黃 婷,梁 亮,耿 笛,李 麗,王李娟,王樹果,羅 翔,楊敏華
·農(nóng)業(yè)信息與電氣技術(shù)·
波段寬度對利用植被指數(shù)估算小麥LAI的影響
黃 婷1,梁 亮1※,耿 笛1,李 麗2,王李娟1,王樹果1,羅 翔3,楊敏華4
(1. 江蘇師范大學(xué)地理測繪與城鄉(xiāng)規(guī)劃學(xué)院,徐州 221000;2. 遙感科學(xué)國家重點(diǎn)實(shí)驗(yàn)室,北京 100101;3. 江西省農(nóng)業(yè)科學(xué)院農(nóng)業(yè)工程研究所,南昌 330000;4. 中南大學(xué)地球科學(xué)與信息物理學(xué)院,長沙 410083)
為了能夠根據(jù)遙感數(shù)據(jù)類型實(shí)現(xiàn)指數(shù)的優(yōu)化選擇進(jìn)而提高葉面積指數(shù)的反演精度,該研究分析了不同波段寬度(5~80 nm)對植被指數(shù)反演葉面積指數(shù)精度的影響。通過比較反演模型的決定系數(shù)均值,篩選出14個(gè)模型精度較高的植被指數(shù),并探討了不同波段寬度的選取對各指數(shù)葉面積指數(shù)反演精度的影響。結(jié)果表明,波段寬度對不同植被指數(shù)的影響可分為3類:1)OSAVI2等指數(shù)波寬越窄,反演精度越高,更適合應(yīng)用于高光譜遙感數(shù)據(jù);2)SR[800,680]等指數(shù)隨著波段寬度的增加,反演精度先升后降,最適波寬為35 nm,適用于中等光譜分辨率的遙感數(shù)據(jù);3)SR[675,700]等指數(shù)隨著波段寬度的增大,反演精度不斷提高,在多光譜數(shù)據(jù)中有更好的應(yīng)用潛力。
波段寬度;植被指數(shù);葉面積指數(shù);PROSAIL模型
葉面積指數(shù)(leaf area index,LAI)可有效反映植被生理生化特性,是植被的重要結(jié)構(gòu)特征參數(shù)之一。快速、無損、精準(zhǔn)地監(jiān)測冬小麥關(guān)鍵生育期的葉面積指數(shù)對準(zhǔn)確掌握長勢動(dòng)態(tài)、水肥調(diào)控、災(zāi)害監(jiān)測和產(chǎn)量預(yù)測等田間生產(chǎn)管理具有重要意義[1-3]。遙感技能以較低的成本獲得LAI的時(shí)空變化信息,目前已成宏觀尺度上獲取這一指標(biāo)最常用的方法[4-6]。
如何更準(zhǔn)確地通過遙感數(shù)據(jù)來估算LAI一直是植被遙感的熱點(diǎn)內(nèi)容之一[1,7-9]。目前,為了提高LAI的反演模型的精度與普適性,研究者一方面不斷地優(yōu)化與改進(jìn)反演策略與算法以降低模型誤差[10-12];另一方面則致力于分析土壤背景[13]、土壤類型[14]、觀測幾何[15]、熱點(diǎn)效應(yīng)[16]等因素在LAI反演過程的作用,以期找出相應(yīng)的方法減少或消除干擾因素的影響。有研究表明,植被指數(shù)類型的選取及其計(jì)算時(shí)波段寬度的選擇也是影響植被參量反演的重要因素[17-19]。目前,植被指數(shù)的選取或僅依靠經(jīng)驗(yàn)值或需要通過對各種指數(shù)進(jìn)行篩選才能確定,增加了反演過程的不確定性與復(fù)雜性。而由于波段寬度的影響,利用同一指數(shù)進(jìn)行LAI反演,其結(jié)果也往往存在較大差異。如Twele等[18]利用NDVI(normalized difference vegetation index),SR(simple ratio),RSR(reduced simple ratio index),NDVIc(corrected normalized difference vegetation index),SAVI2(soil adjusted vegetation index 2)估算森林冠層LAI時(shí)發(fā)現(xiàn),該類植被指數(shù)在窄波段時(shí)可獲得更高的反演精度;王福民等[20]對水稻LAI的反演研究表明,利用波段寬度為15 nm的光譜所計(jì)算的NDVI可取得最佳結(jié)果。由于不同衛(wèi)星數(shù)據(jù)波段寬度不同,這一影響也導(dǎo)致某一研究通過窮盡法篩選出的最優(yōu)指數(shù),在實(shí)際的遙感應(yīng)用過程中通常不具備普適性[21-22]。因此,系統(tǒng)地分析LAI反演時(shí)波段寬度對各植被指數(shù)的影響,探討不同指數(shù)的最適波段寬度,對提高LAI反演精度具有重要意義,是一亟待研究的問題。
