Spectral Mixture Analysis Research Articles

Impervious Surface Extraction by Linear Spectral Mixture Analysis with Post-Processing Model

Accurate estimations of impervious surface areas are essential for urban planning development. Linear spectral mixture analysis (LSMA) is commonly adopted to extract the impervious surface (IS) fraction in a mixed pixel at the subpixel scale. However, owing to errors in the spectra of pure pixels selected from remote sensing images, incorrect fractions of different land cover types often emerge after unmixing. In this study, two Landsat 8 Operational Land Imager (OLI) images—acquired on 20 September 2019 (Path/Row: 121/44) and 14 November 2019 (Path/Row: 122/44)—of Guangzhou and Shenzhen were unmixed by LSMA using spectral indices in endmember selection. A post-processing model using the Dry Bare-soil Index (DBSI) and Normalized Difference Vegetation Index (NDVI) as thresholds was established to improve the IS fraction of the LSMA result. Comparative analysis reveals that LSMA with the post-processing model achieves better performance for IS fraction extraction (R2 = 0.910 and 0.926 and root mean square error [RMSE] = 10.08% and 10.83% for Guangzhou and Shenzhen, respectively), and the distribution of IS is basically consistent with the IS of the actual areas. The post-processing model solves the problem of overestimation of pervious surface and underestimation of impervious surface.

Comparison of Different Multispectral Sensors for Photosynthetic and Non-Photosynthetic Vegetation-Fraction Retrieval

It is very difficult and complex to acquire photosynthetic vegetation (PV) and non-PV (NPV) fractions (fPV and fNPV) using multispectral satellite sensors because estimations of fPV and fNPV are influenced by many factors, such as background-noise interference of pixel-, spatial-, and spectral-scale effects. In this study, comparisons between Sentinel-2A Multispectral Instrument (S2 MSI), Landsat-8 Operational Land Imager (L8 OLI), and GF1 Wide Field View (GF1 WFV) sensors for retrieving sparse photosynthetic and non-photosynthetic vegetation coverage are presented. The analysis employed a linear spectral-mixture model (LSMM) and nonlinear spectral-mixture model (NSMM) to unmix pixels with different spectral and spatial resolution images based on field endmembers; the estimated endmember fractions were later validated with reference to fraction measurements. The results demonstrated that: (1) with higher spatial and spectral resolution, the S2 MSI sensor had a clear advantage for retrieving PV and NPV fractions compared to L8 OLI and GF1 WFV sensors; (2) through incorporating more red edge (RE) and near-infrared (NIR) bands, the accuracy of NPV fraction estimation could be greatly improved; (3) nonlinear spectral mixing effects were not obvious on the 10–30 m spatial scale for desert vegetation; (4) in arid regions, a shadow endmember is a significant factor for sparse vegetation coverage estimated with remote-sensing data. The estimated NPV fractions were especially affected by the shadow effects and could increase root mean square by 50%. The utilized approaches in the study could effectively assess the performance of major multispectral sensors to extract fPV and fNPV through the novel method of spectral-mixture analysis.

Fractional Vegetation Cover Change Detection In Megacities Using Landsat Time-Series Images: A Case Study Of Hanoi City (Vietnam) During 1986-2019

The objective of the study is to assess changes of fractional vegetation cover (FVC) in Hanoi megacity in period of 33 years from 1986 to 2016 based on a two endmember spectral mixture analysis (SMA) model using multi-spectral and multi-temporal Landsat-5 TM and -8 OLI images. Landsat TM/OLI images were first radiometrically corrected. FVC was then estimated by means of a combination of Normalized Difference Vegetation Index (NDVI) and classification method. The estimated FVC results were validated using the field survey data. The assessment of FVC changes was finally carried out using spatial analysis in GIS. A case study from Hanoi city shows that: (i) the proposed approach performed well in estimating the FVC retrieved from the Landsat-8 OLI data and had good consistency with in situ measurements with the statistically achieved root mean square error (RMSE) of 0.02 (R 2 =0.935); (ii) total FVC area of 321.6 km 2 (accounting for 9.61% of the total area) was slightly reduced in the center of the city, whereas, FVC increased markedly with an area of 1163.6 km 2 (accounting for 34.78% of the total area) in suburban and rural areas. The results from this study demonstrate the combination of NDVI and classification method using Landsat images are promising for assessing FVC change in megacities.

