曹中盛,李艷大,葉 春,舒時(shí)富,孫濱峰,黃俊寶,吳羅發(fā)
基于高光譜的雙季稻分蘗數(shù)監(jiān)測模型
曹中盛,李艷大※,葉 春,舒時(shí)富,孫濱峰,黃俊寶,吳羅發(fā)
(江西省農(nóng)業(yè)科學(xué)院農(nóng)業(yè)工程研究所,江西省智能農(nóng)機(jī)裝備工程研究中心,江西省農(nóng)業(yè)信息化工程技術(shù)研究中心,南昌 330200)
旨在闡明雙季稻分蘗數(shù)與冠層反射高光譜間的定量關(guān)系,構(gòu)建基于高光譜的雙季稻分蘗數(shù)監(jiān)測模型?;诓煌?、晚稻品種和施氮水平的田間試驗(yàn),于關(guān)鍵生育期(分蘗期、拔節(jié)期和孕穗期)測定早、晚稻分蘗數(shù),同步使用FieldSpec HandHeld 2型高光譜儀采集早、晚稻冠層反射高光譜數(shù)據(jù),分別利用光譜指數(shù)法和連續(xù)小波變換構(gòu)建新型光譜指數(shù)和敏感小波特征對雙季稻分蘗數(shù)進(jìn)行監(jiān)測,建立雙季稻分蘗數(shù)光譜監(jiān)測模型,并用獨(dú)立試驗(yàn)數(shù)據(jù)進(jìn)行檢驗(yàn)。結(jié)果表明,新型光譜指數(shù)和敏感小波特征對雙季稻分蘗數(shù)的監(jiān)測效果優(yōu)于其他類型光譜參數(shù)(植被指數(shù)和“三邊”參數(shù)),其中位于紅邊區(qū)域的小波特征db7(9,735)監(jiān)測早稻分蘗數(shù)時(shí)表現(xiàn)最優(yōu),監(jiān)測模型2為0.754,模型檢驗(yàn)相對均方根誤差RRMSE為0.128;位于紅邊區(qū)域的小波特征mexh(6,714)監(jiān)測晚稻分蘗數(shù)時(shí)表現(xiàn)最優(yōu),監(jiān)測模型2為0.837,模型檢驗(yàn)RRMSE為0.112。研究結(jié)果可為雙季稻分蘗數(shù)快速無損監(jiān)測和群體質(zhì)量精確調(diào)控提供理論基礎(chǔ)與技術(shù)支持。
雙季稻;分蘗數(shù);高光譜;小波特征;模型
發(fā)展雙季稻生產(chǎn)對于保障國家糧食安全和社會穩(wěn)定具有十分重要的戰(zhàn)略意義。分蘗是水稻生長發(fā)育過程中形成的一種特殊分枝[1],其數(shù)量是表征水稻長勢和群體質(zhì)量優(yōu)劣的重要指標(biāo)[2-3]。因此,快速準(zhǔn)確獲取水稻分蘗動態(tài)對于科學(xué)、實(shí)時(shí)和精確調(diào)控肥水管理,提高群體質(zhì)量具有重要作用[4-6]。水稻分蘗數(shù)觀測的常規(guī)方法,主要依靠人工下田觀測[7],費(fèi)時(shí)耗工,勞動力成本高,不能滿足大范圍水稻分蘗數(shù)快速觀測的需要。因此,探索一種高效、準(zhǔn)確的水稻分蘗數(shù)獲取方法十分必要。近年來,具有快速、無損、準(zhǔn)確等特征的光譜遙感技術(shù)快速發(fā)展,已成為監(jiān)測作物實(shí)時(shí)長勢的一種重要手段。國內(nèi)外許多學(xué)者利用光譜遙感技術(shù)實(shí)現(xiàn)了對水稻葉綠素含量、氮素營養(yǎng)、葉面積指數(shù)和生物量等生長指標(biāo)的快速精確監(jiān)測[8-10]。同時(shí),也有許多學(xué)者利用光譜遙感技術(shù)對作物分蘗數(shù)開展了無損監(jiān)測研究。