Welcome to BEE 4850/5850!

Lecture 01

January 22, 2025

Course Overview

About Me

Instructor: Prof. Vivek Srikrishnan,

Interests:

  • Bridging Earth science, data science, and decision science to improve climate risk management;
  • Unintended consequences which result from neglecting uncertainty or system dynamics.

Meet My Supervisors

My Supervisors

What Do You Hope To Get Out Of This Course?

Take a moment, write it down, and we’ll share!

Course Motivation

Why Does Data Analysis Matter?

  • Scientific insight;
  • Decision-making;
  • Understanding uncertainty

The Ideal

Goal: Obtain data that is so self-evident that you don’t need to do statistics!

XKCD 2400

Source: XKCD 2400

But in Actuality

  • Except in the most controlled experiments, data are noisy;
  • The causes of the data cannot be extracted from the data alone;
  • The reasons for statistical analyses are not found in the data themselves.

Unique/Challenging Features Of Data

There are many features of environmental (and biological!) data which make data analysis interesting and hard.

Extreme Events

Extreme Events

Source: Doss-Gollin & Keller (2023)

Extreme Events

XKCD 2107

Source: XKCD 2107

Correlated Uncertainties

Correlations Projections

Source: Errickson et al. (2021)

Non-Stationarity

Non-Stationary Trends

Source: Fagnant et al. (2020)

Forcing & Model Uncertainty

Different Uncertainties for SLR

Source: Doss-Gollin & Keller (2023)

Deep Uncertainty

Deep Uncertainty for Future Climate Projections

Source: Srikrishnan et al. (2022)

Causation Is Not Association

  • We’ve all heard “correlation (association) does not imply causation”
  • But confounds/noise can mean causation does not imply association

Correlation not Causation Meme

Source: Richard McElreath

Description and Inference are Connected

Except for superficial tasks:

  • Knowing what is important to describe requires a model;
  • Understanding how the data might differ from the population requires a model.

Spidermen Meme

Some Problems With The “Standard” Data Analysis Toolkit

  • Statistical assumptions may not be valid;
  • “Null” vs “Alternative” hypotheses and tests may be chosen for computational convenience, not scientific relevance.
  • Many different substantive models can imply the same statistical model.

Important: “Big” data doesn’t solve the problem!

My Philosophical Position

  • Probability theory helps us deduce logical implications of theories conditional on our assumptions
  • Cannot use an “objective” procedure to avoid subjective responsibility

Bart Statistics Meme

Model-Based Data Analysis

We can (transparently):

  • Examine logical implications of model assumptions (including interventions/out-of-sample generation).
  • Assess evidence for multiple hypotheses by generating simulated data.
  • Identify opportunities to design future experiments or observations to distinguish between competing hypotheses.

Model-Based Data Analysis

Models are how we assess evidence for theories.

Rock Paper Scissors meme

Course Organization

timeline
      Introduction (Weeks 1-2): Overview
                  : Hypothesis Testing and Scientific Inference
      Probability Fundamentals (Weeks 2-5): Prob/Stats "Review"
                                          : Modeling Data-Generating Processes
                                          : Bayesian Statistics
                                          : Time Series
                                          : Model Fitting
      Simulation Methods (Weeks 6-7): Monte Carlo
                                    : Bootstrap
      Model Evaluation (Weeks 8-9): Cross-Validation
                      : Model Selection
      Useful Models (Weeks 10-13): Extreme Values
                                 : Missing Data
                                 : Mixture Models
                                 : Gaussian Processes
      Experimental Design (Weeks 14): Confounds
                                       : Controls and Designing Experiments

Course Policies

Background Knowledge: Computing

  • Basics (at the level of CS 111x)
  • No specific language requirement.
  • Some extra work/effort may be needed if you haven’t coded in a while.
  • May need some additional familiarity with statistical packages (and “light” optimization)

Background Knowledge: Probability/Statistics

  • ENGRD 2700/CEE 3040
  • Summary statistics of data
  • Probability distributions
  • Basic visualizations
  • Monte Carlo basics

Grades

Assessment Weight
Participation 5%
Readings 10%
Quizzes 10%
Literature Critique 15%
Homework Assignments 30%
Term Project 30%

Overall Guidelines

  • Collaboration highly encouraged, but all work must reflect your own understanding
  • Submit PDFs on Gradescope
  • 50% penalty for late submission (up to 24 hours)
  • Standard rubric available on website
  • Always cite external references

Literature Critique

  • Select a paper which involves some type of statistical or data analysis
  • Critique choices: do they support the scientific conclusions?
  • Submit a 2-3 page writeup at the end of the semester
  • If you’re unsure where to look for a paper, talk to Prof. Srikrishnan

Readings

  • Several readings assigned for discussion throughout the semester.
  • Annotation assignments on Canvas: by end of the week.
  • Will discuss paper at start of class next Monday.
  • 5850 students: Write a one-page summary.

