Alternative	Existing BMA area (km²)	Additional core area of 105 subwatersheds (km²)	Additional watershed area selected (km²)	Total area (km²)	# of watersheds selected by BMAS	Total suitability index
Coarse-filter	9,693	9,145	19,240	38,078	185	582.2
Fine-filter	9,693	9,145	50,779	69,617	501	2,980.3
Integrated coarse- and fine-filter	9,693	9,145	56,353	75,191	567	3,245.7

Figure 6. Map of the coarse-filter alternative

Representing the 359 rare species and plant associations that are vulnerable required substantially more area (501 subwatersheds, Figure 7) because the locations of these target elements tend to be widely scattered. Whereas the area of the subwatersheds containing these rare elements was approximately 160% greater than the area for representing GAP alliances, the suitability index was more than 400% greater. Thus even the most suitable subwatersheds containing rare elements (mean index of 5.9) was much less suitable than those for alliances (mean index of 3.1). This result indicates that there are many more options of where alliances can be represented than the relatively few sites available for rare elements. This lower overall site suitability should also not be surprising because many of the rare elements are only found in isolated habitat fragments in cultivated areas.

The integrated coarse- and fine-filter alternative found some efficiencies in representing both sets of elements (481 vulnerable elements) simultaneously (Figure 8). The total area and number of subwatersheds selected were substantially less than the sum of the coarse-filter and fine-filter alternatives taken individually. Mean suitability index (5.7) was similar to that of the fine-filter sites, however, because representing rare elements does not leave much flexibility without selecting very low suitability sites. In fact, 273 of the 359 rare target elements occur at no more sites than required by the representation goals. Thus all their occurrences had to be selected, making 321 of the 567 subwatersheds "irreplaceable."

Figure 7. Map of the fine-filter alternative

The policy question about the relative role of public and private lands in assembling a conservation portfolio was addressed by comparing the results of the integrated coarse- and fine-filter alternative to a variation which strongly favored public lands. The public lands alternative was only slightly less efficient than the baseline alternative. Total area selected was 57,173 km², an increase of 1.5%, and actually selected one less subwatershed. However, the two alternatives shared 530 planning units in common, with only 36 unique to the public lands alternative and 37 unique to the baseline. Further, while the baseline had a public/private mix of 47.5/52.5%, the public land alternative shifted this only to 49.1/50.9%. These findings suggest that even with a very heavy weighting to favor public lands in the BMAS solution, the outcome was quite similar to the baseline. Evidently, the rarer elements occur primarily in habitat remnants on private lands. This makes their planning units irreplaceable in any portfolio with the given representation goals. Hence, for the Columbia Plateau there is relatively little flexibility in achieving this set of goals.

Although reserve selection algorithms produce solutions that may not be the best when scrutinized using better data or additional criteria, they have proven to be powerful indicative tools and very effective at facilitating understanding, group planning and negotiation (Pressey et al. 1995). The planning approach applied in the Columbia Plateau ecoregion guarantees that specific representation goals are achieved. While these goals may only be estimates of the effort needed to conserve biodiversity, they at least are explicit in the prototype process and can be modified in future revisions if new information becomes available. Scoring and expert opinion approaches, in contrast, are good at identifying high quality sites but do not attempt to represent all varieties of biodiversity. They are also labor-intensive and therefore do not lend themselves to exploration of alternatives (e.g., changing the scoring system). We believe that a BMAS type of modeling technique and the expert approach are complementary and can be used effectively in tandem. The expert process integrates knowledge of site quality that is not easily incorporated into the modeling approach. Specifically we recommend:

1. TNC use expert panels to identify a set of core areas that must be protected in any portfolio, followed by the BMAS or a similar model to select additional planning units as needed to meet the representation goals.

The BMAS model can be formulated to cluster additional sites around these core areas and to give extra weight to non-core areas that were nevertheless identified by some of the experts. Alternatively, instead of assigning core areas, the sites identified by many expert groups could be given very high suitability weights to make it more likely, but not guaranteed, that the model will select them.

