The LnGM radon was used on the y axis because the uncertainty
(expressed by the LnGSD) had to be symmetrically distributed either side of each data point for the
WTLS regression analysis. There is little difference between the indoor radon concentrations
predicted by the GM indoor radon – eRa226 and LnGM indoor radon – eRa226 regression lines at
the upper limits of the data used to construct the models. However, at the higher concentrations
which may in some circumstances be encountered above near-surface radioactive waste disposal
sites, the LnGM indoor radon – eRa226 regression models predict higher radon concentrations than
the GM indoor radon – eRa226 models, when the intercepts for both types of models are
unconstrained.
Intercepts of 4-5 Bq m-3 would be expected for all regression analysis models as this is the average
contribution to indoor air from radon in outdoor air and building materials for dwellings in the UK.
The soil gas data vs. indoor radon regression models have intercepts of 4 Bq m-3
for both the INONS
(1-km grouping) and the Derby-Notts Carb-Perm 5-km grouping. In contrast, the intercepts for the
eRa226 (HIRES) vs. indoor radon models have intercepts of 18 Bq m-3 for DINMDMIX and 42 Bq m-3
for DINLM (1 Bq m-3
for ‘all data’).
A key issue is whether to use models with an intercept constrained for theoretical reasons to
5 Bq m
-3
, which represents the average value for the contribution from outdoor radon and building
materials), or unconstrained models based solely on the empirical data. Slopes of the fixed intercept
models are generally steeper than the slopes for unconstrained models. Extrapolation of fixed
intercept models to the likely 226Ra and soil gas radon concentrations that may be encountered in
certain circumstances in future over the LLWR will predict higher indoor radon concentrations than
will extrapolation of unconstrained models. However, models with and without a constrained
intercept for the value of indoor radon gave results that were generally within a factor of two of
each other, suggesting that, allowing for uncertainties on the inputs, such extrapolation of the
constrained models may be justified.
19
Comparison with published data used to derive an average indoor radon : soil radium ratio showed
that the LS models for the relatively impermeable English geological units plot parallel to the
Sheppard et al. (2006) model but below it (i.e. lower indoor radon : soil radium ratios), whilst the
models for the permeable English geological units plot parallel but above the Sheppard et al. (2006)
model (i.e. have higher indoor radon : soil radium ratios).