Releases Notes

ASYNCH release notes provide information on the features and improvements in each release. This page includes release notes for major releases and minor (bugfix) releases. If you are upgrading from an earlier version of ASYNCH, you will find essential information in the Breaking Changes associated with the relevant release notes.

Version 1.7

Released 2026-09-28. A speed release: a new, optional numerical solver for stiff equations. The details are in the Changelog.

Breaking Changes

  • None. The default solver is unchanged; existing configurations give the same results as 1.6.

New Features

  • Numerical solver index 4 (Rodas5P, a Rosenbrock method) for stiff equations: several times faster than the default solver at the same accuracy, with more precise peaks. Use it with tolerances about 100 times smaller than those used with the default solver. It is not available for models solved with algebraic equations; such runs stop with a message. How to use it: in the global file, write 4 instead of 2 on the line after %Numerical solver index and divide the error tolerances by 100; see 2. Running the model, section 2.4, for a worked example (also from Python).

  • Runs start about one second sooner.

Version 1.6

Released 2026-09-27. A quality release: the units and equations of every built-in model were checked systematically, and the inconsistencies found were corrected. The details are in the Changelog.

Breaking Changes

  • Numerical results change for some built-in models, as a consequence of these corrections. The Changelog lists the models concerned and what changed for each. Model 254 gives the same results as 1.5.

  • Peak-flow files (.pea) of the oldest models write the area in km², like all other models.

New Features

  • pip install asynch-hlm installs the ready-made package from PyPI; its page shows the changelog.

  • New tests check the physical consistency of every model (for instance, that the water taken by evaporation matches the potential evaporation, and that every model runs from a completely dry state).

Version 1.5

Released 2026-09-26. Every change, with its effect on numerical results, is in the Changelog; the problems found and fixed are explained in 7. What was fixed, and why it matters.

Breaking Changes

  • Model 254 results change. The baseflow equation of 2015 is restored: a line added in 2021 kept the baseflow state q_b at 0 (S-02). Total discharge changes by at most 0.0003 m³/s in the larger example; the baseflow output was wrong since 2021. Apart from models 105 and 263 (below), all other models give the same results as 1.4.3.

  • Runs stop earlier, with a message, instead of failing silently: a missing or read-only output folder (checked before computing; exit code 1), a numerical solver index other than 0, 1 or 2, a model number without equations (200, 260, 300, 301, 315, 607, 2000), a link with more than 16 upstream links. An initial-state file with fewer values than the model has states gives a warning, and the missing values are 0 (they came from random memory).

  • Model 263 reads 16 values per link from its parameter file. In models 105 and 263, the states that have no equation keep their initial values (their results depended on leftover memory, B-26).

  • C API: Asynch_Get_Num_Links returns unsigned int (it returned unsigned short, wrong above 65 535 links). The library is also built as a shared library, libasynch.so.

  • Python: the old interface (py/, asynchdist.py) is replaced by the package asynch (python/).

New Features

  • A ready-made wheel for Linux (x86-64, glibc 2.31 or newer), attached to the release: pip install it and the Python package, the library and the asynch program are there, with nothing to compile; MPI comes with it (the mpich package of PyPI, with mpiexec).

  • The Python package asynch, installed with pip or make install-python: run a global file (identical output files), advance step by step, read and change states and parameters, add outputs, read and write every file format, and define new models with equations in C, Numba or Python (10. Using ASYNCH from Python, Python API reference). Works with MPI (mpirun -n 4 python3 script.py) and mpi4py.

  • asynch_api.h: a C interface for other languages and for custom models defined outside the source code.

  • make check runs 23 C unit tests, 73 Python tests and every example against the reference results of the original repository (9. Reproducibility and regression testing). GitHub Actions run it on every push.

  • Per-link solver settings (.rkd files) and solver methods 0 and 1 work (they did not, B-14 and B-13).

  • A Dockerfile for a ready-made environment, and this documentation website.

Version 1.4

Breaking Changes

The signature of the function that implements the differential equations has an extra parameter max_num_dim. This is required because data assimilation uses a variable number of equations at links, the maximum number of degree of freedom beeing known at runtime.

typedef void (DifferentialFunc) (
  double t,
  const double * const y_i, unsigned int num_dof,
  const double * const y_p, unsigned short num_parents, unsigned int max_num_dof,
  const double * const global_params,
  const double * const params,
  const double * const forcing_values,
  const QVSData * const qvs,
  int state,
  void *user,
  double *ans);

New Features

This release introduces Data Assimilation.

An additional, more compact HDF5 format for the time series is available, see option 6 of Time Series Location.

Three new models are available:

Constant Runoff Model 195:

  • based on 190 with precipitation forcing replaced by runoff and infiltration forcings to describe spatial variability of such processes, and including an additional state for rainfall accumulation.

Top Layer Model 256:

  • based on 254;

  • has a new state variable ans[7] : accumulated evapotranspiration;

  • has one more toplayer storage to link parameter k_tl = global_params[12];

  • has a flux component from toplayer storage to channel that represents the interflow \(q_{tl} = k_{tl} s_t\).

Top Layer Model 257:

  • based on 256 but channel velocity is spatialized as a function of the Horton order;

  • the Horton order is an additional local parameter.

Version 1.3

This release is the result of a loong run profiling and optimization work.

Memory footprint improvements, better data structure reduce the memory usage typically by a factor two. If you have memory available you may want to increase the number of buffers, see Buffer Sizes.

Performance improvements, ASYNCH runs about 30% faster.

Version 1.2

Breaking Changes

The global file structure has changed. The time of the simulation is now given in absolute time, see Simulation period.

The snapshots in HDF5 format has changed, see Ini HDF5 Files.

New Features

Time series outputs can be written in HDF5 format, see Time Series Location.