Introduction

One of the most vexing concerns of reef aquarists is the problem of water chemistry. These concerns are founded in the knowledge that reef aquaria are really very tiny analogues of natural systems, and because of their small size chemical changes could occur that could adversely impact their livestock. The composition of natural sea water (NSW) is taken as the basis for measuring all changes in the water in aquaria, even though the standard for such natural seawater has historically been from the North Sea and not coral reefs. Chemical concentrations of most materials in seawater are presumed by most aquarists to be "conservative" and unchangeable, consequently, it is presumed that any changes away from the NSW concentrations are detrimental. On the gross scale of the world ocean's such a supposition has some validity, however in the close environment of the reef significant changes in many chemical constituents can and do occur. However, the data detailing with changes in coral reef water chemistry are sparse, and difficult to relate to the standard reef aquarium. Most aquarists, probably correctly, feel that they can't go wrong if they pursue a course of aquarium water chemistry management based trying to maintain "natural levels."

Implicit in such a management plan, although seldom directly stated, is the feeling that organisms somehow "use up," "change," or "consume" many of these chemicals, and in doing so, forever remove the chemicals from that reef aquarium system. This assumption is not completely false, some chemicals are "used up" and removed from the system, but most are not. Organisms are dynamic entities, and while some chemicals are temporarily sequestered away, such chemicals generally remain available in the system due to metabolic turnover. The only real exceptions to this as far as organisms are concerned are those chemicals, such as calcium, which get incorporated into an insoluble matrix. Once the chemical is removed from solution, even if that matrix, such as a coral skeleton or clam shell, remains in the aquarium, it is beyond the use of most organisms.

Another more important variable of these systems is chemical export via some sort of filtration or organism removal. The various types of filtration are widely regarded as very efficient, not at all efficient, or useless, depending on the authority consulted and presumably the whim of the moment. Unfortunately, there are few hard data to support any of the suppositions about export, except that it occurs. I hope to address the question of export methodologies and their effectiveness in a future research project, but for the present, we simply don’t have a good handle on what, or how, various chemicals leave our aquaria.

If we just started with sea water and had to deal with its changes over time, that would make the determination of what is happening in our systems relatively easy. Note that I didn't say, "Easy" but "relatively easy." However, we don't generally start with real seawater. Relatively few aquarists have access to natural seawater, and must use dried artificial mixes and some type of water to make the medium they put into their aquaria. These mixtures may vary significantly in many regards from natural water (See Atkinson and Bingman, 1999: AFM-online). Additionally, and importantly, we continually further alter the chemical constituents of our systems by adding food, and various other additives. The food is added based on our observations and knowledge of the organisms' needs. The other additives are added based on chemical tests, belief in the advertisements of the additive producers, barometric pressure, a toss of the dice, or maybe, just maybe, some rational guesstimate of depletion rates.

This report is the first of several, at least three and maybe more, examining the chemical constituents of marine aquarium water. It is a follow-up to my "Food and Additive Study" done about two years ago. In that project, I examined the chemical constituency of 15 popular foods and additives used by aquarists. The data from that study are available here: AFM-online. Using those data, we can reasonably well determine what chemicals may enter our systems. However, until now, we really didn't have any really good idea of what exactly is in our systems.

This study is an attempt to describe the chemical constituents of an "Average Reef Tank." The goal was to survey as wide a database as possible in the hope that the results of the study would be representative of reef tanks, in general. To that end, I advertised for volunteers willing to pay for the analysis of their tank water. I had hoped to get at least 15 volunteers, I got 18, and we were able to analyze the data from 23 different systems as well as one sample of artificial sea water. The participants had to answer a somewhat detailed questionnaire about their systems, and how those systems were maintained. Those responses and further analyses will constitute the basis for the forthcoming articles. This article will deal with the basic results of the water analyses of these systems.

Materials and Methods

I am an invertebrate zoologist, not an analytical chemist, so I couldn't do the necessary tests. They were done by a commercial analytical laboratory in the Seattle region.

The laboratory was:

AM TEST LABORATORIES, INC.
14603 NE 87th Street
Redmond, WA 98052.

They used two basic sources for analytical techniques:

EPA = Methods for Chemical Analysis of Water and Wastes, 1983 (available from National Technical Information Service, 5285 Port Royal Road, Springfield, VA 22161, Stock No. NTIS PB84-128677);

SM = American Public Health Association. 1992. Standard Methods for the examination of water and wastewater, 18th ed. American Public Health Association, Washington, D.C.

The concentrations were determined by Inductively Coupled Plasma Emission Spectrometry or ICP Scan, EPA method 200.7.

After verification of the participants, the laboratory was contacted and they sent sampling materials to me, these were plastic bottles labeled for each of three necessary samples from each aquarium. In total, each sample consisted of three 500 ml samples of aquarium water, one each for the metals, nitrogen chemistry, and conventional nutrients. After I received the sampling bottles, I prepackaged them and distributed them to the participants along with a questionnaire and instructions. The samples were returned me and then sent to the analytical laboratory. After the analysis, the data were returned to me, and electronically coded for analysis.

There were 24 samples analyzed: 23 aquarium samples, and one sample, of Instant Ocean(tm) (IO), prepared with 0 total dissolved solids RO/DI, mixed to 35 ppt, and stored for two weeks prior to testing in an FDA food-grade container. For comparative analyses, the chemical constituents for NSW were obtained from Weast, 1966.

