(Video tape available)

Courtney opened the session by reviewing the process SEI has developed in these meetings. This has been a cooperative approach, which has focused on the development of tools for evaluating land management options and conservation strategies and for balancing these interests. We have thus far avoided advocacy positions, and tried to develop science-based consensus. So far this has worked well.

The Science Advisory panel was originally drafted to help oversee the Population Viability Analysis (PVA). This meeting is the first time that it has acted to advise on the other analyses. Tom Tuchmann has also asked that the panel review the work of agency staff- the panel is well respected by all parties, and recommendations have been valued.

Courtney then reviewed the science that is being carried out by various groups. Several analyses were identified at an early point in the program:

PVA - This work considers the long term effects of management options on the population of the Marbled Murrelet. In essence this is work that will advise on the potential for 'jeopardy' of various actions: will management options lead to the loss of the population or the species. This is therefore an analysis of potential impact; it is carried out by Applied Biomathematics (AB) and SEI (Courtney). This research has been the focus of two previous panel meetings and the workshop in March.

Habitat analysis - This work informs the PVA, and is essentially an analysis of the amount of habitat or conservation value lost and preserved under different options. This will guide parties in their assessment of 'take'.

The initial task of the habitat analysis is to develop consensus tools that can be used by all parties - in effect to craft a common language with which to discuss the potential effects of management.

Update on PVA
The first panel meeting set out the process, and guided AB and SEI on the development of models. The March workshop helped parameterize these models, and establish a range of alternative values for consideration. The second panel meeting allowed the panel to evaluate how far the science had proceeded, and to make appropriate recommendations. The panel also discussed the extent to which the PVA could help with negotiations; the panel discussed the advisability of deferred options that would allow science to reach stronger conclusions.

The basic demographic models (as presented at the 2nd panel meeting) have been taken as far as possible at this point. We need ancillary studies to reach completion, so that parameter values can be improved, and the range of uncertainty over the models reduced. These studies are mostly nearing completion. These are:

Habitat analyses (today's discussion)
Population trend data
Marine factors presumed to have effect (report here at meeting)
Oil spill data
Analysis of variability and correlation between populations (Hamer)

The PVA is continuing. We anticipate a report becoming available in January or February, using the best available data at that point.

The LIMBS model is a parallel process, using individual based models. Material was forwarded to the panel by the authors (some of whom are present), but will not be discussed at this meeting. If appropriate this may be discussed at future panel meetings.

Where are we in the negotiation process? Some positive developments have occurred. The federal appropriations appear secure. State appropriations are less sure, and may end up as a bond issue. There may also be Congressional hearings on the science results.

Negotiations are proceeding: several ideas have been put forward and sent out as alternative proposals. Three main proposals are under consideration. The MMCA (Marbled Murrelet Conservation Area) approach, put forward by USFWS and others, preserves a majority of the company's old-growth, and substantial areas of second-growth as recovery habitat.

The current company proposal is to defer, for five years, harvest of the eight largest habitat timber stands, while continuing scientific evaluation of murrelet populations (the August HCP).

This proposal derives, in part, from the panel's previous statements that hard decisions might be better deferred.

The 'Toggle proposal' also defers some decisions. A smaller amount of potential habitat (three stands) is set aside in interim, and one of these areas is then subsequently selected to be added to the reserve.

There is some distance between these proposals. We need tools, to help to decide which proposal best meet management objectives. It is not our goal to advocate one proposal over another - instead we should try to develop common language and tools for evaluating these proposals.

Paul Henson agreed with these statements, and emphasized that the political process is ongoing and urgent. It needs to be tied together with science; we are being urged to get the science done as quickly as possible, because we are at a critical stage in negotiation.

Habitat analysis
The work reported today is just one part of one part of the science. It characterizes what is habitat in the bioregion. It is important to state that this is just one part of a prioritization process that will identify where are the most valuable areas. This prioritization will include not just how many birds are detected, but also fragmentation, distance to ocean etc.; these other analyses are not yet well developed. The overall goal is a ranking of areas on a range of parameters; if we can agree on ranking, then the negotiations can focus on where the cut-off for protection will lie.

RSL analysis
CJRalph presented the analyses to date (see handout). The methodology in use was very different from previous (ARCO) studies of similar issues. There are many assumptions in the analyses (treated in report).

There are two basic metrics - straight habitat acreage; habitat weighting by bird detection levels (traditional approach in other animals). This is a part of a larger ongoing landscape analysis (from Coos Bay to Santa Cruz). Another year's work is projected.