如張猛等[11]基于4波段作物光譜監(jiān)測儀構(gòu)建植被指數(shù)OSAVI(650,850)和EVI2(650,850)對冬小麥返青期和起身期的莖蘗數(shù)進(jìn)行有效反演;Scotford等[12]利用植被指數(shù)NDVI對冬小麥的莖蘗密度進(jìn)行估算;吳軍華等[13]利用GreenSeeker光譜儀獲取植被指數(shù)NDVI和RVI對冬小麥生長前期的分蘗動態(tài)進(jìn)行監(jiān)測進(jìn)而實(shí)現(xiàn)氮素營養(yǎng)診斷等。上述研究證明利用光譜遙感技術(shù)可快速監(jiān)測作物分蘗狀況,但是,前人研究所選擇的植被指數(shù)包含波段數(shù)較少,是否探測到反射光譜與分蘗數(shù)之間潛在的最大相關(guān)性尚未可知。此外,僅采用簡單植被指數(shù)監(jiān)測分蘗數(shù)容易受到生育期變化的影響[12],使得已建立的模型多為單生育期模型,適用于不同生育期分蘗數(shù)監(jiān)測的通用模型較少。高光譜數(shù)據(jù)具有光譜分辨率高、波段連續(xù)性強(qiáng)和包含信息量大等優(yōu)點(diǎn)[14-15],能通過光譜指數(shù)法和連續(xù)小波變換等數(shù)據(jù)處理方法對目標(biāo)物信息進(jìn)行準(zhǔn)確探測[16-20]。由于高光譜儀器價(jià)格昂貴,且前人研究主要集中于與作物氮素營養(yǎng)診斷相關(guān)的葉綠素含量、葉面積指數(shù)和生物量等生長指標(biāo)的監(jiān)測,而采用高光譜數(shù)據(jù)構(gòu)建敏感光譜指數(shù)和小波特征監(jiān)測雙季稻分蘗數(shù)的研究鮮有報(bào)道。為此,本研究基于不同早、晚稻品種和施氮水平的田間試驗(yàn),利用實(shí)測高光譜數(shù)據(jù)分別通過光譜指數(shù)法和連續(xù)小波變換提取對早、晚稻不同生育期分蘗數(shù)敏感的光譜指數(shù)和小波特征,并構(gòu)建光譜監(jiān)測模型,以期為雙季稻分蘗數(shù)的快速無損監(jiān)測、群體質(zhì)量的精確調(diào)控提供理論基礎(chǔ)與技術(shù)支持。
于2016年和2017年3月至11月在江西省南昌縣八一鄉(xiāng)(28°33'54'' N,115°57'3'' E)進(jìn)行不同早、晚稻品種和不同施氮水平的田間小區(qū)試驗(yàn)。試驗(yàn)點(diǎn)耕作層土壤含有機(jī)質(zhì)20.62 g/kg、全氮1.93 g/kg、速效磷107.91 mg/kg、速效鉀89.70 mg/kg。采用裂區(qū)設(shè)計(jì),主區(qū)為品種,副區(qū)為氮肥。早、晚稻均設(shè)2個(gè)品種和5個(gè)施氮水平,重復(fù)3次,株行距為14 cm×24 cm,每穴3苗,南北行向,小區(qū)間以埂相隔,埂上覆膜,獨(dú)立排灌,小區(qū)面積21.6 m2。供試早稻品種為中嘉早17(ZJZ17)和潭兩優(yōu)83(TLY83),5個(gè)施氮水平分別為純氮0、75、150、225和300 kg/hm2,3月26日播種,4月25日移栽,7月23日收獲;供試晚稻品種為天優(yōu)華占(TYHZ)和岳優(yōu)9113(YY9113),5個(gè)施氮水平分別為純氮0、90、180、270和360 kg/hm2,6月25日播種,7月26日移栽,11月1日收獲。早、晚稻氮肥用尿素,分3次施用(基肥40%,分蘗肥30%,穗肥30%);早稻配施P2O575 kg/hm2和K2O 90 kg/hm2,晚稻配施P2O560 kg/hm2和K2O 120 kg/hm2,磷肥用鈣鎂磷肥,鉀肥用氯化鉀,全部作基肥施用。其他栽培管理措施同當(dāng)?shù)馗弋a(chǎn)栽培。
1.2.1 冠層反射高光譜數(shù)據(jù)獲取
