Trying to model and invert the data set

We are trying to model and invert the data set and tried to solve but not sure about it. need some better
solution for TDEM 1D inversion.

In attachment, a data file on which we are working right now as a test to better manage the code for both forward modeling and standard inversion. It is a 1D sounding . This data was collected along an ERT profile. The first test we want to do is to be able to model the data that should be recorded if we consider the ERT inversion as the true model, simplified as a 4-layer model (0-7 m : 2000 Ohm.m, 7-20 m: 120 Ohm.m, 20-45 m: 300 Ohm.m, 45- : 15 Ohm.m). We used a 25 x 25 square loop as transmitter/receiver. The theoretical input signal from the device is a square wave (thus a step-like shape) with ratio pulse/pause = 3/1, the dead-time is 4 µs, the intensity of the current was 3.4 A. but the true signal during turning off has a ramp. We would like to see the influence of this on the data and thus model both. We have seen that SIMPEG can model this type of signal, but are not sure how to proceed exactly.

We guess that if we are able to model this, this should get us started.

In attachment, there are some tentative code based on SimPEG published examples. In the first one, we just generate synthetic data that we invert. In the second, we are trying to invert the collected real field data. and

We have a few questions related to those:

  1. We are not sure how we can easily generate a grid which would fit specific boundaries as expected in our case. In the current example, we define 4 homogeneous layers but whose limits do not coincide exactly with the grid.
  2. There are a series of available source in SIMPEG, but we are not sure how to simulate correctly our source, which is a square loop 25 m x 25 m (for both transmitter and receiver). Here we used a circular loop of similar area. Similarly, we are not sure how to specify the receiver loop size.
  3. Our original data (see 1107_4.tem file) are expressed in V/A as recorded by the device (recorded voltage divided by input current intensity) as most TEM device do. However, it seems that SIMPEG is generating the vertical component of the magnetic field as calculated data. Therefore we tried to convert those back to V and Magnetic Field. However, it feels like this can probably be done within SIMPEG ? Is it possible to specify the type of data we want as input/output of the forward model or input of the inversion? Or should we always convert those as we tried here ?

In summary, we could use some help to get started with the proper use of the code.

Thanks in advance

1107_4.pdf (15.1 KB) TRAIL_1.pdf (424.6 KB) TRAIL_3.pdf (169.6 KB) TRAIL_4.pdf (169.5 KB) TRAIL_5.pdf (174.4 KB) TRIAL_2.pdf (426.2 KB)

1 Like

Hi @Arsalan, thanks for your questions! I will try to give you a few pointers to code and examples.

  1. I assume you are starting with TensorMesh meshes. These are the simplest from the user perspective for getting up and running. There are some tutorials available that discuss setting up meshes. For a 3D mesh,

    from discretize import TensorMesh
    mesh = TensorMesh([hx, hy], x0='CC')

    Here hx and hy can be vectors of cell widths. So inputting hx = [1, 2, 3] will generate a mesh whose first cell is 1m wide, the second cell is 2m wide and the third is 3m wide. So if you want to design a mesh that conforms to your layers, then you can refine / adjust as you see fit.

  2. I think @sgkang09 has a few examples. Are you looking to run a 1D simulation or a 3D one. If you are planning to start with 1D, I would recommend taking a look at SimPEGEM1D

  3. If you are using a dbdt receiver, then you can provide voltages. This conversation on input data is quite relevant here (and it sounds like you might be working with a similar survey setup to @jcbarreto!)

I hope this is a helpful starting point. I will circulate this post and see what other relevant examples we can point you to.

1 Like

Hi Iheagy thanks for your reply it is helpful and highly appreciated. And yes please circulate this post.

Thank you


Hi @Arsalan, I’m trying to invert TEMfast data, if you wish, we can talk more about that :blush:

1 Like

Hi jcbarreto thank you . Yes you can ask any question