Using Principal Component Analysis to Investigate Changes in the North American Dipole in Future Simulations
ABSTRACT
The effects of climate change extend beyond global warming due to the interconnectedness of climate and ecosystems. Changes in the atmosphere create cascading consequences that can cause irreversible damage to the environment. However, changes in climate happen over decades, so researchers must predict how the atmosphere will function in different scenarios of the future to avoid further effects of global warming. Thus, six Global Climate Models (GCMs) were analyzed to compare the effects of historically consistent greenhouse gas emissions versus increased emissions on the timing of North American climate dipole (NAD) from 2015-2100. By comparing the predictions for spring and summer of different models, changes in timing could be identified if patterns from summer were found in spring in the future. These comparisons can help contextualize the degree to which fossil fuel emissions must be reduced to prevent a shift in dipole timing. The results of the analysis revealed disagreement across models for the timing of the NAD in future simulations. Models INM-CM5-0, MIROC6, and TaiESM1 had higher correlations with the observed North American Dipole in the summer, as normal timing predicts. However, ACCESS-CM2 and NorESM2-MM had higher correlations in the spring, suggesting the summer dipole would shift into the spring. More analysis is left to be conducted on a larger sample of models, and more up to date data with the release of updated climate models.
INTRODUCTION.
Global Warming is caused by the release of greenhouse gases into the atmosphere. Radiative forcing is the difference between energy entering and leaving the atmosphere [1]. Most of the energy from the sun is reflected from the surface and clouds, but some energy is absorbed by the surface and radiated out as infrared radiation. Greenhouse gases increase global temperatures by absorbing and re-emitting the outgoing radiation from Earth back into the atmosphere [1]. Increased radiative forcings also worsen climate change by causing feedbacks that increase the rate of warming. For example, warmer average temperatures cause more water vapor to be in the atmosphere, which also acts as a radiative forcing agent.
According to the Intergovernmental Panel on Climate Change’s (IPCC) summary for policymakers, global average temperatures have increased by 0.8 to 1.2 degrees Celsius since pre-industrialization (1850) and will reach 1.5 degrees between 2030 to 2052 with current greenhouse gas emissions [2]. These increases in global temperature disrupt ecosystems through changes in temperature, precipitation, and worsening severe weather. Losses to the biodiversity of a region are nearly impossible to fix, and habitat destruction and droughts from climate change could do irreparable damage. The uncertainty in the future of the environment has led scientists to develop climate models that predict how climate change will progress in the future. They also provide methods of finding and testing mitigation strategies that could be used to avoid permanent harm to the environment.
The Moran effect is a phenomenon in which ecological factors, such as reproduction and migration are synchronized with regional atmospheric variability [3]. An example of these patterns of atmospheric variability is a climate dipole; wherein different sides of a continent will experience opposite anomalies in weather [4]. For example, one half will experience anomalously cool and wet summer, and the other half will have a warm and dry summer. Subsequently, these fluctuations coincide with ecological anomalies depending on the sign of the dipole in its region. Increased temperatures and fluctuating weather due to climate change can affect the timing of these patterns to the point of disrupting the ecological processes triggered by them.
An example of the interaction between climate fluctuations and ecological movement is seen in the North American Dipole (NAD), an east-west summer dipole shown in figure 1 as oppositely correlated pressure anomalies. The NAD alternates between warmer and drier summers on one pole, shown in red as higher pressure, and colder and wetter summers on the opposite pole, shown in blue as lower pressure. High pressure systems tend to lead to less precipitation because the higher weight of air leads to less water vapor to rise and form clouds. Warm dry summers tend to cause the boreal coniferous trees to release large seed yields the following summer, called a mast-seeding event [4]. This leads to one side of the continent having more seeds than another. In turn, triggering an irruptive migration of boreal seed-eating animals, such as the pine siskin (Spinus pinus). This is the Moran effect at a continental scale as it is “a climate dipole creating an ecological dipole” [5]. Seed production, and bird migration are critical to the health and balance of the food chain in the boreal forests, and previous research found that under severe increases in greenhouse gas emissions, the NAD could shift into the spring [6].
The NAD triggers mast-seeding at a roughly biennial time scale, however, the events have become more irregular since 1998 [4]. A warm and dry summer tends to lead to more seed production, which is then released the following year. The teleconnections that cause the NAD to appear are not fully understood. The atmospheric circulation that triggers the NAD to form is hypothesized to be from Rossby waves driven by the Madden-Julian Oscillation (MJO), as well as Pacific stationary Rossby waves affected by east Asian monsoonal convection (EAM) [5]. These are complex and variable mechanisms that require sophisticated methods to predict.
