. Novel methods improve prediction of species' distributions from occurrence data. Á/ Ecography 29: 129 Á/151.Prediction of species' distributions is central to diverse applications in ecology, evolution and conservation science. There is increasing electronic access to vast sets of occurrence records in museums and herbaria, yet little effective guidance on how best to use this information in the context of numerous approaches for modelling distributions. To meet this need, we compared 16 modelling methods over 226 species from 6 regions of the world, creating the most comprehensive set of model comparisons to date. We used presence-only data to fit models, and independent presence-absence data to evaluate the predictions. Along with well-established modelling methods such as generalised additive models and GARP and BIOCLIM, we explored methods that either have been developed recently or have rarely been applied to modelling species' distributions. These include machine-learning methods and community models, both of which have features that may make them particularly well suited to noisy or sparse information, as is typical of species' occurrence data. Presence-only data were effective for modelling species' distributions for many species and regions. The novel methods consistently outperformed more established methods. The results of our analysis are promising for the use of data from museums and herbaria, especially as methods suited to the noise inherent in such data improve. J. Elith

Generalized dissimilarity modelling (GDM) is a statistical technique for analysing and predicting spatial patterns of turnover in community composition (beta diversity) across large regions. The approach is an extension of matrix regression, designed specifically to accommodate two types of nonlinearity commonly encountered in large‐scaled ecological data sets: (1) the curvilinear relationship between increasing ecological distance, and observed compositional dissimilarity, between sites; and (2) the variation in the rate of compositional turnover at different positions along environmental gradients. GDM can be further adapted to accommodate special types of biological and environmental data including, for example, information on phylogenetic relationships between species and information on barriers to dispersal between geographical locations. The approach can be applied to a wide range of assessment activities including visualization of spatial patterns in community composition, constrained environmental classification, distributional modelling of species or community types, survey gap analysis, conservation assessment, and climate‐change impact assessment.

Prioritizing areas for conservation requires the use of surrogates for assessing overall patterns of biodiversity. Effective surrogates will reflect general biogeographical patterns and the evolutionary processes that have given rise to these and their efficiency is likely to be influenced by several factors, including the spatial scale of species turnover and the overall congruence of the biogeographical history. We examine patterns of surrogacy for insects, snails, one family of plants and vertebrates from rainforests of northeast Queensland, an area characterized by high endemicity and an underlying history of climate-induced vicariance. Nearly all taxa provided some level of prediction of the conservation values for others. However, despite an overall correlation of the patterns of species richness and complementarity, the efficiency of surrogacy was highly asymmetric; snails and insects were strong predictors of conservation priorities for vertebrates, but not vice versa. These results confirm predictions that taxon surrogates can be effective in highly diverse tropical systems where there is a strong history of vicariant biogeography, but also indicate that correlated patterns for species richness and/or complementarity do not guarantee that one taxon will be efficient as a surrogate for another. In our case, the highly diverse and narrowly distributed invertebrates were more efficient as predictors than the less diverse and more broadly distributed vertebrates.

Summary 1. AusRivAS (Australian River Assessment Scheme) models were developed, using macroinvertebrates as indicators, to assess the ecological condition of rivers in Western Australia as part of an Australia‐wide program. The models were based on data from 188 minimally disturbed reference sites and are similar to RIVPACS models used in Britain. The major habitats in the rivers (macrophyte, channel) were sampled separately and macroinvertebrates collected were identified to family level. 2. Laboratory sorting of preserved macroinvertebrate samples recovered about 90% of families present when 150 animals were collected, whereas live picking in the field recovered only 76%. 3. Reference sites clustered into five groups on the basis of macroinvertebrate families present. Using seven physical variables, a discriminant function allocated 73% of sites to the correct classification group. A discriminant function based on seven physical and two chemical variables allocated 81% of sites to the correct group. However, when the same reference sites were re‐sampled the following year, the nine variable discriminant function misallocated more sites than the seven variable function, owing to annual fluctuations in water chemistry that were not accompanied by changes in fauna. 4. In preliminary testing, the wet season channel model correctly assessed 80% of reference sites as undisturbed in the year subsequent to model building (10% of sites were expected to rate as disturbed because the 10th percentile was used as the threshold for disturbance). Nine sites from an independent data set, all thought to be disturbed, were assessed as such by the model. Results from twenty test sites, chosen because they represented a wide range of ecological condition, were less clear‐cut. In its current state the model reliably distinguishes undisturbed and severely disturbed sites. Subtle impacts are either detected inconsistently or do not affect ecological condition.

Systematic data in the form of collections data are useful in biodiversity studies in many ways, most importantly because they serve as the only direct evidence of species distributions. However, collecting bias has been demonstrated for most areas of the world and has led some to propose methods that circumvent the need for collections data. New methods that model collections data in combination with abiotic data and predict potential total species distribution are examined using 25,111 records representing 5,123 species of plants and animals from Guyana; some methods use the reduced number of 320 species. These modeled species distributions are evaluated and potential high-priority biodiversity sites are selected based on the concept of irreplaceability, a measure of uniqueness. The major impediments to using collections data are the lack of data that are available in a useful format and the reluctance of most systematists to become involved in biodiversity and conservation research.