Evaluation of gap-filling methods for Landsat 7 ETM+ SLC-off image for LULC classification in a heterogeneous landscape of West Africa

ABSTRACT

The Landsat mission which has existed over five decades has remained at the forefront of providing consistent moderate spatial and temporal resolution optical images of the earth. The failure of the scan line corrector (SLC) on board the Landsat 7 Enhanced Thematic Mapper Plus (ETM+) in May 2003 has permanently resulted in data gaps on each Landsat 7 scene. Due to the obvious negative impacts on the image usability, a number of methods have been developed to fill the no-data areas in the image. This study assessed the performance of four Landsat 7 ETM+ SLC-off gap-filling methods in a highly heterogeneous landscape of West Africa for two different seasons (dry and rainy). The methods considered are: (1) Weighted Linear Regression (WLR) integrated with Laplacian Prior Regularization Method (LPRM), (2) Localised Linear Histogram Matching (LLHM), (3) Neighbourhood Similar Pixel Interpolator (NSPI) and (4) Geostatistical Neighbourhood Similar Pixel Interpolator (GNSPI). All the images used were Landsat 7 ETM+ SLC-off images, temporally close and from the same season for each set of time step. Visual comparison, mean, and standard deviations of the histograms of all bands of only the filled areas were used to assess the results. Additionally, overall accuracy (OA), kappa coefficient (κ), and balanced accuracy (BA) per class were used to evaluate a land use/cover (LULC) classification based on the gap-filled images. Visually, all the four methods were able to completely fill the gaps in the Landsat 7 ETM+ SLC-off image. They all look similar and spatially continuous with no anomalies or artefacts on them. The histograms from each band for only the filled areas for all the four methods also gave similar means and standard deviations in most cases. All the four gap-filling methods provided satisfactory results (OA >96% and κ> 0.937 in all methods for images in the dry season and OA >93% and κ> 0.877 for the image in the rainy season) in the land cover classification considering the complexity of the study area. But the GNSPI was superiority in all cases with the highest OA of 97.1% and κ of 0.947 in the dry season and OA of 94.6% and κ of 0.899 in the rainy season. This implies that the GNSPI is more robust in gap filling of Landsat 7 ETM+ SLC-off images than the other three methods in a heterogeneous landscape of West Africa regardless of the season. This study suggests that gap filling of Landsat 7 ETM+ SLC-off images will help to increase the number of Landsat images needed to build time-series data for a data-scarce region such as West Africa.     

A comparison of gap-filling approaches for Landsat-7 satellite data

The purpose of this study is to assess the relative performance of four different gap-filling approaches across a range of land-surface conditions, including both homogeneous and heterogeneous areas as well as in scenes with abrupt changes in landscape elements. The techniques considered in this study include: (1) Kriging and co-Kriging; (2) geostatistical neighbourhood similar pixel interpolator (GNSPI); (3) a weighted linear regression (WLR) algorithm; and (4) the direct sampling (DS) method. To examine the impact of image availability and the influence of temporal distance on the selection of input training data (i.e. time separating the training data from the gap-filled target image), input images acquired within the same season (temporally close) as well as in different seasons (temporally far) to the target image were examined, as was the case of using information only within the target image itself. Root mean square error (RMSE), mean spectral angle (MSA), and coefficient of determination (R²) were used as the evaluation metrics to assess the prediction results. In addition, the overall accuracy (OA) and kappa coefficient (kappa) were used to assess a land-cover classification based on the gap-filled images. Results show that all of the gap-filling approaches provide satisfactory results for the homogeneous case, with R² > 0.93 for bands 1 and 2 in all cases and R² > 0.80 for bands 3 and 4 in most cases. For the heterogeneous example, GNSPI performs the best, with R² > 0.85 for all tested cases. WLR and GNSPI exhibit equivalent accuracy when a temporally close input image is used (i.e. WLR and GNSPI both have an R² equal to 0.89 for band 1). For the case of abrupt changes in scene elements or in the absence of ancillary data, the DS approach outperforms the other tested methods.     

Comparison of data gap-filling methods for Landsat ETM+ SLC-off imagery for monitoring forest degradation in a semi-deciduous tropical forest in Mexico

A number of methods to overcome the 2003 failure of the Landsat 7 Enhanced Thematic Mapper (ETM+) scan-line corrector (SLC) are compared in this article in a forest-monitoring application in the Yucatan Peninsula, Mexico. The objective of this comparison is to determine the best approach to accomplish SLC-off image gap-filling for the particular landscape in this region, and thereby provide continuity in the Landsat data sensor archive for forest-monitoring purposes. Four methods were tested: (1) local linear histogram matching (LLHM); (2) neighbourhood similar pixel interpolator (NSPI); (3) geostatistical neighbourhood similar pixel interpolator (GNSPI); and (4) weighted linear regression (WLR). All methods generated reasonable SLC-off gap-filling data that were visually consistent and could be employed in subsequent digital image analysis. Overall accuracy, kappa coefficients (κ), and quantity and allocation disagreement indices were used to evaluate unsupervised Iterative Self-Organizing Data Analysis (ISODATA) land-cover classification maps. In addition, Pearson correlation coefficients (r) and root mean squares of the error (RMSEs) were employed for estimates agreement with fractional land cover. The best results visually (overall accuracy > 85%, κ < 9%, quantity disagreement index < 5.5%, and allocation disagreement index < 12.5%) and statistically (r > 0.84 and RMSE < 7%) were obtained from the GNSPI method. These results suggest that the GNSPI method is suitable for routine use in reconstructing the imagery stack of Landsat ETM+ SLC-off gap-filled data for use in forest-monitoring applications in this type of heterogeneous landscape.     

Assessment of atmospheric correction methods for Landsat TM data applicable to Amazon basin LBA research

Atmospheric correction is an important preprocessing step required in many remote sensing applications. The authors are engaged in the project 'Human Dimensions of Amazonia: Forest Regeneration and Landscape Structure' in NASA/INPE's Large Scale Biosphere-Atmosphere Experiment in Amazonia (LBA) programme. This project requires use of corrected Landsat TM data since research foci integrate ground-based data and TM to: (1) measure and model biomass; (2) classify multiple stages of secondary succession; (3) model land cover/land use changes; and (4) derive spectral signatures consistent across different study areas. The 30+ scenes of TM data are historic and lack detailed atmospheric data needed by physically-based atmospheric correction models such as 6S (Second Simulation of the Satellite Signal in the Solar Spectrum). Imagebased DOS models are based on image measurements and explored in this article for application to LBA study areas. Two methods using theoretical spectral radiance and image acquisition date respectively were used to convert TM DN values to at-satellite radiance. Three image-based models were employed using each method to convert at-satellite radiance to surface reflectance. Analyses of these six different image-based models were conducted. The Improved Imagebased DOS was the best technique for correcting atmospheric effects in this LBA research with results similar to those obtained from physically-based approaches.     

National Park vegetation mapping using multitemporal Landsat 7 data and a decision tree classifier

Decision tree classifiers have received much recent attention, particularly with regards to land cover classifications at continental to global scales. Despite their many benefits and general flexibility, the use of decision trees with high spatial resolution data has not yet been fully explored. In support of the National Park Service (NPS) Vegetation Mapping Program (VMP), we have examined the feasibility of using a commercially available decision tree classifier with multitemporal satellite data from the Enhanced Thematic Mapper-Plus (ETM+) instrument to map 11 land cover types at the Delaware Water Gap National Recreation Area near Milford, PA. Ensemble techniques such as boosting and consensus filtering of the training data were used to improve both the quality of the input training data as well as the final products.Using land cover classes as specified by the National Vegetation Classification Standard at the Formation level, the final land cover map has an overall accuracy of 82% (κ=0.80) when tested against a validation data set acquired on the ground (n=195). This same accuracy is 99.5% when considering only forest vs. nonforest classes. Usage of ETM+ scenes acquired at multiple dates improves the accuracy over the use of a single date, particularly for the different forest types. These results demonstrate the potential applicability and usability of such an approach to the entire National Park system, and to high spatial resolution land cover and forest mapping applications in general.     

Landsat7的自动云量评估

介绍了Landsat7的一种自动云量评估（Automatic Cloud Cover Assessment,简称ACCA）算法。该算法通过对每一景ETM＋图像数据的两遍扫描进行云量评估。第一遍扫描,利用景中地物的反射率和温度特性与判断该景是否有云,若有云,则建立一个可靠的云罩;第二遍扫描,利用云的温度特性来识别该景中剩余的云。     

Comparing ten classification methods for burned area mapping in a Mediterranean environment using Landsat TM satellite data

Various methods have been developed during the past three decades to improve the classification accuracy in burned area mapping using satellite data captured by different sensors. In this article, we compare ten such classification approaches using Landsat Thematic Mapper (TM) imagery on three Mediterranean test sites by evaluating the classification accuracy using (i) a traditional pixel-based approach, (ii) the concept of the Pareto boundary of efficient solution and (iii) linear regression analysis. Additionally, we make a discrimination of errors depending on their distribution and causal factor. The classification approaches compared resulted in not statistically significant differences in the accuracy of the burned area maps. Differences between the methods were also observed when considering the accuracy along the edges of the burned patches; however, again these were not statistically significant. The findings of our study in a Mediterranean environment clearly demonstrate that, for the selection of the most suitable classification approach, other factors could be given more weight, such as computational resources, imagery characteristics, availability of ancillary data, available software and the analyst's experience. Maybe the most important finding of our work is that the variance imposed by the methods is less than the variance imposed by factors differentiated locally in the three study sites since the between-group variance of the overall accuracy is higher than that of the within groups.     

Comparison of automatic thresholding methods for snow-cover mapping using Landsat TM imagery

Knowledge of snow cover is essential to understanding the global water and energy cycle. Thresholding the normalized difference snow index (NDSI) image is a method frequently used to map snow cover from remotely sensed data. However, the threshold is dependent on the scenario and needs to be determined accordingly. In this study, nine automatic thresholding methods were tested on the NDSI. Comparisons of the automatic thresholding methods, optimal threshold, and support vector machine (SVM) classification show that Otsu's and Nie's methods appear to be the most robust among the nine automatic thresholding methods, achieving comparable accuracies with the latter two approaches. In addition, NDSI from the digital number (DN) can be an efficient substitution for NDSI obtained from atmospherically or topographically corrected data, with similar accuracy.     

美国陆地卫星7号的数据产品分类和格式

扼要地介绍美国陆地卫星７号数据产品的分类及所采用的数据格式，详细地描述了各种格式的内部结构。     

Evaluating MODIS data for mapping wildlife habitat distribution

Habitat distribution models have a long history in ecological research. With the development of geospatial information technology, including remote sensing, these models are now applied to an ever-increasing number of species, particularly those located in areas in which it is logistically difficult to collect habitat data in the field. Many habitat studies have used data acquired by multi-spectral sensor systems such as the Landsat Thematic Mapper (TM), due mostly to their availability and relatively high spatial resolution (30 m/pixel). The use of data collected by other sensor systems with lower spatial resolutions but high frequency of acquisitions has largely been neglected, due to the perception that such low spatial resolution data are too coarse for habitat mapping. In this study we compare two models using data from different satellite sensor systems for mapping the spatial distribution of giant panda habitat in Wolong Nature Reserve, China. The first one is a four-category scheme model based on combining forest cover (derived from a digital land cover classification of Landsat TM imagery acquired in June, 2001) with information on elevation and slope (derived from a digital elevation model obtained from topographic maps of the study area). The second model is based on the Ecological Niche Factor Analysis (ENFA) of a time series of weekly composites of WDRVI (Wide Dynamic Range Vegetation Index) images derived from MODIS (Moderate Resolution Imaging Spectroradiometer – 250 m/pixel) for 2001. A series of field plots was established in the reserve during the summer–autumn months of 2001–2003. The locations of the plots with panda feces were used to calibrate the ENFA model and to validate the results of both models. Results showed that the model using the seasonal variability of MODIS-WDRVI had a similar prediction success to that using Landsat TM and digital elevation model data, albeit having a coarser spatial resolution. This suggests that the phenological characterization of the land surface provides an appropriate environmental predictor for giant panda habitat mapping. Therefore, the information contained in remotely sensed data acquired with low spatial resolution but high frequency of acquisitions has considerable potential for mapping the habitat distribution of wildlife species.     

Regional-scale boreal forest cover and change mapping using Landsat data composites for European Russia

Boreal forests are a critical component of the global carbon cycle, and timely monitoring allows for assessing forest cover change and its impacts on carbon dynamics. Earth observation data sets are an important source of information that allow for systematic monitoring of the entire biome. Landsat imagery, provided free of charge by the USGS Center for Earth Resources Observation and Science (EROS) enable consistent and timely forest cover updates. However, irregular image acquisition within parts of the boreal biome coupled with an absence of atmospherically corrected data hamper regional-scale monitoring efforts using Landsat imagery. A method of boreal forest cover and change mapping using Landsat imagery has been developed and tested within European Russia between circa year 2000 and 2005. The approach employs a multi-year compositing methodology adapted for incomplete annual data availability, within-region variation in growing season length and frequent cloud cover. Relative radiometric normalization and cloud/shadow data screening algorithms were employed to create seamless image composites with remaining cloud/shadow contamination of less than 0.5% of the total composite area. Supervised classification tree algorithms were applied to the time-sequential image composites to characterize forest cover and gross forest loss over the study period. Forest cover results when compared to independently-derived samples of Landsat data have high agreement (overall accuracy of 89%, Kappa of 0.78), and conform with official forest cover statistics of the Russian government. Gross forest cover loss regional-scale mapping results are comparable with individual Landsat image pair change detection (overall accuracy of 98%, Kappa of 0.71). The gross forest cover loss within European Russia 2000-2005 is estimated to be 2210 thousand hectares, and constitutes a 1.5% reduction of year 2000 forest cover. At the regional scale, the highest proportional forest cover loss is estimated for the most populated regions (Leningradskaya and Moskovskaya Oblast). Our results highlight the forest cover depletion around large industrial cities as the hotspot of forest cover change in European Russia.     

Spectral mixture analysis for bi-sensor wetland mapping using Landsat TM and Terra MODIS data

Spatial and temporal resolution is essential for understanding the spatial and temporal characteristics and dynamics of wetland ecosystems. However, single satellite imagery with both high spatial resolution and high temporal frequency is currently unavailable. Instead, the development of a bi-sensor monitoring technique utilizing spatial details of middle-to-high resolution data and temporal details of coarse spatial resolution data is highly desirable. For the initial work on our time-series bi-sensor wetland mapping, the applicability of multiple endmember spectral mixture analysis (MESMA) using single-date bi-sensor imagery with different orbiting periods was investigated. Landsat-5 Thematic Mapper (TM) and Terra Moderate Resolution Image Spectrometer (MODIS) data were utilized in the Poyang Lake area in China and the Great Salt Lake area in the USA to examine three decisive elements in utilizing MESMA: (1) the method of optimal endmember selection; (2) the threshold between two- and three-endmember models; and (3) the treatment of shade fractions. As a result, we found that (1) the number of spectra for an endmember spectrum similar to other endmember spectra meeting the modelling restrictions of maximum and minimum land-cover fractions and root mean square error (RMSE) within a class (In_CoB), the number of spectra for an endmember spectrum similar to other endmember spectra meeting the modelling restrictions outside of a class (Out_CoB), the ratio of In_CoB to Out_CoB multiplied by the inverse number of spectra within the class (CoBI) and the endmember average RMSE (EAR) were optimal endmember selection methods for the TM maps, whereas CoBI, EAR and minimum average spectral angle (MASA) were optimal endmember selection methods for the MODIS maps; (2) the MODIS maps were more sensitive to change in the two- and three-endmember modelling thresholds than the TM maps; and (3) the addition of shade fractions to dark water fractions were an appropriate shade treatment. This research demonstrated how MESMA can be applied for multi-scale mapping of wetland ecosystems, how the difference in observation dates between the TM and MODIS data affects the agreement in land-cover fractions and how spectral similarity between dark water and shade affects the agreement in land-cover fractions.     

Comparison of regression and geostatistical methods for mapping Leaf Area Index (LAI) with Landsat ETM+ data over a boreal forest

This study compared aspatial and spatial methods of using remote sensing and field data to predict maximum growing season leaf area index (LAI) maps in a boreal forest in Manitoba, Canada. The methods tested were orthogonal regression analysis (reduced major axis, RMA) and two geostatistical techniques: kriging with an external drift (KED) and sequential Gaussian conditional simulation (SGCS). Deterministic methods such as RMA and KED provide a single predicted map with either aspatial (e.g., standard error, in regression techniques) or limited spatial (e.g., KED variance) assessments of errors, respectively. In contrast, SGCS takes a probabilistic approach, where simulated values are conditional on the sample values and preserve the sample statistics. In this application, canonical indices were used to maximize the ability of Landsat ETM+ spectral data to account for LAI variability measured in the field through a spatially nested sampling design. As expected based on theory, SGCS did the best job preserving the distribution of measured LAI values. In terms of spatial pattern, SGCS preserved the anisotropy observed in semivariograms of measured LAI, while KED reduced anisotropy and lowered global variance (i.e., lower sill), also consistent with theory. The conditional variance of multiple SGCS realizations provided a useful visual and quantitative measure of spatial uncertainty. For applications requiring spatial prediction methods, we concluded KED is more useful if local accuracy is important, but SGCS is better for indicating global pattern. Predicting LAI from satellite data using geostatistical methods requires a distribution and density of primary, reference LAI measurements that are impractical to obtain. For regional NPP modeling with coarse resolution inputs, the aspatial RMA regression method is the most practical option.     

Assessment of the distribution of aerosols in the area of Athens with the use of Landsat Thematic Mapper data

Landsat high resolution images, at the spectral interval 0.45-0.52 mu m, are used in an effort to qualitatively assess distribution of aerosols in the greater Athens area. It is found that the spatial distribution of aerosols in Athens can be described by calculating the differences in optical density between two images, the one referring a pollution-free day and the second reflecting a day with an atmospheric pollution episode. Results are supported by a positive correlation between the differences in optical densities for the 'polluted' and 'pollution free' images, and the measured levels of sulphur dioxide and smoke in various city areas. Finally according to the above methodology, aerosols exhibit their highest concentration mostly in the industrialized region of Athens, to the west of the city's centre, as well as in a limited area in the centre of Athens.     

Mineral imaging with Landsat Thematic Mapper data for hydrothermal alteration mapping in heavily vegetated terrane

A mineral imaging methodology, which involves processing of Landsat Thematic Mapper (TM) images and integration of ground data, is tested in the Baguio district of the Philippines to map hydrothermally altered zones in heavily vegetated terranes. Based on published reflectance spectra, two band ratio images are created and input into principal components analysis to map each predominant hydrothermal alteration mineral into separate mineral images. Digitized map data of known hydrothermal alteration zones are used for identifying training pixels for the known alteration zones. The mineral images and the training pixels are used in a supervised classification to map hydrothermally altered zones; classification accuracy reaches 69%. Inclusion of an image of a digital elevation model improves the classification accuracy to 82%. The mineral imaging methodology proved more successful in remote mapping of known hydrothermally altered zones in the Baguio district than remote mapping of limonitic and clay alteration using previously developed techniques.     

Accuracy of forest mapping based on Landsat TM data and a kNN-based method

A multi-source forest inventory (MSFI) method has been developed for use in the Norwegian National Forest Inventory (NFI). The method is based on a k-nearest neighbour rule and uses field plots from the NFI, land cover maps, and satellite image data from Landsat Thematic Mapper. The inventory method is used to produce maps of selected forest variables and to estimate the selected forest variables for large areas such as municipalities. In this study, focus has been on the qualitative variables ‘dominating species group’ and ‘development class’ because these variables are of central interest to forest managers. A mid-summer Landsat 5 TM scene was used as image data, and all NFI plots inside the scene were used as a reference dataset. The relationship between the spectral bands and the forest variables was analysed, and it was found that the levels of association were low. A leave-one-out method based on the reference dataset was used to estimate the pixel-level accuracies. They were found to be relatively low with 63% agreement for species groups. An independent control survey was available for a municipality and estimates from the MSFI were compared to it. The levels of error were quite high. It was concluded that the large area estimates were biased by the reference dataset.     

Landsat 8 OLI satellite data for mapping of the Posidonia oceanica and benthic habitats of coastal ecosystems

The benthic seabeds and seagrass ecosystems, in particular the vulnerable Posidonia oceanica (PO), are increasingly threatened by climate change and other anthropogenic pressures. Along the 8000 km coastline of Italy, they are often poorly mapped and monitored to properly evaluate their health status. Thus to support these monitoring needs, the improved capabilities of the Landsat 8 Operational Land Imager (OLI) Earth Observation (EO) satellite system were tested for PO mapping by coupling its atmospherically corrected multispectral data with near-synchronous sea truth information. Two different approaches for the necessary atmospheric correction were exploited focusing on the Aerosol Optical Depth (AOD) and adjacency noise effects, which typically occur at land–sea interfaces. The general achievements demonstrated the effectiveness of High Resolution (HR) spectral responses captured by OLI sensor, for monitoring seagrass and sea beds in the optically complex Tyrrhenian shallow waters, with performance level dependent on the type of applied atmospheric pre-processing. The distribution of the PO leaf area index (LAI) on different substrates has been most effectively modelled using on purpose developed spectral indices. They were based on the coastal and blue-green OLI bands, atmospherically corrected using a recently introduced method for AOD retrieval, based on the Short Wave Infrared (SWIR) reflectance. The alternative correction method including a less effective AOD assessment but the removal of adjacency effects has proven its efficacy for improving the thematic discriminability of the seabed types characterized by different PO cover–substrate combinations.     

Vegetation water content mapping using Landsat data derived normalized difference water index for corn and soybeans

Information about vegetation water content (VWC) has widespread utility in agriculture, forestry, and hydrology. It is also useful in retrieving soil moisture from microwave remote sensing observations. Providing a VWC estimate allows us to control a degree of freedom in the soil moisture retrieval process. However, these must be available in a timely fashion in order to be of value to routine applications, especially soil moisture retrieval. As part of the Soil Moisture Experiments 2002 (SMEX02), the potential of using satellite spectral reflectance measurements to map and monitor VWC for corn and soybean canopies was evaluated. Landsat Thematic Mapper and Enhanced Thematic Mapper Plus data and ground-based VWC measurements were used to establish relationships based on remotely sensed indices. The two indices studied were the Normalized Difference Vegetation Index (NDVI) and the Normalized Difference Water Index (NDWI). The NDVI saturated during the study period while the NDWI continued to reflect changes in VWC. NDWI was found to be superior based upon a quantitative analysis of bias and standard error. The method developed was used to map daily VWC for the watershed over the 1-month experiment period. It was also extended to a larger regional domain. In order to develop more robust and operational methods, we need to look at how we can utilize the MODIS instruments on the Terra and Aqua platforms, which can provide daily temporal coverage.     

Improving tropical forest mapping using multi-date Landsat TM data and pre-classification image smoothing

The present study explores the possibility of using Landsat imagery for mapping tropical forest types with relevance to forest ecosystem services. The central part in the classification process is the use of multi-date image data and pre-classification image smoothing. The study argues that multi-date imagery contains information on phenological and canopy structural properties, and shows how the use of multi-date imagery has a significant impact on classification accuracy. Furthermore, the study shows the value of applying small kernel smoothing filters to reduce in-class spectral variability and enhance between-class spectral separability. Making use of these approaches and a maximum likelihood algorithm, six tropical forest types were classified with an overall accuracy of 90.94%, and with individual forest classes mapped with accuracies above 75.19% (user's accuracy) and above 74.17% (producer's accuracy).     

Chlorophyll distribution in Lake Kinneret determined from Landsat Thematic Mapper data

Chlorophyll distribution in Lake Kinneret was estimated at a time of low chlorophyll concentrations (3-7 mgm^?3). Landsat Thematic Mapper (TM) data were acquired three days after the acquisition of high spectral resolution radiometric measurements in the range 400 to 750 nm, chlorophyll and suspended matter concentrations, and Secchi disk transparency at 22 stations. The radiometric data were used to create an algorithm for estimation of chlorophyll concentration from the TM data. The radiance in channel TM3 (620-690 nm) was primarily dependent upon non-organic suspended matter concentration. Radiance in this channel was substracted from radiance in TM1 (450-520 nm) to correct for the additional radiance caused by scattering of non-pigmented suspended particles and (TM1 – TM3)/TM2 was found to be a useful index for estimating chlorophyll concentration. The concentrations calculated from atmospherically corrected TM data were compared to chlorophyll extracted from lake water samples. The estimation error of chlorophyll concentration was less than 0.85 mgm^?3.