© 2006 Nature Publishing Group
follows a species–area relationship. Last, we have assumed that the
amount of resources invested in a region is directly proportional to
the p robability of species persisting. These assumptions can be
relaxed w ithin the framework that we present.
There is a need for computationally feasible and understandable
algorithms that can deliver near-optimal solutions for large con-
servation resource allocation problems. The simple heuristics that we
explore were developed to solve a properly formulated problem,
perform surprisingly well relative to the optimal SDP solution, and
are superior to simple ranking approaches. Minimizing short-term
loss most closely approximates the optimal allocation schedule and
maximizing short-term gain is close to optimal despite ignoring
threat, although it underperforms if threat levels are very different.
Under extreme uncertainty, maximizing short-term gain is the most
risk-averse approach as it provides a buffer against an uncertain
We recommend that conservation organizations maximize short-
term gain, unless the regions of concern have similar endemic species
richness and ver y different levels of threat. In such circumstances,
resource allocation should minimize short-term loss. We argue that
conservation investments should be evaluated as any investment is
evaluated: that is, with a clearly deﬁned objective and an assessment
of how well the returns from the investment meet this objective.
Responsible conservation organizations and international agencies
should consider embracing a dec ision theoretic approach when
scheduling the allocation of conservation resources.
The ﬁrst step in formulating the conservation resource allocation problem is to
deﬁne a quantiﬁable objective. Our objective is to maximize the number of
endemic species remaining across all regions when habitat conversion ceases
because there is no unreserved or unconverted land (se e Supplementary
Methods ‘Problem formulation’). We assume that, for each region, the number
of endemic species conserved per unit area is a monotonically decreasing
function of the area reserved. Therefore, our conservation returns in each region
diminish with increasing investment. We model this relationship using a species–
(see Supplementary Methods (ii)). In principle, any relationship
between area and endemic species conserved could be used. The next step in
formulating the problem is determining what actions are possible, in this case
what fraction of the budget to allocate to each region each year. Our decisions are
thus constrained by a ﬁxed annual budget. Although we recognize that funds for
conservation can be directed towards many kinds of activity (for example,
restoration programs, the purchase of forestry concessions and species recovery
programs), we focus on the acquisition of land for reservation. We assume that
each land parcel can be classiﬁed as reserved, available for reservation or
converted (anthropogenically altered and assumed no longer suitable habitat
for the endemic species of the region). Threat is modelled by assuming that a
constant proportion of available parcels in each region are converted each year.
To incorporate the uncertainty associated with parcel loss
tary Methods (iii)), conversion is represented as a stochastic process with a
binomial distribution. We estimate the cost of reservation in each region using
(see Supplementary Methods (iv)).
We use SDP and two myopic heuristics to determine how many parcels to
reserve in each region each year. We compare these results to a random
acquisition process. Once the optimal solution is obtained, we forward-simulate
the resulting a cquisition schedule 10,000 times for each parameter set to
calc ulate the expected number of species conserved. The solutions found
using SDP are optimal in the face of uncertainty
. Owing to an exponen-
tially increasing state space, however, ‘the curse of dimensionality’ limits SDP to
problems with few regions. The maximize short-term gain heuristic selects
parcels for reservation that result in the greatest increase in the number of
endemic species conserved. This heuristic ignores threat and is myopic, con-
sidering only the short-term future when selecting the next parcel to reserve and
not all possible futures as the SDP algorithm does. The minimize short-term loss
heuristic is also myopic: it selects parcels that will minimize the expected loss of
species from the system in the next time step (see Supplementary Methods (i)).
Received 27 September; accepted 24 October 2005.
1. Possingham, H. P., Andelman, S. J., Noon, B. R., Trombulak, S. & Pulliam, H. R.
in Research Priorities for Conservation Biology (eds Orians, G. & Soule
225–-244 (Island Press, Washington DC, 2001).
2. Pimm, S. L. et al. Can we defy nature’s end? Science 293, 2207–-2208 (2001).
3. Dalton, R. Biodiversity cash aimed at hotspots. Nature 406, 818 (2000).
4. Myers, N., Mittermeier, R. A., Mittermeier, C. G., da Fonseca, G. A. B. & Kent, J.
Biodiversity hotspots for conservation priorities. Nature 403, 853–-858 (2000).
5. Stattersﬁeld, A. J., Crosby, M. J., Long, A. J. & Wege, D. C. Endemic Bird Areas
of the World: Priorities for Biodiversity Conservation (Birdlife International,
Cambridge, UK, 1998).
6. Olson, D. M. & Dinerstein, E. The global 200: priority ecoregions for global
conservation. Ann. MO Bot. Gard. 89, 199–-224 (2002).
7. Halpern, B. S. et al. Gaps and mismatches between global conservation
priorities and spending. Conserv. Biol. (in the press).
8. Bellman, R. Dynamic Programming (Princeton Univ. Press, Princeton, NJ, 1957).
9. Wilson, K. A. et al. Measuring and incorporating vulnerability into conservation
planning. Environ. Manag. 35, 527–-543 (2005).
10. Orme, C. D. L. et al. Global hotspots of species richness are not congruent with
endemism or threat. Nature 436, 1016–-1019 (2005).
11. Balmford, A., Gaston, K. J., Rodrigues, A. S. L. & James, A. Integrating costs of
conservation into international priority setting. Conserv. Biol. 14, 1–-9 (2000).
12. Moore, J., Balmford, A., Allnutt, T. & Burgess, N. Integrating costs into
conservation planning across Africa. Biol. Conserv. 117, 343–-350 (2004).
13. Frazee, S. R., Cowling, R. M., Pressey, R. L., Turpie, J. K. & Lindenberg, N.
Estimating the costs of conserving a biodiversity hotspot: a case-study of the
Cape Floristic Region, South Africa. Biol. Conserv. 112, 275–-290 (2003).
Figure 3 | Comparison of conservation performance of heuristic algorithms
to the optimal solution.
Two hypothetical regions are compared that differ
in relative threat and number of endemic species. The performance of
each heuristic is the percentage of endemic species not reserved relative to
the optimum solution. a, Performance of the heuristic that maximizes
short-term gain as compared with the optimal SDP solution. b, Performance
of the heuristic that minimizes short-term loss as compared with the optimal
SDP solution. The heuristic that minimizes short-term loss outperforms the
heuristic that maximizes short-term gain for most parameter sets.
Figure 4 | Proportion of ﬁve priority regions from Southeast Asia reserved
The average proportion of the total area reserved is
determined using two resource allocation heuristics. a, Heuristic that
maximizes short-term gain. b, Heuristic that minimizes short-term loss. The
annual budget is US$ 1 million and assumes no pre-existing reserves. In
both cases resources are initially allocated completely to Sulawesi and
investment is directed to southern peninsular Malaysia only after
investment has occurred in Sumatra, Java/Bali and Borneo.
NATURE|Vol 440|16 March 2006 LETTERS