Chase terrain suitability study map

Jeff Duda

site owner, PhD
Staff member
Supporter
I spent a few weeks trying to accomplish this with the help of a few classmates with no success. I had the idea of using GIS to produce a map (particularly using ESRI's ArcMap software) that shows the quality of terrain in certain areas of the US for storm chasing. There are definitely some qualities about the topography in certain areas of the country that make it better or worse for chasing, and most people agree on them. The biggest ones I think of are:
-flatness of the terrain
-openness (land use)
-road network
-severe weather/tornado climatology

It was my goal to find a way to "quantify" the qualities listed above and map them to show which areas of the country are better or worse for chasing. Obviously some areas should stick out, but the true ideas behind this are 1) evaluate the suitability for chasing on a very detailed scale and 2) identify "sleeper" areas - i.e., little known areas that may remain unknown to many chasers that are actually great terrain for chasing.

I have found resources for the data needed to create such a map. They include the National Elevation Dataset, a 1 arcsecond data set containing elevation for all practical areas of the US (see http://seamless.usgs.gov/), a land cover data set with the same resolution as that of the NED (see also this site), and road shapefiles across the country (available in many locations, including www.grlevelxstuff.com and http://www.meteor.iastate.edu/~slincoln/GRS/shapefiles/).

My plan would be to set up a grid and for each point on that grid to assign the variable (for the sake of the topic, called "storm chasing index" - SCI), using the formula:
SCI = A*B*C*D, where
A represents the quality of the topography (i.e., terrain flatness)
B represents the land use
C represents the quality of the road network, and
D represents severe weather climatology,
and each of A through D are essentially normalized to range from 0 to 1, where one would define either 0 as best or worst quality for that element. The resulting value of SCI would be between 0 and 1 and that would be what is plotted.

Regarding A: Things such as the average elevation, standard deviation of elevation, and average slope across a bin of data are good mathematical ways to evaluate the flatness of the land (low standard deviation and slope would be better). Things such as the first or second partial derivative of elevation with respect to direction could also be beneficial, as it would indicate the quantity of terrain flipping, or "hilliness".

Regarding B: Given the different land use categories (e.g., 1 for water, 2 for urban, 3 for grassland), one could assign a normalized value to each category which represents the quality of chasing on that land use value (can't chase on water and chasing in urban areas would be very hard, and chasing in grasslands/prairies, or temperate forest would be easier).

Regarding C: this one would probably be the hardest to actually do, but in each grid box, one could compute the weighted sum of the length of road (the weights could be assigned based on road type: gravel road vs. two lane paved road vs. four-lane free access highway vs. four lane controlled access highway), the number of road intersections, and perhaps count the number of consecutive miles of straight road segments.

Regarding D: One cool use of ArcMap is the "density" function in the spatial analyst tool extension of the software. Simply download the tornado tracks, touchdown points, lift points, etc data from SPC's SVRGIS site and run the tool at whatever resolution you want to create a map of tornado climatology (or other severe event climatology if you wish) that can be applied to the grid.

The reason I and my partners have been unable to complete this task is because the machines we are trying to run this on are simply not capable of handling the large amount of data required to do this. The NED data comes in 1 deg lat by 1 deg lon pieces, and even those small chunks are very large (on the order of 100 MB). Thus the tools simply fail due to lack of memory. But I'm sure ArcMap is not the only GIS software out there that can do this, nor is this the only method to quantify storm chasing terrain. I'm now turning to the Stormtrack community to see if anyone knows a better way to do this or has the technology to do it as I've tried.
 
Why not do it by county? It wouldn't be as detailed as you might like, but it would be much easier to calculate. Essentially assign a letter to each county in the country and then color code them on a map.

Actually...then after you do that, someone could build a GR placefile that color codes the bad counties :)
 
UC Berkely has developed an application that runs as a screensaver on volunteer's computers to perform chunks of very large computations. The volunteers get to choose what they run including weather data, SETI radio signals, and polling data.

From what I understand it is open to Masters and Doctorate level research projects, some commercial entities, and to further scientific knowledge when intensive data crunching is required.

