Scaling the walls of history: The perils and pitfalls of multy-scale land cover comparison

The problem addressed is this: most environmental issues require context to solve them. Is the ocean getting warmer? Is the desert growing? Is the forest declining? Solution: measure the temperature / size / leaf area. But such measurements only have significance if there are other comparable historical measurements to compare them too. This paper is about that word comparable. Can we really compare landscape generalisations gathered at different times and at different spatial scales? Today we have the ability to produce land cover maps at a very high spatial resolution (grid cell sizes of 10, 5 or even 1 m). Historically, data has typically been collected at coarser spatial scales (grid cell sizes of 50, 100 or even 1000 m). To facilitate comparison, modern data is often re-scaled to match the historical data. To evaluate the validity of this process, a series of synthetic landscapes were created. These landscapes included
the full range of possible dispersion from a random spatial distribution of scene elements to a highly clustered spatial pattern. Each simulated landscape was firstly classified and then degraded to four levels of generalisation (simulating a range of spatial resolutions). In parallel, the process was reversed and the simulated landscapes were degraded and then classified. The resultant classifications were then compared. In all cases the integrity of the data was best preserved when the image was more highly spatially autocorrelated. Changing the spatial scale (i.e. degrading) of classifications resulted in a rapid decline in information content, particularly in more random landscapes. The implications of these results are then discussed.