Testing-the-waters: prediction of high flow events in blanket bog catchments using MLA for automated event water sampling

Peatlands are estimated to cover approximately 20% of Ireland, with blanket bogs accounting for 13%. High-resolution hydrological monitoring is key to establishing baselines for determining impacts of blanket bog rewetting. Blanket bogs are prone to flashy runoff regimes, where storm flow accounts for the bulk of runoff and aqueous carbon export. However, their short time to peak can lead to delays in capturing entire event hydrographs. In addition, their often isolated location can hinder both access to and remote communication with devices. Machine learning algorithms (MLA) offers a dynamic, data driven method for catchment monitoring. High-resolution runoff, precipitation and barometric pressure data were collected over two years at Letterunshin, Co. Sligo. These data are used to develop a binary classification model using Support Vector Machines (SVM) to predict the occurrence of high flow events. This prediction together with our telecommunications network serves to trigger an autosampler array for targeted event sampling. Here, we present the process of data preparation, parameter selection and the development of a SVM algorithm with performance accuracy > 90%. This research underpins the management of water resources derived from blanket bogs, their runoff and water quality monitoring applications within an Irish context.