Bayesian modeling of velocity break points in GNSS time series and the effect of noise on their estimation: Did velocity anomalies in the Krafla volcanic system, north Iceland, precede the Bárðarbunga-Holuhraun 2014-2015 rifting episode?

Correct estimation of the timing of velocity changes (break points) and associated uncertainties in ground deformation observed with Global Navigation Satellite System (GNSS) coordinate time series is crucial for understanding various Earth processes and how they may couple with each other. To simultaneously estimate break points, velocity changes and their uncertainties, we implement Bayesian modeling with Markov Chain Monte Carlo algorithm for GNSS time series. As the presence of white noise (WN) and time-correlated flicker noise (FLN) in GNSS time series was found to affect velocity estimation, synthetic data experiments are first conducted to investigate their effect on break point estimation. The results indicate that reliable estimates are obtained only when the value of velocity change is larger than FLN amplitude. With the presence of WN and FLN, whose amplitudes are one twentieth and one fourth of the velocity-change value, the estimation bias and uncertainty are <0.5 mm/yr and ~5 mm/yr for velocity change, and <30 d and ~100 d for break point, respectively. In this case the uncertainty is one magnitude larger than that with only the presence of WN. Then the proposed method is applied to model two velocity changes detected manually during 2014-2015 at the Krafla volcanic system, North Volcanic Zone (NVZ), Iceland. Similar accuracy and precision as the synthetic data experiments can be expected in east component of the real data as the velocity-change values are 6.9-16.5 times of the WN amplitudes and 2.5-4.0 times of the FLN amplitudes from preliminary analysis. Considering the uncertainty estimated with 95% confidence interval, the first break point at the three continuous GNSS stations in the Krafla area suggests a change in extension pattern across the NVZ prior to the beginning of a major rifting episode that started on 16 August 2014 at the Bárðarbunga volcanic system, which is ~130 km south of Krafla. The first break point at KRAC station in the Krafla caldera occurs on 2-4 July 2014, with 95% confidence interval being 4 May to 13 August 2014. The first velocity change is about 7.6 to 9.8 mm/yr to the west with its uncertainty ranging from 4.5 to 14.4 mm/yr. The velocities approximately resume to the original level after the second change at the end of 2014 or early 2015, whose chronological relationship with the end of Bárðarbunga-Holuhraun episode cannot be asserted because of uncertainties. The results may indicate coupling of activities between the volcanic systems in the NVZ via processes not well understood. Further work is needed to confirm these results and their significance.