Application of spectral mixture analysis to Amazonian land-use and land-cover classification

Abundant vegetation species and associated complex forest stand structures in moist tropical regions often create difficulties in accurately classifying land-use and land-cover (LULC) features. This paper examines the value of spectral mixture analysis (SMA) using Landsat Thematic Mapper (TM) data for improving LULC classification accuracy in a moist tropical area in Rondônia, Brazil. Different routines, such as constrained and unconstrained least-squares solutions, different numbers of endmembers, and minimum noise fraction transformation, were examined while implementing the SMA approach. A maximum likelihood classifier was also used to classify fraction images into seven LULC classes: mature forest, intermediate secondary succession, initial secondary succession, pasture, agricultural land, water, and bare land. The results of this study indicate that reducing correlation between image bands and using four endmembers improve classification accuracy. The overall classification accuracy was 86.6% for the seven LULC classes using the best SMA processing routine, which represents very good results for such a complex environment. The overall classification accuracy using a maximum likelihood approach was 81.4%. Another finding is that use of constrained or unconstrained solutions for unmixing the atmospherically corrected or raw Landsat TM images does not have significant influence on LULC classification performances when image endmembers are used in a SMA approach.     

基于ENVISat ASAR影像的湿地植被冠层下淹水范围提取以扎龙保护区为例

以扎龙自然保护区湿地为例,结合ENVISat ASAR多极化(HH/HV)雷达影像与传统的光学影像Landsat TM (band1~5,7),分析雷达影像后向散射系数与Landsat TM影像不同波段反射率在淹水植被、非淹水植被、明水面和裸土不同地表覆被类型的差异。选择训练样本,采用分类回归树(Classification and Regression Tree,CART)模型,分别对两种影像进行分类,可视化表达湿地植被淹水范围空间分布情况。基于实测的植被冠层下淹水范围与非淹水范围样本点对两种数据源的分类结果进行精度验证。结果表明:HH/HV极化影像中,植被覆盖下水体的后向散射系数与其他地表覆被类型有明显区别,分类结果总精度为79.49%,Kappa系数为0.70,湿地植被淹水范围提取精度较高。而TM影像分类结果中,由于部分地区植被覆盖水体,淹水植被分类误差较高。将雷达影像引入沼泽湿地研究,提高了植被淹水范围提取效果,为有效分析湿地生态水文过程提供基础,对湿地水资源合理利用及生物多样性保护具有重要意义。     

Linear spectral mixture analysis of Landsat TM data for monitoring invasive exotic plants in estuarine wetlands

This study assessed the feasibility of spectral mixture analysis (SMA) of Landsat thematic mapper (TM) data for monitoring estuarine vegetation at species level. SMA modelling was evaluated, using χ² test, by comparing SMA fraction images with a precisely classified QuickBird image that has a higher spatial resolution. To clearly understand the strengths and weaknesses of SMA, eight SMA models with different endmember combinations were assessed. When the TM data dimension for SMA and the endmember number required were balanced, a model with three endmembers representing water and two vegetation types was most accurate, whereas a model with five endmembers approximated the actual surface situation and generated a relatively accurate result. Our results indicate that an SMA model with appropriate endmembers had relatively satisfactory accuracy in monitoring vegetation. However, errors might occur in SMA fraction images, especially in models with an inappropriate endmember combination, and the errors were mainly distributed in areas filled with water or near water. Therefore, short vegetation usually submerged during high tide tended to be poorly predicted by SMA models. These results strongly suggest that tide water has a great influence on SMA modelling, especially for short vegetation.     

