Skip to contents

Changelog

Source: NEWS.md

nert 1.0.0

First stable release. The package has matured through the 0.0.x development line and is presented here with its complete user-facing surface — 11 exports covering 14 TERN datasets via a unified read_tern() dispatcher and the batch primitive collect_tern_data() — along with the audit-driven polish accumulated during pre-CRAN review.

Breaking changes

  • read_smips() argument day has been renamed to date, matching the unified argument naming used across all read_*() functions. Existing code calling read_smips(day = ...) will fail with an “unused argument” error; replace day = with date =. The rename was introduced when read_smips() was reintroduced as a thin wrapper over read_tern("SMIPS", ...) (PR #26) but was not surfaced in NEWS.md at the time; this 1.0.0 entry corrects the omission per #35.

New features

  • Retry behaviour is now configurable via package options nert.max_tries (default 3L) and nert.initial_delay (default 1L, seconds). Set globally with, e.g., options(nert.max_tries = 5L, nert.initial_delay = 2L) in your .Rprofile to apply across every read_*() and collect_tern_data() call. Closes #20.

Performance

  • collect_tern_data() is now vectorised across locations. An earlier implementation looped over (location, dataset, date) tuples and called read_tern() + terra::extract() once per single point, redownloading the same Cloud Optimised GeoTIFF (COG) once per location. The refactored version plans a list of work items (one item per unique COG), opens each COG once via /vsicurl/, and calls terra::extract() once with all coordinates passed as a single SpatVector. For M locations and a work-item budget K, the function now performs K reads and K extracts rather than M × K of each.

Bug fixes

  • collect_tern_data() now predeclares every output column at NA at planning time. A per-COG fetch failure leaves the affected column at NA without removing it from the result, so the column count and schema are invariant under network failure (non-SLGA static datasets no longer disappear from the result on a transient fetch error).
  • collect_tern_data() now always emits lon and lat columns, including for single-location calls (previously omitted when only one location was passed), standardising the output schema across call shapes.
  • .check_collection_agreement() (SMIPS date validator) now uses the documented 2015-11-20 lower bound (was 2005) and emits grammatical, dated error messages. The operator-precedence bug that caused non-totalbucket collections to bypass the lower-bound check is fixed.
  • read_asc(confusion_index = NULL) previously silently took the Confusion-Index branch via isFALSE(NULL) == FALSE. The argument is now strictly required to be a single non-NA logical.
  • .read_cog(max_tries = 0L) previously returned NULL silently because the loop body never executed; now errors with a clear message. Non-integer-coercible max_tries (e.g. "two") is also rejected.
  • .tern_dispatch_id() now rejects vector input rather than silently collapsing to the first element. The error message points users to collect_tern_data() for multi-dataset calls.
  • read_phenology() with a missing year now surfaces the documented 2003–2018 message rather than R’s generic missing-argument message (year defaults to NULL). The validator also rejects non-integer values (e.g. 2018.5) and vector years.
  • Two vignette runnability defects fixed: the agricultural vignette previously called read_smips(day = ...) (now date = ...) and contained an injected collect_tern_data() section whose output schema clobbered the chunk-to-chunk smips_data variable used by later analytics; replaced with a corrected, eval=FALSE showcase placed after the analytics. The intro vignette’s two remaining day = ... calls were also corrected.
  • Prose corrections in vignettes: Yield_tha → Yield_Tha (matches the data column name); Ecological → Ecosystem (TERN’s actual expansion); vaporisation → volatilisation (consistent with the earlier paragraph and chemistry); visualalising → visualising; SMindex scale described as “0–100” rather than “0–1” to match the COG.
  • grain Rd description corrected: Variety factor has 8 levels (matches levels(grain$Variety)), not 10.

Documentation

  • All portal.tern.org.au/metadata/TERN/<UUID> Rd references migrated to geonetwork.tern.org.au/... URLs; several portal URLs returned HTTP 404 as of 2026-05, while GeoNetwork resolves all UUIDs. The dead www.clw.csiro.au/aclep/... URL replaced with the active esoil.io/TERNLandscapes/... equivalent. urlchecker::url_check() is now clean.

Internal / API changes

  • The max_tries and initial_delay arguments of all public read functions (read_tern(), read_aet(), read_asc(), read_canopy_height(), read_phenology(), read_slga(), read_smips(), read_soil_diversity()) now default to NULL. When NULL, values are resolved at call time from getOption("nert.max_tries") and getOption("nert.initial_delay"). This is non-breaking: existing user code that passes explicit integers (e.g., read_tern("CANOPY", max_tries = 5L)) continues to work unchanged and still overrides the option.

