Abstract
The thesis proposes the development of a global burned area detection algorithm for optical sensors of moderate spatial resolution. Throughout the thesis, four algorithms have been developed and their respective products have been generated on a global scale. Each of them has played a complementary role to the other algorithms, either as an improved version or as an adaptation of the same algorithm to different sensors. The derived products have been validated globally and intercomparisons with other existing products have been carried out. In addition, to confirm the stability of the spatiotemporal patterns, the products have been applied to answer different scientific questions related toanomalies in burned area trends in different parts of the world. The work has been funded and developed under the Fire Disturbance (FireCCI) project of the Climate Change Initiative (CCI) programme of the European Space Agency (ESA) and the Copernicus Climate Change Service (C3S) of the European Commission (EC). The author of this work has also received funding from the Spanish Ministry of Science, Innovation and Universities, through an FPU grant.
When this thesis was first proposed in 2018, there was only a single global burned area product that provided a long and consistent time series. This was the National Aeronautics and Space Administration’s (NASA) MCD64A1 product, which was being operationally produced and provided global burned area information at 500 m since November 2000. On the European side, there were two products available, FireCCI41 and GIO_GL1_BA, but these products provided a time series that was either too short (FireCCI41) or with low reliability. In any case, all three products, including MCD64A1, were far from meeting the requirements set by users in terms of commission and omission errors. In this
context this thesis aimed to advance in the knowledge of global burned area mapping
algorithms and the generation of global products that met or were significantly closer to the users' expectations. To this end, this thesis has used information coming from sensors that had not been used before to generate global burned area products. This information includes the high-resolution bands at 250 m from the Moderate Resolution Imaging Spectroradiometer (MODIS), the Ocean and Land Colour Instrument (OLCI) bands and SYNERGY bands, as well as active fires or hotspots (HS) from MODIS and the Visible Infrared Imaging Radiometer Suite (VIIRS). In this last case, it was the first time that they have been used globally to generate this type of product. Thus, four global burned area products have been generated and made available to the scientific community. These products have been duly analysed and validated to confirm their quality and robustness.
In order to explain this process, the thesis has been structured in eight chapters: an introduction, six publications in international journals (one of them under review at the time of writing) and conclusions.
The first chapter is an introduction that describes the state of the topic at the time the thesis was proposed. This includes a description of the role of fire in the earth system through its interactions with humans, climate, and the carbon cycle. The importance of DEVELOPMENT OF A GLOBAL BURNED AREA MAPPING ALGORITHM FOR MODERATE SPATIAL XIV RESOLUTION OPTICAL SENSORS
having products that provide long time series of global burned area maps to understand these interactions is described as well as the state of the development of these products at that time. Besides, this chapter provides an overview of the thesis by defining the objectives and achievements.
The second chapter presents the first global burned area product of this thesis, called FireCCI50, that was generated within the FireCCI project and was based on the MODIS red and near-infrared (NIR) bands and the HS of the same sensor. The hybrid algorithm used to generate this product was the first to use the high-resolution bands of MODIS and, consequently, FireCCI50 provided the highest spatial resolution (250 m) among existing burned area products for the period 2001-2016. The algorithm was an adaptation of its predecessor FireCCI41, the only product that had been developed so far within the FireCCI project. The burned area detection was based on monthly composites of daily images that were obtained considering the spatial and temporal distance to the HS. The
algorithm had two phases. The first identified those pixels with a high probability of being burned (seed pixels), thus reducing commission errors. The second step applied a contextual growing from these seeds to detect the entire burned patch, reducing omission errors. The FireCCI50 product proved to have a higher detection capability for small fires (<100 ha) than other global products. This article was published in the journal Earth System Science Data with an impact index of 11.333 in Journal Citation Reports (JCR).
The third chapter describes an improved version of the previous product, called
FireCCI51, which overcame the limitations found in its predecessor. A thorough analysis of the results of FireCCI50 showed that the product contained significant border effects in several areas of the globe, which were a consequence of the methodology used to obtain thresholds for burned area detection. This methodology was highly dependent on the distribution of the tiles, i.e. the processing units of about 1100 x 1100 km that had been used to optimise the global processing, leading to significant changes in the thresholds of
adjacent tiles. This resulted in artificial breaks in the fires between tiles. In addition, the window filter applied to limit the excessive growing of burned area patches was found to generate artificial rectangular patterns. It should be noted that the commission and omission errors could be improved as well, taking into account the requirements of the scientific community. Thus, a novel methodology was implemented in FireCCI51 that computed adaptive thresholds for each fire based on spatiotemporal HS clusters. This new methodology reduced border effects, eliminated rectangular patterns, improved detection of small fires and improved omission errors. In fact, global estimates of burned area increased up to an average of 4.63 million km2, 22% more than FireCCI50. The time series was also extended to 2018 and later to 2020. This article was published in the journal Remote Sensing of Environment with an impact index of 10.164 in JCR.
