Advantages and issues of applying Machine learning based denosing on inversions of the DPRK nuclear tests

We apply a current state-of-the-art machine learning based denoising algorithm on the seismological and hydroacoustic waveform records of the selected DPRK nuclear tests. We use the DeepDenoiser algorithm to reduce the noise present in the waveform records of the larger DPRK nuclear tests. The denosing of waveform records using machine learning has obvious advantages on the picking of phases and signal detection but the question is if the currently available techniques can be used beyond that. We investigate the impact the denoising has on the source mechanism inferences by comparing the seismic moment tensor inversion results of original and denoised data. Because of the good signal to noise ratio and as the source type is well known we can in this cases establish if the so denoised waveforms can be used for further source analysis. We find that care needs to be taken using the modified waveform data but also find promising results hinting at possible further use the technique in the future for standard analyses. We further investigate if the application of the chosen denoising algorithm allows for the better resolution of the seismic moment tensor of the smaller DPRK nuclear tests.