于早、晚稻分蘗期、拔節(jié)期和孕穗期釆用美國 Analytical Spectral Dvice 公司的 FieldSpec HandHeld 2 型便攜式高光譜儀(波長范圍325~1 075 nm,采樣間隔1.4 nm,分辨率3 nm)測定每個(gè)小區(qū)的冠層反射光譜,測試后光譜分辨率經(jīng)儀器自帶軟件重采樣為1 nm。選擇晴朗、無風(fēng)或微風(fēng)天氣的10:00~14:00進(jìn)行測定,測量時(shí)探頭垂直向下,距離冠層正上方1 m,視場角為25°,視場面積約0.15 m2。測量過程中,根據(jù)天氣變化及時(shí)進(jìn)行標(biāo)準(zhǔn)白板校正。每個(gè)小區(qū)測量3個(gè)點(diǎn),每點(diǎn)重復(fù)測量5次,對所采集的數(shù)據(jù)進(jìn)行差異顯著性分析后,取平均值作為該小區(qū)的測量值。
1.2.2 分蘗數(shù)觀測
與冠層反射光譜觀測同步,通過人工計(jì)數(shù)觀測每穴水稻的分蘗數(shù)(tiller number,TN),選擇每個(gè)小區(qū)進(jìn)行光譜獲取的3個(gè)觀測點(diǎn),每點(diǎn)選擇20穴植株進(jìn)行觀測,對所采集的數(shù)據(jù)進(jìn)行差異顯著性分析后,取平均值作為該小區(qū)TN觀測值。
光譜指數(shù)法和連續(xù)小波變換是目前處理高光譜數(shù)據(jù)的2種常用方法。前者通過比較高光譜數(shù)據(jù)中不同波段組合與目標(biāo)物之間的相關(guān)性提取最優(yōu)光譜指數(shù),后者利用不同縮放尺度和平移的小波函數(shù)對高光譜數(shù)據(jù)進(jìn)行變換篩選敏感小波特征。本研究采用這2種方法提取最優(yōu)光譜指數(shù)和敏感小波特征來監(jiān)測雙季稻TN,所有參數(shù)構(gòu)建和數(shù)據(jù)運(yùn)算均采用Matlab 2014a軟件自編程進(jìn)行。
1.3.1 光譜指數(shù)法
最優(yōu)光譜指數(shù)基于實(shí)測高光譜數(shù)據(jù)進(jìn)行提取,提取時(shí)以簡單植被指數(shù)為基礎(chǔ)形式比較所有波段組合與TN之間的相關(guān)性。本研究選擇的2種光譜指數(shù)分別為歸一化光譜指數(shù)(normalized spectral index,NDSI)和比值光譜指數(shù)(ratio spectral index,RSI)。NDSI和RSI篩選時(shí),分別以式(1)和式(2)進(jìn)行波段兩兩組合,擬合所有波段組合與TN之間的線性相關(guān)關(guān)系,并繪制擬合關(guān)系決定系數(shù)(coefficient of determination,2)二維分布圖,以2前2%的波段組合分布區(qū)域?yàn)槊舾袇^(qū)域,確定每個(gè)敏感區(qū)域內(nèi)具有最大2的波段組合為最優(yōu)光譜指數(shù)。
NDSI和RSI光譜指數(shù)計(jì)算方程如下
式中()和()為NDSI和RSI中包含的不確定波段,為波段()的反射率。
1.3.2 連續(xù)小波變換
連續(xù)小波變換通過平移和縮放的母小波函數(shù)與反射光譜進(jìn)行卷積運(yùn)算,得到不同縮放尺度()和平移量()的小波特征W(,)。本文通過和兩兩組合獲得不同母小波函數(shù)ψ,(),利用不同ψ,()對反射光譜進(jìn)行變換獲得不同小波特征W(,),然后擬合不同W(,)與TN之間的線性相關(guān)關(guān)系,最后基于擬合方程的2提取敏感小波特征。本研究選擇db7和mexh兩種母小波函數(shù)對光譜進(jìn)行變換,為減少數(shù)據(jù)運(yùn)算量,采樣步長為2[19-20]。
