A cloud detection method based on a time series of MODIS surface reflectance images

The Moderate Resolution Imaging Spectroradiometer (MODIS)-Terra surface reflectance product (MOD09A1), with bands 1 to 7, is a gridded, eight-day composite product derived from the MODIS-Terra top of atmosphere reflectance swaths. It performs cloud detection and corrects for the effects of atmospheric gases and aerosols. The cloud mask (CM) algorithms for MODIS are based on empirical thresholds on spectral reflectance and brightness temperature. Since the spatial resolution of the thermal band is 1000 m, while that of MOD09A1 is 500 m, many undetected and false clouds are observed in MOD09A1. These errors always result in temporal and spatial inconsistencies in higher-level products. In this paper, a cloud detection algorithm (TSCD) based on a MOD09A1 time series is introduced. Time series cloud detection (TSCD) algorithm is based on the relative stability of ground reflectance and the sudden variations in reflectance that result from cloud cover. The algorithm first searches the clear-sky reference data, and then discriminates clouded and unclouded pixels by detecting a sudden change of reflectance in the blue wavelength and spectral correlation coefficient at the pixel level. Compared with cloud cover assessments obtained from MODIS' original CM, TSCD provides similar or better discrimination in most situations when the land surface changes slowly.     

Forest cover classification using Landsat ETM+ data and time series MODIS NDVI data

Forest cover plays a key role in climate change by influencing the carbon stocks, the hydrological cycle and the energy balance. Forest cover information can be determined from fine-resolution data, such as Landsat Enhanced Thematic Mapper Plus (ETM+). However, forest cover classification with fine-resolution data usually uses only one temporal data because successive data acquirement is difficult. It may achieve mis-classification result without involving vegetation growth information, because different vegetation types may have the similar spectral features in the fine-resolution data. To overcome these issues, a forest cover classification method using Landsat ETM+ data appending with time series Moderate-resolution Imaging Spectroradiometer (MODIS) Normalized Difference Vegetation Index (NDVI) data was proposed. The objective was to investigate the potential of temporal features extracted from coarse-resolution time series vegetation index data on improving the forest cover classification accuracy using fine-resolution remote sensing data. This method firstly fused Landsat ETM+ NDVI and MODIS NDVI data to obtain time series fine-resolution NDVI data, and then the temporal features were extracted from the fused NDVI data. Finally, temporal features combined with Landsat ETM+ spectral data was used to improve forest cover classification accuracy using supervised classifier. The study in North China region confirmed that time series NDVI features had significant effects on improving forest cover classification accuracy of fine resolution remote sensing data. The NDVI features extracted from time series fused NDVI data could improve the overall classification accuracy approximately 5% from 88.99% to 93.88% compared to only using single Landsat ETM+ data.     

Dating flowering cycles of Amazonian bamboo-dominated forests by supervised Landsat time series segmentation

Bamboo-dominated forests are unusual and interesting because their structure and biomass fluctuate in decades-long cycles corresponding to the flowering and mortality rhythm of the bamboo. In southwestern Amazonia, these forests have been estimated to occupy an area of approximately 160 000 km², and a single reproductively synchronized patch can cover up to thousands of square kilometers. Accurate mapping of these forests is challenging, however: the forests are spatially heterogeneous, with bamboo densities varying widely among adjacent sites; much of the area is inaccessible, so field verification of bamboo presence is difficult to obtain and georeferenced records of past flowering events virtually non-existent; and detectability of the bamboo by remote sensing varies considerably during its life cycle. In this study, we develop a supervised time series segmentation approach that allows us to identify both the presence of bamboo forests and the years in which the bamboo flowering and subsequent mortality have occurred. We then apply the method to the entire Landsat TM/ETM+ archive from 1984 to the end of 2018 and validate the classification by visual interpretation of very high resolution imagery. Collecting accurate ground reference data of bamboo presence and bamboo mortality timing is notably difficult in these forests, and we therefore developed a methodology that takes advantage of imperfect reference data obtained from the Landsat time series itself. Our results show that bamboo forests can be differentiated from non-bamboo forests using any of the infrared bands, but band 5 produces the highest classification accuracy. Interestingly, there appears to be a temporal difference in the spectral responses of the three infrared bands to bamboo flowering and mortality: near infrared (band 4) reflectance reacts to the event earlier than shortwave infrared (bands 5 and 7) reflectance. The long Landsat TM/ETM+ archive allows our methodology to detect some areas with two mortality events, with a theoretical maximum interval of 29 years. Analysis of these pixels with repeated mortality confirms that the life cycles of the local bamboo species (Guadua sarcocarpa and G. weberbauerii) last typically 28 years.     

Probabilistic mapping of flood-induced backscatter changes in SAR time series

The information content of flood extent maps can be increased considerably by including information on the uncertainty of the flood area delineation. This additional information can be of benefit in flood forecasting and monitoring. Furthermore, flood probability maps can be converted to binary maps showing flooded and non-flooded areas by applying a threshold probability value p_F = 0.5. In this study, a probabilistic change detection approach for flood mapping based on synthetic aperture radar (SAR) time series is proposed. For this purpose, conditional probability density functions (PDFs) for land and open water surfaces were estimated from ENVISAT ASAR Wide Swath (WS) time series containing >600 images using a reference mask of permanent water bodies. A pixel-wise harmonic model was used to account for seasonality in backscatter from land areas caused by soil moisture and vegetation dynamics. The approach was evaluated for a large-scale flood event along the River Severn, United Kingdom. The retrieved flood probability maps were compared to a reference flood mask derived from high-resolution aerial imagery by means of reliability diagrams. The obtained performance measures indicate both high reliability and confidence although there was a slight under-estimation of the flood extent, which may in part be attributed to topographically induced radar shadows along the edges of the floodplain. Furthermore, the results highlight the importance of local incidence angle for the separability between flooded and non-flooded areas as specular reflection properties of open water surfaces increase with a more oblique viewing geometry.     

Land use intensity trajectories on Amazonian pastures derived from Landsat time series

Monitoring changes in land use intensity of grazing systems in the Amazon is an important prerequisite to study the complex political and socio-economic forces driving Amazonian deforestation. Remote sensing offers the potential to map pasture vegetation over large areas, but mapping pasture conditions consistently through time is not a trivial task because of seasonal changes associated with phenology and data gaps from clouds and cloud shadows. In this study, we tested spectral-temporal metrics derived from intra-annual Landsat time series to distinguish between grass-dominated and woody pastures. The abundance of woody vegetation on pastures is an indicator for management intensity, since the duration and intensity of land use steer secondary succession rates, apart from climate and soil conditions. We used the developed Landsat-based metrics to analyze pasture intensity trajectories between 1985 and 2012 in Novo Progresso, Brazil, finding that woody vegetation cover generally decreased after four to ten years of grazing activity. Pastures established in the 80s and early 90s showed a higher fraction of woody vegetation during their initial land use history than pastures established in the early 2000s. Historic intensity trajectories suggested a trend towards more intensive land use in the last decade, which aligns well with regional environmental policies and market dynamics. This study demonstrates the potential of dense Landsat time series to monitor land-use intensification on Amazonian pastures.     

Rapid detection of new and expanding human settlements in the Limpopo province of South Africa using a spatio-temporal change detection method

Recent development has identified the benefits of using hyper-temporal satellite time series data for land cover change detection and classification in South Africa. In particular, the monitoring of human settlement expansion in the Limpopo province is of relevance as it is the one of the most pervasive forms of land-cover change in this province which covers an area of roughly 125 000 km². In this paper, a spatio-temporal autocorrelation change detection (STACD) method is developed to improve the performance of a pixel based temporal Autocorrelation change detection (TACD) method previously proposed. The objective is to apply the algorithm to large areas to detect the conversion of natural vegetation to settlement which is then validated by an operator using additional data (such as high resolution imagery). Importantly, as the objective of the method is to indicate areas of potential change to operators for further analysis, a low false alarm rate is required while achieving an acceptable probability of detection. Results indicate that detection accuracies of 70% of new settlement instances are achievable at a false alarm rate of less than 1% with the STACD method, an improvement of up to 17% compared to the original TACD formulation.     

The ecological fallacy in a time series context: evidence from Spanish regional unemployment rates

The ecological fallacy (EF) is a common problem regional scientists have to deal with when using aggregated data in their analyses. Although there is a wide number of studies considering different aspects of this problem, little attention has been paid to the potential negative effects of the EF in a time series context. Using Spanish regional unemployment data, this paper shows that EF effects are not only observed at the cross-section level, but also in a time series framework. The empirical evidence obtained shows that analytical regional configurations are the least susceptible to time effects relative to both normative and random regional configurations, while normative configurations are an improvement over random ones.

Updating Landsat time series of surface-reflectance composites and forest change products with new observations

The use of time series satellite data allows for the temporally dense, systematic, transparent, and synoptic capture of land dynamics over time. Subsequent to the opening of the Landsat archive, several time series approaches for characterizing landscape change have been developed, often representing a particular analytical time window. The information richness and widespread utility of these time series data have created a need to maintain the currency of time series information via the addition of new data, as it becomes available. When an existing time series is temporally extended, it is critical that previously generated change information remains consistent, thereby not altering reported change statistics or science outcomes based on that change information. In this research, we investigate the impacts and implications of adding additional years to an existing 29-year annual Landsat time series for forest change. To do so, we undertook a spatially explicit comparison of the 29 overlapping years of a time series representing 1984–2012, with a time series representing 1984–2016. Surface reflectance values, and presence, year, and type of change were compared. We found that the addition of years to extend the time series had minimal effect on the annual surface reflectance composites, with slight band-specific differences (r ≥ 0.1) in the final years of the original time series being updated. The area of stand replacing disturbances and determination of change year are virtually unchanged for the overlapping period between the two time-series products. Over the overlapping temporal period (1984–2012), the total area of change differs by 0.53%, equating to an annual difference in change area of 0.019%. Overall, the spatial and temporal agreement of the changes detected by both time series was 96%. Further, our findings suggest that the entire pre-existing historic time series does not need to be re-processed during the update process. Critically, given the time series change detection and update approach followed here, science outcomes or reports representing one temporal epoch can be considered stable and will not be altered when a time series is updated with newly available data.     

Assessment of the abnormal growth of floating macrophytes in Winam Gulf (Kenya) by using MODIS imagery time series

The objective of this research study is to assess the capability of time-series of MODIS imagery to provide information suitable for enhancing the understanding of the temporal cycles shown by the abnormal growth of the floating macrophytes in order to support monitoring and management action of Lake Victoria water resources.The proliferation of invasive plants and aquatic weeds is of growing concern. Starting from 1989, Lake Victoria has been interested by the high infestation of water hyacinth with significant socio-economic impact on riparian populations.In this paper, we describe an approach based on the time-series of MODIS to derive the temporal behaviour, the abundance and distribution of the floating macrophytes in the Winam Gulf (Kenyan portion of the Lake Victoria) and its possible links to the concentrations of the main water constituencies.To this end, we consider the NDVI values computed from the MODIS imagery time-series from 2000 to 2009 to identify the floating macrophytes cover and an appropriate bio-optical model to retrieve, by means of an inverse procedure, the concentrations of chlorophyll a, coloured dissolved organic matter and total suspended solid.The maps of the floating vegetation based on the NDVI values allow us to assess the spatial and temporal dynamics of the weeds with high time resolution.A floating vegetation index (FVI) has been introduced for describing the weeds pollution level.The results of the analysis show a consistent temporal relation between the water constituent concentrations within the Winam Gulf and the FVI, especially in the proximity of the greatest proliferation of floating vegetation in the last 10 years that occurred between the second half of 2006 and the first half of 2007.The adopted approach will be useful to implement an automatic system for monitoring and predicting the floating macrophytes proliferation in Lake Victoria.     

Development of time series stacks of Landsat images for reconstructing forest disturbance history

Abstract

Forest dynamics is highly relevant to a broad range of earth science studies, many of which have geographic coverage ranging from regional to global scales. While the temporally dense Landsat acquisitions available in many regions provide a unique opportunity for understanding forest disturbance history dating back to 1972, large quantities of Landsat images will need to be analysed for studies at regional to global scales. This will not only require effective change detection algorithms, but also highly automated, high level preprocessing capabilities to produce images with subpixel geolocation accuracies and best achievable radiometric consistency, a status called imagery-ready-to-use (IRU). This paper describes a streamlined approach for producing IRU quality Landsat time series stacks (LTSS). This approach consists of an image selection protocol, high level preprocessing algorithms and IRU quality verification procedures. The high level preprocessing algorithms include updated radiometric calibration and atmospheric correction for calculating surface reflectance and precision registration and orthorectification routines for improving geolocation accuracy. These automated routines have been implemented in the Landsat Ecosystem Disturbance Adaptive System (LEDAPS) designed for processing large quantities of Landsat images. Some characteristics of the LTSS developed using this approach are discussed.     

A new approach for surface water change detection: Integration of pixel level image fusion and image classification techniques

Normally, to detect surface water changes, water features are extracted individually using multi-temporal satellite data, and then analyzed and compared to detect their changes. This study introduced a new approach for surface water change detection, which is based on integration of pixel level image fusion and image classification techniques. The proposed approach has the advantages of producing a pansharpened multispectral image, simultaneously highlighting the changed areas, as well as providing a high accuracy result. In doing so, various fusion techniques including Modified IHS, High Pass Filter, Gram Schmidt, and Wavelet-PC were investigated to merge the multi-temporal Landsat ETM+ 2000 and TM 2010 images to highlight the changes. The suitability of the resulting fused images for change detection was evaluated using edge detection, visual interpretation, and quantitative analysis methods. Subsequently, artificial neural network (ANN), support vector machine (SVM), and maximum likelihood (ML) classification techniques were applied to extract and map the highlighted changes. Furthermore, the applicability of the proposed approach for surface water change detection was evaluated in comparison with some common change detection methods including image differencing, principal components analysis, and post classification comparison. The results indicate that Lake Urmia lost about one third of its surface area in the period 2000–2010. The results illustrate the effectiveness of the proposed approach, especially Gram Schmidt-ANN and Gram Schmidt-SVM for surface water change detection.     

Prototyping an artificial neural network for burned area mapping on a regional scale in Mediterranean areas using MODIS images

Each year thousands of ha of forest land are affected by forest fires in Southern European countries such as Spain. Burned area maps are a valuable instrument for designing prevention and recovery policies. Remote sensing has increasingly become the most widely used tool for this purpose on regional and global scales, where a large variety of techniques and data has been applied. This paper proposes a semiautomatic method for burned area mapping on a regional scale in Mediterranean areas (the Iberian Peninsula has been used as a study case). A Multi-layer Perceptron Network (MLPN) has been designed and applied to MODIS/Terra Surface Reflectance Daily L2G Global 500m SIN Grid multitemporal composite monthly images. The compositing criterion was based on maximum surface temperature. The research covered a six year period (2001–2006) from June to September, when most of the forest fires occur. The resulting burned area maps have been validated using official fire perimeters and compared with MODIS Collection 5 Burned Area Product (MCD45A1). The MLPN shown as an effective method, with a commission error of 29.1%, in the classification of the burned areas, while the omission error was of 14.9%. The results were compared with the MCD45A1 product, which had a slightly higher commission error (30.2%) and a considerably higher omission error (26.2%), indicating a high underestimation of the burned area.     

长时间序列多源遥感数据的森林干扰监测算法研究进展

时空意义明确的森林干扰和恢复信息是评价森林生态系统碳动态的关键因素之一。然而由于诸多的现实困难,多尺度的森林干扰定量化时空信息相对缺乏。Landsat数据具备光谱、时间和空间分辨率上的优势,以及可以免费获取的特点,使其成为主要的长时间序列动态监测的遥感数据源之一,为长时间周期内提供具有合适的空间细节和时间频率的森林干扰信息成为可能。特别是基于Landsat时间序列堆栈(LTSS)的森林干扰自动分析算法的出现,更为森林生态系统的近实时监测提供强有力的工具。本文全面评述了长时间序列遥感数据准备和预处理技术以及国内外基于遥感数据源的多时相森林干扰监测方法,重点分析了基于Landsat的多种指数监测和自动化方法的优缺点,并总结了其与多源数据结合的扩展应用,最后就现有方法与国内外新的数据、技术手段的关联进行了展望,以期为推广中国本土卫星影像应用于森林干扰监测提供理论借鉴。     

遥感时间序列影像变化检测研究进展

同一区域、不同时期大量历史数据的积累,以及同一区域能够方便地获取高时间分辨率遥感数据,使遥感时间序列影像变化检测成为近年来遥感技术与应用的研究热点。本文系统总结和评述了当前遥感时间序列影像变化检测的相关研究进展和应用状况,在阐明遥感时间序列分析的意义,以及时间序列影像在变化检测中的优势的基础上,从非遥感领域时间序列变化检测方法出发,针对遥感时间序列影像变化检测的需求,明确和归纳了遥感时间序列变化检测的问题与类型,并对当前最新研究进行了综述,总结了各种方法的优点与不足,重点介绍了基于经验模态分解的遥感时间序列影像异常信息检测方法和基于隐马尔可夫模型的土地利用/覆盖变化检测方法,以期能够为相关研究提供参考。最后总结了该研究领域的发展趋势和存在问题,并对今后的研究工作和未来发展方向进行了展望。     

Land cover classification of VHR airborne images for citrus grove identification

Managing land resources using remote sensing techniques is becoming a common practice. However, data analysis procedures should satisfy the high accuracy levels demanded by users (public or private companies and governments) in order to be extensively used. This paper presents a multi-stage classification scheme to update the citrus Geographical Information System (GIS) of the Comunidad Valenciana region (Spain). Spain is the first citrus fruit producer in Europe and the fourth in the world. In particular, citrus fruits represent 67% of the agricultural production in this region, with a total production of 4.24 million tons (campaign 2006-2007). The citrus GIS inventory, created in 2001, needs to be regularly updated in order to monitor changes quickly enough, and allow appropriate policy making and citrus production forecasting. Automatic methods are proposed in this work to facilitate this update, whose processing scheme is summarized as follows. First, an object-oriented feature extraction process is carried out for each cadastral parcel from very high spatial resolution aerial images (0.5 m). Next, several automatic classifiers (decision trees, artificial neural networks, and support vector machines) are trained and combined to improve the final classification accuracy. Finally, the citrus GIS is automatically updated if a high enough level of confidence, based on the agreement between classifiers, is achieved. This is the case for 85% of the parcels and accuracy results exceed 94%. The remaining parcels are classified by expert photo-interpreters in order to guarantee the high accuracy demanded by policy makers.     

A spatiotemporal mining framework for abnormal association patterns in marine environments with a time series of remote sensing images

A spatiotemporal mining framework is a novel tool for the analysis of marine association patterns using multiple remote sensing images. From data pretreatment, to algorithm design, to association rule mining and pattern visualization, this paper outlines a spatiotemporal mining framework for abnormal association patterns in marine environments, including pixel-based and object-based mining models. Within this framework, some key issues are also addressed. In the data pretreatment phase, we propose an algorithm for extracting abnormal objects or pixels over marine surfaces, and construct a mining transaction table with object-based and pixel-based strategies. In the mining algorithm phase, a recursion method to construct a direct association pattern tree is addressed with an asymmetric mutual information table, and a recursive mining algorithm to find frequent items. In the knowledge visualization phase, a “Dimension–Attributes” visualization framework is used to display spatiotemporal association patterns. Finally, spatiotemporal association patterns for marine environmental parameters in the Pacific Ocean are identified, and the results prove the effectiveness and the efficiency of the proposed mining framework.     

The ‘Stained Glass Procedure’, a new method to compare classification performance of images acquired with different pixel sizes

Objective comparison of classification performance of earth observation images, acquired at different spatial resolutions (e.g. NOAA-AVHRR, IRS-MOS, IRS-WiFS, Landsat-TM, IRS-LISS), is complicated because both class definition and training site selection are hampered by the inherent scale differences. This paper presents a new, generic method to compare the information content of such a set of images, the “Stained Glass Procedure”. It overcomes the stated problems by computing the scale-dependent, internal spectral variation in an image and by using this as an indicator for land cover information. The Stained Glass Procedure creates segments in the images and calculates the internal spectral variation in a high-spatial-resolution image for each segment. For each image from the set the average variance, weighted to area, is calculated. The Stained Glass Procedure can be used to predict the performance of sensors that are not available, yet, or to roughly determine the optimal spatial resolution for the classification of a specific area.The procedure was applied to images with pixel sizes ranging from 23 to 1100 m. Classification detail of Envisat-MERIS (300 m pixel size), not included in the image set, could be predicted accurately using the Stained Glass Procedure.The Stained Glass Procedure applies one procedure to all images, without any subjective decision during the analysis, thus offering a method to compare images with different pixel sizes in terms of classification detail that is truly objective.     

Evaluation of different machine learning methods for land cover mapping of a Mediterranean area using multi-seasonal Landsat images and Digital Terrain Models

Land cover monitoring using digital Earth data requires robust classification methods that allow the accurate mapping of complex land cover categories. This paper discusses the crucial issues related to the application of different up-to-date machine learning classifiers: classification trees (CT), artificial neural networks (ANN), support vector machines (SVM) and random forest (RF). The analysis of the statistical significance of the differences between the performance of these algorithms, as well as sensitivity to data set size reduction and noise were also analysed. Landsat-5 Thematic Mapper data captured in European spring and summer were used with auxiliary variables derived from a digital terrain model to classify 14 different land cover categories in south Spain. Overall, statistically similar accuracies of over 91% were obtained for ANN, SVM and RF. However, the findings of this study show differences in the accuracy of the classifiers, being RF the most accurate classifier with a very simple parameterization. SVM, followed by RF, was the most robust classifier to noise and data reduction. Significant differences in their performances were only reached for thresholds of noise and data reduction greater than 20% (noise, SVM) and 25% (noise, RF), and 80% (reduction, SVM) and 50% (reduction, RF), respectively.     

Evaluation of diverse classification approaches for land use/cover mapping in a Mediterranean region utilizing Hyperion data

Information on Earth's land surface cover is commonly obtained through digital image analysis of data acquired from remote sensing sensors. In this study, we evaluated the use of diverse classification techniques in discriminating land use/cover types in a typical Mediterranean setting using Hyperion imagery. For this purpose, the spectral angle mapper (SAM), the object-based and the non-linear spectral unmixing based on artificial neural networks (ANNs) techniques were applied. A further objective had been to investigate the effect of two approaches for training sites selection in the SAM classification, namely of the pixel purity index (PPI) and of the direct selection of training points from the Hyperion imagery assisted by a QuickBird imagery and field-based training sites. Object-based classification outperformed the other techniques with an overall accuracy of 83%. Sub-pixel classification based on the ANN showed an overall accuracy of 52%, very close to that of SAM (48%). SAM applied using the training sites selected directly from the Hyperion imagery supported by the QuickBird image and the field visits returned an increase accuracy by 16%. Yet, all techniques appeared to suffer from the relatively low spatial resolution of the Hyperion imagery, which affected the spectral separation among the land use/cover classes.