TY - JOUR

T1 - Reducing Uncertainty in Ecosystem Service Modelling through Weighted Ensembles

AU - Hooftman, Danny

AU - Bullock, James

AU - Jones, Laurence

AU - Eigenbrod, Felix

AU - Barredo, Jose

AU - Forrest, Matthew

AU - Kinderman, George

AU - Thomas, Amy

AU - Willcock, Simon

PY - 2022/2

Y1 - 2022/2

N2 - Over the last decade many ecosystem service (ES) models have been developed to inform sustainable land and water use planning. However, uncertainty in the predictions of any single model in any specific situation can undermine their utility for decision-making. One solution is creating ensemble predictions, which potentially increase accuracy, but how best to create ES ensembles to reduce uncertainty is unknown and untested. Using ten models for carbon storage and nine for water supply, we tested a series of ensemble approaches against measured validation data in the UK. Ensembles had at minimum a 5-17% higher accuracy than a randomly selected individual model and, in general, ensembles weighted for among model consensus provided better predictions than unweighted ensembles. To support robust decision-making for sustainable development and reducing uncertainty around these decisions, our analysis suggests various ensemble methods should be applied depending on data quality, for example if validation data are available.

AB - Over the last decade many ecosystem service (ES) models have been developed to inform sustainable land and water use planning. However, uncertainty in the predictions of any single model in any specific situation can undermine their utility for decision-making. One solution is creating ensemble predictions, which potentially increase accuracy, but how best to create ES ensembles to reduce uncertainty is unknown and untested. Using ten models for carbon storage and nine for water supply, we tested a series of ensemble approaches against measured validation data in the UK. Ensembles had at minimum a 5-17% higher accuracy than a randomly selected individual model and, in general, ensembles weighted for among model consensus provided better predictions than unweighted ensembles. To support robust decision-making for sustainable development and reducing uncertainty around these decisions, our analysis suggests various ensemble methods should be applied depending on data quality, for example if validation data are available.

KW - Carbon

KW - Committee averaging

KW - Prediction Error

KW - Accuracy

KW - United Kingdom

KW - Validation

KW - Water supply

KW - Weighted averaging

U2 - 10.1016/j.ecoser.2021.101398

DO - 10.1016/j.ecoser.2021.101398

M3 - Article

VL - 53

JO - Ecosystem Services

JF - Ecosystem Services

SN - 2212-0416

M1 - 101398

ER -