Using multispectral information and machine learning to improve daytime fire radiative power estimation from METImage measurements
Fire radiative power (FRP), which represents the instantaneous rate of radiative energy emitted by fires, has been successfully used to characterize fire events. A widely used FRP retrieval method is based on the midwave infrared (∼4 μm) radiance difference between the fire pixel and the background....
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-12-01
|
| Series: | Ecological Informatics |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S157495412500367X |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1849233532336996352 |
|---|---|
| author | Yingxin Gu Ivan Csiszar Marina Tsidulko Wei Guo |
| author_facet | Yingxin Gu Ivan Csiszar Marina Tsidulko Wei Guo |
| author_sort | Yingxin Gu |
| collection | DOAJ |
| description | Fire radiative power (FRP), which represents the instantaneous rate of radiative energy emitted by fires, has been successfully used to characterize fire events. A widely used FRP retrieval method is based on the midwave infrared (∼4 μm) radiance difference between the fire pixel and the background. The METImage sensor onboard the new MetOp-Second Generation satellite missions is an advanced multispectral imaging radiometer covering 0.44 μm to 13.34 μm in wavelength. The 500 m spatial resolution measurements from METImage will enable fire detections at a higher sensitivity than the current medium resolution sensors flown on missions on the mid-morning polar orbit (e.g. Terra MODIS, Sentinel-3 SLSTR). However, METImage does not have a dedicated “fire” band and large fires are expected to trigger saturation in the 4 μm band. In this study, we used a machine learning (ML) regression tree (RT) method and Visible Infrared Imaging Radiometer Suite (VIIRS) M-band data as a proxy of METImage to examine the potential and limitations of METImage FRP retrievals when band saturation occurs in the 4 μm band. The approach is based on first estimating actual 4 μm radiances for saturated measurements, followed by the traditional FRP retrieval using those estimated radiances. 13 VIIRS M-band daytime data, which have the corresponding METImage bands, and related geometry information were obtained from the large fire events in California (8021 samples in total). METImage band saturation levels were applied to the 13 VIIRS band radiances for those 8021 samples to generate the proxy METImage data. 4172 samples were saturated at the 4 μm band which were used as the radiance ML RT model training and testing data. The developed radiance ML model was validated using independent fire sample data collected from USA and Australia. Results indicate that the radiance ML RT models can successfully predict the 4 μm radiance for a wide range of conditions, with a 0.97 correlation coefficient between the predicted and the actual radiance and an average prediction error of 17.8 %. The derived saturated daytime FRPs varied based on the multispectral information, resulting in an average error of 21.7 % to predict reference VIIRS M-band FRP retrievals. |
| format | Article |
| id | doaj-art-32f4495fea534e5bbab9d14483dcae23 |
| institution | Kabale University |
| issn | 1574-9541 |
| language | English |
| publishDate | 2025-12-01 |
| publisher | Elsevier |
| record_format | Article |
| series | Ecological Informatics |
| spelling | doaj-art-32f4495fea534e5bbab9d14483dcae232025-08-20T05:05:57ZengElsevierEcological Informatics1574-95412025-12-019010335810.1016/j.ecoinf.2025.103358Using multispectral information and machine learning to improve daytime fire radiative power estimation from METImage measurementsYingxin Gu0Ivan Csiszar1Marina Tsidulko2Wei Guo3Science and Technology Corporation at NOAA NESDIS Center for Satellite Applications and Research, 5830 University Research Court, College Park, MD 20740, USA; Formerly I.M. Systems Group, Inc. (IMSG), USA; Corresponding author at: Science and Technology Corporation at NOAA NESDIS Center for Satellite Applications and Research, College Park, MD 20740, USA.NOAA NESDIS Center for Satellite Applications and Research, 5830 University Research Court, College Park, MD 20740, USAScience and Technology Corporation at NOAA NESDIS Center for Satellite Applications and Research, 5830 University Research Court, College Park, MD 20740, USA; Formerly I.M. Systems Group, Inc. (IMSG), USAScience and Technology Corporation at NOAA NESDIS Center for Satellite Applications and Research, 