GitHub

Validation and Diagnostics

ClimateTools can produce corrected scenarios, regridded fields, and large catalogs of indices, but those products still need validation before they should be trusted in downstream analysis.

Validate Every Stage

In practice, it helps to validate at three stages:

  1. after dataset preparation and subsetting
  2. after regridding
  3. after bias correction

Regridding Checks

After regridding, confirm that:

  • the output grid is the one you intended
  • coastlines and masks look reasonable
  • missing-value behavior is acceptable
  • regional means or representative slices remain physically plausible

Useful quick checks:

  • map a single time slice before and after regridding
  • compare a regional average on the source and target grid

Bias-Correction Checks

After bias correction, compare the corrected product with the observational reference over the calibration period.

Useful diagnostics include:

  • mean bias
  • standard deviation
  • annual maxima and minima
  • wet-day frequency or dry-spell duration
  • representative grid-cell time series

Examples:

obs_txx = tx_max(obs)
qq_txx = tx_max(qq)

obs_wet = wetdays(obs_pr; thresh=1.0)
qq_wet = wetdays(qq_pr; thresh=1.0)

TVC-Specific Checks

When using tvc, inspect not only marginal behavior but also variability and persistence.

Recommended checks:

  • variance at the calibration scale
  • annual maxima or heat-wave-related indices
  • lag behavior or persistence-sensitive metrics at representative points

Missing-Value Diagnostics

Many climate-processing failures show up as unexpected NaN propagation.

Check:

  • whether the calibration domain contains missing values in key regions
  • whether regridding has increased masked areas
  • whether a bias-correction method returned all-NaN values for some cells because the calibration sample was insufficient

Validation Is Workflow-Dependent

The right diagnostic depends on the application.

  • For mean climate, annual means and standard deviations may be enough.
  • For extremes, focus on maxima, wet-day intensity, spell duration, and upper-tail behavior.
  • For persistence-sensitive studies, include TVC-style variability checks and duration indices.