Multi-sensor tracking of pyroconvection reveals discrepancies between satellite cloud-top detection and convective dynamics

Abstract. Clouds generated by intense wildfires can evolve from pyrocumulus (pyroCu) to pyrocumulonimbus (pyroCb), injecting smoke into the upper troposphere and occasionally the stratosphere. Automated satellite-based detection of pyroCb is therefore critical for monitoring extreme fire–atmosphere interactions, yet such approaches inherently rely on cloud-top properties rather than the underlying convective dynamics.

Here, we evaluate the pyroCb detection algorithm of Peterson et al. (2017), originally developed for GOES observations over North America, in the Australian context using Himawari-8 multispectral imagery during the 2019–2020 Black Summer bushfires. Satellite-derived classifications are assessed against independent observations from operational weather radar and total lightning networks within a multi-sensor framework implemented in the PyroScope platform. An object-based tracking approach is used to compare the temporal evolution of radar- and satellite-derived pyroconvective features.

Satellite observations reliably capture the onset of deep pyroconvective development, showing strong agreement with radar-derived echo-top heights and lightning activity during the early phase. However, a systematic divergence emerges as events evolve, with satellite-derived objects increasingly dominated by horizontally advected anvil structures rather than the actively convecting core. This behaviour is reflected in increasing centroid displacement, decreasing spatial overlap, and an apparent overestimation of plume extent and persistence in satellite retrievals.

To investigate this limitation, we introduce an additional spectral test to separate detached or decaying anvils (class 4) from intense pyroCb (class 5), and explicitly evaluate the impact of this classification on object-based metrics. Restricting the analysis to class 5 reduces the spatial extent and persistence of satellite-derived objects, but does not fully eliminate the divergence with radar observations at later stages, highlighting the inherent limitations of an anvil-based detection paradigm.

These results demonstrate that satellite-based pyroCb detection should be interpreted as a cloud-top representation of pyroconvection rather than a direct measure of the actively convecting core. Multi-sensor integration is therefore essential to distinguish between initiation, mature, and decay phases of pyroconvection. The object-based tracking framework developed here provides a foundation for future studies of pyroCb lifecycle, including the quantification of growth, decay, and potential precursors to pyroCb formation.