易秋香
基于Sentinel-2多光譜數(shù)據(jù)的棉花葉面積指數(shù)估算
易秋香
(1. 中國科學院新疆生態(tài)與地理研究所,荒漠與綠洲生態(tài)國家重點實驗室,烏魯木齊 830011;2. 新疆維吾爾自治區(qū)遙感與地理信息系統(tǒng)應用重點實驗室,烏魯木齊 830011;3. 中國科學院大學,北京 100049)
棉花葉面積指數(shù)(leaf are index, LAI)的快速、準確獲取對棉花長勢監(jiān)測、發(fā)育期診斷、面積提取以及產(chǎn)量估算等遙感監(jiān)測具有重要意義。該研究利用2017年和2018年的Sentinel-2多光譜衛(wèi)星數(shù)據(jù)及大面積田間試驗觀測獲取的棉花不同發(fā)育期LAI實測數(shù)據(jù),構建了基于單波段反射率及各類植被指數(shù)的棉花不同發(fā)育期及全發(fā)育期LAI估算模型,并采用留一驗證(LOOCV, leave-one-out cross validation)和交叉驗證對模型精度進行了檢驗。結果表明:1)對于單波段反射率,基于中心波長為842 nm波寬為145 nm的B8近紅外波段對不同發(fā)育期LAI估算精度最優(yōu)均方根誤差(RMSE, root mean square error, RMSE=0.378);2)對于各類植被指數(shù),花蕾期(20170616)和花鈴期(20170802)時增強植被指數(shù)(EVI, enhanced vegetation index,)表現(xiàn)最佳(RMSE分別為0.352和0.367),開花期(20180623)時校正土壤調(diào)節(jié)植被指數(shù)(MSAVI2, modified soil adjusted vegetation index 2,)估算精度最高(RMSE=0.323);3)單波段反射率和各類植被指數(shù)對全發(fā)育期LAI的估算均要優(yōu)于對單個發(fā)育期LAI的估算,其中基于IRECI指數(shù)的(inverted red-edge chlorophyll index)全發(fā)育期LAI估算模型精度最佳,LOOCV檢驗RMSE=0.425,交叉檢驗RMSE=0.368;將基于IRECI的全發(fā)育期LAI估算模型應用到單個發(fā)育期LAI估算并與各單個發(fā)育期LAI估算模型精度對比,發(fā)現(xiàn)交叉驗證RMSE平均值僅比LOOCV驗證RMSE平均值高0.07,反映了全發(fā)育期LAI估算模型良好的普適性。該研究為農(nóng)作物LAI估算提供了新的數(shù)據(jù)選擇,完善了Sentinel-2衛(wèi)星數(shù)據(jù)在LAI估算中的應用領域。
作物;遙感;模型;Sentinel-2多光譜衛(wèi)星;棉花;葉面積指數(shù);植被指數(shù)
葉面積指數(shù)LAI(leaf area index)最早由Watson[1]提出,定義為單位土地面積綠色葉片的單面面積總和。葉面積指數(shù)是蒸散發(fā)、光能利用率、產(chǎn)量估算及作物發(fā)育期診斷、地球化學元素循環(huán)等研究中的重要參數(shù)[2],它影響葉片及冠層的諸多過程[3-5],如冠層葉綠素含量、光合速率、碳、氮循環(huán)以及作物鮮、干生物量、作物發(fā)育期等。對于農(nóng)作物及森林等葉面積指數(shù)的遙感監(jiān)測已有大量研究[6-7]。但這些研究中提出的各類算法面對不斷更新改進的新的遙感數(shù)據(jù),其應用精度仍需進行討論驗證[6,8]。
過去的幾十年,隨著遙感技術的飛速發(fā)展,很多新型的中分辨率對地觀測衛(wèi)星也已投入了研究應用,如2013年2月11日發(fā)射的Lsandsat-8衛(wèi)星,以及由歐空局于2015年6月23日發(fā)射的Sentinel-2A和2017年3月7日發(fā)射的Sentinel-2B多光譜遙感衛(wèi)星。相比Landsat-8衛(wèi)星數(shù)據(jù),Sentinel-2多光譜遙感衛(wèi)星從705 ~ 865 nm特有3個紅邊參數(shù)波段,空間分辨率20 m,同時具有空間分辨率為10 m的3個可見光波段以及1個近紅外波段,以及空間分辨率為60和20 m的近紅外和短波紅外波段等,共計13個波段,并且隨著Sentinel-2A和Sentinel-2B雙軌衛(wèi)星的同時運行,Sentinel-2多光譜衛(wèi)星的時間分辨率縮短至5 d,這為遙感衛(wèi)星數(shù)據(jù)在農(nóng)業(yè)遙感中的應用提供了更多的可能性。
Sentinel-2數(shù)據(jù)可正式獲取應用前,已有很多研究通過模擬Sentinel-2衛(wèi)星數(shù)據(jù)探討了其在農(nóng)業(yè)[9-13]以及森林生態(tài)[14]中的應用,證實了紅邊參數(shù)波段的有效性,并且Frampton等[11]基于模擬的Sentinel-2數(shù)據(jù)提出了用于估算作物葉綠素和葉面積指數(shù)LAI的IRECI(inverted red edge chlorophyll index)和S2REP(Sentinel-2 red-edge position)紅邊參數(shù)植被指數(shù)。在Sentinel-2數(shù)據(jù)可下載獲取后,其逐漸在作物分類及面積提取[15]、生物量估算[16]、作物葉綠素[17]以及水分信息遙感監(jiān)測[18]等領域展開了應用。其中,利用Sentinel-2多光譜衛(wèi)星數(shù)據(jù)進行葉面積監(jiān)測也有相關研究,主要包括Korhonen等[19]開展的森林葉面積指數(shù)估算、Clevers等[17]基于Sentinel-2數(shù)據(jù)的土豆葉面積指數(shù)及葉綠素含量監(jiān)測以及蘇偉等[20]利用Sentinel-2數(shù)據(jù)進行的玉米葉面積指數(shù)估算。這些研究所構建的基于Sentinel-2的估算模型,證實了Sentinel-2多光譜數(shù)據(jù)用于LAI估算的可行性,但研究中所構建的LAI估算模型均是針對特定研究區(qū)域、研究對象的經(jīng)驗模型,無法直接用于棉花LAI估算。對于Sentinel-2數(shù)據(jù)用于棉花LAI的估算還有待探討。
目前Sentinel-2多光譜數(shù)據(jù)用于棉花LAI遙感監(jiān)測還未有相關研究。已有的棉花葉面積指數(shù)的遙感監(jiān)測研究所采用的數(shù)據(jù)源多是地面高光譜數(shù)據(jù),如早期由石河子大學柏軍華[21-22]、王登偉等[23]開展的基于高光譜數(shù)據(jù)的棉花葉面積指數(shù)反演以及近期馬文君[24]、張卓然[25]的棉花生理生化參數(shù)高光譜反演模型研究。航空及航天遙感數(shù)據(jù)方面,柏軍華等[21]利用Landsat5數(shù)據(jù)提取的NDVI、PVI以及EVI植被指數(shù)對石河子地區(qū)148團場的棉花葉面積指數(shù)進行了反演,田明璐等[26]利用低空無人機成像光譜影像獲取的植被指數(shù)進行了棉花葉面積指數(shù)空間分布制圖。同樣,這些研究中,不論是利用地面高光譜數(shù)據(jù)還是航空航天數(shù)據(jù),所構建的LAI估算模型大多是經(jīng)驗模型,無法直接套用在新的數(shù)據(jù)源上。
本研究擬通過大面積田間實測LAI和Sentinel-2多光譜衛(wèi)星數(shù)據(jù),構建基于Sentinel-2多光譜衛(wèi)星的單波段反射率以及多種植被指數(shù)的棉花LAI估算模型,在精度檢
驗的基礎上,對比不同輸入變量對不同發(fā)育期以及全生育期棉花LAI的估算精度,研究結果將進一步豐富棉花LAI估算方法,提高估算精度。
利用2018年6月23日的Sentinel-2多光譜影像數(shù)據(jù)的B4,B3和B2波段的真彩色合成影像圖反映研究區(qū)樣點分布,如圖1。研究區(qū)位于新疆維吾爾自治區(qū)北部石河子墾區(qū),地處天山北麓中段,古爾班通古特大沙漠南緣,地理坐標位于84°58'~86°24'E,43°26'~45°20'N。在研究區(qū)內(nèi),依據(jù)棉花種植面積,選取面積超過50 hm2的棉花種植區(qū)作為樣點,共選取了4個樣點地進行觀測點布設。為了豐富觀測樣本取值范圍,選取的LAI觀測樣點中包含了不同品種(新陸早61、棉74和棉64)、不同播種日期(2017年4月20日、23日及25日)的田塊,依據(jù)樣點面積大小在每個樣點布設8~15個不等的觀測點。
注:P1-P4為觀測樣點
分別于2017年6月14日棉花花蕾期(16個樣點)、8月3日花鈴期(26個樣點)以及2018年6月21日開花期(28個樣點)進行了棉花LAI實地測定。LAI實地測定利用LAI-2000冠層分析儀進行。3次觀測共計獲取有效樣本數(shù)70個,LAI取值最大值6.67,最小值1.44,平均值4.13,標準差1.41。
通過歐洲航空局的數(shù)據(jù)共享網(wǎng)站( https://scihub. copernicus.eu/dhus/#/home)下載獲取與地面觀測時間同步的Sentinel-2多光譜衛(wèi)星的Level-1C級影像數(shù)據(jù),成像時間分別為2017年6月16日、2017年8月2日以及2018年6月23日。所有遙感影像數(shù)據(jù)均為已經(jīng)過輻射校正和幾何校正處理的Level-1C大氣上層表觀反射率。本研究采用SNAP-Sen2Cor軟件對影像數(shù)據(jù)進行大氣校正,并通過最近鄰插值法,將大氣校正后的各波段重采樣至10 m后用于研究區(qū)各觀測點反射率提取及植被指數(shù)計算。Sentinel-2多光譜數(shù)據(jù)不同波段中心波段分布及空間分辨率信息見表1。
Sentinel-2多光譜衛(wèi)星在可見光及近紅外波段的分布,可實現(xiàn)多種植被指數(shù)的計算。本研究除了探討Sentinel-2多光譜衛(wèi)星的單波段反射率用于LAI估算的潛力外,還通過計算獲取了LAI估算研究中的各類植被指數(shù),包括土壤校正型植被指數(shù),如SAVI,MSAVI等;大氣校正型植被指數(shù),如EVI,GARI等;紅邊參數(shù)植被指數(shù),如S2REP,REIP,IRECI等;葉綠素含量植被指數(shù),如PSSRa,MCARI等;以及傳統(tǒng)的近紅外植被指數(shù),如NDVI,GNDVI,DVI,RVI等,共計17個植被指數(shù)。各類植被指數(shù)基于Sentinel-2數(shù)據(jù)各波段的計算公式見表2。
表1 Sentinel-2數(shù)據(jù)主參數(shù)
表2 所采用植被指數(shù)及基于Sentinel-2數(shù)據(jù)的計算方法
采用決定系數(shù)R,均方根誤差RMSE(root mean square error ),平均偏差,以及實測值與估算值擬合趨勢線的斜率()和截距()進行模型精度檢驗。其中,RMSE數(shù)值直接體現(xiàn)模型估算誤差的多少,其單位與實測值單位相同;平均偏差是用百分比來表示總體的估算值偏離實測平均值的程度,其值為正值時,總體上高估,為負值時,總體上低估;決定系數(shù)2、斜率和截距主要體現(xiàn)估算值和實測值之間變化趨勢的吻合程度,決定系數(shù)2值介于0~1之間,2接近1,斜率()接近1,截距()接近0時,估算值與實測值吻合最佳。各指標的計算公式如下:
對Sentinel-2多光譜衛(wèi)星各波段反射率在不同發(fā)育期以及不同LAI值時的變化特征進行分析,如圖2所示。圖2b中LAI值選取本研究中LAI最大值(LAI=6.67)、最小值(LAI=1.44)以及接近平均值的樣本所對應的Sentienl-2多光譜反射率進行對比。從圖2a可見,隨著棉花從花蕾期(20170616)到開花期(20180623)到花鈴期(20170802)的發(fā)育進程,Sentinel-2多光譜衛(wèi)星可見光波段(B1~B4,443 nm~665 nm)反射率逐漸減小,紅邊參數(shù)波段(B5~B7,705 nm~783 nm)及近紅外波段(B8~B8a,842 nm~865 nm)反射率逐漸增大。圖2b,不同LAI取值的Sentinel-2多光譜衛(wèi)星反射率與不同發(fā)育期光譜反射率變化呈現(xiàn)相同規(guī)律,隨著LAI值從1.44增大至6.67,可見光波段反射率減小,紅邊參數(shù)波段及近紅外波段反射率增大,由此可得出,LAI與可見光波段反射率變化呈負相關關系,與紅邊參數(shù)波段及近紅外波段反射率呈正相關關系。
圖2 不同觀測時期以及不同LAI值Sentinel-2反射率特征
各變量包括Sentinel-2多光譜數(shù)據(jù)的單波段反射率以及各類植被指數(shù)與棉花不同發(fā)育期LAI相關關系分析見表3。
表3 Sentinel-2單波段反射率及植被指數(shù)與實測LAI相關關系
注:*表示相關系數(shù)顯著性水平為0.001 (160.0010.708;=26,0.001=0.588;=28,0.001=0.57;=70,0.001=0.38)。
Note: * represents significant level at 0.001 (160.0010.708;=26,0.001= 0.588;=28,0.001=0.57;=70,0.001=0.38).
