Forecasting groundwater anomaly in the future using satellite information and machine learning

• Published in: Journal of hydrology (Amsterdam) Vol. 612; p. 128052
• Main Authors: Soltani, Keyvan, Azari, Arash
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.09.2022
• Subjects: Climate change; GRACE and GRACE-FO; GroundWater Anomaly (GWA); Group Method of Data Handling (GMDH); Lake Urmia Basin (LUB); Terrestrial Water Storage Anomaly (TWSA); Climate change; GRACE and GRACE-FO; Terrestrial Water Storage Anomaly (TWSA); GroundWater Anomaly (GWA); Group Method of Data Handling (GMDH); Lake Urmia Basin (LUB)
• ISSN: 0022-1694; 1879-2707
• DOI: 10.1016/j.jhydrol.2022.128052

[Display omitted] •A machine learning model based remote sensing and terrestrial data was developed.•Fill the data gap in GRACE and GFO satellites data by machine learning.•Soil Moisture, temperature and evapotranspiration was estimated by the CanESM2.•Terrestrial water storage and groundwater anomaly were predicted for the future. By applying Gravity Recovery and Climate Experiment (GRACE) and GRACE Follow-On (GFO) satellites, this study estimates the value of Terrestrial Water Storage Anomaly (TWSA) and GroundWater Anomaly (GWA) in Lake Urmia Basin (LUB), Iran. TWSA changes are first calculated using GRACE/GFO satellites where the best Gaussian Filter Radius in Kilometer (GFRK) with the least error is obtained. To provide a model and a practical convenient equation for modeling the value of TWSA in the study area, six parameters affected the value of TWSA is used. These data are obtained from two data centers: Global Land Data Assimilation System (GLDAS) and fifth generation (V5) European Center for Medium-Range Weather Forecasts (ECMWF) Reanalysis (ERA5). This information is obtained in the Google Earth Engine (GEE) environment by developing JavaScript code for the study area and on a monthly basis for 107 points that have a suitable distribution in the area. The TWSA value is also calculated using GRACE/GFO data satellites for 10 pixels that completely cover the area. Next, Machine Learning (ML) is used to model the TWSA value. The best model is obtained for 2–6 inputs in calculating the TWSA value. The results indicate that the use of three inputs of Soil Moisture in 10–40 cm (SM2), Average Temperature (AT), and Evapotranspiration (ET) present the best performance compared to other models. Finally, GRACE/GFO missing data are obtained using this equation. To calculate the TWSA value in the future, three input parameters are downscaled using the second-generation Canadian Earth System Model (CanESM2) to predict the TWSA value. The results of climate change show that the value of TWSA in the two periods 2021–2040 and 2041–2060 has a reduction compared to the average measurement period. Also, due to TWSA changes, the value of GWA would be drastically reduced in the future.