Mapping Urban Impervious Surfaces by Using Spectral Mixture Analysis and Spectral Indices

Impervious surface is the major component of urban areas, and it has been widely considered as the key for assessing the degree of urban sprawl. While the effectiveness of applying spectral mixture analysis (SMA) and spectral indices in mapping urban impervious surface has been proved, most studies have relied either on SMA or spectral indices without considering both. In this study, the SMA and spectral indices were integrated together to map impervious surfaces distributions in both Milwaukee County in the Wisconsin State and Fayette County in the Kentucky State. Specifically, spectral indices were used for identifying major land covers. Two-dimensional feature space plots were generated by calculated spectral indices images for endmember selection and extraction. Linear constrained SMA was finally applied to quantify the fractional impervious surfaces. Research results indicate that the proposed method has achieved a promising accuracy, and better performance was achieved in less developed areas than the developed areas. Moreover, a comparative analysis shows that the proposed method performs better than the conventional method in both the whole study area and the developed areas, and a comparable performance has been achieved in the less developed areas.

How to account for endmember variability in spectral mixture analysis of night-time light imagery?

ABSTRACTNight-Time light imagery has become a very popular data source for monitoring the intensity of human activity in urban environments. Subpixel information is required in many applications, however, the widely used low-spatial-resolution night-time light imagery suffers from the mixed-pixel problem. In this paper, using the Visible Infrared Imaging Radiometer Suite (VIIRS) data, we presented the first spectral mixture analysis (SMA) on night-time light imagery. Specifically, we proposed to define two endmembers (light and dark) for endmember selection. In order to address the severe endmember variability problem caused by the various light sources and intensities, we adopted the Bayesian SMA (BSMA) method which is based on Bayes’ theorem. The results indicate that the approach can obtain subpixel light fraction accurately with an overall root mean square error (RMSE) of 0.17 and a coefficient of determination (R2) of 0.95. Although BSMA achieved similar results with the traditional linear SMA, BSMA allows one to determine the uncertainty of the estimated fraction as a distribution.

SPECTRAL CHARACTERIZATION OF A CLOSED CANOPY AND OPEN CANOPY FOREST IN NORTHERN SIERRA MADRE NATURAL PARK

Abstract. Forest lands play crucial roles in nutrient recycling and climate regulation. The change of closed canopy forests to open canopy forests may indicate disturbance within the closed canopy forest. Within the local context of the Philippines, few studies have been conducted to monitor changes in closed canopy forest lands. Efforts to do so are limited by the spatial extent, remoteness and ruggedness of closed canopy forests. Satellite imagery can cover the spatial extent of forest lands as well as provide constant revisit periods for monitoring. However, while multispectral imaging can detect changes in land cover, it has limitations when detecting the subtler change from closed canopy to open canopy forest cover. This study aims to provide baseline spectral characterization of a closed canopy forest in the Philippines. For this study, a hyperspectral sensor (EO1-Hyperion) with 198 band channels ranging from 426.82 nm to 2395.50 nm and a pixel size of 30 m was used to characterize the spectral variations of closed canopy forest, open canopy forest, shrubs and cropland in Northern Sierra Madre, Philippines. Multiple endmember spectral mixture analysis (MESMA) was employed to sort the image into classes as well as to characterize intra-spectral variations among the identified classes. Spectral library endmembers were assembled, optimized and used to classify the image. The spectral libraries were optimized by using Endmember Average Root Mean Square Error (EAR) , Minimum Average Spectral Angle (MASA) and Iterative Endmember Selection (IES). Results overall agreement is 0.56 for EAR and IES and kappa coefficient is at 0.4.

