The construction of deep railway tunnels requires the prediction of na
tural temperatures at depth. Geothermal data for the Alps are presente
d and principles of previously employed methods to predict temperature
s, using Andreae's analytical approach, are discussed. We then use a f
inite element numerical model based on pure conduction to calculate te
mperatures at depth. This method allows rock heterogeneity and anisotr
opy to be taken into account. This model is applied to the Maurienne-A
mbin tunnel project, a 55 km long tunnel between St-Jean-de-Maurienne
(France) and Susa (Italy), which will be the longest tunnel for the pl
anned TGV (high speed train) Lyon-Torino link. Data from several deep
boreholes (10 total, with 3 > 1000 m) are used to provide essential pa
rameters for the model, i.e.: - geological structure; - geothermal gra
dients; - rock conductivities from cores; - geothermal deep heat flow.
Modelling is done in two dimensions, but the effect of surface topogr
aphy (3D) is considered. Results are given in the form of a geothermal
cross-section along the tunnel axis that provides maximum temperature
s and lengths of zones of high temperature encountered (for instance,
zones where theta is greater than or equal to 40 degrees C). In genera
l, differences between calculated and measured temperatures are less t
han 1 degrees C at great depth. At shallow depth, differences are some
times higher and probably best explained by water circulation connecte
d to the surface. The modelling of temperatures, in relation to the ge
ological structure, rock properties, and geothermal data for this area
, appears to be a very useful tool for comparing alternative routes fo
r deep tunnel projects and, during construction, to predict potential
local geological or hydrological anomalies.