Skip to content

Numerical verification

Verification checks the implementation against its equations, published results, and independent calculations. Validation evaluates model predictions against observations from a real application.

Reference comparison notebooks

  1. Baum and Srivastava–Yeh equation sets: compares pressure and moisture conversions, water-table rise, stability, and early-time response under identical inputs.
  2. Gardner retention-curve comparison: compares the constitutive relation with approximate points from Baum et al. (2010), Figure 4, alongside Figure 9 of Baum et al. (2008).
  3. Space-time profile verification: compares Qva and Qtb pressure-head and factor-of-safety profiles with Baum et al. (2010), Figure 10, including numerical checkpoints.
  4. Srivastava–Yeh homogeneous soils: reproduces wetting and drainage for two soils (Figures 1–4) and their basal outflow (Figure 5), with an independent finite-difference comparison.

Each notebook contains its inputs, computed figures, reference comparisons, and interpretation. Readings from published figures are approximate; the Srivastava–Yeh numerical errors are measured against the independent solver. The Qva/Qtb cases use baum_etal; the homogeneous-soil cases use srivastava_yeh. See the equation comparison.

Automated checks

Tests cover analytical profiles and basal flux, rainfall conversions and infiltration limits, antecedent histories, storage integration, column geometry, material mapping, and agreement between spatial and individual-column runs.

Basal-flux tests also use reference outputs from the TRIGRS v2.1.0 Fortran routines under controlled forcing and early-time settings. Reproduction scripts and source checksums are in research/trigrs-reference/; the independent Srivastava–Yeh solver is defined directly in the homogeneous-soil notebook.

Antecedent water-balance checks

Diagnostics compare effective rainfall with allocated drainage and reconstructed profile-water change. The residual reports numerical and model approximations; exact conservation is not imposed. Basal storage is represented in the reconstructed profile and must not be added as a separate water store. Storage above the fillable pore volume is flagged separately from rainfall rejected by the infiltration cap. See diagnostic fields and water-table assumptions.

Scope

These checks support the selected equations and reference cases, rather than field predictive skill or equivalence with the complete TRIGRS model. The Cerro Bravo application demonstrates deterministic scenarios using real slope-unit geometry; it does not compare predictions with an observed landslide inventory. Full-model validation requires such observational evidence.