Monitoring Post-Fire Recovery of Chaparral and Conifer Species Using Field Surveys and Landsat Time Series

Recovery trajectories derived from remote sensing data are widely used to monitor ecosystem recovery after disturbance events, but these trajectories are often retrieved without a precise understanding of the land cover within a scene. As a result, the sources of variability in post-disturbance recovery trajectories are poorly understood. In this study, we monitored the recovery of chaparral and conifer species following the 2007 Zaca Fire, which burned 97,270 ha in Santa Barbara County, California. We combined field survey data with two time series remote sensing products: the relative delta normalized burn ratio (RdNBR) and green vegetation (GV) fractions derived from spectral mixture analysis. Recovery trajectories were retrieved for stands dominated by six different chaparral species. We also retrieved recovery trajectories for stands of mixed conifer forest. We found that the two remote sensing products were equally effective at mapping vegetation cover across the burn scar. The GV fractions (r(78) = 0.552, p < 0.001) and normalized burn ratio (r(78) = 0.555, p < 0.001) had nearly identical correlations with ground reference data of green vegetation cover. Recovery of the chaparral species was substantially affected by the 2011–2017 California drought. GV fractions for the chaparral species generally declined between 2011 and 2016. Physiological responses to fire and drought were important sources of variability between the species. The conifer stands did not exhibit a drought signal that was directly correlated with annual precipitation, but the drought likely delayed the return to pre-fire conditions. As of 2018, 545 of the 756 conifer stands had not recovered to their pre-fire GV fractions. Spatial and temporal variation in species composition were important sources of spectral variability in the chaparral and conifer stands. The chaparral stands in particular had highly heterogeneous species composition. Dominant species accounted for between 30% and 53% of the land cover in the surveyed chaparral patches, so non-dominant land cover types strongly influenced remote sensing signals. Our study reveals that prolonged drought can delay or alter the post-fire recovery of Mediterranean ecosystems. It is also the first study to critically examine how fine-scale variability in land cover affects time series remote sensing analyses.

A spatial pattern analysis of forest loss in the Madre de Dios region, Peru

Over the past decades, the Peruvian Amazon has experienced a rapid change in forest cover due to the expansion of agriculture and extractive activities. This study uses spectral mixture analysis (SMA) in a cloud-computing platform to map forest loss within and outside indigenous territories, protected areas, mining concessions, and reforestation concessions within the Madre de Dios Region in Peru. The study area is focused on key areas of forest loss in the western part of the Tambopata National Reserve and surrounding the Malinowski River. Landsat 8 Operational Land Imager and Landsat 7 Enhanced Thematic Mapper Plus surface reflectance data spanning 2013–2018 were analyzed using cloud-based SMA to identify patterns of forest loss for each year. High-resolution Planet Dove (3m) and RapidEye (5m) imagery were used to validate the forest loss map and to identify the potential drivers of loss. Results show large areas of forest loss, especially within buffer zones of protected areas. Forest loss also appears in the Kotsimba Native Community within a 1 km buffer of the Malinowski River. In addition to gold mining, agriculture and pasture fields also appear to be major drivers of forest loss for our study period. This study also suggests that gold mining activity is potentially not restricted to the legal mining concession areas, with 49% of forest loss occurring outside the mining concessions. Overall accuracy obtained for the forest loss analysis was 96%. These results illustrate the applicability of a cloud-based platform not only for land use land cover change detection but also for accessing and processing large datasets; the importance of monitoring not only forest loss progression in the Madre de Dios, which has been increasing over the years, especially within buffer zones, but also its drivers; and reiterates the use of SMA as a reliable change detection classification approach.

Improved mapping and understanding of desert vegetation-habitat complexes from intraannual series of spectral endmember space using cross-wavelet transform and logistic regression

Desert vegetation-habitat complexes in dryland systems are fragile ecosystems with complex vegetation-habitat feedback, and have significant implications for natural environment protection and global climate change mitigation. However, a spatial-detailed and high-precision remote sensing method for the identification of desert vegetation-habitat complexes and characterization of their biophysical processes remain scarce. Here, we developed an innovative cross-wavelet transform (XWT)-based approach coupled with logistic regression to extract critical vegetation-habitat interaction characteristics in order to identify, map, and understand their complex ecological processes. Fine intraannual profiles between the green vegetation (GV) fraction and habitat fractions including dark material (DA), saline land (SA), sand land (SL) were unmixed by Multiple Endmember Spectral Mixture Analysis (MESMA) from 16-period Gaofen-1 (GF-1) wide field of view (WFV) images in Minqin County, after which XWT was performed to extract feedback characteristics as feature parameters. Major principal components (PCs) were obtained from those feature parameters to reduce dimensions and solve multicollinearity, logistic regression was applied for mapping. The results demonstrate that the proposed procedure efficiently reproduced desert vegetation-habitat complexes with high accuracy (overall accuracy: 87.33%; Kappa coefficient: 0.86) in the entire Minqin County, representing a 3.42% overall accuracy increase relative to a previously published decision tree (DT) method. The new method also had a lower quantity and allocation disagreement. Moreover, this procedure not only achieved comparable accuracy to that of an optimized Support Vector Machine (SVM) and superior to a Convolutional Neural Network (CNN)-based U-net model, but also explored biophysical processes and complex relationships with better interpretability. Therefore, the developed approach has the potential for accurately monitoring the highly heterogeneous dryland landscape and characterizing the land degradation processes in the spectral endmember space of fine spatial-temporal remote sensing data.

