Contents Lists Available At ScienceDirect International .

3y ago
41 Views
2 Downloads
1.10 MB
8 Pages
Last View : 17d ago
Last Download : 2m ago
Upload by : Annika Witter
Transcription

International Journal of Applied Earth Observation and Geoinformation 31 (2014) 78–85Contents lists available at ScienceDirectInternational Journal of Applied Earth Observation andGeoinformationjournal homepage: www.elsevier.com/locate/jagEarly detection of crop injury from herbicide glyphosate by leafbiochemical parameter inversionFeng Zhao a, , Yiqing Guo a , Yanbo Huang b , Krishna N. Reddy b , Matthew A. Lee b ,Reginald S. Fletcher b , Steven J. Thomson babSchool of Instrumentation Science and Opto-electronics Engineering, Beihang University, Beijing 100191, PR ChinaUSDA-Agricultural Research Service, Crop Production Systems Research Unit, 141 Experiment Station Road, Stoneville, MS 38776, USAa r t i c l ei n f oArticle history:Received 23 December 2013Accepted 12 March 2014Keywords:Crop injuryGlyphosateFoliar biochemistrySensitivity analysisModel inversionHyperspectrala b s t r a c tEarly detection of crop injury from herbicide glyphosate is of significant importance in crop management.In this paper, we attempt to detect glyphosate-induced crop injury by PROSPECT (leaf optical PROpertySPECTra model) inversion through leaf hyperspectral reflectance measurements for non-GlyphosateResistant (non-GR) soybean and non-GR cotton leaves. The PROSPECT model was inverted to retrievechlorophyll content (Ca b ), equivalent water thickness (Cw ), and leaf mass per area (Cm ) from leaf hyperspectral reflectance spectra. The leaf stress conditions were then evaluated by examining the temporalvariations of these biochemical constituents after glyphosate treatment. The approach was validated withgreenhouse-measured datasets. Results indicated that the leaf injury caused by glyphosate treatmentscould be detected shortly after the spraying for both soybean and cotton by PROSPECT inversion, withCa b of the leaves treated with high dose solution decreasing more rapidly compared with leaves leftuntreated, whereas the Cw and Cm showed no obvious difference between treated and untreated leaves.For both non-GR soybean and non-GR cotton, the retrieved Ca b values of the glyphosate treated plantsfrom leaf hyperspectral data could be distinguished from that of the untreated plants within 48 h after thetreatment, which could be employed as a useful indicator for glyphosate injury detection. These findingsdemonstrate the feasibility of applying the PROSPECT inversion technique for the early detection of leafinjury from glyphosate and its potential for agricultural plant status monitoring. 2014 Elsevier B.V. All rights reserved.IntroductionGlyphosate drift has been of particular concern recently becauseit can cause injury or mortality to off-target sensitive nonGlyphosate-Resistant (non-GR) crops (Ding et al., 2011). For theearly detection of crop injury from off-target glyphosate drift, portions of the visible and near-infrared reflectance spectra are idealindicators of stress because stress-induced changes of leaf interiorstructure and growth status could alter the spectrum from that ofa healthy leaf (Huang et al., 2012).Foliar biochemical properties represent the growth status ofplants, and they are good indicators of glyphosate-induced leafinjury (Reddy et al., 2000, 2010; Koger et al., 2005). For thepurpose of detecting the crop injury caused by glyphosate drift,traditional methods of directly measuring the leaf biochemicalparameters in vivo are labor- and time-intensive and cannot meet Corresponding author. Tel.: 86 10 82315884.E-mail address: zhaofeng@buaa.edu.cn (F. 303-2434/ 2014 Elsevier B.V. All rights reserved.requirements for rapid and large-scale monitoring. Several studies have attempted to develop indirect approaches for detectingcrop stress (e.g. water-stress and nitrogen-stress) with hyperspectral reflectance data (Barnes et al., 1992; Carter, 1994; Filella andPeñuelas, 1994). Recently, these indirect approaches have beenintroduced for detection of glyphosate-induced crop injury by thebiological remote sensing community. For example, in an airborneremote sensing experiment, Huang et al. (2010) assessed damageto cotton caused by spray drift from aerially applied glyphosate bymapping the NDVI (Normalized Difference Vegetation Index) imageof the experimental area. More recently, Huang et al. (2012) usedhyperspectral reflectance data to distinguish the glyphosate injuredsoybean and cotton leaves from the healthy ones by calculatingthe NDVI, RVI (Ratio Vegetation Index), SAVI (Soil Adjusted Vegetation Index), and DVI (Difference Vegetation Index) of each leaf. Ina greenhouse experiment, Yao et al. (2012) found that hyperspectral imaging of plant canopy was a useful tool for early detectionof soybean injury due to glyphosate application, and that spectral derivative indices proved to be a good indicator for glyphosateinjury. As these vegetation indices were not specifically designed