  • New internal helper .init_nert_options() (in R/zzz.R) populates the package options on .onLoad(). Pattern adapted from the rOpenSci read.abares package.

  • sf removed from Imports (zero references in package code; legacy scaffolding from a never-delivered helper).

  • nlme demoted from Imports to Suggests; it is used only by the agricultural vignette, which already declares %\VignetteDepends{nlme}. The @import nlme directive that pulled it into the namespace without using any of it was removed.

  • tests/testthat/test-validation.R and a substantially expanded test-collect_tern_data.R cover the validation paths and the work-item planner (37 + 9 deterministic offline tests).

  • Test coverage uplift. Eight new/expanded offline test files (test-url-snapshots.R, test-read_aet.R, test-read_slga.R, test-read_soil_diversity.R, test-read_canopy_height.R, test-read_phenology.R, test-collect_tern_data_mocked.R, test-read_asc.R) lift package coverage from ~58% to ~83%, with every read_*() reader at 100%. Mocking is done at the internal .read_cog() binding via testthat::local_mocked_bindings(); test-url-snapshots.R pins every URL template via expect_snapshot().

  • Bumped testthat (>= 3.0.0) -> testthat (>= 3.2.0) in Suggests for local_mocked_bindings() support.

nert 0.0.4

Bug fixes

  • AET URL prefix corrected from /landscapes/aet/v2_2/ to /model-derived/aet/v2_2/ in .make_aet_url(). TERN’s bucket reorganisation moved CMRSET v2.2 under the model-derived classification (CMRSET is a CSIRO-modelled output, not a satellite landscape product); the legacy path returns 301 → 401 because GDAL /vsicurl/ does not propagate the apikey:KEY@host userinfo across the redirect, surfacing in R as [rast] file does not exist. Verified against the live TERN COG listing.

  • read_slga() now applies toupper() to the attribute argument before rlang::arg_match() (rather than inside it), fixing the ! 'arg' must be a symbol, not a call. error reported when running the help example.

  • .TERN_ALIASES casing bug in .tern_dispatch_id() resolved (incomplete rename surfaced during PR #30; live name is .tern_aliases).

  • collect_tern_data() now correctly expands depth = "all" for the eight SLGA datasets (AWC, CLY, SND, SLT, BDW, PHC, PHW, NTO) into the six GlobalSoilMap depth intervals, producing one column per (dataset, depth) named e.g. AWC_000_005. Previously the literal string "all" was passed straight through to .read_tern_slga() which rejected it via rlang::arg_match(); the resulting error was silently swallowed and the eight SLGA datasets disappeared from the result. Returning all 14 datasets at all SLGA depths now yields 60 columns (1 date + 6 SMIPS variants + ASC + AET + 8 SLGA x 6 depths + SOILDIV + CANOPY + PHENOLOGY).

  • collect_tern_data() now surfaces fetch errors via cli::cli_warn() instead of silently dropping the failed dataset’s column. Failed cells contain NA; the warning identifies the dataset, the date (where applicable), and the underlying error message. Pairs with the SLGA fix above to ensure the column count of the result matches the verbose datasets-table announcement.

Internal changes

  • Removed unused .create_sf() helper and its tests. Originally added as scaffolding for an extract_aet() function that was never delivered.

  • Removed unused .check_not_example_api_key() helper and its test.

  • withr added to Suggests for test fixtures.

  • tests/testthat/setup.R loads TERN_API_KEY from ~/.Renviron so the network-dependent tests run during R CMD check when the user has the key configured locally. CI continues to skip the network tests when no repository secret is configured — intentional behaviour.

  • roxyglobals and moodymudskipper/devtag integrated for @autoglobal and @dev roxygen tag support; pkgdown reference index now builds without internal-helper warnings.

  • Authors@R: Max Moldovan ORCID added.

nert 0.0.3.9000

New features

  • Addition of collect_tern_data() for batch extraction of TERN datasets at one or more locations over a date range. Returns a data.table with one row per date per location. Supports all 14 datasets, multi-location coordinate input (lon/lat vectors or data.frame), and column expansion for multi-layer products (SLGA depths, SMIPS variants).

  • Addition of collect_tern_data vignette with worked examples.

nert 0.0.2.9000

New features

nert 0.0.1.9000

Minor changes

Bug fixes

  • Spelling corrections