Abstract
This thesis describes a two-year case study monitoring the densities of four species of common farmland birds on sheep and beef farms on New Zealand's South Island. The main objectives of the study were twofold: i) to establish baseline population estimates and investigate their seasonal dynamics and habitat associations in order to add to the understanding of ecology and ecosystem roles of these species; ii) to estimate the performance of line transect distance sampling for bird monitoring on farmlands, and evaluate the suitability of this approach in a the establishment of a national New Zealand bird monitoring programme for common and widespread species. Lastly I present a simulation-based power analysis of an experiment using the bird estimation methods intended to test the relative influences of habitat quality and predation by introduced mammals on farmland bird populations. The study formed part of the ARGOS project (Agriculture Research Group on Sustainability) which examines the environmental, social and economic sustainability of New Zealand’s farming systems. This transdisciplinary investigation compares environmental impacts of organic, Integrated Management, and conventional farming.
Population densities of Skylark Alauda arvensis, Common Blackbird Turdus merula, Song Thrush Turdus philomelos, and Australian Magpie Gymnorhina tibicen were monitored on twelve pastoral farms located between the Banks Peninsula (Canterbury) and Owaka (Southland). Each farm was visited nine to ten times between November 2005 and August 2007. Birds were counted on ten 500m unbounded line transects per visit and farm while recording detection distances and several field parameters, and densities per farm and species were estimated in the distance sampling modelling program Distance™ 6.0. Multiple covariate modelling was used to incorporate the influence of the field parameters on detection probability (Chapter One), and the specific effect of individual covariates on the detection function was examined (Chapter Two). Model averaging was performed with the aid of Akaike scores. The effects of habitat parameters on species detectability and estimated density were modelled using Hierarchical General Linear Models.
The average estimated detection probability (0.53) and average precision of detectability and density estimates (coefficients of variation of 0.12 and 0.21 respectively) compared well with international results. Farm-level woody vegetation cover emerged as the main driver of detectability for all four species, and few seasonal or geographical effects were found. Covariates of detectability were found to play a major role in improving model fit (especially time since sunrise and wind speed), but there were few clear directional trends within species. The effects of observer's varying monitoring experience and broad habitat type were the most consistently observed covariates of detectability. This indicates a need to include additional factor interactions when estimating bird density. In general, the distance sampling approach was successful in producing unbiased density estimates. An assessment of linear correlations between raw index counts and distance sampling estimates demonstrated that raw count data would be insufficiently accurate for magpies. However the less involved indexing method could produce reliable relative estimates for skylarks and thrushes, and also for blackbirds provided that vegetation cover was factored into the estimates. For these species, such uncorrected count methods could be used for the comparison of populations between farms using organic, Integrated Management or conventional farming systems, as I found no detectability differences between farm management types. Nor was there any indication of differences in bird densities between farms these management systems (Chapter Three). This outcome contrasts to findings of several international studies.
Percentage of woody vegetation was found to have the strongest effect on bird numbers, being positively correlated with thrush and blackbird and negatively with skylark densities; the percentage of introduced species within woody vegetation negatively affected blackbird numbers. Thrushes and blackbirds also showed strong seasonal population dynamics, part of which could be traced to seasonal changes in availability for detection. In contrast, the abundance of magpies did not vary seasonally or with percentage of woody vegetation.
Average densities for skylarks, blackbirds, thrushes and magpies were 0.53, 0.41, 0.23 and 0.18 birds per hectare respectively. These numbers were generally lower than those derived from other ARGOS surveys carried out on the same farms using a different modelling approach (a global detection model with post-stratification instead of farm-specific models). The discrepancies can be traced to specific differences in methodology and farm coverage. My survey method achieved higher precision than the ARGOS approach, but is probably less suited to the monitoring of uncommon species and the efficient tracking of long-term trends. The abundances of skylarks, blackbirds and thrushes in my study were much higher than recorded in the United Kingdom.
While the effect of habitat composition as examined in my study is regarded as the main determinant of bird biodiversity on farms, predation by small introduced mammalian predators has also been shown to have a strong influence on bird populations. I tested the statistical power of a proposed experiment to disentangle the relative influences of habitat quality and predation on the density of breeding birds (Chapter Four). Bird densities on a group of matched farm pairs with "simple" versus "complex" habitat structures would be monitored while imposing a predator press (sustained predator control) on half the farms for a number of breeding seasons in a Before-After-Control-Impact (BACI) design. I developed a simplified computer model of bird population dynamics that simulated differential recruitment on a pair of control and treatment farms, incorporating several types of stochastic variations to approximate the expected accuracy of estimating bird densities in the field. The simulation predicted that an effect of predator control would be detected at the 5% signficance level in 75% of the cases provided that four farm pairs were monitored and bird population estimate precision was below 40%. The simulation demonstrated that improved precision of bird density estimation has a large effect on experimental power to learn how farm bird communities could best be restored. This experiment could be carried out during the duration of a standard PhD thesis project or as an addendum to a larger ongoing monitoring effort.
Implementation of this kind of experiment and longitudinal monitoring of farmland bird abundance would be one of many benefits for New Zealand. Comprehensive nationwide monitoring schemes exist in several countries, but no such programme yet exists in New Zealand. The results of my case study are combined with the findings of two workshops held in 2004 and 2005 that explored options for the establishment of a national monitoring scheme in Chapter Five. The programme would likely be volunteer-based and should cover as much of New Zealand's land area as possible and include all terrestrial habitats and species. A random stratified sampling strategy employing line or point transects with distance band suitable for analysis with distance sampling methods should be employed. I argue the advisability of using finer-grained distance bands to enable reasonable modelability of detections, and recommend the use of a double sampling approach to calibrate density estimates and the implementation of a pilot study to establish needed sampling effort and test field method feasibility. A programme built along these lines could provide valuable long-term population trends, early warning of species declines or pest eruptions, insights into ecosystem health, and will benefit conservation interests. Judging from the results of my case study, distance sampling would be the recommended method to use in this context.