factorial-design.html
It is not so surprising that there are so many factors have been proposed to control the population of AFA in Upper Klamath Lake, but that all of them seem to be simultaneously important. This cyanobacterial strain seems to be pushing the limits against photosynthesis, nutrition, inhibitors and predation. This is truly a case of a weed species, a species that if it were commercially valuable in its own right would be adding value to the ecosystem rather than detracting from the value. In the Upper Klamath Lake ecosystem, AFA is causing damage because it degrades water quality and disrupts the lake environment for other species such as native fish.
The list of individual factors which have been proposed to control AFA include:
light availability - high turbidity and high algal absorption
nutrient availability - phosphorus loading, phosphorus recycling from sediments, iron, nitrogen and nitrogen fixation
predation - variable predation by zooplankton on the filament and flake forms
humics and UV - humic acids released from in-lake and watershed marshes inhibit AFA growth, this inhibition may depend on UV light,
temperature - temperature optima for growth
There are several models for control that are being considered. Some of these individual factors have been proposed to be a major, overriding control (such as P loading) and others have been combined into multi-factor models (such as the growth dynamic explanation that invokes early season inhibition by humics that allows predation to keep the population from reaching a critical point). It is important that we compare these models with a degree of, what Scheffer (1998) calls "modesty". Scheffer considers it very possible that
"the fact that the modelled mechanisms can be shown to operate in the field is not a sufficient basis either for concluding that it offers the appropriate explanation in that specific case as the modelled mechanism may well be acting in concert with other, possibly more important ones." (page xviii)
Before we consider a set of experiments that are designed to test the which of these factors is "most important" in different circumstances, let us consider the broader problem of what the different models mean, what evidence we would gain by testing each model, and how the results could be used in lake management decisions. For the purpose of this consideration, the phosphorus loading model and the multiple-stable state models will be compared. There are other models that could be introduced into this discussion, but the comparison of even just these two models is complicated enough. Each model has several key features that will be stated in an over simplistic way here. The point of this comparison is to demonstrate how the choice of model leads to experimental approaches and has management implications.
The P-loading model is based on a large number of observations in other lakes that have show an empirical relationship between P-loading and chlorophyll. Specific lakes are put in context of this general relationship. In this approach the overall loading of the lake is important and this drives the decision to look at the average water chemistry and biology as accurately as possible. Stations are usually chosen in main lake locations, and these locations are compared for consistency. Some times, stations that are continually outliers in the data set are abandoned, justified by the interest in whole lake processes. The main evidence is presented in the form of lake averages. This scientific approach has been used successfully to provide evidence that decreased P loading will lead to decreased lake algal chlorophyll and has supported management decisions to decrease P-loading.
The multiple-stable-state model for shallow lakes doesn't refute that P-loading is important but addresses the path by which the lake will shift from turbid to clear (and back). Instead of focusing on the bulk and average lake chemistry and biology, this model looks for positive and negative feedback features of the lake ecosystem that tend to stabilize different algal communities. A simple example of this model is that shallow lakes that have a large amount of emergent vegetation are stable because of feedbacks between fish, zooplankton, and algae that limit algae and P recycling, and that this same lake could also be stable in a configuration that had no emergent vegetation, high algae and fish that continually disturb sediment, leading to higher P recycling rates. The feedback cycles can be thought of as mechanisms that will push the lake toward each end of the continuum depending some threshold condition. It is much more difficult to provide evidence that there is a threshold and what it might be in any given lake. If there is a threshold, however it means that simply reducing the P-loading may take an even longer time to have an effect that calculated from average lake residence times and, that other factors and situations may help the lake jump from one stable-state to another. Multiple stable-states are considered to be an important consideration for lake managers. Carpenter (2001) states
"using economic optimization criteria, it can be shown that even if the probability of multiple states is low, the implications for policy choice may be profound. The heavy impact of multiple state on policy results from the high cost of restring the ecosystem if it is shifted into and undesirable stable state."(pg 358)
In comparison to the P-loading model evidence - Pelican Bay /transects
management - heterogeneity of sites
Both models describe the situation in UKL to some extent. Neither model is right or wrong. Instead of validating one and disproving either model, it seems that the crucial task in the management of UKL and Agency Lake should be to study the lakes with approaches that provide us with the most useful information, i.e. information that can be used to make management decisions. The ecological community has been wrestling with problems similar to this for decades and have come up with suggestions for approaches that rely on multiple models, strong inference and multiple scales of inquiry. One useful tool in this inquiry is to employ a factorial design to look at potential controlling factors over one the processes at one scale. However, these factorial designs need to be incorporated into a set of experiments across different scales.
