No abstract

We explored the effects of land-cover configuration, body size and trophic diversity in determining avian species richness on Prince Edward Island, Canada. Data on avian species richness were obtained from the Maritime Breeding Bird Atlas data. Prince Edward Island was divided into 97 sampling cells of 10 × 10 km. Land-cover metrics were calculated using a forest inventory database, Fragstats and ArcView version 8.1. The relationships between avian species richness and explanatory variables were explored using correlation analysis, mixed forward-backward stepwise analysis, generalized linear models and Akaike's information criterion. Models predicted between 27% and 63% of the variability in species richness, attributing substantial explanatory power to both the average body size and the range of body size spanned by the avian community. The body-size frequency distribution showed that avian communities were dominated by species weighing between 50 and 80 g. Habitat metrics associated with forests were more important to the avifauna than those related to agriculture. Avian species richness also decreased with both the fragmentation and isolation of wetlands. The total area covered by the human infrastructure land-cover and its subdivision were also important. Clearly, body size plays a key role in determining the diversity of birds on Prince Edward Island. In particular, species weighing 50-80 g appear to have sufficient resources to be successful on Prince Edward Island's landscapes. Our findings also highlighted the importance of controlling the expansion of human infrastructure and both the fragmentation and reduction in size of wetlands to maintain avian species richness patterns.

El presente artículo tiene como objetivo elegir, bajo un determinado escenario, el mejor algoritmo supervisado de machine learning para localizar un terminal que soporte wifi. Se usa un dataset que cuenta con 2000 registros de Received Signal Strength Indicator (RSSI), obtenidos de 7 puntos de acceso (AP), los cuales se cargan en 8 algoritmos supervisados de machine learning. Luego se elige el algoritmo que realiza la predicción más precisa, incluso cuando se cuenta con un menor número de AP. La mayor precisión se logra con el algoritmo naive Bayes, tanto para el caso de 7 AP (99 % de precisión) como para cuando se cuenta con un número menor de AP. Asimismo, se observa que los algoritmos basados en redes neuronales presentan el peor rendimiento. Finalmente, se proponen trabajos futuros para continuar con la investigación sobre el tema de localización de dispositivos wifi en interiores.