A novel deep learning-based approach for reconstruction of historical long-term high-quality gridded meteorological dataset

초록

Climate change presents a significant challenge, impacting both the environment and society. Addressing this challenge effectively necessitates in-depth analysis of climate trends and extreme events, relying on long-term, high-resolution meteorological data. However, such data is often lacking in many regions globally, with dense observation networks only recently becoming available. To tackle this issue, our study proposes a novel framework utilizing deep learning to reconstruct high-quality gridded meteorological data for historical periods. Specifically, our approach involves training deep learning models to bridge the gap between gridded products derived from sparse networks, which have existed from the past to the present, and gridded products from dense networks representing recent periods. This training enables us to simulate a gridded product for historical periods of interest using the trained deep learning model. To demonstrate and test our approach, we reconstruct the gridded daily meteorological data for the historical period (1973-1997) over South Korea, where a dense network has been in operation since the late 1990s. The simulated daily datasets by the developed models can capture complex geographic and topographic effects reasonably well, providing a realistic depiction of resulting climate variability. Furthermore, the extreme analysis suggests that the models accurately represent detailed variations between grid cells, contrasting with the limitations of the one derived from sparse observation network, which suffers from gridding artifacts. These findings emphasize the capability of the framework developed in this study to reconstruct high-resolution, high-quality datasets for historical periods without dense observation networks. This indicates the potential of our approach to produce long-term, high-quality gridded meteorological data crucial for hydrological models and climate change analyses, including extreme events.

키워드