Novel approaches for filling gaps in the spatial wind field in the coastal regions of India

Coastal regions in India possess an exceptional wind energy potential, exceeding 8,000 MW, with wind speeds ranging from 6.8 to 7.1 m/s. However, these areas face critical data gaps in wind monitoring networks due to sparse instrumentation, station failures, and disruptions from tropical cyclones that frequently impact India's eastern coast. Accurate, high-resolution wind field data is essential for renewable energy planning, infrastructure resilience assessment, and identifying optimal sites for wind farm development in cyclone-vulnerable regions. This study presents novel approaches for filling spatial wind field gaps. We used two approaches based on Multiple-Point Statistics (MPS), which reconstructs wind patterns by learning spatial relationships from training images, and Deep Learning (DL) using ConvLSTM2D neural networks. We apply these methods to ERA5 reanalysis data at 25 km resolution spanning the Andhra Pradesh region. Two gap scenarios were tested: (i) systematic contiguous gaps, and (ii) random scattered gaps using MPS and DL methods. Preliminary results indicate that the MPS approach yields a Pearson correlation of 0.40 with a mean absolute error (MAE) of 0.42 m/s for contiguous gaps and a Pearson correlation (r) of 0.97 with an MAE of 0.34 m/s for random gaps. The DL method for both random and contiguous gaps exhibit better performance, with r > 0.998 and MAE < 0.16 m/s. Ground-based validation with operational wind farm data remains necessary to confirm site-specific accuracy for practical wind energy applications. These gap-filled wind datasets enable the identification of optimal wind farm locations and support climate risk assessments for existing renewable infrastructure and enhance resilience planning against tropical cyclone hazards.

Keywords: Wind field, Multiple-point statistics, Deep learning, Renewable energy.