Discrimination of hoary cress and determination of its detection limits via hyperspectral image processing and accuracy assessment techniques

doi:10.1016/j.rse.2005.04.004

Remote Sensing of Environment

Volume 96, Issues 3–4, 30 June 2005, Pages 509-517

https://doi.org/10.1016/j.rse.2005.04.004 Get rights and content

Abstract

This study documents successful discrimination of hoary cress (Cardaria draba) in southwestern Idaho using hyperspectral imagery to a maximum producer's accuracy of 82% for infestations with greater than 30% cover. Different hyperspectral processing parameters were evaluated and compared, including data transformations, endmember selection, classification algorithms, and post-classification accuracy assessment methods. In this study, the Spectral Angle Mapper (SAM) and Mixture Tuned Matched Filtering (MTMF) classification algorithms performed equally. Minimum Noise Fraction (MNF) data transformation generated producer's accuracies 23% higher than did similar classifications using Principal Components Analysis (PCA) transformed data. Two hoary cress endmembers derived from different vegetative regimes were necessary for successful classification. Finally, this study documents a methodology comparing incremental map accuracies to optimize classifier performance and determine the detectable limits of hoary cress. Detection limits using hyperspectral imagery were as low as 10% cover over a 3 m × 3 m pixel using a mesic vegetative regime endmember. However, for management level use of the imagery, both a mesic and a xeric endmember were necessary for the 30% cover threshold.

Introduction

Hoary cress (Cardaria draba) is an invasive plant that is listed as a noxious weed in the State of Idaho (Prather et al., 2002). It is a 1–2 ft tall rhizomatous perennial that thrives in disturbed semiarid landscapes. During bloom, hoary cress exhibits dense white flowers that give the plant a flat and mat-like appearance. Management of hoary cress is moderately difficult, often requiring the use of herbicides. The weed spreads rapidly via anthropogenic and wildlife vectors, and as a result land managers have the need to understand its distribution to aid in management decisions.

The objectives of this study are: (1) to map the distribution of hoary cress using hyperspectral imagery acquired over a study area in southwestern Idaho, and (2) to evaluate and compare different hyperspectral analysis parameters for hoary cress discrimination. This study was completed under a NASA sponsored initiative to bring science and technology applications into an operational context. The methods and discussion presented herein are intended to serve as guide for hyperspectral image processing techniques for vegetative species discrimination.

The study area (− 111°20′W, 43°30′N) is located in relatively flat terrain that is heavily utilized for agriculture in Ada County, Idaho (Fig. 1). Ten-mile Creek, several ditches and canals, and a well-developed network of roads provide transport vectors for hoary cress, which forms infestations diverse in both size (contiguous hectares to smaller than 10 m²) and percent canopy cover (herein referred to as ‘percent cover’ or ‘cover’). Vegetation in the study area is locally variable, including mesic (e.g. bluegrass (Poa spp.) and orchardgrass (Dactylis glomerata L.)), and xeric (e.g. cheatgrass (Bromus tectorum L.), blue mustard (Chorispora tenella (Pall.) D.C.) and wild rose (Rosa multiflora Thunb. ex Murr)) vegetative regimes. Typically, infestations of hoary cress with high cover (∼ 100%) are correlated with mesic regimes, while xeric regimes host smaller infestations with lower cover (∼ 60% to 80%).

