HeatStressR 2.3.0

HeatStressR 2.3.0 improves maintainability, parallel execution, aggregate API ergonomics, and scientific preprocessing transparency.

Highlights

• Centralizes Liljegren forcing, status, dewpoint-policy, and humidity preprocessing across local and PSOCK execution paths.
• Replaces foreach/doParallel backend manipulation with cross-platform base parallel::parLapply() PSOCK workers and removes both package dependencies.
• Makes heat_indices(tas, hurs) select the five indices supported without wind while preserving the existing seven-index default when wind is supplied.
• Adds row-level wind_clamped, radiation_clamped, radiation_zeroed_below_horizon, and dewpoint_adjusted diagnostics.
• Preserves legacy public argument order, output structure, NA alignment, dewpoint semantics, and established numerical tolerances.

Validation

The merged release commit passed the complete R CMD check matrix on Linux, Windows, and macOS, including R 4.1, release, oldrel, and devel.

This release is currently available on GitHub only and has not been submitted to CRAN.

Full changelog: v2.2.1...v2.3.0

HeatStressR 2.2.1 — First CRAN Release

• Examples completing within five seconds are now unwrapped and run as
  ordinary examples.
• Remaining executable long-running examples now use \donttest{} instead of
  \dontrun{}.

Available on CRAN: HeatStressR 2.2.1

HeatStressR 2.2.0

Added

• heat_indices() calculates multiple requested non-Liljegren indices from
  aligned observations, reusing shared validation and vapour-pressure work.
  wbgt.Bernard is available as an explicit opt-in when dew point is supplied.
• Reproducible benchmarks cover optimized non-Liljegren paths across cool,
  mixed, and hot inputs from 1 through 1,000,000 rows.

Changed

• wbgt.Bernard() uses a vectorized, bracketed bisection solver in place of
  one stats::optimize() call per valid row. Its public arguments and
  preprocessing behavior are unchanged.
• hi() evaluates the Rothfusz polynomial only for rows where it applies.
  It now returns degrees Celsius, matching its Celsius input and the rest of
  the package. To reproduce a previous Fahrenheit result, use
  hi(tas, hurs) * 1.8 + 32.
• Shared vapour-pressure work and small vector-allocation improvements reduce
  repeated computation in closed-form indices. Existing exported function
  names, formal arguments, positional calling conventions, and legacy
  unequal-length behavior of tashurs2vap.pres() and dewp2hurs() are
  retained.

End-to-End Parallel Liljegren WBGT

• Multi-worker wbgt.Liljegren() execution now uses a temporary
  foreach/doParallel PSOCK backend. Each worker calculates solar geometry
  and completes its WBGT shard end-to-end; the caller's existing foreach
  backend is restored after the call.
• Parallel benchmark runners now compare a fixed total row count across worker
  counts. The worker sweep accepts LILJEGREN_PARALLEL_ROWS (default
  1,000,000) rather than increasing rows per worker.
• Reduced the benchmark suite to the maintained three-way pre-fork comparison
  and fixed-workload internal-worker sweep, with refreshed result records.

Vectorized Liljegren Solver with Timestamp-Aware Solar Geometry

• wbgt.Liljegren() now defaults to engine = "batch"; select
  engine = "scalar" for the prior reference execution path. Batch retains
  automatic scalar fallback for unresolved rows.
• wbgt.Liljegren() and calZenith() now accept solar_time = "timestamp"
  or "date_noon". The inherited hour argument remains a compatibility
  alias.
• Removed the fork-only gmt_offset option. Supply timezone-aware timestamps
  or convert local observations to UTC before calculating solar geometry.
• Removed the fork-only averaging_period option. HeatStressR evaluates the
  supplied timestamp as an instantaneous observation; callers align
  interval-mean or accumulated data before calculation.
• wbgt.Liljegren() now accepts direct_fraction, the direct share of total
  supplied shortwave radiation. It may be one value or a row-aligned vector
  and defaults to 0.8.

Parallel batch calls optimization

• Parallel batch calls with diagnostics = FALSE now return compact worker
  summaries rather than row-level solver metadata. Final values and aggregate
  numerical-failure warnings are unchanged, while PSOCK result transfer is
  substantially reduced.
• Liljegren performance runners now time the public default
  diagnostics = FALSE; residual, fallback, and worker-parity validation is
  performed separately outside timed sections.
• Documented how to choose external foreach parallelism versus in-package
  workers, including the requirement to avoid nested worker pools.
• Updated documentation generation to roxygen2 8.0.0 and bound affected
  roxygen blocks to their exported public functions.

Vectorized Solar Geometry: Timestamp-Term Caching for Faster Multi-Location WBGT

2.1.3 — 2026-07-22

• wbgt.Liljegren() now caches timestamp-only solar-time terms and reuses
  them across coordinate pairs, reducing repeated solar-geometry work while
  retaining aligned vector results.
• Refreshed the Liljegren coordinate benchmarks, including isolated
  fixed/grouped/unique coordinate workloads and timestamp-cache E2E results.
• The unique-coordinate Liljegren E2E benchmark now uses the full 129,024-row
  fixture, matching the fixed- and grouped-coordinate workloads. Recorded
  results and reproduction commands have been updated accordingly.

Vectorize Liljegren coordinates

wbgt.Liljegren() now accepts scalar or row-aligned lon and lat. Solar geometry is computed once per distinct coordinate pair before either scalar or batch heat-balance solving begins.

Fixed broken relative links during CRAN packaging

v2.1.1

chore: fixed relative links, update version and NEWS for 2.1.1 release

HeatStressR 2.1.0 — Multi-core support; prep for CRAN

Added

• wbgt.Liljegren() batch execution accepts an explicit workers count.
  Values above one use cross-platform PSOCK workers while retaining output,
  warning, and diagnostic ordering. Worker count is user-selected within the
  currently permitted worker limit; it is not automatically reduced based on
  workload size except that it is capped at the number of input rows.
• New internal parallel execution layer (R/wbgt-parallel.R) providing
  max_liljegren_workers(), validate_workers(), split_liljegren_chunks(),
  solve_liljegren_batch_raw_chunk(), and solve_liljegren_parallel().
• Benchmark harnesses for multicore performance evaluation:
  benchmarks/benchmark-liljegren-workers-1-to-6x87600.R (sweeping worker
  counts with configurable ROWS_PER_WORKER) and
  benchmarks/benchmark-liljegren-parallel.R (annual workload
  profiling). Results recorded on Apple M2 Max.
• DESCRIPTION now lists parallel as an explicit import.

CRAN preparation

• Standardised licence metadata and package authorship.
• Added package and Liljegren-model citations.
• Added runnable Liljegren examples and expanded installed documentation.
• Clarified that HeatStressR is an R implementation of the Liljegren method,
  not a claim of improvement over the original Liljegren program.
• Made worker-limit detection respect CRAN check core limits.
• Added CRAN submission notes and related regression tests.

Changed

• wbgt.Liljegren() with engine = "batch" automatically falls back to
  multi-worker PSOCK parallel execution when workers > 1, splitting rows
  across chunks and combining results with preserved diagnostic attributes.
• Worker count validation enforces finite integer input and caps at the
  currently permitted worker count; no workload-size heuristic is applied
  beyond the row-count cap.

Fixed

• Batch worker count is now capped at the detected logical CPU limit to
  prevent oversubscription on multi-socket or hyperthreaded systems.