Environmental Sustainability in Practice

Errors in Spatial Data Analyses

Information from geospatial analyses is increasingly being used to inform decision making about environmental sustainability issues, such as planning and ecosystem monitoring. Accordingly, the thematic information derived from the results of geospatial analyses must be accurate. Unfortunately, there is always inherent error in information derived from spatial analyses. That is to say, nothing is perfect. To that end, individuals conducting geospatial analyses need to recognize the sources of error, minimize this error as much as possible, and convey to the end user of the information how much confidence they should have in using the derived information (Jenzen, 2016).

Assessing the accuracy of geospatial analyses may take one of two forms: qualitative assessment or quantitative assessment. Qualitative assessment may include visual examination of the result and making judgements based on personal knowledge of a study area. In contrast, quantitative assessment typically involves the generation of statistics and indices to interpret and determine the accuracy of results and outputs.

As previously discussed, image classification is a popular image analysis technique in the field of remote sensing. Given that the results of image classification involve the production of land-use/land-cover maps that may be used in environmental decision making, performing an accuracy assessment on image classification results is of paramount importance.  

The most common method of determining the accuracy of an image classification result is selecting a sample of pixels from a classified image and comparing these sample pixels with known pixel classes from reference data (e.g., these reference data would normally be collected during a field study).

If 85% or more of the sample pixels represent the known classes from the reference data (e.g., a sample image pixel was classified as "water" and based on the reference data, that pixel is indeed water), it can be said that the map is accurate. Ideally, we want to be as accurate as possible and the closer to 100% accuracy, the better.

Simply put, the map user wants to know that if he/she were to take the map out into the field to where the map says they should find "water", what is the likelihood that there will actually be water at this location?  

This page has paths: