Luo_data-model assimilation in ecology

Data-Model Assimilation in Ecology
History, present, and future
Yiqi Luo
University of Oklahoma
Outline
1. Historical Perspective
2. Present opportunities
3. Future prospects
Historical Perspective
Data-model Assimilation
Process
thinking
Information
contained in
data
Synthesis
and
prediction
Approaches to scientific research
Experiment (observation)
Model (Theory)
Data
Processes thinking
Theory delineates possibilities
Empirical studies discriminate the actualities
Robert May 1981
Approaches to scientific research
Model – Theory
Experiment
Data
Processes thinking
Simple
model
Simple ecological models
(1800s-1950s)
1. Growth models
Logistic growth equation – Pierre Verhulst 1838
2. Competition model – Lotka-Volterra model 1925,1926
3. Predation model
Merits
Generalizations that sum up many measurements of
attribute and, within limits, can be used for predictions.
Weakness
No much information on mechanisms or processes
Approaches to scientific research
Model – Theory
Experiment
Data
Probability
Statistic
analysis
Processes thinking
Simple
model
Statistical analysis (1600s-)
1654 – Pascal developed mathematics of probability
1805 – A-M Legendre – Least square method
1877-1889 – F. Galton – regression and correlation
1919 – R.A. Fisher – ANOVA
1960s- Ecology literature
Analysis, interpretation, and presentation of masses
of numerical data.
Approaches to scientific research
Model – Theory
Experiment
Data
Probability
Statistic
analysis
Processes thinking
Simple
model
Systems
analysis
Systems analysis
1. First described by Heraclitus in 6th century
BC
2. Active research tools in 1930s-40s
3. Used in ecology in 1950s–60s by Odum,
Watt, and many others.
Holistic analysis on structure and behavior of a
system as a whole.
Approaches to scientific research
Model – Theory
Experiment
Data
Probability
Statistic
analysis
Processes thinking
Simple
model
Systems
analysis
Simulation
model
Simulation model
(1960s- present)
1. Forrester, J.W. 1961 Industry Dynamics
2. De Wit in Netherlands, 1960s – 90s
3. Applications in ecology 1960s – pres
4. Example: CENTURY
Uses
1. Synthesis and integration of data
2. Predicting the behavior of ecosystems
3. Hypothesis generation for study design
4. Policy making.
Simulation model (cont.)
Challenges
• Low confidence on
model output
• Model validation and
testing against data
• Transparency and
amenability to
analysis.
Approaches to scientific research
Model – Theory
Experiment
Data
Simple
model
Probability
Statistic
analysis
Processes
thinking
Systems
analysis
Simulation
model
Baysian
analysis
Data-model assimilation
Simulation model vs. data-model assimilation
Simulation modeling
Data-model fusion
Multiple
Datasets
Parameter
estimates from
literature
Inverse
model
Simulation
model
Simulation
(forward)
model
Model prediction
Model predictions
Inverse
modeling
Forward
modeling
Techniques of Optimization
in Data-model Assimilation
Deterministic inversion
1. Steepest descending
2. Newton method –Isaac Newton (1711)
3. Newton-Gauss method
4. Levenburg-Marquardt algorithm (1944, 1963)
Stochastic inversion
1. Bayesian inversion – Thomas Beyes (1701 – 1761)
2. Markov Chain Monte Carlo – Metropolis-Hastings
(1950s)
3. Simulated annealing (Kirkpatrick et al. 1983)
4. Genetic algorithms (Goldberg 1989)
Potential Uses of the
Data-model fusion
Use of both process thinking and information
contained in data towards a global synthesis.
1.
2.
3.
4.
5.
Parameter estimation
Test of model structure
Uncertainty analysis
Evaluation of sampling strategies
Forecasting
Present Opportunities
FLUXNET
A worldwide network with over 400 tower sites operating
on a long-term and continuous basis, supplemented with
data on site vegetation, soil, hydrologic, and
meteorological characteristics at the tower sites.
TERACC
A worldwide network with over 100 manipulative experimental sites
to study impacts of global change factors on ecosystem processes.
Long Term Ecological Research (LTER) Network
LTER Network
established in 1980,
has 26 sites, and
involves more than
1800 scientists and
students
investigating
ecological
processes over long
temporal and broad
spatial scales.
Synthesis across
sites is one of the
major challenges for
LTER
NEON
Transformational research for a
data-rich era
Characteristics
Data-poor era
data-rich era
Activities
Major effort
Informatics
Objectives
Motives
Service to society
Data collection
Measurements
Spreadsheet
Discovery
Curiosity-driven
Long-term
Data processing
Theory development and test
Eco-informatics
Forecasting
Decision making
Real-time
Future prospects
NEON and other
sensor networks
Theory
Real-time data strings
ecological
models
Eco-informatics
Data-model fusion
Ecological
forecasting
Preparation for
catastrophe
Resource
management
Decision
making
Future research
1. Eco-informatics is not only about acquisition,
analysis and synthesis, and dissemination of
data and metadata but also include model
assimilation to generate data products.
2. Streamline real-time data collection, QA/QC,
and data-model assimilation and data products.
3. Test theory for model development.
4. Support decision making processes
Summery
1. Data and model are two foundational
approaches to scientific inquiry about natural
world.
2. Data-model assimilation combines the bests
from both approaches
3. As we enter a data-rich era, data-model
assimilation becomes an essential tool of
ecological research.
4. Data-model assimilation ultimately help
ecological forecasting to best serve the society