Trends across years from per-year fingerprints #129

Activity

1. aivanchenk

   added

   enhancementNew feature or requestNew feature or request

   performance

   visualization

   on Sep 21, 2026
2. aivanchenk commented on Oct 22, 2026

   

   aivanchenk

   on Oct 22, 2026

   CollaboratorAuthor

   More actions

   @lazarusA I have a question on the forecast part in here. How extensive and deep should it be? So the forecast options that I was looking into are:
   ────────────────────────────────────────
   Option: Theil–Sen trend
   What it does: Straight-line trend for each month × hour cell, extended forward
   Good: Not thrown off by odd years like 2003 or 2018; simple to explain; works with 21 years
   Weak: Only straight lines; assumes the past trend continues
   ────────────────────────────────────────
   Option: Least-squares trend (I am familiar with it from university)
   What it does: Same straight line, ordinary fit
   Good: Simplest, standard error ranges
   Weak: One odd year can tilt the line
   ────────────────────────────────────────
   Option: ARIMA / Prophet
   What it does: Statistical model on the monthly series
   Good: Handles year-to-year carry-over
   Weak: Prophet needs a Python or R server; ranges widen fast, one of the most complex to implement
   ────────────────────────────────────────
   Option: Real model data (CMIP6 land-carbon output)
   What it does: Actual climate-model projections, run under different emission scenarios
   Good: The only option that's a real forecast
   ────────────────────────────────────────
   I would appreciate your feedback here. How you were doing your visualization forecasts before?