A causal flow approach for the evaluation of global climate models

Abstract

Global climate models (GCMs) are generally used to forecast weather, understand the present climate, and project climate change. Their reliability usually rests on their capability to represent climatic processes, and most evaluations directly measure the spatiotemporal agreement of scalar climate variables. However, climate naturally involves complex interactions that are hard to infer and, therefore, difficult to evaluate. Climate networks (CNs) have been used to infer flows of mass and energy in the complex climate system. Here, an Evaluation of Models by Causal Flows (EMCaF) is proposed. EMCaF focuses on the assessment of properties about mass and energy flows in the CNs derived from GCMs. First, causal CNs are inferred from GCMs, and then the capabilities to reproduce characteristic transfer flows are assessed with reference models. A more in‐depth feature is the possibility to assess how climate change disturbs CNs properties. In addition to the quantitative difference between modelled and observed values taken into account in standard evaluations, the EMCaF approach aims to assess the weaknesses and strengths of GCMs to represent climate mechanisms and processes that couple different components of the climate system. The comparison of models through this approach allows having complimentary feedback on model evaluations to understand possible causes of errors and enable a judgement based on processes. The approach is illustrated by evaluating one GCM and subsequently assessing changes of its CNs under future climate projections. Results show that known climatic patterns are assimilated and that causal strength patterns are likely to agree with the wind magnitude as a transfer factor. Significative issues are then explored, showing the capabilities of the approach and allowing understand fundamental structures in transport flows, compare their properties, and assess changes in the future. Different alternatives and considerations in each step of the approach are discussed to expand its applicability.

Dynamic Bayesian Networks for Evaluation of Granger Causal Relationships in Climate Reanalyses

“…Probabilistic graphical models provide a natural and intuitive framework with which to describe the complicated interactions between climate processes. As a result, they are increasingly being applied for the purposes of studying potential causal relationships and for model evaluation (Nowack et al., 2020; Vázquez‐Patiño et al., 2020). In this latter application, models are generally evaluated on the basis of structural comparisons between graph structures inferred from observations and from model runs.…”

“…manuscript submitted to Journal of Advances in Modeling Earth Systems (JAMES) Di Capua et al, 2020;Harwood et al, 2021) and models (Deng & Ebert-Uphoff, 2014;Ebert-Uphoff & Deng, 2017), and, most recently, for model evaluation (Vázquez-Patiño et al, 2020;Nowack et al, 2020).…”

“…All rights reserved. for model evaluation (Vázquez-Patiño et al, 2020;Nowack et al, 2020). In this latter application, models are generally evaluated on the basis of structural comparisons between graph structures inferred from observations and from model runs.…”

“…As a result, graphical models based on DAGs provide a powerful tool for studying causal relationships (Pearl, 1995), and form a useful basis for other forms of causal inference (Greenland & Brumback, 2002). Bayesian, or belief, networks (BNs), as such models are usually called, have therefore received increasing attention from the climate science community (Ebert‐Uphoff & Deng, 2012a), having variously been used for forecasting and risk assessment based on expert systems (e.g., Abramson et al., 1996; Boneh et al., 2015; Catenacci & Giupponi, 2009, 2013; Leonard et al., 2014; Peter et al., 2009), for learning independence relationships and possible causal interactions in observations (Di Capua et al., 2020; Ebert‐Uphoff & Deng, 2012b; Harwood et al., 2021; Horenko et al., 2017; Kretschmer, Coumou, et al., 2016; Kretschmer, Runge, & Coumou, 2017; Li et al., 2018; Pfleiderer et al., 2020; Runge, 2015, 2018a, 2018b; Runge, Bathiany, et al., 2019; Runge, Nowack, et al., 2019; Runge, Petoukhov, Donges, et al., 2015; Runge, Petoukhov, & Kurths, 2014; Saggioro et al., 2020; Samarasinghe, Deng, & Ebert‐Uphoff, 2020; Samarasinghe, McGraw, et al., 2019) and models (Deng & Ebert‐Uphoff, 2014; Ebert‐Uphoff & Deng, 2017), and, most recently, for model evaluation (Nowack et al., 2020; Vázquez‐Patiño et al., 2020).…”

A concerted research effort to advance the hydrological understanding of tropical páramos

“…Climatic changes will lead to changes in vegetation and soils, which may propagate to the water cycle. Ecological niche models have been used to model and project the distribution of Andean biomes and species under future climate scenarios in the tropical Andes (Tovar, Arnillas, et al, 2013; Vázquez‐Patiño, Campozano, Mendoza, & Samaniego, 2020). Results project a median loss of páramos extent of 31.4% by 2039 under emissions scenario A1B, mainly due to a lack of available space for upslope migration (Tovar, Arnillas, et al, 2013).…”