Integrating interferometric SAR data with levelling measurements of land subsidence using geostatistics

Differential Synthetic Aperture Radar (SAR) interferometric (D-InSAR) data of ground surface deformation are affected by several error sources associated with image acquisitions and data processing. In this paper, we study the use of D-InSAR for quantifying land subsidence due to groundwater extraction. We model the data as the sum of a trend, a zero-mean stochastic process and white noise. A geostatistical approach combines D-InSAR subsidence data with in situ levelling measurements. The objective of this paper is to correct the errors contained in the D-InSAR measurements by using measurements as the ground data to improve their accuracy. Discrepancies between the true subsidence values and original D-InSAR measurements are analysed at levelling points using variograms and predicted at unvisited points using kriging. The integrated measurements are obtained by subtracting the predicted errors from the original D-InSAR measurements. The proposed method is applied to data collected in the Tianjin (China) area where land subsidence occurs due to groundwater extraction. Results demonstrate the capability of the D-InSAR technique for detecting subsidence at the centimetre level, and of using a limited number of levelling points to improve the accuracy of D-InSAR deformation measurements provided the coherence of images used is high enough. Discrepancies between the true subsidence values and D-InSAR measurements are quantified using the root mean square error (RMSE). RMSE for the original data was equal to 2.8, and 0.8 for the integrated data, whereas the mean error was equal to 2.1 and 0.0, respectively.