Water level bias and drift, against observations and against the model's own starting point

Reports two independent drift signals:

drift_m_per_yr - obs-based: slope of the (sim - obs) residual over time, at whatever observed dates exist. Needs real observations, and enough of them to trust a slope.
sim_drift_m_per_yr - model-only: slope of the raw simulated LKE_lvlwtr series itself (get_var with use_obs = FALSE, i.e. referenced to the hypsography's own zero, not any observation-derived elevation offset) against time, over every simulated day in the post-spin-up window. Available for every lake regardless of observation coverage, so a lake that is silently draining or filling under its own simulated water balance shows up here even with zero level observations to compare against.

Usage

diag_water_balance(
  aeme,
  model,
  drift_ok_m_per_yr = 0.05,
  bias_ok_m = NULL,
  min_n = 20
)

Arguments

aeme: Aeme object; already built and run for model (see build_aeme/run_aeme).
model: character; single model code to diagnose (see check_model).
drift_ok_m_per_yr: numeric; below this absolute slope, classified "static_offset" (or "ok") rather than "drift".
bias_ok_m: numeric or NULL; below this absolute mean bias (in metres), classified "ok" regardless of drift. When NULL (default), derived from the lake's maximum depth as 5% of max depth, clamped to \([0.1, 0.5]\) m, so the threshold scales sensibly across shallow and deep lakes without changing the absolute-unit semantics.
min_n: integer; below this many observations, the obs-based signal is not trusted (even if computed) and the model-only signal decides the classification instead.

Value

list with mean_bias, drift_m_per_yr, n, span_yr, sim_drift_m_per_yr, sim_change_m, sim_span_yr, classification (one of "ok", "static_offset", "drift", "ok_no_obs", "sim_drift", "no_data"), low_confidence, detail.