式中()為母小波函數(shù),ψ,()為經(jīng)過平移和縮放后的母小波函數(shù),為縮放尺度,為平移量(亦為波長),W(,)為小波特征。
利用2017年試驗(yàn)數(shù)據(jù)提取最優(yōu)光譜指數(shù)和敏感小波特征并構(gòu)建監(jiān)測模型,利用2016年試驗(yàn)數(shù)據(jù)對其進(jìn)行檢驗(yàn)。通過計(jì)算預(yù)測值和實(shí)測值之間的相對均方根誤差(relative root mean square difference,RRMSE)檢驗(yàn)光譜參數(shù)及監(jiān)測模型的精準(zhǔn)度[19];計(jì)算噪聲指數(shù)(noise equivalent,NE)評估光譜參數(shù)在不同TN下的敏感性[21],NE值越低,表明光譜參數(shù)在對應(yīng)TN值下越敏感。同時(shí),采用前人構(gòu)建的與作物長勢密切相關(guān)的植被指數(shù)和“三邊”參數(shù)進(jìn)行比較分析。
式中為樣本數(shù)量,P為預(yù)測分蘗數(shù)(tiller number per hill,TN),O為實(shí)測TN,d(SP)/d(TN)為光譜參數(shù)(spectral parameter,SP)與TN之間最佳擬合方程的一階導(dǎo)數(shù)。
冠層光譜反射率與分蘗數(shù)之間的相關(guān)性在早、晚稻中表現(xiàn)基本相同(圖1)。但不同波段范圍的相關(guān)性差異較大,可見光區(qū)域內(nèi)反射率與TN呈負(fù)相關(guān),近紅外區(qū)域內(nèi)反射率與TN呈正相關(guān)。
圖1 早、晚稻冠層反射光譜與分蘗數(shù)之間的相關(guān)性
早、晚稻NDSI和RSI的敏感區(qū)域均集中在近紅外和紅邊波段組合區(qū)域。其中,早稻NDSI的敏感區(qū)域?yàn)椋海?): 835~1 075 nm,(): 700~735 nm)(圖2a),RSI的敏感區(qū)域?yàn)椋海╥)((): 700~735 nm,(): 830~1 075 nm)和(ii)((): 830~1 075 nm,(): 700~740 nm)(圖 2b);晚稻NDSI的敏感區(qū)域?yàn)椋?): 745~910 nm,(): 720~765 nm)(圖2c),RSI的敏感區(qū)域?yàn)椋╥)((): 715~760 nm,(): 740~920 nm)和(ii)((): 745~900 nm,(): 725~765 nm)(圖2d)。
與早稻TN相關(guān)性較高的3個(gè)光譜指數(shù)分別為NDSI(975,714)、RSI(971,718)和RSI(720,985),其中,NDSI(975,714)與早稻TN之間線性回歸方程的2最大,其值為0.724,確定為監(jiān)測早稻TN的最優(yōu)光譜指數(shù)(表1)。與晚稻TN相關(guān)性較高的3個(gè)光譜指數(shù)分別為NDSI(800,738)、RSI(736,798)和RSI(788,738),其中,RSI(788,738)與晚稻TN之間線性回歸方程的2最大,其值為0.792,確定為監(jiān)測晚稻TN的最優(yōu)光譜指數(shù)(表1)。
圖2 歸一化光譜指數(shù)和比值光譜指數(shù)兩波段組合(λ(x)和λ(y))與分蘗數(shù)的線性回歸關(guān)系決定系數(shù)等勢圖
表1 基于光譜指數(shù)法構(gòu)建的最優(yōu)光譜指數(shù)
注:為光譜指數(shù),為分蘗數(shù)。
Note:is the spectral index,is the tiller number.