Global Climate Models (GCMs) allow scientists to perform experiments on past and future climate by predicting the circulation of Earth’s climate, like the MJO and EAM, and how they will change in the future. The Coupled Model Intercomparison Project phase 6 (CMIP6) is an international effort to compile and document the results of experiments of GCMs from research institutes across the world [7]. Individual models have biases that cause them to over- or under-predict certain factors, like precipitation for example, which hinder their accuracy of modeling the real world [8]. By comparing results across many models, they provide a more holistic picture that helps arrive at a consensus of the outlooks for climate. Another significant use of models is in comparing severities of climate change by modeling increased or decreased carbon emissions.
The Shared Socioeconomic Pathways (SSPs) are a way of categorizing the projections for the future in climate models. The pathways range from 1-5, with SSP1 representing a future with the least fossil fuel emissions, and SSP5 being a scenario of severe dependence on fossil fuel emissions [2]. SSP scenarios also include radiative forcing measured in W/m2 that represents difference of energy entering and leaving the atmosphere. For example, SSP5-8.5 (8.5 is the radiative forcing) will have a disproportionate amount of energy entering the atmosphere versus leaving, due to the higher amount greenhouse gases trapping the energy.
Previous research found that models in CMIP6 predicted the NAD to shift into the spring months under severe greenhouse gas emissions in SSP5-8.5 (SSP585), a worst-case scenario [6]. However, no experiments have been conducted to investigate how more moderate scenarios for future fossil fuel emissions will affect the timing of the NAD. SSP2-4.5 (SSP245) reflects a future where emissions progress at a constant rate that they are now, and SSP3-7.0 (SSP370) mirrors a future where emissions continue to increase, but not as severely as higher SSPs [2]. Climate dipoles are common patterns in the atmosphere, so understanding the evolution of one could be applied to ecosystems world-wide, not just North America.
Climate change is expected to alter the atmospheric pattern that regulate interactions within ecosystem, and severe fossil fuel emission scenarios have been shown to shift the seasonal timing of the North American Dipole, a prominent mode of atmospheric circulation. However, it remains unclear what threshold of warming is sufficient to trigger such a shift in the NAD, and how this, in turn, could destabilize the ecological processes that depend on its variability. The objective of this research is to investigate the evolution of the North American Dipole (NAD) by analyzing and comparing the predictions of climate models to find potential shifts in seasonal timing and the minimum climate forcing required to trigger them. We hypothesize that the models will predict the NAD to shift into the spring months under the increased emission scenario of SSP370, while there will be no shift in SSP245 because such drastic changes in timing would require vast changes to the upstream teleconnections driving the NAD, requiring more significant warming.
MATERIALS AND METHODS.
Data.
Historical observational data of monthly averaged 500-hPa geopotential height was obtained from the European Center for Medium-range Weather Forecasts (ECMWF) Reanalysis version 5 (ERA5), which contained measurements from 1950 to 2019 on a 31 km grid [9]. For analysis of future projections, six climate models from the Coupled Model Intercomparison Project phase 6 (CMIP6) were used [7]. These models, also utilized by previous work among a broader ensemble of 14, were selected for their strong correlation with the observed summer North American Dipole (NAD) [6]. For each model, the data was downloaded from the Copernicus climate data store, and the same average monthly geopotential height at 500-hPa was the unit used to measure the NAD in figure 1. Our subset of six models was chosen based on data availability and computational constraints. Each model was analyzed over its historical simulation period (1850-2014), and two future scenarios (2015-2100) representing different fossil fuel emission trajectories: Shared Socioeconomic Pathway 2-4.5 (SSP245), reflecting a moderate, historically consistent emission path, and Shared Socioeconomic Pathway 3-7.0 (SSP370), representing a future with increased emissions, but not as severe as the higher SSPs. These were chosen because they could help find the minimum forcings required to cause a shift in the timing of the NAD. The extreme carbon emissions of SSP585 are likely to cause a shift [6], so investigating SSP245 and SSP370 could provide insights into the effects of more moderate increases in radiative forcing.
Principal Component Analysis.
