Modelling inundated area in wetlands combining satellite and hydrological data: A comparison of classical methods and machine learning algorithms

Wetlands, which are systems with significant environmental value, can be very sensitive to global change. The inundated area of wetlands, reflecting water quantity, stands as a key variable in the decision-making process for evaluating sustainable management strategies in these ecosystems.
Satellite optical sensors are effective for regional and global surface water monitoring. However, depending on the satellite, they may not offer a comprehensive long-term time series of inundated areas to study the effects of global change. This limitation arises from factors such as presence of clouds, sensor failure, low revisit time or spatial resolution, or recent launch. 
We propose leveraging hydro-climatological data to enhance and complement satellite-observed inundated area dynamics. In this approach, we evaluate the effectiveness of classical methods such as ARIMA and multiple regression models, along with advanced techniques like artificial autoregressive neural networks and other machine learning algorithms. The goal is to integrate covariate information and simulate extensive and continuous inundated area time series. This methodology is valuable not only for filling gaps in observational data but also for projecting the impacts of climate change on inundated area in wetlands.
The suggested methodology was implemented in the Lagunas de Ruidera wetland area in south-eastern Spain. This region exhibits a significant natural interplay between groundwater and surface water, highlighting a conflict between groundwater-dependent ecosystems and groundwater extraction for irrigation. From January to June, the average observed inundated area is approximately 4.3 km². In summer, there is a reduction of about 13% in the surface water area, which is subsequently recovered during the autumn.

Acknowledgments: This research has been partially supported by the project SIGLO-PRO (PID2021-128021OB-I00) funded by the Spanish Ministry of Science, Innovation and Universities and the project C17.i7.CSIC – CLI 2021-00-000 funded byEuropean Union NextGenerationEU/PRTR.