圖3為早、晚稻光譜經(jīng)連續(xù)小波變換后的敏感小波特征分布(彩色部分代表2前2%),早稻光譜經(jīng)db7母小波函數(shù)變換后,敏感小波特征主要集中在可見光和紅邊區(qū)域,敏感區(qū)域內(nèi)2較高的小波特征為db7(9,395)和db7(9,735)(圖3a);經(jīng)mexh母小波函數(shù)變換后,敏感區(qū)域主要集中在紅光到近紅外之間區(qū)域,敏感區(qū)域內(nèi)2較高的小波特征為mexh(6,714) 和mexh(7,709)(圖3b)。晚稻反射光譜經(jīng)db7母小波函數(shù)變換后,敏感小波特征在可見光、紅邊和近紅外區(qū)域均有分布,敏感區(qū)域內(nèi)2較高的小波特征為db7(6,844)、db7(7,693)和db7(8,720)(圖 3c);經(jīng)mexh母小波函數(shù)變換后,敏感區(qū)域主要集中在紅邊和近紅外區(qū)域,敏感區(qū)域內(nèi)2較高的小波特征為mexh(4,794)、mexh(5,706)和mexh(6,714)(圖3d)。綜合比較上述小波特征與早、晚稻分蘗數(shù)之間線性擬合方程的2,最終確定db7(9,735)和mexh(7,709)為早稻分蘗數(shù)的敏感小波特征,其2分別為0.754和0.757(表2);db7(8,720)和mexh(6,714)為晚稻分蘗數(shù)的敏感小波特征,其2分別為0.836和0.837(表2)。
圖3 不同小波特征與分蘗數(shù)之間線性回歸方程決定系數(shù)等勢圖
表2 敏感小波特征與分蘗數(shù)之間的相關(guān)關(guān)系
注:為小波特征,為分蘗數(shù)。
Note:is the wavelet feature,is the tiller number.
表3中,監(jiān)測早稻TN時(shí),植被指數(shù)中的歸一化紅邊指數(shù)(NDRE)表現(xiàn)最優(yōu),其建模2為0.683,檢驗(yàn)RRMSE為0.150;“三邊”參數(shù)中的紅邊面積(SD)與TN之間的相關(guān)性較高,其建模2為0.656,檢驗(yàn)RRMSE為0.153;最優(yōu)光譜指數(shù)NDSI (975,714)監(jiān)測早稻TN的效果較NDRE和SD明顯提高,其建模2為0.724,檢驗(yàn)RRMSE為0.151;小波特征在監(jiān)測早稻TN時(shí)表現(xiàn)較好,其中,以db7 (9,735)表現(xiàn)最優(yōu)(2= 0.754,RRMSE = 0.128)。監(jiān)測晚稻TN時(shí),植被指數(shù)中的NDRE表現(xiàn)最優(yōu),其建模2為0.718,檢驗(yàn)RRMSE為0.181;“三邊”參數(shù)中的紅邊振幅(D)表現(xiàn)較優(yōu),其建模2為0.586,檢驗(yàn)RRMSE為0.212;新建比值光譜指數(shù)RSI(788,738)效果較優(yōu),其建模2為0.792,檢驗(yàn)RRMSE為0.142;利用小波特征監(jiān)測晚稻TN時(shí),敏感小波特征mexh (6,714)的表現(xiàn)進(jìn)一步提高,其建模2為0.838,檢驗(yàn)RRMSE為0.112。圖4和圖5為幾個(gè)表現(xiàn)較優(yōu)的光譜參數(shù)監(jiān)測早、晚稻TN時(shí)的表現(xiàn)。
表3 基于不同光譜參數(shù)的早、晚稻分蘗數(shù)監(jiān)測模型構(gòu)建與檢驗(yàn)
圖4 光譜參數(shù)與分蘗數(shù)在建模數(shù)據(jù)集中的相關(guān)性
圖5 基于不同光譜參數(shù)的早、晚稻分蘗數(shù)監(jiān)測模型預(yù)測值與實(shí)測值1∶1關(guān)系圖