The fourth chapter deals with the operational product C3SBA10 (now called C3SBA11) which was developed for the C3S. The FireCCI51 product was well received by the scientific community, who soon demonstrated the potential of this new product at 250 m and its capabilities to answer scientific questions. However, the FireCCI51 was not XV DEVELOPMENT OF A GLOBAL BURNED AREA MAPPING ALGORITHM FOR MODERATE SPATIAL RESOLUTION OPTICAL SENSORS
conceived as an operational product, so its processing was limited to a specific period, which may or may not be extended. Thus, C3S considered FireCCI51 mature enough to be implemented within its processing chain, which aims to ensure the operational provision of several essential climate variables, including Fire Disturbance. This meant that FireCCI51, which had been designed for the MODIS NIR band, had to be adapted to OLCI, a European sensor on board Sentinel-3 that has spectral bands in the visible and NIR. As C3S is a European service and the Sentinel missions aim to ensure operational earth observation in the future, the adaptation was necessary. Consequently, the FireCCI51 algorithm was adapted to work with the NIR at 300 m of OLCI. The main objective of this new C3SBA11 product was to ensure consistency with FireCCI51, so that the scientific community could combine both products to obtain longer time series.
In this sense, the results showed a Pearson correlation above 0.95 in the global burned area estimates. This article was published in the journal Remote Sensing with an impact index of 4.848 in JCR.
The fifth and final chapter on algorithm development aims to evaluate the impact of including the SYNERGY shortwave infrared (SWIR) bands and VIIRS high-resolution HS in the burned area algorithm. The idea was to analyse whether the inclusion of the SWIR, which is a spectral region quite sensitive to the burned signal, and the VIIRS highresolution HS, which has an enhanced ability to detect small fires, could improve FireCCI51 results and overcome some of its limitations. Analyses of FireCCI51 results within this thesis, but also by the scientific community, identified two key limitations. On the one hand, the temporal reporting accuracy of FireCCI51, i.e. the ability to determine the date on which a pixel was burned, was still limited, with only 18% of the cases showing an accuracy lower than 1 day. On the other hand, border effects, although they
were considerably reduced since FireCCI50, were still present in some specific regions of the Earth. To overcome these limitations, the new algorithm created a monthly composite using a multi-temporal index that was calculated based on SWIR. Then, using an improved thresholding process, which built over the experience of FireCCI51, the new algorithm was able to detect burned area more accurately. This new algorithm improved the commission and omission errors of FireCCI51, eliminated border effects and considerably improved temporal reporting accuracy. The results for 2019 have been published through the FireCCIS310 product, although the idea is to extend it into the future. This article is under review in the journal Remote Sensing of Environment with an impact index of 10,164 in JCR.
Chapters 6 and 7 describe two applications of the FireCCI51 product, which play a key role in assessing the quality of the products generated in this thesis. These studies intended to evaluate the robustness of the FireCCI51 burned area estimates and, thus, aimed to answer scientific questions related to anomalies in the spatiotemporal patterns of burned area.
The first of these applications (Chapter 6) analysed in which extent the fire season of 2019 was anomalous for the Amazon and, in general, for the tropical areas of South DEVELOPMENT OF A GLOBAL BURNED AREA MAPPING ALGORITHM FOR MODERATE SPATIAL XVI RESOLUTION OPTICAL SENSORS
America. These regions are a fundamental part of the earth system, due to their high biodiversity. Therefore, the media, guided by information coming from near-real-time HS products, raised the alarm about the large number of active fires in the Amazon. However, active fires provide an incomplete picture of the burned area. Thus, the study analysed the 19-year time series provided by FireCCI51 to assess whether the situation experienced in 2019 was anomalous or not. The results showed that the burned area in Bolivia,
Paraguay and Venezuela was indeed above normal, but not in Brazil. This article was published in the journal Remote Sensing with an impact index of 4.848 in JCR.
The second application (Chapter 7) aimed to analyse the extreme fire season that took place in eastern Australia in late 2019 and early 2020. The fundamental issue raised in this study was the inability of Australian national structures to effectively monitor the
extreme fires that took place in the east of the country. Therefore, FireCCI51 combined with other historical databases was used to quantify the anomaly of these extreme fires relative to the historical period. The analysis showed that almost 20% of all Australian Eucalyptus Forests burned in 2019-2020 compared to an average of 2% in the 2001-2018 period, meaning 7.59 times the average burned area. This article was published in the journal Nature with an impact index of 49.962 in JCR.
Finally, the eighth chapter presents the most important conclusions reached throughout the thesis. It also describes some limitations that have been found, which, in turn, show the way forward for future studies.
An epilogue has been added at the end of the thesis with a list of scientific papers that have used the products developed in this thesis as input, confirming thus their usefulness and acceptance by the scientific community. Download statistics of the products have also been collected.
| Date of Award | 2022 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Emilio Chuvieco Salinero (Supervisor) |