Interpreting Climate Change Through Data Science

Over the past few decades, the records for global mean annual temperature have steadily increased. With every additional increment in temperature, there is an exponential increase in accumulated heat in the atmosphere to cause changes in local and global weather patterns. As a consequence, during this period we witnessed a steady increase in weather-related disasters like prolonged droughts, heavier precipitation and flooding, severe tropical cyclones, and more recently with the sudden spike in wildfires across California. The long term impacts of climate change include shifting weather patterns, coastal erosion, and sea level rise.

Carbon dioxide (CO₂) is an important heat trapping (greenhouse) gas, which is released through human activities and burning fossil fuels. According to the United Nations Intergovernmental Panel on Climate Change (IPCC) report, since the beginning of the industrial revolution, the CO₂ concentrations have risen from 280 ppm (parts-per-million) to 407 ppm as of January 2018. During the same time period, the global mean temperature increased approximately 1⁰C above the pre-industrial baseline. In other words, the additional 125 ppm of CO₂ into the earth’s atmosphere corresponds to a rise in global temperature by around 1⁰C.

Anything to do with climate change is strongly debated, and the warming effects of CO₂ is no exception. Contrary to the general belief that CO₂ drives increase in global temperature, some argue that changes in temperature causes changes in atmospheric CO₂ — not the other way around. While examining past climate change using ice cores, scientists observed that CO₂ lags temperature. This has led some to argue that the CO₂ lags disproves the warming effects of CO₂. In reality, both sides of the argument hold water as there exists a positive feedback loop between CO₂ and global temperature. In other words, warming causes more CO₂ and more CO₂ causes warming. This positive feedback loop further exacerbates the effects of global warming.

Mauna Loa Observatory, Hawaii:

Atmospheric CO2 (ppm) @ Mauna Loa, Hawaii [1958–2020]Mauna Loa Observatory, Hawaii, is the oldest continuous CO₂ monitoring station, and has become the global standard for CO₂ levels due to its location far away from any continent. The above graph shows monthly atmospheric CO₂ levels measured at Mauna Loa Observatory. The x-axis represents time and the y-axis represents monthly CO₂ levels measured in parts-per-million (ppm). The CO₂ concentration has steadily increased over time, and the present levels are higher than it has been for several million years.

Forecasting Future CO₂ Levels:

To limit global warming to 1.5 ⁰ C, global carbon emissions need to fall by a staggering 45% by 2030 from the 2010 levels (IPCC, United Nations)

In this study, let’s leverage the advances in data-driven modeling to forecast future CO₂ levels. Once modeled, the predicted future CO₂ levels can be used to assess how far we are in achieving IPCC emission goals, and what that means in terms of global temperature increase. Time series prediction problems are a difficult type of predictive modeling problem. Unlike regression predictive modeling, time series also adds the complexity of a sequence dependence among the input variables.

Decomposition of CO₂ Time series:

Observing the atmospheric CO₂ time series from Mauna Loa, we could see there is an increasing uptrend with a strong seasonality. The CO₂ annual highs are during the Spring (May) when melting snow cover exposes soil microbes to release an abundance of CO₂. During Autumn (September) the CO₂ levels are typically low as the atmospheric CO₂ is being siphoned off by a burst of summer plants. Time series decomposition techniques allow us to deconstruct a time series into its individual component parts. The idea behind decomposition techniques is that any series can be represented as a sum (or a product) of its components: a)Level, b)Trend, c)Seasonality, and d) Residuals. To perform this data decomposition, we will use seasonal_decompose function in statsmodel packageas shown below:

Decomposition of Time Series — Additive Model : y(t) = Trend(t)+Seasonality(t)+Residual(t)The above plot shows the breakdown of the timeseries to their individual components. Since the seasonal variations are of the same magnitude across time, our choice of additive model decomposition is good.

Trend: Refers to the natural progression of a time series to relatively higher or lower values over a long period of time. With the CO₂ data, we see an uptrend.

Seasonality: Repeating patterns within a fixed time period. Seasonality can be weekly, monthly, or yearly.

Residual: Represents random and irregular (outliers and missing values) influences that are not consistent with the rest of the data.

Building a Forecast Model:

There are several time-series forecast methods and packages available, but Facebook’s Prophet is a quick and reliable way of arriving at very good results for that initial pass at modeling/forecasting time series data. Prophet is an additive regression model where nonlinear trends are fit by breaking down the time series into trend, seasonality and residual components. For details on the math behind Prophet, please read this paper.

From the Mauna Loa observatory, we have nearly 62 years of monthly data between March 1958 and January 2020. The monthly data between March 1958 and October 2013 is used for “training” the model, and the remaining dataset (November 2013 — January 2020) is used for “validating” the model. Once validated, the model is used to forecast CO₂ levels for the next 50 years.

CO₂ forecast for next 50 years until 2070The plot above shows the observed, estimated, and the forecast values of CO₂, along with the 95% confidence interval. In general, the FB Prophet model was very robust in capturing the longer term trends and seasonality. The 2015 Paris Agreement, the latest international climate treaty, is aiming to keep the global temperature increase this century well below 2 degrees C, and hopefully limit it to 1.5 degrees, below the pre-industrial levels.

At the current rate of growth in CO₂, levels will hit 530 ppm within 50 years, putting us on track to reach temperature boosts of perhaps more than 3 degrees C.

Relationship to RCP Pathways:

To standardize climate research and policy making, the Intergovernmental Panel on Climate Change (IPCC) has set up four scenarios indicative of different climate futures. The four Representative Concentration Pathways (RCP) are commonly referred as RCP2.6, RCP4.5, RCP6.0 and RCP8.5. The numerical values of the RCPs (2.6, 4.5, 6.0 and 8.5) refer to the target radiative forcing (measured in watts per meter square) in the year 2100. These scenarios are designed to cover from very high (RCP8.5) through to very low (RCP2.6) future greenhouse concentrations. By forecasting future CO₂ levels, we can estimate corresponding change in surface temperature and the rate of energy change per unit area. These correlated measures aid in defining and benchmarking different RCP scenarios.

Closing Thoughts:

Clearly countries around the world are not doing enough to reach climate change goals, and are operating business-as-usual. With increasing temperatures, we can expect to see the frequency and severity of extreme weather-related disasters to increase, melting glaciers and coastal inundation causing mass migrations, droughts and fire destroying rainforest and global food supply chains. More needs to be done to combat climate change and build a low-carbon future. There is a general perception that low-carbon future means losing out on conventional jobs and stronger economy. Contrary to the general belief, companies like Tesla who are at the forefront of the fight against carbon emissions, and showing a way to getting it done.

 

Original post: https://towardsdatascience.com/interpreting-climate-change-through-data-science-321de6161baf

Leave a Reply

Your email address will not be published. Required fields are marked *