Bias in avian sampling effort due to human preferences: An analysis with catalonian birds (1900 - 2002)

Aims: To investigate the sources of spatial bias in the bird records of Catalonia from 1900 to 2002, with the aim of providing generalized recommendations for using other databases, and setting up broad inventory projects. The paper examines the influence of environmental variables, human distribution and ornithological preferences of birdwatchers on past avian sampling effort in Catalonia, a contrasting region in the north east of Spain. Location: Catalonia (Spain). Methods: The relationship between time (field days) devoted to sampling birds in 10 x 10 km UTM squares (from the records of VertebraCat database, across 5 study periods from 1900 to 2002) and a set of environmental, human distribution and bird species richness variables was analysed. These relationships were analysed by means of Partial Least Squares Regression (PLSR) and multiple regression analyses. Results: A partial least squares regression analysis accounted for 39.4 % of the spatial variation in time devoted to avian sampling per UTM 10 x 10 km square in the most recent (1983 - 2002) period. The major pattern shows that large visit frequencies were mainly associated with coastal areas with a large cover of wetlands, marshes and sand dunes, high human population density, dense transportation, and also high covers of irrigated croplands, urban and industrial environments. Another group of highly visited UTM squares was also covered with a large surface of protected areas, mostly located in mountainous Catalonian sectors, as opposed to those UTM squares in lowland areas mainly covered by non-irrigated extensive croplands. Catalonian ornithologists and birdwatchers also showed an uneven preference for different groups of bird species, as they mainly relied on migratory and endangered species. Conclusions: The illustrated biased pattern of ornithological field work in Catalonia casts doubts about the usefulness of biodiversity indices obtainable from databases of observation records without a randomstratified sampling approach. However, these problems might be overcome by including a variable of sampling bias such as the number of records or days of field work.