Lo et al : Modeling performance of an airborne lidar survey system for anchovy 



279 



point where the mean may fall or what differences 

 might exist between species. 



Detection of schools during the night would im- 

 prove if maximum schooling depth and a were cor- 

 related as Hunter and Nicholl (1985) speculated. 

 They determined the visual threshold for schooling 

 in northern anchovy (6xlO"i^W/cm-) and suggested 

 that maximum nighttime depth was a function of 

 the ability offish to see one another. They estimated 

 that the visual threshold for schooling would occur 

 at 38 m during a full moon and at 30 m on a starlit 

 night where chlorophyll was 0.2 mg Chla/m'^ and at 

 8 m (starlit) and 20 m (full moon) when chlorophyll 

 was 2.0 mg Chla/m^. If Hunter and Nicholl (1985) 

 are correct, then the maximum lidar detection depth 

 should increase as the schooling depth increases at 

 night, as long as survey flights are made under the 

 same moon phase. This relationship between visual 

 threshold and moon phase also indicates that it may 

 also be important to exclude survey nights during a 

 full moon — a rule long observed by pilots who locate 

 schools for the fishing industry. 



It may be possible in practice to detect schools 

 somewhat deeper than those that our model indi- 

 cates because the model estimates the detection of a 

 single pulse at one range gate or depth. In practice, 

 a lidar will generate a composite image of a school 

 derived from a number of such pulses over a range of 

 gates (depths) analogous to an echogram trace. Such 

 a composite image produced from multiple returns 

 and gates can be more readily separated from back- 

 ground noise than can a single pulse, but such a 

 separation involves a more complex, and at the pres- 

 ent time, somewhat more qualitative discrimina- 

 tion process. Signal-processing algorithms can be 

 developed for this application, but their performance 

 would depend on the exact algorithm used. More 

 accurate estimates of detection depth would depend 

 upon the development of such signal-processing algo- 

 rithms. Development of such algorithms is one of the 

 most promising directions for future research on fish- 

 eries lidar. Their development would greatly improve 

 both the accuracy and precision of future lidar surveys 

 for fisheries, as well as reduce the work in processing 

 images. Similarly, a more thorough understanding of 

 the causes of the observed variation in the vertical 

 distribution offish could improve survey accuracy and 

 precision. The phase of the moon, time of day, mixed 

 layer depth, temperature, location of forage, fish size, 

 season, and spawning habitats, may all influence 

 where in the water column a school may be found. 



It seems unlikely that depth of detection will be 

 greatly improved by increasing sensitivity or power 

 of a lidar system over the basic radiometric system 

 used in our model. Our analysis indicated that an 



order of magnitude increase in equivalent laser power 

 (laser power plus sensor changes) would gain about 

 10 m in detection depth. Such a change would require 

 a custom, rather than an "off-the-shelf laser, which 

 would cost around a million dollars, in addition to 

 associated costs, including a larger aircraft to satisfy 

 the new power and weight requirements. In addition, 

 increasing the depth of penetration by 10 or 20 m, 

 on the average, would not increase the numbers of 

 schools detected by more than about 10% during the 

 day because school distributions tend to be skewed 

 with a long tail extending to depths far beyond the 

 practical limits of lidar detection in coastal waters. A 

 10-m gain would be more significant during the night 

 but may not be worth the additional cost. 



We have treated the failure of a lidar to count deep 

 schools as a potential bias, which is true unless an 

 unbiased estimate of the mean vertical distribution 

 of schools exists for the particular survey region and 

 season and an appropriate statistical model is used for 

 the sui-vey. When these conditions are met, the fail- 

 ure of a lidar to count deep schools becomes a matter 

 of precision rather than bias. An unbiased estimate 

 of the mean vertical distribution of schools could be 

 estimated from data generated by lidar and acoustic 

 surveys for the same region because by combining the 

 two surveys, one corrects for the vertical bias in each. 

 The appropriate statistical model for a lidar survey 

 would be one based on line transect theory (Buckland 

 et al., 1993). Line transect theory usually deals with 

 encounter rates on the horizontal plane, and animals 

 are assumed to be uniformly distributed in space. In 

 the case of lidar, we turned the model on its side 

 and used an average vertical distribution of anchovy 

 schools in the survey area. An empirically derived ver- 

 tical distribution does not seem to be subject to any 

 more bias than a uniform, horizontal distribution, one 

 that is commonly assummed in line transect surveys. 



To provide indices of relative abundance based on 

 airborne lidar is an important fishery application 

 that is less demanding than that of estimating bio- 

 mass. For an index of abundance, the extent to which 

 schools are available for counting is not a major con- 

 cern. Lidar seems uniquely well-suited for taking an 

 inventoi-y of the juveniles of small pelagic fishes (pre- 

 recruits) because they are extremely patchy and tend 

 to inhabit shallow water near the coast in areas dif- 

 ficult to sample with a research vessel. Lidar sur- 

 veys can provide useful indices of adult biomass as 

 well. Aerial observations (Lo et al., 1992) and passive 

 imaging ( Nakashima, 1990; Nakashima and Borstad^) 



Nakashima. B. S., and G. A. Borstad. 1993. Detecting and 

 measuring pelagic fish schools using remote sensing tech- 

 niques. ICES Report CM. 1993/B:7, session T, Fish Capture 

 Committee, 18 p. 



