PaperlayerView exact revision on arXiv

Source-linked AI summary

Statistical physics approaches to the complex Earth system

Jingfang Fan, Jun Meng, Josef Ludescher, Xiaosong Chen, Yosef Ashkenazy, Jurgen Kurths, Shlomo Havlin, Hans Joachim Schellnhuber

arXiv:2009.04918v1physics.soc-phcond-mat.stat-mechphysics.data-anphysics.geo-ph

TL;DR

The review addresses how statistical physics and complex-network methods can improve understanding and prediction of disruptive climate and earthquake phenomena. It synthesizes critical phenomena, network theory, percolation, tipping-point analysis, and entropy across Earth-system applications, reporting improved predictive performance while recognizing scope boundaries and event-definition assumptions.

  • Problem

    Extreme climate events and earthquakes remain difficult to understand and predict because Earth systems involve nonlinear feedbacks, complex structures, and debated early-warning signals.

  • Method

    The review synthesizes statistical-physics and complex-systems approaches, including critical phenomena, networks, percolation, tipping points, and entropy, for Earth-system applications.

  • Results

    The reviewed approaches improve predictive performance, including one-year-ahead El Niño forecasts with 70% accuracy and 4% false alarms.

  • Takeaways & Limitations

    Integrating statistical physics with Earth-system science provides new insights into Earth-system dynamics and supports prediction of high-impact disruptive events.

  • Takeaways & Limitations

    The review focuses on the Earth’s surface systems, including climate, relief, and earthquakes, rather than the whole planetary interior.

Abstract

from arXiv · show

Global climate change, extreme climate events, earthquakes and their accompanying natural disasters pose significant risks to humanity. Yet due to the nonlinear feedbacks, strategic interactions and complex structure of the Earth system, the understanding and in particular the predicting of such disruptive events represent formidable challenges for both scientific and policy communities. During the past years, the emergence and evolution of Earth system science has attracted much attention and produced new concepts and frameworks. Especially, novel statistical physics and complex networks-based techniques have been developed and implemented to substantially advance our knowledge for a better understanding of the Earth system, including climate extreme events, earthquakes and Earth geometric relief features, leading to substantially improved predictive performances. We present here a comprehensive review on the recent scientific progress in the development and application of how combined statistical physics and complex systems science approaches such as, critical phenomena, network theory, percolation, tipping points analysis, as well as entropy can be applied to complex Earth systems (climate, earthquakes, etc.). Notably, these integrating tools and approaches provide new insights and perspectives for understanding the dynamics of the Earth systems. The overall aim of this review is to offer readers the knowledge on how statistical physics approaches can be useful in the field of Earth system science.

1. Introduction

Earth system science treats Earth as an integrated system of interacting natural and human components. This review examines how statistical physics and complex-network approaches support understanding and prediction across climate, earthquakes, and related Earth-system phenomena.

  • Earth comprises interacting geosphere, atmosphere, hydrosphere, cryosphere, and biosphere components with nonlinear feedbacks.
  • ESS developed through conceptualizations of Earth as a biosphere-influenced, self-regulating, complex system and later incorporated human–nature co-evolution.
  • ESS combines observations, modeling and simulations, and assessments and syntheses, while instrumental records extend only about one-to-two centuries into the past.
  • The review applies critical phenomena, complex networks, percolation, tipping-point analysis, and entropy to climate, Earth relief, and earthquakes, excluding the whole planetary interior.
  • Statistical physics addresses Earth-system challenges by analyzing many interacting components and measurable macroscopic behavior in inherently stochastic systems.

2. Methodology

The methodology introduces complex networks as graph-based representations for studying Earth-system structure, dynamics, and function. Climate networks map geographical locations to nodes and inter-location similarity or causality to links.

  • Complex networks represent systems through non-trivial topological features beyond regular lattices and random Erdős-Rényi graphs.
  • Climate networks use geographical locations or grid points as nodes and similarities or causalities between time series as links with strengths.
  • Climate-network analysis represents interrelationships among global locations and has been used to analyze, model, understand, and predict climate phenomena.
  • The review compares a regular 2D square lattice, an Erdős-Rényi network, and a scale-free network as basic model-network structures.
  • Graph-theoretic network analysis defines vertices and edges, also called nodes and links, with N vertices and M edges.

The Adjacency Matrix

The adjacency matrix encodes network connectivity and can represent directed, undirected, unweighted, and weighted links. Degree-based quantities derive network size and connectivity from this matrix.