對比各類光譜變量在不同發(fā)育期的表現(xiàn)可發(fā)現(xiàn),相比于棉花花蕾期(20170616)以及花鈴期(2070802),開花期(20180623)時各類變量與LAI相關性最好,這一時期的Sentinel-2多光譜衛(wèi)星的單波段反射率以及各類植被指數(shù)(除S2REP、REIP和MTCI)與LAI的相關性均達到0.001極顯著相關水平。這可能是由于棉花花蕾期時,冠層還未封閉,Sentinel-2多光譜反射率中還包含有土壤光譜信息,而棉花花鈴期時,雖然葉面積指數(shù)達到最大,此時葉片色素含量降低以及棉桃形成對光譜信息產(chǎn)生影響,進而導致這2個時期的很多變量與LAI的相關關系未能達到極顯著相關。進一步分析發(fā)現(xiàn),對于單波段反射率,紅邊參數(shù)波段B6、B7以及近紅外波段B8和B8a與LAI相關性均達到0.001極顯著水平,并且3個發(fā)育時期均是中心波長為842 nm波寬為145 nm的B8波段表現(xiàn)最佳,這與已有研究證實的LAI估算的敏感波段主要集中在紅邊參數(shù)波段及近紅外波段并且與這些波段的反射率呈極顯著線性相關關系的結論一致[42]。對比各類植被指數(shù)的表現(xiàn),EVI在花蕾期及花鈴期均表現(xiàn)最佳,與LAI呈0.001極顯著正相關關系,對于開花期MSAVI2表現(xiàn)最佳,總生育期,紅邊參數(shù)植被指數(shù)IRECI與LAI相關性最好,具有最大的相關系數(shù)。各單波段反射率以及各類植被指數(shù),與總樣本LAI相關性,均達到極顯著相關??偘l(fā)育期表現(xiàn)優(yōu)于單個發(fā)育期,主要與總樣本分析中,LAI的取值分布相比單個發(fā)育期的LAI取值更為寬泛,涵蓋范圍更大有關,同時,對于S2REP和REIP這兩類隨發(fā)育期變化而變化的紅邊參數(shù)指標指數(shù),也具有了更多的取值分布,因此與LAI的相關性會明顯優(yōu)于在單個發(fā)育期時表現(xiàn)。此外,觀察NDVI的表現(xiàn),可發(fā)現(xiàn)NDVI在花蕾期、開花期以及總發(fā)育期時與LAI相關性均達到0.001極顯著相關,但隨著棉花LAI增大到花鈴期達到最大時,NDVI與LAI的相關系數(shù)降低未能達到極顯著相關,這很大原因是由于NDVI隨著LAI值增大,會達到飽和從而對LAI值變化的敏感性降低。
通過前文的相關性分析,挑選出與單個發(fā)育期LAI及總發(fā)育期LAI均達到0.001極顯著相關水平的光譜變量,將這些光譜變量作為自變量(),LAI作為因變量(),進行進一步建模對比分析。建模過程中,對比了線性、一元二次以及指數(shù)等模型的決定系數(shù),發(fā)現(xiàn)線性模型總體上決定系數(shù)最大,并且模型結構最為簡單,在此只列出基于各變量的LAI線性估算模型表達式及其決定系數(shù)2,如表4。
表4 基于單波段反射率及植被指數(shù)的LAI估算模型
注:為光譜變量,為葉面積指數(shù)LAI。
Note:is spectral variable,is leaf area index (LAI).
從表4可見,各種類型的植被指數(shù)均有與LAI相關性達到0.001極顯著代表性植被指數(shù),如常規(guī)植被指數(shù)中,有DVI,土壤背景校正型植被指數(shù)中有MSAVI2,大氣校正型植被指數(shù)有EVI,紅邊參數(shù)植被指數(shù)有IRECI指數(shù),并且總體上,基于植被指數(shù)的LAI估算模型的決定系數(shù)都略高于基于單波段反射率的LAI估算模型。對比不同發(fā)育期表現(xiàn)最佳的LAI估算模型的輸入變量,發(fā)現(xiàn)對于單波段反射率,基于近紅外B8波段的LAI估算模型在不同發(fā)育期,相比其他單波段反射率,均具有最大決定系數(shù);對于植被指數(shù)變量,在棉花花蕾期及花鈴期均是EVI指數(shù)表現(xiàn)最佳,在開花期,MSAVI2表現(xiàn)最佳;總體模型中,單波段反射率B8a和IRECI植被指數(shù)具有最大判定系數(shù)。IRECI指數(shù)與LAI的極顯著相關關系與提出該指數(shù)的研究結論一致[11]。分析IRECI指數(shù)所包含的各波段位置(表2)可發(fā)現(xiàn),該指數(shù)包含了Sentinel-2的2個分別位于705 nm和740 nm的紅邊參數(shù)波段(B5和B6),以及位于植被光譜反射率最大的近紅外783 nm處的B7波段和反射率最小的紅光區(qū)域665 nm處的B6波段,利用近紅外B7波段減去紅光B6波段的處理削弱了LAI值較大時紅光區(qū)域飽和的影響,同時通過相除的運算強化了2個紅邊參數(shù)與LAI的相關性。
對表4中,不同發(fā)育期及總發(fā)育期LAI估算模型中決定系數(shù)最佳的模型進行進一步精度檢驗。精度檢驗采用留一驗證(LOOCV, leave-one-out-cross-validation)和交叉驗證2種方式展開。留一驗證是將樣本數(shù)個樣本作為訓練樣本,剩余的一個樣本作為檢驗樣本,并運行次,尋求最小驗證誤差,該方法可充分利用測定數(shù)據(jù),尤為適用于樣本數(shù)據(jù)較少的情況。該方法也應用在棉花色素估算的研究中[43],關于該方法的詳細介紹可參考Shao[44],不再贅述。本研究中基于留一驗證構建的線性模型自變量與表4中決定系數(shù)最佳的模型的自變量相同,因變量為LAI,利用MatlabR2012b中的程序代碼實現(xiàn)。交叉檢驗針對總樣本展開,表4中的總發(fā)育期LAI估算模型是基于總樣本的2/3(46個樣本)構建的,將剩余的1/3(24個樣本)用于對該模型進行交叉檢驗,此外,同時利用各單個發(fā)育期時的樣本對基于總樣本構建的LAI估算模型進行交叉檢驗,探討該模型的普適性。各精度檢驗指標的自檢驗及交叉檢驗結果見表5。
