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.