The conventional recognition of nighttime fires typically relies on infrared brightness temperature data, which often presents issues such as limited accuracy and challenges in identifying small fires. On the other hand, low-light detectors excel at capturing bright targets in settings of low illumination or during night conditions, making their observational data a valuable supplement to nighttime fire recognition. Consequently, the integration of low-light-assisted infrared technology in nighttime fire recognition holds considerable research significance. In this context, we introduce a novel fire recognition algorithm named FRJLI (nighttime tiny fires recognition by joint low-light-assisted infrared remote sensing data). This algorithm aims to integrate low-light data that eliminates interference from urban lights into fire recognition processes and establish thresholds for both low-light and infrared data to enhance the detection accuracy of small nighttime fires.

Given the heightened intensity and destructiveness of forest and grassland fires compared to fires in other vegetation types, our investigation delves into the atypical behavior exhibited by the visible infrared imaging radiometer suite (VIIRS) data within the medium-resolution infrared channel (M-band) and the low-light channel (DNB) of the U.S. Next Generation Meteorological and Environmental Satellite (NPP) when facing forest and grassland fires. Our methodology involves fusing VIIRS DNB data to extract monthly city light background information, projecting both M-band and DNB data simultaneously to the study area, preprocessing the projected remote sensing data to derive standardized data, and executing multiband threshold discrimination, absolute fire recognition, and contextual discrimination on the processed data to culminate in a comprehensive joint low-light and infrared nighttime fire recognition process.

By implementing the FRJLI algorithm on forest fire in the Republic of Korea and grassland fire in Mongolia, we daily map out the distribution areas of these fires (Fig. 5). Our evaluation process focuses on two key aspects: first, a false color image that integrates low-light radiation values with mid-infrared brightness temperatures; second, the utilization of vegetation indices for a more accurate depiction of the affected fire zones. Ensuring the accuracy of our recognition outcomes, we visually compare the recognition results obtained through the FRJLI algorithm with those yielded by the NASA official algorithm, the MODIS Collection4 algorithm, and the FILDA algorithm (Fig. 9). The FRJLI algorithm demonstrates remarkable consistency with the identification outcomes and false color imagery, enabling the detection of minor fires at the fire line periphery. In a detailed analysis, the identification results from all four algorithms are scrutinized in terms of quantity and area coverage (Fig. 10). The findings affirm that the FRJLI algorithm not only identifies a greater number of fires but also offers superior quality compared to other methods, thus providing crucial technical support for more efficient and precise fire detection processes. Furthermore, an innovative examination of the correlation and sensitivity discrepancies between low-light and infrared data in the daily identification of fires is provided (Fig. 12).This analysis confirms the general patterns observed in fires, validates the trend accuracy of the FRJLI algorithm’s identification outcomes, and highlights its ability to identify colder and smaller fires in contrast to NASA’s findings. Significantly, this study concludes that low-light data is more responsive to the fire’s burning status, while infrared data is more adept at revealing fire trends, showcasing the FRJLI algorithm’s capability to leverage the complementary strengths of low-light and infrared fire detection techniques. Finally, through the insights gleaned, we speculate on and verify the varying states of fire identification achieved by the FRJLI algorithm (Figs. 14 & 15). These figures vividly portray the algorithm’s advantages in accurately identifying fire quantities, pinpointing fire centers and boundaries, as well as capturing critical trends in fire-related data.

Taking into account the peculiar behavior exhibited in mid-infrared brightness temperature, the discrepancy in mid-infrared and long-wave infrared brightness temperatures, and variations in low-light radiation values during fires, we leverage the available data to introduce a novel algorithm for nocturnal tiny fire recognition through joint low-light-assisted infrared technology. Our methodology involves merging monthly city light data with low-light information to mitigate city light interference in low-light fire detection. By leveraging both low-light and infrared data concurrently for fire recognition, we aim to enhance the detection accuracy of small fires, including those concealed in shaded areas. Experimental validation is performed on forest fire occurring in March 2022 in the Republic of Korea and grassland fire in April 2022 in Mongolia, successfully enabling the identification of colder and smaller fires. The proposed algorithm significantly advances the capability to detect these colder and smaller fires, thereby enhancing the quantity and quality of nighttime fire recognition. Furthermore, it offers more precise and timely insights into fire location, fire center coordinates, fire line positions, and trend analysis, making it particularly valuable for forest and grassland resource protection applications. This innovative approach holds immense potential and practical value in bolstering fire management strategies for forest and grassland ecosystems.