Fusion of multi-source data and artificial intelligence for land subsidence evaluation

Excessive groundwater extraction in Iran, especially in the Neyshabur region, has led to widespread land subsidence (LS), posing a serious threat to the environment and infrastructure. This study aims to investigate LS and assess the consistency of multi-source data to improve the accuracy and reliability of the results through data fusion. The datasets include Sentinel-1 satellite images (2014 to 2020), global positioning system (GPS) time series, precision leveling, gravity recovery and climate experiment (GRACE) and its follow-on GRACE-FO satellite data, and piezometric well data. The results show that the maximum LS rate in Neyshabur is 21 ± 2 cm per year, consistent with other observations. The linear trend in the GPS data is -8 cm per year, and the maximum value recorded by precision leveling is -15 cm per year. The annual decrease in groundwater storage was 0.13 km³, and the maximum water table decline recorded in piezometric wells was 2.5 m/year. The correlation between GPS and leveling data exceeds 0.9. To assess the LS risk, an index was developed using optimization algorithms including genetic algorithm (GA), genetic programming (GP), particle swarm optimization (PSO) and firefly algorithm (FA). The correlations of these methods with the data are 0.80, 0.82, 0.73 and 0.65, respectively, with GP showing the highest agreement. These results confirm the high capability of combining multi-source data and artificial intelligence in monitoring and sustainable water management.