As is described in the Chapter 2.1, "To establish methodologies for evaluating predictions" is defined as one of the goals of the Phase I. This goal is set to specify criteria and detailed methodologies for evaluating predictions made in the Phase I by comparison with the data obtained in the Phase II.

In order to evaluate prediction appropriately, it is required to clearly define what and how to evaluate, where and how to obtain data. Taking it into consideration, overseas precedent cases are examined in the Phase I. The outlines are as follows.

1) Geological investigations
Predictions are evaluated by comparing with geology and geological structures encountered in research drifts in the Phase II.

A precedent about methods of prediction and its evaluation is found in the HRL project in Sweden⁹⁷⁾.

In the HRL project of Sweden, geological models for each of following scale are constructed : facilities scale (5001,000m), block scale (50100m) and minute scale (510m).
facilities scale : fracture zone
block scale : lithofacies, fracture zone, rock properties, fracture properties
minute scale : rock properties, fracture properties

For these scale, predictions are made on the following items.
lithofacies : ratios of constituent lithofacies, location of facies boundaries, frequency of facies boundaries, mineral composition, alteration
fracture zone : location, strike, dip, width, RQD (for large-scale fracture zone only)
rock properties : matrix porosity, density
fracture properties : number of fracture systems, general trend, spacing between fractures, trace length, filling minerals

Predicted values include confidence interval and certainty. These are determined by experts by taking sampling locations into consideration. When measured values drop within confidence interval, the prediction is assessed as valid. The validity of the prediction is quantitatively expressed by the following equation.

Percentage of absolute error = (absolute value of difference between a predicted value and a measured value) measured value.

2) Hydrogeological investigations

Predictions are evaluated by comparing data obtained in the Phase II with a hydrogeological model and the results of groundwater flow simulation carried out based on the hydrogeological model.

Methods of prediction and evaluation are shown by the HRL project in Sweden which carries out a groundwater flow simulation on the basis of a hydrogeological model. In the project, major water channels are deterministically integrated in the model during surface-based investigation stage⁹⁸⁾. Here, the prediction is evaluated by the amounts of various groundwater inflows. Specifically, total inflow into research drifts, inflow from water channels along research drifts (inclined drifts and spiral ramp) and inflow into advance-installed weirs. However, the verification is limited to be qualitative.

As for the MIU Project, following two things could be used for the evaluation.
amount of groundwater inflow into research drifts
change in water pressures in the existing boreholes with the progress of research drifts' excavation.
Concerning the amount of inflow, (a) total amount, (b) the amount from expected water channels and (c) the amount from the sections set in advance for inflow measurement are thought to be possible to predict. For measurement, following procedure is expected : a sheet is sandwiched between drift wall and concrete lining, and groundwater flowing between the sheet and the drift wall is gathered at certain depths, and finally measure the amount of the gathered water. In fracture zones where a lot of water inflows are expected, several antecedent boreholes are excavated for measuring the water inflow amount by crosshole hydraulic test. After these measurements, drift walls are grouted due to necessity of construction work. The predictions are evaluated by comparing with the measurement data.

Amounts of drawdown in the existing boreholes with the advance of excavation of research drifts are measured by the MP system. The MP system is installed in these boreholes prior to the excavation. Intervals for water pressure measurement are determined by the result of long-term pumping tests carried out prior to the shaft excavation and its predictive simulation. The prediction is evaluated by comparing the amount of drawdown in the existing boreholes with the results of the predictive simulation. This evaluation is expected to be done as quantitatively as possible.

3) Hydrochemical investigations

The spatial distribution of hydrochemical properties is predicted using the predictive results of distributions of hydrogeological properties. The prediction is evaluated by comparing the predicted and measured data obtained during the drift excavation.

One of the phenomena originated in the research drifts excavation is a pressure drop of groundwater and the resultant variations in groundwater hydrology⁹⁸⁾. It is presumed that the distribution of hydrochemical properties of groundwater are so closely related to groundwater hydrology. Therefore, a change in groundwater hydrology could cause a change in the distribution of hydrochemical properties of groundwater. Besides, the drop in water level causes air invasion into the rock mass, in turn resulting in a change in redox potential in geological environments⁹⁹⁾.

To predict changes originated in the research drifts excavation, it is necessary to extract items that should/could be predicted in the Phase I. These items are as follows.
Items that should be predicted :
All items planned to analyze using boreholes excavated in the Phase I
Items that could be predicted :
Based on the 3-D distribution of hydrochemical properties obtained in the Phase I, items that could be predicted are extracted. For example, components with no significant difference between surface water and groundwater are excluded. Also, components whose analytical quality are not guaranteed are excluded.

4) Rock mechanical investigations

The predictions are evaluated by understanding behaviors of the surrounding rock mass and change in their physical properties during the excavation of underground caverns (e.g. shaft, research drifts). The distribution of RMR (Rock Mass Rating : one of the rock classification methods) is predicted in overseas underground facilities as one of the case studies for predicting the physical properties distribution of the rock mass. However, there is no predictive analysis of the rock mass aiming at the whole facilities.
Following predictions are obtained through modeling and simulation in the Phase I.
3-D distribution of rock mechanical properties of the rock mass (physical/mechanical properties and initial stress states of rocks)
deformational behaviors of the rock mass surrounding research drifts generated by their excavation
distribution of the damaged zone caused by stress change/concentration in the rock mass surrounding research drifts
The first one is predicted by the construction of rock mechanics conceptual model. The latter two are predicted by the results of the predictive simulations. All of these predictions are evaluated as follows.

Data needed for evaluation of the prediction
a) Data for evaluating rock mechanics conceptual model
Rock mechanics conceptual model is constructed by interpreting the data obtained from boreholes and rock samples. Therefore, following in-situ data obtained during excavation is necessary.
physical properties (physical/mechanical properties of rocks and initial stress distribution)
b) Data for evaluating the results of predictive simulations
Following results of the predictive simulations are necessaary.
Behaviors of the rock mass surrounding research drifts (displacements of rock matrix and fractures and the total displacements)
the extent of stress changes caused by stress redistribution
the extent of damage (plastic domain) by stress concentration in the rock mass

Evaluation methods of predictive results
a) evaluation of rock mechanics conceptual model
Physical property zones in the model, physical property and the initial stress distribution (given as a boundary condition) given to the individual zones are compared with the measured data for evaluation. As for the initial stress distribution, it might be greatly affected by the Tsukiyoshi Fault depending on the location of research drifts. Therefore, the results of predictive simulation of the 3-D stress distribution carried out before research drift excavation are evaluated.
b) evaluation of the results of predictive simulations
When the rock mechanics conceptual model is evaluated appropriate in a), the data described in b) are compared with the measured data.
c) re-evaluation of the results of predictive simulations
When the rock mechanics conceptual model is not evaluated appropriate in a), the rock mechanics conceptual model is re-constructed based on the result of a). And the predictive simulation is carried out using the model, and its results are evaluated. When there are qualitative discrepancies between the results of the predictive simulations and the measured data, the cause should be examined. In this case, the discrepancy is thought to be originated in factors that are not considered in the model. Therefore, retrial of the predictive simulation are carried out if possible. When only quantitative discrepancies are found, individual physical property should be rechecked by the data used in a) to carry out a retrial of the predictive simulation.

Following is made possible by the above-mentioned process.
validity of a series of process from investigations, model construction to a predictive simulation
required data and its precision acquired in the Phase I
optimization of the investigation methodologies for model construction