APIS - Ecology

Overview

Biomass has been measured by a wide variety of methods and sampling designs. For that reason, it would be beneficial to review some techniques for sampling biomass (both destructive and nondestructive), sampling designs, and sizes for biomass samplers. In addition, some consideration should be given for sample processing and data analysis.

Methods

Biomass sampling can be done either by destructive methods, where plant material is actually collected from the site and weighed, or by nondestructive techniques, in which an alternate measure related to weight (such as length, height, or biovolume) has been calibrated using a subsampling of destructive plant samples measuring weight, and the two quantitative variables related by regression. In either case, some destructive biomass samples must be taken. Samples may either be taken directly by the individual, using an appropriately-sized quadrat (sample size guide), or using a specially-designed sampling device, generally for the purpose of collecting samples from a boat (e.g., Osborne 1984).

The best method for collecting samples is directly, by the observer. These samples may be taken by wading in shallow lake areas and streams (Madsen and Adams 1988) or by SCUBA divers in deeper lake environments (Photo 1) (Downing and Anderson 1985). Only plants observed to be rooted within the quadrat should be sampled, and all plants must be harvested completely from within the quadrat. Whether or not root material is to be harvested should be decided beforehand. If only shoot material (i.e., above-ground material) is desired, plants need to be carefully clipped or broken at the sediment surface by the sampler/diver. If roots are desired, then the sampler will carefully harvest all root material attached to the plants, either by digging with his/her hands or using an implement. If possible, the same individuals should be used repeatedly for sampling, and sampling personnel must discuss sampling techniques before beginning a sampling program to maintain consistency. Direct sampling by an individual is demonstrably superior to sampling by indirect, mechanical means if only because an individual is on hand to decide what is within or outside of the quadrat, and to ensure that no material is lost.

Photo 1

Under many circumstances, however, direct sampling is not feasible due to unsafe diving conditions, lack of experienced divers, or poor visibility. Under these conditions, a biomass sampler may be utilized. Many dredges commonly used for sampling surface sediments have been utilized, such as the Eckman and Ponar, but they have generally been found to be inferior for sample collection (Westlake 1969). Due to the volume of biomass in dense plant growths, a specialized sampler is needed. Table 2 presents a brief list of just a few samplers. These samplers follow two general patterns: dredges and corers. The dredge uses mechanical jaws to collect a sample within the enclosed box. These samplers generally lack precision, in that large amounts of extra material might be harvested if stems are intermixed, or material around the edges may not be harvested. In addition, sticks or other foreign objects may clog the mechanism, causing the loss of plant material. Also, these samplers do a poor job harvesting root material. Core samplers generally work better, are more precise, and may actually sample root material better than direct quadrat methods. However, these samplers are of necessity small in area, requiring a large number of samples (see Quadrat Size). Core samplers are typically difficult to operate in clay substrates (Sutton 1982), and course sands. They work best in soft, fine sediments up to fine sands, and in shallower depths (i.e., less than 3 m). At greater depths (i.e., 4 m), the mechanism is ungainly and difficult to operate (Fornwall and Hough 1990). In addition to these corers, a simple cylindrical core sampler with stoppers on each end may be used by SCUBA divers for core sampling (Madsen 1990).

Photo 1

Many individuals investigating species composition and abundance of aquatic plants in lakes simply estimate the abundance of plants visually. Although this technique might be quite suitable for initial surveys or quick assessments, it cannot take the place of a quantitative analysis, particularly when evaluating the effectiveness of a particular control method. Some have taken this type of visual assessment one step further, either by visual observations of abundance in a given area (Sheldon and Boylen 1978) or actual sampling of plant material, which is then assessed as to amount (Schloesser and Manny 1984). These observation techniques also require a "calibration" of the observer to quantitative biomass estimates. Although this might sound like a viable alternative, the human eye is yet too qualitative to be used to quantify material. These methods improve on simple visual assessments only thorough the collection of information at many sites around the lake of study.

Appropriate nondestructive biomass techniques must substitute an equally quantitative parameter in place of the destructive sampling of a unit area of plant material, not merely rely on a visually-estimated, qualitative judgment. For experimental purposes, either the fresh weight, length and density of shoots, or fresh volume of plants could be measured before, during, and after the experiment (Pine et al. 1989). These methods would be especially useful to evaluate effectiveness of herbicides in the laboratory or in situ, where potted plants could be used. However, these methods would not be useful for estimating biomass of intact plants in situ.

