Scholarlink

<p>Sulfate scattering is an indicator of diverse aspects of atmospheric interactions and climate change dynamics. In this study, we attempted to use Machine Learning frameworks to sieve the vital factors that can be used in studying sulfate scattering using multiple variables that originate from the surface and that are already suspended in the atmosphere. We used 32 features to predict the sulfate scattering tendency using 19 machine learning models. Out of 19 models, we filtered the models Linear Regression, Extra Trees Regressor, Random Forest Regressor, and Extreme Gradient Boosting due to their relatively better evaluation metrics. We proposed a technique called the ML—Averaged Weighing Method for Sieving Important Predictors (MAWSIP) method which can sieve the vital indicators both across models and variables based on relative weights. We used MAE, MSE, RMSE, R2, RMSLE, and MAPE as evaluation metrics to filter appropriate models. We report the order of priority as TAS (0.40) &gt; BCSAOT500 (0.36) &gt; SST500 &amp; OCSAOT550 (0.27) &gt; ASAT_m (0.22) across the models that may influence sulfate scattering and LR (0.80) &gt; ET (0.12) &gt; RF (0.16) &gt; XGB (0.13) across the features. We presented the spatial distribution of multiple variables used in this work that may affect the sulfate scattering using GIS maps. We conclude that Total Aerosol Scattering (TAS) and Black Carbon Scattering are important variables to comprehend sulfate scattering on a global scale.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Machine learning based sieving of conducive factors of sulfate scattering from a pool of multiple variables of the atmospheric strata using MAWSIP

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年9月25日

查看全文
  • Jagadish Kumar Mogaraju

摘要

Sulfate scattering is an indicator of diverse aspects of atmospheric interactions and climate change dynamics. In this study, we attempted to use Machine Learning frameworks to sieve the vital factors that can be used in studying sulfate scattering using multiple variables that originate from the surface and that are already suspended in the atmosphere. We used 32 features to predict the sulfate scattering tendency using 19 machine learning models. Out of 19 models, we filtered the models Linear Regression, Extra Trees Regressor, Random Forest Regressor, and Extreme Gradient Boosting due to their relatively better evaluation metrics. We proposed a technique called the ML—Averaged Weighing Method for Sieving Important Predictors (MAWSIP) method which can sieve the vital indicators both across models and variables based on relative weights. We used MAE, MSE, RMSE, R2, RMSLE, and MAPE as evaluation metrics to filter appropriate models. We report the order of priority as TAS (0.40) > BCSAOT500 (0.36) > SST500 & OCSAOT550 (0.27) > ASAT_m (0.22) across the models that may influence sulfate scattering and LR (0.80) > ET (0.12) > RF (0.16) > XGB (0.13) across the features. We presented the spatial distribution of multiple variables used in this work that may affect the sulfate scattering using GIS maps. We conclude that Total Aerosol Scattering (TAS) and Black Carbon Scattering are important variables to comprehend sulfate scattering on a global scale.