I’ve always liked the saying “the perfect is the enemy of the good.” I was reminded of this recently when asked for best practices for combining lidar data sets. I immediately recommended that they make sure the data sets were in the same datum, including having used the same geoid model, and that they needed to start from the point cloud. When I stopped my train of thought and listened long enough to understand their goals, I realized I was telling them to do a lot of work that may not lead to increased value. It was going to be closer to perfect, but it wasn’t more useful than the easier “good enough” approach.
Missing the Forest for the Trees
So, what were they trying to do? In this particular case, they wanted to derive tree canopy heights from multiple lidar data sets, some of which were overlapping. So, they would need to make a ground surface and a top of canopy surface. The field office staff were the most likely to have to do the work. That amounted to a lot of different people distributed across the country with only a few common software options. My approach would certainly have worked, but while they probably could have found a way to get all the point clouds into the same projection and the free VDatum program could handle getting the geoid models consistent, was there really a reason to go to that effort? It would increase the consistency of the combined point cloud and allow the creation of a better DEM, right? Well, yes, however, all of that is essentially at the 10 cm level, while figuring out the tops of the trees is probably closer to the meter level uncertainty or more.
Focus on What Matters Most
It may seem obvious, but sometimes you have to do a bit of sensitivity analysis to figure out what’s worth paying attention to. That didn’t take much in the case above, but our assumptions on where to put our money to improve a situation aren’t always right. For example, we used to think we’d get the best bang for our buck in flood mapping by investing in modeling and engineering improvements. That’s what the NRC study team looking at improving flood map accuracy thought they’d find when they started. Instead, they found that improving the input topography was the most important thing to do.
How Different Are Geoid Models Anyway?
I started thinking about how much change there is between geoid models when I saw some recent lidar data released that used GEOID96 to convert to NAVD88 orthometric heights instead of the current GEOID12a. Was that good enough? That depends on what you want to use it for, but first I just wanted to get an idea of how different the models were. I grabbed the grids out of VDatum and converted them to Imagine files using the gdal_translate program that comes with the open source GDAL package. That made it easy to pull into a GIS package and do the grid subtraction. You can see the spatial distribution of the change (new minus old) in the first image.
I decided I only cared about the absolute value, not the direction of change, so I calculated that and redid the colors for the second image. Turns out that new data was on Long Island, which you can see is a less than 5 cm change, or less then the random error of the data. For most purposes, that’s probably good enough. However, you can also see lots of areas of the U.S. that are much larger changes and even a small systematic offset can make a big difference if you’re calculating volumes, as Keil Schmid pointed out in his post. It all depends on the use. For my forestry colleagues, it still wouldn’t have been worth worrying about since any NGS geoid would have been good enough, and that’s all they need. On the other hand, if you’re doing something that has to satisfy everyone’s real or imagined use, you may find that only near perfection is good enough. In that case, you have my sympathy. In all cases, it’s important to first determine your true needs and then figure out how you’re going to get there within budget.