光譜參數(shù)在不同TN下的敏感性也是評估其優(yōu)劣的重要指標(biāo)。不同光譜參數(shù)監(jiān)測早稻TN時(shí)NE值相對較低且差異不明顯(圖6),表明利用光譜參數(shù)監(jiān)測早稻TN時(shí)受飽和效應(yīng)影響較小。光譜參數(shù)NE在晚稻中的變化趨勢較在早稻中的變化趨勢差異表現(xiàn)明顯,幾個(gè)精準(zhǔn)度較高的光譜參數(shù)中,D的NE值較其他光譜參數(shù)高,監(jiān)測晚稻TN時(shí)易出現(xiàn)飽和現(xiàn)象。而NDRE、RSI(788,738)、db7(8,720)和mexh(6,714)的NE值較低,能有效緩解飽和。綜上所述,本研究提取到的最優(yōu)光譜指數(shù)和敏感小波特征在監(jiān)測早、晚稻分蘗數(shù)時(shí)敏感性較強(qiáng)。其中,又以敏感小波特征db7(9,735)和mexh(6,714)表現(xiàn)最優(yōu)。
江西、湖南等地是中國重要的雙季稻主產(chǎn)省份,發(fā)展雙季稻生產(chǎn)對保障中國糧食安全與社會穩(wěn)定意義重大。分蘗數(shù)是表征雙季稻長勢和群體質(zhì)量優(yōu)劣的重要指標(biāo),快速準(zhǔn)確獲取雙季稻分蘗動態(tài)對實(shí)現(xiàn)肥水的精確管理尤其重要。
生育進(jìn)程和氮肥供應(yīng)是造成早、晚稻分蘗數(shù)發(fā)生改變的主要原因。水稻生育前中期的分蘗發(fā)生與葉片生長呈正相關(guān)[22-23],水稻植株氮素營養(yǎng)也對分蘗發(fā)生起促進(jìn)作用[22],而葉綠素含量是指示植株氮素營養(yǎng)狀況的重要參數(shù)[24]。因此,該階段分蘗數(shù)的變化與水稻群體葉片面積、覆蓋度和葉片葉綠素含量的變化密切相關(guān)。本研究通過分析冠層反射光譜與早、晚稻分蘗數(shù)之間的相關(guān)關(guān)系,發(fā)現(xiàn)可見光波段范圍內(nèi),早、晚稻冠層反射率與分蘗數(shù)之間呈負(fù)相關(guān),這與前人研究發(fā)現(xiàn)的葉綠素含量與可見光光譜反射率的相關(guān)性相一致[25]。表明分蘗數(shù)變化引起可見光反射光譜的改變與色素含量的變化有間接關(guān)聯(lián)。在近紅外波段范圍內(nèi),早、晚稻冠層光譜反射率與分蘗數(shù)之間呈正相關(guān),這與前人研究葉面積指數(shù)與冠層反射光譜的變化結(jié)論相一致[26]。主要原因在于植被對近紅外光線具有較強(qiáng)的反射能力[27],分蘗數(shù)增加導(dǎo)致葉面積增大引起了近紅外冠層反射率上升。因此,在雙季稻生長前期,分蘗數(shù)和葉綠素含量變化因氮素作用表現(xiàn)出一致性,使得分蘗數(shù)變化能夠引起可見光反射率發(fā)生改變。另外,分蘗數(shù)增加直接導(dǎo)致雙季稻葉面積指數(shù)增大、覆蓋度升高,使得冠層結(jié)構(gòu)復(fù)雜度增大,最終引起近紅外反射率的升高。
植被指數(shù)是監(jiān)測作物生長指標(biāo)最常用的一類光譜參數(shù),具有波段數(shù)量少、構(gòu)型簡單、計(jì)算方便等優(yōu)點(diǎn)。本研究通過實(shí)測高光譜數(shù)據(jù)篩選出的光譜指數(shù)NDSI(975,714)和RSI(788,738)在監(jiān)測早、晚稻分蘗數(shù)時(shí)具有較強(qiáng)的準(zhǔn)確性和敏感性。分析2個(gè)光譜指數(shù)的波段構(gòu)成,發(fā)現(xiàn)其均包含有紅邊波段。前人研究表明,紅邊波段不僅能反映作物葉片的色素含量,同時(shí)也與作物群體內(nèi)部結(jié)構(gòu)狀況存在相關(guān)性[28-29]。因此,包含有紅邊波段的光譜參數(shù)能對其進(jìn)行精確反演。此外,紅邊波段較可見光波段具有更強(qiáng)的抗飽和能力[30],進(jìn)一步提高了高分蘗數(shù)下光譜指數(shù)的監(jiān)測能力。
作物冠層高光譜反射光譜通常包含有豐富的光譜信息,但不同信息之間相互干擾則會影響光譜監(jiān)測的精度。連續(xù)小波變換是近年來興起的一種處理高光譜數(shù)據(jù)的有效方法,與直接利用反射光譜監(jiān)測生長指標(biāo)相比,連續(xù)小波變換能對反射光譜按照不同尺度進(jìn)行分解,進(jìn)而通過提取吸收特征獲取目標(biāo)物信息[17,31]。本研究篩選到的兩個(gè)敏感小波特征db7(9,735)和mexh(6,714)的中心波段均位于紅邊區(qū)域,該區(qū)域波段與作物冠層結(jié)構(gòu)和葉綠素含量均具相關(guān)性,因此可對分蘗數(shù)進(jìn)行監(jiān)測,且能緩解飽和影響。此外,連續(xù)小波變換的尺度選擇也是提高其估算精度的重要原因。本研究中,可用于分蘗數(shù)監(jiān)測的敏感小波特征尺度較高(29和26)。在使用較高尺度母小波函數(shù)對光譜進(jìn)行變換時(shí),使用的光譜信息量較光譜指數(shù)增大,提高了分蘗數(shù)監(jiān)測的精度,這與前人研究生物量估算時(shí)的表現(xiàn)相一致[20]。
當(dāng)然,本研究基于有限的試驗(yàn)資料對模型進(jìn)行了初步檢驗(yàn),且不同早、晚稻品種的分蘗特性的不一致可能會導(dǎo)致模型的實(shí)用性不廣泛。因此,今后需要在江西、湖南等雙季稻主產(chǎn)區(qū)采用多年多點(diǎn)試驗(yàn)資料對模型進(jìn)行測驗(yàn)和完善,并對不同早、晚品種的分蘗特性進(jìn)行深入研究。
本研究基于實(shí)測高光譜數(shù)據(jù),利用光譜指數(shù)法和連續(xù)小波變換篩選最優(yōu)光譜指數(shù)和敏感小波特征,構(gòu)建早、晚稻分蘗數(shù)光譜監(jiān)測模型。光譜反射率與早、晚稻分蘗數(shù)之間的相關(guān)性在可見光部分為負(fù)相關(guān),在近紅外區(qū)域?yàn)檎嚓P(guān)。篩選出來的最優(yōu)光譜指數(shù)和敏感小波特征對分蘗數(shù)的監(jiān)測精度高,較前人開發(fā)出來的植被指數(shù)和“三邊”參數(shù)監(jiān)測效果好。其中,小波特征db7(9,735)監(jiān)測早稻分蘗數(shù)時(shí)表現(xiàn)最優(yōu),監(jiān)測模型為TNearly=3.632×db7(9,735)+7.318,建模2為0.754,模型檢驗(yàn)RRMSE為0.128;小波特征mexh(6,714)監(jiān)測晚稻分蘗數(shù)時(shí)表現(xiàn)最優(yōu),監(jiān)測模型為TNlate=-15.351×mexh(6,714)+8.173,建模2為0.837,模型檢驗(yàn)RRMSE為0.112。
[1]Li Xueyong, Qian Qian, Fu Zhiming, et al. Control of tillering in rice[J]. Nature, 2003, 422(6932): 618-621.
[2]Gurdev K. Green revolution: Preparing for the 21stcentury[J]. Genome, 1999, 42(4): 646-655.
[3]Conway G, Gary T. Feeding the world in the twenty-first century[J]. Nature, 1999, 402(Suppl 6761): 55-58.
[4]汪洋. 氮素營養(yǎng)對水稻分蘗的產(chǎn)量異質(zhì)性影響及調(diào)控[D]. 武漢:華中農(nóng)業(yè)大學(xué),2017. Wang Yang. Heterogeneity and Regulations of Rice Tillers Yield by Nitrogen Nutrition[D]. Wuhan: Huazhong Agricultural University, 2017. (in Chinese with English abstract)
[5]Huang Min, Yang Cailing, Ji Qiumei, et al. Tillering responses of rice to plant density and nitrogen rate in a subtropical environment of southern China[J]. Field Crops Research, 2013, 149: 187-192.
