Discussion on Assessing Predictability of Environmental Time Series With Statistical and Machine Learning Models - Open Access

Aug 3, 2026

Upload by Rowan

Building on Bonas et al. (2024), the discussion examines how statistical and machine learning models relate when analyzing environmental time series. It highlights key characteristics of such data, including the need for multivariate modeling and explicit handling of spatial dependence. The argument centers on the role of statistical inference in environmental studies beyond forecasting, proposing statistical and ML as complementary frameworks that can improve predictive accuracy while strengthening interpretability. A case study is used to illustrate these points.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages6

Reading Time15 min

AccessFree to download

Introduction

Environmental datasets for model comparison

Background: statistical time-series models

Background: machine learning time-series models

Evaluation approach (sliding window prediction)

Environmetrics DISCUSSION OPEN ACCESS Discussion on Assessing Predictability of Environmental Time Series With Statistical and Machine Learning Models Francesco Finazzi1 | Jacopo Rodeschini2 | Lorenzo Tedesco1 1 Department of Economics, University of Bergamo, Bergamo, Italy | 2 Department of Engineering and Applied Sciences, University of Bergamo, Bergamo, Italy Correspondence: Francesco Finazzi ([francesco.finazzi@unibg.it](francesco.finazzi@unibg.it)) Received: 5 November 2024 | Revised: 7 January 2025 | Accepted: 15 January 2025 Funding: This research was supported by the National Recovery and Resilience Plan (NRRP), Mission 4 Component 2 Investment 1.3, under Call for tender No. 341 of 15/03/2022, funded by the Italian Ministry of University and Research and the European Union’s NextGenerationEU initiative. Award Number: PE 00000018, Concession Decree No. 1558 of 11/10/2022, CUP F83C22001720001, under the project titled Growing Resilient Inclusive and Sustainable (GRINS) . Keywords: environmental modelling | forecasting | time series | uncertainty ABSTRACT Building on the insights from Bonas et al. (2024), we explore the relationship between statistical and machine learning models in the analysis of environmental time series. We specifically address the unique challenges of environmental time series data, including the need to consider the multivariate approach and account for spatial dependence. Emphasizing the importance of various types of statistical inference in environmental studies—not limited to forecasting—we propose that viewing statistical and machine learning approaches as complementary rather than alternative methods can unlock innovative modeling strategies that enhance both predictive accuracy and interpretive power. To illustrate these concepts, we present a case study that highlights the key points raised in the discussion. 1 | Introduction We would like to acknowledge the authors for their significant contribution to the examination of the relationship between statistical and machine learning (ML) models within the context of environmental time series. Their work raises an important question: should traditional statistical methods in environmental analysis be replaced by ML approaches? This question is followed by a comprehensive literature review that reveals a growing interest within the environmental statistics community in exploring the adaptability and potential of flexible ML models. The question is addressed by analyzing two distinct environmental datasets. The first dataset comprises daily measurements of PM2.5 (fine particulate matter smaller than 2.5 microns) concentrations from 12 air pollution monitoring stations (without any missing data) in the Golden Horseshoe region of Toronto, Canada. The data spans from January 1, 2009 to December 31, 2019, and allows for the evaluation of both predictive accuracy and uncertainty quantification, as the seasonal and environmental complexities provide a robust testing ground for comparing statistical and machine learning models. The second dataset consists of hourly meteorological data from a citizen science weather station near South Bend, Indiana, USA. The station records continuous measurements of temperature and wind speed, aggregated to an hourly level from July 20, 2020 to December 31, 2021 . This dataset features clear seasonal patterns, with temperature following typical mid-latitude seasonal variation and wind speed showing increased variance in fall and winter. This dataset provides another opportunity to assess model performance in capturing these dynamics, particularly in time series with shorter temporal resolution. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. © 2025 The Author(s). Environmetrics published by John Wiley & Sons Ltd. Environmetrics, 20

FAQs

What is the central question about statistical and machine learning models in environmental time series?

The discussion asks whether traditional statistical methods for environmental analysis should be replaced by machine learning approaches, or whether they can be used together effectively.

What challenges specific to environmental time series does the discussion emphasize?

It emphasizes multivariate considerations and accounting for spatial dependence, reflecting the structure of environmental measurements.

How are the models evaluated for predictability in the case study?

The discussion describes comparing forecasting performance using a sliding window prediction evaluation scheme and related uncertainty quantification needs.

Aug 3, 2026

Upload by Rowan