Development of a globally calibrated stochastic weather generator with the pattern scaling technique to study future climates

Wilson Kemsley, Sarah (2022) Development of a globally calibrated stochastic weather generator with the pattern scaling technique to study future climates. Doctoral thesis, University of East Anglia.

[thumbnail of SWK 230301 PhD final thesis.pdf]
Download (11MB) | Preview


The most sophisticated tools for studying future climates are General Circulation Models (GCMs). However, GCMs have biases, coarse resolution, and have not simulated all relevant scenarios. Stochastic weather generators (SWGs) produce synthetic daily time series of weather variables that, when calibrated with local observational data, can address the GCM bias and scale issues. If their parameters are perturbed using specific simulations, projections are still limited to scenarios simulated by the parent GCM. To solve this, the pattern scaling (PS) technique is applied for the first time to estimate SWG parameter perturbations for future scenarios or warming levels, by diagnosing GCM-simulated dependence of SWG parameters on global mean surface temperature (GMST).

A daily SWG is developed for multiple weather variables and calibrated using observations from a range of climates. It uses a Markov-chain gamma-distribution model for precipitation and a regression model conditioned on precipitation for temperatures. A new assessment of Markov-chain model order across Köppen climate regimes finds that optimal model order primarily depends on performance metric. Another innovation is a generalised linear model for mean wind speed, with further regression models for humidity and radiation. The model is collectively referred to as the Globally Calibrated stochastic Weather Generator (GCWG).

Response patterns of the key precipitation and temperature parameters are diagnosed globally from GCM projections to demonstrate how to combine PS with the GCWG. For example, linear regression has been used to determine the response of first-order Markov probabilities to increasing GMST, with R2 scores typically exceeding 0.5 over land. Mean daily minimum temperatures show larger increases with GMST than maximums (increasing at 3.4°C/°C compared to 1.3°C/°C). GCWG parameters are hence perturbed by GCM-scale responses to construct local-scale time series for a range of scenarios with the potential to emulate those that have not yet been simulated by GCMs.

Item Type: Thesis (Doctoral)
Faculty \ School: Faculty of Science > School of Environmental Sciences
Depositing User: Nicola Veasy
Date Deposited: 23 Mar 2023 14:35
Last Modified: 23 Mar 2023 14:42

Actions (login required)

View Item View Item