Conference paper
Quantum machine learning models for enhanced weather prediction
A. Tiwari, S. Kumar, P. Tyagi, S. Siyanwal, S. Sakhuja
Overview
We apply five quantum machine learning models (QGRU, QLSTM, QNN, VQC and QSVM) to meteorological time-series data from the ERA5 reanalysis dataset, for both regression (forecasting) and classification tasks.
Problem / motivation
Whether current QML models can be used for weather prediction, and where hardware limits and noise constrain them.
Methods
Meteorological features from ERA5 single-level data for Safdarjung, Delhi, are preprocessed and fed to quantum recurrent (QGRU, QLSTM), variational (VQC, QNN) and kernel (QSVM) models. Each model is evaluated on regression and classification tasks, and compared with classical baselines: LSTM and SVM (built by me) and GRU and ANN (built by S. Sakhuja).
Results / outcomes
The models reach reasonable accuracy on both prediction and classification, best on binary classification. Hardware limitations and noise limit scalability and generalisation.
My role
I contributed equally to the code, data, algorithm and comparisons.