Data-model comparison using fuzzy logic in paleoclimatology.

Until now, most paleoclimate model-data comparisons have been limited to simple statistical evaluation and simple map comparisons. We have applied a new method, based on fuzzy logic, to the comparison of 17 model simulations of the mid-Holocene (6 ka BP) climate with reconstruction of three bioclimatic parameters (mean temperature of the coldest month, MTCO, growing degree-days above 5 °C, GDD5, precipitation minus evapotranspiration, P−E) from pollen and lake-status data over Europe. With this method, no assumption is made about the distribution of the signal and on its error, and both the error bars related to data and to model simulations are taken into account. Data are taken at the drilling sites (not using a gridded interpolation of proxy data) and a varying domain size of comparison enables us to make the best common resolution between observed and simulated maps. For each parameter and each model, we compute a Hagaman distance which gives an objective measure of the goodness of fit between model and data. The results show that there is no systematic order for the three climatic parameters between models. None of the models is able to satisfactorily reproduce the three pollen-derived data. There is larger dispersion in the results for MTCO and P−E than for GDD5. There is also no systematic relationship between model resolution and the Hagaman distance, except for P−E. The more local character of P−E has little chance to be reproduced by a low resolution model, which can explain the inverse relationship between model resolution and Hagaman distance. The results also reveal that most of the models are better at predicting 6 ka climate than the modern climate.