F. Zhao et al. / International Journal of Applied Earth Observation and Geoinformation 31 (2014) 78–85for crop injury detection and therefore less effective, spectral feature extraction methods were introduced (Zhao et al., 2014).Current efforts depend primarily on constructing vegetationindices or spectral features that potentially relate to glyphosateinduced crop stress. But these methods are not physically-basedand may not be applied effectively over a wide range of species.Physically-based radiative transfer models that quantitativelyrelate foliar biochemical properties to reflectance spectra caninherently provide more consistent results over multiple speciesand have the potential of improving detection of glyphosateinduced crop injury.In this study, we attempted to detect glyphosate-inducedleaf injury through quantitative estimation of foliar biochemicalcontents from leaf hyperspectral reflectance measurements. Thiswas accomplished by inversion of a physically based radiativetransfer model, PROSPECT (leaf optical PROperty SPECTra model)(Jacquemoud and Baret, 1990; Fourty et al., 1996; Jacquemoudet al., 1996, 2000; Feret et al., 2008). To obtain a more accurateresult, we applied an improved procedure for model inversion toimprove the retrieval accuracies of the foliar biochemical parameters: chlorophyll content (Ca b , chlorophyll a b content, in unitof g/cm2 ), equivalent water thickness (Cw , mass of water per leafarea, in unit of g/cm2 ), leaf mass per area (Cm , mass of dry matter per leaf area, in unit of g/cm2 ), and leaf structural parameter(N, number of compact layers specifying the average number ofair/cell walls interfaces within the mesophyll). In order to evaluatethe effectiveness of the proposed inversion procedure, correlationscalograms of retrieved versus measured values were plotted forCa b , Cw , and Cm , respectively. Glyphosate-induced leaf injury wasthen analyzed by examining temporal variations of these retrievedbiochemical parameters after leaf treatment at high-dose, low-doseand no glyphosate. Finally, advantages and potential of this proposed method were discussed.ExperimentThe experiment was conducted in a greenhouse located atthe USDA-Agricultural Research Service, Crop Production SystemsResearch Unit, Stoneville, Mississippi on December 17–20, 2012,and repeated February 4–7, 2013. The crops were planted in potsusing a Completely Randomized Design (CRD), and growing conditions for the plants set temperature to 23.9 C in the daytime and21.1 C at night. Four weeks after planting, the plants were treatedand the leaves of them were measured for spectral reflectanceexperiment. The four week schedule to spray glyphosate was determined by weed scientists to simulate the situation in field toeffectively control weeds.In each experiment, 36 pots of non-GR cotton (cultivar FM955LL)and 36 pots of non-GR soybean (cultivar SO80120LL) were usedto obtain leaf reflectance spectra and foliar biochemical properties. For each crop, we divided the pots randomly into 3 treatmentgroups: 12 plants were sprayed with 0.433 kg ae/ha solution ofglyphosate (0.5X group; X 0.866 kg ae/ha, which is the label rate ofglyphosate); another 12 plants were sprayed with half of the 0.5Xdose (0.25X group); the remaining 12 plants were used as controlswith no glyphosate treatment (CTRL group). Glyphosate solutionswere prepared using a commercial formulation of the potassiumsalt of glyphosate (Roundup WeatherMax, Monsanto AgriculturalCo., St. Louis, MO), and applied using a CO2 -pressurized backpacksprayer that delivered 140 L/ha of spray solution at 193 kPa. Afterthe glyphosate spraying, leaf reflectance and biochemical parameters (Ca b , Cw , and Cm ) of three plants for each group were measuredat 6, 24, 48, 72 Hours After the Treatment (HAT) to study plantresponse to glyphosate.79Leaf reflectance measurements were acquired by using an ASDintegrating sphere apparatus coupled with the ASD FieldSpec 3Hi-Res spectroradiometer (ASD Inc., Boulder, CO., USA), yieldinga 1-nm spectral resolution in the visible to near-infrared range(400–2500 nm). Connected with the integrating sphere, SpareLamps (Qty 2, Osram #64225, 6 V, 10 W) provides a collimatedbeam as the light source, which illuminates the sample or the Reference Standard.The reflectance of leaf sample was measured following the procedure described in the manual of ASD integrating sphere (ASD Inc.,2008) in which three measurements are required: sample measurement (Is ), stray light measurement (Id ), and Reference Standardmeasurement (Ir ). These spectra were collected in raw DN (Digital Number) mode. An integration time of 544 ms was used forall the measurements. With the known reflectance of the Reference Standard, Rr , the reflectance of the sample for a given centerwavelength and spectral bandpass, Rs , is calculated as follows:Rs (Is Id )RrIr Id(1)One of the lowermost trifoliate leaves for soybean and twinleaves for cotton was selected for the measurements of thereflectance. These leaves were identified before the glyphosatetreatment to make sure leaves at the same position of each plantwere used for all four days. The leaves were large enough to coverthe port of the integrating sphere. The location of the leaf samplechanged three times during the measurement (avoiding main veinsof the leaf in the port) to acquire the mean spectrum of the leaf.After the leaf reflectance measurement, the leaf sample’s areawas immediately measured using a LI-COR 3100 Area Meter (LICOR, Inc., Lincoln, NE, USA). The sample was then dropped into avial with DiMethyl SulfOxide (DMSO) and covered with aluminumfoil. After 24 h in the dark environment, the solution was used forchlorophyll analysis using a Shimadzu UV160U Spectrophotometer(Shimadzu Corp., Kyoto, Japan). In order to calculate Cw and Cm , theremaining leaves of the plants were scanned to determine the leafarea and weighed to measure their fresh weights. Then they wereoven-dried at 45–50 C for 48 h, and reweighed to determine thedry weights. The mean values and ranges of Ca b , Cw , and Cm overthese two experiments are summarized in Table 1.MethodsAn improved approach for PROSPECT inversion was implemented for enhanced retrieval accuracy of leaf biochemicalparameters. The PROSPECT model was first used to generate anartificial dataset, which would be used in sensitivity analysis; inthis case a sensitive wavelength region was selected for eachinput parameter of PROSPECT. Based on the sensitivity analysisresult, each parameter was assigned a specific merit function onits sensitive wavelength region, and a global optimization algorithm was used to retrieve these parameters. Finally, the accuracyof the inversion process was evaluated by comparing the retrievedand measured values. After the leaf biochemical parameters wereretrieved by model inversion, glyphosate-induced leaf injury wasanalyzed by examining the temporal variations of these retrievedvalues. The schematic representation of the injury detection process is shown in Fig. 1.Artificial data generationWhen N, Ca b , Cw , and Cm are determined, leaf hemispherical reflectance spectra in the wavelength band of 400–2500 nmcan be simulated by PROSPECT. The model was first calibrated using the method given by Feret et al. (2008) andLi and Wang (2011) with the data of CTRL groups. For