本研究將利用地面實(shí)測數(shù)據(jù)集對LAI反演時(shí)較常用的植被指數(shù)進(jìn)行分析,研究各不同波段寬度對LAI反演精度的影響,從而為LAI估算時(shí)不同光譜分辨率傳感器指數(shù)的選擇以及遙感估算小麥LAI時(shí)植被指數(shù)的篩選(即針對不同遙感數(shù)據(jù)源來選擇合適的植被指數(shù))提供依據(jù)。
本研究數(shù)據(jù)來自國家農(nóng)業(yè)信息化工程技術(shù)研究中心開展的“作物田間信息獲取與基于影像GIS的快速診斷系統(tǒng)”農(nóng)學(xué)遙感實(shí)驗(yàn)。試驗(yàn)區(qū)(40°10'31"N~40°11'18"N,116°26'10"E~116°27'05"E)占地面積 167 hm2,海拔高度30~100 m,種植作物為冬小麥(圖1)。為保證小麥LAI值有較大變化范圍以便開展農(nóng)學(xué)遙感分析,在試驗(yàn)基地的24個(gè)小區(qū)(面積60 m×60 m)內(nèi)分別進(jìn)行了氮脅迫與水脅迫試驗(yàn)。2類脅迫各設(shè)置6個(gè)處理(施氮量:0~375 kg/hm2,級差75 kg;澆水量:0~1 125 m3/hm2,級差225 m3),每一處理包括2個(gè)重復(fù)(圖1)。
利用的FieldSpec Pro FR 地物光譜儀(ASD公司生產(chǎn),光譜范圍350~2 500 nm;350~1 000 nm分辨率為3 nm,采樣間隔為1.4 nm;1 000~2 500 nm分辨率為10 nm,采樣間隔為2 nm)對生長季(拔節(jié)后至孕穗前)的小麥進(jìn)行光譜采集。光譜采集在風(fēng)力小于3級,無卷云與濃積云的晴朗天氣下進(jìn)行,時(shí)間范圍規(guī)定為北京時(shí)間10:00~15:00,以保證有較高的太陽高度角。傳感器探頭(25° 視場角)垂直向下,高度保持在冠層上方1.3 m附近,每一樣本重復(fù)測量10次取均值,且每半小時(shí)用參考板對儀器進(jìn)行一次校正,以消除環(huán)境變化所帶來的影響。光譜采集的同時(shí)進(jìn)行農(nóng)學(xué)采樣,以干重法測定LAI值,即取50~100片葉進(jìn)行面積測量后,烘干稱重,建立干質(zhì)量與葉面積之間的相關(guān)模型,然后再根據(jù)被測對象的干質(zhì)量反推葉面積,并采用激光葉面積儀(Cl-203型)進(jìn)行矯正。在試驗(yàn)期間,在小麥的水、氮脅迫區(qū)內(nèi)同步采樣6次(采樣日期分別為4月11日、4月21日、5月4日、5月13日、5月23和6月3日),在大田均布點(diǎn)上同步采樣一次(4月11日),共獲取有效樣本139份。圖2為各小麥樣本的光譜反射率曲線圖。
圖1 研究區(qū)概況
圖2 小麥冠層的光譜曲線
葉面積指數(shù)與利用地表反射率計(jì)算的植被指數(shù)之間有很強(qiáng)的相關(guān)性,經(jīng)驗(yàn)反演方法則通過建立LAI和植被指數(shù)之間的某種函數(shù)關(guān)系能夠較好的估算出LAI。但植被指數(shù)的選擇通常不唯一,目前對于最適合葉面積反演的植被指數(shù)還沒有一致的結(jié)論[1,7]。本研究在前人的研究基礎(chǔ)上,選擇了28個(gè)LAI反演較常用的植被指數(shù)[4],用于不同波段寬度下植被指數(shù)與葉面積指數(shù)的相關(guān)關(guān)系研究(表1)。在建模時(shí),將植被指數(shù)作為自變量,將實(shí)測LAI作為因變量,分別利用線性回歸、指數(shù)回歸、對數(shù)回歸、多項(xiàng)式回歸和冪函數(shù)回歸建立植被指數(shù)和LAI的曲線擬合模型,并采用均方根誤差(root mean square error,RMSE)和決定系數(shù)(coefficient of determination,R)這2個(gè)統(tǒng)計(jì)量作為模型精度評價(jià)指標(biāo),從而篩選出R最大,RMSE最小的最優(yōu)曲線擬合模型,各植被指數(shù)的最佳擬合模型如表2所示。
表1 本研究選用的植被指數(shù)
注:為光譜反射率,下標(biāo)為波長。
Note:is the spectral reflectance, the subscript is the wavelength.
