Seasonal Prediction of Summer Precipitation in the Middle and Lower Reaches of the Yangtze River Valley: Comparison of Machine Learning and Climate Model Predictions

He, Chentao; Wei, Jiangfeng; Song, Yuanyuan; Luo, Jianbin

doi:10.3390/w13223294

Cited by 8 publications

(3 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…AI methods for weather forecasting can be classified into machine learning-and deep learning-based. Furthermore, machine learning (ML) architectures, applied to weather forecasting, can be grouped into static and dynamic (recurrent) in terms of whether the prediction is generated sequentially [48]- [51], or not. Static ML architectures include clustering (K-means, PCA) [52], [53], artificial neural networks (ANN) [49], [54], [55], graph neural networks (GNN) [56], clustering + neural networks [57], [58], decision trees such as Gradient Boosting (XGBoost [59], AdaBoost [50], CatBoost [50], [60], [61]), and Random Forest [62], [63].…”

Section: Related Workmentioning

confidence: 99%

A Novel Transformer Network With Shifted Window Cross-Attention for Spatiotemporal Weather Forecasting

Bojesomo,

AlMarzouqi,

Liatsis

2024

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

Earth Observation is a growing research area that can capitalize on the powers of AI for short time forecasting, a now-casting scenario. In this work, we tackle the challenge of weather forecasting using a video transformer network. Vision transformer architectures have been explored in various applications, with major constraints being the computational complexity of attention and the data-hungry training. To address these issues, we propose the use of Video Swin-Transformer, coupled with a dedicated augmentation scheme. Moreover, we employ gradual spatial reduction on the encoder side and crossattention on the decoder. The proposed approach is tested on the Weather4Cast2021 weather forecasting challenge data, which requires the prediction of 8 hours ahead future frames (4 per hour) from an hourly weather product sequence. The dataset was normalized to 0-1 to facilitate the use of the evaluation metrics across different datasets. The model results in an MSE score of 0.4750 when provided with training data, and 0.4420 during transfer learning without using training data, respectively.

show abstract

Section: Related Workmentioning

confidence: 99%

A Novel Transformer Network With Shifted Window Cross-Attention for Spatiotemporal Weather Forecasting

Bojesomo,

AlMarzouqi,

Liatsis

2024

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

show abstract

“…Liu et al [16] conducted flood risk assessments for 2030 and 2050 using a new watershed-scale framework, exploring changes in future flood risks. He et al [17] used machine learning methods to predict floods in a certain region in 2020 with small errors.…”

Section: Introductionmentioning

confidence: 99%

Influence of Multi-Cross Structures on the Flood Discharge Capacity of Mountain Rivers in the Yellow River Basin

Song

Zhang

et al. 2023

Water

View full text Add to dashboard Cite

This study investigates the impact of cross structures on flood occurrences in mountainous rivers. The governing equations of open channel flow were formulated based on the Saint-Venant equations. The open channel was segmented, and a node equation was established at each section’s connection point. An overflow model of bridges and weir dams was also developed. The physical model of the open channel was simplified and modeled using actual building data and model calculation requirements. The study found that the primary impact of weirs and bridges on the open channel was the backwater effect on the water level. The influence of these structures on the water level in the Huang Stream urban section in the Yellow River Basin was assessed under various working conditions. The results showed that deleting the #1 weir could reduce the maximum backwater height by 1.14 m, and deleting the #2 weir could reduce it by 1.09 m. While reducing the weir height significantly decreased the backwater range and height, it did not enhance the river’s flood discharge capacity. The Huang Stream contains 17 bridges, 13 of which could potentially affect flood discharge. The eight flat slab bridges in the submerged outflow state had a significant impact on flood discharge, with a maximum water level change of 0.51 m. Conversely, the three single-hole flat slab bridges in the free outflow state downstream had a negligible impact on flood discharge. The study found that bridges had a greater influence on flood discharge capacity than weirs. This research provides valuable insights for the reconstruction of cross structures in mountainous rivers and for managing flood discharge capacity and flood control.

show abstract

“…The extension of MLR is stepwise regression. The subsets of meteorological factors that are most important for predicting rainfall are automatically chosen by this empirical technique [17,18].…”

Section: Introductionmentioning

confidence: 99%

A Comparison of Machine Learning Models for Predicting Rainfall in Urban Metropolitan Cities

Kumar,

Kedam,

Sharma

et al. 2023

Sustainability

View full text Add to dashboard Cite

Current research studies offer an investigation of machine learning methods used for forecasting rainfall in urban metropolitan cities. Time series data, distinguished by their temporal complexities, are exploited using a unique data segmentation approach, providing discrete training, validation, and testing sets. Two unique models are created: Model-1, which is based on daily data, and Model-2, which is based on weekly data. A variety of performance criteria are used to rigorously analyze these models. CatBoost, XGBoost, Lasso, Ridge, Linear Regression, and LGBM are among the algorithms under consideration. This research study provides insights into their predictive abilities, revealing significant trends across the training, validation, and testing phases. The results show that ensemble-based algorithms, particularly CatBoost and XGBoost, outperform in both models. CatBoost emerged as the model of choice throughout all assessment stages, including training, validation, and testing. The MAE was 0.00077, the RMSE was 0.0010, the RMSPE was 0.49, and the R2 was 0.99, confirming CatBoost’s unrivaled ability to identify deep temporal intricacies within daily rainfall patterns. Both models had an R2 of 0.99, indicating their remarkable ability to predict weekly rainfall trends. Significant results for XGBoost included an MAE of 0.02 and an RMSE of 0.10, indicating their ability to handle longer time intervals. The predictive performance of Lasso, Ridge, and Linear Regression varies. Scatter plots demonstrate the robustness of CatBoost and XGBoost by demonstrating their capacity to sustain consistently low prediction errors across the dataset. This study emphasizes the potential to transform urban meteorology and planning, improve decision-making through precise rainfall forecasts, and contribute to disaster preparedness measures.

show abstract

Seasonal Prediction of Summer Precipitation in the Middle and Lower Reaches of the Yangtze River Valley: Comparison of Machine Learning and Climate Model Predictions

Cited by 8 publications

References 34 publications

A Novel Transformer Network With Shifted Window Cross-Attention for Spatiotemporal Weather Forecasting

A Novel Transformer Network With Shifted Window Cross-Attention for Spatiotemporal Weather Forecasting

Influence of Multi-Cross Structures on the Flood Discharge Capacity of Mountain Rivers in the Yellow River Basin

A Comparison of Machine Learning Models for Predicting Rainfall in Urban Metropolitan Cities

Contact Info

Product

Resources

About