劉 軻,黃 平※,任國業(yè),周清波,李源洪,王 思,董秀春
(1. 四川省農(nóng)業(yè)科學院遙感應用研究所/農(nóng)業(yè)部遙感應用中心成都分中心,成都 610066;2. 中國農(nóng)業(yè)科學院農(nóng)業(yè)資源與農(nóng)業(yè)區(qū)劃研究所/農(nóng)業(yè)部農(nóng)業(yè)信息技術重點實驗室,北京 100081)
·農(nóng)業(yè)信息與電氣技術·
基于冠層反射率模型的作物參數(shù)多階段反演方法研究進展
劉 軻1,黃 平1※,任國業(yè)1,周清波2,李源洪1,王 思1,董秀春1
(1. 四川省農(nóng)業(yè)科學院遙感應用研究所/農(nóng)業(yè)部遙感應用中心成都分中心,成都 610066;2. 中國農(nóng)業(yè)科學院農(nóng)業(yè)資源與農(nóng)業(yè)區(qū)劃研究所/農(nóng)業(yè)部農(nóng)業(yè)信息技術重點實驗室,北京 100081)
利用遙感手段,基于冠層反射率(canopy reflectance,CR)模型反演農(nóng)作物參數(shù)具有經(jīng)濟、高效、普適性好的特點,是智慧農(nóng)業(yè)快速、精確監(jiān)測區(qū)域尺度農(nóng)情信息的理想方法。然而,CR模型反演過程受“病態(tài)反演”問題影響。針對此,前人提出了多階段目標決策(multi-stage,sample-direction dependent,target-decisions,MSDT)法和面向?qū)ο螅╫bject-based)反演法。分別依據(jù) C R模型參數(shù)的敏感性和不確定性,以及作物參數(shù)的空間分布特征,將反演過程劃分為若干階段,每階段只反演部分參數(shù),前階段反演結(jié)果作為后階段反演的先驗知識,以此減少 C R模型參數(shù)優(yōu)化的不確定性,改善“病態(tài)反演”問題。該文系統(tǒng)總結(jié)了MSDT法與面向?qū)ο蠓囱莘?,將其歸納為統(tǒng)一的“多階段反演”方法,并提出概念模型?;诖?,總結(jié)、討論了多階段反演中如下三方面共性問題,試分析可能的解決途徑:1)多階段反演決策還需要廣泛比較、科學論證與改進,以確保其合理性和有效性;未來研究中,應將MSDT法與面向?qū)ο蠓囱莘椒ㄓ袡C結(jié)合,在統(tǒng)一的多階段反演技術框架下,制定更加合理的反演決策方法。2)CR模型的參數(shù)化精度可能影響多階段反演;未來應嘗試利用“天空地一體化”遙感技術和尺度轉(zhuǎn)換方法獲取先驗知識,提高CR模型參數(shù)化精度。3)多階段反演過程中,反演誤差逐級傳遞;未來研究中,一方面應嘗試識別并糾正前階段反演中的誤差,另一方面應合理利用前階段反演結(jié)果,避免前階段反演誤差影響后階段的反演。
遙感;模型;作物;多階段目標決策;面向?qū)ο螅欢嚯A段反演;作物參數(shù)
中國正經(jīng)歷從傳統(tǒng)農(nóng)業(yè)向現(xiàn)代農(nóng)業(yè)的歷史轉(zhuǎn)變[1]。建立智慧農(nóng)業(yè)系統(tǒng),實現(xiàn)農(nóng)業(yè)生產(chǎn)、流通、銷售和管理等環(huán)節(jié)的智能化,是發(fā)展現(xiàn)代農(nóng)業(yè)的重要手段[2]。為此,需要準確了解農(nóng)作物生理生化參數(shù)及其動態(tài)變化。其中,葉面積指數(shù)(leaf area index,LAI)與葉片葉綠素含量(leaf chlorophyll content,Cab)尤其受到重視。LAI通常定義為單位地表面積上單面葉面積的總和[3],表征植被的空間分布和密度[4],反映了植被的生產(chǎn)力[5]、地氣間能量和物質(zhì)交換[6]、太陽能及水分利用效率[7]等生態(tài)功能。Cab定義為單位面積或單位質(zhì)量的葉片中葉綠素a和葉綠素b含量之和。Cab與葉片氮素狀況密切相關[8],反映了葉片光合能力[9],是農(nóng)作物水、肥狀況的重要指標[10-11]。因此,LAI與Cab是很多陸表過程模型的輸入?yún)?shù)[12-13],在農(nóng)作物長勢監(jiān)測[14-15]、病蟲害監(jiān)測[16]和產(chǎn)量估測[17-18]等領域具有廣泛的應用潛力,是智慧農(nóng)業(yè)管理和生產(chǎn)決策的重要依據(jù)。
遙感技術以其經(jīng)濟、高效、大尺度、長時序的優(yōu)點,成為區(qū)域至全球尺度LAI和Cab監(jiān)測的主要手段,已形成一系列 L AI遙感數(shù)據(jù)產(chǎn)品[19-21],但其空間分辨率多為千米級,時間分辨率多為8~10 d,難以滿足智慧農(nóng)業(yè)需求。利用遙感手段反演農(nóng)作物參數(shù)的方法大致可分為基于統(tǒng)計模型和基于冠層反射率(canopy reflectance,CR)模型的估測方法。其中,后者對地面實測數(shù)據(jù)依賴較少,各因素的影響機理明確,更具有普適性[12,22]。然而,其估測精度受“病態(tài)反演”問題的嚴重影響[23]。約束“病態(tài)反演”問題,提高植被生理生化參數(shù)反演的精度與穩(wěn)定性一直是定量遙感領域關注的熱點、難點之一。前人研究通過模型耦合[24-25]、引入先驗知識[26-28]、引入多角度遙感數(shù)據(jù)和多源數(shù)據(jù)[4,29-30],以及改進反演策略[25,31-32]等方面來約束病態(tài)反演問題。其中,為合理分配和利用先驗知識與觀測數(shù)據(jù),或增加額外的空間約束條件約束“病態(tài)反演”問題,國內(nèi)外研究者將常規(guī)的單一階段反演發(fā)展為多階段反演,提出了多階段目標決策(multi-stage,sample-direction dependent,target-decisions,MSDT)法和面向?qū)ο螅╫bject-based)反演法。本文旨在系統(tǒng)地整理和總結(jié)MSDT法和面向?qū)ο蟮姆囱莘椒?,總結(jié)其共性,將其置于統(tǒng)一的“多階段反演”概念模型之下,分析其技術難點和可能的解決途徑。以期為進一步改進多階段反演方法理清思路,為智慧農(nóng)業(yè)系統(tǒng)快速、精確地獲取農(nóng)作物生理生化參數(shù)提供技術參考。
CR模型參數(shù)眾多,而與之相關的先驗知識往往難以獲得,致使模型參數(shù)化存在嚴重的不確定性,不同的參數(shù)取值組合往往得到相似的光譜反射率。因此,反演過程中,難以得到CR模型變量的唯一最優(yōu)解,造成了“病態(tài)反演”問題。針對此問題,李小文等[33]提出了MSDT反演法。本文歸納了前人基于MSDT法反演林地和農(nóng)作物參數(shù)的相關研究,見表1。
表1 基于多階段目標決策法的植被參數(shù)反演研究Table 1 Studies on retrieving vegetation variables based on MSDT method
1.1 MSDT法原理和流程