The comparison of these two models is really the same as looking for an alternative stable state. Instead of a linear regression of chlorophyll vs. P-loading (Figure 1), we would be looking for evidence that there are two stable states (Figure 2). The implications are profound as mentioned above. The linear regression implies a decrease in P-loading would drive a decrease in Chla whereas the multiple stable state model would imply that we need to decrease the P-loading to a very low level in order to facilitate the community switch which has clearer water as one of its attributes.
The tools for discriminating between these two general models include the observations of large scale perturbations (either experimental or natural), experiments that focus on particular scales and processes, analysis of long term data sets and models. The large perturbations are essential for determining the realm of possible responses whereas the individual experiments at any one scale are essential for providing information that can be used for prediction and testing (Kitchell et al 1988). In this context, factorial experiments are used to determine rates of responses, the shape of the response curve, and the interactions between system components. This means that the factorial experiments assume that the endpoints of the experiments are within the realm of possible lake reactions, and then characterize the processes that lead to that change. Factorial experiments are, necessarily, at a limited time and space scale.
Even though it is impossible to do a whole lake experiment with Upper Klamath Lake, there are several natural perturbations that are on the scale that is needed to help test for alternative stable states. The first perturbation, as mentioned in our original proposal, is the emergence of the lake from the winter. During this time there are large scale changes in temperature, ice cover, water mixing, sediment suspension, algal growth conditions, and predation. This annual perturbation sets the scene for the rest of the growing season. Other perturbations could be a particularly low or high water year. Changes in the circulation, input, output and lake depth that might accompany these abnormal years could have a large effect on the overall response of the lake's communities.
A good scientific approach would embed factorial experiments in the context of larger scale processes and the comparison of at least two models for community response. In the case of UKL, the factorial experiments would be very valuable to assess the rates of response of different communities during the different stages of the annual cycle.
We propose the following experimental design to take place over three or more years.
Transects between the marsh areas and mid-lake stations are a crucial piece of work that will put factorial experiments in a spatial and temporal context. Water from the lake or marsh endpoints. The transects would be taken at multiple times during each year. It is crucial that these transects include a bay that has an aberrant behavior as predicted from one or the other model, specifically Pelican Bay is a crucial piece of evidence for evaluating the multiple stable state model.
Enclosure experiments during different times of the year to examine the response of the natural populations over a period of about a week to ten days maximum. These enclosures should hold about 1000 liters and both include and exclude the sediments.
Factorial experiments would be carried out in 5 to 10 liter containers over three to five days. The convenience of this size of container would allow multiple conditions (up to 30 simultaneously) and extensive monitoring. We propose that each container be sampled for chemical and biological parameters on a daily basis and that each container be continuously monitored with either a dissolved oxygen or fluorescence data-logger.
The analysis of the long term data set as described in our original proposal would be used to help evaluate the likelihood of the two lake models.
The schedule for this work would start with the large scale processes as examined by the transects and enclosure work over the entire season. This work would help establish the realm of in-lake possible states. During the second and third years the transect work would continue but might focus more on specific events, such as particularly high or low flows coming out of input rivers or marshes. Factorial experiments would be added to complement the enclosure work. These experiments would manipulate the pH, humic content, and temperature of sample of the lake community.