Calculating Metrics for Spatial Data Quality

-

 Spatial Data Quality...

What is spatial accuracy? Why do we care?



When you build maps, or work with any kind of job in which you are presenting data, you want to be accurate and precise. Accuracy is essentially the inverse of error, or lack of error in a dataset. In regards to GIS, we are primarily focused on spatial accuracy. This is the positional accuracy of your point data. 


If you have an xyz coordinate, such as a crime was committed at this LAT and LON, or address, we want that location to be correct when reporting it. You don't want to present to your boss at a marketing firm that Sally bought your product at a certain store, but in reality it was another store located 5 miles away because your coordinate data or spatial accuracy was full of errors! Think about SWAT teams responding to an active shooter and they end up at the wrong school. Or even something more simple such as trying to drive to a certain location and your GPS takes you to the wrong place. 


There are other kinds of accuracy...

Temporal accuracy: Time related 


Thematic accuracy: This is usually looked at as metrics or statistics such as mean, median , mode, max, min, etc...


You need to be accurate when portraying data to your boss, your company, and to stakeholders being affected by your maps and products. 

Below is an example of calculating the accuracy and precision of an average point to a real located reference point using a GPS device. The real point is true data and is highly accurate. We are calculating how many points taken with different projections and comparing their distances to our average point and the real point


Average Point Longitude: 82.3188985°W (-82.31)
Average Point Latitude: 27.8258524°N (27.83)
Average Elevation In Meters: 28.5 Meters
Horizontal Precision (68%) in Meters: 5.3 Meters
This included approximately 34 data points within the 68th percentile buffer or the 5.3 meter buffer.
Vertical Precision (68%) in Meters: 29.45 or 29.5 Meters
This Included approximately 35 data points for the 68th percentile or vertical height of 29.5 meters



         The horizontal distance at a 68th percentile that I calculated from the average point was 5.3 meters. I drew the measurement from the real point to my average point and the distance was 3.37 meters, which is a significant difference of 1.9 meters. The average elevation or altitude for the real point was 22.58 meters, and the average of my points were 28.54 meters. That is a difference of approximately 6 meters. 

        The real point being collected using GPS equipment is going to be more accurate and precise in regard to its xyz location and elevation data. The GPS device has a stronger cellular connection using satellites to pinpoint a more accurate location. 


1.      The metrics are the following…

a.      Horizontal Accuracy:

                                                  i.      Mean: 4.6

                                                ii.      Max: 20.2

                                              iii.      Min: 0.0001

                                              iv.      50th Percentile / Median: 2.47

                                                v.      68th Percentile: 5.3

                                              vi.      95th Percentile = 13.3

b.      Horizontal Precision (Reference Point Real Data):

                                                  i.      Mean: 5.4

                                                ii.      Max: 20.9

                                              iii.      Min: 0.84

                                              iv.      Median / 50th: 3

                                                v.      68th: 6.01

                                              vi.      95th: 11.98

Results: There is slight variation such as the mean or average is lower for my data versus the real point from the reference point. My maximum value is less, and my minimum value is less. This also shows slight variations in my percentiles in the data distribution. 

The real point has 34 waypoints within it’s 68th percentile buffer of 6.01 meters. My accuracy average point contained 34 points in its 68th percentile buffer of 5.3 meters. So despite having differences in variation, both points included the same number of waypoints within the 68th percentile buffer.

c.       Vertical Accuracy:

                                                  i.      Max: 47.2

                                                ii.      Min: 11.4

                                              iii.      Mean: 28.5

                                              iv.      50th Percentile / Median: 27

                                                v.      68th Percentile: 29.45

                                              vi.      95th Percentile: 46.5

d.      Vertical Precision (Real Point Data Reference):

                                                  i.      Max: 47.2

                                                ii.      Min: 11.4

                                              iii.      Mean: 28.5

                                              iv.      50th Percentile / Median: 27

                                                v.      68th Percentile: 29.45

                                              vi.      95th Percentile: 46.5

                                            vii.      My vertical accuracy compared to the precision of the true data point was the same data.

I can say that my accuracy of my average point was fairly close differing the average distances of over 5 meters to the real point or reference point. My precision was strong seeing that 34 points fell within the 68th percentile buffer which is a strong grouping and well over the majority of the waypoints. So despite my accuracy being a bit off my precision was high. 

BIAS



Yes there is bias in my analysis because my horizontal accuracy and precision have some variation in distances to the waypoints which is to be expected. The average can vary based on the coordinate system, the user, or what locator is being used to derive the xyz point data. 


Check out this awesome video on spatial accuracy from using a smartphone to do GIS field mapping. This is a great overview of why spatial accuracy and precision matters. 



Don't Forget To Subscribe!


Comments

Post a Comment

Popular posts from this blog

TIN's vs. DEM's: 3D Elevation Models In Arc GIS Pro

-
Image
 Elevation Models For GIS Today we are going to run through some analysis of comparing Digital Elevation Models (DEM's) to Triangular Irregular Networks (TIN's) using Arc GIS Pro.   One of the main differences between TIN's and DEM's, is that a DEM file is raster based, whereas a TIN elevation model is vector based. In GIS, we can use TIN's and DEM's to analyze elevation of local areas or global even to conduct slope analysis, aspect ratio, terrain mapping, and solve complex problems.  Using a DEM to Develop a Ski Run Suitability Map In the example below, we are identifying where an engineering firm would place a ski run for a resort in an area based on elevation, slope, and aspect parameters. We used the weighted overlay tool to combine 3 reclassified rasters of elevation at a high altitude, slope at a high pitch, and aspect for a SW positioning with given percentages of weight.   25% aspect  40% elevation  35% slope This was our result: As you can see, we c...
Read more

Isarithmic Mapping: Annual Rainfall in Washington

-
Image
  Isarithmic Mapping Today we are going to talk about Isarithmic Mapping! This type of mapping is a way to explain smooth or continuous phenomenon such as a heat map, weather, temperature or rainfall.  There are two types of these kind of maps, Isopleth and Isometric . .. Isopleth Maps : These are made from perceptual poi nt data such as the population of an area or the rate of crime in an area. These maps should be normalized if using raw data and should come from averages, densities, and ways to standardize the data. Such as the number of Covid cases per every 10,000 people, rather than coloring polygons of counties based on the number of covid cases.  Isometric Maps : These come from true point data, such as crime locations as points at a certain address or X and Y coordinate The two rainfall maps of Washington below are Isopleth maps as they are using perceptual data of areas regarding their average rainfall in the state of Washington. This data is already standardi...
Read more

Ward 7 Public Schools in Washington D.C.

-
Image
  Hello everyone! Today I decided to experiment with a completely new color scheme for my mapping lab assignment.   I went with purple!   I am not sure why, but it just seemed right at the time... The map above is unique as today we are going to focus on our main talking points staying within the realm of Cartography. In particular, we are going to talk about Visual Hierarchy, and the map making process following Gestalt's Principles. As a GIS analyst, we must look at making maps and visuals as a graphic design process as well as a scientific process. This means that maps need to have a theme, a feel, and most importantly provide context to the map reader in a way that makes sense.  Before we dive into the map above, lets explain what visual hierarchy is and the Gestalt Principles! Gestalt's Theory is as humans, we see or perceive the whole of something before we actually make out the individual parts. When you see a landscape, you see the landscape first, then the ...
Read more