“Old Texts, New Tech, Better Theory”: Applying Machine Learning to Textual Weather Data from Historical Ship Logbooks
Bibliographic record
Abstract
During oceanic expeditions, pre-modern sailors meticulously recorded information about their longitude and latitude, the local wind conditions, and the state of the sea. For a long time, prior to precision instrumentation, sailors provided qualitative recordings of wind speed instead of quantitative (e.g.: “light breeze” instead of 5 meters/second). For that reason, this textual data requires additional processing before being usable for comparison with modern instrumental data or reanalysis products. In particular, the phrases used in wind descriptions can be classified using the Beaufort Wind Force Scale (codified in 1805), that consists of thirteen base wind force levels assigned a numerical value. Manually categorizing all the distinct and unique variations on the wind information can be ambiguous and time consuming. Because of historical weather data’s importance for climate science, we investigated if machine learning could speed up this process while producing accurate results.Using a novel dataset of >100,000 (sub)daily maritime weather recordings from historical whaling ship logbooks housed across New England archives and covering the period 1820-1890, here we show that k-means nearest neighbors and density based spatial clustering models, while efficient, generate outputs with reduced accuracy when compared to the data classified by humans. However, there is a noticeable improvement in the quality of the clustering when we introduce the Beaufort Wind Force Scale’s thirteen categories as starting centroids. These results show that machine learning could be a useful tool for wind term processing and that well-placed human input aids in the accuracy of outcomes. Therefore, cross-validation methods are employed to help with the interpretability of the machine models utilized. Additionally, various neural network clustering models are evaluated regarding their efficacy, such as a two sliding windows text GNN-based (TSW-GNN) model, since its graph-based approach has demonstrated improved accuracy in classifying textual data as compared to language representation models.
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.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.013 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.001 |
| Bibliometrics | 0.004 | 0.003 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.002 | 0.002 |
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 itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".