Sample synthetic dataset

This direcotry contains synthetic dataset for testing RF_INV. Note that the dataset here is paricuraly 'easy' dataset that are tuned such that one can obtain resonable solutions quickly. For application to real dataset, you will need ~100 times more iterations. DON'T copy & paste the parameter file here for your dataset without noticing this.

• Uniform reference velocity model (Vp = 5.0 km/s, Vs = 2.89 km/s)
• Vp is fixed
• Noise level (sigma) is fixed
• Multiple inputs with different ray parameters

Usage

1. mkdir rslt
2. mpirun -np 20 ../bin/rf_inv params.in
3. python ../util/plot.py params.in
4. display rslt/plot01.png or display rslt/plot02.png

Output example

Implication from this experiment

We see dual peaks in the posterior probability distribution of the number of layer interfaces (the topmost panel in the figure). The second peak at k = 9 is due to overfit to noise. This non-uniquenss of the solutions is always problematic for receiver function inversion! One may mitigate this by employing more tight prior constraints.

True model

See true.true.velmod for the true velocity model.