When working with the financial data mid-course, we approached the question of forecasting: if we know at least a fragment of the past of a time series, will we be able to predict what the future values might be? This module exposes us to an ambitious mathematical construct called an ARIMA model that does three things at once: attempts to fit a regression on autocorrelations, calculates differences between consecutive values instead of looking at the values as such (possibly more than onces), and observes how model error evolves over time. With the combined (assumed) powers of these three elements, a single-point forecast can be produced.
Not all time series data is equally "predictable". There may be periodic (seasonal) variations, the values may be growing or shrinking over time, and there may be sources of noise, bias, or error in the measurement. The ARIMA model has three parameters that one can adjust to try to counter for such phenomena in the observed data. It is not a simple task, however, to find the best combination.
This module will help you do the following:
After watching the video, find online a visualization of historial data of a climate-related phenomenon of your choice (options include coastal flooding, polar ice caps, coral reef bleaching, athmospheric gas composition, sea temperatures, crop yield etc.). Sketch on top of the visualization your own attempted forecast of what you think will happen over the next three decades. You do not need to employ any particular method, your best guess is enough. Write down, however, a paragraph or two to describe what in the data makes you think that might be a realistic future.
After this module, you should be familiar with the following concepts:
Remember that you can always look concepts up in the glossary. Should anything be missing or insufficient, please report it.