5830 University Research Court, College Park, MD 20740, USA; Formerly I.M. Systems Group, Inc. (IMSG), USAFire radiative power (FRP), which represents the instantaneous rate of radiative energy emitted by fires, has been successfully used to characterize fire events. A widely used FRP retrieval method is based on the midwave infrared (∼4 μm) radiance difference between the fire pixel and the background. The METImage sensor onboard the new MetOp-Second Generation satellite missions is an advanced multispectral imaging radiometer covering 0.44 μm to 13.34 μm in wavelength. The 500 m spatial resolution measurements from METImage will enable fire detections at a higher sensitivity than the current medium resolution sensors flown on missions on the mid-morning polar orbit (e.g. Terra MODIS, Sentinel-3 SLSTR). However, METImage does not have a dedicated “fire” band and large fires are expected to trigger saturation in the 4 μm band. In this study, we used a machine learning (ML) regression tree (RT) method and Visible Infrared Imaging Radiometer Suite (VIIRS) M-band data as a proxy of METImage to examine the potential and limitations of METImage FRP retrievals when band saturation occurs in the 4 μm band. The approach is based on first estimating actual 4 μm radiances for saturated measurements, followed by the traditional FRP retrieval using those estimated radiances. 13 VIIRS M-band daytime data, which have the corresponding METImage bands, and related geometry information were obtained from the large fire events in California (8021 samples in total). METImage band saturation levels were applied to the 13 VIIRS band radiances for those 8021 samples to generate the proxy METImage data. 4172 samples were saturated at the 4 μm band which were used as the radiance ML RT model training and testing data. The developed radiance ML model was validated using independent fire sample data collected from USA and Australia. Results indicate that the radiance ML RT models can successfully predict the 4 μm radiance for a wide range of conditions, with a 0.97 correlation coefficient between the predicted and the actual radiance and an average prediction error of 17.8 %. The derived saturated daytime FRPs varied based on the multispectral information, resulting in an average error of 21.7 % to predict reference VIIRS M-band FRP retrievals.http://www.sciencedirect.com/science/article/pii/S157495412500367XMETImageVIIRSMachine learning (ML)Regression tree (RT) modelMultispectral informationFire radiative power (FRP) |
| spellingShingle | Yingxin Gu Ivan Csiszar Marina Tsidulko Wei Guo Using multispectral information and machine learning to improve daytime fire radiative power estimation from METImage measurements Ecological Informatics METImage VIIRS Machine learning (ML) Regression tree (RT) model Multispectral information Fire radiative power (FRP) |
| title | Using multispectral information and machine learning to improve daytime fire radiative power estimation from METImage measurements |
| title_full | Using multispectral information and machine learning to improve daytime fire radiative power estimation from METImage measurements |
| title_fullStr | Using multispectral information and machine learning to improve daytime fire radiative power estimation from METImage measurements |
| title_full_unstemmed | Using multispectral information and machine learning to improve daytime fire radiative power estimation from METImage measurements |
| title_short | Using multispectral information and machine learning to improve daytime fire radiative power estimation from METImage measurements |
| title_sort | using multispectral information and machine learning to improve daytime fire radiative power estimation from metimage measurements |
| topic | METImage VIIRS Machine learning (ML) Regression tree (RT) model Multispectral information Fire radiative power (FRP) |
| url | http://www.sciencedirect.com/science/article/pii/S157495412500367X |
| work_keys_str_mv | AT yingxingu usingmultispectralinformationandmachinelearningtoimprovedaytimefireradiativepowerestimationfrommetimagemeasurements AT ivancsiszar usingmultispectralinformationandmachinelearningtoimprovedaytimefireradiativepowerestimationfrommetimagemeasurements AT marinatsidulko usingmultispectralinformationandmachinelearningtoimprovedaytimefireradiativepowerestimationfrommetimagemeasurements AT weiguo usingmultispectralinformationandmachinelearningtoimprovedaytimefireradiativepowerestimationfrommetimagemeasurements |