Section snippets

Previous work and technical background

Because different vegetation types have different spectral characteristics (Everitt et al., 2001, Gates et al., 1965, Kondrat'yev & Fedchenko, 1981), many studies have utilized remote sensing techniques to generate distribution maps for invasive species rather than relying solely on the use of a field crew (e.g. Everitt et al., 1995, Glenn et al., 2005, Parker-Williams & Hunt, 2002). In general, studies using multispectral data for species distribution mapping have found limited success, while

Data collection

Hyperspectral data was acquired by the HyMap sensor (operated by HyVista, Inc.) over the study area (1.75 by 22 km; 3 m spatial resolution) on May 21, 2003, while hoary cress was blooming (Fig. 1). The hyperspectral sensor collected 126 contiguous spectral bands representing the intensity of reflected solar radiation between the wavelengths of 450 nm and 2500 nm, with radiometric resolutions of ∼ 15 nm (Kruse et al., 2000). Concurrent with image acquisition in 2003, Global Positioning System

Processing

Data reduction transformations demonstrated large accuracy differences when classifying with SAM using the mesic endmember. The PCA classification resulted in producer's, user's, and overall accuracies of 31%, 96%, and 46%, respectively (Table 2; SAM/PCA/M). The same classification applied to the MNF transformed data generated producer's, user's, and overall accuracies of 54%, 96%, and 63%, respectively (Table 2; SAM/MNF/M). While significance testing between the respective Kappa values

Conclusions

This study successfully discriminated hoary cress in Ada County, Idaho, with producer's, user's, and overall accuracies of 82%, 79%, and 86%, respectively, for infestations with at least 30% cover. The most successful classification approach utilized MNF transformed reflectance data using two endmembers for hoary cress. In this study, the SAM (threshold of 1 radian) and MTMF (MF values greater than 0 and infeasibility values less than 15) comparison performed nearly identically (maximum 3%

Acknowledgements

This project was funded by the NASA BAA program (BAA-01-0ES-01; NAG13-02029). We thank Mr. William Graham of NASA Stennis for his continued support. Additional funding was contributed by the Idaho National Laboratory (INL) through the Idaho State University–INL Partnership for Integrated Environmental Analysis Education Outreach Program. The authors would like to acknowledge the Weed Control crews from Ada County and Bonneville County, Idaho, as well as Charlla Adams, Mark Strom and Ellen

References (31)

R.J. Aspinall
Use of logistic regression for validation of maps of the spatial distribution of vegetation species derived from high spatial resolution hyperspectral remotely sensed data
Ecological Modeling
(2002)
R.G. Congalton
A review of assessing the accuracy of classifications of remotely sensed data
Remote Sensing of Environment
(1991)
G.M. Foody
Status of land cover classification accuracy assessment
Remote Sensing of Environment
(2002)
N.F. Glenn et al.
Hyperspectral data processing for repeat detection of small infestations of leafy spurge
Remote Sensing of Environment
(2005)
E.B. Knipling
Physical and physiological basis for the reflectance of visible and near-infrared radiation from vegetation
Remote Sensing of Environment
(1970)
F.A. Kruse et al.
The spectral image processing system (SIPS)—Interactive visualization and analysis of imaging spectrometer data
Remote Sensing of Environment
(1993)
A. Parker-Williams et al.
Estimation of leafy spurge cover from hyperspectral imagery using Mixture Tuned Matched Filtering
Remote Sensing of Environment
(2002)
D.A. Roberts et al.
Mapping chaparral in the Santa Monica Mountains using multiple endmember spectral mixture models
Remote Sensing of Environment
(1998)
S.V. Stehman et al.
Design and analysis for thematic map accuracy assessment: Fundamental principles
Remote Sensing of Environment
(1998)
E. Underwood et al.
Mapping nonnative plants using hyperspectral imagery
Remote Sensing of Environment
(2003)

R.J. Aspinall et al.

Considerations in collecting, processing, and analyzing high spatial resolution hyperspectral data for environmental investigations

Journal of Geographic Systems

(2002)

J.W. Boardman

Leveraging the high dimensionality of AVIRIS data for improved sub-pixel target unmixing and rejection of false positives: Mixture tuned matched filtering

D. Chen et al.

The effect of training strategies on supervised classification at different spatial resolutions

Photogrammetric Engineering and Remote Sensing

(2002)

R.G. Congalton et al.

Assessing the accuracy of remotely sensed data: Principles and practices

(1999)

J.C. Davis

Statistics and data analysis in geology

(1986)

Cited by (75)

Evaluation of subpixel unmixing algorithms in mapping the porphyry copper alterations using EO-1 Hyperion data, a case study from SE Iran
2018, Remote Sensing Applications: Society and Environment
An EO‐1 Hyperion scene was used to assess various subpixel unmixing algorithms in a study area at the central Iranian volcano-sedimentary complex, where abundant porphyry copper deposits like Sarcheshmeh, Darrehzar, and Sereidun as well as numerous occurrences of copper exist. Different subpixel unmixing algorithms such as matched filtering (MF), constrain energy minimization (CEM) and mixture tune matched filtering (MTMF) were applied to map the sericite occurrences within alteration systems associated with porphyry deposits. The output results were evaluated by semi-quantitative XRD analysis of corresponding field samples and their accuracies were assessed using R² coefficients and deviation errors. Generally, MF results provided a moderate consistency with the corresponding XRD abundances following a positive relation. The CEM algorithm provided weak results for Hyperion data. MTMF had the best performance and this is definitely because of segregating power in the mixture tuning, which can distinguish high infeasible samples with unreliable fractions of target mineral.
A non-intrusive image analysis technique for measurement of heterogeneity in grass species around tree vicinity in a green infrastructure
2018, Measurement: Journal of the International Measurement Confederation
Spatial heterogeneity of vegetation growth is important for maintenance of urban green space. It also governs the differential settlement of foundation of buildings and performance of biofiltration units. The objective of this study is to analyse the heterogeneity in vegetation density and shoot growth of a grass around a tree vicinity. A novel non-intrusive image analysis approach was designed and developed for quantifying heterogeneity in vegetation growth. A commercially available unmanned air vehicle (UAV; PHANTOM 3 STANDARD) was utilized to capture images. Vegetation density from these captured images were quantified using a public domain image processing program ImageJ. Atmospheric parameters were monitored by micro-climate monitoring system for interpreting vegetation growth. It is found that, at a given radial distance from stem of tree, vegetation density range is found more heterogenous than shoot growth. The basic assumption of symmetricity around tree vicinity as adopted in previous models for root water uptake is found to be not true. Variation of rainfall is one of the main reason causing heterogeneity in grass growth around tree vicinity. Heterogeneity in vegetation growth more prominent near the tree vicinity than away from it. An increase in vegetation density is found within 2 m radial distance in both sides of tree stem due to presence of shredded leaves from tree during winter.
Detecting bugweed (Solanum mauritianum) abundance in plantation forestry using multisource remote sensing
2016, ISPRS Journal of Photogrammetry and Remote Sensing
Citation Excerpt :
Remote sensing technologies thus offer a viable option for obtaining information related to the location, distribution and spatial extent of IAPs (Lass et al., 2005; Miao et al., 2006). Many applications in remote sensing have investigated the detection and mapping of IAPs (Fuller, 2005; Glenn et al., 2005; Lawrence et al., 2006; Mundt et al., 2005; Laba et al., 2008 Peerbhay et al., 2015). For example, Fuller (2005) employed a back-propagation neural network (NN) classifier to successfully map the invasive paperbark tree using IKONOS imagery (4 m) and obtained an accuracy of greater than 85% when combined with a landscape fragmentation analysis.
The invasive weed Solanum mauritianum (bugweed) has infested large areas of plantation forests in KwaZulu-Natal, South Africa. Bugweed often forms dense infestations and rapidly capitalises on available natural resources hindering the production of forest resources. Precise assessment of bugweed canopy cover, especially at low abundance cover, is essential to an effective weed management strategy. In this study, the utility of AISA Eagle airborne hyperspectral data (393–994 nm) with the new generation Worldview-2 multispectral sensor (427–908 nm) was compared to detect the abundance of bugweed cover within the Hodgsons Sappi forest plantation. Using sparse partial least squares discriminant analysis (SPLS-DA), the best detection results were obtained when performing discrimination using the remotely sensing images combined with LiDAR. Overall classification accuracies subsequently improved by 10% and 11.67% for AISA and Worldview-2 respectively, with improved detection accuracies for bugweed cover densities as low as 5%. The incorporation of LiDAR worked well within the SPLS-DA framework for detecting the abundance of bugweed cover using remotely sensed data. In addition, the algorithm performed simultaneous dimension reduction and variable selection successfully whereby wavelengths in the visible (393–670 nm) and red-edge regions (725–734 nm) of the spectrum were the most effective.