An appropriate technique for estimating biomass or biovolume of intact plants within the lake is echolocation or echosounders (Maceina and Shireman 1980, Maceina et al. 1984, Duarte 1987, Thomas et al. 1990). Echosounders are readily available, relatively inexpensive, and provide a method to rapidly cover large areas. The data can be used directly, as biovolume of aquatic plants (Thomas et al. 1990), or can be calibrated to biomass by some minimal biomass sampling (Duarte 1987). The major drawback to this method is that plant species cannot be differentiated or quantified, so additional sampling must occur to glean this information.

Sampling Design

One critical aspect of biomass sampling is sampling program design. Consideration must be given to distribution and placement of samples, selection of sample locations, size of samples, and number of samples taken. Although there is no single best method for any of the above factors, biomass sampling can be planned in a logical manner to maximize data quality and minimize time or effort required.

Studies of plant biomass in lakes have used sampling at both discrete representative sites, as in Carpenter and Gasith (1978) and Anderson and Kalff (1986), and widely distributed biomass sample sites around the lake's littoral zone (Anderson, 1978; Lillie, 1988; Engel, 1990). Both of these methods have limitations; neither is appropriate to all situations. If your lake is small enough, it can, in essence, be a single sample site. However, the size of most lakes dictates a limit on the number of locations to be sampled. Also, sampling around an entire lake introduces additional variation that may obscure other trends. If the chief objective is to evaluate whole-lake nuisance problems, or factors contributing to nuisance problems, then whole-lake sample distribution is appropriate, but may require far more samples than otherwise necessary. If the chief objective is to evaluate specific control techniques, then selection of specific study sites is more appropriate. At a minimum, a treated and an untreated site should be studied in this case, but the use of several treated and untreated sites to replicate your results is desirable.

Figure 1

Next, the method of selecting the location of samples need to be determined. In Fig. 1, some possible options for sample location are indicated. In the first, or regular distribution, samples are evenly spaced from each other. Although this may appear favorable, two considerable problems arise that weigh against this option. The first is that samples may unwittingly be placed along some regularly-spaced phenomenon, thus biasing the results. Secondly, and more important, most statistical tests assume that the samples were selected at random, and biomass will follow a normal distribution of points. Because of this crucial assumption, a regular distribution of samples is generally not performed. The next option is a completely random distribution of sample locations within the sample site. Although this scheme complies with the critical requirement of randomness noted above, sample locations may not evenly cover the entire study site. Despite this, many investigators utilize random distribution, including a critical review by Downing and Anderson (1985). The last scheme generally utilized is the stratified-random distribution. In this scheme, the study area is subdivided into equal areas, or strata. These divisions may be based on geographical areas, depth ranges, or other combinations of criteria. Then, one or more random samples are selected from each strata. This sampling scheme is useful for either whole-lake or stream sampling (Madsen and Adams 1988) or specific sites (Madsen and Adams 1989, Madsen et al. 1988). Since it both complies with the need for random samples, and evenly distributes samples around the geographic area or other strata, it is the method that I prefer in most instances.

The following discussion will center on attributes of the sample itself. Each sample constitutes the total weight of plant material harvested from a given unit surface area of the bottom. The unit area sampled, and thus the device that delimits this area, is referred to as the quadrat.

In most cases, samples will be relocated for each new sample event, because biomass sampling is by nature destructive. Also, most studies are looking at the biomass of plants in the natural environment. However, it would be negligent to totally neglect the use of permanent quadrats, and only discuss relocated quadrats. Permanent quadrats, in which the same particular sample sites are examined before and after treatments, may be quite useful in some instances, particularly in conjunction with the study of herbicide treatments (Hollingsworth 1978) or Biocontrol treatments. In these instances, the use of nondestructive biomass techniques should probably be employed in conjunction with untreated plots.

Early in the development of quantitative vegetation assessment methods, the optimal shape of the quadrat was a hotly debated item. Quadrat shapes ranged from circular, to square, then rectangular to long, thin strips. With the development of the strip quadrat, debates also raged as to how the quadrat should be oriented. Most have now forgotten this debate, and use whatever shape is convenient. Circular quadrats reduce the amount of edge to area sampled, thus reducing one source of error - which plants are in, and which are out of the quadrat. Circular quadrats could be made by cutting the end off of various-sized PVC pipes, or other inflexible tubes. However, most quadrats now used for biomass sampling are square, as they are the easiest to construct. Square quadrats can easily be constructed of lengths of small-diameter PVC pipe cut to length, and connected using right-angle pipe joints (Photo 2). Another quadrat design is constructed of galvanized sheet metal, approximately 30 cm in height, constructed as an open-ended box. This design is especially helpful in cutting through root tissue.

