No Slide Title

Download Report

Transcript No Slide Title

Exploratory methods to analyse output from
complex environmental models
Adam Butler, Biomathematics and Statistics Scotland
www.bioss.ac.uk/staff/adam.html
ICMS/SPRUCE workshop, March 2007
Statistical post-processing
Use conventional statistical methods - including smoothing
techniques - to analyse outputs from process-based models
“Static”, in contrast with “dynamic” emulation/GP approaches
Provides an approach to the exploratory analysis of aspects
of uncertainty and inadequacy in highly complex models, by
allowing us to make use of information from a small # of runs
Past trends in North Sea storm surges .
Butler et al. : to appear in JRSS C
Analyse output from a single run of a storm surge model reconstructed North Sea surge elevations for the years 1955-2000
Compare spatial and temporal trends in storm surge magnitude
and frequency with those seen in observational sea level data
Analysis based on an extreme value model
Use nonparametric regression (local likelihood) to allow
parameters to vary over space and time in a smooth way
Projecting trends in global vegetation .
Doherty et al. : in preparation
Quantify impact of climate uncertainty upon projections generated
by the Lund-Potsdam-Jena Dynamic Global Vegetation Model
Run LPJ once using gridded climate data for the 20th century
(“control run”), then eighteen more times using climate projections
for the 20th and 21st centuries generated by ensemble runs from
nine different Atmosphere-Ocean General Circulation Models
Ignore inadequancies of LPJ; focus on using the climate model
projections to predict future values of the control run
Annual global vegetation carbon
Calibration period (20th century)
xC = control run of LPJ
yiC = LPJ run using i-th ensemble
Prediction period (21st century)
yiP = LPJ run using i-th ensemble
Prediction
Compute discrepancies zkC = ykC - xC
Fit a parametric model to ziC
Use it to compute predictive distribution of ziP
Predict control run to be xP = yKP - zKP,,
where run K selected with probability wK
The ALARM project
Integrated European Union research project to develop a set
of tools for biodiversity risk assessment
>50 organisations, >200 scientists and social scientists...
BioSS provides statistical support for the project
Focus is on assessing risks from multiple environmental
pressures at multiple scales, and on risk communication…
Evidence comes from species atlas data, local experimental
data, process-based models and expert opinion