Thermodynamically constrained retrieval algorithm to estimate subpixel fire properties

Satellite-based measurements have been widely used for estimating fire-emitted pollutants based on the parameters of either burned area or fire radiative power (FRP). Fire-related remote sensing additionally requires information on active fire areas and fire temperature at a subpixel scale, as well as the combustion phases (i.e., smoldering and flaming) to infer the plume injection height and to understand the mechanics of resulting atmospheric processes like pyro-convection. The FRP is as a key indicator of fire intensity that is frequently retrieved using infrared signals. The fire properties at a subpixel level, including the effective fire temperature and fire area, can be retrieved by the bi-spectral method. However, these approaches normally neglect the heat transport phenomena and subsequently fail to characterize the fire area that could be composed of different combustion phases (e.g. smoldering, flaming, or a combination of the two). Neglecting the phenomena of heat transfer leads to mis-estimation of the actual fire area and its associated emission profile for combustion products that are a function of the combustion phase. To address this challenge, this work presents a new approach to resolve the effective temperature variation inside each fire pixel using a semi-empirical heat-transfer algorithm. This algorithm utilizes radiance observations from geostationary satellites as inputs. With the aid of fine-spatial resolution spaceborne and airborne observations, we evaluated and validated the fire retrieval performance on western US wildfires corresponding to the 2019 season. Our results show that FRP obtained through this heat transfer method exhibits a stronger linear correlation with those retrieved from airborne measurements. Moreover, by analyzing the temperature variation curve obtained using this method, it is possible to further retrieve the fire area under different combustion conditions within the fire pixels.