Technical Note: Temporal disaggregation of spatial rainfall fields with generative adversarial networks

Abstract

Abstract. Creating spatially coherent rainfall patterns with high temporal resolution from data with lower temporal resolution is necessary in many geoscientific applications. From a statistical perspective, this presents a high- dimensional, highly underdetermined problem. Recent advances in machine learning provide methods for learning such probability distributions. We test the usage of generative adversarial networks (GANs) for estimating the full probability distribution of spatial rainfall patterns with high temporal resolution, conditioned on a field of lower temporal resolution. The GAN is trained on rainfall radar data with hourly resolution. Given a new field of daily precipitation sums, it can sample scenarios of spatiotemporal patterns with sub-daily resolution. While the generated patterns do not perfectly reproduce the statistics of observations, they are visually hardly distinguishable from real patterns. Limitations that we found are that providing additional input (such as geographical information) to the GAN surprisingly leads to worse results, showing that it is not trivial to increase the amount of used input information. Additionally, while in principle the GAN should learn the probability distribution in itself, we still needed expert judgment to determine at which point the training should stop, because longer training leads to worse results.

DownScaleBench for developing and applying a deep learning based urban climate downscaling- first results for high-resolution urban precipitation climatology over Austin, Texas

“…At what resolution this effect would become visible can only be determined by experimentation (Song et al., 2019 ). Second, the method only considers spatial information, whereas localized heavy extremes have temporal disaggregation (Scher & Peßenteiner, 2021 ), which is not considered in this study. In terms of the general limitations of this study, which is often the case with many downscaling exercises, is that lack of reference high-resolution dataset to validate the generated values.…”

Latent diffusion models for generative nowcasting and uncertainty quantification of precipitation fields

“…This has most often been achieved with Generative Adversarial Networks (GANs; [15]), which consist of two simultaneously-trained neural networks: a discriminator that is trained to distinguish real samples that belong to the training dataset from generated samples, and a generator that is trained to produce samples that "fool" the discriminator, thus learning to produce samples that resemble those in the training set. GANs have been used to create precipitation fields in applications such as postprocessing and downscaling [16,17,18], precipitation estimation from remote sensing measurements [19,20] and disaggregation [21]. The state of the art in generative nowcasting is, to our knowledge, presently Deep Generative Models of Rainfall (DGMR) [22], which uses a conditional GAN with a regularization term to incentivize the model to produce forecasts close to the true precipitation.…”

The effect of spatial–temporal characteristics of rainfall on urban inundation processes

“…As the number of extreme rainfall events grows globally, the spatial-temporal heterogeneity of rainfall has become the focus of study. Many in-depth researches dealt with topics as the spatial-temporal characteristics of rainfall erosion force (da Silva et al, 2020;Wang et al, 2013), the relationship between the spatial-temporal characteristics of rainfall and flood characteristics (Villarini et al, 2011), the inversion of the spatial-temporal characteristics of rainfall (Fuentes et al, 2008;Scher & Peßenteiner, 2021;Zhou et al, 2018), the trend of spatial-temporal characteristics of rainfall in a certain area (Jung et al, 2017;Mamunur Rashid et al, 2015;Varouchakis et al, 2018), and the impact of spatialtemporal resolution of rainfall on the accuracy of numerical simulation (Patil et al, 2011;Zhou et al, 2018), and so on.…”