Skip to content

Phenology/NDVI-driven ET factor for temperature–phenology decoupling #26

Description

@awickert

Motivation

ThornthwaiteChang2019 derives ET from temperature alone, so it cannot represent seasonality when the actual ET cycle is out of phase with temperature — most clearly in cold-region forests, where canopy leaf-out lags spring warming by weeks. Thornthwaite ramps ET up with the spring temperature rise before transpiration actually begins, over-estimating early-spring ET; in snowmelt basins this consumes the spring freshet, and the annual water-balance scaling redistributes the bias into other seasons. (Documented as a warning under evapotranspiration_method in configuration.rst; origin: notes/HANDOFF_thornthwaite_ET_phenology.md. On the Crow Wing a prescribed leaf-out factor fixed the spring residual and made the melt factor identifiable.)

Proposal

An optional ET factor that follows observed vegetation green-up (e.g. an NDVI- or remote-sensing-based phenology signal) rather than temperature, so the seasonal ET shape tracks actual leaf-out instead of the temperature curve. This would remove the phasing bias structurally rather than papering over it with et_scale (an annual scalar that can only trade the bias between seasons).

Open design questions: prescribed monthly phenology shape vs. forced from an NDVI time series; how it composes with the existing water-balance correction and et_scale; basin-mean vs. land-cover-resolved.

Relationship to #22 (land-cover DDF prior) — kept separate deliberately

This and #22 both draw on remote-sensing / land-cover input, so it's tempting to fold them into one "forcing/priors from land cover" effort. We're keeping them as separate issues: the melt-factor prior (#22) and a phenology-driven ET formulation are distinct developments that will likely walk different paths (different signals, different parts of the model, different validation), and yoking them risks one blocking the other. Cross-referencing, not merging.

Status

Future work — flagged from the Crow Wing finding, not scheduled.

🤖 Generated with Claude Code

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    priors-from-dataDeriving model parameters/priors from external data (terrain, soil, land cover, remote sensing)

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions