What are the challenges of using machine learning to predict environmental outcomes?

1 Data quality and availability

One of the main challenges of using ML to predict environmental outcomes is the quality and availability of data. Environmental data are often noisy, incomplete, inconsistent, or outdated, which can affect the accuracy and reliability of ML models. Moreover, environmental data are often scattered across different sources, formats, and scales, which can make it difficult to integrate and access them. To overcome these challenges, environmental engineers need to use appropriate data preprocessing techniques, such as cleaning, imputing, transforming, and aggregating data, to ensure that the ML models have sufficient and relevant data to learn from.

2 Model complexity and interpretability

Another challenge of using ML to predict environmental outcomes is the complexity and interpretability of the models. ML models can have many parameters, layers, or features, which can make them difficult to understand and explain. This can pose problems for environmental engineers who need to communicate their results to stakeholders, regulators, or policy makers, who may require transparency and accountability for the ML models. Moreover, complex ML models can also be prone to overfitting or underfitting, which can compromise their generalization and robustness. To overcome these challenges, environmental engineers need to use appropriate model selection, validation, and evaluation techniques, such as cross-validation, regularization, or sensitivity analysis, to ensure that the ML models are fit for purpose and can be justified.

3 Uncertainty and variability

A third challenge of using ML to predict environmental outcomes is the uncertainty and variability of the data and the models. Environmental data are often affected by natural or human-induced variability, such as seasonal, spatial, or temporal changes, which can influence the ML models' performance and predictions. Moreover, ML models are often based on assumptions, simplifications, or approximations, which can introduce uncertainty and error in the predictions. To overcome these challenges, environmental engineers need to use appropriate uncertainty quantification and propagation techniques, such as confidence intervals, error bars, or Monte Carlo simulations, to ensure that the ML models' predictions are reliable and robust.

5 Here’s what else to consider

This is a space to share examples, stories, or insights that don’t fit into any of the previous sections. What else would you like to add?