Satellite-derived mapping of kelp distribution and water optics in the glacier impacted Yendegaia Fjord (Beagle Channel, Southern Chilean Patagonia)

Giant kelp Macrocystis pyrifera is a brown alga with extensive global distribution, however, recent evidence suggests that its dynamics presents high degree of regional variability. In southern Chilean fjord region, largely unexplored kelp forests are currently being threatened by global change and human impacts. High-resolution satellite (Sentinel-2) imagery was used to describe temporal and spatial distribution patterns of kelp beds in Yendegaia Fjord (Beagle Channel) using Spectral Mixture Analysis (SMA), and to characterize water optical gradients of this habitat strongly influenced by river runoff from a melting glacier. The suitability of SMA for kelp classification was contrasted with other vegetation indices (NDVI, EVI, FAI). Validation was made using drone aerial photographs of kelp canopies. Different analysis tools resulted in up to 35% difference in kelp coverage estimation. The overall accuracy (66–82%) of kelp classification followed an order FAI < EVI < NDVI < SMA. Omission error of SMA and lower coincidence with vegetation indices occurred in pixels with low kelp pixel abundance (<0.50). Based on SMA, the lowest kelp abundance was observed in the river mouth with high turbidity, increasing towards the Beagle Channel. The highest kelp abundance was observed in late summer, but otherwise no clear seasonal patterns could be observed. Water turbidity presented both spatial and seasonal variation. Strong particle sedimentation (leading to light attenuation, interference with remote detection of kelps, and even to their detachment due to substrate quality) and tidal fluctuations in glacier-impacted fjord-type environments can be identified as key features affecting both the kelp population dynamics as well as their remote sensing. Also, low sun elevation at high latitudes in mid winter produces uncertainties in image analyses. In all, the remote sensing approach used in the present study can be regarded as a useful tool to map and monitor kelps forests from a remote region.

Assessing the impact of endmember variability on linear Spectral Mixture Analysis (LSMA): A theoretical and simulation analysis

Abstract: The extensive existence of mixed pixels in satellite imagery is challenging for accurate land surface interpretation and parametric inversion. Spectral mixture analysis (SMA) addresses the issue by providing valuable sub-pixel information. However, it is affected by endmember variability exhibiting diverse spectral characteristics for the same endmember, inducing substantial uncertainties in model estimations. A number of approaches have been proposed to reduce the uncertainties, but theoretical explanations of how endmember variability affects model estimation and how these approaches perform successfully have not yet been derived. In this study, error propagation caused by endmember variability in linear SMA (LSMA) was investigated using theoretical analysis and simulation experiments. The major findings are as follows: (1) the impact of endmember variability on LSMA unmixing error depends on the interactions between deviation signal and gain vectors in a multiplicative way. (2) Unmixing error originates from the deviation signal that is governed jointly by spectral variability within an endmember class (i.e., intra-class variability) and endmember abundances, while spectra magnitude and spectral similarity among different endmember classes (i.e., inter-class variability) can amplify or reduce the deviation signal. (3) The typical approaches for mitigating endmember variability could not only change the deviation signal but also affect gain vectors, resulting in improved performance of LSMA to some extent. Based on these results, we recommend Multiple Endmember Spectral Mixture Analysis (MESMA) to be used in most applications considering its robustness in mitigating endmember variability. The results of this study will be of benefit to the application of LSMA in practice.