Photo 2

A less trivial subject is sample size. Traditionally, quadrat sizes were developed for terrestrial studies, and generally were oriented to studies of plant cover and frequency rather than biomass. No comprehensive quantitative study of aquatic plant biomass sample quadrat size was published until 1985, when Downing and Anderson (1985) published a thought-provoking, and time-saving, study on quadrat sizes for aquatic macrophyte biomass sampling. For the most part, quadrats in popular use were much larger than required for sampling aquatic plant biomass, and investigators tended to take too few samples.

In their study, Downing and Anderson (1985) sampled macrophytes at a range of biomass levels with a range of quadrat sizes. They found that quadrat size did not affect either accuracy or precision of the estimates, although the smallest quadrats were somewhat difficult to place in very dense plant stands. The number of samples needed to adequately sample low biomass stands was very sensitive to quadrat size, but relatively insensitive at very high stand plant biomass (Fig. 2). However, large quadrats take much longer to sample and sort. Even though small quadrats require more samples to be taken, they are easily and rapidly handled. Therefore, sampling with small quadrats was much more efficient at all biomass levels than using the large quadrats typically utilized (Fig. 3). Although this general recommendation might suggest the smallest quadrat conceivable, such as the 100 cm² quadrat utilized in the above study, other limits might occur. Aquatic plants often grow in clumps on a small scale, but appear more homogeneous as the scale, or quadrat size, is increased. At small quadrat sizes, the diver or sampler might be biased toward an opening in the canopy. A slightly larger sampler will avoid these biases. Therefore, I would suggest avoiding the smallest quadrat, and utilizing one approximately 0.1 m² in size for most applications, although they do require more handling time. However, they still save a great deal of time over the traditional 0.25 m² to 1 m² quadrats.


Figure 2	Figure 3

One final consideration in sample design is sampling frequency. Samples should be placed temporally to best capture the data sought. As in sample size and number, the fewest possible times to get the needed data should be used. Sample timing should either fit the experiment to be performed (i.e., twice if measuring the effectiveness of harvesting - once after, and once before the harvest); or, timed to follow a noticeable increase in plant biomass through growth. A survey of 69 plant biomass studies performed showed a bimodal distribution of sampling frequencies (Fig. 4). A large number of studies sampled biomass only once per year, as if only interested in annual variations, or twice per year, as in examining the before-and-after effects of treatments. Studies examining trends on a more seasonal basis tended to sample bimonthly, monthly or biweekly. Very few studies took weekly biomass samples, since it is unlikely that noticeable differences will occur from 1 week to the next. One interesting trend in examining the seasonal studies is that studies in the southern portion of the continent tended to use the bimonthly frequency more, while those in northern areas chose biweekly sampling frequencies. This trend may be due less to the growth patterns of plants than to the difficulty in sustaining a biweekly sampling frequency in the longer growing season of the south. A reasonable compromise on sample frequency is to sample more frequently while plant biomass is increasing (or decreasing) rapidly, and less frequently during periods of stable biomass.

Figure 4

Sample Processing

Once samples are collected, they must be appropriately processed (Photo 3). Sample processing is possibly the most time-consuming aspect of biomass collection, yet often gets the least attention. Poorly processed samples may compromise the usefulness of the data collected. Therefore, attention to the rigor of sample processing will ensure that time spent is not wasted. Sample processing can be separated into three stages: biomass sample separation and cleaning, drying, and weighing.

Photo 3

The individual biomass samples should be processed as soon after collection as possible. If they are not processed immediately, they should be kept cool or refrigerated, otherwise decomposition may ensue. Decomposition of samples not only affects the weight of the sample, but makes subsequent processing unpleasant. Samples to be processed should be thoroughly washed or cleaned to remove extraneous material. If root material is not to be included in the sample, underground material should be consistently removed from the shoots. Although not always pertinent to most studies, species should be separated at this point. Although it is more time consuming to separate samples into individual species groups, more information pertinent to management will be gathered if this is done. At a minimum, dominant species (particularly the most common target species) should be separated. If species separations are done, collect voucher specimens for future identifications and reference, and use suitable taxonomic keys for reference (e.g., Fassett 1957, Gleason and Cronquist 1963; and other local floras). For more information on appropriate taxonomic precautions to take, see the paper by Hellquist in this volume.

Drying of samples is the next stage in processing. Since ovens are expensive and have limited space for large numbers of samples generated by most biomass studies, some investigators use fresh (or "wet") weight, where plants are not dried in an oven. However, even these samples must be prepared before weighing. Surface water must be removed, either by drip-drying in racks, or using a spin-dryer. Spin-drying has been found to be more effective in removing water, producing more consistent results (Westlake 1965). Spin-dryers utilized for this purpose range in size from lettuce centrifugal dryers to the spin cycle of an old clothes washer. If fresh weights are used, spin-drying preparatory to weighing is mandatory. Even with spin drying, fresh material will dry in the air rapidly, giving highly variable results. Some studies have also utilized air-drying, but this technique also is highly variable, and not as consistent as spin-drying fresh material. In all cases, oven drying to determine dry weights gives much more consistent results, and is the preferable pretreatment before weighing.

