Edited by: Tomas Halenka, Charles University, Czechia
Reviewed by: Sara del Rio Gonzalez, Universidad de León, Spain; Mahdi Haddad, Algerian Space Agency, Algeria
This article was submitted to Interdisciplinary Climate Studies, a section of the journal Frontiers in Earth Science
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
Numerous studies have indicated that El Niño and the southern oscillation (ENSO) could have determinant impacts on remote weather and climate using the conventional correlation-based methods, which however, cannot identify the cause-and-effect of such linkage, and ultimately determine a direction of causality. This study employs the vector auto-regressive (VAR) model estimation method with the long-term observational sea surface temperature (SST) data and the NCEP/NCAR reanalysis data to demonstrate the Granger causality between ENSO and other climate attributes. Results showed that ENSO as the modulating factor can result in abnormal surface temperature, pressure, precipitation and wind circulation remotely, not vice versa. We also carry out the global climate model sensitivity simulations using the parallel computing techniques to double confirm the causality relations between ENSO and abnormal events in remote regions. Our statistical and climate model-based analyses may enrich our current understanding on the occurrences of extreme events worldwide caused by different ENSO strengths through teleconnections.
El Niño and the southern oscillation (ENSO) is a local phenomenon of the variation in sea surface temperature (SST) and air pressure across the equatorial eastern Pacific Ocean. ENSO is strongly linked to remote weather and climate far away over other parts of the world through the atmospheric “teleconnection.” Strong El Niño events have the potential to temporarily increase global mean sea level (
Over the past several decades, ENSO has been found as one of the most dominating climate factors that impacts remote weather and climate through the atmospheric “teleconnection” using the conventional correlation-based methods (
Granger causality method (
In this study, we aim to determine the spatio-temporal causality relationships between ENSO and abnormal events in remote regions, and to provide some valuable insights for the prediction of several extreme weather/climate events under different ENSO backgrounds. We hypothesize that ENSO is one of the modulating factors of the extreme weather and climate events, and the causality can be statistically demonstrated using observational datasets and can be consistently simulated using climate models. We note this study greatly extends our previous work at
This paper is structured in the following sections. Section “Materials and Methods” lists the datasets used for the study. Section “Results” introduces the Granger causality methods and global climate model simulations. Section “Discussion” reports the main findings from our study, followed by section “Conflict of Interest” that discusses and concludes the study.
For this study, we use the Hadley centre sea ice and sea surface temperature data (HadiSST). The HadISST data utilize both
Same as the ENSO indices defined by National Oceanic and Atmospheric Administration (NOAA), we use the SST in the Niño 3.4 region (5°S-5°N, 170°W-120°W) to derive the ENSO index (
The running 3-month mean SST anomaly for the Niño 3.4 region from 1950 to 2017 with the 1950–2000 as the base period. Unit of SST anomaly is degree Celsius.
The NCEP/NCAR reanalysis I data are employed in this study. This reanalysis data is produced using a state-of-the-art analysis/forecast system that performs data assimilation using past observational data from 1948 to the present. The data span from 1948 to present at the 2.5° × 2.5° latitude-longitude resolution with 17 vertical levels (
For flooding and drought extreme events, we use the Global Precipitation Climatology Project (GPCP) version 2.3 precipitation data from 1979 to the present at the 2.5° × 2.5° latitude-longitude resolution (
In this study, we use two statistic methods (Granger causality method and Maximum lag correlation) and a global climate model (Community Atmospheric Model) to investigate the global impacts of ENSO on the climate variables.
Granger causality could be calculated using different approaches such as vector autoregressive model (VAR), Graphical Lasso and SIN methods (
An autoregressive (AR) model is usually used to measure the dependency of a variable on its own previous values.
