Elsevier

Remote Sensing of Environment

Volume 110, Issue 1, 14 September 2007, Pages 18-28
Remote Sensing of Environment

Early fire detection using non-linear multitemporal prediction of thermal imagery

https://doi.org/10.1016/j.rse.2007.02.010Get rights and content

Abstract

This paper presents a sub-pixel thermal anomaly detection method based on predicting background pixel intensities using a non-linear function of a plurality of past images of the inspected scene. At present, the multitemporal approach to thermal anomaly detection is in its early development stage. In case of space-borne surveillance the multitemporal detection is complicated by both spatial and temporal variability of background surface properties, weather influences, viewing geometries, sensor noise, residual misregistration, and other factors. We use the problem of fire detection and the MODIS data to demonstrate that advanced multitemporal detection methods can potentially outperform the operationally used optimized contextual algorithms both under morning and evening conditions.

Section snippets

Introduction and background

Early detection of fires from space-borne measurements is important from operational and economical perspectives, due to the need to monitor vast territories, the high rate of fire spread, substantially higher costs of fighting large fires, as well as profound consequences of biomass burning for climate, global carbon budgets, ecosystem functions, and other environmental costs. There are many factors currently limiting application of satellite remote sensing for early fire detection, among

Multitemporal non-linear model of background dynamics

Consider a scene S consisting of M pixel locations. The value of intensity measured for a given band at a spatial location s and a time moment t is denoted by w(s,t). We also let Wt denote the band image collected at time t. The images in the sequence are obtained via a spatio-temporal observation process whose analytical form is unknown.

The background object intensity observed at a time t and location s can be represented as a function of intensities that were observed at s previously, at P

Application for MODIS thermal image sequence

The experimental work described in this Section has two objectives. The first objective is to evaluate the image prediction quality (goodness of fit) in real conditions using satellite remote sensing data from the MODerate Resolution Imaging Spectroradiometer (MODIS) instrument. Secondly, we compare the hot anomaly tests constructed by the non-linear DDM to the equivalent contextual tests currently used for fire detection, and to other methods.

Goodness of fit

Fig. 2 displays the sequences of determination coefficients r2 and the sequences of the r.m.s.e. values calibrated to brightness temperature. Note that the values of r2 are high for both thermal bands, R4 and R11, both evening and morning times. The background prediction performance does not deteriorate with the range of prediction, which manifests itself in the temporal consistency of r2 and r.m.s.e. The determination coefficients are similar between the evening and morning images. Fig. 3

Discussion

The objective of the prediction model is not to merely detect objects with changed internal properties. In fact, under real conditions after extended surveillance periods the object properties at most pixel locations are found to have changed since the basis times. When this is the case, the goal is to discriminate between two classes of changes: “natural” changes and anomalous ones. The natural changes are due to the combined effects of object internal properties and various external factors.

Conclusion

The problem of fire detection from satellite observations has received considerable attention in remote sensing literature. The most prominent and optimized algorithms utilize spectral or spatial information, or both. Unfortunately, little progress, if any, has been reported about using temporal dimension of space-borne data for fire detection. In this paper, we presented a multitemporal algorithm for detecting hot anomalies. This algorithm, non-linear DDM, was applied to MODIS thermal image

Acknowledgments

We wish to acknowledge NASA support under subcontract on grant #NNG04GK34G, “Studies of biosphere–atmosphere interactions with a GCM with MODIS Spectral Resolution’’. We thank the anonymous reviewers whose comments substantially improved the manuscript. We also thank Mr. George Scheer and Mr. Lawrence Ross for systems administration and other computation support that enabled us to analyze these data.

References (15)

There are more references available in the full text version of this article.

Cited by (44)

  • A self-adaptive wildfire detection algorithm by fusing physical and deep learning schemes

    2024, International Journal of Applied Earth Observation and Geoinformation
  • eDaRT: The Ecosystem Disturbance and Recovery Tracker system for monitoring landscape disturbances and their cumulative effects

    2020, Remote Sensing of Environment
    Citation Excerpt :

    A subset of training images to serve as basis images. The term “basis images” (Koltunov et al., 2009, 2016; Koltunov & Ustin, 2007) stands for images combined by the anomaly detection methods described in section 2.4 to model background signal in pixels of the inspection images and estimate anomalous change relative to the time period and vegetation conditions spanned by the basis images. Segmentation maps to stratify anomaly detection by vegetation type, usually obtained using Gaussian mixture clustering in the bitemporal Landsat spectral space formed by two seasonally most distinct basis images.

  • A simple method based on the thermal anomaly index to detect industrial heat sources

    2018, International Journal of Applied Earth Observation and Geoinformation
  • The development and first validation of the GOES Early Fire Detection (GOES-EFD) algorithm

    2016, Remote Sensing of Environment
    Citation Excerpt :

    Using a physical approach, Koltunov et al. (2009) derived the ranges of weather and sensor characteristics and the surface thermophysical properties, under which radiance can be accurately modeled by the linear model (1) with eight basis images. In general, more complex practical scenarios can be modeled by increasing the number of basis times, using a non-linear background model, or both (Koltunov and Ustin, 2007), at the cost of extra computational power. We see from the discussion in the previous paragraph that the term “multitemporal” that we use to characterize the DDM-based background modeling should not be interpreted as if only the temporal dimension from the pixel in question is used to estimate W(s, t).

  • A comparison of four relative radiometric normalization (RRN) techniques for mosaicing H-res multi-temporal thermal infrared (TIR) flight-lines of a complex urban scene

    2015, ISPRS Journal of Photogrammetry and Remote Sensing
    Citation Excerpt :

    Thus, TIR imagery are increasingly being used for urban and rural temperature studies (Carnahan and Larson, 1990; Streutker, 2002), urban heat island analysis (Gallo and Tarpley, 1996; Hafner and Kidder, 1999; Gonzalez et al., 2005; Gluch et al., 2006), forest fire detection (Giglio et al., 2008), surface energy flux estimation (Wan and Li, 1997; Voogt and Grimmond, 2000; French et al., 2005), building waste heat analysis (Hay et al., 2011) and others. Although Warner and Chen (2001), Koltunov and Ustin (2007), and Scheidt et al. (2008) and others, have applied radiometric normalization techniques on moderate to low resolution (60 m to 1 km) TIR imagery, we recognize an emerging need to evaluate the utility of relative radiometric normalization techniques on H-res TIR imagery (Hay et al., 2011; Rahman et al., 2013). Based on this summary, the objective of this paper is to operationally evaluate the applicability of existing relative radiometric normalization techniques on high spatial resolution TIR flight-lines to produce a seamless mosaic of a large study area.

  • Semi-supervised change detection method for multi-temporal hyperspectral images

    2015, Neurocomputing
    Citation Excerpt :

    Change detection has been an important topic in multi-temporal remote sensing data [1]. Multi-temporal remote sensing datasets, acquired on the same land-cover area from the same (or different) sensors at different times, are the key data for change detection in many application domains, such as urban expansion monitoring [2,3], disaster monitoring [4], glacial ablation and resource management [5]. Although the multi-temporal multi-spectral imagery has been applied widely for a good performance in change detection domains, there exists a limitation that the limited spectral information cannot represent the objects very well.

View all citing articles on Scopus
View full text