  • The adjacency matrix A records whether an edge connects vertices i and j, with entries equal to 1 or 0.
  • Weighted networks assign real-valued strengths or weights to adjacency-matrix entries, while directed networks encode edge direction.
  • A vertex degree k_i is the number of edges connected to vertex i and can be expressed using the adjacency matrix.
  • For an undirected network, the number of edges equals half the sum of all vertex degrees, and the mean degree is 2M/N.
  • Directed networks distinguish in-degree from out-degree according to incoming and outgoing edges.

Degree Distributions

Degree distributions describe how connectivity is distributed across network vertices. The review contrasts homogeneous Erdős-Rényi networks with heterogeneous scale-free networks characterized by power-law degree distributions.

  • The degree distribution P(k) is the fraction of nodes having degree k, or k connections.
  • Erdős-Rényi networks have Poisson degree distributions determined by the average degree and are typically relatively homogeneous.
  • Real networks often deviate from the Poisson law and exhibit approximate power-law degree distributions.
  • Scale-free networks are defined by power-law degree distributions, with typical exponent values in the range 2 ≤ λ ≤ 3.

Clustering Coefficient

The clustering coefficient measures how strongly nodes cluster locally through triangles and connected triples. Local and whole-network coefficients quantify the likelihood that neighboring nodes are connected, with values from 0 to 1.

  • Clustering Coefficient: The clustering coefficient measures local connectivity through the number of triangles relative to connected triples.Each triangle is counted three times because it contains three centered connected triples.
  • Clustering Coefficient: A local coefficient C_i counts connected neighbor pairs around vertex i and divides by all neighbor pairs.For vertices of degree 0 or 1, C_i is set to 0 because both counts are zero.
  • Clustering Coefficient: The whole-network clustering coefficient is obtained by averaging local coefficients, and clustering lies between 0 and 1.
  • Subgraphs and Communities: Subgraphs are graphs whose nodes and edges are contained in a larger graph, including cycles, trees, and complete subgraphs.Networks also contain components and tightly connected, potentially overlapping communities.
  • Subgraphs and Communities: Figure 4 illustrates tree, fully completed, and overlapping k-clique community subgraphs at k = 4.

Network Models

The review introduces network models and a climate-network workflow for converting climatic time series into weighted or directed links. It also describes alternative dependence measures, including event synchronization and mutual information, for interpreting climate-system dynamics.

  • Network Models: The Erdős-Rényi model connects each possible edge independently with probability p and has Poissonian degree distributions.Its fixed-edge and fixed-probability graph ensembles are treated as analogous statistical-physics ensembles.
  • Network Models: The Barabási-Albert model grows networks by adding nodes and preferentially attaching them to higher-degree nodes, producing a power-law degree distribution.
  • Network Models: The Watts-Strogatz model interpolates between a regular lattice and a random graph by randomly rewiring lattice edges.
  • Climate Network Construction: Climate-network construction defines spatial nodes, selects and preprocesses climatological time series, analyzes links, and interprets network characteristics through climate dynamics.Nodes commonly represent longitude-latitude locations, while variables can include temperature, precipitation, wind, or sea-surface temperature.
  • Climate Network Construction: Figure 5 summarizes the methodology for constructing a climate network from climatic time series.
  • Pearson Correlation Climate Network: Pearson correlation climate networks use time-delayed cross-correlations to define weighted and directed links between climate-record nodes.The link direction follows the lag of the strongest peak or lowest valley in the cross-correlation function.
  • Climate Network Construction: Links can be retained only when their weights exceed a significant threshold, which helps overcome strong autocorrelation in the data.
  • Alternative Dependence Measures: Alternative climate-network methods use event timing or mutual information to quantify synchronization and nonlinear dependence between time series.Mutual information is zero exactly when the variables are independent and can also be computed with a time lag.

Percolation Order Parameter

The percolation order parameter P∞ is the fraction of sites in the infinite cluster and distinguishes finite-cluster and spanning regimes. Its behavior under decreasing occupation links system collapse to resilience.

  • Percolation Order Parameter: P∞ is the fraction of sites belonging to the infinite cluster and serves as the percolation order parameter.
  • Percolation Order Parameter: For p < pc, only finite clusters exist and P∞ = 0; for p > pc, an infinite cluster forms and P∞ becomes nonzero.
  • Percolation Order Parameter: When p decreases from 1 through node or link removal, the system collapses after removing a fraction 1 −pc, which measures resilience to collapse.

Cluster Size Distribution

Near the percolation threshold, finite-component sizes follow a scaling form and develop a power-law tail. The characteristic component size diverges as the occupation probability approaches pc.

