AMEMR 2005
In June 2005 Plymouth Marine Laboratory organised and hosted the first international symposium titled Advances in Marine Ecosystem Modelling Research (AMEMR). This provided a novel forum for presentation and discussion of all aspects of model based marine ecosystem research, encompassing numerical, conceptual, mathematical and statistical approaches. The symposium attracted over 160 abstract submissions and nearly 200 delegates from across the globe specialising in a diverse range of modelling techniques and research foci. The symposium was supported by a number of UK national and international organisations including PML, NERC, GLOBEC, SCOR, IMBER and GEOHAB.
The need for such a symposium stems from the development of earth system science over the last few years. Marine, coastal and estuarine ecosystems are complex assemblages of biota, chemical processes and physical dynamics that are influenced by both climate and human activities. Interactions between these system components are often non-linear and comprise varied feedback mechanisms. As a result, changes in species distributions and seasonal responses are rarely fully understood. It is becoming clear that improved management of the marine system requires an appreciation of both ecosystem structure and function. Modelling provides a key scientific technique by which we can elucidate the workings of the marine system and predict its evolution in both the short and long term. Our ambition was that this symposium will contribute to the next generation of model based exploration by providing scientists and students an opportunity to discuss and contrast recent advances, outstanding problems and future requirements.
Three key challenges for ecosystem modellers were identified during the conference; how do we balance increasing model complexity to represent all relevant processes with the subsequent increase in uncertainty in parameterisation, how can we re-establish the relationship between modellers and experimentalists and how can be better understand, quantify and reduce model errors? The primary role of models in ecosystem science is to provide a simplification of complex reality. However it is apparent that a deeper understanding of the marine biogeochemical system is required and to describe the multidimensional behaviour of ecosystems and their interaction with many interlinked biogeochemical cycles, the degree of elaboration may have to grow substantially more. A fundamental issue therefore, is to establish appropriate levels of complexity that will enable ecosystem models to have most predictive skill while also providing scientific insight.
However, the continual development and enhancement of coupled hydrodynamic-ecosystem models has reached the point where the outputs of such models are themselves of a complexity and volume that they require simplification in order to be understood. Validation is a non-trivial exercise as errors can derive from both models and real-world observations. Model errors may derive from inaccuracies in process descriptions, parameterisation, initialisation and forcing functions. Errors in real-world observations may arise from basic measurement error, inappropriate scales of sample dispersion (for example data that are over influenced by small-scale processes not included in the model) or lack of replication. A crucial issue is balancing precision (how well the model fits the data) with trend (how well the model reproduces observed trends). For example, small differences in the timing of an event can lead to large errors in terms of precision even when the trend is well reproduced. The choice of error statistic is crucial, and a comprehensive validation process must consider several. Taking all of this into account, it is surprising that model evaluation, when it is attempted, is often qualitative and largely subjective.
The need for such a symposium stems from the development of earth system science over the last few years. Marine, coastal and estuarine ecosystems are complex assemblages of biota, chemical processes and physical dynamics that are influenced by both climate and human activities. Interactions between these system components are often non-linear and comprise varied feedback mechanisms. As a result, changes in species distributions and seasonal responses are rarely fully understood. It is becoming clear that improved management of the marine system requires an appreciation of both ecosystem structure and function. Modelling provides a key scientific technique by which we can elucidate the workings of the marine system and predict its evolution in both the short and long term. Our ambition was that this symposium will contribute to the next generation of model based exploration by providing scientists and students an opportunity to discuss and contrast recent advances, outstanding problems and future requirements.
Three key challenges for ecosystem modellers were identified during the conference; how do we balance increasing model complexity to represent all relevant processes with the subsequent increase in uncertainty in parameterisation, how can we re-establish the relationship between modellers and experimentalists and how can be better understand, quantify and reduce model errors? The primary role of models in ecosystem science is to provide a simplification of complex reality. However it is apparent that a deeper understanding of the marine biogeochemical system is required and to describe the multidimensional behaviour of ecosystems and their interaction with many interlinked biogeochemical cycles, the degree of elaboration may have to grow substantially more. A fundamental issue therefore, is to establish appropriate levels of complexity that will enable ecosystem models to have most predictive skill while also providing scientific insight.
However, the continual development and enhancement of coupled hydrodynamic-ecosystem models has reached the point where the outputs of such models are themselves of a complexity and volume that they require simplification in order to be understood. Validation is a non-trivial exercise as errors can derive from both models and real-world observations. Model errors may derive from inaccuracies in process descriptions, parameterisation, initialisation and forcing functions. Errors in real-world observations may arise from basic measurement error, inappropriate scales of sample dispersion (for example data that are over influenced by small-scale processes not included in the model) or lack of replication. A crucial issue is balancing precision (how well the model fits the data) with trend (how well the model reproduces observed trends). For example, small differences in the timing of an event can lead to large errors in terms of precision even when the trend is well reproduced. The choice of error statistic is crucial, and a comprehensive validation process must consider several. Taking all of this into account, it is surprising that model evaluation, when it is attempted, is often qualitative and largely subjective.