MSDT法認為遙感數(shù)據(jù)中不敏感波段無法為反演提供有效信息。因此,不應使用所有的遙感數(shù)據(jù)來反演所有的未知參數(shù),而應分階段反演。首先利用部分遙感數(shù)據(jù)來反演對其最敏感,最不確定的參數(shù)。在這部分參數(shù)敏感性降低后,再用遙感數(shù)據(jù)的其他子集反演其他敏感參數(shù)。采用上述方法,通過合理地分割遙感觀測數(shù)據(jù)集與待反演參數(shù)集,減少反演過程中各參數(shù)的相互干擾,降低物理模型反演的不確定性,從反演過程中挖掘先驗知識。
MSDT法將CR模型的反演流程拆分為多個階段。每階段反演前,首先評估CR模型各參數(shù)在各波段、各觀測方向(有多角度觀測數(shù)據(jù)時)的敏感性與不確定性,使用最敏感波段和觀測方向的數(shù)據(jù)反演最不確定的參數(shù),每階段反演不超過 4 個參數(shù)。前階段反演結(jié)果作為后階段反演的先驗知識,降低這部分參數(shù)的不確定性。而后,重新評估各參數(shù)的敏感性,“用觀測數(shù)據(jù)的另一子集反演另一部分參數(shù)”[33]。直到反演結(jié)果滿足要求,或遙感數(shù)據(jù)已經(jīng)得到充分利用,而反演結(jié)果沒有明顯收斂為止[33-34]。
MSDT反演中,先驗知識的提取與利用方法尚未統(tǒng)一。高峰等[34]使用左、右兩個高斯函數(shù)擬合前一階段的反演結(jié)果,采用前一階段的反演結(jié)果作為該參數(shù)的期望值;70%~82%置信度水平下的(最大似然值-左高斯分布標準差)至(最大似然值+右高斯分布標準差)作為參數(shù)取值的軟邊界限制。朱小華等[40]采用反演結(jié)果80%置信區(qū)間的最大值和最小值作為該參數(shù)在下一階段的軟邊界限制。馮曉明和趙英時[37]為提高效率,經(jīng)試驗調(diào)整,確定每階段參與反演參數(shù)的不確定性減為原來的20%。
1.2 CR模型參數(shù)敏感性與不確定性分析方法
模型參數(shù)敏感性分析的基本原理是將模型輸出結(jié)果的不確定性依據(jù)一定準則分配到各個模型參數(shù)中,從而找出關鍵控制參數(shù),識別各參數(shù)的相對重要性,預測當某一參數(shù)輸入發(fā)生變動(即存在“不確定性”)時,對模型輸出結(jié)果帶來的影響[43]。MSDT法中,參數(shù)的敏感性和不確定性是確定多階段反演順序的依據(jù)。李小文等[33]建議采用不確定性與敏感性矩陣(uncertainty and sensitivity matrix,USM)評價CR模型參數(shù)的敏感性,矩陣元素定義為
式中S(i,j)是第i個方向上第j個參數(shù)在特定波段的敏感性;ΔBRDF(i,j)為第i個方向的其他參數(shù)固定為期望時,第j個參數(shù)取最小、最大值造成的模擬光譜反射率的之差;BRDFexp(i)為所有參數(shù)固定為期望值時第i個方向的模擬光譜反射率。大量反演研究(表1)與模擬試驗[44]表明了基于USM的MSDT反演決策的可靠性。
此外,拓展傅里葉幅度敏感性檢驗(extended Fourier amplitude sensitivity test,EFAST)方法[43,45]也廣泛應用于模型參數(shù)敏感性評價[46-49]。EFAST方法假設模型輸出總方差V由獨立參數(shù)xi引起的方差Vi和各參數(shù)間交互作用引起的方差共同構成。參數(shù)xi的一階和多階敏感性指數(shù)分別定義為
式中Vij為參數(shù)xi通過參數(shù)xj所貢獻的方差,即xi與xj的耦合方差;Vijm為參數(shù)xi通過參數(shù)xj和xm所貢獻的方差;Vij至 V12…i…k表示各參數(shù)間的耦合方差。Si是參數(shù) xi的一階敏感性指數(shù),反映了xi對V的直接貢獻;Sij和 Sijm等是參數(shù)xi的多階敏感性指數(shù),反映了xi與其他參數(shù)的交互作用對V的間接貢獻。當考慮參數(shù)間的交互作用時,將各階敏感性指數(shù)之和定義為xi的總敏感性指數(shù)STi(式3),用于評價參數(shù)xi在模型中的敏感性。
一方面,由式(2)、(3)可見,EFAST方法不僅考慮CR模型參數(shù)自身變化對模擬反射率的影響,也考慮了參數(shù)間的交互作用,因此,其評價結(jié)果比局部敏感性分析方法(如USM)更全面和客觀[47]。另一方面,在應用中,基于EFAST方法的CR模型參數(shù)敏感性分析通常按照圖1所示的流程實現(xiàn)。此外,EFAST方法要求抽取參數(shù)組合的樣本量大于參數(shù)個數(shù)的65倍[47]。以ACRM模型為例,欲分析其中11個主要參數(shù)[46]的敏感性,則應抽取多于715組參數(shù)取值組合代入ACRM模型,生成對應的模擬光譜以供分析。由此可見,EFAST方法的過程復雜,計算量大。未來研究中,應針對具體的CR模型,定量比較EFAST與USM等局部敏感性分析方法用于MSDT反演的優(yōu)勢與不足,進而深入探索將EFAST方法用于MSDT法的可行性與必要性。
圖1 使用EFAST方法的CR模型參數(shù)敏感性分析流程Fig.1 Process of sensitivity analysis on CR models parameters using EFAST method
1.3 MSDT反演法的優(yōu)勢與局限性
針對“病態(tài)反演”問題,國內(nèi)外研究者提出了基于貝葉斯網(wǎng)絡的混合反演法[50]、基于數(shù)據(jù)同化的反演方法[51]和多階段目標決策(MSDT)等多種反演方法。與其他方法相比,MSDT法:1)合理分割遙感數(shù)據(jù)集與待反演參數(shù)集,更有利于先驗知識的合理利用和觀測數(shù)據(jù)的有效分配[40];有助于減少各個參數(shù)在反演過程中的相互影響[35],適于農(nóng)作物LAI與Cab同步反演的需要。2)由表1可見,MSDT法的適用性、有效性已在基于不同的CR模型和植被種類的研究中得到驗證。
然而,其局限性在于:1)研究數(shù)量較少,筆者僅發(fā)現(xiàn)20a來的9篇相關文獻(表1),其用于CR模型反演的有效性尚待廣泛驗證;2)依賴多角度或多尺度遙感數(shù)據(jù),限制了MSDT法的適用范圍,例如,表1列舉的多數(shù)研究均基于多角度遙感數(shù)據(jù)開展;3)其技術方法仍處于探索階段,遠未成熟,例如MSDT法和面向?qū)ο蟮姆囱莘椒ㄍ玫浇厝幌喾吹亩嚯A段反演方案(詳見3.1節(jié)),又如MSDT法反演過程受CR模型參數(shù)的初始期望和取值范圍影響顯著(詳見3.2節(jié))。未來研究中,一方面,應深入驗證和改進其技術方法,特別是其反演決策的確定方法,研究如何避免CR模型參數(shù)初始期望對反演過程的影響。另一方面,除多角度和多尺度遙感數(shù)據(jù)外,應嘗試其他類型的多維度遙感數(shù)據(jù)(如超光譜和高光譜數(shù)據(jù))用于MSDT反演的可行性以及多源數(shù)據(jù)在MSDT反演中協(xié)同應用的可行性,以拓展MSDT法的適用范圍。
農(nóng)田生態(tài)系統(tǒng)存在明顯的空間特征:1)同一田塊內(nèi)作物品種,生長階段,水、肥管理水平等因素基本一致,因此作物參數(shù)差異較??;相反,田塊之間作物參數(shù)差異較大[52]。2)地統(tǒng)計理論表明,鄰近像元比距離較遠的像元表現(xiàn)出更多的相似性,因此,多數(shù)情況下,農(nóng)作物參數(shù)在空間上平滑地變化[53]。
