Background: Smouldering peatland wildfires can last for months and create a positive feedback for climate change. These flameless, slow-burning fires spread horizontally and vertically and are strongly influenced by peat moisture content. Most models neglect the non-uniform nature of peat moisture.

Aims: We conducted a computational study into the spread behaviour of smouldering peat with horizontally varying moisture contents.

Methods: We developed a discrete cellular automaton model called BARA, and calibrated it against laboratory experiments.

Key results: BARA demonstrated high accuracy in predicting fire spread under non-uniform moisture conditions, with >80% similarity between observed and predicted shapes, and captured complex phenomena. BARA simulated 1 h of peat smouldering in 3 min, showing its potential for field-scale modelling.

Conclusion: Our findings demonstrate: (i) the critical role of moisture distribution in determining smouldering behaviour; (ii) incorporating peat moisture distribution into BARA’s simple rules achieved reliable predictions of smouldering spread; (iii) given its high accuracy and low computational requirement, BARA can be upscaled to field applications.

Implications: BARA contributes to our understanding of peatland wildfires and their underlying drivers. BARA could form part of an early fire warning system for peatland.