Comparing on site human and video counts at Igarapava fish ladder , South-eastern Brazil

On site human observations and video images were collected and compared at the window of the Igarapava Dam fish ladder (IDFL), rio Grande , Southeastern Brazil, between March 1 and June 30, 2004. We conducted four experiments with two humans (Observer 1 and Observer 2) observing fish passage in the IDFL window while a Sony 3CCD video camera (Observer 3) recorded fish passage at the same time. Experiments, each one hour in length, were distributed throughout the diel cycle using full spectrum lights. We identified fish species, the number of individuals for each species, and the real time that they passed. Counts from each human observer were compared to the video counts. The fish species most commonly observed in the window were curimba (Prochilodus lineatus), mandi-amarelo (Pimelodus maculatus), piau-três-pintas (Leporinus friderici) and ferreirinha (Leporinus octofasciatus). The number of species and individuals were indistinguishable for the three observers. But, the number of species and individuals were significantly different among experiments. Thus, the three observers register the same number of species and count the same number of individuals even when these two response variables differ significantly among experiments. Based on these results, we concluded that the video count was an accurate method to assess fish passage at the IDFL.

Brazilian reservoir fisheries management actions are in general not successful because of shortage of scientific information, limitations of financial and human resources, and absence of monitoring (Agostinho et al., 2004).Monitoring fish passage in hydroelectric power plants is of paramount importance in evaluating the contribution of fish pass mechanisms in hydroelectric power plants for upstream reservoir fisheries.
The rio Grande, rio Paraná basin, is an important economic resource in Southeastern Brazil.In general, its fish stocks are low and some of its species are listed as threatened with extinction (Brazil, 2004).The 12 dams built along the 1,390 km of the river have undoubtedly contributed to its present poor fisheries status (Santos & Formagio, 2000).Igarapava Dam, located 501 km from the mouth, was completed in 1998, and is the 5 th dam upstream from the confluence with the rio Paraná.To allow upstream fish passage around Igarapava Dam, a fish ladder began operation in 1999.A fish counting window near the top of the ladder allows visual counts of fish as they exit the ladder.Unpublished data on fish passage have been gathered using on-site visual counts and video-images counts.
Estimates of the number and species of fish in a fish ladder through direct counts are generally considered as absolute, as they only happen once and cannot be tested for accuracy, nor is independent confirmation of species identification possible.The degree of confidence that may be placed in such estimates is impossible to estimate both for the number of fish passing and for individual specimen identification (Hatch et al., 1994, Hatch et al., 1998) Recent progress with video technology has allowed automation and reduced the need for visual counting carried out by observers in situ (Travade & Larinier, 2002).Timelapse video systems have been used to record migration of adult Pacific salmon (Hatch et al., 1994).This technique provides the opportunity to calculate variance and place confidence bounds on estimates.Video technology provides a permanent record of fish passage that can be reviewed multiple times by different readers to obtain accurate species and number of fish using the fish pass.Video fish counting can also reduce data gathering costs and increase the amount of data collected compared with on-site counting (Hatch et al., 1998).In most situations, time-lapse recorded video tapes of fish passage contain only a small percentage of frames of actual fish images.However, a counter must review all tape frames, no matter how many fish are present (Hatch et al.,1994).
Computerized image processing techniques have many potential applications for the analysis of fish passage (Travade & Larinier, 2002).Some authors believe that in the near future it may be possible for computers to count and identify fish with speed and accuracy at least as good as human counters.Eventually, computers may even be trained to recognize, classify and count fish (Nery, 2004).
This paper tested the feasibility of using video technology for estimating and evaluating fish passage at Igarapava Dam fish ladder (IDFL) on the rio Grande.The objectives of this study were to: i) document and count fish-ladder pas-sage using video technology; ii) compare fish counts generated from video recordings and on-site observers counts; iii) test species recognition between the two methods The study was conducted using images acquired in the IDFL sub sampled window whose dimensions were 1.17 m in width x 0.50 m in height x 1.00 m in depth (Fig. 1).The images were obtained between March 1 st and June 30 th , 2004.
On-site counts and video camera.Four experiments were conducted at different times of the day, starting respectively at 15:45, 19:25, 12:17 and 19:58.Two human observers (Observer 1 and Observer 2) made on-site observations of fish passing through the window while a Sony 3CCD video camera recorded fish passage at the same time (Observer 3 identified fish recorded).The experiments conducted during the day employed full spectrum lamps (CORALIFE 10,000K VHO 95W, High-Intensity Purified Super Daylight Lamp) whereas the night experiments were performed under a General Electric lamp, model Chill Chaser Deluxe, infrared and heatlamp (Hiebert et al., 2000).This lamp provided infrared and some visible light.The water turbidity was not measured in our experiments since it was clear enough not to interfere with the counts (Fig. 2).For each experiment and for the three observers we recorded the number of species and the number of individu- als passing by the IDFL window.During recording, the film record registered the time and date on each frame of video tape, providing a record of the exact time each fish passed through the counting slot.The species were identified by specialists that were the human observers.
All of the recorded films were watched by Observer 3 with the Ulead VideoStudio 6 SE Basic Program linked with the video camera.
Statistical analyses.We used a Mixed Model, 1-Way Analysis of Variance (ANOVA) (SAS, 1999) We executed an ANOVA for each of two independent variables: Observer and Experiment.We used a Mixed Model because the observers were selected from a large pool of fisheries professionals and cameras.But, the experiments were fixed with respect to time and space.

