Somodi et al. (2012) Biological Conservation

Imelda Somodi, Andraž Čarni, Daniela Ribeiro, Tomaž Podobnikar (2012): Recognition of the invasive species Robinia pseudacacia from combined remote sensing and GIS sources. Biological Conservation, 150 (1) 59–67. DOI 10.1016/j.biocon.2012.02.014


Abstract
Monitoring the spread of invasive species is crucial for nature conservation; however regularity can only be assured if cost-effectiveness can be achieved. We aimed at testing low-cost remote sensing sources and simple methodology for recognising the invasive species Robinia pseudacacia and thus founding a monitoring scheme. A study area with mixed wooded stands containing R. pseudacacia has been selected for this purpose in NE Slovenia. Four different sources (Landsat ETM and airborne orthophotos from summer and spring) were tested together with a filtering for forested areas. Filtering was based either on Landsat information or on a forest polygon layer as alternatives. Generalised linear models were constructed in a training window within the study area to establish a statistical rule of recognition for the species based on spectral information. Models were tested both within and outside the training window for accuracy. As means of accuracy assessment both the well-established AUC and the specially adapted Jaccard index have been applied.
The best and most reliable recognition was achieved by using the spring orthophoto, in which the species was captured in flower, combined with a GIS filtering by a forest vector layer. The superiority of this combination was especially striking when tested over the full study area. The Jaccard index appeared to be more sensitive in discrimination between models. Thus we conclude that even spectrally less detailed data sources may provide a basis for successful monitoring if the phenology of the target species is also considered.


Keywords
AUC, Cost-effectiveness, GLM, Landsat, Orthophoto, Phenology

Aszalós et al. (2012) European Journal of Forest Research

Réka Aszalós, Imelda Somodi, Kata Kenderes, János Ruff, Bálint Czúcz, Tibor Standovár (2012): Accurate prediction of ice disturbance in European deciduous forests with generalized linear models: a comparison of field-based and airborne-based approaches. DOI 10.1007/s10342-012-0641-6


Abstract
We analyzed an ice disturbance event of deciduous forests in Hungary by Generalized Linear Models (GLM). Two statistical models were generated: the first model was based on a disturbance map created from a series of aerial photographs, and the second model was based on a map created by half-year-long intensive field work. The second map was considered as the reference map of ice disturbance. Our hypothesis was that the predictive power of the field-based statistical model would be significantly higher than that of the aerial photo-based model on the reference map. Elevation, slope, aspect, mixture ratio of beech, height of the dominant tree species and their interactions were used in the two (aerial photo- and field-based) GLMs as explanatory variables. The accuracy of the models was measured by the AUC (Area under the ROC curve) values. Sensitive area maps of ice disturbance were generated by both models. Our hypothesis was definitely rejected. Both models performed high predictive accuracy (median AUC > 0.9) with no significant difference in the prediction capacity regarding the reference ice disturbance pattern. Our study demonstrates that ice damage can effectively be predicted if remote sensing interpretation is coupled with GLM as predictive model.

 

Keywords
Forest damage, GLM, Susceptibility assessment, Probability map, Variable interactions