Large eddy simulations of a partially premixed flame are performed with the purpose of predicting the reacting flow in a swirl-stabilised, low emissions industrial gas turbine combustor. The corresponding sub-grid scale turbulence-chemistry interactions are modelled using a probability density function transport equation, which is solved by the stochastic fields method. A 15-step reduced, but accurate, methane mechanism including 19 species is employed for the description of all chemical reactions. The test case involves a combustor with complex geometry and simulations are carried out for two different combustor operating conditions. Overall, results of the velocity, temperature and species mass fractions (including carbon monoxide) as well as the instantaneous thermochemical properties are shown to be in good agreement with experimental data, demonstrating the capabilities of the applied stochastic fields method. The inclusion of wall heat transfer in the combustion chamber is found to improve temperature and species predictions, especially in the near-wall regions. Comparisons between an oscillating and a 'stable' flame case furthermore highlight the influence of experimentally observed thermo-acoustic instabilities on the scalar fluctuations near the combustor centreline. None of the default model parameters were adjusted and the results showcase the accuracy and flexibility of the present large eddy simulation method for an application to complex, partially premixed combustion problems; this being particularly important for the designers of new generation low emission gas turbine combustors. (C) 2019 The Combustion Institute. Published by Elsevier Inc. All rights reserved.