然而,常規(guī)CR模型反演中,對影像內(nèi)同一種作物的不同田塊應用同一套“折衷”的CR模型參數(shù)化方案,造成了一定的模型參數(shù)化誤差。并且,常規(guī)CR模型反演方法逐像元獨立進行反演,忽略了像元之間的空間關系,未能充分利用遙感數(shù)據(jù)所蘊含的空間信息。針對此,Atzberger[31,52]提取田塊和臨近像元的空間特征來約束農(nóng)作物參數(shù)的遙感反演,提出了面向?qū)ο蠓囱莘?。相關研究(見表2)可大致分為2類。一類是基于統(tǒng)計模型的面向?qū)ο蠓囱莘ǎ豪锰飰K的空間特征作為額外的LAI光譜特征參量,參與建立LAI的統(tǒng)計預測模型[52];另一類是基于CR模型的面向?qū)ο蠓囱莘ǎ禾崛√飰K的空間特征(主要是田塊內(nèi)不敏感參數(shù)及其期望)作為CR模型反演的額外約束條件[31]。限于本文主題為 CR模型多階段反演,筆者僅關注基于CR模型的面向?qū)ο蠓囱莘?。本文總結(jié)了前人基于面向?qū)ο蟮姆囱莘椒ǚ囱萘值睾娃r(nóng)作物參數(shù)的相關研究,見表2。
表2 基于面向?qū)ο蠓囱莘ǖ霓r(nóng)作物參數(shù)反演研究Table 2 Studies on retrieving crop variables based on object-based method
2.1 面向?qū)ο蟮腃R模型反演方法原理和流程
面向?qū)ο蟮腃R模型反演方法將農(nóng)作物幾何結(jié)構、生理生化參數(shù)和土壤背景狀況較均一的若干臨近像元定義為一個“對象”。針對每個對象確定獨立的參數(shù)化方案,并分階段反演。首先,通過預反演獲取對象的空間特征,即對象中具有一定空間分布規(guī)律的植被或環(huán)境參數(shù)期望值和取值范圍,例如對象內(nèi)空間異質(zhì)性較小的SAIL模型熱點因子(Hot)、Cab、葉片干物質(zhì)含量(Cm)和土壤反射率(αsoil)等。以此作為先驗知識,進而在對象內(nèi)部逐像元反演空間異質(zhì)性較顯著的參數(shù)和感興趣參數(shù),如LAI。通過這種方法,面向?qū)ο蟮姆囱莘椒ɡ棉r(nóng)作物參數(shù)的空間分布規(guī)律和遙感數(shù)據(jù)中的空間信息為CR模型反演提供額外的約束條件,改善模型參數(shù)優(yōu)化中的不確定性。
面向?qū)ο蠓囱莸牟僮髁鞒倘缦?。第一步,在中、高分辨率遙感影像中劃分對象,確保對象內(nèi)部農(nóng)作物參數(shù)和土壤背景狀況較均一。對象一般是獨立田塊,也可以是鄰近像元組成的移動窗。此步驟可利用目視解譯、遙感影像分類或空間分割實現(xiàn)。第二步,預反演,即從對象中提取對象空間特征作為后續(xù)反演的先驗知識。隨著對象劃分方法的不同,預反演過程可能包含一個或多個反演步驟。例如,Laurent等[32]首先以LAI、Cab、覆蓋度(Cv)、葉片含水量(Cw)、Cm、葉肉結(jié)構參數(shù)(N)、棕色葉片占總LAI比例(fB)為自由變量,進行田塊尺度4SAIL2+MODTRAN4 耦合模型的預反演。除LAI、Cab外其他參數(shù)的反演結(jié)果將作為下一步反演的先驗知識。Atzberger和Richter[31]假設Hot、Cab、Cm和αsoil在3×3像元組成的移動窗內(nèi)一致。據(jù)此,在移動窗內(nèi)優(yōu)化上述參數(shù),再以其反演結(jié)果為先驗知識,在田塊尺度內(nèi)優(yōu)化平均葉傾角(θl)。第三步,逐像元反演。即以對象空間特征反演結(jié)果作為先驗知識,在特定對象范圍內(nèi)逐像元反演LAI和Cab。例如,Laurent等[32]在田塊內(nèi)的逐像元反演中,將Cv、Cw、Cm、N、fB固定為預反演結(jié)果。Atzberger和 R ichter[31]以預反演獲取的 H ot、Cab、Cm、αsoil為先驗知識,在田塊范圍內(nèi)逐像元反演LAI。
2.2 對象空間特征的提取與利用方法
面向?qū)ο蟮姆囱莘椒ɑ趯ο髢?nèi)各個像元的光譜特征提?。ǚ囱荩ο罂臻g特征,其方法可大致歸為如下三類。1)求對象的平均光譜特征。Laurent等[32,54]利用對象內(nèi)所有像元的平均光譜反射率反演對象空間特征。2)求對象的整體代價函數(shù)。Atzberger和Richter[31]以對象內(nèi)所有像元代價函數(shù)之和作為對象空間特征反演的代價函數(shù)。3)從對象內(nèi)各像元的反演結(jié)果中篩選合理的參數(shù)組合。Houborg等[25]假設,LAI反演結(jié)果達到邊界值的像元數(shù)量如果超過對象內(nèi)總像元數(shù)的2%,則判定該參數(shù)組合“不合理”,予以剔除。
為了利用預反演獲取的對象空間特征,降低對象內(nèi)逐像元LAI和Cab反演的不確定性,表2中基于CR模型的研究均利用預反演結(jié)果作為對象內(nèi)對應參數(shù)的期望值。同時,CR模型參數(shù)間的相互關系也可作為反演的限制條件。例如,為了克服LAI對Cab反演的影響,Laurent等[54]基于LAI-Cab的統(tǒng)計關系,依據(jù)預反演的LAI值確定Cab的取值范圍:當LAI<4時,Cab最小值為10 μg/cm2;當LAI≥4時,Cab最小值為15 μg/cm2。
2.3 面向?qū)ο蠓囱莘椒ǖ膬?yōu)勢與局限性
面向?qū)ο蟮腃R模型反演方法起步晚,就本文的統(tǒng)計而言,相關研究少,但發(fā)展迅速。前人研究(見表2)表明,該方法對于約束“病態(tài)反演”現(xiàn)象,提高農(nóng)作物參數(shù)反演的精度和穩(wěn)定性效果良好。然而,其方法遠未成熟,還存在一些技術難點。1)CR模型參數(shù)的空間約束條件設置缺乏試驗依據(jù),甚至僅僅是“合理”的假設。例如,Atzberger和Richter[31]假設9像元移動窗內(nèi)Hot、Cab、Cm、αsoil一致;田塊內(nèi)θl一致;而Laurent等[54]則假設田塊內(nèi) Cv、fB、Cw、Cm、N一致。不同研究的作物參數(shù)空間分布特征各異,且都缺乏田間測量數(shù)據(jù)支持。未來研究首先應基于大量的,具有統(tǒng)計意義的地面觀測數(shù)據(jù),詳細探索特定作物的各個參數(shù)在田塊內(nèi)部和田塊之間的空間分布規(guī)律,為合理設置空間約束條件提供依據(jù)。其次,遙感獲取的某些光譜特征參量(如植被指數(shù)[55-56]、紅邊參量[57-58]、敏感波段的光譜反射率[59-60]、小波變換得到的特征量[61]等)與CR模型某些參數(shù)顯著相關。在面向?qū)ο蟮姆囱葜?,可以基于這些參量與模型參數(shù)的相關關系,利用光譜數(shù)據(jù)和(或)相關遙感數(shù)據(jù)產(chǎn)品,考察模型參數(shù)的空間分布,為合理設置空間約束條件提供依據(jù)。2)一方面,對象空間特征提取結(jié)果缺乏驗證,易誤導對象內(nèi)逐像元LAI和Cab反演。另一方面,對象空間特征在后續(xù)逐像元反演中的最佳表達與利用方法尚缺乏科學對比,詳見3.3節(jié)。
