CadNSDI data accuracy

[jrbouldin]

I don't know if anyone here works with these data, but I've had questions and concerns about their spatial accuracy for a while now, and before it, the GCDB data. I find these BLM spatial data sets to be very poorly documented, a situation which, in terms of using them for analyses, makes me uncomfortable and/or aggravated. I don't typically find this situation the case with other federal agency data (USGS, NOAA etc).

A major drawback of original GLO plat maps is that they are not spatially geo-referenced--users who need that must do it themselves. The question is how. The only data I've ever been aware of that can be used for this purpose, are the BLM GCDB (former) and CadNSDI (current) data. As far as I can tell, the latter is derived heavily, maybe even entirely, from the former--just packaged in a different format at a different site. The CadNSDI are the cadastral component of the National Spatial Data Infrastructure and are downloaded for all the twps of a given state, en masse. The data contain a simple point layer as well as several different resolutions of parcel data (townships, sections, subsections).

As far as I can tell there is no meaningful and available data description, catalog or dictionary for the CadNSDI data. By this I mean a document that fully and clearly explains exactly how the data were produced. In science we would call that a data description paper but it could also be called a white paper or technical spec document or similar term--the sort of thing you would find describing satellite imagery data production, for example. Highly detailed and thorough.

Not finding such, I went back to the old California GCDB data manual I had from 20 years ago, even though I knew there was a lot of holes and/or general weirdness in that document; it didn't help much. All I can effectively glean is that two algorithms were used at different times, to model the point locations and their expected spatial errors for a given township--referred to as "GMM" and "PCCS". There is a very brief and nearly worthless discussion of the differences in intent and procedure of the two, nowhere near detailed and thorough enough to actually explain such, given that there are also no references to external publications describing and/or evaluating the two techniques. They have the look of being in-house BLM routines with god-only- knows what origin and statistical basis; a strong appearance of being ad-hoc and poorly evaluated tools.

When I georeference a given township by pinning an image raster thereof to the four twp corner points given in the CadNSDI data, I need some definite and trustworthy estimate of the likely spatial error in each of those four points. Instead what I find in the data are single declared values, for which I have no idea how they were generated, in terms of what the controlling points were and the statistical modeling method used, which is the whole ball of wax. One doesn't even know what parameter the stated value refers to. The 95% confidence interval maybe? Or maybe 99%? Or, maybe one (or maybe two or three) standard deviations from the mean? Aside from any biases that might very well exist in any given twp--and which I have strong reason to believe do exist and are widespread across large blocks of townships--this lack of clarity on the uncertainty parameter weakens the data, and potentially greatly depending on one's purpose.

To try to get some kind of insight I started looking at easily available high res imagery--recent Google Earth imagery which I know is generally accurate to within a few meters, with ~1-5 meter resolution. I figured that of all the cadastral survey points in the state, the one that I could most trust would be accurately located in the data would be the Mount Diablo Initial Point. The National Geodetic Survey location for that point is given to six decimal degrees, which at ~38 degrees N lat equates to roughly 0.1 meter resolution in both the lat and long directions.

So, I imported the NGS coordinates for the MDIP into Google Earth to see how it compared to the value given by CadNSDI, as well as to the octagon tower on the summit of Mount Diablo--the MDIP is in the center of that tower.

And this is what I get: https://drive.google.com/file/d/1y7r0Et ... sp=sharing

The yellow pushpin on the right is the NGS-given location of the MDIP. The center of the octagon tower is just W of it, at ~ 2.5 m, so the imagery in this region is off by about that much. The yellow push pin in the parking lot near the car, is the given CadNSDI SW corner location for T01NR01E MDM (point 100100). It is about 25 meters west and 3-4 meters S of the center of the octagon, for about a 26 meter error overall. This far exceeds the error inherent to the imagery.

So...what exactly am I supposed to make of this? What sort of confidence is anyone supposed to place in the CadNSDI data generally, if it cannot even give an accurate location for the Mount Diablo IP? How far off is it going to be for any random point in say Modoc, or Inyo, county? Nor do I have any reason to believe things would be better on the Humboldt or San Bernardino meridians. And there is the matter of the many large, multi-twp areas in which many to all CadNSDI section lines do not align with visible roads, ag field boundaries, apparent fence lines, vegetation and other on-ground boundaries, being systematically displaced from them over large areas.

Note that this is not really a Syndicate-twp-specific question because the errors in their twps range from enormous to non-intelligible/fabricated dwarfing the error in the CadNSDI data. The issue is with non-fabricated twp maps having smaller amounts of error.

Any thoughts on this, specific or general, would be very much appreciated--thanks.
Jim Bouldin

[Jim Frame]

I thought I might be able to help, as I knew a couple of guys who worked for BLM in Sacramento putting coordinates on township plats. I thought one of them might still be with us, but when I looked for his contact information just now I learned that he died last spring. So I got nothin'. :(

[jrbouldin]

Thanks Jim, I appreciate the consideration. The only Sacramento BLM person I corresponded with two decades ago regarding GCDB data was named Bill Twenhofel (sp?) I believe. But that was mainly just about how to obtain digital copies of the data, not basics or details of their origin/creation.

I also wanted to add that these errors are not due to re-projection mistakes from one projection to another (the native CadNSDI data format is in a California Albers projection on NAD83 whereas Google Earth uses lat/long and WGS84). I checked my reprojection of the data against GE's and the values were identical.