Share

Model Skepticism

Introduction

I tend to meet people with opposite views on models. Either they see the model as a miraculous black box that could predict / simulate the wildest scenarios, without paintaken field work, or they are highly skeptical about the usefulness of models at all. After all, what you put in is what you get out. I tried to present some of my thoughts on modeling below. I'll be pleased to receive your comments at the bottom of this page.

Modeling paradigm

Simulation modeling is a way to describe current understanding of a system in mathematical equations. The predictive accuracy of biological models is often low. It is questionable if biological models will ever mimic reality closely. Biological systems are highly complex and therefor any model simulating these systems tend to be inherently complex. Furthermore, there is a statistical relation between number of input parameters and number of calculations and the accuracy of the model. Numerical errors can easily 'grow'. Uncertainty in the accuracy of each parameter may easily grow each timestep.

What than is the usefulness of simulation models?

First of all, the model description in itself is useful. Modeling develops our thinking about a system. The inaccuracy and the often criticized fudge factors indicate gaps in our knowledge and could be seen as 'research questions' generated by the modeling exercise. Models have had great influence on our thinking and on the direction of research.

Biological system are often too complex to simply think them through. Computer models aggregate our knowledge and allow the researcher to think at a higher abstraction level, without loosing quantitative information. Our minds have great ability to simplify complex problems by splitting them up in a number of smaller problems. Computer programming language (designed by humans) supply the programmer with constructs to modularize a program. Notably, object oriented programming has been designed to closely mimic human perception of the world. However, the difference between computers and our brain is that our brains tends to overlook the details when considering higher aggregation levels thus making it difficult to quantitatively asses a problem. Rather, humans tend to prefer qualitative assessments. Therefor simulation models are complementary to our thinking.

Models stimulate quantitative research. There is a strong bias towards qualitative research in life sciences. The anova is the preferred statistical tool and quantitative measurements are often presented in qualitative terms: "significant differences".

The importance of quantitative research has grown with the emphasis on a holistic approach. The relative new discipline of Ecology is an example in which a holistic world view is more dominant than a reductionist world view. The study of ecosystems lays emphasis on interdependence and interactions which only exist in the system as a whole. Interesting enough, traditional research methods are not well adapted for a holistic approach. Simulation modeling is an attempt to study a system as a whole. Interestingly, a paradox arises here in which an in nature extremely reductionist method (see previous paragraph) has become a prime holistic research tool.

Till now I emphasized the importance of models as research tool. How useful are models as decision support systems (DSS)? Since the predictive nature of biological simulation models is low, they may not be very good decision support systems. In fact, I know of only 1 success story in which a relative simple mineralization model aided precision fertilization better than continued N content measurements of the leafs (PRI, unpublished). However, although skepticism about the usefulness of models as DSS is grounded with examples, models may still give better advise than no advise at all. Current examples are the climate change models that predict a wide range of scenarios for the future, however all have 1 thing in common, the earth's climate is warming up. These models, together with empirical data, function as a strong incentive to change our lifestyle for the good.

Examples in which model were wrongly ignored are sometimes overlooked. A typical example is the breeding effort for up regulation of P uptake carriers which has not let to higher yields on P deficient soils. Barbers relative simple nutrient uptake model predicted over 20 years ago that P uptake was not limited by the carriers but by the slow diffusion of P in the soil.