由上文總結(jié)可見,MSDT法和面向?qū)ο蠓囱莘ǖ幕舅悸范际菍R模型反演過程劃分為多個階段,前一階段反演結(jié)果作為后階段反演的先驗知識。兩種方法區(qū)別在于反演決策的依據(jù)不同(見圖2)。MSDT法利用CR模型參數(shù)的敏感性決定反演流程;而面向?qū)ο蟮姆囱莘椒ɡ脜?shù)的空間分布規(guī)律決定反演流程。因此,本文將兩種方法統(tǒng)稱為“多階段反演方法”,其概念模型見圖2。
圖2 多階段反演法概念模型Fig.2 Conceptual framework of multi-step inversion
3.1 反演決策合理性和有效性
合理的反演決策是多階段反演達到預期目標的前提。然而,前人研究提出的多階段反演決策還存在不足之處。
就反演順序而言,MSDT法認為應當優(yōu)先利用部分觀測數(shù)據(jù)反演對其最敏感,最不確定的參數(shù),待其不確定性降低后,再用觀測數(shù)據(jù)的另一子集反演另一部分參數(shù)[33-34,37,40]。而面向?qū)ο蟮姆囱萃谧畛醯念A反演階段將CR模型所有敏感參數(shù)設為自由變量,提取田塊內(nèi)空間異質(zhì)性較小的,往往也是相對不敏感的參數(shù)特征[24-25,31-32,54]??梢?,從光譜特征、空間特征兩方面挖掘先驗知識的不同考慮得到了截然相反的多階段反演方案,這顯然是不合理的。然而,國內(nèi)外多階段反演方法研究剛剛起步,目前仍缺乏對其反演決策合理性和有效性的比較研究和科學論證。未來應依據(jù)反演試驗結(jié)果,系統(tǒng)、科學地比較多階段反演方案,將MSDT法與面向?qū)ο蠓囱莘椒ㄓ袡C結(jié)合,在統(tǒng)一的多階段反演技術框架下,制定更加合理的反演決策方法。
3.2 CR模型參數(shù)化精度對多階段反演的影響
多階段反演過程中,當CR模型參數(shù)的期望值顯著偏離真值時,一方面,如果未參與反演的參數(shù)被固定于錯誤的期望值,將造成模型參數(shù)化誤差,誤導反演過程;另一方面,如果MSDT法的待反演參數(shù)取值范圍不當,將造成敏感性與不確定性評價結(jié)果有誤,導致反演順序決策錯誤,致使反演失敗[34,62]??梢姡瑓?shù)的期望值和取值范圍對多階段反演有明顯影響。為了提高模型參數(shù)化精度,前人從以下方面獲取關于CR模型參數(shù)的先驗知識,1)參數(shù)的空間分布規(guī)律。例如,Houborg等[25]假設土壤反射率在小區(qū)域內(nèi)一致,利用 L AI<0.5的像元反演土壤反射率參數(shù)(s1和s2),作為后續(xù)反演的先驗知識。然而,不同田塊間由于耕作方法和灌水量的不同,土壤反射率往往存在明顯差異。一個田塊上獲取的s1和s2能否用于另一田塊仍缺少科學論證。2)多尺度遙感數(shù)據(jù)。例如,馮曉明與趙英時[37]利用小波變換,將MISR(multiangle imaging spectro radiometer)多角度遙感影像分解為4個不同尺度,分別進行MSDT反演。以大尺度反演結(jié)果作為先驗知識參與小尺度反演。然而,大尺度反演時,仍缺少輔助CR模型參數(shù)化的先驗知識。3)地面觀測數(shù)據(jù)。例如,朱小華等[40]繼承了“大尺度反演為小尺度反演提供先驗知識”的思路,并以地面實測地表參數(shù)作為大尺度(1 km分辨率)反演的先驗知識。然而,由于點狀地面數(shù)據(jù)與面狀遙感數(shù)據(jù)的尺度差異,地面數(shù)據(jù)可能無法反映1 km尺度上的真實情況[63]。加之將實測數(shù)據(jù)用于最大尺度的模型參數(shù)化,使得尺度效應最大化。與廣泛采用的,以高分辨率遙感數(shù)據(jù)作為橋梁,將地表觀測數(shù)據(jù)向中、低分辨率進行尺度轉(zhuǎn)換的技術路線[64-66]相矛盾。4)遙感產(chǎn)品(例如MODIS LAI產(chǎn)品)。Houborg等[25,67]利用MODIS LAI數(shù)據(jù)產(chǎn)品輔助CR模型參數(shù)化。然而,首先,現(xiàn)有LAI產(chǎn)品的空間分辨率普遍較低[19],與小尺度研究廣泛采用的米級至30 m級遙感數(shù)據(jù)之間存在尺度效應;其次,LAI遙感產(chǎn)品普遍存在一定的系統(tǒng)誤差,若缺少地面測量數(shù)據(jù)則無法加以驗證和消除[68]。
綜上所述,多階段反演受CR模型參數(shù)的初始期望和取值范圍影響顯著,依賴先驗知識。參數(shù)的空間分布規(guī)律、同區(qū)域不同尺度反演結(jié)果、地面實測數(shù)據(jù)或遙感數(shù)據(jù)產(chǎn)品四方面信息都在一定程度上反映了地表植被參數(shù)的真實狀況,但都有其各自的局限,也就是說,往往缺乏全面、可靠的關于CR模型參數(shù)的先驗知識。未來研究中,一方面,應深入探索CR模型參數(shù)化誤差對多階段反演的影響,探索減小上述影響的反演策略。另一方面,近年來尺度轉(zhuǎn)換與多尺度驗證方法逐漸發(fā)展完善[63,69-70];高空間分辨率遙感數(shù)據(jù),特別是無人機遙感數(shù)據(jù)逐漸普及[71],為地面實測數(shù)據(jù)與遙感數(shù)據(jù)協(xié)同應用提供了一定的方法和物質(zhì)基礎。嘗試利用高分辨率衛(wèi)星和無人機遙感數(shù)據(jù),并結(jié)合上述四方面先驗知識,實現(xiàn)“天空地一體化”的農(nóng)情信息獲取,進而利用尺度轉(zhuǎn)換方法從中挖掘先驗知識,是提高CR模型參數(shù)化精度的一個值得嘗試的研究方向。
3.3 反演誤差逐級傳遞
多階段反演過程中,先驗知識逐步積累,反演誤差也逐級傳遞。前一階段反演和先驗知識提取的準確性影響后續(xù)各階段反演。因此,1)應嘗試驗證每一階段的反演精度,發(fā)現(xiàn)并修正可能存在的系統(tǒng)誤差和極端異常值。以往研究缺乏對前一階段反演結(jié)果的驗證方法。未來研究中,可以嘗試利用適量的地面觀測數(shù)據(jù)、光譜特征參量和相關遙感產(chǎn)品,識別并糾正前階段反演中的誤差。例如,F(xiàn)ang和Liang利用土壤反射率指數(shù)(soil reflectance index,SRI)簡化了GeoSAIL模型中土壤反射率的參數(shù)化[55]。在多階段反演中,則可以嘗試將利用SRI識別土壤反射率反演的極端異常值。2)后階段反演應當合理利用前階段反演結(jié)果,避免反演誤差的影響。由于前一階段反演的隨機誤差和田塊內(nèi)部空間異質(zhì)性不可避免,導致即使前階段沒有明顯反演錯誤,其反演結(jié)果與真實值仍存在一定差異。因此,如果基于前階段反演結(jié)果,過度限制后階段反演中參數(shù)(特別是較敏感參數(shù),例如,基于近紅外波段反演時的LAI、比葉重(specific leaf weight,SLW)和平均葉傾角[46])的取值范圍,那么,其參數(shù)化誤差將顯著影響后續(xù)反演。未來研究應針對各個參數(shù)的敏感性和空間異質(zhì)性,分別制定多階段參數(shù)優(yōu)化方案。先驗知識可靠性無法保障時,應該避免固定敏感或空間異質(zhì)性強的參數(shù),而應結(jié)合前期反演所得到的期望和方差,在不引入顯著參數(shù)化誤差的情況下,謹慎縮小其不確定性范圍。例如,Atzberger和Richter[31]假設3×3像元移動窗內(nèi)Hot、Cab、Cm和αsoil一致,但允許上述參數(shù)在田塊范圍內(nèi)有所變化。在限制“病態(tài)反演”問題的同時考慮對象內(nèi)部客觀存在的參數(shù)空間異質(zhì)性,防止先驗知識的誤差誤導反演,防止反演誤差累積。
