Dynamic Resource Allocation in Conservation Planning

AAAI, 2011.

Cited by: 37|Bibtex|Views144
EI
Other Links: dblp.uni-trier.de|academic.microsoft.com
Weibo:
We proved the surprising result that this simple opportunistic policy is competitive with a clairvoyant solution that is informed in advance of the budget in each timestep and when each patch will be available

Abstract:

Consider the problem of protecting endangered species by selecting patches of land to be used for conservation purposes. Typically, the availability of patches changes over time, and recommendations must be made dynamically. This is a challenging prototypical example of a sequential optimization problem under uncertainty in computational ...More

Code:

Data:

0
Introduction
  • One key challenge in computational sustainability is to allocate resources in order to optimize long-term objectives.
  • An archetypal application is conservation planning: managers recommend patches of land for conservation in order to achieve long-term persistence of endangered species.
  • In this and similar applications, the authors typically have to make decisions over time: Financial resources are periodically made available and should be used effectively.
  • The authors prove a surprising fact: Under some natural conditions, a simple policy, that in every round of the decision making process opportunistically allocates the budget given the current reserve and current resources, attains a performance which is competitive with the optimal
Highlights
  • One key challenge in computational sustainability is to allocate resources in order to optimize long-term objectives
  • Our intent during the workshops was to formally capture this uncertainty, via inter-expert variation, so that it could be reflected in the predictive patch dynamics models, and conservation recommendations could be obtained that are robust to this uncertainty; to do so we used a modified Delphi process for expert elicitation (c.f., Vose, 1996)
  • We vary the budget from 0 to 60 km2, and compare the optimized reserves with random selection, as well as selecting patches according to decreasing area
  • We considered the problem of protecting rare species by recommending patches of land for conservation
  • In order to cope with changing availability of patches, we proposed an opportunistic policy for dynamically making recommendations
  • We proved the surprising result that this simple opportunistic policy is competitive with a clairvoyant solution that is informed in advance of the budget in each timestep and when each patch will be available
Methods
  • The authors generate contiguous candidate patches from the parcels by a region growing process, which picks a random parcel as seed, and iteratively grows the patch up to a random size.
  • This growth process is randomly biased to avoid complex boundaries.
  • Instead of using the algorithm described by Sviridenko [2004] for solving Problem (2), the authors use a faster algorithm of Leskovec et al [2007] that carries theoretical guarantees
Results
  • 4.1 Reserve Design Case Study

    The authors are conducting a computational study in collaboration with the US Fish and Wildlife Service Washington Office, in Washington State, USA.
  • The target species are Taylor’s checkerspot (TCS; Euphydryas editha taylori), Mazama pocket gopher (MPG; Thomomys mazama), and streaked horned lark (SHL; Eremophila alpestris strigata)
  • As part of this effort, the authors held elicitation workshops to garner the input of biologists with expertise on the target taxa and the South Puget Sound prairie ecosystem.
  • The goal of these workshops was to parameterize patch dynamics models for each of the species.
Conclusion
  • The authors considered the problem of protecting rare species by recommending patches of land for conservation.
  • In order to cope with changing availability of patches, the authors proposed an opportunistic policy for dynamically making recommendations.
  • The authors proved the surprising result that this simple opportunistic policy is competitive with a clairvoyant solution that is informed in advance of the budget in each timestep and when each patch will be available.
  • The authors conducted a detailed case study of conservation planning for three rare taxa in the Pacific Northwest of the United States.
  • The authors believe that the results provide interesting insights for dynamic/adaptive optimization, and could be useful for other applications such as influence maximization over networks
Summary
  • Introduction:

    One key challenge in computational sustainability is to allocate resources in order to optimize long-term objectives.
  • An archetypal application is conservation planning: managers recommend patches of land for conservation in order to achieve long-term persistence of endangered species.
  • In this and similar applications, the authors typically have to make decisions over time: Financial resources are periodically made available and should be used effectively.
  • The authors prove a surprising fact: Under some natural conditions, a simple policy, that in every round of the decision making process opportunistically allocates the budget given the current reserve and current resources, attains a performance which is competitive with the optimal
  • Methods:

    The authors generate contiguous candidate patches from the parcels by a region growing process, which picks a random parcel as seed, and iteratively grows the patch up to a random size.
  • This growth process is randomly biased to avoid complex boundaries.
  • Instead of using the algorithm described by Sviridenko [2004] for solving Problem (2), the authors use a faster algorithm of Leskovec et al [2007] that carries theoretical guarantees
  • Results:

    4.1 Reserve Design Case Study

    The authors are conducting a computational study in collaboration with the US Fish and Wildlife Service Washington Office, in Washington State, USA.
  • The target species are Taylor’s checkerspot (TCS; Euphydryas editha taylori), Mazama pocket gopher (MPG; Thomomys mazama), and streaked horned lark (SHL; Eremophila alpestris strigata)
  • As part of this effort, the authors held elicitation workshops to garner the input of biologists with expertise on the target taxa and the South Puget Sound prairie ecosystem.
  • The goal of these workshops was to parameterize patch dynamics models for each of the species.
  • Conclusion:

