Rule-based predictive models are not cost-effective alternatives to bird monitoring on farmland

Abstract

Monitoring bird populations is labour-intensive and requires expert knowledge. Development of more cost-effective methods of bird species monitoring is therefore desirable. Simple rule-based (RB) models were developed that predicted the presence of birds in lowland arable farmland in the breeding season. These were based on the presence of key habitat elements required by the birds identified from the literature. Habitat data and bird data were collected independently from 525 transects each of which ran along a field margin. Model performance was tested in five species: grey partridge Perdix perdix, skylark Alauda arvensis, linnet Carduelis cannabina, corn bunting Miliaria calandra and yellowhammer Emberiza citrinella. RB models were not good predictors of species presence. Model performance did not vary appreciably between early or late breeding season or according to geographic region. Skylark had the best performing model, but even in this case the predictive power was low. Prediction of species presence from habitat survey data is therefore unlikely to be a cost-effective method of monitoring birds on British farmland.