Abstract
Simultaneous use of multiple independent data sets can improve constraints on earthquake source-model parameters. However, the ways in which data sets have been combined in the past are manifold and usually qualitative. In this paper we present a method to combine geodetic data in source model estimations, which includes characterizing the data errors and estimating realistic model-parameter uncertainties caused by these errors. We demonstrate this method in a case study of the June 2000 Kleifarvatn earthquake, which occurred on Reykjanes Peninsula in Iceland. We begin by showing to what extent additional data can positively influence the source modelling results, by combining both GPS and descending-orbit InSAR data, which were used in two earlier studies of that event, with InSAR data from an ascending orbit. We estimate the data error covariances of the InSAR observations and base the data weights in our model-parameter optimization on the corresponding data variance-covariance matrix. We also derive multiple sets of synthetic data errors from the estimated data covariances that we use to modify the original data to generate numerous data realizations. From these data realizations we estimate the model-parameter uncertainties. We first model the Kleifarvatn earthquake as a simple uniform-slip fault and subsequently as a fault with variable slip and rake. Our fault model matches well with the field observations of coseismic surface ruptures and its near-vertical dip (83 degrees) agrees with the regional faulting style as well as with aftershock locations. The two published source models of the event, on the other hand, both differ from our model as well as differing for one another. These studies, which were based on the descending InSAR data alone (the first study) and on that same data and GPS data (the second study), both neglect correlations in the InSAR data and do not report model-parameter uncertainties. Therefore, to compare these results with our model, we simulate the earlier model estimation set-ups and provide realistic estimates of the model-parameters uncertainties for these cases. We then discuss the significance of the difference between the existing fault models and demonstrate that both the inclusion of additional independent data as well as the covariance-based data weights improve the model-parameter estimation.