80F. Zhao et al. / International Journal of Applied Earth Observation and Geoinformation 31 (2014) 78–85Table 1Leaf biochemical data from greenhouse-measured datasets (December 2012 and February 2013). The maximum, minimum, and mean values of leaf chlorophyll content(Ca b ), water content (Cw ), and dry matter content (Cm ) of soybean and cotton acquired in these experiments are shown in the 3552.808619.1068.1611Cw (g/cm2 1750.01350.02460.0177Cm (g/cm2 0300.00240.00420.00302Ca b ( g/cm )Min.Max.MeanaCTRL group contains leaves with no glyphosate treatment; 0.25X group contains leaves treated with 0.217 kg ae/ha solution of glyphosate; 0.5X group contains leavestreated with 0.433 kg ae/ha solution of glyphosate.Table 2Sensitive wavelength regions for PROSPECT input parameters.ParameterNCa bCwCmSensitive wavelength region760–1300 nm400–760 nm1900–2100 nm2100–2300 nmartificial data generation, ranges of Ca b , Cw , and Cm weredefined as 2.8086–19.106 g/cm2 , 0.0098–0.0267 g/cm2 , and0.0018–0.0045 g/cm2 , respectively, since they could cover all thegreenhouse-measured values presented in Table 1. N was assigneda reasonable range of 1–4, which could describe a wide range ofmesophyll structures of different leaf species (Jacquemoud andBaret, 1990). One thousand combinations of the parameters wererandomly selected from these ranges as the inputs and 1000reflectance spectra were produced by model simulation. All combinations of parameters with the corresponding reflectance spectraformed our artificial dataset, which would be used in sensitivityanalysis of PROSPECT.Sensitivity analysis of PROSPECTThe method of EFAST (Extended Fourier Amplitude SensitivityTest), which was proposed by Saltelli et al. (1999), was used forsensitivity analysis of PROSPECT in our study. The artificial datasetpreviously simulated by PROSPECT was used as input data. EFASTallows the simultaneous computation of the first order and thetotal sensitivity indices for a given input variable. The first ordersensitivity index gives the independent effect of the correspondingparameters, while the total sensitivity index contains both independent effect of each parameter and the interaction effects withthe others (Saltelli et al., 2008).The result of PROSPECT sensitivity analysis showed that eachparameter had its own comparatively sensitive spectral band.Within wavelengths between 760 and 1300 nm, N was the crucialparameter that contributed more than 90% uncertainty of the outputs. For shorter wavelengths in the visible band of 400–760 nm,Ca b had the greatest influence. Compared with N and Ca b , Cw andCm were more sensitive in the short-wave infrared band. It couldbe seen that Cw is the most sensitive parameter in 1900–2100 nm,while Cm was relatively more sensitive in 2100–2300 nm comparedwith others wavelengths. There was little difference between thefirst order and total sensitivity indices in most cases ( 5%), suggesting that interaction effects among different parameters weresmall. The sensitive bands for these parameters are summarized inTable 2. More details about the PROSPECT sensitivity analysis withthe method of EFAST could be found in Zhao et al. (2014).Model inversion approachFig. 1. Schematic representation of the leaf injury detection process.Many approaches have been used in previous research forPROSPECT inversion, almost all of which retrieve input parameters by minimizing a single merit function defining on the entireoptical domain from 400 to 2500 nm with a classical optimization algorithm (most of them are local optimization algorithmslike the downhill simplex) (Fourty et al., 1996; Jacquemoud et al.,1996; Feret et al., 2008; Romero et al., 2012). With these traditionalapproaches, the algorithms always choose downhill direction ineach step, in order to reach a nearby solution as quickly as possible.