For the determination of oven-dry weight, plants should be dried to a consistent weight - that is, until all water has been removed from the sample. This may take 24 to 48 hours, depending on the incubation temperature selected. The oven of preference is a forced-air oven, that forcibly circulates air around the samples. This ensures that dry air circulates around the samples, more rapidly removing water. The temperature at which drying is done depends on the oven available, but 105 °C has been a standard (Westlake 1965). However, many other temperatures have been used. In a brief overview of 56 papers, the most commonly-used temperature was 105 °C, but other popular drying temperatures were 60, 70 and 80 °C (Fig. 5). In most of the studies using a drying temperature of less than 105 °C, the plant tissue was to be used in subsequent analyses, such as for tissue nutrients or carbohydrate content. The bottom line is that dry weights should be done in an oven with at least 50 °C temperatures that has good air circulation, and samples should be dried for enough time that they achieve constant weight.

Figure 5

Weighing of plant samples should be done on an adequately accurate and calibrated scale. The weigher should ensure that the scale is not affected by air movement or motion. In recording weights of plants, take only a reasonable number of significant figures. The plants should be weighed directly from the oven, not left out overnight to pick up water from the air. Plants left out may pick up an additional 5 to 10% in weight.

In most cases, dry weight of the plant is adequate in comparing biomass values between lakes or different times of year, without additional processing. However, some lakes are so high in dissolved calcium that a significant proportion of the plant's weight might be encrusted calcium carbonate, or marl. A correction to dry weight can be determined by subsequent combustion in a high temperature oven at 550 °C, leaving the residual ash without the organic carbon. Ash content of plants typically runs from 15 to 25% of dry weight, but may be in excess of 50% in highly calcareous lakes and streams (Westlake 1965). "Ash-free dry weight" is reported in many studies (Westlake 1965), but is generally not needed for most studies.

Data Analysis

One aspect of data analysis to consider is whether an adequate number of samples has been collected. Downing and Anderson (1985) have indicated an approximation of the number of samples required based on the size of the sampler and the expected biomass at the sample site. I would also recommend that an assessment be made after the first samples are collected, or from a pilot project of a few biomass samples. Calculating an adequate sample size from a known mean and standard deviation from a group of samples requires assumptions that would make a true statistician cringe. However, for the purpose of avoiding a more likely mistake of undersampling at the risk of making an error of precision statistically, I suggest that the following equation be used as a crude estimate of minimum adequate number of samples on the total mean biomass:

N = S / (0.1 x X)2

where N is the minimum sample size, S is the standard deviation of biomass samples, and X is the mean of biomass. This equation is derived by solving the equation for standard error of the mean with the assumption that an error of 10% of the mean is acceptable (Wonnacott and Wonnacott 1977). Other methods exist for deciding on adequate sample numbers that are more complicated, but also statistically more rigorous (see paper by Spencer and Whitehand in this volume). The number of samples required will change over the season, as the biomass of the community increases (Fig. 6). Typically, more samples are needed early in the year, before plants become dense. Variance of biomass samples generally decreases as plant density increases. Variance also depends on the distribution of plants.

Figure 6

Caution is needed during statistical analysis of biomass data. Most parametric statistical tests, such as ANOVA and the student's t-test, require that data fit a normal distribution (Fig. 7). However, most biomass data sets are not normally-distributed, and may exhibit distributions more similar to other distributions such as the lognormal or Poisson. Biomass data from a dense bed of Myriophyllum spicatum in Lake George demonstrates this tendency. Both data for all sample dates (Fig. 8, top) and for one date at maximum biomass (Fig. 8, bottom) exhibited significant deviation from a normal distribution, as indicated by a rankit plot. Such data can be used in parametric statistics if it is mathematically transformed using a constant function. For instance, the data for one date (Fig. 8, bottom) fitted a normal distribution once transformed by log10. Other transformations used for biomass data include loge and the fourth-root (Downing and Anderson 1985). In addition, nonparametric tests that do not require normally-distributed data should be considered for the statistical analysis of biomass data. Such tests include the rank sum test for two groups, and the Kruskal-Wallis One Way Analysis of Variance for three or more study groups.


Figure 7	Figure 8

When expressing the mean for biomass samples, always also include some measure of variability to assist in comparing different groups. The standard error of the mean is particularly appropriate for this purpose. When reporting on data collection, the number of samples always should be indicated either with the methods, or in the discussion of the results.