表5 模型精度檢驗結果
從表5可見,LOOCV精度檢驗結果中,不論對于單個發(fā)育期還是總發(fā)育期,基于植被指數(shù)變量的LAI估算模型相比基于單波段反射率的LAI估算模型,具有更大的決定系數(shù),更小的RMSE和Bias值;同時可見,總發(fā)育期LAI估算模型,相比單個發(fā)育期,估算精度更高,其中基于紅邊葉綠素指數(shù)(IRECI)的LAI估算模型具有最大的決定系數(shù)0.908,最小平均Bias值0.001%,并且由其得到的LAI估算值與實測值之間的斜率為0.908,最接近1,說明估算值與實測值之間的擬合較好;進一步將基于IRECI的總發(fā)育期LAI估算模型應用各單個發(fā)育期LAI估算,各發(fā)育期交叉驗證RMSE與各單個發(fā)育期基于植被指數(shù)的LAI估算模型LOOCV驗證RMSE相比,平均值略高出0.07;對總發(fā)育期LAI估算模型采用交叉驗證時,各精度檢驗指標優(yōu)于其LOOCV檢驗,具有更大的2=0.951,以及更接近1的斜率=0.945,該結論更好地說明了總發(fā)育期LAI估算模型的普適性。
通過上述分析,基于Sentinel-2多光譜遙感數(shù)據(jù)的B8單波段反射率及EVI和MSAVI2植被指數(shù)對棉花不同發(fā)育期LAI估算精度最高,總生育期的IRECI植被指數(shù)具有最好的反演精度,因此利用這些模型進行棉花不同發(fā)育期LAI遙感制圖,形成研究區(qū)LAI空間分布圖,如圖3。由圖清晰可見,總體上,各類模型均能反映出不同時段研究區(qū)LAI的主要取值分布,在6月中旬研究區(qū)的LAI取值分布主要在2.5左右(圖3a, b, c),至6月下旬大部分地區(qū)LAI取值介于2.5~4.0之間(圖3d, e, f),到8月初研究區(qū)大部分地區(qū)LAI取值大于4.0,有小部分地區(qū)LAI達到7.0左右(圖3g, h, i);對比不同估算參數(shù)的表現(xiàn),相對于B8單波段反射率及全生育期的IRECI指數(shù),EVI植被指數(shù)可更清晰地區(qū)分植被區(qū)域和非植被區(qū)域的LAI取值,這一點反映了EVI植被指數(shù)用于LAI空間監(jiān)測的優(yōu)勢。
圖3 基于Sentinel-2數(shù)據(jù)的LAI遙感反演制圖
本研究利用Sentinel-2多光譜遙感衛(wèi)星數(shù)據(jù)及棉花實測LAI數(shù)據(jù)探討了Sentinel-2多光譜衛(wèi)星數(shù)據(jù)用于棉花LAI估算的特點和精度。主要得到以下結論:
1)對于Sentinel-2各單波段反射率,基于B8近紅外波段(842 nm)的線性模型對LAI估算精度最高。通過分析Sentinel-2各單波段反射率及各類植被指數(shù)與LAI的相關關系發(fā)現(xiàn),Sentinel-2多光譜衛(wèi)星的2個紅邊參數(shù)波段(B6和B7)及近紅外波段(B8和B8a)與不同發(fā)育期LAI的相關性均達到0.001的極顯著相關水平,相關系數(shù)均大于0.7,并且不同發(fā)育期的最佳LAI估算模型均基于近紅外B8波段,LOOCV檢驗的決定系數(shù)R均大于0.564。
2)基于Sentinel-2衛(wèi)星各波段構建的植被指數(shù)中,大氣校正指數(shù)EVI、土壤背景校正指數(shù)MSAVI2及紅邊參數(shù)植被指數(shù)IRECI構建的LAI估算模型表現(xiàn)最佳?;赟entinel-2波段的各類指數(shù)與LAI的相關分析結果表明,植被指數(shù)尤其是大氣校正指數(shù)EVI、土壤背景校正指數(shù)MSAVI2及紅邊參數(shù)植被指數(shù)IRECI對LAI的估算精度最高,均達到極顯著相關;其中由IRECI構建的總發(fā)育期LAI線性估算模型表現(xiàn)最佳,相關系數(shù)為0.953,由其構建的總LAI估算模型的LOOCV檢驗以及交叉檢驗的決定系數(shù)R及預測值與實測值之間擬合的斜率均大于0.9,說明LAI估算值與實測值擬合較好。
3)鑒于該研究中實測數(shù)據(jù)覆蓋3個不同發(fā)育期,對于下一步工作,一方面需要進一步細化棉花不同發(fā)育期葉面積指數(shù)監(jiān)測模型,另一方面,可以考慮在對基于全發(fā)育期數(shù)據(jù)的LAI估算模型普適性驗證的基礎上,將該估算模型進一步應用于與LAI直接相關的管理參數(shù)的估算中,如與LAI密切相關的棉花長勢監(jiān)測、棉花施肥用量監(jiān)測等領域。
[1] Watson D J. Comparative physiological studies in the growth of field crops I. variation in net assimilation rate and leaf area between species and varieties, and within and between years[J]. Annals of Botany, 1947, 11(1): 41-76.
