void ModelGravityDynamic::BuildGMatrix(ModelGravityStatic * modelGravityStatic,
SeismicParametersHolder & seismicParameters)
{
// Building gravity matrix for each time vintage, using updated mean Vp in generating the grid.
double gamma = 6.67384e-11; // units: m^3/(kg*s^2)
Simbox * fullSizeTimeSimbox = modelGeneral_ ->getTimeSimbox();
// Use vp_current, found in Seismic parameters holder.
FFTGrid * expMeanAlpha = new FFTGrid(seismicParameters.GetMuAlpha()); // for upscaling
FFTGrid * meanAlphaFullSize = new FFTGrid(expMeanAlpha); // for full size matrix
int nx = fullSizeTimeSimbox->getnx();
int ny = fullSizeTimeSimbox->getny();
int nz = fullSizeTimeSimbox->getnz();
double dx = fullSizeTimeSimbox->getdx();
double dy = fullSizeTimeSimbox->getdy();
int nxp = expMeanAlpha->getNxp();
int nyp = expMeanAlpha->getNyp();
int nzp = expMeanAlpha->getNzp();
int nxp_upscaled = modelGravityStatic->GetNxp_upscaled();
int nyp_upscaled = modelGravityStatic->GetNyp_upscaled();
int nzp_upscaled = modelGravityStatic->GetNzp_upscaled();
int upscaling_factor_x = nxp/nxp_upscaled;
int upscaling_factor_y = nyp/nyp_upscaled;
int upscaling_factor_z = nzp/nzp_upscaled;
// dimensions of one grid cell
double dx_upscaled = dx*upscaling_factor_x;
double dy_upscaled = dy*upscaling_factor_y;
int nx_upscaled = modelGravityStatic->GetNx_upscaled();
int ny_upscaled = modelGravityStatic->GetNy_upscaled();
int nz_upscaled = modelGravityStatic->GetNz_upscaled();
int N_upscaled = nx_upscaled*ny_upscaled*nz_upscaled;
int N_fullsize = nx*ny*nz;
int nObs = 30;
G_ .resize(nObs, N_upscaled);
G_fullsize_.resize(nObs, N_fullsize);
// Need to be in real domain for transforming from log domain
if(expMeanAlpha->getIsTransformed())
expMeanAlpha->invFFTInPlace();
if(meanAlphaFullSize->getIsTransformed())
meanAlphaFullSize->invFFTInPlace();
float sigma_squared = GravimetricInversion::GetSigmaForTransformation(seismicParameters.GetCovAlpha());
GravimetricInversion::MeanExpTransform(expMeanAlpha, sigma_squared);
GravimetricInversion::MeanExpTransform(meanAlphaFullSize, sigma_squared);
//Smooth (convolve) and subsample
FFTGrid * upscalingKernel_conj = modelGravityStatic->GetUpscalingKernel();
if(upscalingKernel_conj->getIsTransformed() == false)
upscalingKernel_conj->fftInPlace();
upscalingKernel_conj->conjugate(); // Conjugate only in FFT domain.
// Need to be in FFT domain for convolution and subsampling
if(expMeanAlpha->getIsTransformed() == false)
expMeanAlpha->fftInPlace();
expMeanAlpha->multiply(upscalingKernel_conj); // Now is expMeanAlpha smoothed
FFTGrid * upscaledMeanAlpha;
GravimetricInversion::Subsample(upscaledMeanAlpha, expMeanAlpha, nx_upscaled, ny_upscaled, nz_upscaled, nxp_upscaled, nyp_upscaled, nzp_upscaled);
upscaledMeanAlpha->invFFTInPlace();
float x0, y0, z0; // Coordinates for the observations points
int J = 0; // Index in matrix counting cell number
int I = 0;
float vp;
double dt;
double localMass;
double localDistanceSquared;
for(int i = 0; i < nObs; i++){
x0 = observation_location_utmx_[i];
y0 = observation_location_utmy_[i];
z0 = observation_location_depth_[i];
J = 0; I = 0;
// Loop through upscaled simbox to get x, y, z for each grid cell
for(int ii = 0; ii < nx_upscaled; ii++){
for(int jj = 0; jj < ny_upscaled; jj++){
for(int kk = 0; kk < nz_upscaled; kk++){
double x, y, z;
vp = upscaledMeanAlpha->getRealValue(ii,jj,kk);
int istart = ii*upscaling_factor_x;
int istop = (ii+1)*upscaling_factor_x;
int jstart = jj*upscaling_factor_y;
int jstop = (jj+1)*upscaling_factor_y;
int kstart = kk*upscaling_factor_z;
int kstop = (kk+1)*upscaling_factor_z;
double x_local = 0;
double y_local = 0;
double z_local = 0;
double dt_local = 0;
//Find center position of coarse grid cell using indices of fine grid and averaging their cell centers.
