Reconstructing missing InSAR data by the application of machine leaning-based prediction models: a case study of Rieti

Remote sensing is undoubtedly one of the most practical approaches used in many fields such as structural health monitoring. The most notable deal of this method, however, is the data which are either missing or not given by satellites. In this study, an attempt has been made to propose machine learning (ML)-based models for reconstructing missing InSAR data (i.e., buildings' displacement) to monitor their performance more properly. To this end, buildings located in the historical center of Rieti, Rome (Italy), are considered as case studies. Displacement of the points situated on different buildings' roof, given by remote sensing, is utilized for training the relationship between inputs and outputs to the models. The input variables were points' coordinates, height, and soil condition, while the cumulative displacement was the target output. Tree-based techniques namely decision tree, random forest and XGBoost were implemented for developing prediction models. The accuracy of the models was assessed through common performance metrics and Taylor diagram, and the most accurate model was introduced accordingly. The results demonstrated high capability of the tree-based methods for estimating the displacement of buildings. The proposed prediction models could be used for either predicting or reconstructing missing displacement of buildings at any specific point which ease structural performance monitoring remarkably.

Automated summary

In this study, machine learning models were proposed to reconstruct missing InSAR data for monitoring the performance of buildings. Tree-based techniques were implemented and assessed for accuracy of displacement estimation. The proposed prediction models can be used for predicting or reconstructing missing displacement of buildings, facilitating structural performance monitoring.