A General Guide for Deriving Abundance Estimates from Hydroacoustic Data















Uncertainty in acoustic surveys

The accuracy and precision of the acoustic survey is the combination of uncertainties from many sources (Simmonds et al. 1992, Aglen 1994, Rose et al. 2000, Demer 2004).  We have discussed many of these errors.  What follows is the discussion on uncertainty in Rudstam et al. (2008b). 

There are errors associated with the calibration and operation of the instrument, with weather conditions associated with bubble attenuation and transducer motion, with hydrographic conditions and the associated changes in absorption along a survey, with differences between the survey area and the area occupied by the fish stock, with patchiness, and with the behavior of the fish.  Errors associated with fish distribution and behavior may be the largest contributor.  Fish behavior leads to uncertainty through changes in tilt angle and its effects on TS, avoidance or attraction to survey vessels, acoustic shadowing in dense schools, vertical migration in and out of the surface and bottom dead zones, and larger migrations out of or within the survey area.  In addition, identification of the target species either through echograms or trawl samples, and inclusion of non-target species can bias results.  These errors can be difficult to quantify and often rely on the experience of the biologist conducting the survey.  If using trawl samples, errors associated with trawl catchability add to the uncertainty of the acoustic estimates. 

Estimates of the combined error require estimates of the error in each of the components.  Some of the errors are additive (contribution from fish in the dead zones, fish avoidance, noise) whereas others are multiplicative (calibration, TS, attenuation in schools).  Also, some errors are known to bias the results in a particular direction.  For example, transducer movement will on the average decrease the measured echo level (Simmonds et al. 1992).  If we can make reasonable estimates of the distribution of the different error terms, it is possible to calculate the combined uncertainty in the estimate.  This is seldom done in the Great Lakes and errors reported as uncertainty are generally limited to uncertainty associated with spatial sampling and patchiness.  Simmonds et al. (1992) reviewed errors and biases and suggested that a typical coefficient of variation (SE/mean) is in the order of 26% for relative estimates (based on sa alone) and 35% for absolute estimates (based on both sa and σbs).  They suggest that most uncertainty in relative estimates is due to spatial sampling and that uncertainty in TS may be of equal importance for absolute estimates with an unknown and potentially large component associated with avoidance.  However, the main sources of uncertainty may vary between surveys. Rose et al. (2000) compared sources of variance for surveys with high and low assumptions on uncertainty of collection parameters.  Not surprisingly, spatial variance in measured Sv values dominated, but variance in individual species identification, detectability, and target strength were also important when uncertainty in those parameters increased.