All data was linearly detrended prior to analysis, meaning that linear trends in the data were removed. For example, if a measurement increased over decades, the increase would be removed because the focus of this research is on the fluctuations in the climate rather than long term trends. The atmospheric circulation patterns were analyzed with principal component analysis (PCA), also known as empirical orthogonal function analysis (EOF) to extract the leading modes of atmospheric variability [10]. This is done by finding uncorrelated linear combinations of the variables to reduce the dimensionality of the complex datasets. The code to run the PCA and plot the results was written in Python. The datasets were analyzed separately in March through May (Spring) and June through August (Summer) to compare the results of the two seasons. The NAD was characterized as the second principal component of 500-hPa geopotential height (PC2) in the domain of 35° to 65°N and 170° to 340°E, in accordance with previous research [5], [6]. The spatial pattern of the observed NAD was made with the ERA5 data, and the CMIP6 models were used to study how their NAD seasonal timing varies in the future depending on the different fossil fuel emission scenarios. The model circulation patterns were spatially correlated with the observed NAD by interpolating the loading values onto the ERA5 grid and calculating the Pearson’s correlation between them.
RESULTS.
The panel plots of INM-CM5-0 (Figure 2a) and NorESM2-MM (Figure 2b) visualize the results of the PCA, and they show how different models predict the outcome of the NAD in different scenarios. The red and blue represent opposite anomalies in atmospheric pressure with red being higher and blue being lower. These results of these models were chosen because they clearly show the differences in the predictions the models had for the future NAD. Figure 2a shows INM-CM5-0 predicting a noticeable dipole in the summer in the historical simulation, and both future emissions scenarios. It also shows a monopole, meaning just one color covering North America, in the spring in the historical and SSP245 plots. However, the SSP370 plot shows a dipole in the spring, as well as in the summer. Conversely, figure 2b shows NorESM2-MM predicting a dipole in every scenario except in SSP245 during the summer, which has a dipole.
The correlation coefficients seen in figure 3 represent how well the model’s spatial pattern correlates with the observed pattern of the ERA5 observed NAD, shown in figure 1. Thus, the higher the correlation is, the more likely it is that the NAD will appear in that simulation. The blue bars represent how well each model’s historical simulation was correlated with the observed historical NAD. The model that predicted the NAD the most in its historical simulation was INM-CM4-8, while INM-CM5-0 had the least. A high correlation in the historical summer simulation would increase trust in the results of the model because it has shown to be representative of the real climate.
INM-CM5-0, MIROC6, and TaiESM1 all had higher correlations in the summer than the spring in both scenarios. ACCESS-CM2 and NorESM2-MM had higher correlations in the spring in both scenarios. However, each model seems to have different outcomes for the NAD, with three predicting it to stay in the summer, INM-CM5-0, MIROC6, and TaiESM1. Others, like ACCESS-CM2 and NorESM2-MM, have higher correlations in the spring. INM-CM4-8 was the only model where the correlations did not align between SSP245 and SSP370.
DISCUSSION.
The results of the PCA were highly variable between each model. A shift in timing of the NAD would manifest as a higher correlation in the spring than the summer. As seen in figure 2b, NorESM2-MM has a visible dipole in all simulations except SSP245 in the summer. Additionally, its correlation coefficients in figure 3 suggest a shift into the spring, as SSP245 and SSP370 have higher correlations in the spring than the summer. However, the shift seems to be more drastic in SSP245 than SSP370, despite the increased radiative forcing simulated in the latter. This contradicts the established relationship of increased carbon emissions corresponding with a shift into the spring, and it contributes to the conflicting nature of the results.
As demonstrated, the incongruity of the results on the models along with the small sample size makes drawing conclusions on the general outlook of the NAD in moderate emission scenarios unclear. The hypothesis was neither supported nor refuted. The different emissions pathways had little to no effect on the timing or appearance of the NAD, and there is an apparent lack of consensus about the future of the NAD between the models. This could be due to biases present in the climate models that influenced their predictions. A possible modification to the methods would be a debiasing of the models. Linear detrending was performed to attempt to mitigate some of the biases, but in-depth debiasing, or bias correction, of datasets is very resource- and time-intensive, which went beyond the scope of the project.
Additionally, the small sample size of six climate models analyzed limits the generalizability of findings. Past research used the same models in a larger ensemble of fourteen models, which found that 9/14 predicted the NAD to shift into spring under severe fossil fuel emissions in Shared Socioeconomic Pathways 5-8.5 (SSP585) [6]. Their larger sample size allowed for stronger evidence for a potential shift for the North American climate patterns if greenhouse gas emissions increase at an uncontrolled rate, whereas this study aimed to determine if moderate increases also caused the same shift. However, the sample used in this study was limited due to a lack of availability for climate data caused by increased restrictions on the climate change research and data availability.
