Dataset and code for: A Hybrid Simulation-Machine Learning Approach to Optimising the Operating Parameters of a Clay-based Emitter Across Soil Textures and Maximum Root Depths

This repository contains the numerical simulation dataset and Python source code associated with the study on Self-regulating, Low Energy, Clay-based Irrigation (SLECI) system performance, surrogate modeling, and multi-objective optimization. This dataset and codebase provide a reproducible pipeline for modeling complex, non-linear subsurface soil-water dynamics using machine learning (ML) surrogate models (Extra Trees, CatBoost, and LightGBM) and conducting multi-objective Pareto optimization across varying maximum root depths (Zr, max) Repository Contents 1. Simulation Dataset (/data) Physics-based Simulation Data: Numerical soil-water distribution data generated across various operating pressure heads (He), emitter installation depths (De), and maximum root depths (Zr, max). Target Hydraulic & Performance Variables: Includes cumulative emitter discharge (VQe), horizontal wetting front distance (dx), volumetric soil moisture content (θv), water application efficiency (εa), soil water distribution uniformity (CUθv), and the ratio of effective infiltrated volume (rv). 2. Python Scripts (/code) Scripts to train and evaluate ML surrogate models (Extra Trees, CatBoost, LightGBM) for hydraulic variables and performance indicators. Code for executing simultaneous maximising (εa, CUθv, rv) via an NGSA-II multi-objective optimisation to generate 3D Pareto frontiers and identify ideal compromise (knee) points across root depths. Code for performing omnibus ANOVA and post-hoc Tukey HSD pairwise comparisons on Pareto-optimal parameter sets. System Requirements & Dependencies The scripts are written in Python 3.13.9. The required libraries include: numpy pandas scikit-learn catboost lightgbm scipy matplotlib / seaborn How to Use Clone or extract the repository contents. Place the simulation dataset inside the designated /data directory. Run the model training script to reproduce surrogate model results, or execute the optimization script to evaluate Pareto trade-offs across root depths.