Introduction

HydPy is an interactive framework for developing and applying different types of hydrological models, originally developed at the Ruhr-University Bochum for specific research purposes. Later it was extended on behalf of the German Federal Institute of Hydrology to be applicable in practical applications like runoff forecasting in large river basins. Now it is being maintained by Björnsen Consulting Engineers.

HydPy is intended to be a modern open source software, based on the programming language Python, commonly used in many scientific fields. Through using different well-established Python libraries and design principles, we target high quality and transparency standards. To avoid writing model cores (like lland_v1) in a more native programming language, HydPy includes a Cython based mechanism to automatically translate Python code to C code and to compile it.

HydPy has no graphical user interface. Instead, it is thought to be applied by executing Python scripts. These scripts help to increase the reproducibility of studies performed with HydPy because they can be easily shared and repeatedly executed. This approach facilitates discussing possible weaknesses of HydPy and its implemented models and comparing different methodical approaches (e. g. different strategies to calibrate model parameters). However, if you are not an experienced hydrologist with basic programming skills, you may need some help to become acquainted with HydPy.

We host HydPy in a GitHub repository and everyone is allowed to download, modify, and use it. However, when passing the (possibly modified) code to third parties, one has to be aware that the selected GNU Lesser General Public License 3 cannot be changed to a “less open source” license. If you, for example, implement a new model into HydPy, you can be sure that all possible further developments of your model code are still open source and the mentioned third parties are allowed to pass this modified source code to you.

HydPy offers many functionalities to make the implemented models as transparent and reliable as possible. For this reason, the online documentation is automatically updated for each new HydPy version and includes different documentation test mechanism ensuring that HydPy is working as expected and that the documentation is up-to-date with it.

See for example the documentation of the (very simple) method Calc_NKor_V1. The text describes what the method does and what input data it requires. It is comprehensive but, as in common documentations, technical reports and scientific articles, could be outdated or be wrong in other ways. This is not the case for the example calculation shown in the green box. This example is actual Python code that shows how method Calc_NKor_V1 can be used and how different input values (for variables Nied and KG) result in different output values (for variable NKor). Each time a new HydPy version is pushed into the GitHub repository, automatic test routines on Travis CI and AppVeyor are trigged. The new HydPy version is rejected, if the actual Calc_NKor_V1 method does not result in the exact same output values as given in the last line of the example.

Such basic “unit tests” should provide a good basis for discussing the proper implementation of certain hydrological processes. But they are no proof a complete model is actually working well. Therefore HydPy also offers some “integration test” functionalities.

Each integration test should demonstrate how a certain model could be set up meaningfully. Ideally, the model configuration should be varied to show different aspects of its functionality. See e.g. example 13 of the documentation on model dam_v001, which discusses the implemented flood retention routine. Here, example calculations are performed for a period of 20 days, and for each day all input and output values, as well as all internal states (e.g. the WaterVolume), are tabulated. Again, Travis CI checks that all of these values are exactly recalculated by each new HydPy version. Additionally, the tabulated values are shown in a Bokeh plot, which is also updated for each new HydPy version automatically. You can click on the variables and zoom into some details you are actually interested in.

If there were some methodical or technical flaws in the retention routine of dam_v001, you would have good chances to find them when reading the documentation critically. You could tell us about your finding via a GitHub issue, allowing us or others to read (and at best solve) the problem. Or you could try to solve it on your own and offer your solution as a Pull Request. You could also add a new test to the documentation files to prove that something goes wrong and offer it via a Pull Request, which would enable Travis CI to reject future HydPy versions that still contain this flaw.

We hope to have made clear that the design of HydPy focusses on open collaboration in order to improve existing and to develop better models. The Development section offers more information on how to actually participate in the further development of HydPy. Section Model Collection lists all models implemented so far. Sections Core Tools covers the basic functionalities of the HydPy framework.