Related Experiment Videos
Delta feature and random forest-enhanced LSTM-attention forecasts with probabilistic postprocessing for rainfall
Seyed Mohammad Miri1, Mohammad Reza Kavianpour2, Mohamad Javad Alizadeh3
1Faculty of Civil Engineering, K.N. Toosi University of Technology, Tehran, Iran. m.miri@email.kntu.ac.ir.
Scientific Reports
|May 27, 2026
Summary
Accurate rainfall forecasting is essential for flood control and water management. This study introduces a hybrid Attention-Long Short-Term Memory (Attention-LSTM) model that improves predictions by incorporating data from neighboring stations and temporal gradients, enhancing extreme rainfall forecasts.
Area of Science:
- Meteorology
- Data Science
- Environmental Science
Background:
- Accurate rainfall forecasting is critical for effective flood control and water resource management.
- Deep learning, high-performance computing, and IoT advancements are enhancing rainfall prediction capabilities.
- Existing models often struggle with spatial dynamics and extreme event underestimation.
Purpose of the Study:
- To develop an improved rainfall forecasting framework by integrating spatial and temporal data.
- To enhance the modeling of spatial propagation and regional rainfall dynamics using neighboring station data.
- To refine extreme rainfall value forecasting through a hybrid deep learning model with probabilistic postprocessing.
Main Methods:
- Leveraged data from a target station and seven neighbors, incorporating six meteorological variables.
- Introduced delta-based features for temporal gradients and time-lagged features for dependencies.
- Developed and evaluated a hybrid Attention-Long Short-Term Memory (Attention-LSTM) framework, including probabilistic postprocessing with a Weibull distribution.
Main Results:
- The hybrid Attention-LSTM framework achieved satisfactory predictions for general and extreme rainfall (correlation ≈ 0.69).
- Feature selection methods like PCA, correlation analysis, and Random Forest identified informative inputs.
- Probabilistic postprocessing refined initial underestimation of extreme rainfall events, though high-intensity convective events remained partially underestimated.
Conclusions:
- The proposed Attention-LSTM framework effectively models spatial and temporal rainfall dynamics for improved forecasting.
- Probabilistic postprocessing is crucial for refining extreme value predictions and addressing data imbalance.
- While general and extreme rainfall forecasts are satisfactory, further improvements are needed for high-intensity convective events.
Related Concept Videos
01:12Precipitation Processes
The experimental conditions in a gravimetric analysis should be optimized to maximize the particle size and purity of the obtained precipitate. Ideally, the concentration of the precipitating reagent should be low with effective stirring to maintain low relative supersaturation for the growth of large crystals. In homogeneous precipitation, the precipitant is slowly generated by a chemical reaction in the solution to avoid local reagent excesses. For example, urea decomposes gradually to...
01:17Precipitation and Co-precipitation
Precipitation and coprecipitation methods can be used to separate a mixture of ions in a solution. In qualitative inorganic analysis, ions that form sparingly soluble precipitates with the same reagent are separated based on the differences in solubility products. For example, consider the separation of Cu(II) and Fe(II) ions by precipitation as insoluble sulfides. First, copper(II) sulfide is precipitated by the addition of acidic H2S, where the dissociation of H2S is suppressed. Adding H2S...
03:10Precipitation Reactions
In a precipitation reaction, aqueous solutions of soluble salts react to give an insoluble ionic compound – the precipitate. The reaction occurs when oppositely charged ions in solution overcome their attraction for water and bind to each other, forming a precipitate that separates out from the solution. Since such reactions involve the exchange of ions between ionic compounds in aqueous solution, they are also referred to as double displacement, double replacement, exchange reactions, or...
01:19Precipitation Titration: Endpoint Detection Methods
In argentometric precipitation titrations, endpoints can be detected visually by the Mohr, Volhard, and Fajans methods. In the Mohr method, adding a soluble chromate indicator gives an initial yellow color to the analyte solution. As the titrant is added, the first excess of silver ions forms a red silver chromate precipitate, marking the endpoint. The solution pH should be maintained at about 8 by adding solid CaCO3.
In the Volhard method, a standard excess of AgNO3 is first added to the...
In the Volhard method, a standard excess of AgNO3 is first added to the...
01:03Precipitation Gravimetry
Precipitation gravimetry is based on converting an analyte into a sparingly soluble precipitate, which is separated by filtration and weighed. An ideal precipitate should be pure, insoluble, of known composition, and easily filtered from the reaction mixture.
In determining nickel by gravimetric analysis, a precipitant of ethanolic dimethylglyoxime is added to a hot nickel salt solution. This is quickly followed by the dropwise addition of dilute ammonia solution until precipitation occurs. A...
In determining nickel by gravimetric analysis, a precipitant of ethanolic dimethylglyoxime is added to a hot nickel salt solution. This is quickly followed by the dropwise addition of dilute ammonia solution until precipitation occurs. A...
01:03Prediction Intervals
The interval estimate of any variable is known as the prediction interval. It helps decide if a point estimate is dependable.
However, the point estimate is most likely not the exact value of the population parameter, but close to it. After calculating point estimates, we construct interval estimates, called confidence intervals or prediction intervals. This prediction interval comprises a range of values unlike the point estimate and is a better predictor of the observed sample value, y.
The...
However, the point estimate is most likely not the exact value of the population parameter, but close to it. After calculating point estimates, we construct interval estimates, called confidence intervals or prediction intervals. This prediction interval comprises a range of values unlike the point estimate and is a better predictor of the observed sample value, y.
The...