[2] Aparicio N, Villegas D, Araus J L, et al. Relationship between growth traits and spectral vegetation indices in durum wheat[J]. Crop Science, 2002, 42(5): 1547-1555.
[3] Baret F, Jacquemoud S, Guyot G, et al. Modeled analysis of the biophysical nature of spectral shifts and comparison with information-content of broad bands[J]. Remote Sensing of Environment, 1992, 41(2): 133-142.
[4] Clevers J G P W, De Jong S M, Epema G F, et al. MERIS and the red-edge position[J]. International Journal of Applied Earth Observation and Geoinformation, 2001, 3(4): 313-320.
[5] Ye X J, Sakai K, Okamoto H, et al. A ground-based hyperspectral imaging system for characterizing vegetation spectral features[J]. Computers and Electronics in Agriculture, 2008, 63(1): 13-21.
[6] Gitelson A A. Wide dynamic range vegetation index for remote quantification of biophysical characteristics of vegetation[J]. Journal of Plant Physiology, 2004, 161(2): 165-173.
[7] Zarco-Tejada P J, Ustin S L, Whiting M L. Temporal and spatial relationships between within-field yield variability in cotton and high-spatial hyperspectral remote sensing imagery[J]. Agronomy Journal, 2005, 97(3): 641-653.
[8] Delegido J, Fernandez G, Gandia S, et al. Retrieval of chlorophyll content and LAI of crops using hyperspectral techniques: application to PROBA/CHRIS data[J]. International Journal of Remote Sensing, 2008, 29(24): 7107-7127.
[9] Delegido J, Verrelst J, Alonso L, et al. Evaluation of sentinel-2 red-edge bands for empirical estimation of green LAI and chlorophyll content[J]. Sensors, 2011, 11(7): 7063-7081.
[10] Clevers J G P W, Gitelson A A. Remote sensing of crop and grass chlorophyll and nitrogen content using red-edge bands on Sentinel-2 and -3[J]. International Journal of Applied Earth Observation and Geoinformation, 2013, 23: 344-351.
[11] Frampton W J, Dash J, Watmough G, et al. Evaluating the capabilities of Sentinel-2 for quantitative estimation of biophysical variables in vegetation[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2013, 82: 83-92.
[12] Verrelst J, Rivera J P, Veroustraete F, et al. Experimental sentinel-2 LAI estimation using parametric, non-parametric and physical retrieval methods – a comparison[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2015, 108: 260-272.
[13] Sibanda M, Mutanga O, Rouget M. Examining the potential of Sentinel-2 MSI spectral resolution in quantifying above ground biomass across different fertilizer treatments[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2015, 110: 55-65.
[14] Majasalmi T, Rautiainen M. The potential of Sentinel-2 data for estimating biophysical variables in a boreal forest: a simulation study[J]. Remote Sensing Letters, 2016, 7(5): 427-436.
[15] 朱琳. 基于Sentinel多源遙感數(shù)據(jù)的作物分類及種植面積提取研究[D]. 楊凌: 西北農(nóng)林科技大學, 2018. Zhu Lin. Study on Crop Classification and Area Extraction Based on Multi-source Remote Sensing Data of Sentinel[D]. Yangling: Northwest A&F University, 2018. (in Chinese with English abstract)
[16] 鄭陽, 吳炳芳, 張淼. Sentinel-2數(shù)據(jù)的冬小麥地上干生物量估算及評價[J]. 遙感學報, 2017, 21(2): 318 – 328. Zheng Yang, Wu Bingfang, Zhang Miao. Estimaing the above ground biomass of winter wheat using the Sentinel-2 data[J]. Journal of Remote Sensing, 2017, 21(2): 318-328. (in Chinese with English abstract).
[17] Clevers J G P W, Kooistra L, Brande van den M M. Using Sentinel-2 data for retrieving LAI and leaf and canopy chlorophyll content of a potato crop[J]. Remote Sensing, 2017, 9(405): 1-15.
[18] Rozenstein O, Haymann N, Kaplan G, et al. Estimating cotton water consumption using a time series of Sentinel-2 imagery[J]. Agricultural Water Management, 2018, 207: 44-52.
[19] Korhonen L, Hadi P P, Rautiainen M. Comparison of Sentinel-2 and Landsat 8 in the estimation of boreal forest canopy cover and leaf area index[J]. Remote Sensing of Environment, 2017, 195: 259-274.
[20] 蘇偉, 侯寧, 李琪, 等. 基于 Sentinel-2 遙感影像的玉米冠層葉面積指數(shù)反演[J]. 農(nóng)業(yè)機械學報, 2018, 49(1): 151-156. Su Wei, Hou Ning, Li Qi, et al. Retriving leaf area index of corn canopy based on Sentinel-2 remote sensing image[J]. Transactions of the Chinese Society for Agricultural Machinery, 2018, 49(1): 151-156. (in Chinese with English abstract).
