No abstract

Despite numerous studies in statistical downscaling methodologies, there remains a lack of methods that can downscale from precipitation modeled in global climate models to regional level high resolution gridded precipitation. This paper reports a novel downscaling method using a Generative Adversarial Network (GAN), CliGAN, which can downscale large-scale annual maximum precipitation given by simulation of multiple atmosphere-ocean global climate models (AOGCM) from Coupled Model Inter-comparison Project 6 (CMIP6) to regional-level gridded annual maximum precipitation data. This framework utilizes a convolution encoder-dense decoder network to create a generative network and a similar network to create a critic network. The model is trained using an adversarial training approach. The critic uses the Wasserstein distance loss function and the generator is trained using a combination of adversarial loss Wasserstein distance, structural loss with the multi-scale structural similarity index (MSSIM), and content loss with the Nash-Sutcliff Model Efficiency (NS). The MSSIM index allowed us to gain insight into the model’s regional characteristics and shows that relying exclusively on point-based error functions, widely used in statistical downscaling, may not be enough to reliably simulate regional precipitation characteristics. Further use of structural loss functions within CNN-based downscaling methods may lead to higher quality downscaled climate model products.

Abstract. A Himalayan cloudburst event, which occurred on 3 August 2012 in the Uttarkashi (30.73 • N, 78.45 • E) region of Uttarakhand, India, was analyzed. The near-surface atmospheric variables were analyzed to study the formation, evolution, and triggering mechanisms of this cloudburst. In order to improve upon the understanding provided by the observations, numerical simulations were performed using the Weather Research and Forecasting (WRF) model, configured with a single domain at 18 km resolution. The model was tuned using variation of different parameterizations (convective, microphysical, boundary layer, radiation, and land surface), and different model options (number of vertical levels, and spin-up time), which resulted in a combination of parameters and options that best reproduced the observed diurnal characteristics of the near-surface atmospheric variables. Our study demonstrates the ability of WRF in forecasting precipitation, and resolving synoptic-scale and mesoscale interactions. In order to better understand the cloudburst, we configured WRF with multiply nested two-way-interacting domains (18, 6, 2 km) centered on the location of interest, and simulated the event with the best configuration derived earlier. The results indicate that two mesoscale convective systems originating from Madhya Pradesh and Tibet interacted over Uttarkashi and, under orographic uplifting and in the presence of favorable moisture condition, resulted in this cloudburst event.

No abstract

Abstract. Despite the high historical losses attributed to flood events, Canadian flood mitigation efforts have been hindered by a dearth of current, accessible flood extent/risk models and maps. Such resources often entail large datasets and high computational requirements. This study presents a novel, computationally efficient flood inundation modeling framework (“InundatEd”) using the height above nearest drainage (HAND)-based solution for Manning's equation, implemented in a big-data discrete global grid system (DGGS)-based architecture with a web-GIS (Geographic Information Systems) platform. Specifically, this study aimed to develop, present, and validate InundatEd through binary classification comparisons to recently observed flood events. The framework is divided into multiple swappable modules including GIS pre-processing; regional regression; inundation models; and web-GIS visualization. Extent testing and processing speed results indicate the value of a DGGS-based architecture alongside a simple conceptual inundation model and a dynamic user interface.

No abstract

Cowichan Lake lamprey ( Entosphenus macrostomus) is a threatened species resident to Mesachie Lake, Cowichan Lake, and adjoining Bear Lake and their major tributaries in British Columbia. Decreases in trapping success have created concerns that the population is declining. Some potential threats include water use, climate change, and management actions. Owing to the absence of long-term data on population trends, little information is available to estimate habitat quality and factors that influence it. We sought to fill this gap by examining associations between habitat area and variables representing suspected key drivers of habitat availability. Critical habitat areas were imaged using an unmanned aerial vehicle over a period of three years at three sites at Cowichan Lake and a subsequent habitat area was classified. Meteorological and anthropogenic controls on habitat area were investigated through automatic relevance detection regression models. The major driver of habitat area during the critical spawning period was water level during the storage season, which also depends on the meteorological variables and anthropogenic control. It is recommended that regulation of the weir should aim to ensure that the water level remains above the 1 m mark, which roughly equates to the 67% coverage of water on the habitat area used for spawning.

Despite numerous studies in statistical downscaling methodology, there remains a lack of methods that can downscale from precipitation modeled in global climate models to regional level high resolution gridded precipitation. This paper reports a novel downscaling method using a Generative Adversarial Network (GAN), CliGAN, which can downscale large-scale annual maximum precipitation given by simulation of multiple atmosphere-ocean global climate models (AOGCM) from Coupled Model Inter-comparison Project 6 (CMIP6) to regional-level gridded annual maximum precipitation data. This framework utilizes a convolution encoder-dense decoder network to create a generative network and a similar network to create a critic network. The model is trained using an adversarial training approach. The critic uses the Wasserstein distance loss function and the generator is trained using a combination of adversarial loss Wasserstein distance, structural loss with the multi-scale structural similarity index (MSSIM), and content loss with the Nash-Sutcliff Model Efficiency (NS). The MSSIM index allowed us to gain insight into the model’s regional characteristics and shows that relying exclusively on point-based error functions, widely used in statistical downscaling, may not be enough to reliably simulate regional precipitation characteristics. Further use of structural loss functions within CNN-based downscaling methods may lead to higher quality downscaled climate model products.