Mapping Solanum mauritianum plant invasions using WorldView-2 imagery and unsupervised random forests
2016, Remote Sensing of Environment
The accurate detection and mapping of plant invasions is important for an effective weed management strategy in forest plantations. In this study, the utility of WorldView-2 was investigated to automatically map the occurrence of Solanum mauritianum (bugweed) found as an anomaly in forest margins, open areas and riparian zones. The unsupervised methodology developed, proved to be an effective and an accurate framework in detecting and mapping the invasive alien plant (IAP). Using the random forest (RF) proximity matrix, similarity measures between pixels were successfully transformed into scores (Eigen weights) for each pixel using eigenvector analysis. Neighbourhood windows with minimum variance revealed the most important information from localized surrounding pixels to detect potential anomalous pixels. Bugweed occurrence in forest margins, open areas and riparian zones were successfully mapped at accuracies of 91.33%, 85.08%, and 67.90%, respectively. This research has demonstrated the unique capability of using an automated unsupervised RF approach for mapping IAPs using new generation multispectral remotely sensed data.
A meta-analysis of remote sensing research on supervised pixel-based land-cover image classification processes: General guidelines for practitioners and future research
2016, Remote Sensing of Environment
Citation Excerpt :
Feature extraction is usually done either by selecting the most discriminating bands from existing bands (Chan & Paelinckx, 2008; Clark, Roberts, & Clark, 2005; Thenkabail, Enclona, Ashton, & Van Der Meer, 2004), or by transforming data to a different, more efficient feature space. Linear transformation methods such as Principal Component Analysis (PCA) (Clark et al., 2005; Fauvel, Benediktsson, Chanussot, & Sveinsson, 2008) and Minimum Noise Fraction (MNF) (Mundt et al., 2005; Yang, Everitt, & Johnson, 2009) were the most common techniques of dimensionality reduction observed in the reviewed articles. This general categorization of enhancement methods made it possible to have a sufficiently large sample of articles to conduct statistical tests.
Classification of remotely sensed imagery for land-cover mapping purposes has attracted significant attention from researchers and practitioners. Numerous studies conducted over several decades have investigated a broad array of input data and classification methods. However, this vast assemblage of research results has not been synthesized to provide coherent guidance on the relative performance of different classification processes for generating land cover products. To address this problem, we completed a statistical meta-analysis of the past 15 years of research on supervised per-pixel image classification published in five high-impact remote sensing journals. The two general factors evaluated were classification algorithms and input data manipulation as these are factors that can be controlled by analysts to improve classification accuracy. The meta-analysis revealed that inclusion of texture information yielded the greatest improvement in overall accuracy of land-cover classification with an average increase of 12.1%. This increase in accuracy can be attributed to the additional spatial context information provided by including texture. Inclusion of ancillary data, multi-angle and time images also provided significant improvement in classification overall accuracy, with 8.5%, 8.0%, and 6.9% of average improvements, respectively. In contrast, other manipulation of spectral information such as index creation (e.g. Normalized Difference Vegetation Index) and feature extraction (e.g. Principal Components Analysis) offered much smaller improvements in accuracy. In terms of classification algorithms, support vector machines achieved the greatest accuracy, followed by neural network methods. The random forest classifier performed considerably better than the traditional decision tree classifier. Maximum likelihood classifiers, often used as benchmarking algorithms, offered low accuracy. Our findings will help guide practitioners to decide which classification to implement and also provide direction to researchers regarding comparative studies that will further solidify our understanding of different classification processes. However, these general guidelines do not preclude an analyst from incorporating personal preferences or considering specific algorithmic benefits that may be pertinent to a particular application.
