All elevation data shows a hill in the middle of the job

Hmmm. That's a weird one...

A hump in the center of the map is caused by a camera calibration that we calculate based on the images being slightly off. In an attempt to keep the map photo layer lining up as well as possible it can make the model exhibit the curvature you are seeing here.

You certainly have enough overlap, so that isn't the problem.

There isn't really anything that can be done to warp the model while keeping the map accurate.

Have you tried any other methods of processing it to see how it compares?

Since the majority of jobs coming into our system are only for a few different cameras we are looking at ways of checking our calculated calibrations based on historical data but are still currently doing a fresh one for every job to keep things general.

Well, early on I tried processing a job like this through some other software and got the same hump, but that was a early version of the app.  I originally tried processing jobs with GCP, but I haven't tried that in a long time.  Do you think that would help?

We stumbled over the same problem very recently. According to MME there's not much that can be done to prevent this. GCPs would not solve this problem, as they are only used to localise better the map thats already generated from the photos. We try now to remove this hill-artefact by removing it using our own measured GCPs. This can be done in principle by fitting a low-order polynomial surface to the GCPs and also to the DEM at these GCP locations. The difference between the two surfaces is then subtracted from the DEM. Works of course only if the GCPs cover reasonably well the whole area under consideration. 

What software are you using to create the polynomial and adjust the dem?

I use Mathematica to pull out the numerical elevation data grid from the GeoTIFF file. Then within Mathematica it is very easy to fit the 2d polynomial and subtract it from the numerical values. I guess this would also be very easy in Matlab.  

The more overlap that is present in the source imagery the better chance we will have to calculate an accurate camera calibration to work with. We optimize for XY accuracy first (for mapping) and then fit the Z (for the 3D models).

The quality of the calibration does have a bit to do with the types of features that are available for us to work with as well. Lots of small evenly distributed features will give us better features to work with than a flat uniform surface.

I wanted to circle back on this and see if there has been any progress on your end.  I have upgraded to a Phantom 4 Pro in hopes of getting better data.  But I still get the hill on this site.  Here is a recent version with the new drone.  https://www.mapsmadeeasy.com/maps/public/4ffa8c0725c446fab0c7543e3d098ca1 . Notice that the elevation data is wrong, and that there is a big hump in the middle of the property.

The P4P with 80% overlap won't do this. It looks like you are using less overlap than that. 

I'm fairly certain that this https://www.mapsmadeeasy.com/maps/public/a32a3ef032e7484390a80303449e64a4 was done with the Phantom 4 pro with 80% overlap and I get a similar result.  It seems like there is some sort of calibration error that is always there.  Also, any advice on the bad elevation data?

We derive a custom camera calibration for every job we run. This calibration depends on a lot of things to happen properly. Lots of overlap is one of them and the job you just linked to does seem to have plenty. We have to optimize for something and generally we try to make sure the photo layer is spatially accurate. There can be artifacts in the elevation layer due to this optimization. That being said, I am a bit surprised that the linked job has this curvature. We have a lot of people doing large areas like that and they come out nice and flat with the Phantom 4 Pro camera. If you can, try things with a different P4P aircraft/camera to see if it comes out any better. We have seen cameras that just don't do as well as others in the same model line... This is why we do a custom calibration every time but it is possible that some camera's just won't calibrate as well as others.

Mheimann's answer above described a way to manually correct the data.