Drear All
Hoping you are doing great,
I’m conducting spatial analysis on Expressway in Malaysia, however, I don’t know how to determine the length of the hotspots because the clustering of hot spots using Get G* method with fixed bandwidth 1000 m, sometimes hotspots come in one isolated location from the others on the network as on the attached image.
So, you may someone can explain to me how I can measure and determine the length of hotspots?
Would you please help me in this regard?
--
Thank you
Best Regards
Fathi Salam Mohammed Alkhatni
you could write a small python script for this, to calculate the distance between neighboring hotspots.
It is giving you the totally correct answer for that algorithm - basically, that there are no 'hot spots'. Because you want, or expect there to be 'hot spots' does not mean they actually exist.
First, you read and understood what the algorithm actually expects and how it works: https://pro.arcgis.com/en/pro-app/latest/tool-reference/spatial-statistics/h-how-hot-spot-analysis-getis-ord-gi-spatial-stati.htm and https://pro.arcgis.com/en/pro-app/latest/tool-reference/spatial-statistics/modeling-spatial-relationships.htm ? Especially "Best practice guidelines"?
I can't see the actual properties of your input data, or the detailed processing parameters, or attributes and size of your data table ( your could always zip the shapefile files and post them on here ), so I'll make some assumptions.
You may be solving an essentially one dimensional problem with a two dimensional tool. The first clue would be if your source data was in a linear reference system, like milepost markers. Almost all the spatial statistics are expecting some distribution of points across both x and y. Here, there is no 'y' just an 'x'. And further, on an expressway with lane separation, there might be two 'spaces', one for northbound, and one for southbound. ( there is an option of using this tool on a network, but I'm not sure that relieves the assumption ).
The other expectation is an implicit assumption that the data points are samples on some sort of continuous surface. 'Population' as density, might be regarded as continuous on a national, county, or even city scale, but become just random points below that. Then there is the sample interval, if that is too small, the samples are 'peaks' awash in an ocean of nulls or zeroes, and the algorithm never incorporates the values of neighboring points. Or if all the points are concentrated along a linear feature. Basically, these 'heat maps' are a visualization of clustering of some sort.
It is possible to cluster along one axis, but it has it's own difficulties.
Best I can do without details.
Dear Dr. Patrick
First, I’d like to express to you my pleasure in helping me in this matter. I have attached the file so you may check the pattern and data features.
As you can see from the hotspots map that some hotspots are isolated and some of them are in a group. I have read some articles that measured the length of the hotspots when they used Getis-Ord Gi* tool in case they want to evaluate or compare the spatial tools to each other using the so-called predictive accuracy index (PAI). So, I’m not sure how did they measure and calculate the distances of hot spots along the roadway?
As in this study applied PAI: https://www.sciencedirect.com/science/article/abs/pii/S0001457519305901
In addition, I have difficulties with identifying the identical distance threshold of the spatial tools.
Calculate Distance Band from Neighbor Count tool used first to calculates the minimum, the maximum, and the average distance of a feature to the specified number of the nearest neighbor at least one neighbor. However, the condition of using at least 8 neighbors when the Input Field (i.e. sever inx = Severity index weight for each crash) is skewed which I'm not sure about that? what do you think should I use 1 or 8 neighbors?
When I use Incremental Spatial Autocorrelation Tool to identify distance bend for a crash data point, If I checked and unchecked Row Standardization gives me different results which one should I consider use?
Best Regards
See 'Selecting a conceptualization of spatial relationships: Best practices' at https://pro.arcgis.com/en/pro-app/latest/tool-reference/spatial-statistics/modeling-spatial-relationships.htm - "... For some applications, spatial interaction is best modeled in terms of travel time or travel distance. If you are modeling accessibility to urban services, for example, or looking for urban crime hot spots, modeling spatial relationships in terms of a network is a good option. ..." and https://pro.arcgis.com/en/pro-app/latest/tool-reference/spatial-statistics/updatedgeneratenetworkswm.htm
If you see https://solutions.arcgis.com/defense/help/incident-analysis/workflows/incidents-by-loc.htm ( this is a dead end, no longer maintained ).
This seems to be subsumed under Summarize Incident Count (Crime Analysis and Safety) https://pro.arcgis.com/en/pro-app/latest/tool-reference/crime-analysis/summarize-incident-count.htm Basically, I edited the highway line into a single continuous line, found it's length, then created points along it approx every kilometer, then used those points to split the line into approx 1 km segments which were used to gather the incident counts using 'Summarize ...'. This interval is somewhat arbitrary, a Goldilocks problem - basically by inspection of the histogram.
I don't speak the language, so I wasn't able to divide the accidents into two categories of the opposing directions of travel.
There are many ways to approach this depending on what you are using the results for - just visualization, or additional computation.
There are also ways to get an metrics using raster neighborhood operations. This is vehicles involved.
More important, did it actually answer your question on some ways to go about it, LoL?
- Badges
- Science topic
- Similar topics
- Instrumentation Engineering
- Instruments