[21] 柏軍華, 李少昆, 王克如, 等. 棉花葉面積指數(shù)冠層反射率光譜響應及其反演[J]. 中國農(nóng)業(yè)科學, 2007, 40(1): 63 – 69. Bai Junhua, Li Shaokun, Wang Keru, et al. The response of canopy reflectance spectrum for the cotton LAI and LAI inversion[J]. Scientia Agricultura Sinica, 2007, 40(1): 63-69. (in Chinese with English abstract)
[22] 柏軍華. 基于LAI的棉花產(chǎn)量近地遙感模型研究[D]. 石河子: 石河子大學, 2005. Bai Junhua. Field Remote Sensing Model Research in Cotton Yield Based on LAI[D]. Shihezi: Shihezi University, 2005. (in Chinese with English abstract)
[23] 王登偉, 黃春燕, 馬勤建, 等. 棉花高光譜植被指數(shù)與LAI和地上鮮生物量的相關分析[J].中國農(nóng)學通報, 2008, 24(3): 426-429. Wang Dengwei, Huang Chunyan, Ma Qinjian, et al. Relationships analysis between cotton leaf area index, aboveground fresh biomass and hyperspectral vegetation indices[J]. Chinese Agricultural Science Bulletin, 2008, 24(3): 426-429. (in Chinese with English abstract)
[24] 馬文君. 棉花生理生化參數(shù)高光譜反演模型研究[D]. 楊凌: 西北農(nóng)林科技大學, 2016. Ma Wenjun. Study on Hyperspectral Estimation Models for Cotton Physiological and Biochemical Parameters [D]. Yangling: Northwest A&F University, 2016. (in Chinese with English abstract)
[25] 張卓然. 棉花高光譜特征及其農(nóng)學參數(shù)遙感反演研究[D]. 楊凌: 西北農(nóng)林科技大學, 2018. Zhang Zhuoran. Cotton Hyperspectral Characteristic and Hyperspectral Estimation of Its Agricultural Parameters[D]. Yangling: Northwest A&F University, 2018. (in Chinese with English abstract)
[26] 田明璐, 班松濤, 常慶瑞, 等. 基于低空無人機成像光譜儀影像估算棉花葉面積指數(shù)[J].農(nóng)業(yè)工程學報, 2016, 32(21): 102-108. Tian Minglu, Ban Songtao, Chang Qinrui, et al. Use of hyperspectral images from UAV-based imaging spectroradiometer to estimate cotton leaf area index[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2016, 32(21): 102-108. (in Chinese with English abstract)
[27] Tucker C J. Red and photographic infrared linear combinations for monitoring vegetation[J]. Remote Sensing of Environment, 1979, 8(2): 127-150.
[28] Jordan C F. Derivation of leaf-area index from quality of light on the forest floor[J]. Ecology, 1969, 50(4): 663-666.
[29] Pearson R L, Miller L D. Remote mapping of standing crop biomass for estimation of the productivity of the short grass prairie[C]. Proc. 8th Int. Symp. Remote Sens. Environ. Ann Arbor, Michigan USA, 1972, 3: 1355.
[30] Crippen R E. Calculating the vegetation index faster[J]. Remote Sensing of Environment, 1990, 34(1): 71-73.
[31] Clevers J G P W. The application of a weighted infrared-red vegetation index for estimating leaf Area index by correcting for soil moisture[J]. Remote Sensing of Environment, 1989, 29(1): 25-37.
[32] Deering D W, Rouse Jr J W, Hass R H, et al. Measuring forage production of grazing units from Landsat MSS data[C]. Proc. Tenth Int. Symp. Remote Sens. Environ. Ann Arbor, Michigan USA, 1975, 22: 1169 -1178.
[33] Gitelson A A, Merzlyak M N. Remote sensing of chlorophyll concentration in higher plant leaves[J]. Advances in Space Research, 1998, 22(5): 689-692.
[34] Huete A R. A soil-adjusted vegetation index (SAVI)[J]. Remote Sensing of Environment, 1988, 25(3): 295 – 309.
[35] Qi J, Chehbouni A, Huete A R, et al. A modified soil adjusted vegetation index[J]. Remote Sensing of Environment, 1994, 48(2): 119-126.
[36] Huete A, Didan K, Miura T, et al. Overview of the radiometric and biophysical performance of the MODIS vegetation indices[J]. Remote Sensing of Environment, 2002, 83(1/2):195-213.
[37] Gitelson A A, Kaufman Y J, Merzlyak M N. Use of a green channel in remote sensing of global vegetation from EOS-MODIS[J]. Remote Sensing of Environment, 1996, 58(3): 289-298.
[38] Guyot G, Baret F, Major D. High spectral resolution: determination of spectral shifts between the red and the near infrared[C]. Proc. 16th Int. Congress Photogramm. Remote Sens. Kyoto, Japan, 1988.
[39] Dash J, Curran P J. Evaluation of the MERIS terrestrial chlorophyll index (MTCI)[J]. Advances in Space Research, 2007, 39(1): 100-104.
