Mbi Notes




Lou Gross

  • Big picture: big data, computational challenges.
  • NEON, DataONE, ADW, iDigBio
  • “Convergence” as the new interdisciplinary (National Academies)
  • Rise of synthesis centers
  • “Enabling architecture for next gen life science research” – National Academies report Lou Gross (2013)
  • Comp Science for Natural Resource Management – Fuller, Wang, Gross (2007)

  • language barriers: What’s a model? Mouse, drosophila? logistic, ricker? GIS map layers? …

Contraints frequently dominate, not the control or the state equation.

Let stakeholders make their own rankings. Scenario analysis vs optimal control. Uncertainies! “Relative assessment protocol” Fuller, Gross, Duke-Sylvester, Palmer. “Testing the robustness of management descions under uncertainty” (ATLSS modeling)


notes from my breakout session:


  • Sensitivity analysis of Scenario Rankings
  • What does a Resilience approach add
  • Generalities

Tools for decision under uncertainty

  • optimization
  • threshold planning
  • scenario planning
  • resilience thinking

(Polasky et al 2011 TREE)


  • Info gap “theory”
  • Satisfiability / mini-max
  • model approximation methods
  • dynamic programming

Leading Challenges

  • To what extent are these approaches different sides of the same coin?
  • Are there truly non-optimization based approaches?
  • Almost-optimal approaches
  • Including constraints

what we do well

  • Optimize easy problems
  • open loop


  • Starting to: simulate optimal solutions to simple problems under more realistic circumstances
  • Starting to find multiple “optima”
  • feedback control (SDP)
  • Dual-control / adaptive control / active adaptive management / learning
  • large state space

Open challenges / what we do poorly

  • Dual control under parameter uncertainty (without restrictive assumptions on parameters)
  • high-dimensional problems
  • multiple stake-holders / game-theory solutions (outside fisheries)
  • mapping between control and implementation (partial controllability)
  • large action space
  • (multiple) delayed effect actions

Open challenges: Multiple stake-holder games

  • beyond 2-player differential games (with feedback)
  • simultaneous player actions

Open challenges: adaptive management timescales

  • frequency of revisiting decisions
  • biological timescales
  • political timescales

Known nuisances

  • curse of dimensionality
  • data collection methods
  • numerical methods
  • local vs global

Missed things

  • spatial data, using rich data under the curse of dimensionality


Writing out proof-of-principle interface to the dataone REST API, see rdataone and Introduction to the package

Key things:

  • We can accomplish handling of certificates from httr, just add config = list(sslcert = ); see ?httr::config, e.g. archive a file with:
                                        config=config(sslcert = "/tmp/x509up_u1000"))

Posting new data requires writing a system metadata XML file. Currently have a crude minimal version of this, write_sysmeta.R, should see how dataone package is handling this.