Performance Evaluation and Comparison of Modified Spectral Mixture Analysis Method for Different Images of Landsat Series Satellites

Urban impervious surface is considered one of main factors affecting urban heat island and urban waterlogging. It is commonly extracted utilizing the original linear spectral mixture analysis (LSMA) model. However, due to the deficiencies of this method, many improvements and modifications have been proposed. In this paper, a modified dynamic endmember linear spectral mixture analysis (DELSMA) model was introduced and tested in Zhengzhou, China, using different images of Landsat series satellites. The accuracy and performance of DELSMA model was evaluated in terms of R M S E , r and R 2 . Results show that (1) the DELSMA model performed equally well for Landsat-5 Thematic Mapper (TM) and Landsat-7 Enhanced Thematic Mapper (ETM+) images, and obtained better accuracy by using Landsat-8 Operational Land Imager (OLI) than Landsat TM/ETM+; (2) the DELSMA model achieved a better performance than the original LSMA model consistently, using images of Landsat from different sensors. Based exclusively on the overall accuracy reports, the DELSMA model proved to be a more efficient method for extracting impervious surface. Our study will provide a reliable method of impervious surface estimation for the urban planner and management in monitoring urban expansion, revealing urban heat island, and estimating urban surface runoff, using time-series Landsat imagery.

Using spectral Geodesic and spatial Euclidean weights of neighbourhood pixels for hyperspectral Endmember Extraction preprocessing

Spectral Mixture Analysis is one of the fundamental subjects encountered when dealing with remotely sensed hyperspectral images. Its goal is to identify constituent elements of mixed-pixels called Endmembers (EMs) and their associated abundance maps. In this paper, a novel Geodesic and Euclidean distances-based preprocessing (GEPP) is addressed which is coupled with the classical spectral-based EM Extraction algorithms (EEs). It combines both spatial and spectral information utilizing two approaches with the purpose of searching for spectrally pure and spatially homogenous pixels that may be identified as the EM candidates in subsequent EEs. GEPP reduces EE processing time by introducing a new correlation coefficient similarity function (CCSF) on the spectrally pure and spatially homogenous pixels pick up with the help of spectral weighting computations, unsupervised Fuzzy C-means (FCM) clustering algorithm and a spatial neighbourhood system using Markov Random Field (MRF) so that processing a large amount of mixed and heterogeneous pixels developed by EEs is avoided. Moreover, CCSF exploits the spatial Euclidean and novel spectral Geodesic weights to compute the final mean vector which is able to improve recognition of spatially homogenous regions that are highly spectrally correlated such that it leads to better results of unmixing accuracy. According to experimental results on three synthetic and four real hyperspectral scenes, hyperspectral unmixing outcomes are relatively improved in terms of SAD and RMSE-based error metrics and higher computation speed can be realized by our proposal in comparison with the state-of-the-arttechniques.

Frost Damage Assessment in Wheat Using Spectral Mixture Analysis

Frost damage to broadacre crops can cause up to an 85% loss in productivity. Although growers have few options for crop protection from frost, a rapid method for assessing frost-induced sterility would allow for timely management decisions (e.g., cutting for hay and altering marketing strategies). Spectral mixture analysis (SMA) has shown success in mapping landscape components and was used with hyperspectral data collected on the canopy, heads, and leaves of wheat at different sites to determine if this could quantify frost damage. Spectral libraries were assembled from canopy components collected from local field sites to generate spectral libraries for SMA from which a series of fraction sets was derived. The frost (Fr) fraction was then used to estimate final yield as a means of measuring frost damage. The best-fitting Fr fractions to yield were derived from the same data set as the source Fr spectra, and these ranged over R2 = 0.58–0.75 at the canopy scale. It was clear that spectral signatures need to be collected at scale to assess frost damage. While Fr fractions were able to estimate yield there was no “universal” endmember set from which a Fr fraction could be derived. The normalized difference vegetation index (NDVI) was not able to estimate frost damage consistently. Future work requires determining whether there is a “universal” set of endmembers and a minimum set of targeted wavebands that could lead to multispectral instruments for frost assessment for use in ground and aerial sensors.