表2 植被指數(shù)的最佳擬合模型
為研究不同波段寬度對植被指數(shù)反演LAI精度的影響,需獲得利用不同波段寬度的冠層反射率計(jì)算出的植被指數(shù),由于傳感器各通道受元器件特性的制約,每個(gè)通道在特定光譜區(qū)間對不同光譜輻射的響應(yīng)能力不同,為了能夠更加接近衛(wèi)星傳感器所接收的輻射信號。本研究將地面實(shí)測小麥的冠層光譜反射率(350~2 500 nm)根據(jù)式(1)和式(2)模擬生成不同波段寬度時(shí)的葉片反射率[19,38]。冠層光譜的初始波段寬度設(shè)置為5 nm,并以5 nm為步長逐步增至80 nm,逐一計(jì)算不同波寬下的植被指數(shù)。其中,參與植被指數(shù)計(jì)算的波長設(shè)置為波段寬度拓展時(shí)的中心波長,同時(shí),中心波長的反射率為根據(jù)式(1)模擬生成的光譜反射率。這樣的多波段寬度設(shè)置既保證了波段寬度的豐富度和連續(xù)性,又模擬了絕大多數(shù)傳感器的光譜通道寬度,有助于確定不同傳感器數(shù)據(jù)源反演葉面積指數(shù)時(shí)的最佳植被指數(shù),從而提高LAI反演精度。
為了能定量地比較和評估植被指數(shù)對波段寬度的敏感度,本研究將波段寬度為5~80 nm時(shí)計(jì)算的植被指數(shù)值與實(shí)測原始波段寬度(1 nm)時(shí)計(jì)算的植被指數(shù)值進(jìn)行比較,根據(jù)式(3)定義敏感度系數(shù)Var計(jì)算方法如下[19]
同時(shí),定量分析植被反射光譜對理化參數(shù)的敏感性是遙感反演理化參數(shù)含量的前提[39]。本研究采用靈敏度系數(shù)LAI定量描述光譜指數(shù)對LAI的敏感性,其計(jì)算公式如下[18]
圖3 植被指數(shù)的最佳擬合模型的精度
圖4為篩選出的14個(gè)植被指數(shù)對波段寬度的敏感度系數(shù)Var變化圖。從圖中可以發(fā)現(xiàn),植被指數(shù)的Var與波段寬度基本呈正相關(guān)關(guān)系,而Var越大說明植被指數(shù)對波段寬度的抗干擾性越差,反之則越好。上述研究表明,波段寬度是影響植被指數(shù)的重要因素之一,且隨著波段寬度的增加,本研究所篩選的植被指數(shù)受波段寬度的干擾越大。
圖4 各植被指數(shù)對波段寬度的敏感度
根據(jù)式(4)計(jì)算的敏感度系數(shù)LAI結(jié)果如圖5所示,Carte2指數(shù)的敏感度系數(shù)與波段寬度呈正相關(guān)關(guān)系。與之相反的是,OSAVI2、Carte3、NDCI、SR[752,690]、SR[800,680]、NDVI705、SR[750,550]、SR[750,700]、SR[675,700]、Datt3、Carte4、SR[750,710]隨著波段寬度的增加敏感度度系數(shù)LAI呈下降趨勢。而RI1dB的敏感度系數(shù)曲線雖總體呈上升趨勢,但在30~60 nm之間存在明顯的波谷。因此,植被指數(shù)在各波段寬度下對LAI的敏感度曲線變化趨勢,同樣說明了波段寬度也會(huì)造成植被指數(shù)對LAI的響應(yīng)程度發(fā)生改變。因此,經(jīng)過對Var和LAI隨波段寬度變化的初步分析可以猜測,波段寬度是影響LAI估算精度的重要因素,且波段寬度對不同植被指數(shù)估算LAI精度的影響趨勢可能是不同的,可能存在以下2種情況:波段寬度越大,效果越好;波段寬度越小,效果越差,但波段寬度對利用植被指數(shù)進(jìn)行LAI估算的具體影響仍需進(jìn)一步的研究分析。
圖5 各波段寬度下植被指數(shù)對LAI的敏感度
圖7~圖10是不同指數(shù)所建LAI反演模型2隨波段寬度變化而變化情況。根據(jù)2的變化趨勢,各指數(shù)可分為4類:1)所建反演模型2隨著波段寬度增加不斷降低,即所用波段寬度越窄越合適,這類指數(shù)可稱之為窄波段指數(shù);2)2隨著波段寬度的增加先升后降,變化曲線存在明顯峰值,可稱之為中波段指數(shù);3)2隨著波段寬度增加而升高,即在本研究的分析范圍內(nèi)(波段寬度≤80 nm),波段越寬越合適,可稱之為寬波段指數(shù);4)R隨著波段寬度的增加先下降后上升再下降的植被指數(shù)。
圖6 敏感度系數(shù)均值
圖7為窄波段植被指數(shù)OSAVI2、NDCI、SR[752,690]、SR[750,700]和Carte2所建LAI估算模型的R隨波段寬度增加的變化圖。由圖可知,隨著波段寬度的增加,窄波段植被指數(shù)所建模型R的變化趨勢基本相同,均呈現(xiàn)下降趨勢。說明波段寬度越窄,由這類指數(shù)所構(gòu)建模型估算LAI的能力越好。因此,利用高光譜遙感數(shù)據(jù)進(jìn)行LAI估算時(shí),可優(yōu)先考慮窄波段植被指數(shù),以期獲得更好的估算結(jié)果。從圖中可以發(fā)現(xiàn),指數(shù)OSAVI2所建LAI反演模型的R始終最高,其次為NDCI、SR[752,690]、Carte2和SR[750,700]。同時(shí),值得注意的是,窄波段植被指數(shù)OSAVI2和NDCI隨著波段寬度的增加其R的波動(dòng)較小,因此,當(dāng)OSAVI2和NDCI在不同的遙感數(shù)據(jù)源下估算LAI的差異可能較小。但SR[752,690]、Carte2、SR[750,700]隨波段寬度的增加,R下降趨勢明顯,說明這兩個(gè)指數(shù)所構(gòu)建的LAI反演模型極易受波段寬度的影響,在使用不同傳感器數(shù)據(jù)源估算LAI時(shí),其結(jié)果可能差異較大。因此,綜合植被指數(shù)的敏感度分析及R隨波寬變化的結(jié)果,可確定OSAVI2和NDCI為LAI高光譜反演時(shí)的優(yōu)選指數(shù)。
圖7 窄波段植被指數(shù)LAI估算模型R2隨波寬的變化
圖8 中波段植被指數(shù)LAI估算模型R2隨波寬的變化
圖9為寬波段植被指數(shù)所建LAI反演模型R隨波段寬度的變化圖。寬波段植被指數(shù)SR[750,550]、SR[675,700]的顯著特點(diǎn)是其LAI反演模型R與LAI呈正相關(guān)關(guān)系,而植被指數(shù)SR[750,710]和RI1dB在波段寬度20~60 nm之間有所波動(dòng),但總體呈現(xiàn)上升趨勢,本研究將這2個(gè)指數(shù)也劃分為寬波段植被指數(shù)。隨著波段寬度的增大,寬波段植被指數(shù)所建模型的估算能力越好(R越大),說明寬波段植被指數(shù)在反演LAI時(shí),波段寬度越大越合適。因此當(dāng)利用多光譜數(shù)據(jù)進(jìn)行LAI估算時(shí),寬波段植被指數(shù)可能發(fā)揮出更好的估算潛力。同時(shí)由圖可知,當(dāng)波段寬度小于40 nm時(shí),SR[750,550]所建模型R始終高于SR[675,700],波段寬度大于40 nm時(shí),SR[750,550]的R始終低于SR[675,700],說明利用單一波段寬度比較不同植被指數(shù)反演LAI的能力往往存在局限性,且波段寬度對不同植被指數(shù)的影響程度也有所差異。同時(shí),對比SR[675,700]、SR[750,550]、SR[750,710]和RI1dB所建模型R的變化曲線,SR[750,550]、SR[750,710]和RI1dB的曲線變化趨勢較為平緩,反演LAI的能力較為穩(wěn)定。值得注意的是,雖然SR[675,700]是作為高光譜指數(shù)所提出[30],但本研究分析表明,波段寬度對SR[675,700]具有較大影響,當(dāng)所采用的遙感數(shù)據(jù)光譜通道小于20 nm 時(shí)(如Hyperion與CHRIS等高光譜數(shù)據(jù)),該指數(shù)并非估算LAI的優(yōu)選指數(shù);當(dāng)數(shù)據(jù)的光譜通道大于50 nm時(shí)(如SPOT和Landsat OLI),該指數(shù)則具有較高的反演精度,是進(jìn)行LAI估算的優(yōu)選指數(shù)。
圖9 寬波段植被指數(shù)LAI估算模型R2隨波寬的變化
圖10為Carte3與Carte4所建LAI估算模型的R隨波段寬度的變化圖。Carte3與Carte4的R隨波段寬度變化的趨勢較為相似,均呈現(xiàn)先下降在上升再下降的變化趨勢,但該類植被指數(shù)所建模型在波段寬度5~80 nm之間估算能力較為穩(wěn)定,其最大R與最小R的差值均小于0.003。因此,當(dāng)利用Carte3與Carte4在不同遙感數(shù)據(jù)下進(jìn)行LAI的估算時(shí),LAI估算結(jié)果可能差距較小,估算精度較為穩(wěn)定,可忽略波段寬度的影響。
圖10 Carte3和Carte4的LAI估算模型R2隨波寬的變化
本研究結(jié)果表明,波段寬度是影響LAI反演精度的重要因素之一,王福民等[20]對水稻的分析表明,當(dāng)波段寬度取值為15 nm時(shí),NDVI可取得最佳結(jié)果,而劉玉琴等[7]的分析則表明窄波段寬度下選用的植被指數(shù)能更好地實(shí)現(xiàn)草地LAI的反演。目前,大量研究利用高光譜數(shù)據(jù)下的植被指數(shù)進(jìn)行LAI估算[40-42],但大多研究僅針對一個(gè)或少數(shù)的幾個(gè)植被指數(shù),各植被指數(shù)受波段寬度的影響尚未得到系統(tǒng)的梳理。本研究選取了28個(gè)常用于LAI研究的植被指數(shù),并將波段寬度的設(shè)置為5~80 nm之間連續(xù)的16種波段寬度,較為全面的探討了各類植被指數(shù)估算LAI能力隨波段寬度的變化趨勢,可為各指數(shù)合適波段寬度的選擇提供參考。另,進(jìn)一步分析表明,由于波段寬度對不同植被指數(shù)的影響大小存在差異(如當(dāng)波段寬度小于45 nm時(shí),SR[750,550]的模型估算精度明顯優(yōu)于SR[675,700],但當(dāng)波段寬度大于45 nm時(shí),結(jié)果卻恰恰相反),故在以提高參量反演精度為目標(biāo)的指數(shù)篩選擇優(yōu)過程中,不但要考慮植被指數(shù)的種類,還需要綜合考慮波段寬度的影響。
本研究植被指數(shù)所建模型R隨波段寬度的變化趨勢主要有以下3種情況:1)OSAVI2和NDCI等所建模型R隨著波段寬度增加不斷降低,最適波段寬度越窄越好;2)Datt3和SR[800,680]等所建模型R隨著波段寬度增加先升后降,最適波段寬度位于R峰值處;3)SR[750,550]和SR[675,700]所建模型R隨著波段寬度增加不斷增加,最適波段寬度越寬越好。目前,寬波段/窄波段植被指數(shù)通常為利用寬波段/窄波段傳感器數(shù)據(jù)可以計(jì)算得到的植被指數(shù)[43]。其中,NDVI705、SR[800,680]和SR[750,710]等植被指數(shù)通常被定義為高光譜/窄波段植被指數(shù)[44],但分析表明,NDVI705在波段寬度為35 nm時(shí)具有更好的LAI估算能力,SR[750,710]在寬波段處所建模型R更大,說明部分高光譜指數(shù)的植被指數(shù)在多光譜可能存在很強(qiáng)的應(yīng)用潛力。這一結(jié)果可為不同遙感數(shù)據(jù)類型下植被指數(shù)的優(yōu)選提供指導(dǎo)。