Using AR(s) to denote an autoregressive model of order s, then the AR(s) on a series
The vector autoregressive model is a particular case of the autoregressive model: VAR is used when we have more than one variable. Therefore, we will have the autoregressive model [Eq. (3)] on a vector. Given two time series
The VAR can be applied to test the Granger causality of
In this study, we use the VAR package in Python to implement the vector autoregressive model. The lag order
In [Eq. (5)],
Another way to use the
To compare with and to complement the Granger causality model, we also calculate the maximum lag correlation (i.e., cross correlation) between ENSO index and climate variables. It provides the maximum correlation coefficients between ENSO and climate variable and the corresponding lag time. The lag correlation coefficient between two series
Based on physical hypothesis and sophisticated schemes, climate model is frequently used to find out the impact of a causing factor as effects on other parameters. For this study, series of sensitivity simulations are carried out with the global climate model forced by different simulated ENSO-like SST patterns to see the corresponding responses of atmospheric fields. The climate model we use in this study is the Community Atmospheric Model (version 5.3, CAM5.3) with the CAM5 standard parameterization schemes (
First, we determine the cause-and-effect relation between ENSO and SAT on the global scale using the VAR method for Granger causality model. As the significant differences between
The global distribution of the maximum lag correlation between ENSO index and surface temperature (
Maximum lag correlation between ENSO index and surface air temperature over land.
In this study, we also investigate the causality relation between ENSO and SLP on the global scale for their spatio-temporal patterns. Comparing
The global distribution of the maximum lag correlation between ENSO index and SLP (
Maximum lag correlation between ENSO index and sea level pressure.
To explore the relationship between extreme flooding and drought with ENSO, we analyze the causality relation between ENSO and surface precipitation on the global scale. As the comparison between
The global distribution of the maximum lag correlation between ENSO index and surface precipitation (
Maximum lag correlation between ENSO index and precipitation.
Occurrence of different climate events strongly depends on the large-scale atmospheric circulation. Mid-tropospheric (500 hPa) vertical pressure velocity is widely used as a proxy for the large-scale tropical circulation (
The global distribution of the maximum lag correlation between ENSO index and 500 hPa vertical velocity (
Maximum lag correlation between ENSO index and 500 hPa vertical velocity.
In this study, we use statistical methods, namely the VAR method for Granger causality model, and global climate model simulations to investigate ENSO causality as one of the modulating climate factors that cause the anomalies in surface air temperature, precipitation, surface pressure, and vertical wind over remote regions through teleconnection with lagged temporal variability. We analyzed different observational data, reanalysis data, and model data to comprehensively investigate the global impacts of ENSO. The Granger causality analysis was able to clearly show ENSO as a cause instead of an effect to influence the remote climate variables and thus cause extreme weather events such as flooding, drought, extreme heat, and cold, etc. Our model simulations using the CAM5.3 also successfully simulated ENSO’s remote impacts on other weather variables, consistent to the findings from observational evidence. Besides, all the source codes used in this study can be found on Github
For future work, we will compare the Granger causality on multiple variables for intercomparison, and we will add analysis on the impact of ENSO on clouds and aerosol using NASA satellite remote sensing data. We plan to use the Granger causality methods to predict climate variations and other interesting economic factors, such as crop yield and wheat stock price, under different ENSO backgrounds. At the same time, we plan to explore more efficient use of high performance parallel computing in the studies that use much broader big data of satellite observations with higher spatial and temporal resolutions.
The datasets analyzed for this study can be found in HadISST at
HS, JT, JH, and PG worked on the implementation and experiment. All authors worked on the research topic, approach, plan, and approved the manuscript.
JH was employed by the company Science Systems and Applications, Inc. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
This work was supported by the grant CyberTraining: DSE: Cross-Training of Researchers in Computing, Applied Mathematics, and Atmospheric Sciences using Advanced Cyberin-frastructure Resources from the National Science Foundation (Grant No. OAC–1730250). The hardware in the UMBC High Performance Computing Facility (HPCF) was supported by the U.S. National Science Foundation through the MRI program (Grant Nos. CNS–0821258, CNS–1228778, and OAC–1726023) and the SCREMS program (Grant No. DMS–0821311), with additional substantial support from the University of Maryland, Baltimore County (UMBC). See