Pseudo-absence Data Research Articles

A novel presence-only validation technique for improved Steller sea lion Eumetopias jubatus critical habitat descriptions

We used published information about foraging behaviour, terrestrial resting sites, bathymetry and seasonal ocean climate to develop hypotheses relating life-history traits and physical variables to the at-sea habitat of a wide-ranging marine predator, the Steller sea lion Eumetopias jubatus. We used these hypotheses to develop a series of habitat models predicting the probability of sea lions occurring within a 3 × 3 km 2 grid in the Gulf of Alaska and the Bering Sea. We compared these deductive model predictions with opportunistic at-sea observations of sea lions (presence-only data) using (1) a likelihood approach in a small area where effort was assumed to be uniformly distributed and (2) an adjusted skewness (Skadj) test that evaluated the distribution of the predicted values associated with true presence observations. We found that the Skadj statistic was comparable to the likelihood test when using pseudo-absence data, but it was more powerful for assessing the relative performance of the different predictive spatial models across the entire study area. The habi- tat maps we produced for adult female sea lions using the deductive modelling approach captured a higher proportion of presence observations than the current habitat model (critical habitat) used by fisheries managers since 1993 to manage Steller sea lions. Such improved predictions of habitat are necessary to effectively design, implement and evaluate fishery mitigation measures. The deductive approach we propose is suitable for modelling the habitat use of other age and sex classes, and for integrating these age/sex-class-specific models into a revised definition of critical habitat for Steller sea lions. The skewness test provides a means of comparing the relative performance of such models, using presence-only data. The approach can be readily applied to other central-place foragers.

GIS-based modeling of Javan Hawk-Eagle distribution using logistic and autologistic regression models

Few attempts have been made to model the distribution of the Javan Hawk-Eagle (Spizaetus bartelsi) based on predictions formulated from habitat requirements in West Java or throughout Java Island. This paper proposes a new approach to predicting probability models of Javan Hawk-Eagle (JHE) distribution using the application of logistic regression (LR) and autologistic regression (ALR) coupled with RAMAS GIS, and creating pseudo-absence data using a normalized difference vegetation index from remote sensing data. Habitat requirements of 11 nest-sites in Gunung Gede-Pangrango National Park (TNGP) and its surrounding areas were analyzed and quantified, and the model was validated in southern parts of West Java. The final LR model was used as the starting point for fitting ALR models that account for spatial autocorrelation through the addition of an autocovariate variable using several different neighborhood sizes ranging from 450m (15×15 moving window size, or equal to 20.25ha) to 1500m (50×50 moving window size, or equal to 225ha) using 300m interval. The best model was the ALR model with a 1500m autocovariate that agreed with the distance between nests in TNGP and the mean home range size in Java. This model showed a significant increase in overall accuracy and successfully removed misclassified pixels. Based on our results, we recommend five strategies for the management and conservation of JHE habitat, including integrated conservation management and increased regulation of wildlife reserves.

More complex distribution models or more representative data?

Distribution models for species are increasingly used to summarize species’ geography in conservation analyses. These models use increasingly sophisticated modeling techniques, but often lack detailed examination of the quality of the biological occurrence data on which they are based. I analyze the results of the best comparative study of the performance of different modeling techniques, which used pseudo-absence data selected at random. I provide an example of variation in model accuracy depending on the type of absence information used, showing that good model predictions depend most critically on better biological data.

Assessing the effects of pseudo-absences on predictive distribution model performance

Modelling species distributions with presence data from atlases, museum collections and databases is challenging. In this paper, we compare seven procedures to generate pseudo-absence data, which in turn are used to generate GLM-logistic regressed models when reliable absence data are not available. We use pseudo-absences selected randomly or by means of presence-only methods (ENFA and MDE) to model the distribution of a threatened endemic Iberian moth species ( Graellsia isabelae). The results show that the pseudo-absence selection method greatly influences the percentage of explained variability, the scores of the accuracy measures and, most importantly, the degree of constraint in the distribution estimated. As we extract pseudo-absences from environmental regions further from the optimum established by presence data, the models generated obtain better accuracy scores, and over-prediction increases. When variables other than environmental ones influence the distribution of the species (i.e., non-equilibrium state) and precise information on absences is non-existent, the random selection of pseudo-absences or their selection from environmental localities similar to those of species presence data generates the most constrained predictive distribution maps, because pseudo-absences can be located within environmentally suitable areas. This study shows that if we do not have reliable absence data, the method of pseudo-absence selection strongly conditions the obtained model, generating different model predictions in the gradient between potential and realized distributions.