Data infrastructures and spatial models for biodiversity assessment and analysis: applications to vertebrate communities.

In conservation biology the computation of biodiversity maps, based on statistical models is a central concern. These maps, produced with objective and repeatable methods are an essential tool for conservation and monitoring programs as well as for landuse planning. Since the computation of biodiversity maps requires complex and time consuming procedures for data processing and analysis, it is necessary to design methods for homogeneous, scalable and repeatable data management and analysis. Moreover, the huge volume of data used in ecological modelling requires suitable software architectures to store, analyze, retrieve and distribute information in order to support research and management actions in due time. First of all we developed an analysis system (SOS - Species Open Spreader) providing statistical and mathematical models to predict species distribution in relation to a set of predictive environmental and geographical variables The system is composed of a module for data input/output toward and from the GIS and of a package of scripts for the application of different modelling techniques. At present, three statistical techniques are integrated in SOS: Logistic Regression Analysis (LRA), Environmental Niche Factor Analysis (ENFA) and flexible Discriminant Analysis with method BRUTO. Furthermore, two empirical spatial methods of analysis are available within SOS: Habitat Suitability Index (HSI) and Spatial Overlay. The system is designed to work with the GIS (Geographical Information System) soft-ware GRASS and the statistical environment R, coupled together through the SPGRASS6 library. Three different outputs are expected: text and graphical outputs with statistical results and suitability maps. Second, we tested the use of spatial Database Management Systems (Spatial DBMS) to handle wildlife and socio-economic data and we developed a web database application to provide facilities for database access. The information system was built for the Meru district (Tanzania) in the context of an Italian cooperation project of land use planning in Maasai rural areas. We tested two di_erent solutions: SpatiaLite and PostgreSQL-PostGIS; they both offer advanced technical facilities and spatial extensions to analyze spatial data. SpatiaLite is a new solution and offers the main advantages to consist of a unique file and to present a user-friendly interface, which make it the best solution for many applications. in spite of this we used PostgreSQL-PostGIS since it represents a well-established information system supported by libraries for web applications development. We applied SOS to three case studies at different spatial scale: Brescia plain (small scale), Mount Meru region - Tanzania (medium scale) and Lombardy region (big scale) in order to produce maps of species potential distribution and biodiversity maps for planning and management. We applied logistic regression analyses to compute models and ROC analysis for classification performance evaluation. The automation of processes through SOS gave us the possibility to build models for a large number of vertebrate species. The analysis produced very reliable results at middle and big scale while regression methods did not converge at small scale. This is probably due to habitat homogeneity and to the use of environmental variables with an insufficient level of detail. The potential distribution and biodiversity maps produced also had in all cases an applicative use in fact we used mammal species models computed for Mt. Meru region to produce a map of biodiversity within the area: this map represents an informative base for land use planning at village level within a cooperation project for Maasai economic development and environmental redemption. Amphibians and reptiles models, computed for Lombardy, represent a good informative base for planning management actions in the region.