Invariant features are needed for atmospheric normalization of image pairs. Powerful statistical approaches now exist, which are designed to isolate unchanged pixels based on quantitatively evaluating the spectral correlation of pixels in image pairs. This suggests that it should be possible to obtain similar results following an analytical path, and our hypothesis is that the derivation of an analytical procedure will yield some physical insight that is not directly accessible with a stochastic approach. In this paper, we derive an analytical formula that relates pseudoinvariant features (PIFs) to the radiometric properties of the scenes. The formula is then inverted to yield an estimate of the ratio of transmission spectra of the two images given the path radiance for each scene and a set of invariant features, The normalized spectral transmission ratio proved to be quite stable over the 700-1800 nm but was less reproducible outside of that range. The absolute (nonnormalized) values of the transmission ratio implied a disparity in the spectral transmission suggesting that the September sky was clearer in visible light, whereas the August sky was clearer in the infrared. This leads to a discussion of the sensitivity of PIF reflectance to the viewing and illumination angles, suggesting that there may be an inherent uncertainty related to the bidirectional reflectance distribution function of the targets comprising the PIF pixels.
The grassland in China occupies more than 40% of its rural land area. However, grassland degradation has been a serious problem in recent years. Thus, a policy of returning cultivated land into grassland is enacted. An object-oriented image classification using different feature objects was adopted to classify grassland and a hierarchy of layers in different years for change detection was deployed in this paper to monitor land cover changes. An experiment was conducted in Hulun Buir Meadow in Inner Mongolia, China. The experiment shows that the accuracy of classification obtained by the object-oriented method is much higher than that of the traditional unsupervised ISODATA classification. Grassland protection action is taking effect maintaining a sustainable use of grassland ecosystem.
Goal of this work is to investigate the effects of temporal misalignments between multispectral (MS) and panchromatic (Pan) observations when they are fused together to yield a pansharpened product. Conversely from the case in which spatial misalignments are present between MS and PAN images, for which the performances of component substitution (CS) fusion methods are recognized better than multiresolution analysis (MRA) schemes, both quantitative and qualitative results show that multitemporal misalignments are better compensated by MRA rather than by CS methods.
The majority of phenological studies have focussed on extracting critical points, i.e. phenological metrics such as start-of-season, in the seasonal growth cycle. These metrics do not exploit the full temporal detail of time series, depend on their definition or threshold, and are influenced by disturbances. Here, we evaluated a robust phenological change detection ability of a method for detecting abrupt, gradual, and phenological changes within time series. BFAST, Breaks For Additive Seasonal and Trend method, integrates the decomposition of time series into trend, seasonal, and remainder components with methods for detecting change within trend and seasonal (i.e. phenology) component. We tested BFAST by analysing 16-day MODIS NDVI composites (MOD13C1 collection 5) between 2000-2009 covering Australia. This illustrated that the method is able to detect the timing of major phenological changes within time series while accounting for abrupt disturbances and gradual trends. It was also shown that the phenological change detection is influenced by the signal-to-noise ratio of the time series. The BFAST method is a generic change detection method which can be applied to any time series data. The methods are available in the BFAST package for R from CRAN (http://CRAN.R-project. org/package=bfast).
Automated change analysis of multi-temporal SAR images is a challenging task due to the inherent noisiness of SAR imagery and the variability of the backscattering coefficient to the acquisition angle. Several methods have been proposed in the literature to improve the change detection performances with respect to the classical method based on the Log-Ratio operator. In this paper a pixel change feature is proposed and tested on true Cosmo-SkyMed detected images for damage assessment applications. The method does not require any despeckling pre-processing and is robust both to the acquisition noise and to possible variation of the acquisition angle in the two observations.
The development of new technologies and tools for as-much-as-possible automatic multi-temporal data analysis has been a goal for most of the institutions that aim at promoting the use of satellite data in different application domains. In the framework of the Support by Pre-classification to specific Applications Project, started in 2008, the European Space Agency has requested the development of a specific platform, named Multi-sensor Evolution Analysis (MEA), with the scope of demonstrating that long term satellite datasets coming from different sensors can be accessed and exploited in almost real time (few seconds) from a web application as user interface. The MEA system has been implemented based on 15 years of global (A)ATSR data (1 km resolution), together with 5 years of regional AVNIR-2 data (10 m resolution), with the final aim of permitting on-the-fly Land Use / Land Cover Change analysis. Moreover, a modified version of MEA has been set-up to permit the multi-temporal analysis of pollution maps coming from satellite observations and ground measurements, demonstrating the generality of the pursued approach. The present work aims at introducing the basis of the MEA system, describing the two implementations for land cover and pollution multi-temporal analysis, including external validation activities being performed for the first application by third parties.
Near shore bathymetry of Lake Nasser, Egypt was derived by fusing shoreline contours from 58 Landsat images spanning the years 1998–2003 with water levels from the various satellite radar altimeters operated by US and European agencies. A least-square fit is made on paired water area (from Landsat) with water levels (from the altimeters) observations. The fitted function is then used to assign a relative depth to each Landsat image shoreline. A series of shorelines are interpolated into depth contours at 1 m intervals. The bathymetry resulting from this process has rmse ∼10 cm.
Remote sensing data integrated with GIS techniques were used together with socio-economic data in a post-classification analysis to map the spatial dynamics of land use/cover changes and identify the urbanization process in Al Ain resort city, United Arab Emirates. Land use/cover statistics, extracted from Landsat Multi-spectral Scanner (MSS). Thematic Mapper (TM) and Enhanced Thematic Mapper plus (ETM +) images for 1972. 1990 and 2000 respectively, revealed that the built-up area has expanded by about 170.53km2. The city was found to have a tendency for major expansion in four different directions: North, North-east, southeast, and south-west. GIS overlay analysis of multi-temporal satellite data helped us tracking the different classes' trajectories by adopting a GIS coding system unique to each class.
In order to study anomalies and trends of the land surface phenology for different vegetation types in the world and more specifically in tropical regions, the design of a phenological reference dataset is investigated. The main objective is to get a description of the seasonal behaviour and interannual variations in order to study anomalies and potential trends of the vegetation. The work was based on time series acquired during the last 10 years by the SPOT VEGETATION sensor with a 1km spatial resolution. The NDVI was used as an indicator of the vegetation growing cycle. Daily surface reflectance values were composited into decades to reduce clouds and haze effects, using the mean compositing algorithm. The decadal NDVI values were spatially averaged for each pixels belonging to a similar vegetation type and temporally for the 10 years of data. The result is a smooth profile representing the seasonal reference pattern as well as the interannual variability inherent to a specific vegetation type.
The interannual variability of NDVI (STD(t)) was calculated for each semi-monthly interval over the period 1982-2006, using GIMMS NDVI images of Alberta, Canada. Forested areas usually show maximum interannual variability in spring and fall (temperature dependence), while grasslands have maximum variability in summer (moisture dependence). In moister areas., grasslands show less summer variability and approach the forest pattern. Croplands mimic the temporal pattern of grasslands located in the same ecoregion. In the ecotone between naturally forest and naturally grassland ecoregions, crops show greater summer variability than their nearby grasslands, indicating a greater sensitivity by crops than by grasslands to moisture stress. This pattern divergence may be used to show crop particularly sensitive to drought; this would be particularly useful where detailed local meteorological and crop data are not compiled. Changes in patterns over time can also help plan agricultural adaptation to climate change in a spatially complete form.
Timely and accurate information on the location and the extent of land use types is high up the agenda of several governmental and scientific organizations. Remote sensing, through image classification at the sub-pixel level, is an attractive source of this type of information. The remote sensing community has recognized the multilayer perceptron (MLP) as a popular machine learning technique for performing land use classifications, both at the pixel and at the sub-pixel level. However, theoretical advances in the machine learning community are not easily adopted by the classification practice. An example is the continued use of the gradient descent algorithm for MLP training. In this paper, the accuracy of this standard first order learning algorithm was compared to that of five alternative, second order learning algorithms for performing a sub-pixel classification of land use in Flanders. The result are clear: all second order algorithms perform markedly better than gradient descent, thereby illustrating the importance of translating theoretical advances in MLP training to the classification practice.
Satellite Image Time Series are becoming increasingly available and will continue to do so in the coming years thanks to the launch of space missions which aim at providing a coverage of the Earth every few days with high spatial resolution. In the case of optical imagery, it will be possible to produce land use and cover change maps with detailed nomenclatures. However, due to meteorological phenomena, such as clouds, these time series will become irregular in terms of temporal sampling and one will need to compare irregularly sensed time series. In this paper, we present an approach to satellite image time series analysis which is able to both deal with irregularly sampled series and to capture distorted behaviors. We present the Dynamic Time Warping from a theoretical point of view and illustrate its abilities for satellite image time series clustering.
This paper presents some exploratory results of the FP-7 MOCCCASIN project that aims to MOnitor Crops in Continental Climates through ASsimilation of Satellite Information. MOCCCASIN is a collaborative project which focuses on improving the monitoring of winter-wheat and forecasting of winter-wheat yield in Russia by combining modelling techniques with satellite data assimilation [1]. In continental climate, winter wheat is particularly affected by low temperatures during the winter which determine whether rapid regrowth is possible in spring. A pre-requisite to use satellite earth observation to characterize the effect of winter kill on wheat is to determine if the multi-annual seasonal variability over the entire growing season can be grasped by remote sensing indicators. The results over an exploratory study site in Tula region for 5 years (2005-2009) demonstrate that it was possible to retrieve crop status indicators using an approach combining radiative transfer modeling and neural networks which could inform on where winter kill has stricken.
