Syllabus#

This is the course handout for Objective Analysis — a graduate-level introduction to the statistical and mathematical tools used in the atmospheric and oceanic sciences.

How to read these notes#

Throughout the notes, two types of highlighted boxes are used:

Figure / in-class demonstration

Green boxes mark figures, in-class demonstrations, or Python/Matlab examples to be shown during lecture. They are placeholders for visual material that accompanies the written derivations.

Example / deeper dive

Blue boxes contain worked examples, mathematical proofs, or deeper conceptual discussions that go beyond the main narrative. These are important — they are where most of the practice problems and physical intuition live.

Course Outline#

The course runs over 16 weeks, with the mid-term in Week 7 and the final in Week 16. There is no class in the final-exam week.

Chapter 0: The Idea Behind the Course#

Chapter 0: The Idea Behind the Course — read once now, revisit after each Part

Every method in this course is the same move: optimize a functional subject to constraints. Maximizing entropy under known moments gives you the PDF a hypothesis test needs; minimizing a loss gives you the best model, because a loss function is a negative log-likelihood. The Lagrange multiplier that appears in both turns out, in Part III, to be the eigenvalue.

This chapter is the most abstract material of the semester and nothing in Weeks 1-3 depends on it. Each Part opens with a short “The pattern” box pointing back here and filling in that Part’s row.

Part I: Foundations of Statistics#

Week 1-3: Rule101 — Week 1-3: Rule 101

  • The origin of statistics: why the bell curve?

  • Mean, variance and higher moments

  • Basic probabilities; unions; intersections; conditional probabilities; Bayes theorem

  • Tying it together: Moments, PDFs, and Bayes’ Theorem

  • Statistical significance testing

  • Hypothesis testing

  • Monte Carlo and resampling techniques

  • Compositing

  • Other common distributions

  • Non-parametric tests

Part II: Regression & Autocorrelation#

Week 4-6: Regression & AR1 — Week 4-6: Regression & AR1

  • Linear regression: least squares, slope & intercept

  • Theory of correlation: Pearson’s r, Fisher-Z, Spearman’s rank

  • Autocorrelation & effective sample size: AR1/red noise, Leith & Bretherton

  • Multiple regression: generalized normal equations, overfitting


Week 7 — Mid-term exam, covering Parts I and II.


Part III: Seeking Structure in Data#

Week 8-11: Seeking Structure in Data — Week 8-11: EOFs & Clustering

  • Linear algebra review: inner products, covariance matrices, inverse, rank & null space

  • Eigenvalues and eigenvectors: diagonalizing the covariance matrix

  • EOFs via eigenanalysis: maximizing explained variance, orthogonality, principal components

  • EOFs via Singular Value Decomposition, and its equivalence to eigenanalysis

  • EOFs with real data: weighting, standardization, presentation in physical units

  • How many EOFs to retain: North et al. (1982) and degeneracy

  • Cluster analysis: k-means and Lloyd’s algorithm

  • Self-organizing maps (SOMs): training, mapping, quantization & topographic error

Part IV: Time Series Analysis#

Week 12-15: Time Series Analysis — Week 12-15: Spectral Analysis & Filtering

  • Harmonic analysis: Fourier sums, the Nyquist frequency and aliasing

  • The discrete Fourier transform as orthogonal multiple regression

  • The power spectrum: line vs. continuous, and the resolution–reliability trade-off

  • The complex Fourier transform

  • Significance of spectral peaks: red-noise null, the F-test, a priori vs. a posteriori

  • Windows and finite data: convolution, the convolution theorem, boxcar & Hanning, WOSA

  • Filtering: response functions, non-recursive & Lanczos, recursive & Butterworth

  • Cross-spectrum analysis: co-spectrum, quadrature spectrum, coherence² and phase

  • Mixed space-time analysis: propagating waves and phase speed


Week 16 — Final exam. No class this week.