A major assumption is that the number of birds detected in some way reflects the importance of stands - this relationship could even be linear (extreme point of view). Also we assume that sampling and temporal variability in detections at a station doesn't affect conclusions

RSL used three metrics to assign stations to particular habitat types, as explained in materials: Method 3, the proportional method (using GIS) - observations assigned to habitats within 200 m. Method 2 - used most abundant habitat. Method 1 - most complex and likely, using decision tree.

Panel discussion focussed on allocation to habitat types under method 1 - is it reasonable?. Data have been spot-checked in many cases.

Noon stated that this methods assume a priori understanding of what constitutes habitat - this involves big assumptions. Courtney pointed out that this is consistent with other approaches (e.g. PSG protocol).

The methods then take allocation at stations, and expand analyses to stand levels. Data are summarized within stands - not across stands. Stand boundaries are defined on both natural boundaries, and on basis of having enough stations.

Raphael and Kareiva questioned using a model of habitat related to station detection to scale up to evaluate stands. The model could be reliable or unreliable. What proportion of variation in detection levels is explained by the three approaches? Maybe some much better than others? Jodice suggested using estimates based on proportion of visits with detections.

1992-6, old-growth declined in southern Humboldt region by 1 ½ %, residuals by almost 40%, combined - about one third of habitat lost in time period. Little decline has occurred from 1996-97.

Nelson: we need to understand variability within a habitat type (e.g. OG, R1) in terms of actual habitat characters - i.e. platform abundance - do the categories capture much of the info we need? Should we extrapolate up without checking on the ground? Ralph indicated that some of these checks will be carried out - but other areas have been cut. Nelson also queried use of the 200 m radius circle - why? Ralph indicated that this value -could be changed to 50 or some other figure. Miller indicated that detection levels drop off beyond 200m. If we alter this distance we would change detection intensity, but we are not likely to alter relationship among stands.

Raphael: This method assumes detections are proportional to use. We also have to assume the probability of detections is comparable. Ralph: we tried to be conservative by analyzing stands across not just small areas.

Ralph: there are no big differences among the three metrics in the results obtained. If we look just at occupied detections (higher variance) - Headwaters increases in value by 5%. Large areas in State Park decline by 10%.

Noon: cautiously suggests that shouldn't feel comfortable that methods are similar. The methods rely on assumptions common to all methods, concerning detection probability 'little p'
There are three ways it comes in:
Radius - assumes is constant (using distance)
P is constant across sites
P is constant across time at a site

Methods are available to study these assumptions, using covariates in analysis, such as habitat variables. Results of such analyses will determine comfort level with habitat analyses. They might also allow us to determine which way there will be departures (e.g. closed-canopy underestimated, and stands which are heavily closed therefore under-valued).

Presentation of Paul Henson
There are important assumptions in detection methodology, and in applying the methods right now. Eventually we need some method of getting at Murrelet density - there are some studies in place (e.g. radar, tree-climbing, multiple observers at each station) that will get at that. How precise are methods NOW? Methods and data are important to the extent that they differ in allocation of birds to parks or PL lands.

Henson discussed the problems using the Ralph approach and concluded that it was not a reliable methodology to use at this time. He had concerns regarding Ralph's assumptions #'s 1, 2, 5, 6, and 7, but due to time constraints chose not to discuss these at this time. Instead, he focused on providing three examples of applying the Ralph method to see if the results were biologically reasonable.

In example 1, Stand 2 has 546 acres of residual habitat that appears to be of relatively low quality. It had no occupied detections. Yet it had the highest RBV of all stands except for HW/EHS and HRSP. It was higher than all the stands with unentered old growth, such as Stand 3 (Bell Lawrence). Stand 3 has 1030 acres of old growth and had multiple occupied stations, yet it had a lower RBV than stand 2, as did many other higher quality stands. In Henson's opinion this RBV result does not make biological sense.

Example 2 reveals a similar pattern. Larabee Cr. (Stand 20) has 427 acres, mostly Douglas fir, and is the farthest inland of the stands.
It had only one occupied station, yet it ranked out higher than many other higher quality and larger stands with more occupied detections.

In Henson's third example, he compared HRSP with HW. HRSP had approximately 1600 acres of occupied old growth. HW had
approximately twice that (@ 3200 acres). Yet because of extrapolation of "bird value" to large areas within HRSP that did not have occupied detections, HRSP receives a much higher RBV than HW.

