From historical aerial maps to a biotope map: ecologial image processing in GRASS GIS

Biodiversity, of vital importance for the ecosystems productivity and capacity to provide services to the human populations inhabiting them, is also an indicator of the health of these ecosystems. Here we show how changes in biodiversity can be assessed by interpreting historical data and comparing the results to current conditions. The study area is a part of the Monte Baldo, a mountain close to the Garda lake in the Italian Alps, which is called the European botanical garden for its floristic richness. In order to assess biodiversity changes, we processed scanned historical black/white aerial photographs from 1954 in GRASS GIS to obtain a classified map coverage. We used aerial photographs since they provide the basis for developing indices of landscape composition and structure as sensitive measures of large-scale environmental change over relatively long periods of time. However, prior to any exploitation of the contained information, proper image rectification is needed to enable geometrically unbiased application of landscape metrics in order to obtain meaningful results. We used the robust and freely licensed toolchain for orthorectifying images as available in the Open Source Software GRASS GIS which was improved and updated in GRASS GIS 6.4 during the project. In order to take the steep terrain into account, the sun position for the overflight time of each aerial image was calculated using information from the scanned flight annotation panel and, along with a DEM, fed into the topographic correction method available as GRASS GIS Addon (i.topo.corr). In the first place, an illumination map was created from sun position and DEM, secondly the topographic correction performed using the "c-factor" method. The commonly used cosine correction, due to its mathematical nature, tends to overshoot in steep terrain as found in the Monte Baldo zone (above circa 35° slope the cosine correction saturates). Since the quality of the historical aerial photograms is poor, we used a sophisticated segmentation method (r.seg Addon) to smooth likely homogeneous areas. Since single channel classification is challenging, we generated synthetic channels through texture analysis (sum entropy method). The smoothed black/white aerial photographs and the derived texture maps were used as input for a supervised image classification. Training areas were obtainedby manual digitizing and the classification was eventually performed using Sequential Maximum a Posteriori (SMAP) estimation. As a result, a historical land use map from 1954 was obtained, referenced to modern base cartography and usable for the assessment of biodiversity changes.