Integration of models

Instead of aiming to construct the “best” model, researchers working in systems analysis of complex socio-environmental systems tend to use multi-model ensembles.

In a multi-model ensemble each model relies on its own particular set of assumptions and modeling tools, and views the underlying phenomenon from its own specific angle. Moreover, each model usually has its own intrinsic uncertainties attributed to it. Thus the model’s outcomes are commonly represented as random variables (or random processes).

A systems analyst then has to deal with a family of probability distributions providing alternative descriptions of the same object. Developing a well justified tool for integration of different viewpoints into a single picture is a challenge in systems analysis.

There have been several impressive attempts to create a methodology for integration of alternative model-based results. Such attempts rely primarily on appropriately weighing  results provided by a multi-model ensemble, based on assessment of the models’ performance in the past. However, this approach is missing a “physical” justification.

A “physically justified” approach was suggested that does not employ the weighing paradigm and does not use information on the models’ performance in the past [1]. The approach can thus be used in situations, where either the models’ performance is difficult (or impossible) to estimate, or where the future system dynamics is anticipated to be radically different from the one observed in the past.

The methodology relies on the principle of mutual compatibility of prior random estimates. An illustrative case study [2] reconciles two estimates of the Net Primary Production (NPP) of the Russian terrestrial ecosystems, which come from two alternative NPP assessment methods (see Figure 1). This research effort is supported by the FP7-funded project “Knowledge-based Climate Mitigation Systems for a Low Carbon Economy (COMPLEX)” (2012—2016).

Figure 1. The distributions of the NPP values (g C/m2 per year) in the middle taiga (left), southern taiga (middle) and temperate forests (right) bioclimatic zones in Russia, provided by the landscape-ecosystem approach (blue), an ensemble of dynamic global vegetation models, and resulting from posterior integration of those (green). The posterior integrated distribution is sharper than each of the original ones, indicating that these are structurally consistent. Source: [2].

Last edited: 21 May 2014