城市混合像元分解中土壤与不透水面纯像元选取方法的对比研究——以厦门为例

近年来混合像元分解在城市地表组分监测与分析中的应用逐渐成为城市遥感的一个热点。纯像元的选取是混合像元分解过程中的重点和关键所在。以沿海城市厦门为研究对象,根据不同的土壤和不透水面纯像元选取规则,使用2组12种不同的纯像元选取方法对2007年1月8日TM影像进行混合像元分解,对分解结果的模型适宜度进行了比较,并使用2006年12月25日SPOT5高分辨率影像对分解结果的精度进行了比较和评估。结果表明:混合像元分解在纯像元选取时,S端元选取兼顾低反射率裸土和高反射率裸露基岩的纯像元可以整体提高分解的模型适宜度和分解精度;适度提高I分量纯像元中高反射率纯像元的比例有助于改善整体尤其是S、W分量的分解效果。     

Oakwood crown closure estimation by unmixing Landsat TM data

Using an unconstrained least squares solution (LSS) method and an artificial neural network (ANN) algorithm, we estimated oakwood crown closure from a Landsat Thematic Mapper (TM) image of Tulare County, California, USA. Fractions of endmembers (oak crown (f1), grass (f2) and soil (f3)) from mixed pixels were derived from aerial photographs (scale 1?:?40?000) scanned at 1?m ground resolution for training and testing the LSS and ANN algorithms. The aerial photographs were orthorectified using a digital photogrammetric software package with ground control points collected through a differential global positioning system (GPS). The TM image was georeferenced with respect to the corresponding orthorectified aerial photographs. The training and test samples were randomly selected from the TM image and their corresponding fractions of endmembers were derived from the orthophoto. A fourth endmember, shade (f4), was directly extracted from the TM image. Experimental results indicate that the ANN has performed better than the unconstrained LSS. To extract oakwood crown closure in mixed pixels, better results were obtained without using a shade endmember.     

Application of spectral mixture analysis for terrain evaluation studies

In this article, we describe an approach to calculate the spectral mixture within pixels and classify multispectral images. The results are compared with the classified images by traditional supervised rules such as Maximum Likelihood and appreciable results were accomplished. The method considers the number of endmembers that form the scene spectra, followed by the determination of their nature and finally the decomposition of the spectra into their endmembers. The only requirement for this method is a radiometrically corrected image because the endmembers are directly selected from the image. To make the method presented here more efficient, we propose to apply it only to the classes having low accuracy after a traditional supervised classification. Because the land cover classes in this study are related to a geomorphological terrain unit, we propose to mask the terrain units having problematic classes and decompose these units into their endmembers. A geomorphological analysis of the study area (Tonle Sap basin in Cambodia) was made to establish the relationship between land cover, landforms and soils through terrain mapping units. Then we performed a supervised classification of a Landsat Thematic Mapper (TM) image and of the same image merged with a SPOT-panchromatic (PAN) image, based on the land covers corresponding to the terrain mapping units. Then we masked a terrain unit having problematic spectral classes and applied the spectral mixture analysis which allowed an efficient separation of the land cover classes agglomerated in the preliminary classification. The result of this re-classification was re-inserted into the first classification and was compared statistically with the results obtained in the preliminary classification. We consider this procedure an efficient method to improve the results obtained from a supervised classification. The method can separate different land covers that were agglomerated in the preliminary segmentation. In our case, the classification accuracy for the terrain unit used (the fluvial terrace) increases from 62% (using only the TM bands) and 69% (using TM+ SPOT) to 83%.     

Obtaining crop-specific time profiles of NDVI: the use of unmixing approaches for serving the continuity between SPOT-VGT and PROBA-V time series

The study examined the potential of two unmixing approaches for deriving crop-specific normalized difference vegetation index (NDVI) profiles so that upon availability of Project for On-Board Autonomy – Vegetation (PROBA-V) imagery in winter 2013, this new data set can be combined with existing Satellite Pour l’Observation de la Terre – VEGETATION (SPOT-VGT) data despite the differences in spatial resolution (300 m of PROBA-V versus 1 km of SPOT-VGT). To study the problem, two data sets were analysed: (1) a set of 10 temporal NDVI images, with 300 and 1000 m spatial resolution, from the state of São Paulo (Brazil) synthesized from 30 m Landsat Thematic Mapper (TM) images, and (2) a corresponding set of 10 observed Moderate Resolution Imaging Spectroradiometer (MODIS) images (250 m spatial resolution). To mimic the influence of noise on the retrieval accuracy, different sensor/atmospheric noise levels were applied to the first data set. For the unmixing analysis, a high-resolution land-cover (LC) map was used. The LC map was derived beforehand using a different set of Landsat TM images. The map distinguishes nine classes, with four different sugarcane stages, two agricultural sub-classes, plus forest, pasture, and urban/water. Unmixing aiming at the retrieval of crop-specific NDVI profiles was done at administrative level. For the synthesized data set it was demonstrated that the ‘true’ NDVI temporal profiles of different land-cover classes (from 30 m TM data) can generally be retrieved with high accuracy. The two simulated sensors (PROBA-V and SPOT-VGT) and the two unmixing algorithms gave similar results. Analysing the MODIS data set, we also found a good correspondence between the modelled NDVI profiles (both approaches) and the (true) Landsat temporal endmembers.     

Comparison of MODIS and Landsat TM5 images for mapping tempo–spatial dynamics of Secchi disk depths in Poyang Lake National Nature Reserve,China

Landsat has successfully been applied to map Secchi disk depth of inland water bodies. Operational use for monitoring a dynamic variable like Secchi disk depth is however limited by the 16‐day overpass cycle of the Landsat system and cloud cover. Low spatial resolution Moderate Resolution Imaging Spectroradiometer (MODIS) image captured twice a day could potentially overcome these problems. However, its potential for mapping Secchi disk depth of inland water bodies has so far rarely been explored. This study compared two image sources, MODIS and Landsat Thematic Mapper (TM), for mapping the tempo–spatial dynamics of Secchi disk depth in Poyang Lake National Nature Reserve, China. Secchi disk depths recorded at weekly intervals from April to October in 2004 and 2005 were related to 5 Landsat TM and 22 MODIS images respectively. Two multiple regression models including the blue and red bands of Landsat TM and MODIS respectively explained 83% and 88% of the variance of the natural logarithm of Secchi disk depth. The standard errors of the predictions were 0.20 and 0.37 m for Landsat TM and MODIS‐based models. A high correlation (r = 0.94) between the predicted Secchi disk depth derived from the two models was observed. A discussion of advantages and disadvantages of both sensors leads to the conclusion that MODIS offers the possibility to monitor water transparency more regularly and cheaply in relatively big and frequently cloud covered lakes as is with Poyang Lake.     

Identification of central Kenyan Rift Valley Fever virus vector habitats with landsat TM and evaluation of their flooding status with airborne imaging radar

Rift Valley Fever (RVF) is a mosquito-borne virus that affects livestock and humans in Africa. Landsat Thematic Mapper (TM) data are shown to be effective in identifying dambos, intermittently flooded areas that are potential mosquito breeding sites, in an area north of Nairobi, Kenya. Positive results were obtained from a limited test of flood detection in dambos with airborne high resolution L, C, and X band multipolarization synthetic aperture radar (SAR) imagery. L and C bands were effective in detecting flooded dambos, but LHH was by far the best channel for discrimination (p < 0.01) between flooded and nonflooded sites in both sedge and short grass environments. This study demonstrates the feasibility of a combined passive and active remote sensing program for monitoring the location and condition of RVF vector habitats, thus making future control of the disease more promising.     

应用Landsat TM影像估算渤海叶绿素a和总悬浮物浓度

利用23个实测样点的渤海叶绿素a和总悬浮物浓度数据及同步Landsat TM影像数据,分别分析了Landsat TM离水辐射亮度对渤海叶绿素a和总悬浮物浓度的敏感性,选择合适的波段,通过回归分析构建了基于Landsat TM离水辐射亮度的渤海叶绿素a和总悬浮物浓度反演模型。结果表明,TM1、TM2和TM3波段对叶绿素a的敏感性较高,以TM4/TM1和TM3/TM2的对数为自变量,以叶绿素a浓度的对数为因变量的线性估算模型可以有效反演渤海叶绿素a浓度,决定系数R2达到0.97;TM3波段对悬浮物的敏感性最高,以TM2、TM3和TM3/TM2为自变量,以总悬浮物浓度的以10为底的对数为因变量的多元线性模型获得的结果最佳,决定系数R2达到0.91。     