多階段目標決策(multi-stage,sample-direction dependent,target-decisions,MSDT)法與面向?qū)ο蠓囱莘蓺w納為“多階段反演”。其優(yōu)勢在于:1)每階段只反演部分參數(shù),從而減少CR模型參數(shù)優(yōu)化的不確定性,改善“病態(tài)反演”問題。2)從前一階段反演結(jié)果和(或)遙感影像的空間信息中提取先驗知識,更有利于先驗知識的合理利用和觀測數(shù)據(jù)的有效分配。多階段反演法在反演決策合理性和有效性、CR模型參數(shù)化精度對多階段反演的影響,以及反演誤差逐級傳遞方面還存在技術難點。未來研究應重點從以下方面對多階段反演法加以改進。1)將MSDT法與面向?qū)ο蠓囱莘椒ㄓ袡C結(jié)合,在統(tǒng)一的多階段反演技術框架下,制定更加合理的反演決策方法。2)嘗試 “天空地一體化”的農(nóng)情信息獲取,進而利用尺度轉(zhuǎn)換方法從中挖掘先驗知識,提高CR模型參數(shù)化精度。3)嘗試識別并糾正前階段反演中的誤差。4)合理利用前階段反演結(jié)果,避免反演誤差的影響。
[1] 霍建英,史文娟,彭程. 楊凌智慧農(nóng)業(yè)發(fā)展現(xiàn)狀問題及對策[J]. 西安郵電大學學報,2012,17(4):100-103. Huo Jianying,Shi Wenjuan,Peng Cheng. The current problem and strategy of intelligent agriculture in Yangling[J]. Journal of Xi’an University of Posts and Telecommunications,2012,17(4):100-103.(in Chinese with English abstract)
[2] 李道亮. 物聯(lián)網(wǎng)與智慧農(nóng)業(yè)[J]. 農(nóng)業(yè)工程,2012,2(1):1-7. Li Daoliang. Internet of things and wisdom agriculture[J]. Agricultural Engineering,2012 2(1):1-7.(in Chinese with English abstract)
[3] Chen J M,Black T. Defining leaf area index for non-flat leaves[J]. Plant,Cell &Environment,1992,15(4):421-429.
[4] Verrelst J,Romijn E,Kooistra L. Mapping vegetation density in a heterogeneous river floodplain ecosystem using pointable CHRIS/PROBA data[J]. Remote Sensing,2012,4(12):2866-2889.
[5] Bala G,Joshi J,Chaturvedi R K,et al. Trends and variability of AVHRR-derived NPP in India[J]. Remote Sensing,2013,5(2):810-829.
[6] Doughty C E,Goulden M L. Seasonal patterns of tropical forest leaf area index and CO2exchange[J]. Journal of Geophysical Research Atmospheres,2008,113(G1):440-445.
[7] Palmer A R,Fuentes S,Taylor D,et al. Towards a spatial understanding of water use of several land-cover classes:an examination of relationships amongst pre-dawn leaf water potential,vegetation water use,aridity and MODIS LAI[J]. Ecohydrology,2010,3(1):1-10.
[8] Netto A T,Campostrini E,de Oliveira J G,et al. Photosynthetic pigments,nitrogen,chlorophyll a fluorescence and SPAD-502 readings in coffee leaves[J]. Scientia Horticulturae,2005,104(2):199-209.
[9] Houborg R,Cescatti A,Migliavacca M,et al. Satellite retrievals of leaf chlorophyll and photosynthetic capacity for improved modeling of GPP[J]. Agricultural and Forest Meteorology,2013,177(1):10-23.
[10] 李柏貞,周廣勝. 干旱指標研究進展[J]. 生態(tài)學報,2014,34(5):1043-1052. Li Bozhen,Zhou Guangsheng. Advance in the study on drought index[J]. Acta Ecologica Sinica,2014,34(5):1043- 1052.(in Chinese with English abstract)
[11] 王維,陳建軍,呂永華,等. 烤煙氮素營養(yǎng)診斷及精準施肥模式研究[J]. 農(nóng)業(yè)工程學報,2012,28(9):77-84. Wang Wei,Chen Jianjun,Lü Yonghua,et al. Research on nitrogen nutrition diagnosis and precision fertilizing model for flue-cured tobacco[J]. Transactions of the Chinese Society of Agricultural Engineering(Transactions of the CSAE),2012,28(9):77-84.(in Chinese with English abstract)
[12] Atzberger C,Darvishzadeh R,Immitzer M,et al. Comparative analysis of different retrieval methods for mapping grassland leaf area index using airborne imaging spectroscopy[J]. International Journal of Applied Earth Observation and Geoinformation,2015,43:19-31.
