# Employing artificial intelligence to predict δ¹⁸O and δ²H isotope ratios in precipitation in Iraq under changing climate patterns

## TL;DR

This study uses machine learning to predict isotope ratios in Iraqi precipitation, helping manage water resources under climate change.

## Contribution

The study introduces a novel application of random forest machine learning for predicting isotopic signatures in arid regions.

## Key findings

- The random forest model achieved an R² of 0.89, showing strong predictive performance for isotope ratios.
- The model had the lowest MAE (1.39) and RMSE (3.5), indicating high accuracy in predictions.
- AI-based models show potential for reconstructing historical isotopic data and supporting climate assessments.

## Abstract

Understanding precipitation dynamics in arid regions such as Iraq is of paramount importance in hydrological and climatological studies, as it is a key approach to water resources management and climate change adaptation. This study aims to develop a mathematical predictive model for rainfall isotopic values using machine learning techniques. Stable isotope data for oxygen (δ¹⁸O) and deuterium (δ²H) in precipitation were collected from 32 meteorological stations distributed across Iraq over a 14-year period (2010–2024). The dataset also included meteorological parameters for these stations, including precipitation amount, air temperature, relative humidity, and calculated station elevation. Several machine learning algorithms (i.e., SVM, GBR, ANN, CatBoost, XGBoost, and RF) were employed to compare predicted isotopic values with actual readings, accounting for rainfall characteristics and patterns. The results demonstrated that the RF model achieved superior predictive performance, with a calibration coefficient (R²) of 0.89 in the testing set, indicating strong predictive capability. This model also recorded the lowest mean absolute error (MAE) of 1.39 and the lowest root mean square error (RMSE) of 3.5 compared to the other algorithms, reflecting improved predictive accuracy. These findings confirm the effectiveness of integrating machine learning, particularly the RF approach, in enhancing the modeling of isotopic signature predictions in environmental studies. Furthermore, they highlight the potential of AI-based models as powerful tools for reconstructing historical isotopic datasets, supporting climate variability assessment and sustainable water resources management in arid and semi-arid regions.