Comparison of Visual and Digital Aerial Survey Results of Avian Abundance for Round 3, Norfolk Region


Title: Comparison of Visual and Digital Aerial Survey Results of Avian Abundance for Round 3, Norfolk Region
Publication Date:
November 01, 2009
Document Number: STANDBIRD-09
Pages: 1-40

Document Access

Attachment: Access File
(1 MB)


Burt, L.; Rexstad, E.; Buckland, S. (2009). Comparison of Visual and Digital Aerial Survey Results of Avian Abundance for Round 3, Norfolk Region. Report by University of St Andrews. pp 1-40.

This report describes the analysis of two aerial surveys in the Norfolk Round 3 Zone 5 potential offshore wind farm area in the spring of 2009. One survey employed on-board observers and subdivided the study region into 8 blocks or strata. These strata were surveyed over 5 days of surveying during the period 5 March-20 April. A second survey employed digital video technology to collect animal detections. This survey was completed over 3 days.


Abundance estimates were computed using conventional distance sampling estimators as well as density surface models for the visual surveys. Strip transect methods as well as density surface models were used to compute abundance estimates for the digital video surveys. Rather than producing estimates for each species encountered during the surveys, species were grouped into ‘divers’, ‘gulls’, and ‘seabird’ categories. Hence six estimates for each strata were produced for the visual survey data, and five abundance estimates were produced for the digital video surveys (too few divers were detected to produce estimates using density surface modelling).


It was not feasible to estimate numbers of each species group in the entire region, because the surveys were conducted over a protracted survey period, during which bird numbers and distributions can be expected to vary substantially. We therefore used smaller areas corresponding to the strata (blocks) defined in the design of the visual surveys as the basic unit of comparison. As a result, the digital survey effort was broken into smaller blocks than in the original design of this survey. Likewise, the comparison of estimates produced by the two survey methods could only be on a single stratum where surveys by the two teams were conducted at roughly the same time. The correspondence of abundance estimates for three species groups in this stratum was reasonable for gulls, but confidence intervals for the two surveys do not overlap for the ‘seabird’ group.


We measured precision of the visual and digital surveys in two ways. First we selected a stratum (NS1) where the data for the two types of surveys were most comparable. But to approximate the precision for the entire study region, we needed to sum estimates over all strata, even though the survey was carried out over a period >5 weeks. We then needed to further approximate the precision over the entire study region for the digital survey. This involved extrapolating the increase in precision observed for visual surveys from the stratum to the study region to the digital surveys. We urge caution in interpreting the measures of precision at the level of the entire survey region.


Precision of the survey method/analysis method/species group fell broadly into three categories. Conventional distance sampling estimates from visual surveys were most precise with coefficients of variation (CVs) on the order of 0.10. Density surface model estimates of visual surveys of gulls and digital surveys of seabirds along with strip transect digital survey estimates of seabirds produced CVs on the order of 0.30, and all other survey/analysis/species combinations had CVs in the range of 0.45-0.66. Survey/analysis methods that have CVs of 0.30 require an annual change (either positive or negative) of around 11% to ensure that the power to detect this trend within 10 years of annual surveys is to be 0.8. These are quite large absolute changes to the study populations, making those survey/analysis methods quite poor at detecting change in avian populations.


Because of the lack of coordination during the conduct of the surveys, this post-hoc comparison is uninformative from the perspective of comparing survey methods. Movements over the protracted survey period, both within the study region and in or out of it, render dubious any exercise in summing estimates over sub-regions, and in comparing sub-region estimates corresponding to dates that differ appreciably. However, we do have some confidence in the estimates of precision associated with each survey method. Based on the data after filtering and matching from these surveys in the Norfolk Round 3 Zone 5, sufficient power is assured only for avian data derived from visual surveys.


We suspect that both data acquisition methods and survey design can be improved to enhance the precision of abundance estimates using digital techniques. Encounter rate variability (differences in the number of animals detected between transects) drives much of the uncertainty associated with these surveys, and careful attention to survey design can reduce this source of variation that results in these high measures of uncertainty, and consequent poor power to detect change.

Find Tethys on InstagramFind Tethys on FacebookFind Tethys on Twitter
This question is for testing whether or not you are a human visitor and to prevent automated spam submissions.