Scholarlink

<p>Accurate prediction of air humidity is crucial for climate sensitive sectors, including agriculture, water resource management, and urban planning. This study focuses on modeling air humidity as the primary target variable using Artificial Neural Networks (ANN), with air temperature and rainfall serving as the main predictive inputs, based on monthly meteorological data from four stations in Banten Province, Indonesia, covering the period 2016 to 2024. Data preprocessing involved missing value treatment, outlier inspection, and min max normalization. A single multi-output ANN model was developed using pooled data from all stations, employing a feed-forward multilayer perceptron architecture with one hidden layer and the Rectified Linear Unit activation function. To provide comparative insight into model behavior, additional experiments were conducted for rainfall and air temperature prediction; however, these outputs are treated as complementary analyses rather than the central objective of the study. Model performance was evaluated using Mean Absolute Error, Root Mean Square Error, and the coefficient of determination, supported by scatter and residual analyses. The results indicate that air humidity exhibits moderate and stable predictability, with an R squared value of 0.66, reflecting its strong temporal continuity and physical dependence on temperature and rainfall dynamics. In contrast, air temperature achieved higher accuracy, while rainfall prediction showed substantially lower reliability due to its episodic and extreme characteristics. Feature importance analysis reveals that historical humidity values dominate prediction performance, highlighting strong temporal autocorrelation, while temperature and rainfall contribute as physically meaningful secondary inputs. These findings confirm that ANN effectively captures nonlinear relationships governing atmospheric moisture, although limitations related to spatial heterogeneity and extreme precipitation events remain. This study demonstrates the suitability of ANN for humidity modeling in tropical regions and provides a methodological reference for localized climate analysis, while suggesting that future improvements may be achieved through extended datasets, additional meteorological predictors, and hybrid modeling approaches.</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

Modeling air humidity using an artificial neural network based on temperature and rainfall

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月9日

查看全文
  • Dwi Atmanto,
  • Agung Purwanto,
  • Arita Marini,
  • Asep Marfu,
  • Suwaib Amiruddin,
  • Heri Sapari Kahpi

摘要

Accurate prediction of air humidity is crucial for climate sensitive sectors, including agriculture, water resource management, and urban planning. This study focuses on modeling air humidity as the primary target variable using Artificial Neural Networks (ANN), with air temperature and rainfall serving as the main predictive inputs, based on monthly meteorological data from four stations in Banten Province, Indonesia, covering the period 2016 to 2024. Data preprocessing involved missing value treatment, outlier inspection, and min max normalization. A single multi-output ANN model was developed using pooled data from all stations, employing a feed-forward multilayer perceptron architecture with one hidden layer and the Rectified Linear Unit activation function. To provide comparative insight into model behavior, additional experiments were conducted for rainfall and air temperature prediction; however, these outputs are treated as complementary analyses rather than the central objective of the study. Model performance was evaluated using Mean Absolute Error, Root Mean Square Error, and the coefficient of determination, supported by scatter and residual analyses. The results indicate that air humidity exhibits moderate and stable predictability, with an R squared value of 0.66, reflecting its strong temporal continuity and physical dependence on temperature and rainfall dynamics. In contrast, air temperature achieved higher accuracy, while rainfall prediction showed substantially lower reliability due to its episodic and extreme characteristics. Feature importance analysis reveals that historical humidity values dominate prediction performance, highlighting strong temporal autocorrelation, while temperature and rainfall contribute as physically meaningful secondary inputs. These findings confirm that ANN effectively captures nonlinear relationships governing atmospheric moisture, although limitations related to spatial heterogeneity and extreme precipitation events remain. This study demonstrates the suitability of ANN for humidity modeling in tropical regions and provides a methodological reference for localized climate analysis, while suggesting that future improvements may be achieved through extended datasets, additional meteorological predictors, and hybrid modeling approaches.