Precise Rainfall Forecasting is critical for sectors like agriculture, water management, and disaster readiness. Our researchers create cutting-edge models using advanced data analysis and machine learning. By refining accuracy and lead time in predictions, our aim is to improve decision-making and reduce the effects of severe weather occurrences.
Study 1
- Flood damage and loss of life increased due to urbanization, population growth, land use changes, and climate change over 20 years.
- Accurate flood prediction is crucial but current forecasts have biases needing downscaling solutions.
- Study used GEFR2 re-forecast and TIGGE data (NOAA, ECMWF), with TIGGE ECMWF showing better rainfall prediction skills in June-August.
- Statistical downscaling models like SDSM used to enhance rainfall forecast accuracy, especially at different thresholds.
- +24h forecast corrected with predictor relations, GEFR2 model performed better due to longer data availability.
- Four models calibrated to correct 0-100mm and 100-200mm rainfall ranges, improving extreme event detection for early flood alerts.
- Statistical downscaling models have potential to enhance rainfall forecasts for early flood warnings and mid-range river flood predictions in 0-100mm range.
Study 2
- Automated weather classification system analyzed daily weather conditions in Japan.
- Data from two sources used: mean sea-level pressure data from ECMWF Re-Analysis dataset and daily forecast data from TIGGE dataset.
- Identified 11 weather types: anticyclones, cyclones, hybrids, various wind directions.
- Main contributors to total rainfall: cyclones, hybrids, westerly, and northwest winds.
- Applying gamma-based bias correction improved global rainfall forecast accuracy by 10%.
- Specific weather type bias corrections reduced overall rainfall forecast error by up to 20%, with 5-10% reduction in root mean square error.
- Both global and weather type bias corrections improved extreme rainfall predictions, especially for intensities over 100 mm/d.
Study 3
- Online access to sub-seasonal forecasts sparks interest in extreme rainfall prediction and early warnings.
- Developing tropical countries, e.g., Sri Lanka, need effective early warning due to complex weather challenges.
- This study explores benefits of the Sub-seasonal to Seasonal (s2s) database via self-organizing map classification.
- Findings: Teleconnection indexes link to rainfall, heavy rainfall linked to cluster types, s2s forecast performance varies, introduces bias coefficients for basin water volume prediction.
- Study highlights s2s forecast value, calls for real-time data release for crucial early warnings in developing nations like Sri Lanka.