[40] Blackburn G A. Quantifying chlorophylls and caroteniods at leaf and canopy scales: an evaluation of some hyperspectral approaches[J]. Remote Sensing of Environment, 1998, 66(3): 273 – 285.
[41] Daughtry C, Walthall C, Kim M, et al. Estimating corn leaf chlorophyll concentration from leaf and canopy reflectance[J]. Remote Sensing of Environment, 2000, 74(2): 229-239.
[42] Herrmann I, Pimstein A, Karnieli A, et al. LAI assessment of wheat and potato crops by VENμS and Sentinel-2 bands[J]. Remote Sensing of Environment, 2011, 115(8): 2141-2151.
[43] Yi Q X, Jiapaer G L, Chen J M, et al. Different units of measurement of carotenoids estimation in cotton using hyperspectral indices and partial least square regression[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2014, 91: 72-74.
[44] Shao J. Linear model selection by cross-validation[J]. Journal of the American statistical Association, 1993, 88(422): 486-494.
Remote estimation of cotton LAI using Sentinel-2 multispectral data
Yi Qiuxiang
(1.,,,830011,; 2.,830011,3.,100049,)
Rapid and accurate LAI (Leaf Area Index) acquisition is of great significance for remote sensing monitoring of cotton growth, diagnosis of growth stage, extraction of cotton plant area and yield estimation. The present research discussed the characteristics of Sentinel-2 multi-spectral satellite data for remote estimation of cotton LAI. Measured LAI from filed experiments and Sentinel-2 data in 2017 and 2018 were obtained, and LAI estimation model for different and for all growth stages were established basing on single spectral band reflectance on Sentinel-2 and various vegetation index from Sentinel-2 bands. The estimation accuracy of the established LAI models were validated by coefficient of determination (2), RMSE (root mean square error), mean bias, and slope and intercept, using LOOCV (Leave-One-Out-Cross Validation) method and cross validation, respectively. The results showed that: 1) for the single-band reflectance of sentinel-2 multi-spectral satellite data, two red-edge bands of B6 and B7, and two near-infrared bands of B8 and B8a, were all significantly (<0.001) correlated to LAI at all three tested growth stages, i.e. bud stage (16-Jun-2017), and flowering stage (23-Jun-2018), and boll stage (2-Aug-2017), with correlation coefficient greater than 0.7. And when the correlation between LAI and band reflectance were performed using data consist of three growth stages, the correlation coefficient for all tested bands reach significant level (<0.001), and the maximum correlation coefficient was 0.943 of near-infrared narrow band B8a, which center at 865 nm with a wave width of 32 nm. The accuracy of LAI estimation at different development stages was optimized using the near-infrared band B8 which with a central wavelength of 842 nm and a wave width of 145 nm, with all RMSE smaller than 0.465. 2) for seventeen LAI related vegetation indices, including EVI (Enhanced Vegetation Index), MSAVI2 (Modified Soil Adjusted Vegetation Index 2), IRECI (Inverted Red-Edge Chlorophyll Index), etc., most of them were significantly (<0.001) correlated with LAI, especially atmospheric correction index EVI, soil adjusted index MSAVI2, and red-edge index IRECI, and the coefficient of correlation were over 0.8. EVI provided the best result for LAI estimation at bud stage and boll stage, and at flowering stage it consists by MASVI2, with bud stage RMSE=0.352, and boll stageRMSE=0.367 and flowering stage RMSE=0.323, respectively. 3) LAI estimation models for whole growth stages performed better than these for one single growth stage. And the best LAI estimation models for whole growth period using single spectral band reflectance and vegetation index were respectively obtained by near-infrared narrow band B8a and IRECI, with IRECI performed slightly better, which with2=0.908 and RMSE=0.425 for LOOCV, and2=0.951 and RMSE=0.368 for cross validation. Additionally, when apply IRECI-LAI estimation model for whole growth stages on one single growth stage LAI estimation, the accuracy comparison between the IRECI-LAI model and single growth stage LAI models showed that the average cross validation RMSE was only 0.07 greater than the average LOOCV RMSE, indicating the good universality of LAI estimation model for whole growth stages.
crops; remote sensing; models; Sentinel-2 data; cotton; leaf area index (LAI); vegetation index
2019-02-18
2019-06-14
國家自然科學基金(41571428,41871328)
易秋香,副研究員,主要研究方向為定量遙感,農(nóng)業(yè)遙感。Email:yiqx@ms.xjb.ac.cn
10.11975/j.issn.1002-6819.2019.16.021
P237.9
A
1002-6819(2019)-16-0189-09
易秋香. 基于Sentinel-2多光譜數(shù)據(jù)的棉花葉面積指數(shù)估算[J]. 農(nóng)業(yè)工程學報,2019,35(16):189-197. doi:10.11975/j.issn.1002-6819.2019.16.021 http://www.tcsae.org
Yi Qiuxiang. Remote estimation of cotton LAI using Sentinel-2 multispectral data[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2019, 35(16): 189-197. (in Chinese with English abstract) doi:10.11975/j.issn.1002-6819.2019.16.021 http://www.tcsae.org