Predictability of weather and climate

Abstract:

Increasing the resolution of numerical models has played a large part in improving the accuracy of weather and climate forecasts in recent years. Until now, this has required the use of ever more powerful computers, the energy costs of which are becoming increasingly problematic. It has therefore been proposed that forecasters switch to using more efficient ‘reduced precision’ hardware capable of sacrificing unnecessary numerical precision to save costs. Here, an extended form of the Lorenz ‘96 idealized model atmosphere is used to test whether more accurate forecasts could be produced by lowering numerical precision more at smaller spatial scales in order to increase the model resolution. Both a scale-dependent mixture of single- and half-precision – where numbers are represented with fewer bits of information on smaller spatial scales – and ‘stochastic processors’ – where random ‘bit-flips’ are allowed for small-scale variables – are emulated on conventional hardware. It is found that high-resolution parametrized models with scale-selective reduced precision yield better short-term and climatological forecasts than lower resolution parametrized models with conventional precision for a relatively small increase in computational cost. This suggests that a similar approach in real-world models could lead to more accurate and efficient weather and climate forecasts.

Abstract:

Posit numbers, a recently proposed alternative to floating-point numbers, claim to have smaller arithmetic rounding errors in many applications. By studying weather and climate models of low and medium complexity (the Lorenz system and a shallow water model) we present benefits of posits compared to floats at 16 bit. As a standardised posit processor does not exist yet, we emulate posit arithmetic on a conventional CPU. Using a shallow water model, forecasts based on 16-bit posits with 1 or 2 exponent bits are clearly more accurate than half precision floats. We therefore propose 16 bit with 2 exponent bits as a standard posit format, as its wide dynamic range of 32 orders of magnitude provides a great potential for many weather and climate models. Although the focus is on geophysical fluid simulations, the results are also meaningful and promising for reduced precision posit arithmetic in the wider field of computational fluid dynamics.

Abstract:

Interest in the question of how anthropogenic climate change is affecting extreme weather has grown considerably over the past few years - and 2023 has been no exception. This increase in interest has brought a need for robust approaches that are able to quantitatively answer this question rapidly after an event occurs. However, conventional attribution frameworks using statistical or dynamical climate models have been challenged by several recent events that lay well beyond the historical record. While such events have proven difficult to attribute using conventional methodologies, many were surprisingly well forecast by high-resolution numerical weather prediction systems. These systems generally lie at the state-of-the-art in the spectrum of earth system modelling, and their deficiencies are well documented and understood. We suggest that they therefore represent an opportunity for answering attribution — and other weather and climate risk-related — questions, based on models that are demonstrably able to simulate the often non-linear physics of the extremes that we are most interested in. This can increase the confidence in any attributable changes assessed since such changes can be explained in terms of the underlying physical processes. Further, as attribution science extends beyond purely physical assessments and into socioeconomic impacts, this opportunity will grow: weather models are already widely used by risk and emergency management professionals as inputs to hazard models. A final advantage of basing attribution statements on weather forecast models is that it is not only apparent when a forecast model can be used — but also when the model has a crucial deficiency as indicated by a forecast bust. In this case it would be clear that making a quantitative attribution statement would not be appropriate. We have previously used a global high-resolution and coupled ensemble prediction system to quantify human influence on the Pacific Northwest Heatwave and Storm Eunice. Here, we move from event-centric to pseudo-operational experiments. We present a season of perturbed forecasts for attribution, initialised twice per week during the 2022-23 winter in both pre-industrial and future climates, using the same operational ECMWF model as before. A number of high-impact extreme events took place during this winter, and we will present preliminary results from some of these. We suggest that this large set of simulations may be of interest to a wide range of users both inside and outside the attribution community, and we therefore aim to make them publicly available. In addition, we are keen to overcome the limitation imposed by our use of a single model within these experiments, and therefore invite other weather forecasting groups to run comparable experiments.

Abstract:

A fully-coupled general circulation model of intermediate complexity is documented. The study presents an overview of the model climatology and variability, with particular attention for the phenomenology of processes that are relevant for the predictability of the climate system on seasonal-to-decadal time-scales. It is shown that the model can realistically simulate the general circulation of the atmosphere and the ocean, as well as the major modes of climate variability on the examined time-scales: e.g. El Niño-Southern Oscillation, North Atlantic Oscillation, Tropical Atlantic Variability, Pacific Decadal Variability, Atlantic Multi-decadal Variability. We demonstrate the ability of the model in simulating non-stationarity of coupled ocean-atmosphere modes of variability. Potential applications of the model are discussed, with emphasis on the possibility to generate sets of low-cost large-ensemble retrospective forecasts. We argue that the presented model is suitable to be employed in traditional and innovative model experiments that can play a significant role in future developments of seasonal-to-decadal climate prediction.

Abstract:

The widespread destruction incurred by midlatitude storms every year makes it an imperative to study how storms change with climate. The impact of climate change on midlatitude windstorms, however, is hard to evaluate due to the small signals in variables such as wind speed, as well as the high interannual variability in Atlantic storms. Here, we compare multiple severe midlatitude cyclones with both wind and precipitation impacts using forecast-based event attribution. We use a recent version of the ECMWF IFS ensemble prediction system which is demonstrably able to predict the storms, significantly increasing our confidence in its ability to model the key physical processes and their response to climate change. The comparably high resolution of our simulations, and the focus on individual case studies are particularly useful for dynamically driven events like storms. Our approach is able to combine a dynamical analysis of the storm in question with an analysis of past and future changes. Our results confirm trends of increased severity in storm impacts found in climate projections but add reliability to the forecasted structure and impacts of the storm. This indicates that forecast-based attribution is viable for reliable and fast attribution systems.