Research On Forward Modeling And Inversion Of Surface Nuclear Magnetic Resonance For Exploring Groundwater

Since Nuclear Magnetic Resonance (NMR) was discovered in 1946, it has already been in the stage of applying and developing from theoretical research and practice and such technology has been applied into the field of physics, chemistry, biology, medicine science, etc. Extensively, in geosciences field, it is widely used even. It is a new field in the geosciences for adding the new means in groundwater exploring to use NMR technology, because it has filled the gap that looking for groundwater directly in using geophysics method.This dissertation is based on the theory of NMR. By building the single water-saturated layer model and multi-water-saturated layers model, NMR forward calculation is done in the simulated geomagnetic field. Basing on NMR forward calculation, all sorts of elements that influence the initial amplitude of NMR forward signal are researched detailedly. It is found that the initial amplitude of NMR forward signal is associated with the depth of aquifer, the thickness of aquifer, the water content of aquifer, the excitation condition and receiving condition, the conductivity and magnetic permeability of underground medium and geomagnetic field. In research of inverse problem, not only the classical the Tikhonov Regularization algorithm is accepted, but also the Least Squares with a Quadratic Constraint algorithm is firstly introduced. By calculating the forward data from the model and the real data from the field and then comparing the inversion results with models and drilling data, it could be concluded that the inversion effect of the Least Squares with a Quadratic Constraint algorithm is better than the Tikhonov Regularization algorithm. At last, in research of how to improve the inversion accuracy, it is introduced that improving the signal-to-noise ratio by displacing the geomagnetic field with strong artificial magnetic field. It is proved reasonable in the model test that in the condition of different field sources, the model forward result is added noise of different scales and then inversed.