Infrared-Image Processing for the DLR FireBIRD Mission

Abstract

The release of greenhouse gases and aerosols from fires has a large influence on global climate: on average, fires are responsible for up to 30% of anthropogenic $CO_2$ emissions.

The German Aerospace Center (DLR) is operating the “FireBIRD” constellation, which consists of the two satellite missions TET-1 (Technology Test Platform), and BIROS (Bispectral Infrared Optical System) It is dedicated to scientific investigation of the issues involved as well as to early fire detection from space. The satellite and detector approach is based on proven DLR technology achieved during the BIRD (Bispectral Infrad Detection) Mission, which was launched in 2001 and was primarily used for observation of fires and volcanic activity until 2004.The Payload of TET-1 and BIROS has spectral channels in visible (VIS), near infrared (NIR), mid wave (MIR) and a thermal infrared (TIR) channel. The paper is focused on the processing for TET- and BIROS- Fire- BIRD image data. In the FireBird standard processing chain level 1b and 2a data-products are generated automatically for all users after the data reception on ground. The so called fire-radiative-power (FRP) is one of the most important climate relevant parameters witch is estimated by using the bi-spectral method. Two characteristics of the FireBIRD sensors are unique: first, the high radiometric dynamic sensitivity for quantitative evaluation of normal temperatures and high temperature events (HTE) in the same scene. Second, the evaluation of the effective fire area in square meters independent of the recorded number of fire cluster sizes, which is given as the number of pixels per cluster. For certain users, such as firefighters, it is necessary to obtain fire data products (location and temperature) quickly and with minimal delay after detection. In such applications, data processing must take place directly on board the satellite without using a complex processing chain. The paper describes also an alternative fire-detection algorithm witch uses artificial neural networks (deep learning) and will compare it with the standard Level-2 FireBIRD processing.