Troubleshooting

This page collects the most common workflow issues encountered when using ClimateTools.

A Function Cannot Find the Time Dimension

Some functions expect a time axis, while others can also handle Ti.

What to check:

• inspect axes(cube)
• confirm that the dataset was opened as the intended variable
• make sure you did not drop the time axis during slicing

Observations and Simulations Have Different Calendars

Bias-correction workflows often need comparable calibration periods and compatible calendars.

What to check:

• whether one dataset contains leap days and another does not
• whether the calibration period truly overlaps
• whether you are comparing the same variable and units

ClimateTools bias-correction methods handle common leap-day cases, but the input series still need a defensible temporal alignment.

Regridder Fails on a Rotated Grid

If the source grid uses rlon and rlat dimensions or a grid_mapping attribute, do not treat it like a plain regular cube.

Use the dataset-aware path instead:

ds = open_dataset("rotated_model.nc")
regridder = Regridder(ds, :tasmax, target)

Bias Correction Returns Mostly NaN

Possible reasons:

• the calibration sample has too many missing values
• the grids were not aligned before correction
• the time overlap is insufficient
• the variable was passed with incompatible units or wrong semantics

Check the raw data coverage first before changing the correction parameters.

A Polygon Does Not Overlap the Grid

This often means the polygon and grid use different longitude conventions.

What to check:

• whether the grid uses 0 .. 360 but the polygon uses -180 .. 180
• whether the polygon coordinates are ordered correctly as lon-lat

Regridding Looks Wrong Near Coasts

Check:

• whether skipna and na_thres should be adjusted
• whether bilinear interpolation is appropriate for the variable
• whether the source field has a sharp land-sea mask that nearest-neighbor would preserve better

The Output Has the Right Shape but Unexpected Values

Sanity-check the workflow stage by stage:

1. inspect the raw source value at a representative grid point
2. inspect the regridded value at the same location
3. inspect the corrected value
4. compare annual or monthly summaries rather than only the full cube

Still Unsure?

Reduce the problem to a tiny regional subset and a short time period first. That usually reveals whether the issue comes from:

• data loading
• coordinate alignment
• regridding
• bias correction
• interpretation of the derived index