The BMAS modeling approach is a useful exploratory tool for evaluating the consequences of alternative choices in targets, representation goals, suitability factors, and policy questions. The BMAS model is able to sort through large, complex, multi-dimensional data sets to maximize a balance between efficiency and suitability, which would be beyond the capabilities of a human analyst. We emphasize two critical points, however.

2. The set of planning units selected by the model should not be accepted on faith as an ideal portfolio. The model solution only forms a starting point of a portfolio. The planning team members must still apply their own intimate knowledge of specific sites to refine the portfolio, using personal knowledge not explicitly in the GIS database nor incorporated in the BMAS model. Ideally, more testing of model assumptions and parameters will be undertaken to understand their implications and trade-offs.

3. The cartographic representation of the set of planning units (i.e., subwatersheds) is not a map of the precise boundaries of conservation sites. The locational accuracy of the biological data is imperfect and so some target elements may actually occur in adjacent planning units. This selection error could be remedied during the design/implementation phase. The use of standardized planning units is just an artificial convenience for modeling purposes. Therefore the set of planning units merely indicates general locations where appropriate management strategies can be applied. In the extreme, a subwatershed may be selected in which the only strategy required is to monitor the status of a single target element occurrence at a small site within a planning unit. In other words, displaying a portfolio from this prototype planning process as a set of large planning units risks overstating the magnitude of the conservation agenda and alarming other stakeholders. We strongly urge caution in how TNC publicly portrays any portfolio derived in part by BMAS modeling based on subwatersheds or other such planning units. Noss (1996), in commenting on this question, recommended that network proposals be presented to policymakers and the general public as a staged sequence from the current system, through intermediate stages, to the ultimate network. He felt that presenting the ultimate vision was worth the risk because it may stimulate others to conduct more detailed analyses in the design phase.

The basic conservation planning approach, including the BMAS model, has now been applied in the Sierra Nevada Ecosystem Project for the U. S. Forest Service and the Columbia Plateau ecoregion for TNC. Both of these ecoregions were data-rich study sites, having been the focus of major federal assessment efforts. Large GIS databases had already been assembled for most of the essential spatial data layers such as suitability factors and biodiversity target elements. Gap analysis had recently been completed for both these regions which also provided important data. As GAP is implemented in more regions, and as public-domain GIS layers become more commonly available, lack of data for other regions of the nation should diminish as a limitation. Perhaps more significantly, both regions are largely in public ownership with relatively limited area converted to development and cultivation. Thus there remain substantial opportunities for proactive conservation action. Because of the similarity of these two test sites, it remains untested which circumstances would preclude the effective use of the BMAS model. Obviously, in an extreme case of a highly altered ecoregion in which all remaining fragments with rare elements were known, all such sites would be in the portfolio and a more formal modeling approach would add nothing useful. But what circumstances define the threshold at which the BMAS model becomes an effective planning tool? Therefore we recommend:

4. TNC undertake additional pilot studies in ecoregions across a range of management situations, including more highly altered landscapes in the eastern half of the nation.

Perhaps the single greatest failing of the modeling approaches to date is that they tend to treat sites as collections of species or communities without accounting for the viability of the biota at each site or over the set of sites. The goal of representation is only an approximation of the true goal of maintaining viable populations of all native species. Similarly, management status only approximates the actual underlying threats to biodiversity. A second, related problem is that none of the current selection models addresses the problem of the spatial layout of the sites. Since sets of sites may all be influenced, potentially in different ways, by the same large-scale ecological processes (e.g., fire, floods), consideration of spatial layout may relate to the long-term viability of the sites. Another critical issue is whether conservation planning solutions developed at the regional scale using spatially and biologically coarse data provide reliable guidance when viewed from a more local perspective with better data and understanding of ecological processes and management concerns. Our final recommendation is:

5. TNC should conduct or sponsor research to address three issues we consider central to ecoregion-based conservation, namely: 1) development of approaches and techniques for assessing species and community-level viability under a particular conservation scenario, 2) development of improved, multi-objective models for identifying the best set of sites within a region for meeting the stated conservation goal while addressing viability and spatial configuration, and 3) testing of regional viability measures and siting solutions against more detailed information on biotic composition and ecosystem processes to establish the relationship between regional and local conservation measures and approaches.