6.1 Directed movement of fish with respect to the survey tracks
6.2 Avoidance effect
6.3 Overlapping survey layers
6.4 Shallow water
6.5 Water temperature and the propagation of the sonar beam
6.6 Quality of raw material used
6.7 Accuracy of calibration constant
6.8 Biomass species composition
6.9 The actual accuracy problem of acoustic surveys
The principal object of any sampling procedure is to secure a sample which, subject to limitations of size, will reproduce the characteristics of the target population. There are, however, a number of ways in which sources of bias can be inherent in the sampling procedure and affect the accuracy of the calculated estimates. In this chapter we discuss, in summary, the main sources giving rise to bias estimates of acoustic surveys (sonar, echo-integrator).
It has been observed that, a source of bias is generated from the directed movement of fish with respect to the survey tracks. Specifically, the following two sources of bias have been identified,
Bias - 1 |
(= positive bias): If there is a relative movement of the fish
in the same direction as the survey vessel, abundance may be
overestimated. |
|
|
Bias - 2 |
(= negative bias): If fish moving in the opposite direction as
the survey vessel, abundance may be underestimated. |
As we have discussed, there are two types of errors arising from biases due to the avoidance effect:
1. Echo-integrator surveys: The effect of bias forms a constant component of error (underestimation) affecting fish (schools) located in the upper layers of water, close to the research vessel.2. Sonar surveys: The effect of bias forms a constant component of error (underestimation) affecting schools at a close distance from the vessel. Also, schools split when escaping and thus give a biased picture of their actual size.
In an acoustic survey programme based on concurrent sonar surveys and echo-integrator surveys covering the same target population, there is a problem of double counting because of the beam pattern of the two systems. The following graph (Fig. 6.3a) provides a geometrical presentation of the overlapping layers of water by the two systems.
The following layers of water have been defined (Fig. 6.3a) and are covered by the effective beam of the two systems (sonar, echo-integrator).
Layer - A1: covered only by the sonar system (the echo-integrator system is effective from about 10 m below the surface downwards, in this example 5 m, due to vessel draft, plus 6 m before the echo-sounder operates normally)
Layer - A2: covered simultaneously by the sonar and echo-integrator beams
Layer - A3: covered simultaneously by the sonar and echo-integrator beams
Layer - (A4 + A5 + A6) covered only by the ecbo-integrator beams (note: A2 + A3 = A4 + A5)
Figure 6.3a Overlapping survey layers (sonar, echo-integrator systems)
From Figure 6.3a one can see that, the overlapping layers of water covered by both systems are
De-biasing technique:
An unbiased estimate of the survey biomass covered by the above two systems is given by
where
where: estimated total biomass
: estimated sonar total biomass
= estimated echo-integrator total biomass adjusted
and
In order to calculate an estimate of Bee additional information is collected during the acoustic programme on the vertical distribution of the survey stocks. Specifically, the following items of information are collected during the survey operations of an echo-integrator survey: estimated echo-integrator total biomass
: estimated echo-integrator total biomass within the domain ABCD
(i) vertical location of the gravity centre of the survey stocks on an ESDU basisThe above magnitudes (i, ii) are used for de-biasing the calculated estimates of echo-integrator total biomass.
(ii). vertical location of the upper and lower limits of the observed stocks on an ESDU basis
The following graph (Fig. 6.3b) portrays the observed vertical distribution of the survey stocks (Sardina pilchardus) off the Atlantic coast of Morocco, Nov./Dec. 1980. The total biomass estimates produced have been adjusted from the effect of double counting.
Another source of bias inherent in acoustic surveys is attributed to the depth coverage problem. Although the sample area is clearly defined at the design process of an acoustic programme there are cases in which the collection of information from certain domains of the survey population is impossible because of practical limitations. It is well known that research vessels cannot operate for security reasons in very shallow waters and because of that a part of the survey stocks is not covered by the survey. This, in turn, means that the biomass of these stocks is not included in the calculated overall biomass estimates (underestimation).
De-biasing methods have been developed for adjusting the original biomass estimates. Usually, supplementary information is used on statistical estimators(ratio estimates) aiming at adjusting the original totals from the effect of the missing categories of the survey stocks1
1 See Bazigos, G.P. and L. Rijavec, The effect of incomplete survey programmes on the accuracy of fish biomass estimates with Special reference to Southwest India. Rome, FAO, FIRM-IND/75/038, Report (4) (in preparation)Specifically, the following methods are used for de-biasing purposes:
1. Simple ratio estimate
The inshore sector is first divided into two depth domains, say, A2 (n.mi2): 0-15 m, not covered by the survey, A2 (n.mi2): 15-30 m, covered by the survey:
a) Assumption of uniform distribution of fish2. Modified ratio estimateAn estimate of the total biomass
in A1 is given by
![]()
where
b) Assumption of proportional allocation of fish according to the volume of water,(n.mi2): fishing area out of A1
(t/n.mi2): estimated average biomass per n.mi2 in A2.
![]()
where
V1, V2: volumes of water in A1 and A2 respectively: as above
If the assumptions made above concerning the distribution of fish in the inshore sector are not valid, the modified ratio estimate is used for estimating B1,
where
x1 and x2 catches per unit of effort in A1 and A2 respectively.and
as above
The temperature structure of the body of water affects the sonar beam. Strong vertical temperature gradients on the thermocline will bend the sonar beam downwards reducing its horizontal detection range, thus increasing the detection of deeper schools. It is therefore important to keep control of the vertical temperature structure of the survey body of water by using bathythermograph recordings in order to monitor the actual volume of water covered by the sonar.
It was observed that the quality of raw material used strongly affects the accuracy of the calculated estimates.
In echo-integrator surveys the main source of error giving rise to bias is that of decomposition of the total integrator reading into its component groups (all species of fish covered, plankton and larvae, others).
In sonar surveys the main sources of error giving rise to bias are the ones inherent in the identification of the obtained sonar traces and in measuring the dimensions of sonar traces, and mean school weight(s).
Another source of bias arising in echo-integrator biomass estimates can be attributed to systematic errors which might be inherent in calibration experiments and which affect the accuracy of the estimated calibration constant.
The accuracy of biomass species composition is affected by biases inherent in the methods used for the identification of the kind of fish covered by the echo surveys (concurrent fishing surveys, examining echo records).
The total sampling error of biomass estimates in acoustic samples can be considered to consist of two types of errors, biases, if any exist, and the random sampling error. The essence of bias is that it forms a constant component of error which does not decrease, in a large population, as the size of the sample increases, whereas the random sampling error decreases on the average as the size of the sample increases.
In this section we present a block diagram indicating the various factors which might affect the actual accuracy (total sampling error) of biomass estimates. In the diagram the various sources of error have been grouped under the following four headings:
A: Internal factors of a technical natureThe component factors within the above four categories are portrayed in the next block diagram:
B: Semi-internal factors of a technical nature
C: External factors of a statistical nature
D: Semi-external factors of a practical nature