Testing the discrimination and detection limits of WorldView-2 imagery on a challenging invasive plant target
2016, International Journal of Applied Earth Observation and Geoinformation
Citation Excerpt :
Significant differences between the Kappa statistics were tested using a two-tailed Z-test at α = 0.05 (Congalton and Green, 1999). Confusion caused by small plants was investigated using an incremental cover evaluation — ICE (Mundt et al., 2005). An ICE examines omission errors via the Producer’s accuracy metric (Congalton and Green, 1999) at size-class increments.
Invasive plants pose significant threats to biodiversity and ecosystem function globally, leading to costly monitoring and management effort. While remote sensing promises cost-effective, robust and repeatable monitoring tools to support intervention, it has been largely restricted to airborne platforms that have higher spatial and spectral resolutions, but which lack the coverage and versatility of satellite-based platforms. This study tests the ability of the WorldView-2 (WV2) eight-band satellite sensor for detecting the invasive shrub mesquite (Prosopis spp.) in the north-west Pilbara region of Australia. Detectability was challenged by the target taxa being largely defoliated by a leaf-tying biological control agent (Gelechiidae: Evippe sp. #1) and the presence of other shrubs and trees. Variable importance in the projection (VIP) scores identified bands offering greatest capacity for discrimination were those covering the near-infrared, red, and red-edge wavelengths. Wavelengths between 400 nm and 630 nm (coastal blue, blue, green, yellow) were not useful for species level discrimination in this case. Classification accuracy was tested on three band sets (simulated standard multispectral, all bands, and bands with VIP scores ≥1). Overall accuracies were comparable amongst all band-sets (Kappa = 0.71–0.77). However, mesquite omission rates were unacceptably high (21.3%) when using all eight bands relative to the simulated standard multispectral band-set (9.5%) and the band-set informed by VIP scores (11.9%). An incremental cover evaluation on the latter identified most omissions to be for objects <16 m². Mesquite omissions reduced to 2.6% and overall accuracy significantly improved (Kappa = 0.88) when these objects were left out of the confusion matrix calculations. Very high mapping accuracy of objects >16 m² allows application for mapping mesquite shrubs and coalesced stands, the former not previously possible, even with 3 m resolution hyperspectral imagery. WV2 imagery offers excellent portability potential for detecting other species where spectral/spatial resolution or coverage has been an impediment. New generation satellite sensors are removing barriers previously preventing widespread adoption of remote sensing technologies in natural resource management.

View all citing articles on Scopus

View full text

Discrimination of hoary cress and determination of its detection limits via hyperspectral image processing and accuracy assessment techniques

Abstract

Introduction

Section snippets

Previous work and technical background

Data collection

Processing

Conclusions

Acknowledgements

Ecological Modeling

Remote Sensing of Environment

Remote Sensing of Environment

Remote Sensing of Environment

Remote Sensing of Environment

Remote Sensing of Environment

Remote Sensing of Environment

Remote Sensing of Environment

Remote Sensing of Environment

Remote Sensing of Environment

Considerations in collecting, processing, and analyzing high spatial resolution hyperspectral data for environmental investigations

Journal of Geographic Systems

Leveraging the high dimensionality of AVIRIS data for improved sub-pixel target unmixing and rejection of false positives: Mixture tuned matched filtering

The effect of training strategies on supervised classification at different spatial resolutions

Photogrammetric Engineering and Remote Sensing

Assessing the accuracy of remotely sensed data: Principles and practices

Statistics and data analysis in geology