Climate Change: Linear and Nonlinear Causality Analysis
The goal of this study is to detect linear and nonlinear causal pathways toward climate change as measured by changes in global mean surface temperature and global mean sea level over time using a data-based approach in contrast to the traditional physics-based models. Monthly data on potential clim...
Published in: | Stats |
---|---|
Main Authors: | , |
Format: | Text |
Language: | English |
Published: |
Multidisciplinary Digital Publishing Institute
2023
|
Subjects: | |
Online Access: | https://doi.org/10.3390/stats6020040 |
id |
ftmdpi:oai:mdpi.com:/2571-905X/6/2/40/ |
---|---|
record_format |
openpolar |
spelling |
ftmdpi:oai:mdpi.com:/2571-905X/6/2/40/ 2023-08-20T04:07:17+02:00 Climate Change: Linear and Nonlinear Causality Analysis Jiecheng Song Merry Ma 2023-05-15 application/pdf https://doi.org/10.3390/stats6020040 EN eng Multidisciplinary Digital Publishing Institute https://dx.doi.org/10.3390/stats6020040 https://creativecommons.org/licenses/by/4.0/ Stats; Volume 6; Issue 2; Pages: 626-642 climate change global mean surface temperature global mean sea level causal analysis greenhouse gas vector autoregressive model (VAR) vector error correction model (VECM) autoregressive distributed lag model (ARDL) artificial neural network (ANN) Text 2023 ftmdpi https://doi.org/10.3390/stats6020040 2023-08-01T10:04:44Z The goal of this study is to detect linear and nonlinear causal pathways toward climate change as measured by changes in global mean surface temperature and global mean sea level over time using a data-based approach in contrast to the traditional physics-based models. Monthly data on potential climate change causal factors, including greenhouse gas concentrations, sunspot numbers, humidity, ice sheets mass, and sea ice coverage, from January 2003 to December 2021, have been utilized in the analysis. We first applied the vector autoregressive model (VAR) and Granger causality test to gauge the linear Granger causal relationships among climate factors. We then adopted the vector error correction model (VECM) as well as the autoregressive distributed lag model (ARDL) to quantify the linear long-run equilibrium and the linear short-term dynamics. Cointegration analysis has also been adopted to examine the dual directional Granger causalities. Furthermore, in this work, we have presented a novel pipeline based on the artificial neural network (ANN) and the VAR and ARDL models to detect nonlinear causal relationships embedded in the data. The results in this study indicate that the global sea level rise is affected by changes in ice sheet mass (both linearly and nonlinearly), global mean temperature (nonlinearly), and the extent of sea ice coverage (nonlinearly and weakly); whereas the global mean temperature is affected by the global surface mean specific humidity (both linearly and nonlinearly), greenhouse gas concentration as measured by the global warming potential (both linearly and nonlinearly) and the sunspot number (only nonlinearly and weakly). Furthermore, the nonlinear neural network models tend to fit the data closer than the linear models as expected due to the increased parameter dimension of the neural network models. Given that the information criteria are not generally applicable to the comparison of neural network models and statistical time series models, our next step is to examine the robustness ... Text Ice Sheet Sea ice MDPI Open Access Publishing Stats 6 2 626 642 |
institution |
Open Polar |
collection |
MDPI Open Access Publishing |
op_collection_id |
ftmdpi |
language |
English |
topic |
climate change global mean surface temperature global mean sea level causal analysis greenhouse gas vector autoregressive model (VAR) vector error correction model (VECM) autoregressive distributed lag model (ARDL) artificial neural network (ANN) |
spellingShingle |
climate change global mean surface temperature global mean sea level causal analysis greenhouse gas vector autoregressive model (VAR) vector error correction model (VECM) autoregressive distributed lag model (ARDL) artificial neural network (ANN) Jiecheng Song Merry Ma Climate Change: Linear and Nonlinear Causality Analysis |
topic_facet |
climate change global mean surface temperature global mean sea level causal analysis greenhouse gas vector autoregressive model (VAR) vector error correction model (VECM) autoregressive distributed lag model (ARDL) artificial neural network (ANN) |
description |
The goal of this study is to detect linear and nonlinear causal pathways toward climate change as measured by changes in global mean surface temperature and global mean sea level over time using a data-based approach in contrast to the traditional physics-based models. Monthly data on potential climate change causal factors, including greenhouse gas concentrations, sunspot numbers, humidity, ice sheets mass, and sea ice coverage, from January 2003 to December 2021, have been utilized in the analysis. We first applied the vector autoregressive model (VAR) and Granger causality test to gauge the linear Granger causal relationships among climate factors. We then adopted the vector error correction model (VECM) as well as the autoregressive distributed lag model (ARDL) to quantify the linear long-run equilibrium and the linear short-term dynamics. Cointegration analysis has also been adopted to examine the dual directional Granger causalities. Furthermore, in this work, we have presented a novel pipeline based on the artificial neural network (ANN) and the VAR and ARDL models to detect nonlinear causal relationships embedded in the data. The results in this study indicate that the global sea level rise is affected by changes in ice sheet mass (both linearly and nonlinearly), global mean temperature (nonlinearly), and the extent of sea ice coverage (nonlinearly and weakly); whereas the global mean temperature is affected by the global surface mean specific humidity (both linearly and nonlinearly), greenhouse gas concentration as measured by the global warming potential (both linearly and nonlinearly) and the sunspot number (only nonlinearly and weakly). Furthermore, the nonlinear neural network models tend to fit the data closer than the linear models as expected due to the increased parameter dimension of the neural network models. Given that the information criteria are not generally applicable to the comparison of neural network models and statistical time series models, our next step is to examine the robustness ... |
format |
Text |
author |
Jiecheng Song Merry Ma |
author_facet |
Jiecheng Song Merry Ma |
author_sort |
Jiecheng Song |
title |
Climate Change: Linear and Nonlinear Causality Analysis |
title_short |
Climate Change: Linear and Nonlinear Causality Analysis |
title_full |
Climate Change: Linear and Nonlinear Causality Analysis |
title_fullStr |
Climate Change: Linear and Nonlinear Causality Analysis |
title_full_unstemmed |
Climate Change: Linear and Nonlinear Causality Analysis |
title_sort |
climate change: linear and nonlinear causality analysis |
publisher |
Multidisciplinary Digital Publishing Institute |
publishDate |
2023 |
url |
https://doi.org/10.3390/stats6020040 |
genre |
Ice Sheet Sea ice |
genre_facet |
Ice Sheet Sea ice |
op_source |
Stats; Volume 6; Issue 2; Pages: 626-642 |
op_relation |
https://dx.doi.org/10.3390/stats6020040 |
op_rights |
https://creativecommons.org/licenses/by/4.0/ |
op_doi |
https://doi.org/10.3390/stats6020040 |
container_title |
Stats |
container_volume |
6 |
container_issue |
2 |
container_start_page |
626 |
op_container_end_page |
642 |
_version_ |
1774718793305554944 |