288
Colony Monitoring
The estimates of colony size and total chick numbers from the ground surveys and
preliminary visual estimates were also used to assess the accuracy of those estimated by photo-interpretation and for the total number of lying gulls (OA between
85 and 95%). Ground surveys and visual estimates were, in all cases, higher than the
automatic counts on images from UAVs.
Discussion
In this study, we were able to provide quick and accurate estimates of colony size
(lying birds) and productivity (number of chicks). One of the most challenging tasks
in estimating breeding pairs in a bird colony from the ground is the limited horizontal visibility (Sardà-Palomera et al. 2012). Zenithal views from UAVs definitely
contribute to improving colony size estimates. However, birds are often fully or
partially obscured by trees or shrubs, with causes difficulty in their detection and
counting through photointerpretation. This occurs frequently for Ardeids since nests
are often settled in the middle of reed beds or stands of other tall graminoids (Prosper
and Hafner 1996). This constraint has also been found relevant in ungulate surveys
with UAVs (Chrétien et al. 2016). Given these kind of difficulties, only a complete
capture of chicks can eventually provide accurate estimates of productivity, though
these disturbances might also cause chick abandonment (Oro and Tavecchia 2008).
No disturbances were observed for gulls during the UAV flights, suggesting that
our methodology was appropriate for colony monitoring, which agrees with the
results from previous studies (McEvoy et al. 2016; Ratcliffe et al. 2015; Vas et al.
2015). In addition, we employed only three persons to supervise a single whole mission, two of them for flight operations and one for a ground survey of the colony.
Two big challenges arise when working with photographs taken from UAVs . On
the one hand, images are not always acquired in the best conditions. Firstly, wind
gusts can yield blurred images making it difficult to delineate the target features
although it might be avoided by setting a high shutter speed on the camera. Moving
birds also appeared distorted. Secondly, the mosaicking process can randomly select
the image without the bird that was present in the overlapping discarded image, as
reported by Bakó et al. (2014). We experienced both situations in our study, which
increased the omission error.
Although visual photo-interpretation ensures high accuracy in bird delineation,
automatic classification takes less time and produces satisfactory results (AbdElrahman 2005; Grenzdörffer 2013). However, we had to use a size criterion to
discard small classified patches. Other approaches using pattern recognition such as
Object Based Image Analysis (OBIA) may be able to provide better results (see
chapter by Hurford).
R. Díaz-Delgado et al.
Colony Monitoring
The estimates of colony size and total chick numbers from the ground surveys and
preliminary visual estimates were also used to assess the accuracy of those estimated by photo-interpretation and for the total number of lying gulls (OA between
85 and 95%). Ground surveys and visual estimates were, in all cases, higher than the
automatic counts on images from UAVs.
Discussion
In this study, we were able to provide quick and accurate estimates of colony size
(lying birds) and productivity (number of chicks). One of the most challenging tasks
in estimating breeding pairs in a bird colony from the ground is the limited horizontal visibility (Sardà-Palomera et al. 2012). Zenithal views from UAVs definitely
contribute to improving colony size estimates. However, birds are often fully or
partially obscured by trees or shrubs, with causes difficulty in their detection and
counting through photointerpretation. This occurs frequently for Ardeids since nests
are often settled in the middle of reed beds or stands of other tall graminoids (Prosper
and Hafner 1996). This constraint has also been found relevant in ungulate surveys
with UAVs (Chrétien et al. 2016). Given these kind of difficulties, only a complete
capture of chicks can eventually provide accurate estimates of productivity, though
these disturbances might also cause chick abandonment (Oro and Tavecchia 2008).
No disturbances were observed for gulls during the UAV flights, suggesting that
our methodology was appropriate for colony monitoring, which agrees with the
results from previous studies (McEvoy et al. 2016; Ratcliffe et al. 2015; Vas et al.
2015). In addition, we employed only three persons to supervise a single whole mission, two of them for flight operations and one for a ground survey of the colony.
Two big challenges arise when working with photographs taken from UAVs . On
the one hand, images are not always acquired in the best conditions. Firstly, wind
gusts can yield blurred images making it difficult to delineate the target features
although it might be avoided by setting a high shutter speed on the camera. Moving
birds also appeared distorted. Secondly, the mosaicking process can randomly select
the image without the bird that was present in the overlapping discarded image, as
reported by Bakó et al. (2014). We experienced both situations in our study, which
increased the omission error.
Although visual photo-interpretation ensures high accuracy in bird delineation,
automatic classification takes less time and produces satisfactory results (AbdElrahman 2005; Grenzdörffer 2013). However, we had to use a size criterion to
discard small classified patches. Other approaches using pattern recognition such as
Object Based Image Analysis (OBIA) may be able to provide better results (see
chapter by Hurford).
R. Díaz-Delgado et al.