Leaf Litter

When Bad Data Happens to Good Models

Biohabitats own GIS guru, Christine Mielnicki reminds us to keep four fundamental principles in mind when we apply GIS to our work.

Article Index

GIS tools allow us to organize natural resource data and analyze the spatial impacts of alternative management choices fairly rapidly. In the last ten years, advances in computer processing power and technology have moved spatial analysis from the research and academic arena to widespread applications. Within the same time frame, our understanding of the complexity of our natural resources has increased and a shift in resource management goals has occurred. The concept of natural resource management has migrated from a commodities and production approach to an ecosystem and sustainability model.

But what makes a sound GIS conservation model?

Many organizations are faced with limited time, resources and data when it comes to GIS conservation analysis. Most analysis consists of an inventory assessment based map. What happens when we want to go beyond the inventory assessment and answer the questions, What if? Why? Where?  Our first instinct is to search for data and in the end, take what we can get. Even when relative datasets are identified, data scale, incomplete data sets or — worse yet — no metadata or information about the attributes are often the norm. Does this mean we shouldn’t or can’t answer the questions at hand? Probably not, but managing data uncertainties becomes a priority. Just remember Horwood’s Short Laws of Data Processing and Information Systems.

First and foremost, conservation issues are not black and white. There are many shades of grey in between. Often, these shades of grey are defined by the question and objectives of the model. However, many times the model is not constructed to return any response other than black or white. Four common data mistakes that contribute to less robust conservation models are:

  • assigning discrete values instead of continuous values;
  • inappropriate data scale;
  • insufficient ground verification; and
  • outdated datasets.

Assigning discrete values to measure indices such as distances to forested patches or the size of riparian buffers is common practice. Although this simplifies the model, the results are clearly black or white: forested patches are either within 500 meters or they are not; riparian areas are protected by 100 foot buffers. Yet this is not always the case. Certain riparian areas maybe of more value than others or they may be influenced by adjacent land use practices. Will a 100 foot buffer be acceptable for all riparian areas? Probably not. In order to capture the grey zones of conservation concerns, a proximity function that degrades with distance or size may be a better approach to capture the grey zones.

Another common problem is inappropriate data scale. Analysis at the wrong scale can miss important patterns and features. This problem can occur when using land cover data for species habitat analysis. Models built on coarse data scales can lose small but highly valued areas such as pocket wetlands that support unique smaller species. Models built on fine data scales can lose sight of the big picture and not accurately represent areas used by big game as corridors. A solution may be to include land cover data at both scales.

Desktop analysis should never take the place of old fashion field work or ground truthing.  More often than not, we buy into the ideology that technology trumps best professional judgment. The results of a GIS model cannot be interpreted responsibly without a reality check. After all, the output from a model can only be as good as the weakest data put into the model.

Many times, data sets are a few years old and only represent one snapshot in time. Data is often collected by many volunteers over a season. It is important to understand what the attributes collected mean and what standards were followed, if any.  It is important to ask ourselves if these datasets represent current conditions before running any conservation model. Has much changed on the ground between the time the data was collected and today?  How much development has occurred?  Were climate conditions considered normal for the year the data was collected?  How would more current data affect the conservation model?  There are often many contributing factors to consider when determining if data accurately reflects current conditions.

Often, we lose sight of the big picture and get hung up in the details. GIS based conservation analysis can vastly improve the effectiveness with which we make conservation planning decisions.  However, it is important to ensure the model addresses the ecological integrity and biodiversity of the system accurately and the results are interpreted to ensure they represent real world conservation needs.

Got an idea?

Contact The Editor

Sign up for Leaf Litter

Browse by topic

Browse by year