High-Fidelity Exascale-Enabled Infrastructure for analysing the impact of wind farm wakes on wind/sea interactions

About the Project

OVERVIEW AND BACKGROUND. The extraction of energy from the wind yields the formation of low-speed regions (wakes) behind wind farms (WFs). Wakes are particularly persistent offshore [2], and were recently shown to affect the heat exchange between sea and atmosphere, due to reduced convective heat transfer close to the sea surface [1]. With worldwide offshore wind capacity en-route to achieve 2,000+ GW already by 2050, WF wakes may alter ocean dynamics and marine ecosystems to extents comparable to anthropogenic climate change [2]. Evaluating wakes’ environmental impact credibly requires regional- to mesoscale climate simulations with high-fidelity WF parametrizations at temporal and spatial resolution beyond present supercomputers’ capabilities. Using Graphics Processing Unit (GPU) computing [3], this project will develop the code infrastructure to support these simulations on exascale machines, demonstrating prototype physical investigations using the developed technology.

METHODOLOGY. Two community codes for short-to-long term climate modelling are considered: the Weather Research and Forecasting (WRF) model [4], and the Model for Prediction Across Scales (MPAS) [5]. The codes feature similar models of atmospheric physics, but use different numerical methods. WRF uses structured grids with nested domains to increase resolution in WF wake regions, whereas MPAS uses a single unstructured Voronoi grid with controllable local refinement. WRF has state-of-the-art WF parametrisations [6,7] but little GPU work reported; MPAS uses GPU acceleration but has little work reported on WF parameterization.

This research aims at combining the strengths of both codes to develop a reliable exascale-scalable code for the considered problem. The choice of the baseline code for the project’s core development and demonstrations will follow the teaser projects (TPs) below, which offer hands-on training in climate modelling, wind farm aerodynamics and distributed-memory and GPU parallel computing, and assess the codes’ strengths. Following the TPs, the student will focus on specific topics, e.g. improving the overall code GPU framework or optimizing the parallelized WF model in existing GPU framework, depending on the code selected.

The TPs will share one test case, to compare the two codes’ predictive capabilities and computational performance (execution speed) without GPU acceleration. The GPU development work will be performed on Lancaster University’s HEC cluster and the Bede supercomputer [9].

Teaser project 1 (TP1): WRF-based. To investigate and optimize the predictive capabilities of the two WF parametrizations [6,7] in WRF, analyses (TC1) of the North Sea area containing two real WFs [10] will be performed. The capabilities of both models to predict wind turbine (WT) and WF wakes will be optimised using regression methods for the models’ parameters, and lidar and satellite wind speed measurements to steer the optimization. Measured WT power will also be used in the process, as this parameter is affected by wakes. A second test-case (TC2) without WFs will be used to perform parallel profiling studies of WRF, identifying the code’s computationally most intensive parts and familiarising with its structure. These analyses will identify the code sections that would benefit most from GPU acceleration. TC2 will also be used to cross-compare the predictive capability of WRF and MPAS, assessing it by comparing predicted near-sea surface wind speed maps to measurements from satellites and lidars. Boundary and initial conditions for TC1 and TC2 will be taken from the ERA5 global climate reanalysis [8].

Teaser project 2 (TP2): MPAS-based. First, TC2 will be set up and analyzed without GPUs to cross-compare the computational speed and prediction capabilities of wind speed field of MPAS and WRF. Then, more comprehensive TC2-based parametric analyses of the performance of MPAS using different numbers of CPUs and GPUs will be undertaken to study the dependence of the computational performance of the hybrid parallelization on the CPU and GPU counts, and determine the largest achievable acceleration and the corresponding optimal ratio of GPU and CPU counts - an information paramount for exascale porting. These analyses also enable familiarising with the MPAS structure, knowledge needed to optimally merge wind farm models with the MPAS GPU infrastructure.

TP1 OBJECTIVES: A) Familiarise with WRF: assess predictive capabilities of 3D wind fields with/without WFs; analyze/optimize best suited WF parametrization: B) Assess computational performance and estimate potential of GPU acceleration.

TP2 OBJECTIVES: A) familiarise with MPAS: assess predictive capabilities of 3D wind fields; investigate performance of hybrid CPU/GPU parallelisation; B) Investigate optimal integration of WF model into GPU framework.

Informal enquiries and how to apply

Informal enquiries to Dr. M. Sergio Campobasso: m.s.campobasso@lancaster.ac.uk. Applications from UK applicants only should be made to: https://www.exageo.org/apply/ by Friday 22nd May 2026.