Weather baseline: 24-hour temperature regression

What this project does

This project builds a small, reproducible baseline model that predicts daily maximum temperature 24 hours ahead for a single station. It is an educational pipeline that demonstrates how to (1) prepare station-like tabular observations, (2) build a simple regression baseline, (3) evaluate against simple meteorological baselines (persistence and seasonal climatology) using a chronological split, and (4) publish a short, reviewable forecast comparison with clear limitations and links to official warnings.

This is not an operational forecast system and it must not be used for safety decisions. It is a teaching example for final-year students to practise model-building and evaluation that accounts for chronology and extreme-period assessment.

Data

The code constructs a small synthetic station-like dataset (daily observations for ~3 years) as a faithful stand-in for a single weather station. Each row contains a date and simple predictors that would plausibly be available at forecast issue time (previous-day temperature, previous-day humidity, wind speed, and day-of-year). The target is the next-day maximum temperature (24-hour horizon). The dataset is synthetic (generated deterministically from a seed) so the example is fully self-contained and deterministic.

How to build it

Install the required packages listed in requirements.txt. 2. Open project.py and inspect the run(seed=42) function to see data construction, baselines and the regression pipeline (StandardScaler + LinearRegression). 3. Run the script to produce JSON metrics (it prints them). 4. Read tests in test_project.py to understand required metrics and determinism checks.

How to run

Run the module directly with Python 3.11+: python project.py. The script calls run() and prints JSON with evaluation metrics for the held-out chronological test set.

How to test

Run pytest in the project directory: pytest -q. Tests check that run(42) returns expected keys, that values are in sensible ranges, that a conservative R2 floor is passed on the held-out set, and that the main metric is deterministic.

Results

Measured by running python project.py (seed 42) on 2026-10-23; the tests in test_project.py assert the floor.

Metric	Value
horizon_hours	24
model	LinearRegression
n_samples	1200
test_size	240
mae	2.2440
r2	0.8695
persistence_mae	2.3284
climatology_mae	2.2937
extreme_low_mae	3.2226
extreme_high_mae	4.0620
notes	Synthetic station-like data; baselines: persistence (yesterday) and seasonal climatology. See official warnings from national services (e.g. https://mausam.imd.gov.in/) for real decisions.

Limitations

The dataset is synthetic and from one virtual station; real station metadata issues (sensor changes, missing observations, quality control) are not addressed.
The model is a simple linear regression baseline; it is intentionally lightweight for pedagogic clarity.
This does not replace official forecasts or warning services. For real decisions, consult an official meteorological agency and their severe-weather warnings (for India, see the India Meteorological Department: https://mausam.imd.gov.in/).

Next steps

Replace synthetic data with real station observations and ensure each predictor is available at forecast issue time.
Try non-linear models, lagged features, and probabilistic forecasts.
Validate with longer multi-year backtests, cross-validation that respects seasons, and test against official forecasts where available.

Full case study, build steps and evaluation checks: Weather prediction model on networkershome.com

Built by Networkers Home · Learn the skills in the AI Engineer Course · Part of 100 AI projects with code