Global, high-resolution mapping of tropospheric ozone – explainable machine learning and impact of uncertainties

Betancourt, Clara; Stomberg, Timo T.; Edrich, Ann-Kathrin; Patnala, Ankit; Schultz, Martin; Roscher, Ribana; Kowalski, Julia; Stadtler, Scarlet

doi:10.5194/gmd-2022-2

Cited by 3 publications

(1 citation statement)

References 37 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The causes of multi-model diversity highlighted in previous studies (Young et al, 2018;Mortier et al, 2020;Griffiths et al, 2021) can also be elucidated using machine learning. There is an increase in the availability of globally gridded fused model-observation data products (e.g., Randles et al, 2017;Buchard et al, 2017;Inness et al, 2019;Betancourt et al, 2021;van Donkelaar et al, 2021;Betancourt et al, 2022) that can be used as benchmarks in model evaluation of atmospheric composition. Novel aspects of such benchmarks include providing data relevant to health impacts (e.g., DeLang et al, 2021) and using machine learning techniques for global mapping of atmospheric composition (e.g., Betancourt et al, 2022).…”

Section: Machine Learning Where Neededmentioning

confidence: 99%

Interactions between atmospheric composition and climate change – Progress in understanding and future opportunities from AerChemMIP, PDRMIP, and RFMIP

Fiedler

Naik

O’Connor

et al. 2023

Preprint

View full text Add to dashboard Cite

Abstract. The climate science community aims to improve our understanding of climate change due to anthropogenic influences on atmospheric composition and the Earth's surface. Yet not all climate interactions are fully understood and diversity in climate model experiments persists as assessed in the latest Intergovernmental Panel on Climate Change (IPCC) assessment report. This article synthesizes current challenges and emphasizes opportunities for advancing our understanding of climate change and model diversity. The perspective of this article is based on expert views from three multi-model intercomparison projects (MIPs) – the Precipitation Driver Response MIP (PDRMIP), the Aerosol and Chemistry MIP (AerChemMIP), and the Radiative Forcing MIP (RFMIP). While there are many shared interests and specialisms across the MIPs, they have their own scientific foci and specific approaches. The partial overlap between the MIPs proved useful for advancing the understanding of the perturbation-response paradigm through multi-model ensembles of Earth System Models of varying complexity. It specifically facilitated contributions to the research field through sharing knowledge on best practices for the design of model diagnostics and experimental strategies across MIP boundaries, e.g., for estimating effective radiative forcing. We discuss the challenges of gaining insights from highly complex models that have specific biases and provide guidance from our lessons learned. Promising ideas to overcome some long-standing challenges in the near future are kilometer-scale experiments to better simulate circulation-dependent processes where it is possible, and machine learning approaches for faster and better sub-grid scale parameterizations where they are needed. Both would improve our ability to adopt a smart experimental design with an optimal tradeoff between resolution, complexity and simulation length. Future experiments can be evaluated and improved with sophisticated methods that leverage multiple observational datasets, and thereby, help to advance the understanding of climate change and its impacts.

show abstract

Section: Machine Learning Where Neededmentioning

confidence: 99%

Interactions between atmospheric composition and climate change – Progress in understanding and future opportunities from AerChemMIP, PDRMIP, and RFMIP

Fiedler

Naik

O’Connor

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

Explainable Machine Learning Reveals Capabilities, Redundancy, and Limitations of a Geospatial Air Quality Benchmark Dataset

Stadtler

Betancourt

Roscher

2022

MAKE

View full text Add to dashboard Cite

Air quality is relevant to society because it poses environmental risks to humans and nature. We use explainable machine learning in air quality research by analyzing model predictions in relation to the underlying training data. The data originate from worldwide ozone observations, paired with geospatial data. We use two different architectures: a neural network and a random forest trained on various geospatial data to predict multi-year averages of the air pollutant ozone. To understand how both models function, we explain how they represent the training data and derive their predictions. By focusing on inaccurate predictions and explaining why these predictions fail, we can (i) identify underrepresented samples, (ii) flag unexpected inaccurate predictions, and (iii) point to training samples irrelevant for predictions on the test set. Based on the underrepresented samples, we suggest where to build new measurement stations. We also show which training samples do not substantially contribute to the model performance. This study demonstrates the application of explainable machine learning beyond simply explaining the trained model.

show abstract

Global, high-resolution mapping of tropospheric ozone – explainable machine learning and impact of uncertainties

Betancourt¹,

Stomberg

Edrich

et al. 2022

Geosci. Model Dev.

Self Cite

View full text Add to dashboard Cite

Abstract. Tropospheric ozone is a toxic greenhouse gas with a highly variable spatial distribution which is challenging to map on a global scale. Here, we present a data-driven ozone-mapping workflow generating a transparent and reliable product. We map the global distribution of tropospheric ozone from sparse, irregularly placed measurement stations to a high-resolution regular grid using machine learning methods. The produced map contains the average tropospheric ozone concentration of the years 2010–2014 with a resolution of 0.1∘ × 0.1∘. The machine learning model is trained on AQ-Bench (“air quality benchmark dataset”), a pre-compiled benchmark dataset consisting of multi-year ground-based ozone measurements combined with an abundance of high-resolution geospatial data. Going beyond standard mapping methods, this work focuses on two key aspects to increase the integrity of the produced map. Using explainable machine learning methods, we ensure that the trained machine learning model is consistent with commonly accepted knowledge about tropospheric ozone. To assess the impact of data and model uncertainties on our ozone map, we show that the machine learning model is robust against typical fluctuations in ozone values and geospatial data. By inspecting the input features, we ensure that the model is only applied in regions where it is reliable. We provide a rationale for the tools we use to conduct a thorough global analysis. The methods presented here can thus be easily transferred to other mapping applications to ensure the transparency and reliability of the maps produced.

show abstract

Global, high-resolution mapping of tropospheric ozone – explainable machine learning and impact of uncertainties

Cited by 3 publications

References 37 publications

Interactions between atmospheric composition and climate change – Progress in understanding and future opportunities from AerChemMIP, PDRMIP, and RFMIP

Interactions between atmospheric composition and climate change – Progress in understanding and future opportunities from AerChemMIP, PDRMIP, and RFMIP

Explainable Machine Learning Reveals Capabilities, Redundancy, and Limitations of a Geospatial Air Quality Benchmark Dataset

Global, high-resolution mapping of tropospheric ozone – explainable machine learning and impact of uncertainties

Contact Info

Product

Resources

About