580 California St., Suite 400
San Francisco, CA, 94104
Key research themes
1. How can Machine Learning models, particularly Artificial Neural Networks, improve regional and parameter-specific meteorological condition forecasting?
This research theme explores the application of advanced Machine Learning (ML) techniques, specifically Artificial Neural Networks (ANNs) and their variants, to enhance prediction accuracy of local and regional weather variables such as temperature, humidity, and precipitation. Given the complex, nonlinear nature of meteorological phenomena and the limitations of traditional numerical weather prediction (NWP) models in local-scale forecasting, these approaches focus on adapting ML models to regional data characteristics and specific weather parameters to deliver effective predictions in diverse climatic zones.
Key finding: This study demonstrates that a Multilayer Perceptron (MLP) ANN model trained specifically on regional data in Poland's moderate transition climate can effectively capture nonlinear interactions of climatic variables to... Read more
Key finding: Utilizing a comparative analysis of multiple data mining algorithms including Neural Networks, Random Forest, and Support Vector Machines on Indian weather data, this work finds that Artificial Neural Networks, particularly... Read more
Key finding: By applying a Radial Basis Function Neural Network (RBFNN) to one-year hourly humidity and temperature data in Warangal, India, this paper evidences that RBFNN outperforms other ML models in forecasting these parameters. It... Read more
Key finding: This review consolidates evidence from multiple studies confirming that Artificial Neural Networks, due to their adaptive, nonlinear modeling abilities, are highly suitable for weather forecasting tasks characterized by... Read more
Key finding: Introducing InFORM, a time series ARIMA-based intelligent forecasting pipeline for urban climate parameters like temperature, relative humidity, and visibility in Bangalore, India, this study reveals statistical approaches... Read more
2. What are the advantages and limitations of high-resolution numerical weather prediction models for local-scale forecasts in complex terrains, particularly in hazard early warning systems?
This theme investigates the use of dynamical, physics-based high-resolution Numerical Weather Prediction (NWP) models with spatial resolutions down to 1.25 km for hourly forecasting in regions with complex topography. These models are evaluated for their utility in hazard early warning systems (EWS), notably for rainfall-triggered landslides and floods impacting infrastructure such as pipelines. Focus is placed on the model’s spatial accuracy, ability to capture orographic and mesoscale effects, and performance in temperature, wind, and precipitation forecasts in mountainous or narrow-valley settings.
Key finding: The KALM-HD NWP model at 1.25 km resolution delivers reliable hourly forecasts for temperature and wind speed in the Val d’Agri basin, outperforming coarser 5 km grid results. However, challenges remain in accurately... Read more
Key finding: Verification of the NCEP Global Forecast System (GFS) over diverse African pilot regions reveals model biases including temperature inaccuracies, underestimation of relative humidity, and wind speed overestimation. The study... Read more
Key finding: This work applies nonlinear neural network postprocessing to extend the skill of the NOAA/NCEP Climate Forecast System (CFS) for challenging subseasonal (Week 3-4) precipitation and temperature forecasts. It shows that ML... Read more
3. How can advanced time series and fuzzy logic methods improve local meteorological forecasting by capturing nonstationary and chaotic components?
This theme focuses on using advanced signal processing and soft computing techniques, such as time series analysis, spectral decomposition, and Type-2 Fuzzy Logic Systems (FLS), to enhance local meteorological forecasting capabilities. These methods emphasize decomposing meteorological signals into deterministic and chaotic parts, enabling better handling of nonstationarity and noise inherent in meteorological data for improved predictive accuracy on variables like temperature and humidity in shorter-term horizon forecasts.
Key finding: By deploying Type-2 Fuzzy Logic Systems trained on local historical meteorological time series, including frequency domain signal decomposition to separate chaotic and deterministic signal components, this study achieves... Read more
Key finding: This paper presents a spectral analytical framework using Fourier and wavelet transforms to determine optimal sampling frequencies and analyze temporal variability in nonstationary meteorological signals. The approach enables... Read more