Presentation on theme: "Testing models to destruction Neil Crout Environmental Science School of Biosciences University of Nottingham UK."— Presentation transcript:
Testing models to destruction Neil Crout Environmental Science School of Biosciences University of Nottingham UK
Models and falsification Cant prove something is true – So standard approach is to falsify Models are simplified approximations So we know they are wrong – Hard to falsify in a regular way Models are an assembly of relationships – Hard to separate their influence on one another – Hard to falsify in a regular way So, model evaluation is dependent on purpose – Surely correct, but is this sufficient?
Can we falsify the model design through reduction? H = Reducing the relationship to a constant will increase the discrepancy between prediction and observation For each model relationship test the hypothesis, H,
Compare to observations, calculate likelihoods etc
40 other variables <<50% Should we accept the reduction hypotheses? Crop N physiology 50% Vernalisation 99% N-mineralisation Temp/moisture adjustment~99% Canopy Temperature 96% Diurnal adjustment of thermal time 99% Nitrogen Leaching 50%
Co-workers Davide Tarsitano Glen Cox James Gibbons Andy Wood Jim Craigon Steve Ramsden Yan Jiao Tim Reid NMJ Crout, D Tarsitano, AT Wood. Is my model too complex? (2009) Evaluating model formulation using model reduction. Environmental Modelling & Software, 24:1-7 with thanks to BBSRC and Leverhulme