F. Zhao et al. / International Journal of Applied Earth Observation and Geoinformation 31 (2014) 78–85This mode leads to a local, but not necessarily a global minimum inthe inversion process (Nocedal and Wright, 2006).In order to construct an optimization algorithm with high computational efficiency as well as the ability to reach the globalminimum of the merit function, the method of simulated annealing(Metropolis et al., 1953) is imposed on the classical downhill simplex algorithm (Nelder and Mead, 1965) in our study, resulting inan efficient global optimization algorithm. Compared with classicaloptimization algorithms, the proposed algorithm should be morecapable of finding the global minimum instead of the local one.Instead of choose downhill direction in every step, an uphill step isaccepted in this algorithm. The probability of the uphill steps (Pup )is determined by T, a controllable factor analogous to the temperature used in the simulated annealing algorithm. When T reducesto 0, the algorithm naturally converges to the classical downhillsimplex algorithm.The accuracy of PROSPECT inversion could be further improvedthrough assigning a specific merit function for each retrievedparameter, by which the interaction effects between differentparameters could be alleviated (Li and Wang, 2011). Therefore,instead of minimizing the single merit function to determineN, Ca b , Cw , and Cm simultaneously from the spectral bands of400–2500 nm, a specific merit function was assigned for eachparameter defined on its own sensitive spectral band determinedby the sensitivity analysis result, and then the foliar biochemical parameters were retrieved in steps by minimizing their ownspecific merit functions.A detailed description of the inversion approach is as follows:(1) Merit functions definition. For a parameter x of N, Ca b , Cw or Cmwith its sensitive wavelengths set Wx , which contains all wavelengths in its sensitive wavelength region previously shown inTable 2 with a step of 1 nm, its specific merit function is definedasJ(x) (Rmeas ( ) Rmod ( ))2(2) Wxwhere Rmeas ( ) is the measured reflectance at the wavelengthof , and Rmod ( ) is the modeled one.(2) Initial parameter value selection. Select initial guesses (Nguess ,Ca b guess , Cw guess , and Cm guess ) for N, Ca b , Cw and Cm : these values could be determined from a priori information providedthat such information exists, or randomly selected from theirranges we have defined for artificial data simulation. As theoptimization algorithm we used is a global one, the method ofselecting initial values would not affect the retrieval result.(3) Initial T (T0 ) selection. Select T0 for the simulated annealingalgorithm, which should not be too high since it will cost toomuch computational time, and should not be too low so that thealgorithm could have enough time to find the global minimum.T0 was assigned 6 in our study.(4) Parameter retrieval. Retrieve foliar biochemical parameters oneby one, by minimizing their specific merit functions definedin step (1). The retrieval order was determined by their totalsensitivities over all the wavelength bands of 400–2500 nmdetermined by sensitivity analysis results presented previously,with the most sensitive parameter retrieved first. This step consists of four sub-steps:(4.1) N determination. Determine the parameter N by minimizing its specific merit function, while keeping the otherparameters at their initial values.(4.2) Ca b determination. Determine the parameter Ca b byminimizing its specific merit function,

