smash 1.1.0 Release Notes#

The smash 1.1.0 release introduces several new solvers and functionalities, while continuing efforts to improve usability, fix bugs, and enrich the documentation. The highlights are:

  • Hybrid Physics-AI hydrological solvers: spatially distributed neural-augmented algebraic and ODE models

    In addition to regionalization neural networks, the process-parameterization neural networks are integrated into smash for GR-like model structures to refine internal water fluxes, either within algebraic models solving analytical solutions of time-integrated ordinary differential equations (ODEs), or within state-space models that numerically solve the ODEs.

  • Migration of optimization methods to Python

    All optimization methods have been reimplemented in Python, replacing the previous Fortran-based versions. This change enables the use of various adaptive optimizers for all mappings and supports callback signatures for retrieving intermediate results during the calibration process.

  • Convolutional neural networks for regionalization

    Convolutional neural networks (CNNs) are now supported for the HDA-PR (Hybrid Data Assimilation-Parameters Regionalization) algorithm. The smash.factory.Net sub-package has been updated to accommodate a generic neural network configuration for regionalization.

  • Variants of GR-like model structures

    New variants of the GR-like model structures have been introduced, featuring alternative parameterizations of the conceptual reservoirs. These updates also include the incorporation of rainfall intensity and imperviousness.

  • Fix for active and non-active cells

    Issues related to missing physical descriptors on active cells and incorrect handling of boundary conditions outside the calibration domain have been resolved.

  • Expanded user guide

    The user guide has been significantly expanded, now offering step-by-step tutorials ranging from quick-start examples to advanced usage of smash.

  • Expanded contributor guide

    The Contributor Guide has been significantly expanded, now providing more detailed explanations of the smash development process.

  • Complete mathematical documentation

    A more complete mathematical description of all methods used in smash is now available in the Math / Num Documentation.

Contributors#

This release was made possible thanks to the contributions of:

Pull requests merged#