for(int iii=istart; iii<istop; iii++){
for(int jjj=jstart; jjj<jstop; jjj++){
for(int kkk=kstart; kkk<kstop; kkk++){
fullSizeTimeSimbox->getCoord(iii, jjj, kkk, x, y, z);
x_local += x;
y_local += y;
z_local += z; // NB Need time depth mapping!
dt_local += fullSizeTimeSimbox->getdz(iii, jjj);
}
}
}
x_local /= (upscaling_factor_x*upscaling_factor_y*upscaling_factor_z);
y_local /= (upscaling_factor_x*upscaling_factor_y*upscaling_factor_z);
z_local /= (upscaling_factor_x*upscaling_factor_y*upscaling_factor_z);
dt_local /= (upscaling_factor_x*upscaling_factor_y*upscaling_factor_z);
// Find fraction of dx_upscaled and dy_upscaled according to indicies
double xfactor = 1;
if(istop <= nx){ // inside
xfactor = 1;
}
else if(istart > nx){ //outside
xfactor = 0;
}
else{
xfactor = (nx - istart)/upscaling_factor_x;
}
double yfactor = 1;
if(jstop <= ny){
yfactor = 1;
}
else if(jstart > ny){
yfactor = 0;
}
else{
yfactor = (ny - jstart)/upscaling_factor_y;
}
localMass = (xfactor*dx_upscaled)*(yfactor*dy_upscaled)*dt_local*vp*0.5*1000; // units kg
localDistanceSquared = pow((x_local-x0),2) + pow((y_local-y0),2) + pow((z_local-z0),2); //units m^2
G_(i,J) = localMass/localDistanceSquared;
J++;
}
}
}
// Loop through full size simbox to get x, y, z for each grid cell
for(int ii = 0; ii < nx; ii++){
for(int jj = 0; jj < ny; jj++){
for(int kk = 0; kk < nz; kk++){
double x, y, z;
fullSizeTimeSimbox->getCoord(ii, jj, kk, x, y, z); // assuming these are center positions...
vp = meanAlphaFullSize->getRealValue(ii, jj, kk);
dt = fullSizeTimeSimbox->getdz(ii, jj);
localMass = dx*dy*dt*vp*0.5*1000; // units kg
localDistanceSquared = pow((x-x0),2) + pow((y-y0),2) + pow((z-z0),2); //units m^2
G_fullsize_(i,I) = localMass/localDistanceSquared;
I++;
}
}
}
}
G_ = G_*gamma;
G_fullsize_ = G_fullsize_*gamma;
}
PosteriorElasticPDF3D::PosteriorElasticPDF3D(const std::vector<double> & d1, // first dimension of data points
const std::vector<double> & d2, // second dimension of data points
const std::vector<double> & d3, // third dimension of data points
const std::vector<int> & t1, // trend data points
const std::vector<std::vector<double> > & v, // Transformation of elastic variables from 3D to 2D
const double *const*const sigma, // Gaussian smoothing kernel in 2D
int n1, // resolution of density grid in elastic dimension 1
int n2, // resolution of density grid in elastic dimension 2
int n3, // resolution of density grid in the trend dimension
double d1_min,
double d1_max,
double d2_min,
double d2_max,
double t1_min,
double t1_max,
int ind)
: n1_(n1),
n2_(n2),
n3_(n3),
x_min_(d1_min),
x_max_(d1_max),
y_min_(d2_min),
y_max_(d2_max),
z_min_(t1_min),
z_max_(t1_max)
{
// We assume that the input vectors are of the same length
if (d1.size()!=d2.size() || d2.size()!=d3.size() || d3.size()!=t1.size())
throw NRLib::Exception("Facies probabilities: Size of input vectors do not match.");
if (v.size()>2 || static_cast<int>(v[0].size()) != 3 || static_cast<int>(v[1].size()) != 3)
throw NRLib::Exception("Facies probabilities: Transformation matrix v does not have the right dimensions");
v1_.resize(3);
v2_.resize(3);