The findings of this study will help understand the risks posed to the North American Dipole and other similar climate patterns under different scenarios of greenhouse gas emissions. Dipoles are present on every continent, and they can be responsible for timing ecological patterns critical to the region. Therefore, it is important to learn how they could be affected by global warming by using modelling. This study also serves as a demonstration of the ways that climate models could be improved to produce more accurate and reliable predictions to apply to a broader use in weather forecasting and future climate research.
Further studies should be done that include more climate models to be analyzed, and they should be debiased to maximize their reliability. They should also use downscaling to improve the spatial resolution of the data. This process increases the number of data points in the data set, which makes the models more representative of reality. This, combined with bias correction, provides the most confidence in the analysis of global climate models. Another potential modification to the methods would be analyzing the NAD in individual months, instead of averages across spring and summer. This change could help find slight changes in timing, instead of a more drastic shift seen in SSP585. Ultimately, these could provide more detailed information on the interaction of radiative forcing on the NAD, rather than just a binary shift or lack thereof.
New research should also make use of future iterations on climate models as more data is collected, and more accurate models are developed. The Coupled Model Intercomparison Project phase 7 (CMIP7), the next iteration from CMIP6, has been published as of October 2025 [11]. This will provide more up-to-date data to compare across climate models and will aid in further research using global climate models to investigate the progress of global warming and its effects on the circulation of the atmosphere.
CONCLUSION.
In conclusion, the objective of this study was to use climate models to assess how different future pathways for climate change affected the seasonal timing of the summer North American Dipole (NAD). The results of six models’ predictions for the historical climate and future climate were analyzed using Principal Component Analysis (PCA), or Empirical Orthogonal Function analysis (EOF) and compared to the observed historical NAD. After the predictions of the models were visualized, three of them showed the NAD to stay in the summer, and the others showed it shifting into the spring. No conclusions can be supported by the results due to the high variability across models. This shows the difficulty of predicting the future and emphasizes the urgency to both make models more reliable and to take action to reduce carbon emissions to avoid any irreversible damage to the environment.
ACKNOWLEDGMENTS.
This research was made possible by the Research Experiences for High School Students program. We thank Nathaniel Freymeyer, Nicholas Means, and Natalie Wallace for their continued support throughout and after the program.
REFERENCES.
- National Academies of Sciences, Engineering, and Medicine. Radiative Forcing of Climate Change: Expanding the Concept and Addressing Uncertainties. (National Academies of Sciences, Engineering, and Med-icine, 2005)
- IPCC, “Summary for Policymakers” (Cambridge University Press, 2021)
- P. A. P. Moran, The Statistical Analysis of the Sunspot and Lynx Cycles. The Journal of Animal Ecology 18, 115 (1949).
- C. Strong, B. Zuckerberg, J. L. Betancourt, W. D. Koenig, Climatic dipoles drive two principal modes of North American boreal bird irrup-tion. Proc. Natl. Acad. Sci. U.S.A. 112 (2015).
- H. Bai, C. Strong, B. Zuckerberg, Drivers of an Ecologically Relevant Summer North American Dipole. Journal of Climate 36, 2387–2399 (2023).
- H. Bai, C. Strong, J. M. LaMontagne, I. V. Widick, B. Zuckerberg, A North American climate-masting-irruption teleconnection and its change under future climate. Science of The Total Environment 948, 174473 (2024).
- V. Eyring, et al., Overview of the Coupled Model Intercomparison Project Phase 6 (CMIP6) experimental design and organization. Geosci. Model Dev. 9, 1937–1958 (2016).
- D. C. Lafferty, et al., Statistically bias-corrected and downscaled climate models underestimate the adverse effects of extreme heat on U.S. maize yields. Commun Earth Environ 2, 196 (2021).
- H. Hersbach, et al., The ERA5 global reanalysis. Quart J Royal Meteoro Soc 146, 1999–2049 (2020).
- A. Hannachi, I. T. Jolliffe, D. B. Stephenson, Empirical orthogonal functions and related techniques in atmospheric science: A review. Intl Journal of Climatology 27, 1119–1152 (2007).
- J. P. Dunne, et al., An evolving Coupled Model Intercomparison Project phase 7 (CMIP7) and Fast Track in support of future climate assessment. Geosci. Model Dev. 18, 6671–6700 (2025).
Posted by buchanle on Thursday, May 14, 2026 in May 2026.
Tags: climate change, Global Climate Models, North American Dipole