Exploring Patterns · Lesson 1 of 5
Look Before You Model
~12 min
The Concept
The Bluewater Basin dataset is finally clean. Your team has fifteen wells, six rain gauges, twelve stream sensors, and five years of data, all quality-checked, unit-corrected, and gap-documented. It took six weeks. Now your supervisor delivers the surprising instruction: 'Before you run a single model, spend a week just looking at it.'
She tells you about a famous cautionary tale: a research team spent three months building an elaborate regression model to explain nitrate variability in their watershed, published the results, and only later discovered that nitrate at the most important well simply tracked the agricultural calendar, fertiliser application in spring, crop uptake in summer, leaching in autumn. The pattern was visible in the first scatter plot. The elaborate model added nothing.
Exploratory Data Analysis (EDA, meaning just looking closely at your data before doing anything fancy to it) is the discipline of systematically looking at your data, distributions, correlations, seasonal patterns, spatial patterns, before committing to any statistical model. Its purpose is to let the data tell you what questions are worth asking rather than imposing a hypothesis before you know what the data contains.
EDA isn't a preliminary chore. It's where most real discoveries happen. Unexpected patterns, missed cleaning errors, contradictions between stations, these all show up in plots before they show up in models. A model can confirm a pattern you already understand; it cannot discover a pattern you've never looked for.
The Analogy
It is like walking around a new house before you start rearranging furniture. If you shove a couch against the wall before noticing where the windows and doors are, you will probably have to move it again. A few minutes of just looking saves hours of redoing work later.
Why Real Researchers Care
John Tukey, the statistician who coined 'exploratory data analysis' in 1977, wrote that the greatest value of a picture is when it forces us to notice what we never expected to see. This is as true in environmental science today as it was then. Major discoveries, El Niño's global fingerprint, the ozone hole, declining stream flows, were all identified visually in raw data plots before any formal statistical confirmation.
Quick Check
Q1. What is the main purpose of Exploratory Data Analysis (EDA)?
Your Goal
Before this lesson's simulation, write down two patterns you would expect to find in Bluewater Basin's nitrate data based on what you already know about the watershed. Then check whether you find them in the exploration exercises below.
Hint: Think about the agricultural zone, seasonal fertiliser application, and the locations of wells relative to farmland.
Teach It Back
Explain why building a regression model before doing EDA is risky, what specific bad outcome could result?