    The authors considered the problem of protecting rare species by recommending patches of land for conservation.
  • In order to cope with changing availability of patches, the authors proposed an opportunistic policy for dynamically making recommendations.
  • The authors proved the surprising result that this simple opportunistic policy is competitive with a clairvoyant solution that is informed in advance of the budget in each timestep and when each patch will be available.
  • The authors conducted a detailed case study of conservation planning for three rare taxa in the Pacific Northwest of the United States.
  • The authors believe that the results provide interesting insights for dynamic/adaptive optimization, and could be useful for other applications such as influence maximization over networks
Related work
  • Conservation planning. There are several powerful tools available for conservation planning, including Marxan (Ball, Possingham, and Watts 2009) and Zonation (Moilanen and Kujala 2008). However, none of those tools currently implements complex patch dynamics models of species persistence. Also, they do not provide guarantees of (near-) optimality. Perhaps closest in spirit to our work is an approach by Sheldon et al [2010]. They propose a network optimization approach with applications to conservation planning. In their approach, they model the population behavior using the independent cascade model of Goldenberg, Libai, and Muller [2001]. One particular aspect that they consider is the non-submodularity of the reserve design problem in absence of Condition 4. They propose an approach based on Mixed-Integer Programming (MIP) to overcome limitations of the greedy algorithm. However, their work does not handle the dynamic aspects of conservation planning that are the focus of this paper. The independent-cascade model is essential to their approach, and it seems difficult to use it to model complex interactions between species (such as symbiosis or predator-prey relationships), or more complex population dynamics (beyond presence-only population models), which all can be handled by our approach. Furthermore, in most of their experiments conducted on a real reserve design case study, the non-greedy network design approach based on MIP performs comparably to the greedy approach, providing further evidence about the appropriateness of Condition 4. We consider the development of principled dynamic planning approaches that do not rely on Condition 4 an interesting direction of future work.
Funding
  • This research was partially supported by ONR grant N00014-09-1-1044, NSF grants CNS-0932392 and IIS-0953413, the Caltech Center for the Mathematics of Information, and by the US Fish and Wildlife Service
Reference
  • Ball, I.; Possingham, H.; and Watts, M. 2009. Spatial conservation prioritisation: Quantitative methods and computational tools. Oxford University Press. chapter Marxan and relatives: Software for spatial conservation prioritisation.
    Google ScholarLocate open access versionFindings
  • Bateni, M. H.; Hajiaghayi, M.; and Zadimoghaddam, M. 2010. The submodular secretary problem and its extensions. In APPROX.
    Google ScholarFindings
  • Environmental Systems Research Institute. 2008. ArcGIS 9.3 users manual. Technical report, ESRI.
    Google ScholarFindings
  • Goldenberg, J.; Libai, B.; and Muller, E. 2001. Talk of the network: A complex systems look at the underlying process of word-of-mouth. Marketing Letters 12(3):211–223.
    Google ScholarLocate open access versionFindings
  • Golovin, D., and Krause, A. 2010. Adaptive submodularity: Theory and applications in active learning and stochastic optimization. CoRR abs/1003.3967v3.
    Findings
  • Gomes, R., and Krause, A. 2010. Budgeted nonparametric learning from data streams. In ICML.
    Google ScholarFindings
  • Goundan, P. R., and Schulz, A. S. 200Revisiting the greedy approach to submodular set function maximization. Technical report, Massachusetts Institute of Technology.
    Google ScholarFindings
  • Hanski, I. A.; Moilanen, A.; Pakkala, T.; and Kuussaari, M. 1996. The quantitative incidence function model and persistence of an endangered butterfly metapopulation. Conservation Biology 10.
    Google ScholarLocate open access versionFindings
  • Koller, D., and Friedman, N. 200Probabilistic Graphical Models. The MIT Press.
    Google ScholarFindings
  • Leskovec, J.; Krause, A.; Guestrin, C.; Faloutsos, C.; VanBriesen, J.; and Glance, N. 2007. Cost-effective outbreak detection in networks. In KDD.
    Google ScholarLocate open access versionFindings
  • Moilanen, A., and Kujala, H. 2008. ZONATION: Spatial conservation planning framework and software. www.helsinki.fi/bioscience/ConsPlan.
    Findings
  • Nemhauser, G. L.; Wolsey, L. A.; and Fisher, M. L. 1978. An analysis of approximations for maximizing submodular set functions - I. Math. Prog. 14(1):265–294.
    Google ScholarLocate open access versionFindings
  • Sheldon, D.; Dilkina, B.; Elmachtoub, A.; Finseth, R.; Sabharwal, A.; Conrad, J.; Gomes, C.; Shmoys, D.; Allen, W.; Amundsen, O.; and Vaughan, B. 2010. Maximizing the spread of cascades using network design. In UAI.
    Google ScholarFindings
  • Streeter, M., and Golovin, D. 2008. An online algorithm for maximizing submodular functions. In NIPS.
    Google ScholarFindings
  • Sviridenko, M. 2004. A note on maximizing a submodular set function subject to knapsack constraint. Operations Research Letters 32:41–43.
    Google ScholarLocate open access versionFindings
  • Vose, D. 1996. Quantitative Risk Analysis: A Guide to Monte Carlo Simulation Modeling. John Wiley and Son.
    Google ScholarFindings
Full Text
Your rating :
0

 

Best Paper
Best Paper of AAAI, 2011
Tags
Comments