for(int i=0; i<3; i++) {
v1_[i] = v[0][i];
v2_[i] = v[1][i];
}
int dim = static_cast<int>(d1.size());
std::vector<std::vector<double> > x(2);
x[0].resize(dim);
x[1].resize(dim);
//computes x and y from d1, d2 and d3
CalculateTransform2D(d1, d2, d3, x, v);
histogram_ = new FFTGrid(n1_, n2_, n3_, n1_, n2_, n3_);
histogram_->createRealGrid(false);
int rnxp = histogram_->getRNxp();
histogram_->setType(FFTGrid::PARAMETER);
histogram_->setAccessMode(FFTGrid::WRITE);
for(int l=0; l<n3_; l++) {
for(int k=0; k<n2_; k++) {
for(int j=0; j<rnxp; j++)
histogram_->setNextReal(0.0f);
}
}
histogram_->endAccess();
// Spacing variables in the density grid
dx_ = (x_max_ - x_min_)/n1_;
dy_ = (y_max_ - y_min_)/n2_;
dz_ = (z_max_ - z_min_)/n3_;
// Go through data points and place in bins in histogram
for (int i = 0; i < dim; i++) {
//volume->getIndexes(d1[i], d2[i], d3[i], i_tmp, j_tmp, k_tmp);
int i_tmp = static_cast<int>(floor((x[0][i]-x_min_)/dx_));
int j_tmp = static_cast<int>(floor((x[1][i]-y_min_)/dy_));
int k_tmp = t1[i];
// Counting data points in index (i,j,k)
histogram_->setAccessMode(FFTGrid::RANDOMACCESS);
histogram_->setRealValue(i_tmp, j_tmp, k_tmp, histogram_->getRealValue(i_tmp,j_tmp,k_tmp) + 1.0f);
histogram_->endAccess();
}
//multiply by normalizing constant for the PDF - dim is the total number of entries
histogram_->setAccessMode(FFTGrid::READANDWRITE);
histogram_->multiplyByScalar(float(1.0f/dim));
histogram_->endAccess();
if(ModelSettings::getDebugLevel() >= 1) {
std::string baseName = "Hist_" + NRLib::ToString(ind) + IO::SuffixAsciiFiles();
std::string fileName = IO::makeFullFileName(IO::PathToDebug(), baseName);
histogram_->writeAsciiFile(fileName);
}
histogram_->fftInPlace();
double **sigma_tmp = new double *[2];
for (int i=0; i<2; i++) {
sigma_tmp[i] = new double[2];
}
for(int i=0; i<2; i++) {
for(int j=0; j<2; j++)
sigma_tmp[i][j] = sigma[i][j];
}
// Matrix inversion of the covariance matrix sigma
// SINGULAR MATRIX ?!
double **sigma_inv = new double *[2];
for(int i=0; i<2; i++)
sigma_inv[i] = new double [2];
InvertSquareMatrix(sigma_tmp,sigma_inv,2);
FFTGrid *smoother = new FFTGrid(n1, n2, n3, n1, n2, n3);
smoother->createRealGrid(false);
smoother->setType(FFTGrid::PARAMETER);
smoother->setAccessMode(FFTGrid::WRITE);
for(int l=0; l<n3_; l++) {
for(int k=0; k<n2_; k++) {
for(int j=0; j<static_cast<int>(smoother->getRNxp()); j++)
smoother->setNextReal(0.0f);
}
}
smoother->endAccess();
//SetupSmoothingGaussian2D(smoother, sigma_inv, n1, n2, n3, dx_, dy_);
if(ModelSettings::getDebugLevel() >= 1) {
std::string baseName = "Smoother" + IO::SuffixAsciiFiles();
std::string fileName = IO::makeFullFileName(IO::PathToDebug(), baseName);
smoother->writeAsciiFile(fileName);
}
// Carry out multiplication of the smoother with the density grid (histogram) in the Fourier domain
smoother->fftInPlace();
histogram_->multiply(smoother);
histogram_->invFFTInPlace();
histogram_->multiplyByScalar(sqrt(float(n1_*n2_*n3_)));
histogram_->endAccess();
delete smoother;
for(int i=0; i<2; i++) {
delete [] sigma_inv[i];
delete [] sigma_tmp[i];
}
delete [] sigma_tmp;
delete [] sigma_inv;
}