In Henson's opinion, these three examples illustrate that the
detection method can give misleading results by according high value to relatively lower value stands, and vice versa. He suggested that
some reasons for these misleading results may be:

1. that total detections is not really describing the phenonmenon of
interest (i.e., nesting murrelets - see Ralph's assumption #1);

2. there are no precision measures with the data,

3. similarly, there are no hypotheses posed and tested; it is simply
assumed that differences in mean values used to compare populations (i.e., stands) are absolute and therefore different

Henson concluded that it was not a reliable methodology to use at this time to weight the relative value of stands or to estimate the amount of take that would occur under an HCP. The Service recommends using categorical data that describe occupancy and relate this to proximal habitat attributes.

In summary, RSL studies confirm analysis that earlier suggested that the habitat in the park is largely along Bull Creek and Eel River. The remaining habitat up slope is of limited value; there are some birds nesting in there, but we are unsure how many - probably far less than extrapolation methods suggest. There is certainly some relationship between detections and density, but we are unsure what relationship. Lack of precision estimates makes the data hard to evaluate or accept now - normalizing, and transforming the data may help. Before that, however, we need to know if the data are even measuring what we want (nesting density) - radar and nest searches could help. The RSL method is one method of how to rank stands - but it is premature to apply these results in decision-making.

Courtney raised the issue of how to proceed - if these metrics are not comfortable to Henson, what should be used? We cannot avoid making decisions - what is best available metric?

Timeframe for decision-making
There was also discussion on the issue of timeframe, as set by the congressional clock. Frank Bacik was not sure that we even have two months - is a snapshot possible now? The form of the HCP murrelet conservation strategy has to be agreed upon in December or January for the process to meet timelines. The panel should be encouraged to make longer term suggestions - some options allow management decisions to be deferred depending on outcome of science over next several years (as recommended by panel at last meeting). If , however, we are not going with deferred options, we need as much information now as is possible.

Tom Tuchmann: We have till March 1999 to complete process. Working back from this, we have till January or February to make an agreement. We need perspective on what is the range of reasonable management options that science allows. What are areas of agreement and disagreement?

Presentation by Reid

Tom Reid reviewed and presented different management alternatives, their economic costs, and geographic distribution of conservation (see handout). He questioned whether we need ranking - most alternatives don't need rankings in order to evaluate them (e.g. MMCA). Also some stands fall out as obviously more or less important (Headwaters, Park) - do we need greater accuracy? Other stands are more or less equivalent; it will be hard to rank these usefully. It is not even that important given the scale of question we are considering.

What metrics can we use to guide us among alternatives? One possibility is to look at a range of metrics, and to use the ones that we are most comfortable with. Are there even dramatic differences among metrics? Maybe they all give the same answer.

Noon: We should seek the configuration which maximizes objective function (conservation value) given economic constraints.

Courtney: other criteria to be added to objective function:
RBV
Fragmentation edge effects
Stand size
Distance to other reserves
Distance to ocean
Recovery Habitat
Species composition
Quality of habitat

Reid and Gaither presented an analysis of ranking of stands using fragmentation as the criterion. 50 m of the stand is defined as edge; 300' outside this is set up as buffer. This is an exercise - not an analysis that is to be used or where assumptions will be defended . The focus is on the method - is it useful? It's an alternative method, not using detections, therefore avoiding the problems associated with detection methods.

Noon: dollar maximum is the constraint that should be used. It is easy to write a simulation model that looks at all the combinations, and sees which combination comes out the best in terms of the objective function.

Murdoch: there are no real data on 'edge' in the Limbs model.

Cody: this method defines the issues in such a way that define small stands to have little value. The value of core is set at four times that of edge - this is an arbitrary value, not well supported by data.

Raphael: The assumption drives the outcome. Research work in WA and OR doesn't support the idea of the edge effect - human habitation may be more important. The risk of predation is the driving question, not edge. Edge effect is an open question. Cody - even possible that core with edge is accessible.

Gaither: assumptions need to be tested I agree, and there are limitations in the method's justification - but these principles are being put in place - e.g. in MMCA, because we need to make decisions. May be some data coming (Nelson).

Burger: both this approach and that of RSL are deterministic not stochastic. We need to compare analyses for congruence - also to look at sensitivity to adding individual stands - are there thresholds where one stand makes a big difference?

Panel recommendations
These are the logical and essential steps that are necessary for a spatially explicit HCP. This is specifically directed at the next few weeks and months.

