Improving Prediction of Marine Low Clouds Using Cloud Droplet Number Concentration in a Convolutional Neural Network DOI Creative Commons
Yang Cao, Yannian Zhu, Minghuai Wang

et al.

Journal of Geophysical Research Machine Learning and Computation, Journal Year: 2024, Volume and Issue: 1(4)

Published: Nov. 30, 2024

Abstract Marine low clouds play a crucial role in cooling the climate, but accurately predicting them remains challenging due to their highly non‐linear response various factors. Previous studies usually overlook effects of cloud droplet number concentration (N d ) and non‐local information target grids. To address these challenges, we introduce convolutional neural network model (CNN Met‐Nd that uses both local includes N as cloud‐controlling factor enhance predictive ability daily cover, albedo, radiative (CRE) for global marine clouds. CNN demonstrates superior performance, explaining over 70% variance three variables scenes 1° × 1°, notable improvement past efforts. also replicates geographical patterns trends from 2003 2022. In contrast, similar without Met struggles predict long‐term properties effectively. Permutation importance analysis further highlights critical Met‐N 's success. Further comparisons with an artificial (ANN model, which same inputs considering spatial dependence, show performance R 2 values CRE being 0.16, 0.12, 0.18 higher, respectively. This incorporating information, at least on scale, into predictions climate parameterizations.

Language: Английский

Integrating Advanced Machine Learning Models for Accurate Prediction of Porosity and Permeability in Fractured and Vuggy Carbonate Reservoirs: Insights from the Tarim Basin, Northwestern, China DOI
Armel Prosley Mabiala Mbouaki, Zhongxian Cai,

Allou Koffi Franck Kouassi

et al.

SPE Journal, Journal Year: 2025, Volume and Issue: unknown, P. 1 - 27

Published: April 1, 2025

Summary Accurate prediction of porosity and permeability in fractured vuggy carbonate reservoirs is crucial for optimizing hydrocarbon recovery but remains challenging due to their extreme heterogeneity anisotropy. Traditional methods often struggle capture the complex geological variability, leading suboptimal reservoir characterization. To address this, we propose a novel hybrid machine learning (ML) framework that integrates particle swarm optimization (PSO), mixed-effects random forest (MERF), ensemble models, such as light gradient boosting (LightGBM), (XGBoost), (RF). These models were trained validated using leave-one well-out cross-validation (LOO-CV) train-test split method, leveraging geophysical well-log data from Tarim Basin’s reservoirs. Among three PSO-MERF-LightGBM outperformed others, achieving an R² 0.9752 root mean square error (RMSE) 0.0606 R2 0.9983 RMSE 0.00473 during testing. Moreover, model demonstrates exceptional computational efficiency, completing processing just 11 seconds 9 seconds, respectively. This marks significant reduction computation time compared with other making it highly efficient alternative. results confirm its superior ability nonlinear relationships spatial variability. The study how advanced ML techniques can enhance characterization, improving decision-making subsurface resource management. Future research should extend this settings validate broader applicability.

Language: Английский

Citations

0
Improving Prediction of Marine Low Clouds Using Cloud Droplet Number Concentration in a Convolutional Neural Network DOI Creative Commons
Yang Cao, Yannian Zhu, Minghuai Wang

et al.

Journal of Geophysical Research Machine Learning and Computation, Journal Year: 2024, Volume and Issue: 1(4)

Published: Nov. 30, 2024

Abstract Marine low clouds play a crucial role in cooling the climate, but accurately predicting them remains challenging due to their highly non‐linear response various factors. Previous studies usually overlook effects of cloud droplet number concentration (N d ) and non‐local information target grids. To address these challenges, we introduce convolutional neural network model (CNN Met‐Nd that uses both local includes N as cloud‐controlling factor enhance predictive ability daily cover, albedo, radiative (CRE) for global marine clouds. CNN demonstrates superior performance, explaining over 70% variance three variables scenes 1° × 1°, notable improvement past efforts. also replicates geographical patterns trends from 2003 2022. In contrast, similar without Met struggles predict long‐term properties effectively. Permutation importance analysis further highlights critical Met‐N 's success. Further comparisons with an artificial (ANN model, which same inputs considering spatial dependence, show performance R 2 values CRE being 0.16, 0.12, 0.18 higher, respectively. This incorporating information, at least on scale, into predictions climate parameterizations.

Language: Английский

Citations

0