PM <sub>2.5</sub> concentrations based on near-surface visibility in the Northern Hemisphere from 1959 to 2022
Bibliographic record
Abstract
Long-term PM 2.5 data are essential for the atmospheric environment, human health, and climate change. PM 2.5 measurements are sparsely distributed and of short duration. In this study, daily PM 2.5 concentrations are estimated using a machine learning method for the period from 1959 to 2022 in the Northern Hemisphere based on near-surface atmospheric visibility. They are extracted from the Integrated Surface Database (ISD). Daily continuous monitored PM 2.5 concentration is set as the target, and near-surface atmospheric visibility and other related variables are used as the inputs. A total of 80 % of the samples of each site are the training set, and 20 % are the testing set. The training result shows that the slope of linear regression with a 95 % confidence interval (CI) between the estimated PM 2.5 concentration and the monitored PM 2.5 concentration is 0.955 [0.955, 0.955], the coefficient of determination ( R 2 ) is 0.95, the root mean square error (RMSE) is 7.2 µg m −3 , and the mean absolute error (MAE) is 3.2 µg m −3 . The test result shows that the slope within a 95 % CI between the predicted PM 2.5 concentration and the monitored PM 2.5 concentration is 0.864 [0.863, 0.865], the R 2 is 0.79, the RMSE is 14.8 µg m −3 , and the MAE is 7.6 µg m −3 . Compared with a global PM 2.5 concentration dataset derived from a satellite aerosol optical depth product with 1 km resolution, the slopes of linear regression on the daily (monthly) scale are 0.817 (0.854) from 2000 to 2021, 0.758 (0.821) from 2000 to 2010, and 0.867 (0.879) from 2011 to 2022, indicating the accuracy of the model and the consistency of the estimated PM 2.5 concentration on the temporal scale. The interannual trends and spatial patterns of PM 2.5 concentration on the regional scale from 1959 to 2022 are analyzed using a generalized additive mixed model (GAMM), suitable for situations with an uneven spatial distribution of monitoring sites. The trend is the slope of the Theil–Sen estimator. In Canada, the trend is −0.10 µg m −3 per decade, and the PM 2.5 concentration exhibits an east–high to west–low pattern. In the United States, the trend is −0.40 µg m −3 per decade, and PM 2.5 concentration decreases significantly after 1992, with a trend of −1.39 µg m −3 per decade. The areas of high PM 2.5 concentration are in the east and west, and the areas of low PM 2.5 concentration are in the central and northern regions. In Europe, the trend is −1.55 µg m −3 per decade. High-concentration areas are distributed in eastern Europe, and the low-concentration areas are in northern and western Europe. In China, the trend is 2.09 µg m −3 per decade. High- concentration areas are distributed in northern China, and the low-concentration areas are distributed in southern China. The trend is 2.65 µg m −3 per decade up to 2011 and −22.23 µg m −3 per decade since 2012. In India, the trend is 0.92 µg m −3 per decade. The concentration exhibits a north–high to south–low pattern, with high-concentration areas distributed in northern India, such as the Ganges Plain and Thar Desert, and the low-concentration area in the Deccan Plateau. The trend is 1.41 µg m −3 per decade up to 2013 and −23.36 µg m −3 per decade from 2014. The variation in regional PM 2.5 concentrations is closely related to the implementation of air quality laws and regulations. The daily site-scale PM 2.5 concentration dataset from 1959 to 2022 in the Northern Hemisphere is available at the National Tibetan Plateau/Third Pole Environment Data Center (https://doi.org/10.11888/Atmos.tpdc.301127) (Hao et al., 2024).
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 | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| 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 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".