Hi, I am creating time series datasets of an agricultural field to estimate crop height from the DEMs generated monthly.
To measure the monthly changes (growth) in the plant height, I subtract the DEM from the previous month from the DEM of the current month. In processing the datasets I came about some strange observations
In some instances the series indicates negative growth, i.e., the range of values in the later months’ DEMs are even lower than the range in the terrain model. While in truth the crops have been growing taller handsomely through the months.
In some other instances the difference the DEMs indicate herculean growth of several tens of meters, while in truth they grow by a few centimeters month on month.
We practically flew the drone in the same configuration on similar bright sunny days.
Has anyone here come across similar problems help me understand what am I doing wrong?
Thanks
Unfortunately I am not using ground control points. I suspected this might be the cause but couldn’t find any publications that attribute the wild swings in height estimations to non-use of gcps.
Do you know of any published work that makes this conclusion?
You’re looking at greater than 10m error in the vertical for a static end-user GPS unit. Now combine that with movement, and you’re also re-flying the site on different days at different times with different constellation geometries (therefore different DOPs)… Yeah, these are not in need of specific research papers, these are well-known fundamental properties of GPS units. If someone’s giving you trouble when you tell them you need to use GCPs, you can just refer them to the Wiki on GPS. The burden of proof should not lie with you for this.
By having long-baseline GCPs, you’re reducing your vertical error and constraining your reconstruction to known values. You likely might still get some negative values as reconstruction isn’t perfect nor always deterministic, but it should be far more reasonable.
You should also be wary of assuming that your GSD is actually a realistic minimum measurement for your equipment. You might need to test a bit to determine what that is for your collection procedures and instruments. So for example, 4.33cm/px GSD doesn’t mean your vertical resolution is anywhere near that. I believe Stephen suggests half GSD or quarter GSD to be more realistic.