Estimating forest structure in wetlands using multitemporal SAR

Data from 202 forest plots on the Roanoke River floodplain, North Carolina were used to assess the capabilities of multitemporal radar imagery for estimating biophysical characteristics of forested wetlands. The research was designed to determine the potential for using widely available data from the current set of satellite-borne synthetic aperture radar (SAR) sensors to study forests over broad geographic areas and complex environmental gradients. The SAR data set included 11 Radarsat scenes, 2 ERS-1 images, and 1 JERS-1 scene. Empirical analyses were stratified by flood status such that sites were compared only if they exhibited common flooding characteristics. In general, the results indicate that forest properties are more accurately estimated using data from flooded areas, probably because variations in surface conditions are minimized where there is a continuous surface of standing water. Estimations yielded root mean square errors (RMSEs) for validation data around 10 m²/ha for basal area (BA), and less than 3 m for canopy height. The r² values generally exceeded .65 for BA, with the best predictions coming from sample sites for which both nonflooded and flooded SAR scenes were available. The addition of early spring normalized difference vegetation index (NDVI) values from Landsat Thematic Mapper (Landsat TM) improved model predictions for BA in forests where BA levels were <55 m²/ha. Further analyses indicated a very limited sensitivity of the individual SAR scenes to differences in forest composition, although soil properties in nonflooded areas exerted a weak but nevertheless important influence on backscatter.     

Statistical evaluation of remotely sensed data for water quality monitoring

The primary objective of this study was to determine relationships between water quality parameters (WQPs) and digital data from the Landsat satellite to estimate and map the WQP in the Porsuk Dam reservoir. Suspended sediments (SS), chlorophyll a (chl-a), NO₃-N and transmitted light intensity depth (TLID) were the parameters for water quality determination used in this study. Collection of these data, obtained from the General Directorate of State Hydraulic Works (GDSHW) was synchronized with the Landsat satellite overpass of the September 1987. The relationships between the brightness values (BV) of the TM data and WQP were determined. Using the TM data, we developed multiple regression equations to estimate the WQPs, and the validation of these equations was checked by using ANOVA. The effects of SS, NO₃-N and chl-a on TLID were tested not only for ground data, but also for TM datasets. Regression equations were developed for two different datasets and the homogeneity of those equations was tested. Finally, these regression equations evaluated from digital TM data and ground data were applied to map TLID values.     

Cross-sensor comparisons between Landsat 5 TM and IRS-P6 AWiFS and disturbance detection using integrated Landsat and AWiFS time-series images

Routine acquisition of Landsat 5 Thematic Mapper (TM) data was discontinued recently and Landsat 7 Enhanced Thematic Mapper Plus (ETM+) has an ongoing problem with the scan line corrector (SLC), thereby creating spatial gaps when covering images obtained during the process. Since temporal and spatial discontinuities of Landsat data are now imminent, it is therefore important to investigate other potential satellite data that can be used to replace Landsat data. We thus cross-compared two near-simultaneous images obtained from Landsat 5 TM and the Indian Remote Sensing (IRS)-P6 Advanced Wide Field Sensor (AWiFS), both captured on 29 May 2007 over Los Angeles, CA. TM and AWiFS reflectances were compared for the green, red, near-infrared (NIR), and shortwave infrared (SWIR) bands, as well as the normalized difference vegetation index (NDVI) based on manually selected polygons in homogeneous areas. All R ² values of linear regressions were found to be higher than 0.99. The temporally invariant cluster (TIC) method was used to calculate the NDVI correlation between the TM and AWiFS images. The NDVI regression line derived from selected polygons passed through several invariant cluster centres of the TIC density maps and demonstrated that both the scene-dependent polygon regression method and TIC method can generate accurate radiometric normalization. A scene-independent normalization method was also used to normalize the AWiFS data. Image agreement assessment demonstrated that the scene-dependent normalization using homogeneous polygons provided slightly higher accuracy values than those obtained by the scene-independent method. Finally, the non-normalized and relatively normalized ‘Landsat-like’ AWiFS 2007 images were integrated into 1984 to 2010 Landsat time-series stacks (LTSS) for disturbance detection using the Vegetation Change Tracker (VCT) model. Both scene-dependent and scene-independent normalized AWiFS data sets could generate disturbance maps similar to what were generated using the LTSS data set, and their kappa coefficients were higher than 0.97. These results indicate that AWiFS can be used instead of Landsat data to detect multitemporal disturbance in the event of Landsat data discontinuity.     