[13] Zhu Z,Bi J,Pan Y,et al. Global data sets of vegetation leaf area index(LAI) 3g and Fraction of Photosynthetically Active Radiation(FPAR) 3g derived from Global Inventory Modeling and Mapping Studies(GIMMS) Normalized Difference Vegetation Index(NDVI3g) for the period 1981 to 2011[J]. Remote Sensing,2013,5(2):927-948.
[14] Sakamoto T,Shibayama M,Kimura A,et al. Assessment of digital camera-derived vegetation indices in quantitative monitoring of seasonal rice growth[J]. ISPRS Journal of Photogrammetry and Remote Sensing,2011,66(6):872-882.
[15] 宋曉宇,王紀華,黃文江,等. 變量施肥條件下冬小麥長勢及品質(zhì)變異遙感監(jiān)測[J]. 農(nóng)業(yè)工程學報,2009,25(9):155-162. Song Xiaoyu,Wang Jihua,Huang Wenjiang,et al. Monitoring spatial variance of winter wheat growth and grain quality under variable-rate fertilization conditions by remote sensing data[J]. Transactions of the Chinese Society of Agricultural Engineering(Transactions of the CSAE),2009,25(9):155-162.(in Chinese with English abstract)
[16] Das D K,Pradhan S,Sehgal V K,et al. Spectral reflectance characteristics of healthy and yellow mosaic virus infected soybean(Glycine max L.) leaves in a semiarid environment [J]. Journal of Agrometeorology,2013,15(1):36-38.
[17] Fang H,Liang S,Hoogenboom G. Integration of MODIS LAI and vegetation index products with the CSM-CERESMaize model for corn yield estimation[J]. International Journal of Remote Sensing,2011,32(4):1039-1065.
[18] 李樹強,李民贊. 基于車載近地遙感系統(tǒng)的冬小麥生育早期產(chǎn)量估測方法[J]. 農(nóng)業(yè)工程學報,2014,30(3):120-127. Li Shuqiang,Li Minzan. Yield estimation of winter wheat in early growth periods by vehicle-borne ground-based remote sensing system[J]. Transactions of the Chinese Society of Agricultural Engineering(Transactions of the CSAE),2014,30(3):120-127.(in Chinese with English abstract)
[19] 梁順林,張曉通,肖志強,等. 全球陸表特征參量(GLASS)產(chǎn)品算法、驗證與分析[M]. 北京:高等教育出版社,2014.
[20] Justice C O,Townshend J R G,Vermote E F,et al. An overview of MODIS Land data processing and product status [J]. Remote Sensing of Environment,2002,83(1/2):3-15.
[21] Baret F,Hagolle O,Geiger B,et al. LAI,fAPAR and fCover CYCLOPES global products derived from VEGETATION[J]. Remote Sensing of Environment,2007,110(3):275-286.
[22] Yao Y,Liu Q,Liu Q,et al. LAI retrieval and uncertainty evaluations for typical row-planted crops at different growth stages[J]. Remote Sensing of Environment,2008,112(1):94-106.
[23] Combal B,Baret F,Weiss M,et al. Retrieval of canopy biophysical variables from bidirectional reflectance:Using prior information to solve the ill-posed inverse problem[J]. Remote Sensing of Environment,2002,84(1):1-15.
[24] Houborg R,Anderson M,Daughtry C. Utility of an image-based canopy reflectance modeling tool for remote estimation of LAI and leaf chlorophyll content at the field scale[J]. Remote Sensing of Environment,2009,113(1):259-274.
[25] Houborg R,McCabe M,Cescatti A,et al. Joint leaf chlorophyll content and leaf area index retrieval from Landsat data using a regularizedmodel inversion system(REGFLEC)[J]. Remote Sensing of Environment,2015,159:203-221.
[26] Duan S B,Li Z L,Wu H,et al. Inversion of the PROSAIL model to estimate leaf area index of maize,potato,and sunflower fields from unmanned aerial vehicle hyperspectral data[J]. International Journal of Applied Earth Observation and Geoinformation,2014,26(2):12-20.
[27] Darvishzadeh R,Skidmore A,Schlerf M,et al. Inversion of a radiative transfer model for estimating vegetation LAI and chlorophyll in a heterogeneous grassland[J]. Remote Sensing of Environment,2008,112(5):2592-2604.
[28] Si Y,Schlerf M,Zurita-Milla R,et al. Mapping spatiotemporal variation of grassland quantity and quality using MERIS data and the PROSAIL model[J]. Remote Sensing of Environment,2012,121:415-425.
[29] Lauvernet C,Baret F,Hasco?t L,et al. Multitemporal-patch ensemble inversion of coupled surface-atmosphere radiative transfer models for land surface characterization[J]. Remote Sensing of Environment,2008,112(3):851-861.
[30] Bacour C,Jacquemoud S,Leroy M,et al. Reliability of the estimation of vegetation characteristics by inversion of three canopy reflectance models on airborne polder data[J]. Agronomie,2002,22(6):555-565.
[31] Atzberger C,Richter K. Spatially constrained inversion of radiative transfer models for improved LAI mapping from future Sentinel-2 imagery[J]. Remote Sensing of Environment,2012,120:208-218.
[32] Laurent V C E,Schaepman M E,Verhoef W,et al. Bayesian object-based estimation of LAI and chlorophyll from a simulated Sentinel-2 top-of-atmosphere radiance image[J]. Remote Sensing of Environment,2014,140:318-329.
[33] 李小文,高峰,王錦地,等. 遙感反演中參數(shù)的不確定性與敏感性矩陣[J]. 遙感學報,1997,1(1):5-14. Li Xiaowen,Gao Feng,Wang Jindi,et al. Uncertainty and sensitivity matrix of parameters in inversion of physical BRDF model[J]. Journal of Remote Sensing,1997,1(1):5-14.(in Chinese with English abstract)
[34] 高峰,李小文,夏宗國,等. 基于知識的分階段不確定性多角度遙感反演[J]. 中國科學:D輯 地球科學,1998,28(4):346-350.
[35] 顏春燕,劉強,牛錚,等. 植被生化組分的遙感反演方法研究[J]. 遙感學報,2004,8(4):300-308. Yan Chunyan,Liu Qiang,Niu Zheng,et al. Inversion of vegetation biochemicals by remote sensing[J]. Journal of Remote Sensing,2004,8(4):300-308.(in Chinese with English abstract)
[36] Feng X,Chen Y,Zhao Y. Research on the spectraldirectional soil model in a new remote sensing reflectance model of the semiarid landscape[C]//IGARSS 2004. 2004 IEEE International Geoscience and Remote Sensing Symposium,Anchorage,AK,2004,7:4662-4665.
[37] 馮曉明,趙英時. 多角度衛(wèi)星遙感多階段目標決策反演研究[J]. 中國科學:D輯 地球科學,2006,36(7):672-679.
[38] Feng X,Zhao Y. On MSDT inversion with multi-angle remote sensing data[J]. Science in China Series D:Earth Sciences,2007,50(3):422-429.
