Hi everyone,

Like many universities, mine has a degree in data science. And, like many universities, it is very math heavy - numerous courses in calculus, linear algebra, linear programming, etc. The fact is, few people can get through it, and there is some concern that it is so math and computer science heavy that it doesn’t translate to practical use once the students have a degree.

Many of you have seen my Udemy courses on Manifold, SQL, geostatistics, open source, and big data analytics. So, I’ve done pieces of it. Now I’m thinking of offering a more applied course next Spring, something like:

Applied Spatial Data Science

I would probably make Manifold 9 a central feature of the class, and try to make the course attractive to our students in Geography, Biology, the Business School, and even Health Science.

I’ve got some ideas of what I’d cover from a curriculum standpoint, but thought I’d throw the idea out to my Manifold friends. So my question is:

what would you include in a course on spatial data science using Manifold 9?

Remember, this isn’t a GIS class, per se, it would really be a data science course that is GIS focused. So, I’d appreciate any ideas you might have.

Happy to give you some thoughts Art, but could you post a current course breakdown for the spatial data science class. That would help me see the bigger picture. My little world is primarily local government stuff: parcels, roads, addressing, land use, economics (I pull IMPLAN data into Manifold to manipulate it), hazard mapping, historic data, wildlife, lidar, aerials.

Hi Art

For what it is worth, I find this an interesting spatial issue that took me much longer than it should have to work out the reasons for the rainfall variance in the below chart.

CliFlo is the web system that provides access to New Zealand's National Climate Database. This data is taken from the Ranfurly EWS Weather Station (Agent 18593; Lat:-45.12427 Long: 170.10045) from that system.

Data range from 22 Nov 2000 through to 26 April 2021.

As it turns out, the reason for the change was the farmers of the district changed from a dyke flooding irrigation system to pivot irrigation.

Probably a situation that does require local knowledge.

Attachments:
Ranfurly.png

If I wasn't the old man that I am, I would love to have an Applied Spatial Data Science course. I still use 8 for a workhorse and efficiency perspective and beginning to utilize 9 for specific purposes. I know it would behoove me to crash into 9 and take the deep everlasting plunge; but haven't.

What I wish there existed is a manual of best practices in M8 and M9. Yes, yes, yes, rtfm! I have been using using Manifold since 6.5 and for what I know, I get great work out of it but I've also watched this forum since way before the irksome days when I would ask here when 9 would be out and would get my hand slapped. This forum has been a great asset and reading the manual is a great endeavour and I have hundreds of pages of specific instructions from the manual printed out. The manual is detailed and long. What I wish for was more along a conceptual basis, an higher altitude look at concepts to come to understand best for projects and needs, i.e., if a person wanted to use Manifold for a geologic mapping project, or oil and gas landmap, or coal mine planning. I would it would list concepts and tools to use to set up these minimal and maximal ideas of what can be accomplished; it would say start with a quality PLSS landgrid, use your local projection, bring in digital geographic roads and such, topos, lidar, remote sensing, relational databases for 1 to n and n to 1, how to have geologic tops and bottoms, dynamic x sections and contouring when surface and/or subsurface info is augmented. I know how to do most of these, and think I'm hot stuff. But, when I start delving deeper via this forum and actually spending a few days of trial and error in Manifold trying new things, I realize there is so much more to know. Seems there should be a manual that was more conceptual that shows a person, based on the final product wanted, how best to get there from a product and efficiency point of view. So there, not exactly on point but at least tangentially related in learning how to work with spatial data science in a more disciplined and logical way. And Art, thank you and many others I've read over the years, for the wonderful contributions over the last two decades. The manual really is a fine manual, but sometimes it is too granular and detailed IMHO for trying to plan how best to do a project.

Hi Art. Off the cuff:

1. A refresher in basic trigonometry, using examples from TINs and DEMs. Make sure every entry student can do basic surface math. Lengths (including paths) and areas, with a pen or pencil and a calculator. (I would suspect many students can't get this far, and will need to be re-taught.)

2. Snyder. Read the whole thing as a requirement. Pick one projection from each category (conic, azimuthal, orthographic, compromise, etc.). Say why it is your choice, and say something about the limitations of the math behind it. Intelligent questions for extra credit. (What would you like to ask JPS now? Who would he be working for if he were still alive and why.)

3. Python. (IronPython as an option.)

4. Spatial SQL, using Manifold 9 as the model. Parallelise everything, with extra credit for successful experimentation producing unexpected results.

5. A beginner course in topology for advanced students, taking an historical approach, via Gauss, Poincaré. Riemann. Offer credit for external courses in everything from history of science, to integral calculus, to number theory and group theory, to hyperbolic geometry. Including stuff you don't understand (mainly to get that contribution on board).

6. I don't know how essential training in AI modelling is. I fear it may be utterly essential

Amateur thoughts.

Tim

2) yes indeed, and it's now recently been put online with examples: https://lists.osgeo.org/pipermail/proj/2024-March/011341.html

https://github.com/mdsumner

Hi Art, your course sounds exciting! As a retired data scientist, I remember that the spatial aspect of data scientist in my projects did two things:

1. Looking at the data spatially sometimes led to insights that allowed the problem to be solved
2. After a solution was made, the spatial picture helped others to see the solution better than looking at tables, summaries, etc.

Another study is the famous Titanic data science project of showing which passengers were more likely to survive based on their income level, sex, location on the ship (tourist class vs. staterooms), etc. You can use Manifold like you did on your recent post on showing how you organized your GIS schedules.

Got to go now, I will be thinking of other ideas later hopefully. Hope this helps.

Those are good examples. We should touch base offline. I’d like to hear about your certification.

One homework assignment I found interesting was to take Dr John Snow's 1854 cholera map and create a GIS version.

Once they have the original map, create 4 or 5 images seeking to convince an observer of the validity of the conclusion that the Broad Street pump was contaminated. Suggest 3-d, heat map, and other visuals. As a follow up allow the class to select an image and enhance it.

That’s a fun idea. Especially good as many students may not have GIS experience (i.e. business school, health science), so it kills 2 birds with one stone.

Another practical problem I had to solve was to determine how many cattle should be raised on a ranch to qualify for a special appraisal for animal production. Before I looked into the issue the county appraisal district used a standard of 1 Animal Unit per 20 acres. That standard seemed to be used in all the surrounding counties, but many ranchers complained they could only approach the standard during a rainy year. Thus they were having their special appraisals cancelled during our frequent and long running droughts. What I did to help was to import the USDA Soil Survey data into Manifold and slice that layer based on the parcels while retaining ownership information in the parcel layer and soil productivity in the soils layer. There were no eigenvectors or Big Data involved, but there was some logic and arithmetic used to determine how many animal units the land should produce given the varying quality of the soils on any given ranch. The USDA data had productivity expected for rainy, doughty, and average years, so we could adjust on the fly for ranchers to keep their special appraisals without protesting every year.