Use of Landsat TM and EOS MODIS imaging technologies for estimation of winter wheat yield in the North China Plain

This study focuses on the methodologies of winter wheat yield prediction based on Land Satellite Thematic Map (TM) and Earth Observation System Moderate Resolution Imaging Spectroradiometer (MODIS) imaging technologies in the North China Plain. Routine field measurements were initiated during the periods when the Landsat satellite passed over the study region. Five Landsat TM images were acquired. Wheat yields of the experimental sites were recorded after harvest. Spectral vegetation indices were calculated from TM and MODIS images. The correlation analysis among wheat yield and spectral parameters revealed that TM renormalized difference water index (RDWI) and MODIS near-infrared reflectance had the highest correlation with yield at grain-filling stages. The models from the best-fitting method were used to estimate wheat yield based on TM and MODIS data. The average relative error of the root mean square error (RMSE) of the predicted yield was smaller from TM than from MODIS.     

Wetland change detection on the Kafue Flats,Zambia, by classification of a multitemporal remote sensing image dataset

The paper describes a remote sensing change detection approach used to assess change on a section of the Kafue Flats floodplain wetland system in southern Zambia, which is under the pressures of reduced regional rainfall and damming and water abstraction by man. Four images from September 1984 (Landsat MSS), 1988 (Landsat MSS), 1991 (Landsat TM) and 1994 (Landsat TM) were used. Being near-anniversary images, the change detection error introduced by mere seasonal differences was minimized. Following atmospheric correction of the reference (1994) image, the images were radiometrically normalized and geometrically registered to a common map projection. Each image was separately classified into categories of open water, dense green vegetation, sparse green vegetation, very sparse green vegetation, dry and burnt land. Similar, supervised maximum likelihood classification procedures were employed on all images. The classified images produced were analysed for change in each land-cover category by overlaying them in a Geographic Information System (GIS) framework. The results indicated spatial reduction in area of dense green vegetation in upstream sections of the wetland. Inter-image changes in this land-cover class could be explained by the variations in the timing of regulated flood events on the Kafue Flats. The methodology employed appears to be applicable to monitoring southern Africa's inland wetland systems.     

Mapping North African landforms using continental scale unmixing of MODIS imagery

We describe the production of a landform map of North Africa utilizing moderate resolution satellite imagery and a methodology that is applicable for sub-continental to global scale landform mapping. A mosaic of Moderate Resolution Imaging Spectroradiometer (MODIS) apparent surface reflectance imagery was compiled for Africa north of 10° N. Landform image endmembers were chosen to characterize ten different types of vegetated and unvegetated desert surfaces: alluvial complexes, dunes, dry and ephemeral lakes, open water, basaltic volcanoes and flows, mountains, regs, stripped, low-angle bedrock surfaces, sand sheets, and Sahelian vegetation. Multiple Endmember Spectral Mixture Analysis (MESMA) was applied to the MODIS mosaic to estimate landform and vegetation endmember fractions. The major landform in each MODIS pixel was identified based on the majority endmember fraction in two- or three-endmember models. Accuracy assessment was conducted using two data sources: the historic Landform Map of North Africa [Raisz, E. (1952). Landform Map of North Africa. Environmental Protection Branch, Office of the Quartermaster General.] and Landsat Thematic Mapper (TM) data. Comparison with the Raisz landform map gave an overall classification accuracy of 54% with significant confusion between alluvial surfaces and regs, and between sandy and clayey surfaces and dunes. A second validation using 20 Landsat images in a stratified sampling scheme gave a classification accuracy of 70%, with confusion between dunes and sand sheets. Both accuracy assessment schemes indicated difficulty in vegetation classification at the margin of the Sahel. A comparison with minimum distance and maximum likelihood supervised classifications found that the MESMA approach produced significantly higher classification accuracies. This digital landform map is of sufficiently high quality to form the basis for geomorphic studies, including parameterization of the surface in global and regional dust models.     

