[A-GE03_30AM2] Subsurface Mass Transport and Environmental Assessment

[AGE03-06] Geostatistical Interpolation of Thermal Properties of Boring Core Samples

*Hirou MUTO¹, Hirotaka SAITO¹ (1.Tokyo University of Agriculture and Technology)

Keywords:thermal conductivity, kriging, cokriging, sampling intensity, prediction error

Ground source heat pump systems (GSHP) that use ground or groundwater as a heat source can achieve much higher coefficient of performance (COP) than conventional air source heat pump systems. Although use of GSHP systems has been rapidly increasing worldwide, environmental impacts by GSHP systems have not been fully investigated. To rigorously assess GSHP impact on the subsurface environment, instead of relying on "effective" properties, ground thermal properties including thermal conductivity and heat capacity need to be accurately characterized.A geostatistical least-square interpolation method, known as kriging, has been used to characterize the spatial distribution of soil (or ground) physical (both hydrological and thermal) properties in one, two, and three dimensional domains. Kriging can estimate not only the values of an attribute at un-sampled locations accounting for spatial correlations between variables but also their uncertainties in terms of an error variance. Ordinary kriging (OK) which estimates unknown value as a linear combination of neighboring observations is one of the most commonly used kriging estimators. A secondary variable which is spatially cross-correlated with the primary variable can be used to reduce the estimation variance for the primary variable. Such method is known as cokriging. Ordinary cokriging (OCK) is one of the most commonly used cokriging estimator. The objective of this study was to compare OK and OCK in terms of estimating soil thermal properties along 50-m boreholes through the cross validation. Water content and sand content, which are relatively easy to measur, were used as the secondary attributes in cokriging.In this study, undisturbed boring core samples were collected from two 50-m long boreholes at the campus of Tokyo University of Agriculture and Technology in Tokyo. Volumetric heat capacity (HC), thermal conductivity (TC), gravimetric water content (WC) and volumetric sand content (SC) were measured every 10-20 cm along the cores. The impact of sampling intensity on prediction errors were investigated by drawing random subsets of increasing size and using them to predict thermal properties at the remaining locations (jackknife approach). Then, subsets of N data were selected randomly or randomly per 10-m depth from the entire data set. For both sampling approaches, 50 different random subsets were selected to account for sampling fluctuations. Thermal properties at the remaining locations were then predicted.This study showed that increasing the size of the subset leads to smaller mean absolute error. It was also found that kriging with random subsets per every 10-m depth yields lower mean absolute error than that with random subsets. Prediction errrors by OCK were smaller than those by OK when the sampling intensity was the same.