Changes in version 0.2.0                        

Data

  - New bundled dataset accra_rainfall: a real long-term daily rainfall
    and temperature record for Accra, Ghana (1981-01-01
    to 2026-07-20, 16637 days), retrieved from NASA POWER with
    sebkc::weather(). It replaces the simulated record as the manual's
    worked example, so the examples keep working even if the online
    service is unavailable. Load it and drop it onto a project:
    fp$rainfall <- accra_rainfall.

                        Changes in version 0.1.0                        

Pipeline stages

  - flood_extremes() fits the generalized extreme value distribution to
    annual maximum rainfall, both stationary and with a linear trend in
    location, and reports a likelihood-ratio test for changing extremes
    together with design return levels. A dependency-free
    maximum-likelihood engine is used by default; extRemes is used when
    installed and requested.

  - flood_scenario() turns design return levels into a present-day or
    climate-adjusted event by three methods: "delta" (change-factor
    scaling, the default), "trend" (projecting the fitted location trend
    forward) and "cmip6" (reserved for downscaled projections).

  - roughness() assigns Manning's roughness by three methods:
    "constant", "landcover" (lookup from a table, defaulting to
    floodflow_lc_roughness, and user-overridable) and "ndvi" (a
    monotonic empirical function of vegetation index). Works on vectors
    and, where terra is installed, on rasters.

  - flood_runoff() converts rainfall into a discharge series. Potential
    evapotranspiration is computed by the Oudin formula (pet_oudin())
    from temperature and latitude. GR4J via airGR is used when
    installed; otherwise a mass-conserving conceptual fallback with
    routing lag runs.

  - flood_route() provides a five-method routing ladder that turns
    discharge into a routed hydrograph and a water depth:
    "manning-normal", "kinematic", "diffusive", "muskingum-cunge"
    (default) and "dynamic". Depth comes from Manning's equation;
    hydrograph routing uses the numerically stable Muskingum-Cunge
    family, varying its diffusion across the ladder.

  - flood_hydraulics() derives the time-and-motion family from routed
    flow: peak velocity, time of concentration by the Kirpich, Kerby,
    combined Kerby-Kirpich and velocity methods (tc_kirpich(),
    tc_kerby()), channel travel time, and an event-relative
    time-to-peak.

  - flood_uncertainty() applies Generalized Likelihood Uncertainty
    Estimation (GLUE): it samples roughness and width from priors,
    weights each run by its agreement with an observed depth, and
    returns a predictive depth band and inverse parameter estimates.
    Equifinality among parameters is reported rather than hidden.

  - flood_vulnerability() combines hazard, exposure and vulnerability
    into a normalised risk index (Risk = Hazard x Exposure x
    Vulnerability), identifying hotspots where deep flooding meets
    dense, susceptible population.

  - flood_surrogate() trains a fast emulator (random forest via ranger,
    or a log-linear fallback) of the depth model for rapid what-if
    exploration.

  - flood_route() gains an optional hand argument: supply a Height Above
    Nearest Drainage surface (or a DEM) and it produces a spatial
    inundation-depth raster, depth_raster, which flood_map(layer =
    "depth") draws as a real inundation map.

  - flood_map() renders a chosen layer (depth, risk, velocity,
    uncertainty) as an interactive map via tmap or leaflet when
    available, and always returns a tidy value summary so the pipeline
    works without mapping engines.

Infrastructure

  - Added the flood_project object that carries data and results through
    the pipeline, with print() and summary() methods and an
    is_flood_project() predicate.
  - Established the CRAN-ready package skeleton: lean Imports (only
    stats and utils), all modelling engines under Suggests with graceful
    guards, MIT license, testthat (edition 3) test suite, and continuous
    integration.