Mapping and monitoring heterogeneous landscapes: spatial, spectral and temporal unmixing of MERIS data

Our environment is continuously undergoing change. This change takes place at several spatial and temporal scales and it is largely driven by anthropogenic activities. In order to protect our environment and to ensure a sustainable use of natural resources, a wide variety of national and international initiatives have been established. In this context, Earth observation sensors can provide a substantial amount of information about the biotic and abiotic conditions of our planet. For instance, high spatial resolution sensors, like Landsat TM, deliver data that can be used to produce maps of canopy properties and of land cover types. However, the use of this kind of sensors is not feasible for obtaining full coverage of large areas. Furthermore, high spatial resolution sensors generally do not provide sufficient temporal resolution for monitoring vegetation development during the year. This is especially true for areas having severe cloud coverage throughout the year. In this respect, coarse spatial resolution sensors, which deliver nearly daily data, have a higher chance of encountering cloud free areas. This facilitates large scale monitoring studies but at the expense of a lower spatial resolution providing images with potentially many mixed pixels. Recent developments in imaging devices resulted into a new kind of sensor that works at a medium spatial resolution while providing high temporal and spectral resolutions. The MEdium Resolution Imaging Spectrometer (MERIS) aboard the European Space Agency’s ENVISAT platform belongs to this category. MERIS measures the solar radiation reflected from the Earth’s surface in 15 narrow spectral bands and it has a revisit time of 2-3 days. This unprecedented spectral and temporal resolution has resulted in several land, water and atmospheric products. In addition, two vegetation indices have been specifically designed to monitor vegetated canopies using this sensor: the MERIS Terrestrial Chlorophyll index (MTCI) and the MERIS Global Vegetation Index (MGVI). However, the spatial resolution provided by this sensor – 300 m in full resolution (FR mode) – is not sufficient to accurately map and monitor heterogeneous and fragmented landscapes. This is why the synergic use of high spatial resolution and MERIS data is investigated in this thesis. More precisely, the objective of this thesis is to develop a multi-sensor and multi-resolution data fusion approach that allows mapping and monitoring of heterogeneous and highly fragmented landscapes using MERIS data. The Netherlands is selected as study area because of its mixed landscapes where patches of arable land, natural vegetation, forests, and water bodies can be found next to each other. Besides this, The Netherlands also suffers from frequent cloud coverage, which severely hampers operational mapping and monitoring with both high spatial and high temporal resolution. Chapter 1 outlines the challenges of mapping and monitoring heterogeneous and fragmented landscapes using data from the current optical Earth observing missions, sketches the core concepts of data fusion and linear spectral (un)mixing and, after that, lists the research objectives of this PhD thesis. Chapter 2 presents the calibration scheme of the MERIS sensor and subsequently focuses on the smile effect and on the vicarious calibration corrections. The effects of these corrections on regional land cover mapping and vegetation status assessment are studied. Our analysis showed that MERIS delivers data with a very high radiometric quality. Nevertheless, some effects were observed when using MERIS data without the smile correction. Therefore, we recommend to systematically apply all the necessary corrections to generate stable long-term series of MERIS data. Chapter 3 introduces a data fusion technique that can be used to generate images with the spatial resolution provided by Landsat TM and the spectral resolution provided by MERIS. The method is based on the linear mixing model and requires the use of a high spatial resolution dataset (in this case, a Landsat TM image) to downscale the information collected by MERIS. Two parameters need to be optimized in this implementation of the unmixing-based data fusion approach: the number of classes used to classify the TM image and the size of the MERIS neighbourhood used to solve the unmixing equations. A quantitative data fusion quality analysis was used to assist with the identification of the best combination of these two parameters. The results of this analysis demonstrated that it is possible to downscale MERIS FR images to a Landsat-like spatial resolution (25 m). Chapter 4 elaborates in more detail on the potential of MERIS fused images for land cover mapping and vegetation status assessment in heterogeneous and fragmented landscapes. First, the fused images are used to produce land cover maps, which are validated using a high spatial resolution dataset. Then, the fused image with the highest overall classification accuracy is selected as the best fused image. Subsequently, this image is used to compute three vegetation indices: the normalized difference vegetation index (NDVI), which is an indicator of vegetation amount and its ‘greenness’, and the two vegetation indices specifically designed to monitor vegetation status using MERIS data (i.e. the MTCI and the MGVI). Classification results for the best fused image and for the TM image used to downscale MERIS are very similar and when comparing spectrally similar images (i.e. no SWIR bands in the TM image), the results of the best fused image outperform those of TM. With respect to the vegetation indices, a good correlation was found between the NDVI computed from TM and from the best fused image (in spite of their different spectral configuration). For the MTCI and the MGVI, the spatial patterns found when using MERIS fused images were consistent with those found in MERIS. The main advantage of using fused images is the possibility of monitoring individual agricultural fields and small vegetation patches. This is not possible when using the original MERIS FR images. Chapter 5 investigates the use of the unmixing-based data fusion approach to downscale time series of MERIS FR data. In this case, a high spatial resolution land use database is used to characterize the landscape composition. Because of this, only the size of the MERIS neighbourhood used for the unmixing needs to be optimized. In this chapter, the AMORGOS 3.0 software was used to ensure working with the best possible MERIS geo-location values. This allowed an automatic image co-registration and the calculation of the actual ground instantaneous field of view (GIFOV) of each MERIS pixel which, in turn, allows us to determine the real number of TM pixels covered by each MERIS pixel without having to re-project the datasets. Similar to chapter 2, a quantitative data fusion assessment was performed in order to test the validity of the proposed method and to identify the best neighbourhood size. After that, the series of fused images was used to compute the MTCI and the MGVI. Finally, MTCI and MGVI temporal profiles were extracted for the main land cover types present in the study area. Results indicate that the selected data fusion approach can be successfully used to downscale time series of MERIS FR data and, therefore, to monitor vegetation dynamics at high spatial, spectral and temporal resolutions. Chapter 6 contains the final conclusions and gives recommendations for further research. The overall conclusion is that the selected unmixing-based data fusion approach is able to downscale MERIS FR data to a Landsat-like resolution with little compromises on the spectral quality. This allows mapping and monitoring heterogeneous and fragmented landscapes that suffer from frequent cloud coverage. The results presented in this thesis should, therefore, encourage further research on multi-sensor and multi-resolution data fusion approaches as a means to bridge spatial, spectral and temporal scaling gaps in current and future Earth observation missions.