結(jié)合光譜學(xué)的知識可知,波段越窄,定位敏感信息的能力越強(qiáng),波段越寬,則對某一光譜區(qū)域的信息利用得更為充分[45]。對OSAVI2與 NDCI等窄波段指數(shù)的分析表明,這類指數(shù)的反演精度主要取決于能否準(zhǔn)確地定位敏感波,因此波段越窄,精度越高;而SR[750,710]和RI1dB等寬波段指數(shù),反演精度的提高更多地取決于是否充分利用了該光譜區(qū)域的信息,因此波段越寬,信息利用越充分,反演精度也就越高;Datt3與SR[800,680]等中波段指數(shù),則需要尋找敏感波段與光譜信息充分利用兩者之間的平衡點(diǎn),因此出現(xiàn)先升后降,并在35 nm附近存在峰值的情況。
本研究分析了不同植被指數(shù)對LAI與波段寬度的敏感性,研究了利用這些指數(shù)進(jìn)行LAI估算時(shí),波段寬度變化對模型精度的影響,并探討了各類植被指數(shù)所建反演模型R隨波段寬度增加的變化趨勢,為不同數(shù)據(jù)源下LAI反演的指數(shù)選擇提供了參考。文章主要結(jié)論如下:
1)波段寬度是影響LAI反演精度的重要因素之一。分析表明,利用植被指數(shù)進(jìn)行小麥LAI估算時(shí),反演模型的精度不僅與選用的植被指數(shù)有關(guān),而且與計(jì)算該指數(shù)的波段寬度有關(guān)。在利用植被指數(shù)進(jìn)行LAI反演時(shí),應(yīng)根據(jù)傳感器的通道寬度與光譜分辨率選擇最佳的植被指數(shù)。
2)波段寬度的變化對各植被指數(shù)的影響具有明顯差異,可分為4種類型:①窄波段指數(shù),所建反演模型精度隨著波段寬度增加不斷降低,這一類型包括指數(shù)OSAVI2、NDCI、SR[752,690]、SR[750,700]和Carte2;②中波段指數(shù)所建模型精度隨著波段寬度增加先升后降,這一類型主要包括指數(shù)Datt3、SR[800,680]與NDVI705,其最適波寬約為35 nm;③寬波段指數(shù)所建模型精度隨著波段寬度增加而升高,這一類型包括指數(shù)SR[750,550]、SR[675,700]、SR[750,710]和RI1dB;④植被指數(shù)Carte3與Carte4所建模型的R在波段寬度5~80 nm雖然先下降后上升再下降,但在各波段寬度下其估算精度均較為穩(wěn)定,因此可忽略波段寬度對該類植被指數(shù)的影響。
3)研究結(jié)果表明,利用植被指數(shù)進(jìn)行LAI反演時(shí),應(yīng)根據(jù)傳感器的通道寬度與光譜分辨率選擇最佳的植被指數(shù)。其中OSAVI2與NDCI等指數(shù)波寬越窄,LAI反演精度越高,更適合應(yīng)用于高光譜遙感數(shù)據(jù);Datt3等指數(shù)的最適波寬約為35 nm,更適用于中等/多光譜分辨率的遙感數(shù)據(jù);SR[750,710]和RI1dB等指數(shù)波寬越寬,LAI反演精度越高,在多光譜遙感數(shù)據(jù)中有更好的應(yīng)用潛力。
[1]束美艷,顧曉鶴,孫林,等. 基于新型植被指數(shù)的冬小麥LAI高光譜反演[J]. 中國農(nóng)業(yè)科學(xué),2018,51(18):3486-3496. Shu Meiyan, Gu Xiaohe, Sun Lin, et al. High spectral inversion of winter wheat LAI based on new vegetation index[J]. Scientia Agricultura Sinica, 2018, 51(18): 3486-3496. (in Chinese with English abstract)
[2]李衛(wèi)國,趙春江,王紀(jì)華,等. 基于衛(wèi)星遙感的冬小麥拔節(jié)期長勢監(jiān)測[J]. 麥類作物學(xué)報(bào),2007,27(3):523-527. Li Weiguo, Zhao Chunjiang, Wang Jihua, et al. Monitoring the growth condition of winter wheat in jointing stage based on Landsat TM image[J]. Journal of Triticeae Crops, 2007, 27(3): 523-527. (in Chinese with English abstract)
[3]Krishnan P, Sharma R K, Dass A, et al. Web-based crop model: Web Info Crop-Wheat to simulate the growth and yield of wheat[J]. Computers and Electronics in Agriculture, 2016, 127: 324-335.
[4]Liang Liang, Di Liping, Zhang Lianpeng, et al. Estimation of crop LAI using hyperspectral vegetation indices and a hybrid inversion method[J]. Remote Sensing of Environment, 2015, 165(8): 123-134.
[5]Mutanga O, Skidmore A K, Prins H H T. Predicting in situ pasture quality in the Kruger National Park, South Africa, using continuum-removed absorption features[J]. Remote Sensing of Environment, 2004, 89(3): 393-408.
[6]Fang Hongliang. Retrieving leaf area index using a genetic algorithm with a canopy radiative transfer model[J]. Remote Sensing of Environment, 2003, 85(3): 257-270.