The samples constituted independent samples without replication. The lack of replication significantly reduces the options for statistical analyses, however, it was necessary to obtain as wide a sample as possible. Given the cost of the sample analysis, analysis of replicates was simply too expensive. The lack of sample replicates precluded the use of tests such as analyses of variance (ANOVA), and other tests on means and variances.

However, basic descriptive techniques were possible as were many sample-to-sample comparisons. These analyses were dependent on the numerical values of the concentrations of the various constituents. The numerical values ranged from values as small as 0.0001 µg/l to as high as 12000 µg/l. This is a range variation in abundance of 100,000,000 times. Obviously, in any comparison those samples with high numerical values of any chemical could overwhelm the information in those with smaller amounts of chemicals. Consequently, all tests were run twice, once using untransformed data, and the second time with a log normal transformation of the data. A log normal transformation of the data involves taking the value for the datum, adding one to it, and taking the natural logarithm of the sample. Algebraically, this transformation is indicated by the following terminology: ln(x+1), where x = the sample datum. Such a log-normal transformation is commonly done in those analyses where data may very over several orders of magnitude and it reduces the influence of large values and increases the influence of smaller values. The transformation has known properties and quite useful in comparisons involving widely dissimilar data values. In these samples, this transformation reduced the influence of relatively highly concentrated materials, for example Sodium, Sulfur, and Magnesium.

In the initial analyses, the salinity varied widely. I presumed many of the constituent abundances were correlated with salinity; for example, if two batches artificial salt mix was made up, one a concentration of 30 ppt, and the other to a concentration 36 ppt, there would be differences in all of the materials found in the two samples. However, none of these differences would be due to any "in tank" changes or variables. Consequently I normalized all data by adjusting all the samples to the same sodium concentration as found in the tabled value for NSW; this was done by multiplying all concentrations by a constant consisting of the sodium concentration divided by the sodium concentration in NSW. Finally, after the initial analyses were completed, I removed from consideration all chemicals found in NSW but below detection limits in all of the samples.

The elements removed from these analyses were: Beryllium, Cadmium, Chromium, Iron, Lead, Manganese, Mercury, Selenium, Silver and Yttrium.

Descriptive statistics were determined on a spreadsheet, Quattro Pro (Corel, 1997). The relationships between and within the samples were investigated using a series of analytical programs called the Community Analyses System, V. 5.0 (Bloom, 1994). Initially, the samples underwent a process called classification. The index used to classify the samples was the Percent Similarity (or Bray-Curtis) Index.

Bray-Curtis Index

This index compares the relative proportion of each element in each sample by subtracting one from the other. The values for all the elements are summed and subtracted from one. Two samples are considered to be identical if they have a proportional similarity of 1 and completely dissimilar with a value of 0. During the process of "classification," each sample was compared with all other samples and all possible groups of samples. Finally a representation of which samples are most similar to each other is determined. The samples then are arranged in series of groups based on relative similarity. The graphical representations of such relative similarity lineages resulting from these analyses are called dendrograms (see Figures 1-4). Such dendrograms typically show clusters of samples, and hence the whole process of classification is sometimes called "cluster analysis." Such an analysis results in an unambiguous, quantitative and replicable way of showing sample similarity.

Subsequent to the classification were examined by ordination. Ordination is a statistical approach and methodology used to suggest interrelationships between samples. There are a number of different techniques to do this, I used the process called Principal Coordinate Analysis as it has fewer inherent biases than many of the other processes, provided the data are appropriate and the process is done correctly.

In our natural three-dimensional world, objects may be described and compared based on the three principal dimensional axes of length, width, and height. Ordination results in a similar comparison, but in an artificial universe with more than three dimensions. In the case of this study, ordination may be thought of as a process of creating a mathematically defined universe and arraying the samples in it based on their relative chemical concentrations. This universe would not have just the normal three spatial dimensions. Instead, it would have a number of dimensions corresponding to the total number of chemicals or constituents being tested. The data from the samples are placed in the volume created in this universe and manipulated to "explain" most of the differences between them by creating and rotating the dimensional axes to maximize the differences on the first axis, while minimizing those differences on all subsequent ones. Ideally, one tries construct these axes so that most of the information is contained on the first three axes.

Most people, including me, have significant trouble visualizing a fourth, fifth, sixth or seventh dimensional volume (or in the case of this study, a 30 to 40 dimensional volume). Three dimensions work just fine, for visualizations, IF we can get there in the analyses. In diverse data sets such as these, often the sample variability is so high that one really does need to think about the samples in five or more dimensions. In these cases, we are generally out of luck as far as visualizing the samples, and have to deal with paper data. For these ordination analyses, the data were classified using Gower's Distance Index, and the concentrations were zero mean and unit variance transformed.

Gower's Distance Index

Subsequent to Ordination, the centroid of the data in the volume determined by first three primary axes in ordination space was calculated and a non-parametric (Mann-Whitney) 95% Confidence Interval Interval was calculated about that centroid (See Bloom, 1994, for this procedure and for a good description of classification and ordination techniques).

To help interpret the data, particularly the classification and ordination the sample ranks for each constituent were calculated. The top three and the bottom three ranks were generally determined.

Results

A summary of the concentration data is given in Table 1. Although some chemicals, particularly Calcium and Sodium, are generally found in values that approximate those of natural seawater, the data are more notable for the diversity of values than for their consistency. Many of the rarer trace elements, such as Beryllium and Selenium, if present, were below detectable levels. Generally, toxic materials such as Cadmium, Lead and Mercury were also undetectable.

In contrast, the values for other materials, notably Antimony, Cobalt, and Titanium were hundreds of times more concentrated in the average sample than in natural seawater.