Tropical Open-Ocean Aquaculture Modeling: AquaModel Tuning and Validation

Executive Summary

Performance assessment, regional tuning and validation of a software program known as AquaModel were the primary goals of this study. The software was designed for use by governments and industry to predict the sea bottom and water column effects of fish aquaculture. From an industry perspective, it also includes advanced tools to optimize fish production by obviating the usual trial and error method of configuring pen spacing and loading, by estimating optimum fish loading and culture density for growth in relation to currents and ambient oxygen supply.

The benthic submodel of AquaModel software was applied, tuned and validated at the Blue Ocean Mariculture LLC fish farm site near Kona on the big island of Hawai’i (herein “study site”). Fish production is relatively small at present and this factor, combined with the deep water location and moderately strong current velocity and variable directions of flow allows the organic wastes from the farm to be spread over a very large area and readily assimilated into the food web without perturbations. Seven years of field data from five locations was collected by an independent scientist who reports to the State of Hawaii government for this study. To more accurately simulate the fish farm waste production, AquaModel staff created the first physiological and growth model of the cultured fish, Seriola rivoliana (aka Kampachi^TM or Almaco jack) and tuned this submodel to produce the same growth patterns and food conversion ratios seen at the study site.

Model Overview

AquaModel is composed of interlinked submodels of fish physiology, hydrodynamics, water/sediment quality, solids dispersion and assimilation into the aquatic food web. The model simultaneously calculates and displays a time series of water and sediment quality conditions resulting from fish feed ingestion, fish growth, respiration, excretion, and egestion. The user is presented with a 3-dimensional video-like simulation of growth, metabolic activity of caged fish, associated flow and transformation of nutrients, oxygen, and particulate wastes in adjacent waters and sediments. The software is used by government managers and researchers in several locations worldwide and is presently being formally validated in Canada and Chile at five large fish farms. These validation activities have necessitated numerous upgrades and new utilities in AquaModel, some that are described herein.

New Fish Submodel

Solid wastes dynamics in the model are calculated from feed consumed and a small percent of waste, as well as the assimilation efficiency and food conversion ratio. These results were compared to measurements and estimates from the fish farmers as a quality assurance measure. A physiological model of the cultured fish species reared at Blue Ocean Mariculture (Seriola rivoliani, aka “Kampachi^TM”) was created for this project and tested to produce growth and food conversion efficiency results similar to that achieved at the farm site.

Circulation of Study Site and Prior Monitoring Results

Accuracy of aquaculture models is strongly related to the quality of the physical oceanographic inputs, particularly in open ocean conditions where non-tidal forcing factors result in considerable variation of flow rates and directions. Two months of continuous surface to bottom (ADCP) current meter records were collected every 20 minutes at the center of the net pen area lease. Surface currents above submerged net pen depth were strong, averaging about 28 cm s^-1, but these subsurface readings were affected to some degree by backscatter from the water-air interface. Reliable current velocity readings were obtained from about 10 meters depth (top of submerged net-pen depth and below) to a few meters above the bottom averaging about 9 to 13 cm s^-1 (SD range 7 – 9 cm s^-1). Polar current vector diagrams produced by AquaModel indicated good dispersion flows in all directions with dominance to the northeast and southwest at net-pen depth and flowing mainly to north and south nearest the sea bottom. These characteristics indicate suitable conditions for rearing fish and provide regular resuspension of solid wastes on the sea bottom. Resuspension allows for aerobic assimilation of the waste feed and fish feces.

The sea bottom was composed of a thin, coarse-sand layer over hardpan and had very low background (reference station) total organic carbon concentrations (TOC) of about 0.14% (SD = 0.03) as measured over several years of monitoring. There were four reference areas sampled and one near-net-pen location from the center of the aquatic lease area. Field data suggested only a possible increase of about 0.1 to 0.2 %TOC near the center of the net pen locations to values of 0.15 or 0.16 %TOC (SD = 0.05), respectively. Statistical difference (p =0.035, df =6) was found comparing sediment TOC results of annual mean from a reference area to the center of the net pen area. No field data were available from sediments immediately adjacent to the net pens but the model produced estimates for all locations.

