Usage of statistical modeling techniques in surface and groundwater level prediction

Klemen Kenda,Matej Čerin,Dimitris Kofinas,Jože Rožanec,Jože Peternelj,Nikos Mellios

doi:10.2166/aqua.2020.143

Abstract

Abstract The paper presents a thorough evaluation of the performance of different statistical modeling techniques in ground- and surface-level prediction scenarios as well as some aspects of the application of data-driven modeling in practice (feature generation, feature selection, heterogeneous data fusion, hyperparameter tuning, and model evaluation). Twenty-one different regression and classification techniques were tested. The results reveal that batch regression techniques are superior to incremental techniques in terms of accuracy and that among them gradient boosting, random forest and linear regression perform best. On the other hand, introduced incremental models are cheaper to build and update and could still yield good enough results for certain large-scale applications.

Highlights

Water data are becoming increasingly accessible and lowcost
This paper provides a comprehensive overview of the performance of statistical modeling techniques in applications of groundwater and surface water level forecast
Comparison of regression techniques with classification methods on discretized bins reveals that the classification techniques are significantly inferior

Summary

Introduction

Water data are becoming increasingly accessible and lowcost. Investments in improvement of data acquisition and data transfers have enabled significant growth of knowledge-intensive economies (Washburn et al ; Chourabi et al ; Di Nardo et al a). Water utilities worldwide are incorporating – or have already incorporated – the opportunity costs of capital, operation, maintenance, and environmental impacts to the final price under the Water management digitalization process is showing great potential for the usage of modern technologies such as the Internet of Things (IoT) and Artificial Intelligence (AI). The latter can operate as a catalyst for investigating, understanding, forecasting, and optimizing water usage, leakage, fraud, and pollutant detection, flooding and damage prevention and protection (Di Nardo et al b)

Methods

Results

Conclusion