# 096. Burned-area severity benchmark

**Question:** Can recorded conditions estimate burned area beyond a simple baseline?

## Result

The largest computed metric is 6.548 for Extra trees; the smallest is 6.047 for Median baseline. Metric: mae (Hectares MAE).

This models severity conditional on a recorded observation, not fire occurrence. Log predictions can understate rare large areas; compare RMSE as well as MAE.

![Burned-area severity benchmark](outputs/chart.png)

| model | mae | rmse | r2 | bias |
| --- | --- | --- | --- | --- |
| Median baseline | 6.047 | 21.26 | -0.06553 | -5.272 |
| Ridge | 6.381 | 20.93 | -0.03325 | -3.721 |
| Extra trees | 6.548 | 20.84 | -0.02421 | -2.793 |

The chart shows 3 of 3 result rows; the table previews the first 3 in the analysis-defined order. [Download the full result table](outputs/results.csv). Numerical values are computed from the source; missing results stay unavailable.

## Method

Grouped-input regression on log1p area, inverse-transformed before original-unit scoring.

The study uses shared source preparation and reusable statistical routines. Its specific transformations are in [analysis.py](analysis.py), and common model/evaluation code is in [portfolio/methods.py](../../portfolio/methods.py). The [notebook](analysis.ipynb) executes the study and displays the saved results.

## Evaluation

Fixed 80/20 split of unique input groups (seed 42); exact input duplicates stay together. Training observations: 411; test observations: 106. Fixed configurations specified before scoring; no tuning on the holdout.

Target: `area`. Features: x, y, month, day, ffmc, dmc, dc, isi, temp, rh, wind, rain. Model results and row membership are recorded in [evaluation.json](outputs/evaluation.json).

## Decision and limitations

Use disappointing as well as successful benchmarks to define what additional event data are needed.

The sample consists of recorded fire observations, not all places and times at risk. Zero recorded area is not evidence of no ignition. No precise dates or event identifiers are supplied. This is not an emergency forecasting system. 

Related studies may share observations or holdouts. These are focused analytical studies, not independent replications or deployed business systems. Any model refinements informed by these results need new untouched evaluation data. No commercial impact is inferred from an association or backtest.

## Reproduce

From the repository root, after installing `requirements.txt`:

```powershell
python projects/096-burned-area-severity-benchmark/analysis.py
```

Source data are downloaded automatically if absent. Original archives are retained unchanged and checked by SHA-256. The cleaned cache normalizes column names; field-specific changes are visible in [data preparation](../../portfolio/data.py). Runtime evidence is in [receipt.json](outputs/receipt.json).

## Source

[Forest Fires](https://archive.ics.uci.edu/dataset/162/forest+fires), Cortez and Morais (2007). [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/). Source data are transformed and aggregated in this study. [Source provenance](../../data/provenance/fires.json) and [prepared-data audit](../../data/provenance/fires_prepared.json) record the downloaded files, field coverage and hashes.