of Instrumentation Science and Opto-electronics Engineering, Beihang University, Beijing 100191, PR China b USDA-Agricultural ResearchService, Crop Production Systems Unit, 141 Experiment Station Road, Stoneville, MS 38776, USA a r t i c l e i n f o Article history: Received 23 December 2013 Accepted 12 March 2014 Keywords: Crop injury .

Related Documents:

0065-2113 Advances in Agronomy (BOOK SERIES) ScienceDirect 2001 2013 0737-6146 Advances in Anesthesia ClinicalKey 2007 0065-2164 Advances in Applied Microbiology (BOOK SERIES) ScienceDirect 2000 2013 . 0268-2605 Applied Organometallic Chemistry Wiley 1996 2009 0969-8043 Applied Radiation and Isotopes ScienceDirect 1999 2013

ScienceDirect database with the search term {mass spectrometry skin deposition}. A few publications were tracked by recent articles, which were not listed in ScienceDirect but used in this survey. Several publications were listed in ScienceDirect , but were not from peer-reviewed journals and thus not included in this survey. In total, 11,8886

The OPTION section lists repair parts of options. The BREAKDOWN section lists available repair parts of major components. All sections in this manual consist of parts illustrations and corresponding parts lists. The parts illustrations are numbered with Ref numbers which refer to corresponding parts lists. The parts lists include a number of .

ADVANCE STEEL TRAINING GUIDE DAY 4 7 BOM Editor Introduction Advance Steel has a powerful module for reports such as material lists, plate lists, cut lists, assembly lists, bolt lists etc. The preset templates for these reports are available in the BOM Editor. This tutorial describes the BOM Editor components and explains how to customize the templates for content and format.

Special section: Review Neuroplasticity as a function of second language learning: Anatomical changes in the human brain Ping Li*, Jennifer Legault and Kaitlyn A. Litcofsky Center for Brain, Behavior, and Cognition, Center for Language Science, Department of Psychology, Pennsylvania State University, United States article info Article history: Received 18 January 2014 Reviewed 31 March 2014 .

Drilling is considered as a three dimensional cutting process involving complex cutting mechanism because of the complex cutting edge geometries. Chip shape is the most important factor for smooth drilling process. Drilling will be smooth if it forms well broken chips during process [5]. However, most ductile materials such as austenitic

The list dynamically adjusts its size as elements are added or removed Lists have a lot of built-in functionality to make using them more straightforward. Piech Sahami, CS106A, Stanford University Show Me the Lists! Creating lists –Lists start/end with brackets. Elements separated by commas.

Heading 1 Syllabus Heading 2 DL 101 - Introducion to Online Learning Heading 3 Course Informaion: Format lists as proper lists . a primary way for people using screen reading software to navigate a page of text. Lists Format lists as proper lists. Formatting is conveyed to assistive technologies and mobile devices so they can present