5MCPModelNet HydroTuring

This repository contains five MCP-based hydrologic models used in the HydroTuring experiments.

The models were developed across three studies by Wang and Gupta (2024, 2025).
For all three studies, the Leaf River catchment was used as the primary experimental test bed for model development, evaluation, and interpretation.

---

MCP Models

The five models included in this repository are:

1. MCPModel1: $MC\lbrace O_{\sigma}L_{\sigma}^{con}\rbrace$
2. MCPModel2: $MC\lbrace O_{\sigma}L_{\sigma}^{con}M_{\sigma r}^{R}\rbrace$
3. MCPModel3: $MA_{5}$
4. MCPModel4: $MN_{SALO}^{DS}(5)$
5. MCPModel5: $MN_{SALO}^{DI}(3)$

These five models represent progressively different MCP-based representations of catchment-scale rainfall-runoff dynamics developed in our previous studies.

---

Model–Paper Mapping

Model	Model notation	Original study
MCPModel1	$MC\lbrace O_{\sigma}L_{\sigma}^{con}\rbrace$	Wang & Gupta (2024a)
MCPModel2	$MC\lbrace O_{\sigma}L_{\sigma}^{con}M_{\sigma r}^{R}\rbrace$	Wang & Gupta (2024a)
MCPModel3	$MA_{5}$	Wang & Gupta (2024b)
MCPModel4	$MN_{SALO}^{DS}(5)$	Wang & Gupta (2025)
MCPModel5	$MN_{SALO}^{DI}(3)$	Wang & Gupta (2025)

MCPModel1–2

These models were introduced in:

Wang, Y.H. and Gupta, H.V. (2024a).
A mass-conserving-perceptron for machine-learning-based modeling of geoscientific systems. Water Resources Research, 60(4), e2023WR036461.

This study introduced the Mass-Conserving Perceptron (MCP) as a differentiable, mass-conserving architectural unit for modeling dynamical geoscientific systems.

MCPModel3

This model was developed in:

Wang, Y.H. and Gupta, H.V. (2024b).
Towards interpretable physical-conceptual catchment-scale hydrological modeling using the mass-conserving-perceptron. Water Resources Research, 60(10), e2024WR037224.

This study explored how MCP-based architectures can be assembled to construct interpretable physical-conceptual hydrologic models.

MCPModel4–5

These models were developed in:

Wang, Y.H. and Gupta, H.V. (2025).
Using machine learning to discover parsimonious and physically-interpretable representations of catchment-scale rainfall-runoff dynamics. Water Resources Research, 61(12), e2025WR040178.

This study used MCP-based model networks to investigate parsimonious and physically interpretable representations of catchment-scale rainfall-runoff dynamics.

---

Experimental Catchment

The Leaf River catchment serves as the primary experimental catchment underlying the development and evaluation of the five model structures included in this repository.

The HydroTuring experiments use these previously developed MCP architectures as candidate hydrologic model structures for systematic comparison, evaluation, and interpretation.

---

Environment Setup

The code was tested using Python 3.10.

A Conda environment file and a pip requirements file are provided in this repository:

• environment.yml
• requirements.txt

Option 1: Conda

Create the environment using:

conda env create -f environment.yml

Activate the environment:

conda activate hydroturing

Option 2: pip

Alternatively, install the required Python packages using:

pip install -r requirements.txt

The main dependencies include:

• PyTorch
• NumPy
• Pandas
• scikit-learn
• torchvision
• tqdm

---

Repository Structure

A simplified structure of the repository is:

5MCPmodelnet_HydroTuring/
│
├── README.md
├── environment.yml
├── requirements.txt
├── .gitignore
│
├── MCPModel1/
├── MCPModel2/
├── MCPModel3/
├── MCPModel4/
├── MCPModel5/
│
├── MCPBRNN_lib_tools/
└── data/

The exact directory structure may vary slightly among model implementations because the five architectures were developed across different stages of the MCP studies.

---

References

1. Wang, Y.H. and Gupta, H.V. (2024a).
   A mass-conserving-perceptron for machine-learning-based modeling of geoscientific systems. Water Resources Research, 60(4), e2023WR036461.

2. Wang, Y.H. and Gupta, H.V. (2024b).
   Towards interpretable physical-conceptual catchment-scale hydrological modeling using the mass-conserving-perceptron. Water Resources Research, 60(10), e2024WR037224.

3. Wang, Y.H. and Gupta, H.V. (2025).
   Using machine learning to discover parsimonious and physically-interpretable representations of catchment-scale rainfall-runoff dynamics. Water Resources Research, 61(12), e2025WR040178.

License

Code and pretrained model weights in this repository are released under the Apache License 2.0.