Red Tide Semantics and Statistics

Mary C. Christman1, Jacob Tustison2, Karen A. Steidinger2,3

1 University of Florida, Gainesville, FL

2 Florida Fish and Wildlife Conservation Commission, St. Petersburg, FL

3 University of South Florida, St. Petersburg, FL

When monitoring for harmful algal blooms part of the problem is recognizing when a bloom is occurring and what the likely trajectory is. As a result, in coastal monitoring programs for harmful algal blooms, extensive coastlines such as Florida can be problematic for sampling strategy. This is particularly true for species such as Karenia brevis if three dimensional sampling is not practical because of areal expanse, depth, and rapid assessment needs. Until coastal ocean observing systems are in place for harmful algae and/or their toxins, point sampling is the usual strategy of data collection but its usefulness depends on location, depth, and timing. Where point sampling has been particularly effective has been on ship cruises that sampled fixed stations along transects from inshore to offshore, surface to bottom. Such cruises are in the Florida Fish and Wildlife Conservation Commission’s Red Tide Database, e.g., cruises by USFWS, NASA, Florida Department of Natural Resources, Florida Department of Environmental Regulation, Mote Marine Laboratory, and ECOHAB. Compounding the potential difficulties of the choices of sample station location, depth and frequency of sampling are the choices of sample and subsample sizes for determining abundance of cells. For example, in the Florida program, up to a one liter water sample may be collected and fixed with Lugol’s solution then a pooled 3 ml subsample is enumerated to derive an estimate of abundance. The limit of detection for this protocol is 333 cells L-1 and the results are scaled up with the attendant increase in the uncertainty of the estimate. This can be an issue when one thousand cells per liter represents background levels in the Gulf of Mexico and 5000 cells L-1 is the regulatory limit that activates closure of shellfish beds inshore. On the other hand, red tides are detectable from satellites typically at >50000 cells L-1 which is a developing or ongoing bloom. In addition, K. brevis blooms have penetrated coastal waters along the bottom without surface expression until nearshore and so would not be detectable by satellite or by data collected only in surface waters. The Florida Red Tide database, principally from event response cruises, is composed of about 80% surface sample results and about 20% bottom results (>90000 samples If K. brevis is not recorded in surface samples or listed as “present” does that mean that a bloom is not somewhere around, perhaps on the bottom? What is the probability of a K. brevis bloom being somewhere in nearby coastal waters if a sample has a count of 20000 cells L-1? What is the probability of a bloom in nearby coastal waters if surface water from 30 stations had “zero” records? Is a “negative” or “zero” record a result of small sample size, e.g. 1 liter vs 10 liters, or counts less that the detection limit? Is it necessary in the case of K. brevis to monitor at least surface and bottom rather than just surface water samples to detect potential blooms or bloom development or is surface sampling enough depending on what stage of the bloom is being sought and at what distance offshore? What K. brevis surface count estimate is indicative that a bloom is developing? What is the probability, based on a single count or a cluster of counts, that a red tide is present, if a red tide is defined as 10000, 50000 or 100000 cells L-1? To address these questions and others, we apply Bayesian and other i statistical approaches to data from the Florida database and results will be discussed.