VALIDATION OF EXTRACTED ENDMEMBERS FROM HYPERSPECTRAL IMAGES

Abstract. An essential step in the characterization of surface materials using hyperspectral image analysis is image classification using endmembers. Spectral unmixing is the best method for hyperspectral image classification. This method assumes that the pixel-topixel variability in scene results from varying proportions of spectral endmembers. Spectral endmembers can be derived from the imagery or measurements in the laboratory or field. The primary objective of this paper was to assess the ability of extracted endmembers against some different solutions for extraction of endmembers for hyperspectral image classification. In this paper, we compared the Pixel Purity Index (PPI) and the Sequential Maximum Angle Convex Cone (SMACC) as two popular methods of endmember extraction with library and field spectral. We used spectral information divergence for detection desirable endmembers from field spectral. For accuracy assessment of spectral mixture analysis and production of endmember abundance images for each of methods, the linear spectral unmixing algorithm is used. After a comparison between the results of these methods, it has been verified that field spectral have a better classification result in comparing with other endmember extraction methods. Also, the PPI has reliable results as an automatic endmember extraction method in comparing.

A hyperspectral experiment over tropical forests based on the EO-1 orbit change and PROSAIL simulation

ABSTRACTWe designed a unique hyperspectral experiment from the Earth Observing One (EO-1) orbit change to evaluate solar illumination effects over tropical forests in Brazil. Ten nadir-viewing Hyperion images collected over a fixed site and period of the year (July to August) were selected for analysis. We evaluated variations in reflectance and in 16 narrowband vegetation indices (VIs) with increasing solar zenith angle (SZA) from the pre-drift (2004–2008) to the EO-1 drift period (2011–2016). To detect changes in reflectance and shadows, we applied spectral mixture analysis (SMA) and principal component analysis (PCA) and calculated the similarity spectral angle (θ) between the vegetation spectra measured with variable SZA. The magnitude of the illumination effects was also evaluated from change-point analysis and nonparametric Mann-Whitney U tests applied over the time series. Finally, we complemented our experiment using the PROSAIL model to simulate the VIs variation with increasing SZA resultant from satellite drift. The results showed significant changes in Hyperion reflectance and VIs, especially when the EO-1 crossed the study area at earlier times and larger SZA in 2015 (9:05 a.m.; SZA = 59°) and 2016 (8:30 a.m.; SZA = 67°). Compared to the pre-drift period (10:30 a.m.; SZA = 45°), the SZA differences of 14° (2015) and 22° (2016) increased the shade fractions and decreased the vegetation brightness. PCA separated the pre-drift and drift reflectance datasets, showing shifts in scores due to changes in brightness. θ increased with SZA, indicating changes in the shape of the vegetation spectra with drift. For most VIs, the change-point analysis indicated 2015 (SZA = 59°) as the predominant year of detected changes. Compared to the EO-1 original orbit, the Plant Senescence Reflectance Index (PSRI), Anthocyanin Reflectance Index (ARI) and Structure Insensitive Pigment Index (SIPI) presented the largest positive changes during drift, while the Photochemical Reflectance Index (PRI), Visible Atmospherically Resistant Index (VARI) and Enhanced Vegetation Index (EVI) had the largest negative changes. The effect size of the illumination geometry on these VIs was large, as indicated by increasing values of the Cohen’s r metric toward 2016. The anisotropy of the Hyperion VIs was generally consistent with that from PROSAIL in the simulated pre-drift and drift periods. Focusing on structural indices, it affected the relationships between VIs and simulated leaf area index (LAI) at large SZA.

Comparing Remote Sensing Methods for Monitoring Karst Rocky Desertification at Sub-pixel Scales in a Highly Heterogeneous Karst Region

Rugged karst terrain relief that creates shadows in satellite imagery, combined with high karst landscape heterogeneity stand in the way of fractional cover retrieval on karst rocky desertification (KRD) monitoring. In this study, we explored the feasibility of applying multispectral high spatial resolution Advanced Land Observing Satellite (ALOS) imagery for the fractional cover extraction of rocky outcrops. Dimidiate pixel model (DPM) and spectral mixture analysis (SMA) approaches (including simple endmember spectral mixture analysis and multiple endmember spectral mixture analysis) were selected to explore their feasibility for KRD monitoring through accuracy improvement for fraction estimation. Results showed fractional cover retrievals at the sub-pixel scale is essential in highly heterogeneous karst landscapes. Indeed, mixed pixels accounted for 93.7% of the study area in southwest China. Multiple endmember spectral mixture analysis achieved high overall accuracy (80.5%) in monitoring the percentage of rocky outcrop land cover. Furthermore, the predicted exposed bedrock coverage via spectral mixture analysis were similar in sunlit and shadow areas for the same surface types. This reflected that SMA methods could effectively reduce topographic effects of satellite imagery to improve the accuracy of fractional cover extraction at sub-pixel level in heterogeneous and rugged landscapes.

