Skip to content

Relating climate and NDVI

Method guide · this is not an executed cross-source correlation result.

Question

Do vegetation departures covary with a climate variable over a chosen study period? This question requires comparable spatial support and temporal sampling; loading two cubes is not enough.

Data used

Choose a climate noun such as temperature or precipitation, and the vegetation-index noun. Record each source, CRS, units, acquisition dates, pixel support and quality decisions.

Grammar / pipeline

Prepare each field separately, then compare them only after explicitly aligning their observational support. There is no implemented v.correlation_cube verb; see its availability note.

Plain-language interpretation

A climate grid cell and a satellite pixel do not represent the same area. Likewise, a daily climate total and an irregular cloud-filtered acquisition do not represent the same time interval. Standardizing values does not resolve either mismatch.

Analysis

  1. Choose a target CRS and grid. Reproject/resample explicitly with a method appropriate to the quantity; align_cube is not a general-purpose reprojection engine.
  2. Define the temporal support (for example, an antecedent precipitation window for each valid satellite acquisition). Avoid inventing observations across cloud gaps.
  3. Apply quality masks, require enough paired observations and record valid counts. Check dimensions, coordinates and CRS before comparison.
  4. On already harmonized xarray inputs, xr.align(left, right, join="exact") can reject mismatched indexes; xr.corr(left, right, dim="time") computes Pearson correlation. Neither establishes scientific comparability for you.
  5. Plot correlation alongside paired counts, and assess sensitivity to window, aggregation, seasonality and autocorrelation.

Result

No cross-source result is presented on this page. Correlation is descriptive, not evidence of a causal vegetation response. A reproducible analysis must supply the alignment, quality control and paired-count diagnostics above.

Reproduce

Begin with the PRISM recipe and Sentinel-2 recipe. Both require live access. This method guide is not a turnkey notebook and is not certified by the offline vignette runner.

See also