[7]劉玉琴,沙晉明,余濤,等. 基于寬波段和窄波段植被指數(shù)的草地LAI 反演對比研究[J]. 遙感技術(shù)與應(yīng)用,2014,29(4):587-593. Liu Yuqin, Sha Jinming, Yu Tao, et al. Comparing the performance of broad-band and narrow-band vegetation indices for estimation of grass LAI[J]. Remote Sensing Technology and Application, 2014, 29(4): 587-593. (in Chinese with English abstract)
[8]劉振波,鄒嫻,葛云健,等. 基于高分一號WFV影像的隨機(jī)森林算法反演水稻LAI[J]. 遙感技術(shù)與應(yīng)用,2018,33(3):458-464. Liu Zhenbo, Zou Xian, Ge Yunjian, et al. Retrieval rice leaf area index using random forest algorithm based on GF-1 WFV remote sensing data[J]. Remote Sensing Technology & Application, 2018, 33(3): 458-464. (in Chinese with English abstract)
[9]Roosjen P P J, Brede B, Suomalainen J M, et al. Improved estimation of leaf area index and leaf chlorophyll content of a potato crop using multi-angle spectral data-potential of unmanned aerial vehicle imagery[J]. International Journal of Applied Earth Observation and Geoinformation, 2018, 66: 14-26.
[10]Tang S, Chen J M, Zhu Q, et al. LAI inversion algorithm based on directional reflectance kernels[J]. Journal of Environmental Management, 2007, 85(3): 638-648.
[11]Leonenko G, Lows S O, North P R J. Retrieval of leaf area index from MODIS surface reflectance by model inversion using different minimization criteria[J]. Remote Sensing of Environment, 2013, 139(12): 257-270.
[12]Woodgate W, Disney M, Armston J D, et al. An improved theoretical model of canopy gap probability for Leaf area index estimation in woody ecosystems[J]. Forest Ecology and Management, 2015, 358: 303-320.
[13]Huete A R. A soil-adjusted vegetation index (SAVI)[J]. Remote Sensing of Environment, 1988, 25(3): 295-309.
[14]高林,王曉菲,顧行發(fā),等. 植冠下土壤類型差異對遙感估算冬小麥葉面積指數(shù)的影響[J]. 植物生態(tài)學(xué)報(bào),2017,41(12):1273-1288. Gao Lin, Wang Xiaofei, Gu Xingfa, et al. Exploring the influence of soil types underneath the canopy in winter wheat leaf area index remote estimating[J]. Chinese Journal Plant Ecology, 2017, 41(12): 1273-1288. (in Chinese with English abstract)
[15]楊貴軍,黃文江,王紀(jì)華,等. 多源多角度遙感數(shù)據(jù)反演森林葉面積指數(shù)方法[J]. 植物學(xué)報(bào),2010,45(5):566-578. Yang Guijun, Huang Wenjiang, Wang Jihua, et al. Inversion of forest leaf area index calculated from multi-source and multi-angle remote sensing data[J]. Bulletin of Botany, 2010, 45(5): 566-578. (in Chinese with English abstract)
[16]趙娟,張耀鴻,黃文江,等. 基于熱點(diǎn)效應(yīng)的不同株型小麥LAI反演[J]. 光譜學(xué)與光譜分析,2014,35(1):207-211. Zhao Juan, Zhang Yehong, Huang Wenjiang, et al. Inversion of LAI by considering the hotspot effect for different geometrical wheat[J]. Spectroscopy and Spectral Analysis, 2014, 35(1): 207-211. (in Chinese with English abstract)
[17]Zhao Dehua, Huang Liangmei, Li Jianlong, et al. A comparative analysis of broadband and narrowband derived vegetation indices in predicting LAI and CCD of a cotton canopy[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2007, 62(1): 25-33.
[18]Twlele A, Erasmi S, Kappasa M. Spatially explicit estimation of leaf area index using EO-1 Hyperion and Landsat ETM+ data: Implications of spectral bandwidth and shortwave infrared data on prediction accuracy in a tropical montane environment[J]. GIScience & Remote Sensing, 2008, 45(2): 229-248.
[19]Du Huishi, Jiang Hailing, Zhang Lifu, et al. Evaluation of spectral scale effects in estimation of vegetation leaf area index using spectral indices methods[J]. Chinese Geographical Science, 2016, 26(6): 731-744.
[20]王福民,黃敬峰,唐延林,等. 采用不同光譜波段寬度的歸一化植被指數(shù)估算水稻葉面積指數(shù)[J]. 應(yīng)用生態(tài)學(xué)報(bào),2007,18(11):2444-2450. Wang Fuming, Huang Jinfeng, Tang Yanlin, et al. Estimation of rice LAI by using NDVI at different spectral bandwidths[J]. Chinese Journal of Applied Ecology, 2007, 18(11): 2444-2450. (in Chinese with English abstract)
[21]Liang Liang, Di Liping, Huang Ting, et al. Estimation of leaf nitrogen content in wheat using new hyperspectral indices and a random forest regression algorithm[J]. Remote Sens, 2018, 10(12): 1940-1956.
[22]Liang Liang, Qin Zhihao, Zhao Shuhe, et al. Estimating crop chlorophyll content with hyperspectral vegetation indices and the hybrid inversion method[J]. International Journal of Remote Sensing, 2016, 37(13): 2923-2949.
