A Brief Review of Random Forests for Water Scientists and Practitioners and Their Recent History in Water Resources

Abstract

Random forests (RF) is a supervised machine learning algorithm, which has recently started to gain prominence in water resources applications. However, existing applications are generally restricted to the implementation of Breiman’s original algorithm for regression and classification problems, while numerous developments could be also useful in solving diverse practical problems in the water sector. Here we popularize RF and their variants for the practicing water scientist, and discuss related concepts and techniques, which have received less attention from the water science and hydrologic communities. In doing so, we review RF applications in water resources, highlight the potential of the original algorithm and its variants, and assess the degree of RF exploitation in a diverse range of applications. Relevant implementations of random forests, as well as related concepts and techniques in the R programming language, are also covered.

Gap‐filling approaches for eddy covariance methane fluxes: A comparison of three machine learning algorithms and a traditional method with principal component analysis

“…Furthermore, RF is straightforward to use compared to ANN (e.g., no need for stopping rules, no need for scaling, and less parameter requirements; cf. Tyralis, Papacharalampous, & Langousis, ).…”

“…The first row shows the results when machine learning (ML) algorithms were trained using data from all years. The second row is the results when they were trained separately for each year [Colour figure can be viewed at wileyonlinelibrary.com] no need for stopping rules, no need for scaling, and less parameter requirements; cf Tyralis, Papacharalampous, & Langousis, 2019)…”

Estimation of high frequency nutrient concentrations from water quality surrogates using machine learning methods

“…The well-known machine learning algorithm called Random Forests (RF) (Breiman, 2001) averages the result of many independent individual trees with a random subset of data and predictors in order to mitigate the poor performance of any single tree. A comprehensive review of its implementation in the water science field has been recently published (Tyralis et al, 2019). Details about the implementation of decision trees in this work are given in the Annex (section A.2), in case the reader is not familiar with RF.…”

Quantification of predictive uncertainty in hydrological modelling by harnessing the wisdom of the crowd: Methodology development and investigation using toy models

“…The same applies for investigations using large datasets. In fact, only such investigations allow the inspection of the impact of the various modelling 'tricks' to the quantification of predictive uncertainty, as suggested by well-established guidelines for machine learning-based approaches (see e.g., Tyralis et al 2019a, section 1). For extensive discussions on the subject, the interested reader is referred to Papacharalampous et al (2019b, section 5) and the references therein.…”

“…The proposed methodology is characterized by some additional strengths that are also particularly important from a predictive modelling point of view. First, it is computationally convenient in the sense that it can be easily expressed in algorithmic Tyralis et al 2019a). Lastly, it allows the exploitation of the total amount of available information, in the sense that each sister prediction is herein converted into a probabilistic prediction (consisted of several quantile predictions) instead of a single simulation (randomly extracted from its predictive PDF; see the utilization of the meta-Gaussian bivariate distribution model in Montanari and Koutsoyiannis 2012; see also Kelly and Krzysztofowicz 1997).…”