NEWS

---

appc 3.0.0

• Require terra 1.9.46 or newer for the compressed NetCDF extraction fix, and allow completed AQS refreshes to be reused with options(appc.refresh_aqs = FALSE).
• Retry transient EPA API failures and resume incomplete AQS refreshes from harmonized state checkpoints without replacing the last complete annual extract.
• Include Idaho and North Dakota in AQS API refreshes for complete contiguous-US coverage.
• predict_pm25() now accepts geomarker s2cd objects directly, using their stored dates and retaining the existing list-of-tibbles output.
• Release model generation 3 for January 2017 through June 2026 using geomarker 0.2.0, pinned to release commit 373820d0ab8082f856f7bf0f639f3128111d825a.
• Share geomarker-backed predictor assembly between the committed training-data script and prediction, preserving input location/date order and duplicates.
• Keep AQS outcome harmonization and the existing GRF training specification.
• Remove standalone environmental assessment and installation APIs; use geomarker for those functions. Elevation now comes from USGS 3DEP rather than the v2 PRISM source, so the v2 model cannot be reused with these inputs.
• Record model generation separately from package version, preserve training provenance, and require matching metadata for local v3 model artifacts.
• Distribute the trained forest and its 618,918-row training dataset together under rf_pm_v3, with checksums and the exact model/data provenance.
• Retain the January 2017 through June 2026 date range used by v2. Daily LOLO median absolute error is 1.20 micrograms per cubic meter and median Spearman correlation is 0.84, matching the reported v2 values. Median coverage of the reported 95% intervals is 69% (v2: 70%). Refreshed EPA outcomes add 7,346 observations and three locations relative to v2, so the evaluation rows differ.

appc 2.0.1

• Load the PM2.5 model from a portable RDS artifact (#119).

Release notes

appc 2.0.0

• Release a new PM2.5 model and training data covering January 2017 through June 2026 (#118).

Release notes

appc 1.0.2

• Switch model serialization from qs to qs2 (#117).

Release notes

appc 1.0.1

• Add keep_predictors to predict_pm25() to optionally return predictors alongside predictions.
• Validate buffer arguments and improve the README and vignettes.
• Retain the existing model, data, and prediction values.

Release notes

appc 1.0.0

• Retrain the random forest with updated observations, replacing predictions across the entire January 2017 through October 2025 coverage period.
• Update Annual NLCD to include 2024 data, and update HMS smoke, NARR, and gridMET availability for 2025.
• Add traffic summaries by vehicle class and NEI point-source assessments.
• Add AQS Data Mart installation for more recent daily PM2.5 observations.
• Add force_reinstall for refreshing installed source data.
• Fix maximum-date validation, update gridMET download URLs, and use EARTHDATA_USER for NASA authentication.

Release notes

appc 0.6.1

• Fix downloading the latest model when rendering the model-performance vignette.

Release notes

appc 0.6.0

• Add predict_pm25_date_range() and latlon_to_s2_cell().
• Cache the loaded model in memory for repeated predictions within an R session.
• Associate model and training artifacts with the package major version, allowing minor and patch releases to reuse the same model.
• Add configurable minimum and maximum dates to internal input validation.
• Update the NLCD fractional-imperviousness download URL.

Release notes

appc 0.5.1

• Fix the NLCD fractional-imperviousness download URL (#90).

Release notes

appc 0.5.0

• Extend supported prediction dates through the end of 2024; bulk AQS training observations were available through October 2024 at release time.
• Add the newer Annual NLCD product as an assessment function, without including it in the model.
• Distribute MERRA-2 data under dedicated data-release tags rather than package version tags.

Release notes

appc 0.4.0

• Incorporate available 2024 observations and update the trained model.
• Fix MERRA source installation using response URLs.
• Install smoke data from its original source on first use instead of distributing it as a package release asset.

Release notes

appc 0.3.0

• Add gridMET assessment functions and retrain the model using gridMET weather predictors in place of NARR variables, retaining NARR hpbl.

Release notes

appc 0.2.0

• Publish the first stable release with a finalized PM2.5 prediction model.

Release notes

appc 0.1.0

• Publish the initial release for model development.

Release notes