Skip to content

Hybrid Downscaling

Salvador Navas edited this page Jul 2, 2026 · 1 revision

Hybrid Downscaling

Package: pyhydra.climate.hybrid_downscaling

Hybrid downscaling in pyhydra connects statistical event generation with hydrological/hydraulic simulations and flood-map post-processing.

Event Classification

Main APIs:

  • HydrographClassifier
  • find_spatial_arrangement

These tools classify hydrographs and help organize event typologies. They are used in workflows where a large synthetic catalogue is reduced into a smaller set of representative hydrograph families or simulation classes.

Event Selection and Copula Comparison

Main APIs:

  • FloodEventSelector
  • FloodCopulaComparison

These classes support synthetic event selection and comparison of dependence structures. They are useful when a stochastic catalogue must be sampled into representative events for hydraulic simulation.

Hydrograph Reconstruction

Main APIs:

  • HydrographReconstructor
  • maxdiss

maxdiss implements a maximum-dissimilarity style selection routine. The reconstructor supports mapping selected event descriptors back to hydrograph shapes.

Flood-Map Interpolation

Main APIs:

  • FloodMapInterpolator
  • FloodMapInterpolatorCC

These classes interpolate flood maps from simulated examples. FloodMapInterpolatorCC extends the base workflow for climate-change style applications.

Return-Period Maps

Main APIs:

  • pixel_return_period
  • save_return_period_geotiffs
  • DEFAULT_RETURN_PERIODS

These functions derive pixel-wise return-period products from flood-map ensembles and write geospatial outputs using a template raster.

Limitations

Hybrid workflows are case-study dependent. They require consistent event features, hydrograph definitions, simulated flood maps and raster metadata. Before applying them to a new basin, check:

  • feature scaling and event definitions;
  • number and representativeness of hydraulic simulations;
  • raster alignment, nodata values and coordinate reference systems;
  • extrapolation risk beyond the simulated event envelope.

Clone this wiki locally