Countering low visibility in video survey of an estuarine fish assemblage

Abstract

Sampling fish communities in tropical estuaries is inherently challenging due to poor visibility and the potential presence of dangerous fauna. We present two strategies for improving the identification of fishes in a turbid tropical estuary using video. The first was to attract species close to the camera by using two different bait types compared with no bait, and the second involved manipulating footage in the postfilming phase. No significant difference was found in the species richness recorded among camera bait treatments (thawed Australian sardines, canned sardines and unbaited), although baited cameras did detect 13 taxa not observed on the unbaited cameras. Three different image restoration algorithms (histogram equalisation, white balance and contrast-limited adaptive histogram equalisation) were compared in processing 22 instances where fish could not be confidently identified to species or genus level. Of these processed clips, five were able to be identified to species level by a panel of four coauthors. Further, two of the three algorithms yielded higher average confidence values for identification at the order, family, genus and species level than when the unprocessed footage was viewed. Image restoration algorithms can partly compensate for a reduction in image quality resulting from turbidity, addressing a key challenge for video-based sampling in estuaries.