1 ASSESSING THE VALUE OF TYPICAL METEOROLOGICAL YEARS BUILT FROM OBSERVED AND FROM SYNTHETIC DATA FOR BUILDING THERMAL SIMULATION Ricardo Aguiar, Susana Camelo and Helder Gonçalves INETI / ITE - Department of Renewable Energies Estrada do Paço do Lumiar, 1649-038 Lisboa, Portugal ABSTRACT Typical Meteorological Years obtained from obser- ved meteorological records have become the de facto data source when evaluating thermal performance of buildings. However, this data source has various drawbacks, and alternative TMY assembling methods based on statistical and stochastic models seem to have been perfected to a point where they become sound alternatives. Numerical simulations of test cells were performed for a mid-latitude temperate climate, Lisbon. Thermal performance was evaluated using as input long term observed time series (control situation), long term stochastic data, and TMY obtained by classic and by stochastic methods. It is found that stochastic data seems to provide a good source of meteorological data, indeed more flexible and adequate than those series provided by the currently usual approach. INTRODUCTION Typical Meteorological Year type data sets (hereafter, TMY) obtained from observed meteoro- logical records –“classic” TMY– have become the usual data source when evaluating thermal perfor- mance of buildings through numerical simulation. However there are not many sites for which TMY adequate for this purpose exist. This can be blamed mainly on the shortness of observed hourly data in general, and of solar irradiation data in particular. Other drawbacks of this data type can be pointed. Classic TMY rely on selection of monthly series from a pool of observed data – but its limited size is seldom enough to reach the goal of exactly matching TMY and long term statistics, most remarkably when considering all meteorological parameters simul- taneously. This is due to the large interannual va- riability of the meteorological series. For instance, Mediterranean climates display variation coefficients of monthly values is about 10% for solar irradiation, 7% for temperature and 6% for relative humidity. Also, and again due to the necessarily limited size of the observed data pools, a hierarchy of meteorolo- gical parameters and statistical characteristics of inte- rest must be established, therefore they can’t all be represented at the same level. Finally, classic TMY are often built as “general purpose tools”, and thus correctness of mean values of parameters like pre- cipitation and atmospheric pressure – nearly irrele- vant for building simulation – are also put near the top of the selection criteria. This again can cause other parameters that interest more to the thermal si- mulation, like irradiation, to be less well represented. In recent years, and for some regions/climate types, statistic and stochastic models have been developed to a degree of sophistication enough to provide an alternative way of assembling sound synthetic multi- variate meteorological data series – either long term series or TMY (e.g. Knight et al., 1991; Aguiar, 1998a). For convenience, these will be termed “stochastic” hereafter, although in fact many of the models used in time series generators are just regressions, balance, or thermodynamic equations. Stochastic data are based on necessarily simplified models of the time series occurring in Nature (note that some statistical properties are nearly irrelevant to building behaviour due to smoothing effects introduced e.g. by thermal inertia), but have nume- rous advantages in respect to classic TMY. They are free from discontinuities, gaps, and spurious values. They can be produced starting with input climatic monthly data only, widely available from Atlas and meteorological publications. And they can be made to yield at the output exactly these long term monthly data. In previous works by the same authors [2, 3] it was shown that observed and synthetic TMY yield compatible building performance statistics for Medi- terranean climates. However, these studies where not conclusive, as no reference building performance estimates (computed with “true” input data, viz. observed data series spanning many years) were available. This work contributes to clarify the issue of the relative value of classic TMY and stochastic series (long term and TMY), by comparing the internal thermal conditions of a test cell evaluated through