Geophysical Monitoring for Geologic Carbon Storage. Группа авторов

Читать онлайн книгу.

Geophysical Monitoring for Geologic Carbon Storage - Группа авторов


Скачать книгу
one must resort to numerical approaches in order to compute G i (x, y). Even a layered medium requires a numerical approach to compute the Green's function (Wang et al., 2006).

      It is assumed that the fractional volume change given by Δ υ i is constant over the elemental volume V j . The function G ij (x) represents the integral of the Green's function G i (x, y) over V j ,

      (2.10)upper G Subscript i j Baseline left-parenthesis x right-parenthesis equals integral Underscript upper V Subscript j Baseline Endscripts upper G Subscript i Baseline left-parenthesis x comma y right-parenthesis d y period

      where the range change kernel M j (x) is

      (2.12)upper M Subscript j Baseline left-parenthesis x right-parenthesis equals sigma-summation Underscript i equals 1 Overscript 3 Endscripts l Subscript i Baseline upper G Subscript i j Baseline left-parenthesis x right-parenthesis period

      (2.13)r equals upper M v period

      In the inverse problem, observational data are used to estimate the distribution of volume change at depth. The most common approach is a least‐squares formulation in which one seeks a model v that minimizes the sum of the square of the residuals

      (2.14)upper S equals left-parenthesis r minus upper M v right-parenthesis Superscript t Baseline normal dot left-parenthesis r minus upper M v right-parenthesis period

      While it is possible to try to estimate the volume change for a full three‐dimensional source model by minimizing S, the solution will most likely be nonunique or poorly determined. That is, there will be many possible solutions and solving the least‐squares constraint equations will not lead to a stable answer. One can stabilize the solution by adding penalty terms to the misfit function S, representing aspects of the model that are considered undesirable, a procedure known as regularization. For example, the magnitude of deviations from an initial or prior model,

      (2.15)upper M equals left-parenthesis v minus v 0 right-parenthesis Superscript t Baseline dot left-parenthesis v minus v 0 right-parenthesis comma

      is often included as a penalty term. A term penalizing model roughness is also commonly used to regularize the inverse problem. Note that if one takes v 0 equal to zero, the norm penalty will have the undesired effect of biasing the solution to have the largest changes at the shallowest depth. That is, in order to minimize the magnitude of the solution, most significant anomalies are placed close to the surface where they will have the greatest effect on the observations.

      (2.16)upper I equals sigma-summation Underscript i element-of upper B Subscript upper I Baseline Endscripts left-parenthesis normal v Subscript i Baseline minus f nu Subscript i Baseline right-parenthesis squared comma

      to the data misfit function S, where f is a scaling factor to account for the fact that the injection of a cubic meter of water leads to a change that is a fraction of the injected volume. If no such information is available, or if one is uncertain about how to scale the injected fluid volume to fractional volume changes at depth, it is still possible to penalize changes that are far from the well. That is, we can bias the solution to try to put the most significant volume changes near the injection site, where the pressure changes should be the largest. For example, if a well is located at x w , then the penalty function takes the form

      (2.17)upper D equals sigma-summation Underscript i equals 1 Overscript upper N Endscripts upper Delta left-parenthesis x Subscript w Baseline comma i right-parenthesis normal v Subscript i Baseline squared comma

      where Δ(x w , i) is a function measuring the distance between the well and the center of the i‐th grid block.

      One of the advantages of geodetic observations is the relatively dense temporal sampling in comparison with other geophysical methods. For example, SAR data may be gathered at weekly to monthly intervals, leading to a time series of range change for each scatterer. Hence, it is possible to image the time evolution of the volume changes and to try and relate these changes to fluid movement and hydrological properties such as permeability (Vasco, 2004; Vasco et al., 2008; Rucci et al., 2010). The sequence of volume changes within each grid block of the reservoir model can be used to estimate an arrival or onset time, that moment at which the volume of a


Скачать книгу