On the suitability of a convolutional neural network based RCM-emulator for fine spatio-temporal precipitation

100%

On the suitability of a convolutional neural network based RCM-emulator for fine spatio-temporal precipitation

Abstract

High resolution regional climate models are necessary to capture local precipitation but are too expensive to fully explore the uncertainties associated with future projections. To resolve the large cost of regional climate models, a neural network based regional climate model-emulator for the near-surface temperature, at a daily and twelve kilometer-resolution, was proposed. It uses existing regional climate model simulations to learn the relationship between low-resolution predictors and high resolution surface variables. When trained, the emulator can be applied to any low resolution simulation to produce ensembles of high resolution emulated simulations. This study assesses the suitability of applying the regional climate model-emulator for precipitation thanks to a novel asymmetric loss function to reproduce the entire precipitation distribution over any grid point. Under a perfect conditions framework, the resulting emulator shows striking ability to reproduce the regional climate model original series with an excellent spatio-temporal correlation. In particular, a very good behavior is obtained for the two tails of the distribution, measured by the number of dry days and the ninety-ninth quantile. Moreover, it creates consistent precipitation objects even if the highest frequency details are missed. The emulator quality holds for all simulations of the same regional climate model, with any driving global climate model, ensuring transferability of the tool to global climate models never downscaled by the regional climate model. A first showcase of downscaling global climate model simulations showed that the regional climate model-emulator brings significant added-value with respect to the global climate model as it produces the correct high resolution spatial structure and heavy precipitation intensity. Nevertheless, further work is needed to establish a relevant evaluation framework for global climate model applications.

One Introduction

Precipitation is the primary source of accessible freshwater on Earth. It plays a pivotal role in maintaining Earth's system equilibrium, supporting ecosystems, and crucially, sustaining human survival and activities. However, it also harbors the potential for catastrophic events. Intense rainfall can lead to devastating floods and adversely impact agricultural yields. Severe droughts inflict significant damage on ecosystems, agriculture, and access to potable water. Given the contemporary backdrop of global climate change, it is crucial to study potential changes in precipitation patterns and extremes.

The study of precipitation is inherently complex. It is a non-continuous variable, neither in temporal nor spatial terms. Precipitation occurrences are characterized by their frequency and intensity but also by their duration and spatial extent. Investigating precipitation series across diverse temporal and spatial scales is imperative for a comprehensive grasp of their inherent nature. While rainfall or snowfall may be influenced by large-scale atmospheric circulations, they can also manifest as highly localized events due to small-scale physical processes, influenced by local topography or surface heterogeneity, among other factors. Fine spatial and temporal resolution is, therefore, imperative when modeling precipitation and studying its local changes in the context of global climate change.

Undeniably, regional climate models stand out as one of the most widely employed modeling tools today, to fulfill the imperative for precise spatial and temporal resolution in projecting the future dynamics of precipitation. Regional climate models are a specific kind of climate models used to downscale at high-resolution and over a limited domain the low resolution simulations produced with Global Climate Models. Their high computational costs render unfeasible the production of large ensembles of high resolution simulations necessary to address the different sources of uncertainty associated with the local impacts of climate change. To try to address this high-resolution versus large-ensemble dilemma, various papers introduced the concept of emulator for Regional Climate Model as a solution to create large ensembles of high resolution climate projections blending the regional climate model approach with modern machine-learning techniques. In the recent years, many regional climate model-emulators have been proposed, focusing on different variables or different regions, showing the exponential success of the approach.

A regional climate model-emulator for the near-surface temperature for a regional climate model at its full resolution, twelve kilometers, over Europe was introduced. The concept of the regional climate model-emulator involves using machine learning tools to learn the relationship between low-resolution tropospheric variables describing the atmospheric circulation on a specific day and a high-resolution local surface variable, such as daily precipitation. This downscaling function is learnt inside existing regional climate model simulations. The aim is to tackle the cost limitation of regional climate models by mimicking its downscaling function for a specific variable at a low computational cost and then by applying it to any global and low resolution simulation. Regional climate model-emulators are categorized as hybrid downscaling methods because they incorporate both statistical and dynamical downscaling. Using historical and future regional climate model simulations in the training set enables the regional climate model-emulator to learn how this relationship may evolve under changing climate conditions. Moreover, a regional climate model-emulator can also be built over regions with no long series of good quality precipitation records as it relies only on regional climate model simulations. Here, we propose testing if the regional climate model-emulator introduced is suitable for the downscaling of daily precipitation at twelve kilometers and we propose an adaptation to better capture the complexity of this new variable.

