Ecosystem Analysis Using Probabilistic Relational Modeling Bruce DAmbrosio, Eric Altendorf, Jane Jor

1
Ecosystem Analysis Using Probabilistic Relational
ModelingBruce DAmbrosio, Eric Altendorf, Jane
Jorgensen

• Presented by Iulia Oroian and Leonard
  RodrigoTuesday Dec 2nd CSCE 582 Fall
  2003Instructor Dr. Marco Valtorta

2
Definitions

• Ecosystems
• Systems composed of interacting populations of
  organisms and their environment
• Community-level ecosystem model
• An integrated model of the ecosystem as a whole
• Synthetic variables
• Variables derived from observational data
• Aggregator
• A count or value of a specific variable,
  included in the synthetic variable space

3
Goal

• To aid domain scientists in gaining insight into
  data.
• Controlled experimentation in an ecosystem is
  undesirabletherefore it is desirable to create
  comprehensive models from the vast amount of
  observational data available.
• Generally, individual, domain-specific teams
  apply traditional statistical methods to
  investigate correlations among variables in their
  separate datasets.
• Few methods exist for investigating the complex,
  noisy cross-disciplinary interactions that are
  crucial to understanding the ecosystem as a whole.

4
Abstract

• Application of relational model discovery
  methods to building comprehensive ecosystem
  models from data.
• In particular two projects are considered
• - Crater Lake Ecosystem
• - West Nile Virus Disease Transmission
• In both cases the relational probabilistic model
  discovery is applied for building community
  level models of the ecosystems.

5
Project 1 Crater Lake

• Problem
• The NPS is concerned about long-term changes in
  the clarity of Crater Lake, a national park and
  the clearest deep-water lake in the world.
• So far, linking various domain-specific surveys
  into one overall assessment of lake health has
  been lacking.
• Using the relational model discovery methods the
  authors try to derive parameters that account for
  variations in explicit variables, like clarity of
  the lake water.

6
Project 1 Crater Lake

• Data
• Data are obtained from long-term studies of the
  lake (some readings go back to 1880).
• This data have been collected in tables using
  various time and spatial scales.
• For example surface weather condition
  information, phytoplankton densities, weather
  data at altitude.
• Notice that the temporal and spatial granularity
  of the data varies surface weather condition
  information, is available on a daily basis,
  weather phytoplankton densities are measured only
  once or twice a month, and weather data at
  altitude is rarely available.

7
Project 1 Crater Lake

• Method
• A set of temporal units were chosen to frame the
  analysis. For this purpose expert knowledge was
  used.
• These units were time periods corresponding to
  observed patterns of clarity of lake and for
  which data were available
• In the project Jun-Jul, Aug, Sep-Oct

8
Project 1 Crater Lake

• Challenges
• Problem deal with the time, which wasnt
  explicitly reified, therefore constructing paths
  likesecchi.DesDepth.yrSegment.Phyto.density
  was a problem.
• Solution manually add a Season table.
• Problem how to gain scientific insight into data
  
• Solution learning models over not just
  variables in the provided tables, but over their
  parents as well.

9
Project 1 Crater Lake

• A complete schema for
• the data tables related to
• the temporal tables is
• shown in figure 1.

10
Project 1 Crater Lake

• After performing the analysis ( meaning applying
  the relational model discovery method), the
  following essential elements showed in the
  discovered model.

11
Project 1 Crater LakeResults

• One relationship that was discovered is that the
  dominant fish species in gill net catches was
  probabilistically dependent upon
• - Secchi descending depth (water clarity) in the
  current year
• - mean fish weight in the current year
• - descending Secchi depth the previous year
• - dominant fish species two years previous

12
Project 1 Crater LakeResults

• Other findings
• the fact that schools of Kokanee smolts swimming
  at the edges of the lake were preyed upon by
  Rainbow trout and this phenomenon does not occur
  every year. A time lag of two years, discovered
  by the model, is consistent with experts
  observations. The relation between this
  interaction and water quality was previously
  unknown.
• The centrality of water clarity (measured by the
  Secchi DesDepth parameter)
• The lack of a direct relationship between
  Zooplankton count and water clarity.
• These findings suggest that fish attributes may
  serve as a predictor of water clarity.

13
Project 1 Crater LakeResults

• Another important result
• learning models over not
• just the variables in the provide
• tables but over their parents as
• well provide additional insight.
• An example for the
• FishSpecimen table
• is shown in Fig3.

14
Project 2 West Nile Virus

• Data available
• Reports of dead birds testing positive
• Reports of breeding populations of mosquitoes
  testing positive
• Human case reports
• Landscape type

15
Project 2 West Nile VirusDatabase Types

• Static Type
• Presence of permanent mosquito breeding sites
  (tire disposal facilities, etc)
• Landscape type
• Event Type
• Located in place and time
• Birds located testing positive for West Nile
• Mosquitoes testing positive for West Nile

16
Project 2 West Nile VirusModeling Method

• Attempt to create a model of the spread of the
  West Nile Virus in Maryland, 2001
• Selectors are used to relate the correct subset
  of values to other nodes.

17
Project 2 West Nile VirusRelating Different
Databases

• Location and Time are continuous variables
• This is handled by creating a scale. The scale
  is determined by examining previous case studies
  such as the life-cycle of disease-carrying
  mosquitoes and flight distance of competent bird
  hosts.
• In this particular study, the space / temporal
  scale consisted of 5 miles and 1 month.
• Selectors
• Implemented as boolean typestrue for elements in
  the same range, and false for elements outside.

18
Project 2 West Nile VirusModel Fragment
19
Project 2 West Nile ModelResults

• The researchers found that there were
  insignificant cases to effectively use human and
  horses test cases to model the spread of the
  virus
• The model was, however, reasonably accurate, thus
  possibly implying that it is not necessary to
  gather data on insignificant hosts such as horses.

20
Conclusions and Future Work

• Relational probabilistic modeling provides a
  natural framework for investigating ecological
  data.
• Based on the systems relational database the
  methods of relational learning provide the
  opportunity to learn comprehensive models
  directly from the data sources.
• There still are limitations in the current
  synthetic variable construction methods.