Assessing inversion uncertainty from initial model variability: A practical approach for geothermal MT exploration

We propose a simple and computationally efficient approach to evaluate the uncertainty in magnetotelluric (MT) inversion results arising from differences in initial models. Spatially smooth initial resistivity structures are generated by assigning random resistivity values to fixed representative points and applying Kriging-based interpolation. A series of 3D inversions was conducted using these initial models, and the variability in the resulting resistivity distributions was analyzed. Results from field data acquired in the Yuzawa geothermal field (northeastern Japan) demonstrate that even when using the same dataset and inversion parameters, the final resistivity models exhibit meaningful variability depending on the initial model. The selected cases with final RMS values ≤ 2.15 reveal spatial patterns in uncertainty: the smaller-area realizations were entirely encompassed within the larger-area realizations at a depth of 1000 m, indicating a conductive core region that is common to all realizations, while the spatial extent of C1 became more variable among realizations in the eastern region with increasing depth, reflecting greater uncertainty. This spatially variable uncertainty can guide the planning of drilling and geophysical surveys. Furthermore, the distribution of isosurface volumes below a given resistivity threshold enables the construction of optimistic and pessimistic scenarios for reservoir modeling. The proposed approach is scalable and well-suited for practical geothermal applications where efficient and feasible uncertainty assessment is required. This method provides a practical framework for evaluating geothermal potential.