  • Cluster Size Distribution: The finite-component distribution near criticality follows a scaling form in component size s, with n_s denoting the number of components of size s.
  • Cluster Size Distribution: At the percolation threshold, the characteristic size sξ diverges as |p −pc|^−σ and the distribution tail follows a power law with critical exponent σ.

Average Cluster Size

The average cluster size is defined from the finite-cluster size distribution and diverges as the system approaches the critical point.

  • The probability that a site belongs to a cluster of size s is used to characterize the cluster-size distribution.
  • The probability ws(p) describes the size of the finite cluster containing a given site.
  • The probabilities ws(p) over finite-cluster sizes sum to 1.
  • The finite-cluster average size ⟨s(p)⟩ diverges near the critical point.

Correlation Length

The correlation function measures whether two sites belong to the same finite cluster, while the correlation length characterizes their typical separation. As p approaches pc, the correlation length increases and critical exponents describe the transition.

  • The correlation function g(r) gives the probability that a site at distance r from an occupied site belongs to the same finite cluster.
  • At large distances, g(r) has an exponential cutoff governed by the correlation length ξ.
  • The correlation length ξ measures the mean distance between two sites on the same finite cluster.
  • As p approaches pc, ξ increases, while β, γ, and ν describe the critical behavior of the percolation transition.
  • Percolation-cluster structural measurements are introduced as tools for characterizing cluster properties.

Fractal Dimension

Percolation clusters are characterized through fractal dimensions, scaling relations, and structural measurements across length scales. These concepts extend to networks, where renormalization supports fractal analysis, while universality and explosive processes broaden the transition framework.

  • Fractal structure: The fractal dimension df describes how cluster mass M within a sphere changes with radius r, and the infinite cluster is self-similar at pc.
  • Fractal structure: Below the correlation length ξ, clusters exhibit fractal structure; above ξ, they become homogeneous.
  • Fractal structure: The relative size of the giant component P∞ provides another description of percolation-cluster structure.
  • Fractal structure: The cluster fractal dimension relates to critical exponents through df = d − β/ν.
  • Fractal analysis of networks: Network fractal dimensions can be estimated by box counting or cluster growing, with renormalization used because networks are generally not spatially embedded.
  • Fractal analysis of networks: The chemical distance ℓ describes shortest paths within clusters, while dmin characterizes their scaling structure.
  • Scaling theory: Scaling theory relates the cluster-size distribution ns(p) to critical exponents and a rapidly decaying scaling function.
  • Scaling theory: Cluster radius, moments, and the order parameter P∞ are used to derive critical scaling relationships.

Gap Exponents

The review uses percolation theory and tipping-point analysis to characterize abrupt transitions in complex Earth systems and identify indicators that may precede them. It also surveys climate tipping elements and early-warning signals based on recovery, autocorrelation, variance, and flickering.

  • Percolation: The percolation control parameter is link density, while the order parameter is the size of the largest connected cluster.Links are added progressively, and the largest connected component is recorded during system evolution.
  • Gap Exponents: Gap exponents characterize a percolation universality class and remain related to, rather than independent of, standard critical exponents.The review introduces six gap critical exponents and derives relationships between them and standard percolation exponents.
  • Percolation Results: For two-dimensional square-lattice bond percolation, β1 = β2 ≈0.104 and rc(∞) ≈0.5, consistent with known theoretical values.The reported results also give 1/ν1 = 1/ν2 ≈1/ν and df 1 = df 2 ≈1.895.
  • Tipping Points: Tipping points describe situations in which a small change produces large, long-term consequences, often through bifurcations or shifts to alternative attractors.The review applies this concept across climate, ecosystems, social systems, financial markets, and medicine.
  • Climate Tipping Elements: Climate tipping elements are assessed through system features, control parameters, critical values, transition timescales, and key impacts.Policy-relevant criteria additionally include political and ethical time horizons and a corresponding qualitative change in impacts.
  • Early-Warning Signals: Early-warning indicators include increasing lag-1 autocorrelation, variance, and flickering as systems approach critical transitions.Critical slowing down reduces return speed, while autocorrelation and variance increase; flickering is oscillation between two stable states.

3. Applications

Applications of statistical physics and climate-network methods address climate-system structure, El Niño prediction, monsoon timing and rainfall, ENSO impacts, and climate-change responses. Across these cases, network-based indicators and percolation or tipping-point analyses provide reported predictive signals and system-level diagnostics.