Here is the link: http://boinc.berkeley.edu/
 
I've thought about making chase maps myself. A quantitative approach seems like the best way to get an accurate and complete map. One alternative method I thought of would be to use google maps with something like a 10x10 mile grid and have users rate the tiles they are familiar with by clicking on the tile and assigning a score in each category. People's opinions would vary greatly, but averaging the results and maybe throwing out the outliers could yield a decent looking map. It might also be a good way to get your road quality variable without having to jump through some major data crunching hoops. With enough users you'd wind up with a fairly complete map of the Great Plains and Midwest too.
 
I tried the quantitative approach a few years ago but gave up due to a lack of computing power. It was going to be high enough resolution that it could be used for navigational decisions while actively chasing.

Something less adventurous would still be useful (and more practical). It would be nice to combine "bad chase terrain" with "bad roads" in a sort of openstreetmap-py way (I know that idea has been floated before...apparently it didn't go very far?)
 
This is definitely a huge project, but I think it is worth the effort. I am going with a much simpler method using topos, aerials, and mainly Google StreetView to create a point shapefile/placefile for AL, MS, TN, GA, and possibly AR and LA. My chase area is much smaller in scope than you are dealing with, and the number of viewing locations is small as well, so I think I can get away with using this method. I just have to continually go back and increase the density of points as I have time.

I am in GIS so over the next few days I will try to look further into what you are trying to do. My wife will be delivering our child later this month, so as you can imagine my time is limited.
 
I think it would be very difficult to come up with a machine that could handle that level of data, but not impossible. I've actually thought of this on a couple of occassions myself, but just haven't had the time with classes to really look into it more deeply. I think doing it by county would be an excellent method because (A) the amount of data wouldn't be quite as brutal, and (B) warnings are typically issued with reference to counties, and thus are quite important regarding positioning. A few more basic ideas could be obtained in a relatively straightforward, albeit time consuming manner through the use of topo maps, other usgs data, google earth, maps, and streetview, etc.

The math is where I'm out, though. I have no experience in anything above Calc I, so things like partial derivatives will have to be up to people that are much smarter than me. :(
 
Jeff, what kind of things were you trying in ArcMap? Viewsheds along the roadways would be useful to show what one could see from the road, that could narrow down the area you need to crunch numbers in for land use/cover and tornado data.
 
How far did you get into this? Did you manage to compute A, B, C, & D?

It should be doable at say 1km or 10km scale across the CONUS on reasonable desktop hardware. Break the large rasters into smaller tiles if necessary so that they can fit into memory.

It would probably be much easier to do most of this work in a programming langauge instead of inside ArcGIS. Python, IDL or matlab should make the job easy to do.
 
I ran into problems when trying to merge the 1 deg by 1 deg NED data sets together. Those alone are too big to be merged on the machine I was trying to do this. The county idea, and the general idea of upscaling it is where I think I'd have the most luck. My initial intentions were to get things on a larger grid than is available on the NED data (like Zac mentioned - 5 to 10 or more km on a square). However, I'm no GIS or ArcMap expert - more of a novice really. I took one course on it last spring and despite it being a graduate level course, I was rather disappointed in how little about the software or GIS in general I learned. I don't have the time to take another class in it either. The only tool I know of to use is the resampling tool, but that just applies the value of the NED data (like elevation) to the nearest grid point of the grid I'm resampling to. I would like to be able to essentially average the values of the smaller grid data into the larger grid so that I could compute the terms as I had planned. I also agree with Zac in that it may be easier to do the computations in a different language. However, I am rather inexperienced with Python, IDL, and MATLAB, but very experienced with FORTRAN. One of my partners is much more experienced in MATLAB, so perhaps he can help with that.

If anyone understands exactly what I'm trying to do and can produce a sample of it, I'd love to see if it can practically be done. Otherwise, I'll be trying it myself over the next few weeks/months. Thanks for the suggestions so far, everyone. Please keep them coming. I think a very useful product can come out of this.
 
GIS is so cool. Creating the layers will be a pain the a** but if done well it will be worth it. I've got an intro to ArcGIS book here at home from an intro to GIS course I took last semester if you think that will help.
 
Back
Top