A review of machine learning applications in life cycle assessment studies

Life Cycle Assessment (LCA) is a foundational method for quantitative assessment of sustainability. Increasing data availability and rapid development of machine learning (ML) approaches offer new opportunities to advance LCA. Here, we review current progress and knowledge gaps in applying ML techniques to support LCA, and identify future research directions for LCAs to better harness the power of ML. This review analyzes forty studies reporting quantitative assessment with a combination of LCA and ML methods. We found that ML approaches have been used for generating life cycle inventories, computing characterization factors, estimating life cycle impacts, and supporting life cycle interpretation. Most of the reviewed studies employed a single ML method, with artificial neural networks (ANNs) as the most frequently applied approach. Both supervised and unsupervised ML techniques were used in LCA studies. For studies using supervised ML, training datasets were derived from diverse sources, such as literature, lab experiments, existing databases, and model simulations. Over 70 % of these reviewed studies trained ML models with less than 1500 sample datasets. Although these reviewed studies showed that ML approaches help improve prediction accuracy, pattern discovery and computational efficiency, multiple areas deserve further research. First, continuous data collection and compilation is needed to support more reliable ML and LCA modeling. Second, future studies should report sufficient details regarding the selection criteria for ML models and present model uncertainty analysis. Third, incorporating deep learning models into LCA holds promise to further improve life cycle inventory and impact assessment. Finally, the complexity of current environmental challenges calls for interdisciplinary collaborative research to achieve deep integration of ML into LCA to support sustainable development.

Comparing Machine Learning Approaches for Predicting Spatially Explicit Life Cycle Global Warming and Eutrophication Impacts from Corn Production

Agriculture ranks as one of the top contributors to global warming and nutrient pollution. Quantifying life cycle environmental impacts from agricultural production serves as a scientific foundation for forming effective remediation strategies. However, methods capable of accurately and efficiently calculating spatially explicit life cycle global warming (GW) and eutrophication (EU) impacts at the county scale over a geographic region are lacking. The objective of this study was to determine the most efficient and accurate model for estimating spatially explicit life cycle GW and EU impacts at the county scale, with corn production in the U.S.’s Midwest region as a case study. This study compared the predictive accuracies and efficiencies of five distinct supervised machine learning (ML) algorithms, testing various sample sizes and feature selections. The results indicated that the gradient boosting regression tree model built with approximately 4000 records of monthly weather features yielded the highest predictive accuracy with cross-validation (CV) values of 0.8 for the life cycle GW impacts. The gradient boosting regression tree model built with nearly 6000 records of monthly weather features showed the highest predictive accuracy with CV values of 0.87 for the life cycle EU impacts based on all modeling scenarios. Moreover, predictive accuracy was improved at the cost of simulation time. The gradient boosting regression tree model required the longest training time. ML algorithms demonstrated to be one million times faster than the traditional process-based model with high predictive accuracy. This indicates that ML can serve as an alternative surrogate of process-based models to estimate life-cycle environmental impacts, capturing large geographic areas and timeframes.

Estimating Missing Unit Process Data in Life Cycle Assessment Using a Similarity-Based Approach

In life cycle assessment (LCA), collecting unit process data from the empirical sources (i.e., meter readings, operation logs/journals) is often costly and time-consuming. We propose a new computational approach to estimate missing unit process data solely relying on limited known data based on a similarity-based link prediction method. The intuition is that similar processes in a unit process network tend to have similar material/energy inputs and waste/emission outputs. We use the ecoinvent 3.1 unit process data sets to test our method in four steps: (1) dividing the data sets into a training set and a test set; (2) randomly removing certain numbers of data in the test set indicated as missing; (3) using similarity-weighted means of various numbers of most similar processes in the training set to estimate the missing data in the test set; and (4) comparing estimated data with the original values to determine the performance of the estimation. The results show that missing data can be accurately estimated when less than 5% data are missing in one process. The estimation performance decreases as the percentage of missing data increases. This study provides a new approach to compile unit process data and demonstrates a promising potential of using computational approaches for LCA data compilation.

Machine learning for toxicity characterization of organic chemical emissions using USEtox database: Learning the structure of the input space

Toxicity characterization of chemical emissions in Life Cycle Assessment (LCA) is a complex task which usually proceeds via multimedia (fate, exposure and effect) models attached to models of dose-response relationships to assess the effects on target. Different models and approaches do exist, but all require a vast amount of data on the properties of the chemical compounds being assessed, which are hard to collect or hardly publicly available (especially for thousands of less common or newly developed chemicals), therefore hampering in practice the assessment in LCA. An example is USEtox, a consensual model for the characterization of human toxicity and freshwater ecotoxicity. This paper places itself in a line of research aiming at providing a methodology to reduce the number of input parameters necessary to run multimedia fate models, focusing in particular to the application of the USEtox toxicity model. By focusing on USEtox, in this paper two main goals are pursued: 1) performing an extensive exploratory analysis (using dimensionality reduction techniques) of the input space constituted by the substance-specific properties at the aim of detecting particular patterns in the data manifold and estimating the dimension of the subspace in which the data manifold actually lies; and 2) exploring the application of a set of linear models, based on partial least squares (PLS) regression, as well as a nonlinear model (general regression neural network--GRNN) in the seek for an automatic selection strategy of the most informative variables according to the modelled output (USEtox factor). After extensive analysis, the intrinsic dimension of the input manifold has been identified between three and four. The variables selected as most informative may vary according to the output modelled and the model used, but for the toxicity factors modelled in this paper the input variables selected as most informative are coherent with prior expectations based on scientific knowledge of toxicity factors modelling. Thus the outcomes of the analysis are promising for the future application of the approach to other portions of the model, affected by important data gaps, e.g., to the calculation of human health effect factors.

AutoLCA

With increasing public consciousness regarding sustainability, companies are ever more eager to introduce eco-friendly products and services. Assessing environmental footprints and designing sustainable products are challenging tasks since they require analysis of each component of a product through their life cycle. To achieve sustainable design of products, companies need to evaluate the environmental impact of their system, identify the major contributors to the footprint, and select the design alternative with the lowest environmental footprint. In this article, we formulate sustainable design as a series of clustering and classification problems, and propose a framework called AutoLCA that simplifies the effort of estimating the environmental footprint of a product bill of materials by more than an order of magnitude over current methods, which are mostly labor intensive. We apply AutoLCA to real data from a large computer manufacturer. We conduct a case study on bill of materials of four different products, perform a “hotspot” assessment analysis to identify major contributors to carbon footprint, and determine design alternatives that can reduce the carbon footprint from 1% to 36%.