[39] 萬華偉,王錦地,梁順林,等. 聯(lián)合MODIS與MISR遙感數(shù)據(jù)估算葉面積指數(shù)[J]. 光譜學與光譜分析,2009,29(11):3106-3111. Wan Huawei,Wang Jindi,Liang Shunlin,et al. Estimating leaf area index by fusing MODIS and MISR data[J]. Spectroscopy and Spectral Analysis,2009,29(11):3106-3111.(in Chinese with English abstract)
[40] 朱小華,馮曉明,趙英時. 基于LAI空間知識的多尺度多階段目標決策反演[J]. 中國科學:D輯 地球科學,2012,42(2):246-255. Zhu Xiaohua,Feng Xiaoming,Zhao Yingshi. Multi-scale MSDT inversion based on LAI spatial knowledge[J]. Science China:Earth Science,2012,42(2):246-255.(in Chinese with English abstract)
[41] Zhu X,Feng X,Zhao Y. Multi-scale MSDT inversion based on LAI spatial knowledge[J]. Science China:Earth Sciences,2012,55(8):1297-1305.
[42] Ma H,Song J,Wang J,et al. Improvement of spatially continuous forest LAI retrieval by integration of discrete airborne LiDAR and remote sensing multi-angle optical data[J]. Agricultural and Forest Meteorology,2014,189-190(189/190):60-70.
[43] Saltelli A,Tarantola S,Campolongo F. Sensitivity Analysis in Practice:A Guide to Assessing Scientific Models[M]. Chichester:John Wiley &Sons Inc.,2004.
[44] LI X,Wang J,Hu B,et al. On utilization of a priori knowledge in inversion of remote sensing models[J]. Science in China:Series D,1998,41(6):580-585.
[45] Saltelli A,Tarantola S,Chan K P S. A quantitative modelindependent method for global sensitivity analysis of model output[J]. technometrics,1999,44(1):39-39.
[46] 劉軻,周清波,吳文斌,等. 基于多光譜與高光譜遙感數(shù)據(jù)的冬小麥葉面積指數(shù)反演比較[J]. 農(nóng)業(yè)工程學報,2016,32(3):155-162. Liu Ke,Zhou Qingbo,Wu Wenbin,et al. Comparison between multispectral and hyperspectral remote sensing for LAI estimation[J]. Transactions of the Chinese Society ofAgricultural Engineering(Transactions of the CSAE),2016,32(3):155-162.(in Chinese with English abstract)
[47] 何維,楊華. 模型參數(shù)全局敏感性分析的EFAST方法[J].遙感技術與應用,2013,28(5):836-843. He Wei,Yang Hua. EFAST method for global sensitivity analysis of remote sensing model’s parameters[J]. Remote sensing technology and application,2013,28(5):836-843.(in Chinese with English abstract)
[48] Bowyer P,Danson F. Sensitivity of spectral reflectance to variation in live fuel moisture content at leaf and canopy level[J]. Remote Sensing of Environment,2004,92(3):297-308.
[49] Ma C,Li X,Wang S. A Global Sensitivity Analysis of Soil Parameters Associated With Backscattering Using the Advanced Integral Equation Model[J]. IEEE Transactions on Geoscience and Remote Sensing,2015,53(10):5613-5623.
[50] Qu Y,Wang J,Wan H,et al. A Bayesian network algorithm for retrieving the characterization of land surface vegetation [J]. Remote Sensing of Environment,2008,112(3):613-622.
[51] 趙虎,裴志遠,馬尚杰,等. WOFOST 模型同化時序HJ CCD數(shù)據(jù)反演葉面積指數(shù)[J]. 農(nóng)業(yè)工程學報,2012,28(11):158-163. Zhao Hu,Pei Zhiyuan,Ma Shangjie,et al. Retrieving LAI by assimilating time series HJ CCD with WOFOST[J]. Transactions of the Chinese Society of Agricultural Engineering(Transactions of the CSAE),2012,28(11):158-163.(in Chinese with English abstract)
[52] Atzberger C. Object-based retrieval of biophysical canopy variables using artificial neural nets and radiative transfer models[J]. Remote Sensing of Environment,2004,93(1/2):53-67.
[53] Curran P J,Atkinson P M. Geostatistics and remote sensing [J]. Progress in Physical Geography,1998,22(1):61-78.
[54] Laurent V C E,Verhoef W,Damm A,et al. A Bayesian object-based approach for estimating vegetation biophysical and biochemical variables from APEX at-sensor radiance data[J]. Remote Sensing of Environment,2013,139:6-17.
[55] Fang H,Liang S. A hybrid inversion method for mapping leaf area index from MODIS data:Experiments and application to broadleaf and needleleaf canopies[J]. Remote Sensing of Environment,2005,94(3):405-424.
[56] Broge N H,Leblanc E. Comparing prediction power and stability of broadband and hyperspectral vegetation indices for estimation of green leaf area index and canopy chlorophyll density[J]. Remote Sensing of Environment,2000,76(2):156-172.
[57] 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.
[58] Pu R,Gong P,Biging G S,et al. Extraction of red edge optical parameters from Hyperion data for estimation of forest leaf area index[J]. IEEE Trans Geosci Remote Sens,2003,41(4):916-921.
[59] Houborg R,Boegh E. Mapping leaf chlorophyll and leaf area index using inverse and forward canopy reflectance modeling and SPOT reflectance data[J]. Remote Sensing of Environment,2008,112(1):186-202.
[60] 夏天,吳文斌,周清波,等. 冬小麥葉面積指數(shù)高光譜遙感反演方法對比[J]. 農(nóng)業(yè)工程學報,2013,29(3):139-147. Xia Tian,Wu Wenbin,Zhou Qingbo,et al. Comparison of two inversion methods for winter wheat leaf area index based on hyperspectral remote sensing[J]. Transactions of the Chinese Society of Agricultural Engineering(Transactions of the CSAE),2013,29(3):139-147.(in Chinese with English abstract)
[61] Pu R,Gong P. Wavelet transform applied to EO-1 hyperspectral data for forest LAI and crown closure mapping[J]. Remote Sensing of Environment,2004,91(2):212-224.
[62] 李小文,王錦地,胡寶新,等. 先驗知識在遙感反演中的作用[J]. 中國科學:D輯,1998,28(1):67-72.
[63] 彭菁菁,劉強,聞建光,等. 衛(wèi)星反照率產(chǎn)品的多尺度驗證與不確定性分析[J]. 中國科學:地球科學,2015,45(1):66-82. Peng Jingjing,Liu Qiang,Wen Jianguang,et al. Multi-scale validation strategy for satellite albedo products and its uncertainty analysis[J]. Science China:Earth Sciences,2015,45(1):66-82.(in Chinese with English abstract)
[64] Wang K,Liu J,Zhou X,et al. Validation of MODIS global land surface albedo product using ground measurements in a semidesert region on the Tibetan Plateau[J]. Journal of Geophysical Research Atmospheres,2004,109(5):385-399.
[65] 家淑珍,馬明國,于文憑. 黑河中游LAI產(chǎn)品的真實性檢驗研究[J]. 遙感技術與應用,2014,29(6):1037-1045. Jia Shuzhen,Ma Mingguo,Yu Wenping. Validation of the LAI product in Heihe River Basin[J]. Remote Sensing Technology and Application,2014,29(6):1037-1045.(in Chinese with English abstract)
[66] 趙海軍,紀力強. 大尺度生物多樣性評價[J]. 生物多樣性,2003,11(1):78-85. Zhao Haijun,Ji Liqiang. Biodiversity assessment at broad scale[J]. Biodiversity Science,2003,11(1):78-85.(in Chinese with English abstract)
[67] Houborg R,Soegaard H,Boegh E. Combining vegetation index and model inversion methods for the extraction of key vegetation biophysical parameters using Terra and Aqua MODIS reflectance data[J]. Remote Sensing of Environment,2007,106(1):39-58.
