Search

Deep Learning (DL) based downscaling has recently become a popular tool in earth sciences. Multiple DL methods are routinely used to downscale coarse-scale precipitation data to produce more accurate and reliable estimates at local scales. Several studies have used dynamical or statistical downscaling of precipitation, but the availability of ground truth still hinders the accuracy assessment. A key challenge to measuring such a method's accuracy is comparing the downscaled data to point-scale observations, which are often unavailable at such small scales. In this work, we carry out DL-based downscaling to estimate the local precipitation using gridded data from the India Meteorological Department (IMD). To test the efficacy of different DL approaches, we apply SR-GAN and three other contemporary approaches (viz., DeepSD, ConvLSTM, and UNET) for downscaling and evaluating their performance. The downscaled data is validated with precipitation values at IMD ground stations. We find overall reasonably well reproduction of original data in SR-GAN approach as noted through M.S.E., variance statistics and correlation coefficient (CC). It is found that the SR-GAN method outperforms three other methods documented in this work (CCSR-GAN = 0.8806; CCUNET = 0.8399; CCCONVLSTM = 0.8311; CCDEEPSD = 0.8037). A custom V.G.G. network, used in the SR-GAN, is developed in this work using precipitation data. This DL method offers a promising alternative to other existing statistical downscaling approaches. It is noted that superiority in the SR-GAN approach is achieved through the perceptual loss concept, wherein it overcomes the issue of smooth reconstruction and is consequently able to capture better fine-scale details of data considered. [ABSTRACT FROM AUTHOR]

This study focuses on the current problems in earth system science (ESS), where machine learning (ML) algorithms can be applied. It provides an overview of previous studies, ongoing work at the Ministry of Earth Sciences, Government of India, and future applications of ML algorithms to some significant earth science problems. We compare previous studies, a mind map of multidimensional areas related to ML and Gartner's hype cycle for ML in ESS. We mainly focus on the critical components in earth sciences, including studies on the atmosphere, oceans, biosphere, hydrogeology, human health and seismology. Various artificial intelligence (AI)/ML applications to problems in the core fields of earth sciences are discussed, in addition to gap areas and the potential for AI techniques. [ABSTRACT FROM AUTHOR]