How should you set accuracy?

There was a bit of discussion on one of my observations about the accuracy field. What’s the best practice for setting accuracy when the location was set automatically from the photos geotag metadata, when that was from a phone GPS?

Unfortunately my camera app and no camera app that I’ve found will embed the GPS DOP into the EXIF metadata.

I was able to record a GPS trace separately which does record the GPS DOP and I could correlate this back into the EXIF metadata into the standard GPSDOP field.

However when then using this photo as my observation, iNat didn’t try to set the accuracy from the GPSDOP. I realise that DOP is not a direct accuracy in meters, but apparently the app I was using to record it was just setting it based on the error in meters anyway.

Should I still be attempting to set an accuracy per observation derived from the GPS DOP? If iNat could derive it from the GPSDOP EXIF field that would make it much easier to provide this information?

Is there a better way to say the location is from GPS and hence has GPS level accuracy, and not bother with the exact error in meters?

The person who commented in https://www.inaturalist.org/observations/58310748 said they just ignore observations without an accuracy and suggested to just use 10m instead, but that feels wrong to just make up an average number. Maybe if I could say 10m as an average but could be anywhere from 1m to 100m depending on the factors that affect GPS accuracy that would be more helpful (essentially saying it’s GPS accuracy).

3 Likes

There is no single, simple solution for this. Accuracy/precision of an observation depends on multiple factors, including the precision of the GPS point, the observer’s distance from the subject (esp. for telephoto bird observations), and the potential mismatch between the GPS location and the iNaturalist base map. In my personal case, I don’t have a GPS equipped camera and use time-matching of photos to a GPS track recorded with a separate device, so there’s also the factor of my travel speed and the GPS vs camera time mis-match. This means the GPS DOP recorded in EXIF would only be one contributing factor, and shouldn’t be used directly - so don’t worry that your camera doesn’t record it.

When I’m uploading a set of observations, I usually will use rules-of-thumb to set the accuracy for the whole set of observations (via “batch edit”). Rules of thumb for GPS precision are based on sky visibility for a consumer-grade GPS unit (like a phone): 5-10 m on flat, open terrain, 10-20 m under moderate tree cover, 20-30 m in dense forest, 50+ m in canyons or near buildings. Then, if I’m not taking close-ups, I add a guestimate of my average distance to subject. Once I’ve done the batch edit, I go through my observations one-by-one to check that the circle on the map matches my memory of the actual location, and adjust location or accuracy for individual observations to fix any problems.

The goal is to make sure that the accuracy circle has a high probability of including the true location, so if in doubt, err on the high side.

4 Likes

That was a pretty helpful discussion on your observation of some of the issues and options. Best practice is going to vary with individual work flows.

My typical field work flow is DSLR photographs, with separate consumer-grade GPS waypoints or high-density track recording the locations. I tend to work in open desert or high mountains with good views of the sky. With long experience and comparison with imagery, I have found that my GPS coordinates are within 5 meters of true, and that should be the case for any consumer GPS manufactured after about 2010 - again with a clear view of the sky, and sufficient time to acquire the satellite constellation first (usually 30 seconds or less with modern chipsets). Mostly I just keep the GPS on.

My previous GPS model was vintage 2004, and I settled on 20 meters as a reliable error level for that one. If anyone has pre-2001 GPS coordinates, watch out. Selective availability was still turned on then, and errors could be up to 100-200 meters. And of course, in dense canopy or deep canyons, all bets are off.

I apply the appropriate number above to batches of observations without reference to actual recorded DOP values. BUT… I always do a reality check against imagery for each batch of observations, either outside iNat before uploading, or within iNat after upload using the Satellite view on the map and zooming way in. Very occasionally I will find an observation that seems to exceed my standard error, and then I’ll either edit the point closer to the spot I remember, or I’ll increase the “accuracy” distance if I can’t tell for sure.

I rarely use the phone app, but when I do, I put the phone in airplane mode, turn off auto-sync in the app (permanently), run GPS Status (or another GPS app) in the background to keep the phone’s GPS active, and use the in-app camera to photograph the observations. I’ve consistently gotten the same 5 meter (or better) accuracy after verifying against imagery.

4 Likes

I took a look at the HDOP values being saved by my GPS tracking app and it only helped distinguish areas of tree cover (value of 2) from no tree cover (value of 1.5), so while that might help me try to set a value in meters automatically, I’m not sure it’s going to be that helpful actually.

So it sounds like I’ll need to do what I was trying to avoid, making a subjective decision for each (or bulk group) of observations.

I wish I could then say the source of my accuracy is based on this subjective opion. The same way iNat tries to keep track of the source of the coordinates (GPS, manually placed on the map, etc), it might be useful to have something like that for accuracy.

I think that in such case is necessary to use 100 m accuracy: only with this size you can guarantee, that the true point is really inside the circle.