[68] 張仁華,田靜,李召良,等. 定量遙感產(chǎn)品真實性檢驗的基礎與方法[J]. 中國科學:地球科學,2010,40(2):211-222. Zhang Renhua,Tian Jing,Li Zhaoliang,et al. Principles and methods for the validation of quantitative remote sensing products[J]. Science China:Earth Science,2010,40(2):211-222.(in Chinese with English abstract)
[69] Mayr M,Samimi C. Comparing the dry season in-situ leaf area index(LAI) derived from high-resolution RapidEye imagery with MODIS LAI in a Namibian savanna[J]. Remote Sensing,2015,7(4):4834-4857.
[70] Tao X,Yan B,Wang K,et al. Scale transformation of Leaf Area Index product retrieved from multiresolution remotely sensed data:analysis and case studies[J]. International Journal of Remote Sensing,2009,30(20):5383-5395.
[71] Shi Y,Ji S,Shao X,et al. Framework of SAGI agriculture remote sensing and its perspectives in supporting national food security[J]. Journal of Integrative Agriculture,2014,13(7):1443-1450.
Review on multi-stage inversion techniques of canopy reflectance models for retrieving crop variables
Liu Ke1,Huang Ping1※,Ren Guoye1,Zhou Qingbo2,Li Yuanhong1,Wang Si1,Dong Xiuchun1
(1. Institute of Remote Sensing Application,Sichuan Academy of Agricultural Science /Chengdu Branch of Remote Sensing Application Center,Ministry of Agriculture,Chengdu 610066,China;2. Institute of Agricultural Resources and Regional Planning,Chinese Academy of Agricultural Sciences /Key Laboratory of Agri-informatics,Ministry of Agriculture,Beijing 100081,China)
Remote sensing technique is known as an inexpensive and effective tool for retrieving crop variables in a large area. The existing methodologies can be identified into two categories:the methodologies based on statistical predictive models and the methodologies based on canopy reflectance(CR) models inversion. The latter is relatively universal. Thus,it has great potential in wisdom agriculture for crop monitoring in regional scale. However,CR model inversions suffer from the so-called“ill-posed problem”. Therefore,the multi-stage,sample-direction dependent,target-decisions(MSDT) inversion technique and the object-based inversion technique were previously proposed. They are similar in technical routes:the progress of an inversion is partitioned into several stages. In each stage,only a part of variables were estimated. The results of preliminary stages are used as prior knowledge of later stages of inversion. In this way,the uncertainties in parameter optimization are reduced,the ill-posed problem is therefore limited. Concretely speaking,the MSDT method firstly estimates the sensitivity and uncertainties of variables before each stage of inversion. The most sensitive and uncertain variables were firstly retrieved using a subset of remote sensing data which is sensitive to the retrieved variables. The scheme of parameterization is then updated based on the preliminary results. Another subset of sensitive variables was subsequently retrieved using another subset of sensitive data. The object-based inversion defines an “object” as a plot or a gliding window,in which the crop has similar attributes. Such attributes are referred to as “object signatures”. A remotely sensed image is firstly segmented into objects. Within each object,object signatures are firstly retrieved,and used as prior knowledge in subsequent pixel-wise retrieval of spatial heterogeneous or interested variables. In this way,spatial constrains,i.e.,the spatial distribution of variables,are extracted and imposed on the inversion. It can be seen the MSDT and object-based inversion essentially follow the same procedure. The major difference between them is that MSDT method makes the scheme of inversion according to the sensitivity and uncertainty of variables,while object-based inversion is based on the spatial distribution of variables. In this review,MSDT and object-based inversions were summarized into an integrated conceptual framework of “multi-stage inversion”. Based on this framework,the following technical problems and the potential solutions can be summarized as follows. 1) The schemes of MSDT and object-based inversions are practically in conflict. In future studies,multi-step inversion strategies need further comparison,verification and improvement to ensure their rationality and effectiveness. The thoughts of MSDT and object-based inversions should be integrated,to develop more sophisticated inversion schemes under the conceptual framework of multi-step inversion. 2) Multi-step inversions might be significantly affected by the accuracy of preliminary parameterization of CR model. In future studies,the integrated application of multi-sources data could be helpful for CR model parameterization,and for detecting errors in each stage of inversion. For instance,same variables can be retrieved from satellite,aerial and ground remote sensing data,or obtained directly from in-situ measurements and existing remote sensing products. With approaches of scale transformation,the variables retrieved from multi-source data can be compared,in order to obtain prior-knowledge,or detect error in inversions. 3) Multi-step inversions might be distorted by error propagation. In future studies,on the one hand,gross errors and systematic errors should be detected and corrected in each stage of inversion according to the statistical distributions of retrieved variables,or by using multiple data sources. On the other hand,the schemes of multi-step parameter optimization should be customized for each variable according to its sensitivity and spatial heterogeneity. Not to fix sensitive or spatially heterogeneous variables if the accuracy and reliability of prior knowledge or the preliminary inversions could not be guaranteed.
remote sensing;models;crops;MSDT(multi-stage,sample-direction dependent,target-decisions);object-based;multi-stage inversion;crop variables
10.11975/j.issn.1002-6819.2017.01.026
S126;TP79
A
1002-6819(2017)-01-0190-09
劉 軻,黃 平,任國業(yè),周清波,李源洪,王 思,董秀春. 基于冠層反射率模型的作物參數(shù)多階段反演方法研究進展[J]. 農(nóng)業(yè)工程學報,2017,33(1):190-198.
10.11975/j.issn.1002-6819.20 17.01.026 http://www.tcsae.org
Liu Ke,Huang Ping,Ren Guoye,Zhou Qingbo,Li Yuanhong,Wang Si,Dong Xiuchun.Review on multi-stage inversion techniques of canopy reflectance models for retrieving crop variables[J]. Transactions of the Chinese Society of Agricultural Engineering(Transactions of the CSAE),2017,33(1):190-198.(in Chinese with English abstract)doi:10.11975/j.issn.1002-6819.2017.01.026 http://www.tcsae.org
2016-05-17
2016-10-21
四川省財政創(chuàng)新能力提升工程青年基金(2015QNJJ-023);四川省財政創(chuàng)新能力提升工程新興學科專項(2013XXXK-024);四川省財政創(chuàng)新能力工程高新領域擴展專項基金(2016GXTZ-011)
劉 軻,男,四川攀枝花人,助理研究員,主要從事農(nóng)作物參數(shù)遙感反演方法研究,高光譜與無人機遙感影像應用研究。成都 四川省農(nóng)業(yè)科學院遙感應用研究所,610066;Email:billc_st@163.com
※通信作者:黃 平,男,四川通江人,高級會計師。主要從事農(nóng)業(yè)經(jīng)濟研究、智慧農(nóng)業(yè)系統(tǒng)技術集成研究。成都 四川省農(nóng)業(yè)科學院遙感應用研究所,610066;Email:546991325@qq.com