Diagnosing Drainage Problems in Coastal Areas Using Machine‐Learning and Geostatistical Models

Abstract

This study focuses on diagnosing drainage problems in the coastal areas of Iran by using geostatistical methods, support vector machines (SVMs) and the adaptive neuro‐fuzzy inference system (ANFIS). Groundwater level (WD) and quality were monitored at 37 shallow wells scattered over a 25 000 ha area at different times. Using prepared raster maps of pH, ESP, EC and WD by the best method, drainage problems were categorized into eight classes. Both SVM and ANFIS models significantly improved predicted data for pH, ESP, EC and WD compared with geostatistical models, while SVMs provided slightly better results which were used for further analysis. More than 60% of the area needs drainage to lower the groundwater table in pre‐planting and post‐harvest periods, while during the growing seasons, more than about 72% of the area requires drainage for salinity control. Based on the results, identifying drainage problems at basin scale is possible with cost‐efficient machine learning models with minimum time and data requirements and investment for detailed field surveys. Copyright © 2017 John Wiley & Sons, Ltd.

Stone weir scour modelling in curved canals using a weighted regularized extreme learning machine*

“…They came to the conclusion that the bend value of the curved channel significantly influences the dimensions of scour depth. Moreover, numerical models, artificial intelligence (AI) tools and machine learning (ML) algorithms have been extensively used to estimate and model the scour pattern around various structures (Ebtehaj et al ., 2015; Akhbari et al ., 2017; Darzi‐Naftchali et al ., 2017). For instance, Azimi et al .…”