Counts and comparisons.
We found the number of species to be exactly the same (Table 1) (ANOVA: F = 0.00, p = 1.0000) and the number of individuals was statistically indistinguishable among observers (F = 0.04, p > 0.9569).But, the number of species was different among experiments (no statistical test is necessary because all observations for each observer were exactly the same within each experiment, i.e. variance = 0).The number of individuals counted was statistically significantly different among experiments (F = 81.80;p < 0.0001).
The results suggest the number of species passing may vary greatly but the human observers and the video camera register were not different from each other under these circumstances.Thus, the three observers registered the same number of species and counted the same number of individuals even though these two variables differ among experiments.
The technology of video cameras, computers and specialized software used in this study provide credible results compared with the human observers.It allowed monitoring the fish passage and the assurance of counts (Irvine et al., 1991;Hatch et al., 1994).The videotape system can make it easier, faster, and less expensive to count migratory fish passage (Irvine et al., 1991).It is most useful at locations or during times when relatively few fish will be observed per day (Hatch et al., 1998) as it avoids the need for full-time human observers.
Counts from video monitoring were found to be comparable to real-time counts in the fish ladder.The observer could also play the tape back an unlimited number of times and at varying speeds, and show the film to other fisheries professionals to confirm species identification (Hatch et al., 1994;Travade & Larinier, 2002).This was not an option for the human observers in real time at the IDFL window.
The reliability of detection of fish in front of the window depends on various environmental factors including lights, water turbidity, water velocity and the characteristics of fish   species including size, color, speed and swimming behavior (Warren & Pardew, 1998;Travade & Larinier, 2002).In the case of our study, the lights and water turbidity were adequate to compare the observers and review the species of fishes that passed by the window.This study has demonstrated that image acquisition hardware and a video editing software package could be used to automate essential fish-ladder observer functions.It also help to initiate the process for regulatory change that will allow for the deployment of video monitoring devices in lieu of human observers.

Fig. 1 .
Fig. 1.Camera configuration with Sony Camera (DCR-TRV950) in frame center.The two vertically oriented full spectrum lights are adjacent to the window; above is the infrared light bulb.

Fig. 2 .
Fig. 2. Still image capture of two Piaractus mesopotamicus and one Pimelodus maculatus, upstream direction, registered by the video counter.

Table 1 .
Species and number of individual observed in each experiment for two on site human observers (Obs.) and one video count (VC).