1. RSL should not use 'all detections' as the response variable
2. Occupancy is the most biologically relevant variable - use that.
Occupied detections should be used (not nests or other metric based on occupancy). Circling not discussed for possible inclusion
3. Relate the occupancy data to site-specific attributes - this should
be done at the level of the individual survey station.
RSL should explore variation in occupancy on the basis of station
attributes, using regression.
4. Analysis should then scale up from the station level to stand level.

Analysis should examine variation in occupancy as explained by variation in habitat attributes. It should explore variance in per-visit detection likelihood as a function of site attributes - this will get at issue of whether p is constant across time, and across sites. There are two general ways to do this: 1. Using subset of data where there are multiple visits to sites, variable distance methods will determine whether p varies as a function of habitat type. If there is no difference in detection functions with habitat, then safe to make assumption that p is constant across habitats. 2. The alternative approach is multinomial logistic regression; to estimate the probability of making a correct call on occupancy as a function of habitat variables. This gives a station specific per visit detection rate, and hence your likelihood of getting a false negative as a function of habitat type.

Assuming that we solve all these problems - we now have a biologically meaningful metric. We should have looked at all the assumptions that are inherent in this metric e.g. RBV, tested them, and made adjustments on the basis of biases. The next step is a straightforward optimization economic analysis where costs are constraints. Given the likelihood that there are disagreements over what the economic costs really are, we should look at the marginal increase in the metric as a function of economic cost -this is the easy step in the analysis.

Gaffney: what is the basis for bringing in economics, in science or law. Noon: this is not a scientific argument. We are making decisions in the face of uncertainty, and estimating the utility of possible outcomes. Estimating utilities is not a scientific process - it is a value based process. Detrich: we could take this idea, but also add the statutory constraint, so we could do this analysis and see how close we are to that constraint.

Noon: HCP should be an experiment, updated through monitoring - if birds not behaving as envisaged, we need to revisit model.

There was general discussion of the possibility of source-sink dynamics - but no data are readily available. If such dynamics are found, they have implications for release of small stands, and growth of buffers (mitigation).

Kareiva: Our suggestions are ways to examine data to evaluate the proposals that have been put forward. There is lots of value in this data set that can be brought out.

Courtney: even after all these metrics are in, there are still other weightings we might want to add. So how should we do this ? Cody: we would like to use the number of fledged Murrelets. We would then be able to project over next 50 years to look at recovery habitat. Murdoch: this is our idea of what to do with existing data - but there is nothing to stop you building in experience or results based on data. It important to distinguish this proposed analysis - what the data tell you - from 'biological insight'. You should add in these other metrics to weight alternatives.

There was general discussion on how long this new analysis would take to do. Ralph: the work is feasible and valuable - but the entire list is not doable in 4 weeks. Murdoch: someone could do logistic regressions and analysis of p in 4 weeks - but not the economic cost-benefit analysis. Bacik: there is willingness on the part of the company to make this happen. Tuchmann: also wants to see it happen.

There was discussion on the productivity metric. The panel's response was to carry this out on a longer timescale, when there are sufficient data. It can be brought in as a weighting factor. It can also be turned around to ask: how big a difference would there have to be between edge and center to alter results? - we don't have enough data yet to answer this question.

Courtney asked: are the occupancy data the best available scientific information - should decision makers make decisions on that basis? Panel: Yes in the short term, and in terms of take - in terms of sustainability, long-term value might include recovery potential.

Discussion of value of detection versus occupancy data

Burger: work in BC correlates habitat variables with occupied detections or total detections. There is no relationship with all detections, but a good, consistent relationship with occupied detections. He has 7 and 3 years data in two sites - and therefore has some confidence in metric. Burger endorses the panel approach of using occupancy.

Jodice: he is carrying out intensive surveys - within year variability is high. Coefficient of Variance across a summer is 40 to 130% - but if we look at just occupied detections, CV comes way down. If we look at areas where there are known nests, surveys make correct decisions >90 % of time. The surveys tend to get the call on occupancy status right - the variance in number of detections is much higher. Jodice endorses the idea of working on occupancy. On basis of four to six visits, numbers of detections alone could lead you to wrongly suggest that one stand had more birds than another stand.

Murdoch: CJ has wonderful data for getting at what's happening in a stand - Kim Nelson has beginnings of data on productivity. Raphael suggests that variables from logistic regression be forwarded to Kim so that she can apply them in her studies.

Meeting closed at 4 pm