Operational Evaluation of a Wildfire Air Quality Model from a Forecaster Point of View
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Abstract An evaluation of an operational wildfire air quality model (WFAQM) has been performed. Evaluation metrics were chosen through an analysis of interviews and a survey of professionals who use WFAQM forecasts as part of their daily responsibilities. The survey revealed that professional users generally focus on whether forecast air quality will exceed thresholds that trigger local air quality advisories (e.g., an event), their analysis scale is their region of responsibility, they are interested in short-term (≈24 h) guidance, missing an event is worse than issuing a false alarm, and there are two types of users—one that takes the forecast at face value, and the other that uses it as one of several information sources. Guided by these findings, model performance of Environment and Climate Change Canada’s current operational WFAQM (FireWork) was assessed over western Canada during three (2016–18) summer (May–September) wildfire seasons. Evaluation was performed at the geographic scale at which individual forecasts are issued (the forecast region) using gridded particulate matter 2.5 (PM2.5) fields developed from a machine learning–based downscaling of satellite and meteorological data. For the “at face value” user group, model performance was measured using the Peirce skill score. For the “as information source” user group, model performance was measured using the divergence skill score. For this metric, forecasts were first converted to event probabilities using binomial regression. We find when forecasts are taken at face value, FireWork cannot outperform a nearest-neighbor-based persistence model. However, when forecasts are considered as an information source, FireWork is superior to the persistence-based model.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it