Quizzes

  • Assigned on Gradescope (almost) weekly.
  • Short (2-3 problems), focused on material from that week.
  • Due before class next Monday.

Homework Assignments

  • More in-depth problems
  • Roughly 2 weeks to complete
  • Regrade requests must be made within one week
  • 5850 Students: Some extra problems

Term Project

  • Analyze a data set of interest using model(s) of your choice
  • Can work individually or groups of 2
  • Several deliverables throughout the semester
  • Final in-class presentation and report

Attendance

Not required, but students tend to do better when they’re actively engaged in class.

Office Hours

  • Instructor: MW 1-2 PM, 318 Riley-Robb
  • TA: MTh 10:30-11:30 AM, 319 Riley-Robb
  • Almost impossible to find a time that works for all (or even most); please feel free to make appointments as/if needed.

Accomodations

If you have any access barriers in this class, please seek out any helpful accomodations.

  • Get an SDS letter.
  • If you need an accomodation before you have an official letter, please reach out to me ASAP!

Academic Integrity

Hopefully not a concern…

  • Collaboration is great and is encouraged!
  • Knowing how to find and use helpful resources is a skill we want to develop.
  • Don’t just copy…learn from others and give credit.
  • Submit your own original work.

Academic Integrity

Obviously, just copying down answers from Chegg or ChatGPT and passing them off as your own is not ok.

LLMs: Bullshit Generators

Think about ChatGPT as a drunk who tells stories for drinks.

It will give you plausible-looking text or code on any topic, but it doesn’t know anything beyond what it “overheard.”

ChatGPT can be useful for certain tasks (e.g. understanding code errors), but may neglect context for why/when certain information or solutions work.

ChatGPT: The Stochastic Parrot

Must specifically call out where you used ChatGPT in your work (beyond simple referencing; see syllabus for details).

Class Tools

Communications

Use Ed Discussion for questions and discussions about class, homework assignments, etc.

  • Try to use public posts so others can benefit from questions and can weigh in.
  • I will make announcements through Ed.

Email

When urgency or privacy is required, email is ok.

Important

Please include BEE4850 in your email subject line! This will ensure it doesn’t get lost in the shuffle.

Better: Use Ed Discussion and reserve email for matters that are particular urgent and/or require privacy.

Course Website

https://viveks.me/simulation-data-analysis

  • Central hub for information, schedule, and policies
  • Will add link and some information to Canvas (assignment due dates, etc)

Computing Tools

  • Course is programming language-agnostic.
  • Assignments will have notebooks set up for Julia (environments, etc) on GitHub.

Some Tips For Success

  • Start the homeworks early; this gives time to sort out conceptual problems and debug.
  • Ask questions (in class and online) and try to help each other.
  • Give me feedback!

Upcoming Schedule

Next Classes

  • Hypothesis testing and decision-making
  • Generative models

Assessments

Homework 1 available; due next Friday (2/7).

References

References (Scroll for Full List)

Doss-Gollin, J., & Keller, K. (2023). A subjective Bayesian framework for synthesizing deep uncertainties in climate risk management. Earths Future, 11, e2022EF003044. https://doi.org/10.1029/2022ef003044
Errickson, F. C., Keller, K., Collins, W. D., Srikrishnan, V., & Anthoff, D. (2021). Equity is more important for the social cost of methane than climate uncertainty. Nature, 592, 564–570. https://doi.org/10.1038/s41586-021-03386-6
Fagnant, C., Gori, A., Sebastian, A., Bedient, P. B., & Ensor, K. B. (2020). Characterizing spatiotemporal trends in extreme precipitation in Southeast Texas. Nat. Hazards, 104, 1597–1621. https://doi.org/10.1007/s11069-020-04235-x
Srikrishnan, V., Guan, Y., Tol, R. S. J., & Keller, K. (2022). Probabilistic projections of baseline twenty-first century CO2 emissions using a simple calibrated integrated assessment model. Clim. Change, 170, 37. https://doi.org/10.1007/s10584-021-03279-7