Aquatic organisms
are patchily distributed. When using acoustic technology to map
and count animals, patches may include small animals that can be
resolved as individuals, single small or large animals, mixes of
predators and prey, and dense aggregations of animals that can not
be resolved. We were interested in how best to estimate densities
of organisms by combining echo integration with *in situ *target
strengths when individuals can not be resolved.

We simulated spatially random and spatially autocorrelated fish density and length distributions in a matrix of 200 by 200 cells to examine the sensitivity of abundance estimates to organism distribution, choice of acoustic size, and to quantify variance in acoustic-based estimates of density, length, and abundance.

Cells were filled to simulate random distributions, dispersed layers, mixed aggregations of predators and prey, and discrete aggregations of fish. To simulate the lack of isolated targets within cells, target strength data was removed from randomly chosen cells (marked with an x) in 5% increments up to a maximum of 95%. | |

Cells
with individual targets removed |

Four methods were used to choose a representative acoustic size for cells that lacked individual targets:

Results from computer simulations show that it is difficult to estimate fish abundance and maintain an accurate length-frequency distribution. Among acoustic size estimation methods, a weighted-mean from a local search window provided optimal estimates of density, abundance, and length. Below are two examples of abundance estimate simulations. The left panes show original densities of cells in random (null model) and predator-prey (discrete aggregations) distributions. The center panes show the distribution and density of cells with known target strengths after 95% of cell with individual targets have been removed. Right panes show estimated densities in all cells using a nearest neighbor or local-window fill method for cells lacking individual targets.

Normal Uni-Modal Length Distribution | 95% Target Removal | Nearest Neighbor Fill Method | ||||

Null Model | ||||||

Normal Bi-Modal Length Distribution | 95% Target Removal | Local-Window Fill Method | ||||

Discrete Aggregations | ||||||

The Null model (upper left pane) has a random distribution of fish densities with lengths drawn from a uni-modal distribution (i.e. one hump). When densities are estimated using the nearest neighbor method, artificial density structure was created in the data. Patches of animals were created where they did not originally exist. The discrete aggregations model (lower left pane) simulates a mixture of predator and more numerous prey patches with prey also concentrated along a thermocline. Length distributions are chosen from a bi-modal (i.e. two humps) frequency distribution. The local-window fill method (lower right pane) accurately estimated fish densities, total array abundances, and preserved length-frequency distributions, even when only 5% of the cells had individual known target strengths.

**Relevant Publications**

Jech, J.M. and J.K. Horne. Quantifying variability in acoustic estimates of fish | |

density: effects of spatial distribution. ICES Journal of marine Science (in review). |