1.2 Machine learning methods
ML methods are part of artificial intelligence (AI) which is a broad
term for tools that mimic cognitive human capabilities. The use of AI
has rapidly increased in recent years. The number of peer-reviewed
publications across all fields between 2000 and 2019 has grown around 12
times (D. Zhang et al., 2021) and with them, multiple algorithms,
architectures, and tools have been created. Fields in which ML methods
have shown outstanding results include computer vision, speech
recognition, and language processing. Most of these applications use
supervised learning, which identifies a branch of ML that is similar to
RS metamodelling. Supervised ML employs a set of input-output examples,
also known as the labelled training dataset, to calibrate a model by
minimizing the error between the model predictions and the values
assumed as ground truth. This set of algorithms usually increase their
performance at a given task as the amount of labelled examples grows
larger. Due to their successes, supervised ML methods, and in particular
deep learning (DL) and artificial neural networks (ANNs), are widely
employed for surrogate modelling across many fields of science and
engineering (Liu et al., 2021; Peng et al., 2020; Wu et al., 2020).
Although scientific studies on ML applications for water resources date
back to over two decades ago (Maier & Dandy, 2000), Hadjimichael et al.
(2016) noted that this trend is not necessarily witnessed in the urban
water sector.