Spectral unmixing of Landsat Thematic Mapper data

The observed spectral signature of pixels in remote sensing imagery in most cases is the result of the reflecting properties of a number of surface materials constituting the area of a pixel. Despite this knowledge most image classification techniques aim at labelling a pixel according to a singular surface category. An alternative product can be generated using spectral unmixing: a technique that strives to find the surface abundances of a number of spectral components together causing the observed spectral reflectance at a pixel. A stepwise approach to implement spectral unmixing in Landsat Thematic Mapper image analysis is proposed: (1) atmospheric calibration of the image data, (2) preselection of a large number of ‘candidate’ endmembers, (3) reduction to the most important spectral endmembers using spectral angle mapping, (4) finding the relative abundances of the endmembers through spectral unmixing analysis, (5) combining the abundance estimates into a final product comparable to a classified image, and (6) accuracy assessment. A Landsat Thematic Mapper image from southern Spain covering a large peridotite body with adjacent limestone and low-grade metamorphic rocks is used as an example to demonstrate the usefulness of unmixing.     

The effects of endmember selection on modelling impervious surfaces using spectral mixture analysis: a case study in Sydney,Australia

In urban areas, spectral mixture analysis (SMA) is a common technique for deriving the fractions of land covers within a pixel and information on the distribution of impervious surfaces. This study examined how the selection of endmembers affected the quantification of impervious surfaces using TM and ASTER imagery. Multiple subsets of endmembers derived using (1) extreme pixels from a minimum noise fraction (MNF) transformation, and (2) a manual approach using a priori knowledge of the study area were analysed. Two data sets were used to assess accuracy: (1) simulated image data comprising unmixed and mixed pixels of 10 typical and spectrally different urban land covers, and (2) detailed data derived from high-resolution aerial photography. The dimensionality of the imagery limited the number of endmembers, and as a result, unmixed land covers were modelled using multiple endmembers and some cells had abundance values that summed to more than one or were negative. The land covers of red roofs and concrete were the largest contributors to the error in impervious surfaces. The Sequential Maximum Angle Convex Cone (SMACC) endmember model was also used to unmix the images; however, the larger number of endmembers did not resolve the use of multiple endmembers to model the unmixed land covers and the accuracy was similar to that using SMA. The relationship between the pervious fraction estimated using the vegetation endmember and the ground reference data was stronger than that for the impervious fraction, although the fraction was underestimated. The problems in modelling highly variable impervious surfaces with a limited number of endmembers suggest that in urban environments with substantial vegetation, modelling the vegetation component as the inverse of the impervious fraction may lead to improved results.     

Development of principal component analysis applied to multitemporal Landsat TM data

Abstract

A method of extracting different crop information more effectively, by applying principal component analysis to a three-dimensional three-date Landsat TM data space, has been developed. Landsat TM data obtained on 7 June, 9 July and 26 August 1984 were used to examine the approach, and the results fully demonstrated its advantages over other methods of keeping classification accuracy without any calibrations for multitemporal Landsat data.     

Deriving land surface temperature from Landsat 5 and 7 during SMEX02/SMACEX

A sequence of five high-resolution satellite-based land surface temperature (T_s) images over a watershed area in Iowa were analyzed. As a part of the SMEX02 field experiment, these land surface temperature images were extracted from Landsat 5 Thematic Mapper (TM) and Landsat 7 Enhanced Thematic Mapper (ETM) thermal bands. The radiative transfer model MODTRAN 4.1 was used with atmospheric profile data to atmospherically correct the Landsat data. NDVI derived from Landsat visible and near-infrared bands was used to estimate fractional vegetation cover, which in turn was used to estimate emissivity for Landsat thermal bands. The estimated brightness temperature was compared with concurrent tower based measurements. The mean absolute difference (MAD) between the satellite-based brightness temperature estimates and the tower based brightness temperature was 0.98 °C for Landsat 7 and 1.47 °C for Landsat 5, respectively. Based on these images, the land surface temperature spatial variation and its change with scale are addressed. The scaling properties of the surface temperature are important as they have significant implications for changes in land surface flux estimation between higher-resolution Landsat and regional to global sensors such as MODIS.