Evaluation of three methods for manually counting fish in dam turbines using DIDSON

Abstract

Mortality of fish entering the turbine draft tube (DT) from the tailrace is common during turbine startup or dewatering in Brazilian hydropower dams. Mortality can be reduced if such operations occur when few fish are present in the DT, but estimating abundance therein using traditional gears (e.g., cast net, gillnet, hook-and-line) is constrained by flow and access. We deployed a Dual frequency IDentification SONar (DIDSON) to record and count fish in the DT for the first time using an innovative approach. Methods used elsewhere for manual counting fish in DIDSON images generated dissimilar results. Therefore, we compared three methods of counting fish manually in the frames of DIDSON software from two Brazilian dams: numerical counting (number of fish in the frame), qualitative counting (percent of the frame occupied by fish), and intersection counting (number of intersections of the grid frame with fish). We assumed that the numerical counting provided the best estimates of the true value but could not be used in 32.2% of the frames from one of the dams due to the presence of uncountable aggregation of fish, bubble curtain, or poor contrast between fish and the DT floor. Qualitative counting was the most congruent, whereas intersection counting was the least, based on comparisons with numerical counting.