8  Abstraction, Accuracy, Error, and Precision

8.1 Abstraction

GIS organizes geographic information into layers, where each layer represents a particular type of geographic feature (such as roads, rivers, or hospitals). The layer approach necessarily relies on abstraction: the process of simplifying complex real-world features into representations that GIS software can store, visualize, and analyze. Representing every aspect of every real-world feature is impractical because of limitations in data collection, storage, processing, and analysis.

Abstraction always involves decisions, and those decisions will impact the outcome of any GIS analysis using the data. Abstraction is guided by the original purpose of the dataset. The purpose allows data creators to determine what aspects of the real-life feature are important (and therefore should be collected and represented with high accuracy) and which aspects are less important (and can be generalized or omitted). A mismatch between your purpose for using the data and the abstraction decisions made in the data creation process can be a serious issue for the reliability of your results.

Consider the following example of different uses for the same dataset- hospitals in North Carolina:

  • An engineer assessing flood risk at hospitals in North Carolina
  • A public health analyst investigating disease outbreaks across hospitals in North Carolina

What is considered important to these two users differ greatly. The engineer may require detailed building footprints to evaluate structural exposure. The public health analyst may want detailed information on the number of beds and the infectious disease protocols, but just need a rough location for the hospital. If these users were creating the datasets themselves, they would likely look very different. However, most of the time, users aren’t collecting or creating the data themselves, they’re working with data that already exists. As a result, it’s crucial for users to critically evaluate the dataset’s underlying abstraction decisions regarding:

  • Selection - What features are included or excluded?
  • Classification - How are features grouped?
  • Simplification - How much detail of the original feature is retained?
  • Resolution- What is the smallest feature or level of detail that the dataset can meaningfully represent?

8.2 Accuracy and Precision

We also need to examine how the processes of measurement and representation introduce accuracy issues, errors, and varying levels of precision into our datasets.

Accuracy is the degree to which a measured or recorded value matches the true value. Precision refers to the level of detail or consistency with which measurements are recorded. A dataset can be:

  • Accurate but not precise- the measurements are close to the “true” value, but recorded at coarse increments
  • Precise but not accurate- the measurements are highly detailed but consistently off from the true value.

For example, let’s consider temperature data collected at the RDU ASOS station. In this dataset, accuracy would be defined by how close the value measured at the station is to the actual temperature outside. Precision would be, essentially, how many decimal places there are. For example, recording temperature to the nearest whole degree is less precise than recording it to the nearest tenth of a degree.

8.3 Data Quality and Error

Accuracy and precision then inform data quality– which is the relative accuracy and precision of a particular dataset. Error refers to the combined effects of inaccuracies and imprecision

Error in a geographic dataset can take several different forms:

  • Positional error- inaccuracies or imprecision in location. Can be caused by:
    • Measurement limitations- GPS inaccuracy, sensor resolution limits, survey instrument errors
    • Map scale- smaller scales generalize more, reducing positional detail
    • Projection and transformation – converting data between coordinate systems can introduce distortions or small positional shifts if inappropriate projections or transformations are used.
  • Attribute error- inaccuracies or imprecision in the non-spatial data associated with features. Can be caused by:
    • Data entry or transcription mistakes
    • Outdated data- attributes no longer represent current conditions
    • Sensor/device errors- faulty readings from measurement device
    • Instrument limitations – the precision or measurement range of the instrument is inherently limited.
  • Conceptual error- Mismatches between the data’s abstraction and the actual characteristics of what it represents. Can be caused by:
    • Inappropriate classification- categories too broad or narrow
    • Overgeneralization- simplifying complex features beyond what’s appropriate for analysis
    • Temporal mismatch- datasets from different dates are combined
    • Modeling assumptions—for example, representing a floodplain as having fixed boundaries even though it changes over time.

8.3.1 When do issues become errors:

Some problems are always errors– for example, a malfunctioning GPS unit or broken temperature sensor will always produce inaccurate data regardless of how the data is used.

Other potential issues only become errors in certain contexts. For instance:

  • Imprecise data is not inherently “wrong”, but if your analysis requires high precision, that imprecision becomes an error
  • Low resolution is not automatically an error, it just limits the detail of the data. However, if your research question demands final spatial detail, low resolution would be a source of error.

These are examples of context-dependent error, which are situations where the data’s characteristics (precision, resolution, scale) are mismatched with the requirements of the analysis and, therefore, become error.

8.3.2 Error vs. Uncertainty

Uncertainty describes the estimated amount of error that might be present in a dataset. It allows users to judge whether the data is suitable for a specific purpose. When potential error can be quantified, it can often be accounted for in analysis. The bigger problem arises when error cannot be measured (or when we don’t even know it exists), because it cannot be corrected or factored into decision-making.

Example: If a GPS is accurate within ±5 m, that’s uncertainty you can plan for. If the GPS occasionally drifts 30 m without warning, that’s unknown error.

8.4 Case Studies: Why Does Accuracy and Precision Matter

8.4.2 Landslide Risk

You are tasked with assessing landslide risk for a mountainous county: Your available data includes:

  • A digital elevation model
  • A soil type map
  • Rainfall records from a local station network

Consider the following questions:

  • Which parts of the dataset could lead to positional error, attribute error, or conceptual error?
  • Imagine a research question where the dataset limits here might not be a problem. What would that question look like?
  • Imagine a research question where the same dataset limits might cause a problem. What would that question look like?