Numerous studies have proposed statistical downscaling methods to estimate the relationship between large-scale and local-scale variables in observational records. Some recent studies have successfully implemented convolutional neural networks for this purpose. The regional climate model-emulator employed and here is based on a fully convolutional neural network architecture called UNet. It has exhibited an excellent ability to emulate the temperature, notably in reproducing the complex spatial structure and daily variability brought by the regional climate model. However, since precipitation is more challenging to model than temperature, this study proposes to explore the use of the loss function to help the neural network focusing on a specific task, following the spirit of previous work. Here, the challenge relies in the reproduction of the complex precipitation distribution and specifically for the heavy precipitation events. Here, we introduce a novel asymmetric loss function tailored for daily precipitation, which we compare to two classical choices for regression problems that we consider as benchmarks. Other recent studies have proposed different strategies to improve the skills on the reproduction of precipitation such as oversampling approach or generative neural networks.

After assessing the suitability of the RCM-Emulator for precipitation, we propose in this study to profit from the EURO-CORDEX simulations to evaluate the transferability of the tool. Indeed the emulator is trained using a given set of available RCM simulations (driven by a given GCM and RCP scenario) and it is crucial to study its behavior when downscaling other socio-economic scenarios or GCMs. Then, in a first step, we evaluate the emulator in a perfect model framework (presented in Section two point one) regarding all available simulations with the emulated RCM. Then in a final step, we propose a first showcase of application by downsampling GCM simulations.

This paper is organised into four main sections. In Section two, we recall the concept of the RCM-emulator introduced in Doury et al., define the technical aspects related to the neural network and the loss functions, and present the framework of the study, including the data, the target domain, and the associated predictors. Section three presents the detailed evaluation and comparison of the emulators within a perfect model framework, while Section four shows the results of applying the asymmetric emulator to GCM simulations. The concluding section summarizes the paper and initiates the discussion.

Two Methodology

Two point one RCM-emulator concept and calibration process

Two point two Data: the RCM matrix

Two point three. Predictands, predictors and neural network architecture

Two D Variables

Two point four. Loss function for the neural network training

Two point five. Evaluation metrics

Two point five point one. Time series comparison

Two point five point two. Climatological scale metrics

Two point five point three. PDF normalisation

Two point five point four SAL score.

Three Perfect model evaluation

Three point one Comparison of the three emulators

Three point one point two Daily scale analysis

Three point one point three Climatological scale analysis

Three point one point four Object oriented analysis

Three point two. Transferability assessment in perfect model framework

Three point two point one. Present climate

Three point two point two Climate change reproduction

Three point two point three Perfect model evaluation conclusions

Four. Application to GCM predictors

Four point one. Showcases on the large scale differences between RCM and GCM

Four point two Present climate

Four point three Climate change

Five Conclusion

Overview

The study evaluates the appropriateness of a CNN-based emulator for fine spatio-temporal precipitation modeling, highlighting its ability to produce high-resolution data from lower-resolution climate simulations. The results demonstrate notable performance, particularly in capturing precipitation extremes and distributions.

Key Points

1A CNN emulator effectively simulates high-resolution precipitation from lower-resolution models
2The study introduces a specialized asymmetric loss function to enhance precipitation modeling
3Results showcase strong correlation with original RCM outputs, especially in extreme events
4The emulator maintains transferability across different global climate models
5Further evaluation frameworks are necessary for global climate model applications.

Details

Authors: Antoine Doury, D. Samuel Somot, Sebastien Gadat
Category: Environmental Studies

PDF
Environmental Ethics and Philosophy: An Exploration
This document is an academic assignment exploring various concepts related to environmental ethics, ecology, and the philosophical relationship between humans and nature. It consists of multiple questions that require long-form answers discussing key themes and theories.
PDF
Environmental Ethics and Ecosophy: Concepts and Perspectives
This document discusses various concepts related to the environment, including definitions of ecology, ecosystems, and the philosophical framework of ecosophy. It covers the relationship between applied ethics and environmental ethics, the significance of ecosophy, and perspectives from classical Indian philosophy and religious viewpoints.
PDF
Our Planet: From Deserts to Grasslands - Full Episode
This document details various ecosystems and the interconnectedness of wildlife, showcasing topics such as wildlife migration, the impact of climate change, and the resilience of nature. It emphasizes the importance of understanding and supporting the natural world's functionality.
PDF
Types of Soil and Soil Conservation Measures
This document discusses various types of soil, their characteristics, formation processes, and measures for soil conservation to combat erosion. It covers soil types such as laterite and arid soils, and highlights the impact of human activities and natural forces on soil erosion.
PDF
A diversity of sustainable lifestyles in 2050: Future SLIM scenario narratives for deep climate change mitigation
This paper discusses four comprehensive lifestyle scenario narratives, known as the Sustainable Living in Models (SLIM) scenarios, which explore plausible alternative futures where lifestyle changes significantly contribute to climate change mitigation. The narratives emerge from multidisciplinary workshops and aim to inform policymaking by integrating qualitative and quantitative approaches.