This paper proposes an approach to change detection in multitemporal very high geometrical resolution (VHR) SAR images for surveillance applications. The approach takes advantage of 3 concepts: (i) multiscale representation for a preliminary detection of areas showing significant changes in backscattering between the two images (hot spots); (ii) exploitation of prior information about typical usage of zones of interest in the area under control; and (iii) definition of features and change detectors optimized for an effective detection of specific changes in each zone of interest. Here the proposed approach is designed for the solution of surveillance problems. A data set made up of a pair of multitemporal VHR SAR images acquired by the COSMO-SkyMed (CSK®) constellation in spotlight mode on the commercial port of Livorno (Italy) was used. Experimental results point out the effectiveness of the proposed approach.
Environmental change is an important issue in the Andes region. The objectives of this research are to study NDVI dynamics in the Andes region based on time series analysis of SPOT-Vegetation and NOAA-AVHRR, and to recognize to which extent this variability can be attributed to either climatic variability or human induced impacts. Correlation analysis between NDVI and SPI were performed in order to identify the best lag per pixel. Trends in SDVI and SPI were investigated using linear least square regression. Significant vegetation trends are found in 46% of the area. Both NDVI time series lead to different results, but the coupling of vegetation and precipitation is more pronounced for the SPOT-Vegetation data.
The presented study aims at developing methods of a seasonal correction with the help of phenological observations of the German Weather Service (Deutscher Wetterdienst) and spectral field measurements for classifying grassland habitats. Therefore, spectral measurements were taken between 2007 and 2010 in a study heathland area of 60 km2. These measurements were phenological corrected by a long-term time series. Each measurement date was corrected to a phonological date. With this information, the measurements could be used independently to a specific year. Finally, the measurements were combined in a phonological curve per class. This curve was applied to a time-series of RapidEye images to classify grassland habitats. First results indicate that a wide-range phonological curve is required to achieve results with an increasing accuracy.
Construction of the Petrochemical Complex of Rio de Janeiro (COMPERJ) will introduce a new scenario to the Guapi-Mirim Environmental Protection Area (EPA) in the coming years, since it will require constant environmental monitoring so as to portray its ecological evolution. Therefore, the objective of this paper is to perform a multi-temporal analysis of the Guapi-Mirim EPA, using object-based segmentation and classification techniques applied to IKONOS II images, in order to characterize changes in land use and cover types in the investigated site. Two scenes of the IKONOS II sensor acquired on 2006 and 2008 were chosen for the study. Overall results reveal a regeneration stage for the mangrove ecosystem and a stagnation of the urban area growth within the limits of the Guapi-Mirim EPA.
Currently remote sensing, based on satellite images is one of the most important source of information for multitemporal change detection. From all types of satellite images, the multispectral images present the advantage of characterizing the earth surface in different bands; each band provides different and useful information. In this work we propose a new methodology based on linear PCA to extract useful and meaningful information from signals provided by the remote sensing, and based on it, detect temporal changes Experiments based on images of the satellite CBERS-2B corresponding to the urban and peri urban region of Rio Cuarto of Córdoba state in Argentina have given satisfactory results in change detection.
In this paper, a patch based method for multi-temporal analysis of high resolution image is proposed. Conventionally, multi-temporal analysis performed at pixel level suffer from several restrictions, e.g., registration, bi-temporal analysis. To overcome these restrictions, two methods for multi-temporal analysis are proposed at patch level. One is for change detection in time series data by classifying all pairs of patches along time axis in the whole sequence into two classes. Features used for classification are similarity measures based on local statistical models and histogram of local patterns. The other aims at evolution analysis in long image time series. To characterize the evolution patterns, spatio-temporal local pattern features are extracted from time series data. ν-support vector machine (ν-SVM) is applied to classify different kinds of evolution at patch level. Performance is evaluated based on our database produced by iterative classification.
In Europe, the 2003 summer heat wave damaged forested areas. The purpose of this study is to compare two methods to analyse time series of NDVI images for monitoring forest decline. The first method is based on phenological indicator linked to spring vegetation activity, and on the analysis of its trend. The second method (BFAST) allows extracting the trend by decomposition of NDVI time series into trend, seasonal and remainder components. The two approaches show similar results for trends estimates. The main advantage of BFAST is its capability to detect breakpoints in the linear trend which highlights the impact of the exceptional climatic conditions in 2003 on forest stands development.