Modeling Challenge

Because of the naturally low organic carbon content of the sea bottom and the relatively small size of the fish farm and the limitations of a single net pen area sampling location, it was not certain at the outset that the model could produce reasonable results. Typically, aquaculture models are used at or for planning of fish farms that may be much larger than the Blue Ocean Mariculture project. Most other farms are located in shallower water, sometimes with lesser current velocity and this produces a strong benthic-effects signal. Therefore, the signal to noise ratio is high for these other farms, but low by a factor of about 5 to 10 or more for the study site. The model was set up to grow concurrent crops of Kampachi^TM in each cage to a total fish biomass of 590 metric tons, slightly exceeding previous annual production. Figure 1 illustrates one of thousands of frames of the video-like output that the model produces. This one is from near the end of the fish production simulation with maximum fish biomass. The color scale in Figure 1 was adjusted to show an extremely low range of TOC concentrations. Solid green color indicates values of about 0.18 %TOC or about 0.04 %TOC above background, a difference that is similar to the normal error range of a high-precision laboratory analysis.

Model Performance

After calibrating and tuning the AquaModel to regional conditions, it produced background (reference) conditions within >0.001 %TOC of measured, steady-state reference-station values. This is essentially no difference between modeled and measured and certainly not with respect to measurable outcomes in the field. This is noteworthy as other benthic aquaculture models have been unable to maintain background organic carbon steady state concentrations due to resuspension washing TOC out of their modeling domains. With AquaModel, best estimates of the results at the single sampling station nearest the net pens were within<0.0012 %TOC of measured, long-term average results for the best-tuned setup. AquaModel consistently produced slightly higher sediment TOC concentration estimates (<0.02% TOC) at other locations nearer the two largest pens that had no corresponding field data measurements to verify the model predictions at these locations. All of the >250 simulations performed for this study indicated the same spatial pattern of increased TOC, with differing values depending on the calibration settings. None of the TOC concentrations measured or modeled indicated any risk of sea bottom eutrophication or probable significant biological change.

Validation Outcome

This study indicates that the tuned and validated AquaModel program should be sufficiently robust to model other open ocean locations of the leeward shores of the Hawai’ian Islands. The model is designed to work effectively with much higher levels of sediment organic carbon loading from fish farms, but not at grossly eutrophic cage sites in some sheltered, inshore cage locations utilized decades ago. AquaModel use would readily identify such outcomes through observation of several parameters, such as TOC delivery rate to the bottom (“TOC Rate”, in grams carbon per m2 per day) as well as sediment interstitial oxygen and sediment sulfides results. With separate cages that are spaced appropriately, most open ocean locations on the leeward shores of the Hawai’ian Islands that are in sufficient depth of water would not produce eutrophic or even modestly elevated sediment conditions. However, some habitats are considered of special biological significance, where net pen siting should not be considered.

Overview and AquaModel Use in Hawai’i

This evaluation, along with the existing routine monitoring program at the subject site as well as other analyses cited herein, indicate that the fish farm operation is not adversely affecting benthic conditions in the area. The waste tracking utilities of AquaModel applied to this particular site indicate that a small fraction of the waste fish feces reaches locations outside the modeling domain. The estimated loading rate of organic carbon in those locations are so minimal at present that it produces no measurable or even modeling-predicted change in concentration of sediment TOC. The chance of changing the biology of the benthos at these same locations is therefore highly unlikely. In general, small amounts of TOC added to the sea bottom from any source in the marine environment have been found to increase biodiversity and abundance of benthic organisms, but often at nearshore fish farms, these levels are exceeded. AquaModel provides a convenient and relatively accurate means of estimating future carrying capacity for this farm or groups of farms in the future. It also should be used to inform future monitoring efforts, rather than selecting sampling locations through best guess or randomly. Now that regional tuning is complete, configuring and running the model is not difficult for other locations similar to the west coast of the Big Island of Hawai’i and in other similar habitats throughout the region.

AquaModel validation continues at other sites around the world that are larger in fish biomass and more replete with measurement locations in the field. Optimum model calibrations or trends identified in this study were in many cases as expected and occurred in other model validation locations. These findings, combined with prior model use experience and published literature guidance gives us confidence that the validation procedure employed herein is not a product of simple coincidence.