Mapping Land Change

The landscapes around us are ever-changing:  forests may be cut for residential development or they may regrow in abandoned agricultural fields; vegetated ground may be paved for new roads or parking lots; and shorelines may change slowly with sea level rise or abruptly after a hurricane.  Researchers at PIE and all the LTER sites who study such changes on decadal scales need to be able to characterize, track, and accurately quantify them.  To do this, they often use time series of maps – historical to present day - but this approach presents a number of challenging problems.  The much-less-detailed historical maps are difficult to compare to current high-resolution imagery.  With high resolution (pixel-scale) imagery comes a very large amount of data that can become computationally overwhelming.  Correctly identifying important changes happening in small patches may be difficult to discern within a much larger, more stable landscape.  Some landscape changes remain difficult for the technology to see, even though satellite and aerial imagery and resulting maps have improved dramatically.  Changes abound, but how do we recognize, quantify, and finally, predict them?

PIE scientist Gil Pontius and his students at Clark University, MA, develop mathematical techniques to address these challenges, building clear, precise methodologies to apply to a variety of ecological research questions, and to share with colleagues and planners.   In a recent paper (cite and/or link), they describe techniques to reduce the data overload that arises from pixel-scaled imagery by creating two measures, “Incidents” and “States”.  “Incidents” is the number of times a land-cover category assigned to a pixel changes across the time intervals being studied.  In the paper’s example, applied to the Georgia Coastal Ecosystems LTER (GCE), results showed how many times a pixel changed among four categories: Evergreen, Cut, Wet, or Other.   “States” is the number of categories that a pixel represented across the time intervals; if the pixel was categorized as only Evergreen or Cut during the entire interval, then “States” would be 2.  Four combinations of Intervals and States are useful to describe trajectories of change:

  1. Persistence: a pixel remained the same category during all time intervals (Incidents =0, State = 1)
  2. One Incident: a pixel changed during exactly one time interval (Incidents = 1, State = 1)
  3. Toggle: a pixel changed back and forth between two categories during more than one time interval (Incidents > 1, State = 2)
  4. Multiple States:  a pixel experienced more than 2 categories (Incidents >1, States >2)

Results from this approach summarize change at the pixel and landscape levels.  They can be visualized with maps to identify and communicate where and how change is occurring over time, even when the change is small relative to the spatial extent.

At PIE, Pontius and his colleagues are working to apply techniques like these to questions of marsh response to sea-level rise, many of which are challenging to address for the entire marsh without the use of maps and other imagery.  For example, a comparison of historical maps to recent imagery and ground surveys suggests that the number and size of ponds on the high marsh has changed over time.  We also see changes in the marsh shoreline, both on its seaward edge and along creekbanks.  Goals of the current work are to discern, categorize, and quantify these changes in the entire marsh by using current and historical images.  Resulting trajectories can then be used to look for links to sea level rise or other drivers, to address implications for the status of the marsh, and to project the likely future state of the marsh.


Pontius, R.G.,, Jr, R .Krithivasan, L. Sauls, Y. Yan & Y.  Zhang (2017) Methods to summarize change among land categories across time intervals, Journal of Land Use Science, 12:4, 218-230, DOI: 10.1080/1747423X.2017.1338768