[23]Sims D A, Gamon J A. Relationships between leaf pigment content and spectral reflectance across a wide range of species, leaf structures and developmental stages[J]. Remote Sensing of Environment, 2002, 81(2/3): 337-354.
[24]Marshak A, Knyazikhin Y, Davis A B, et al. Cloud-vegetation interaction: Use of normalized difference cloud index for estimation of cloud optical thickness[J]. Geophysical Research Letters, 2000, 27(12): 1695-1698.
[25]McMurtrey III J E, Chappelle E W, Kim M S, et al. Distinguishing nitrogen fertilization levels in field corn (L.) with actively induced fluorescence and passive reflectance measurements[J]. Remote Sensing of Environment, 1994, 47(1): 36-44.
[26]Carter G A. Ratios of leaf reflectance in narrow wavebands as indicators of plant stress[J]. International Journal of Remote Sensing, 1994, 15(3): 697-703.
[27]Gupta R K, Vijayan D, Prasad T S. New hyperspectral vegetation characterization parameters[J]. Advances in Space Research, 2001, 289(1): 201-206.
[28]Main R, Cho M A, Mathieu R, et al. An investigation into robust spectral indices for leaf chlorophyll estimation[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2011, 66(6): 751-761.
[29]Wu Chaoyang, Niu Zheng, Tang Quan, et al. Estimating chlorophyll content from hyperspectral vegetation indices: Modeling and validation[J]. Agricultural and Forest Meteorology, 2008, 148(8/9): 1230-1241.
[30]Chappelle E W, Kim M S, McMurtrey III J E. Ratio analysis of reflectance spectra (RARS): An algorithm for the remote estimation of the concentrations of chlorophyll A, chlorophyll B, and carotenoids in soybean leaves[J]. Remote Sensing of Environment, 1992, 39(3): 239-247.
[31]Haboudane D, Miller J R, Pattey E, et al. Hyperspectral vegetation indices and novel algorithms for predicting green LAI of crop canopies: Modeling and validation in the context of precision agriculture[J]. Remote Sensing of Environment, 2004, 90(3): 337-352.
[32]Gitelson A, Merzlyak M N. Quantitative estimation of chlorophyll using reflectance spectra: Experiments with autumn chestnut and maple leaves[J]. Journal of Photochemistry and Photobiology B: Biology, 1994, 22(3): 247-252.
[33]Gupta R K, Vijayan D, Prasad T S. Comparative analysis of red-edge hyperspectral indices[J]. Advances in Space Research, 2003, 32(11): 2217-2222.
[34]Zarco-Tejada P J, Miller J R. Land cover mapping at BOREAS using red edge spectral parameters from CASI imagery[J]. Journal of Geophysical Research, 1999, 104(D22): 27921-27933.
[35]Vogelmann J E, Rock B N, Moss D M. Red edge spectral measurements from sugar maple leaves[J]. International Journal of Remote Sensing, 1993, 14(8): 1563-1575.
[36]Datt B. A new reflectance index for remote sensing of chlorophyll content in higher plants: Tests usingleaves[J]. Journal of Plant Physiology, 1999, 154(1): 30-36.
[37]Daughtry C S T, Walthall C L, Kim M S, et al. Estimating corn leaf chlorophyll concentration from leaf and canopy reflectance[J]. Remote Sensing of Environment, 2000, 74(2): 229-239.
[38]肖艷芳,周德民,宮輝,等. 冠層反射光譜對植被理化參數(shù)的全局敏感性分析[J]. 遙感學(xué)報(bào),2015,19(3):368-374. Xiao Yanfang, Zhou Deming, Gong Hui, et al. Sensitivity of canopy reflectance to biochemical and biophysical variables[J]. Journal of Remote Sensing, 2015, 19(3): 368 -374. (in Chinese with English abstract)
[39]梁順林. 定量遙感[M]. 北京:科學(xué)出版社,2015.
[40]Haboudane D, Miller J R, Pattey E, et al. Hyperspectral vegetation indices and novel algorithms for predicting green LAI of crop canopies: Modeling and validation in the context of precision agriculture[J]. Remote Sensing of Environment, 2004, 90(3): 337-352.
[41]唐建民,廖欽洪,劉奕清,等. 基于CASI高光譜數(shù)據(jù)的作物葉面積指數(shù)估算[J]. 光譜學(xué)與光譜分析,2015,35(5):1351-1356. Tang Jianmin, Liao Qinhong, LiuYiqing, et al. Estimating leaf area index of crops based on hyperspectral compact airborne spectrographic imager (CASI) data[J]. Spectroscopy and Spectral Analysis, 2015, 35(5): 1351-1356. (in Chinese with English abstract)
[42]吳偉斌,李佳雨,張震邦,等. 基于高光譜圖像的茶樹LAI與氮含量反演[J]. 農(nóng)業(yè)工程學(xué)報(bào),2018,34(3):195-201. Wu Weibin, Li Jiayu, Zhang Zhenbang, et al. Estimation model of LAI and nitrogen content in tea tree based on hyperspectral image[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2018, 34(3): 195-201. (in Chinese with English abstract)
[43]Thenkabail P S, Smith R B, Pauw E D. Hyperspectral vegetation indices and their relationships with agricultural crop characteristics[J]. Remote Sensing of Environment, 2000, 71(2): 158-182.