  • Climate System: Climate models span simplified energy-balance formulations to computationally costly Earth System Models whose outputs remain approximations.The review also identifies absorbed and outgoing radiation, heat capacity, albedo, atmospheric transmissivity, and insolation as parameters in the energy-balance formulation.
  • El Niño–Southern Oscillation: The Oceanic Niño Index defines El Niño and La Niña using three-month running-mean sea-surface-temperature anomalies in the Niño 3.4 region.The Niño 3.4 region extends from 5°S to 5°N and 170°W to 120°W.
  • El Niño–Southern Oscillation: Climate-network approaches forecast El Niño approximately one year ahead using cooperative network modes, percolation-cluster changes, or trends in similarity-based indices.The percolation approach combines climate, network, and percolation theory; another index targets low-frequency temperature-anomaly similarity within the Niño 3.4 region.
  • El Niño–Southern Oscillation: The network-based forecasting methods reported successful El Niño predictions, including the 2018–2019 onset and a magnitude estimate of 1.11±0.23°C versus an observed 0.9°C.The SysSampEn correlation with El Niño magnitude reached r = 0.99, with a one-year prediction horizon and RMSE = 0.23°C for the average individual-dataset forecasts.
  • Indian Summer Monsoon: For the Indian Summer Monsoon, climate-network and nonlinear-dynamics methods identify teleconnections and tipping points between Eastern Ghats and North Pakistan reference regions.The resulting scheme achieved 74% successful onset predictions and 84% successful withdrawal predictions under the review’s stated timing thresholds.

3.2. Earth Geometric Surface Relief

The review applies percolation theory and scaling analysis to Earth’s geometric relief, revealing critical transitions in land and ocean cluster structure. These analyses connect topographic organization to mean sea level, plate-tectonic processes, and multifractal surface properties.

  • Data and approach: ETOPO1 provides a 1 arc-minute global relief model integrating land topography and ocean bathymetry for percolation-based analysis.The dataset contains N = 10800 × 21600 grid points and supports geomorphic and hypsographic analyses.
  • Surface structure: Earth’s topography exhibits self-similarity, long-range correlations, multifractal structure, and a bimodal division between continents and ocean basins.The power spectrum follows a scaling relation, while Hurst exponents differ across bathymetry, continents, and continental margins.
  • Land percolation: The landmass order parameter jumps abruptly at h = 0, coinciding with Earth’s present mean sea level.The susceptibility and correlation length also become divergent there, identifying mean sea level as the critical percolation level.
  • Oceanic and lunar comparisons: Oceanic clusters undergo a discontinuous transition near h = −3640 m, while lunar topography shows a threshold with equal land and ocean areas.The Earth oceanic analysis reports a closely related transition near h = −3621 m.
  • Relief-network transitions: Earth’s relief network exhibits several abrupt, statistically significant phase transitions as occupied nodes are added by elevation.At the land percolation threshold, a critical node near 64.458333°N, 171.141667°W connects the two largest clusters at pc ≈ 0.321.
  • Mechanistic interpretation: Percolation on fractional Brownian surfaces with H = 0.66 was used to investigate the discontinuity observed in Earth’s relief network.This Hurst exponent corresponds to the estimated roughness of Earth’s continental topography.

3.3. Earthquake Systems

This section reviews earthquake occurrence laws, memory analyses, and forecasting approaches grounded in statistical mechanics and ETAS models. It reports scale-invariant seismic patterns, persistent memory across catalogs, and improved forecasting when memory is incorporated.

  • Occurrence laws: Earthquake catalogs record occurrence time, location, magnitude, and depth, enabling statistical characterization of seismicity.These quantities underpin empirical descriptions of earthquake occurrence.
  • Occurrence laws: The Gutenberg–Richter law describes an exponential decline in cumulative earthquake counts with increasing magnitude.For the Israeli catalog, the fitted parameters are a = 5.36 and b = 0.97±0.02.
  • Occurrence laws: Universal inter-event scaling is debated because ETAS analyses find the distribution only approximately universal and of gamma form.The reported qualification concerns the scaling relation associated with earthquake inter-event times.
  • Memory analysis: DFA analyses find long-term memory in earthquake interoccurrence times, with α ≈0.75 across Californian, Israeli, and Italian catalogs.The Californian result is reported for both DFA0 and DFA1 and is independent of threshold and catalog; related analyses extend to other seismic variables.
  • Memory analysis: Conditional-probability analyses show that short recurrence times tend to follow short ones and long recurrence times follow long ones.Israeli and Italian catalogs yield nonzero memory coefficients, including ρ1 = 0.248 and ρ4 = −0.193 for Israel, and ρ1 = 0.260 and ρ4 = −0.153 for Italy.
  • Memory analysis: ETAS models reproduce empirically observed earthquake memory only within narrow parameter ranges, with memory affected by background rate and other model parameters.Smaller Omori exponent p and larger αM produce stronger memory through the ETAS branching ratio.

