top of page

PyroCast-Morpheus: A Mnemonic Counterfactual Digital Twin for Pyrocumulonimbus Wildfire Prediction and Analysis

Avani Patil
31/10/2026

Wildfires continue to increase today as a result of climate change, human activity, and natural causes such as lightning. This has also been seen with a rise of Pyrocumulonimbus (PyroCb) clouds, which are a result of updrafts created from fires and wildfires. This paper describes and analyzes a digital twin model that predicts, simulates, and provides current state, memory retrieval, forecasting, and risk analysis for future scenarios. We trained and validated a digital twin model which processed 10 events across North America as 227 rows of 6-hour interval data. To provide more context for the analysis, the data is composed of ERA5 single level variables, pressure level variables, GOES satellite imagery, vegetation data and terrain data per event. This study is an expansion on current PyroCb prediction models and datasets in order to create a well-rounded pipeline of a digital twin model. Results showed the 6-hour EnKF synchronization reduced state RMSE by an average of 45%. Additionally, the fused memory twin was found to retrieve similar past events to the current wildfire state and had a higher prediction accuracy to the true observed path of wildfire. This digital twin can later be applied to a larger and worldwide dataset and can potentially enable teams to work on ground level operations more efficiently, plan containment, and rescue operations.

bottom of page