[44]葛昊,盧珊,趙云升. 葉片茸毛對葉片反射光譜及高光譜植被指數(shù)的影響研究[J]. 光譜學(xué)與光譜分析,2012,32(2):153-158. Ge Hao, Lu Shan, Zhao Yunsheng. Effects of leaf hair on leaf reflectance and hyperspectral vegetation indices[J]. Spectroscopy and Spectral Analysis, 2012, 32(2): 153-158. (in Chinese with English abstract)
[45]普拉薩德,約翰,阿爾弗雷德. 高光譜植被遙感[M]. 北京:中國農(nóng)業(yè)科學(xué)技術(shù)出版社,2015.
Effects of band width on estimation of wheat LAI using vegetation index
Huang Ting1, Liang Liang1※, Geng Di1, Li Li2, Wang Lijuan1, Wang Shuguo1, Luo Xiang3, Yang Minhua4
(1.,,221000,; 2.,100101,; 3.,,330000,; 4.,,410083,)
To improve the accuracy and universality of the inversion model of the leaf area index, on the one hand, many researchers constantly optimized inversion algorithm, on the other hand, they were committed to analyzing the influence of interference factors such as soil background, soil type, observation geometry and hot spot effected on the inversion process of leaf area index. Band width is generally considered as an important factor affecting the inversion of vegetation parameters. However, there were few studies on the influence of band width on estimating leaf area index. To optimize the selection of vegetation indices based on the type of remote sensing data, the influence of different band widths on the inversion model established by vegetation index was analyzed. Firstly, the spectral reflectance of different band widths was simulated by the measured wheat spectral data set. The initial band width was set to 5 nm and gradually increased to 80 nm in 5 nm steps. On this basis, 28 vegetation indices commonly used for inversion of leaf area indices, such as SR[800680], NDCI and Carte2, were calculated. To select the vegetation index with greater potential to estimate the leaf area index, the mean value of the coefficient of determination was used as a prediction accuracy measure, and 14 vegetation indices such as OSAVI2, Carte3 and SR[800680]were screened out. Then, by analyzing the sensitivity of 14 indices and variation of coefficient of determination to band widths, the influence of band widths on the accuracy of the leaf area index estimated by vegetation indices was discussed. The results indicated that the band width was one of the important factors that affected the accuracy of the inversion of the leaf area index, and the influence of band width on vegetation indices was inconsistent. According to the trend of coefficient determination, the indices were divided into three categories: (1) coefficient of determination of inversion models built by vegetation indices decreased with the increase of band width. This type of indices included OSAVI2, NDVI, SR[752690], SR[750700]and Carte2, which was called narrow-band vegetation index. (2) coefficient of determination rose first and then falls with the increase of band width, and the change curve had an obvious peak value, which was called the mid-band vegetation index. This type of indices included Datt3, SR[800680]and NDVI705. (3) coefficient of determination rose with the increase of band width, which was called broad-band vegetation index. This type of indices included SR[750,550], SR[675,700], SR[750,710]and RI1dB; (4) coefficient of determination of the models built by Carte3 and Carte4 showed a trend of first decreasing, then rising followed by declining, the accuracy of estimating leaf area index was stable at different band widths, and difference between the maximum and minimum of coefficient of determination was less than 0.003, so the influence of the band width on this type of vegetation indices could be ignored. The results of this study indicated that when using vegetation index for inversion of leaf area index, we should also comprehensively consider channel width and spectral resolution of the sensor to select the best vegetation index. Furthermore, when the band width increased from 5 nm to 80 nm, the precision of the leaf area index inversion model of built by narrow-band vegetation index was higher with the narrower band width, and this type of indices was more suitable for hyperspectral remote sensing data. The optimal band width of the mid-band vegetation index was about 35 nm, and this type of indices was more suitable for remote sensing data with medium resolution. The precision of the leaf area index inversion model built by broad-band vegetation index was higher with the wider band width, and this type of indices had better application potential in multispectral remote sensing data. This research provided the basis for the selection of indices using different spectral resolution sensors data during estimation of leaf area index, and screening vegetation indices for wheat leaf area index inversion.
band width; vegetation index; leaf area index; PROSAIL model
黃 婷,梁 亮,耿 笛,李 麗,王李娟,王樹果,羅 翔,楊敏華. 波段寬度對利用植被指數(shù)估算小麥LAI的影響[J]. 農(nóng)業(yè)工程學(xué)報(bào),2020,36(4):168-177. doi:10.11975/j.issn.1002-6819.2020.04.020 http://www.tcsae.org
Huang Ting, Liang Liang, Geng Di, Li Li, Wang Lijuan, Wang Shuguo, Luo Xiang, Yang Minhua. Effects of band width on estimation of wheat LAI using vegetation index[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2020, 36(4): 168-177. (in Chinese with English abstract) doi:10.11975/j.issn.1002-6819.2020.04.020 http://www.tcsae.org
2019-12-16
2020-01-15
遙感科學(xué)國家重點(diǎn)實(shí)驗(yàn)室開放基金(OFSLRSS201804);江蘇省自然科學(xué)基金(BK20181474);國家自然科學(xué)基金(41401473);江蘇高校優(yōu)勢學(xué)科建設(shè)工程資助項(xiàng)目(PAPD)資助
黃 婷,主要研究方向?yàn)橹脖欢窟b感。Email:lllxwjhht@163.com
梁 亮,副教授,博士,主要研究方向?yàn)檗r(nóng)業(yè)遙感。Email:liangliang198119@163.com
10.11975/j.issn.1002-6819.2020.04.020
TP79
A
1002-6819(2020)-04-0168-10