Q Q Q

This material outlines statistical tests and diagrams for evaluating earthquake forecasts, including spatial, count, ROC, and Molchan assessments. It also links forecasting accuracy to spatial-temporal resolution and reports memory-informed ETAS improvements.

  • Forecasting framework: Earthquake forecasting evaluates the occurrence probability within a space-time-magnitude hypercell conditioned on available information.The forecasting probability is obtained by integrating the local conditional rate over the hypercell.
  • Forecasting framework: Smaller forecast hypercells can improve rate evaluation, while the temporal window separates long-, intermediate-, and short-term forecasting.The cited timescales range from decades to centuries for long-term forecasts and from seconds to weeks for short-term forecasts.
  • Forecast evaluation: ROC diagrams compare true-positive and false-positive rates as the discrimination threshold varies, with perfect prediction at TPR = 1 and FPR = 0.The diagonal represents random guessing.
  • Forecast evaluation: Molchan diagrams compare miss rate with the fraction of space-time volume under alarms, with the best prediction at F = 0 and M = 0.A very small ΓM indicates a highly skilled alarm set.
  • Forecast evaluation: The N-test assesses whether the forecasted earthquake count matches the observed count using a generalized Poisson or negative-binomial framework.Its interpretation uses a one-sided test with an effective significance value based on α = 0.05.
  • Forecast evaluation: The S-test evaluates consistency between forecast and observed spatial distributions, where values close to 1 indicate optimal spatial forecasts.Values below the critical significance value indicate inconsistency with the forecast.
  • Memory-informed forecasting: Incorporating memory improves cumulative-event forecasting after the Capitignano main shock by more than 20% within 14 days.A revised ETAS model also significantly improves forecasting for Italian and Southern California catalogs.

4. Conclusions and Perspectives

The review finds that statistical-physics and complex-network approaches improve understanding and forecasting across climate, topography, and earthquakes, while leaving major theoretical and applied questions open.

  • Review contribution: Statistical-physics and complex-network techniques support understanding and prediction across climate variability, Earth relief features, and earthquakes.The review presents these approaches as a relatively novel branch of geophysics applied to multiple Earth subsystems.
  • Open problems: A comprehensive framework remains undeveloped, requiring better physical mechanisms, multilayer climate networks, a universal critical-phenomena theory, and principles for anticipating tipping points and entropy.These open problems concern both the theory and representation of coupled Earth systems.
  • Climate applications: Evolving climate networks and combined approaches can serve as novel predictors that significantly improve forecasts across several weather and climate scenarios.Examples include El Niño–Southern Oscillation, the Indian summer monsoon, extreme rainfall, atmospheric circulation, and Atlantic meridional overturning circulation.
  • Earth relief: Earth-relief analyses identify abrupt network transitions and suggest that long-range correlations may contribute to discontinuities in surface-topography evolution.The framework may also help identify critical nodes relevant to future global change in Earth’s relief network.
  • Earthquakes: DFA and CP memory analyses can significantly improve short-term forecasting rates for real earthquake catalogs, although earthquake prediction remains controversial and challenging.The review also discusses empirical earthquake laws, ETAS, and methods for evaluating earthquake predictions and forecasts.
  • Forecasting scope: The reviewed approaches have demonstrated forecasting applications for El Niño, extreme rainfall, monsoon rainfall, Atlantic circulation collapse, aftershock patterns, and earthquake detection.The review highlights network-based and statistical-physics approaches as useful across diverse complex Earth phenomena.
  • Future directions: Complex-network models are still needed to understand or predict cascading failures in interdependent critical infrastructures affected by extreme climate and weather events.The review identifies climate resilience of societies, ecosystems, and economies as an important area for further work.

Appendix A. Acronyms

Appendix A provides two acronym tables: one for scientific acronyms and one for institutional acronyms.

  • Scientific acronyms: Table A.2 lists scientific acronyms used throughout the paper.The appendix identifies this table as the scientific-acronym reference.
  • Institutional acronyms: Table A.3 lists institutional acronyms used throughout the paper.The appendix identifies this table as the institutional-acronym reference.
Loading 2009.04918v1…