Regional Scale Dryland Vegetation Classification with an Integrated Lidar-Hyperspectral Approach

The sparse canopy cover and large contribution of bright background soil, along with the heterogeneous vegetation types in close proximity, are common challenges for mapping dryland vegetation with remote sensing. Consequently, the results of a single classification algorithm or one type of sensor to characterize dryland vegetation typically show low accuracy and lack robustness. In our study, we improved classification accuracy in a semi-arid ecosystem based on the use of vegetation optical (hyperspectral) and structural (lidar) information combined with the environmental characteristics of the landscape. To accomplish this goal, we used both spectral angle mapper (SAM) and multiple endmember spectral mixture analysis (MESMA) for optical vegetation classification. Lidar-derived maximum vegetation height and delineated riparian zones were then used to modify the optical classification. Incorporating the lidar information into the classification scheme increased the overall accuracy from 60% to 89%. Canopy structure can have a strong influence on spectral variability and the lidar provided complementary information for SAM’s sensitivity to shape but not magnitude of the spectra. Similar approaches to map large regions of drylands with low uncertainty may be readily implemented with unmixing algorithms applied to upcoming space-based imaging spectroscopy and lidar. This study advances our understanding of the nuances associated with mapping xeric and mesic regions, and highlights the importance of incorporating complementary algorithms and sensors to accurately characterize the heterogeneity of dryland ecosystems.

Incorporating band selection in the spatial selection of spectral endmembers

The impact of band selection on endmember selection is seldom explored in the analysis of hyperspectral imagery. This study incorporates the N-dimensional Spectral Solid Angle (NSSA) band selection tool into the Spectral-Spatial Endmember Extraction (SSEE) tool to determine a band set that can be used to better define endmembers classes used in spectral mixture analysis. The incorporation aims to define a band set that improves the spectral contrast between endmembers at each step of the spatial-spectral endmember search and ultimately captures key features for discriminating spectrally similar materials. The proposed method (NSSA-SSEE) was evaluated for lithological mapping using a hyperspectral image encompassing a range of spectrally similar mafic and ultramafic rock units. The band selected by NSSA-SSEE showed a good agreement with known features of scene components identified by experts. Results showed an improvement in the selection of detailed endmembers, endmembers that are similar and that can be significant for mapping. The incorporation of NSSA into SSEE was feasible because both methods are well suited for this process. NSSA is one of the few methods of band selection that is suitable for the analysis of a small number of endmembers and SSEE provides such endmember sets via spatial subsetting. The automated NSSA-SSEE approach can reduce the need for field-based information to guide the feature selection process.

Comparison of Methods for Modeling Fractional Cover Using Simulated Satellite Hyperspectral Imager Spectra

Remotely sensed data can be used to model the fractional cover of green vegetation (GV), non-photosynthetic vegetation (NPV), and soil in natural and agricultural ecosystems. NPV and soil cover are difficult to estimate accurately since absorption by lignin, cellulose, and other organic molecules cannot be resolved by broadband multispectral data. A new generation of satellite hyperspectral imagers will provide contiguous narrowband coverage, enabling new, more accurate, and potentially global fractional cover products. We used six field spectroscopy datasets collected in prior experiments from sites with partial crop, grass, shrub, and low-stature resprouting tree cover to simulate satellite hyperspectral data, including sensor noise and atmospheric correction artifacts. The combined dataset was used to compare hyperspectral index-based and spectroscopic methods for estimating GV, NPV, and soil fractional cover. GV fractional cover was estimated most accurately. NPV and soil fractions were more difficult to estimate, with spectroscopic methods like partial least squares (PLS) regression, spectral feature analysis (SFA), and multiple endmember spectral mixture analysis (MESMA) typically outperforming hyperspectral indices. Using an independent validation dataset, the lowest root mean squared error (RMSE) values were 0.115 for GV using either normalized difference vegetation index (NDVI) or SFA, 0.164 for NPV using PLS, and 0.126 for soil using PLS. PLS also had the lowest RMSE averaged across all three cover types. This work highlights the need for more extensive and diverse fine spatial scale measurements of fractional cover, to improve methodologies for estimating cover in preparation for future hyperspectral global monitoring missions.