Construction of surface air temperature over the Tibetan Plateau based on generative adversarial networks

Abstract

In situ measurements are the most important basis of obtaining precise meteorological datasets. However, it is difficult to accurately extrapolate the plausible meteorological field in certain regions based on in situ measurements alone, especially in areas with complex topography like the Tibetan Plateau (TP). Gridded products, remote sensing, and data assimilation technique overcome this problem but they have their own weaknesses, such as low resolution, huge computation, and time-consuming. Here we applied the state-ofthe-art generative adversarial networks (GAN) in image inpainting to construct the surface temperature over the TP combined with high-resolution China Meteorological Forcing Dataset (CMFD) air temperature product and in situ observations. In this study, the surface air temperature at 2 m dataset with grids of 0.1 over the TP from 1979 to 2020 based on GAN-based model have been generated, which can capture the spatial patterns and trends of surface air temperature over the TP. Moreover, the GAN-based daily temperature has higher correlation coefficients (0.99) and lower root-mean-square errors (1.06 C) than CMFD (0.98, 1.43 C) and reanalysis products when compared with in situ observations. Finally, the temperature fields over the TP from 2019 to 2020 is constructed, whereas CMFD is only available from 1979 to 2018. The results reflect the advantages of GAN on big-data computation and fully utilizing effective information over the TP, and GAN-based temperature dataset can be used to detect climate change over the region. This study demonstrated the potential and applicability of artificial intelligence for developing and extending practicable high-resolution meteorological dataset over regions with sparse stations and rugged topography.

Product Design Process Based on GANs Model

“…It can learn its feature distribution from excellent design cases, and then generate results similar to the input samples. In this paper, the composition, principle, development and application of GAN are described in detail, and the quality of images generated by GAN is guaranteed [9,10], and the degree of customization of image generation content is continuously improved. According to the functions put forward in the third chapter of this paper, this section will realize the establishment of interior design case base based on GAN training image intelligent generation model and meet the adjustment and customization of design case function layout and decoration style.…”

Super-Resolution Reconstruction of Humidity Fields based on Wasserstein Generative Adversarial Network with Gradient Penalty

“…However, there is room for further improvement, especially in complex mountainous areas. Ye Yang et al [31] combined GANs with high-resolution China Meteorological Forcing Dataset (CMFD) temperature products and ground observation-based TP surface temperature dataset. They generated a 0.1° resolution 2m temperature dataset covering the Qinghai-Tibet Plateau from 1979 to 2020.…”