Sie befinden Sich nicht im Netzwerk der Universität Paderborn. Der Zugriff auf elektronische Ressourcen ist gegebenenfalls nur via VPN oder Shibboleth (DFN-AAI) möglich. mehr Informationen...
Accurate groundwater level (GWL) prediction is crucial for the management and sustainable utilization of groundwater resources. This study proposes a method, considering spatial–temporal correlation among geographic multi-feature in data, and Self-Organizing Map (SOM)-based clustering technique to identify and partition spatially connectivity among observation wells. Finally, based on the connectivity results, the observation well dataset is determined as inputs to LSTM for GWL prediction. This approach provides a new idea to enhance the accuracy of existing data-driven methods in karst critical zones characterized by significant spatial heterogeneity in GWL. Comparing with prediction models that solely consider internal data correlations, experiments are conducted in the typical highly spatially heterogeneous karst critical zone of Jinan City, Shandong Province, China. The results show a significant improvement in prediction accuracy when considering spatial connectivity between observation wells based on geographical multi-feature spatial–temporal correlation. Confirming that considering the spatial connectivity of observation wells in GWL prediction methods are more accurate, particularly in areas with significant spatial heterogeneity in karst aquifers.