[6]田廣麗,周毅,孫博,等. 氮素及栽培密度影響水稻分蘗動態(tài)的機(jī)制[J]. 植物營養(yǎng)與肥料學(xué)報(bào),2018,24(4):896-904. Tian Guangli, Zhou Yi, Sun Bo, et al. Effects of nitrogen and transplanting density on the mechanisms of tillering dynamic of rice[J]. Journal of Plant Nutrition and Fertilizers, 2018, 24(4): 896-904. (in Chinese with English abstract)
[7]魏廣彬. 基于同伸關(guān)系的水稻群體莖蘗和葉面積動態(tài)模擬模型[D]. 南京:南京農(nóng)業(yè)大學(xué),2011. Wei Guangbin. Simulation of Dynamics of Tillering and Leaf Area of Rice (L.) Population Based on Synchornouly Emerging Characteristics[D]. Nanjing: Nanjing Agricultural University, 2011. (in Chinese with English abstract)
[8]Wang Wei, Yao Xia, Yao Xinfeng, et al. Estimating leaf nitrogen concentration with three-band vegetation indices in rice and wheat[J]. Field Crops Research, 2012, 129: 90-98
[9]Xu Xiaoqing, Lu Jingshan, Zhang Ning, et al. Inversion of rice canopy chlorophyll content and leaf area index based on coupling of radiative transfer and Bayesian network models[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2019, 150: 185-196
[10]王秀珍,黃敬峰,李云梅,等. 水稻地上鮮生物量的高光譜遙感估算模型研究[J]. 作物學(xué)報(bào),2003,29(6):815-821. Wang Xiuzhen, Huang Jingfeng, Li Yunmei, et al. Study on hyperspectral remote sensing estimation models for the ground fresh biomass of rice[J]. Acta Agronomica Sinica, 2003, 29(6): 815-821. (in Chinese with English abstract)
[11]張猛,孫紅,李民贊,等. 基于4波段作物光譜測量儀的小麥分蘗數(shù)預(yù)測[J]. 農(nóng)業(yè)機(jī)械學(xué)報(bào),2016,47(9):341-347. Zhang Meng, Sun Hong, Li Minzan, et al. Prediction of winter wheat tiller number based on 4-waveband crop monitor with spectral reflectance[J]. Transactions of The Chinese Society of Agricultural Machinery, 2016, 47(9): 341-347. (in Chinese with English abstract)
[12]Scotford I M, Miller P. Estimating tiller density and leaf area index of winter wheat using spectral reflectance and ultrasonic sensing techniques[J]. Biosystems Engineering, 2004, 89(4): 395-408.
[13]吳軍華,岳善超,侯鵬,等. 基于主動遙感的冬小麥群體動態(tài)監(jiān)測[J]. 光譜學(xué)與光譜分析,2011,31(2):535-538. Wu Junhua, Yue Shanchao, Hou Peng, et al. Monitoring winter wheat population dynamics using an active crop sensor[J]. Spectroscopy and Spectral Analysis, 2011, 31(2): 535-538. (in Chinese with English abstract)
[14]Dechant B, Ryu Y, Kang M. Making full use of hyperspectral data for gross primary productivity estimation with multivariate regression: Mechanistic insights from observations and process-based simulations[J]. Remote Sensing of Environment, 2019, 234(1): 111, 435.
[15]田永超. 基于高光譜遙感的水稻氮素營養(yǎng)參數(shù)監(jiān)測研究[D]. 南京:南京農(nóng)業(yè)大學(xué),2008. Tian Yongchao. Monitoring Nitrogen Nutrition Parameters with Hyperspectral Remote Sensing in Rice[D]. Nanjing: Nanjing Agricultural University, 2008. (in Chinese with English abstract)
[16]Fuan T, William P. Derivative analysis of hyperspectral data[J]. Remote Sensing of Environment, 1998, 66(1): 41-51.
[17]李旭青,李龍,莊連英,等. 基于小波變換和BP神經(jīng)網(wǎng)絡(luò)的水稻冠層重金屬含量反演[J]. 農(nóng)業(yè)機(jī)械學(xué)報(bào),2019,50(6):226-232. Li Xuqing, Li Long, Zhuang Lianying, et al. Inversion of heavy metal content in rice canopy based on wavelet transform and BP neural network[J]. Transactions of The Chinese Society of Agricultural Machinery, 2019, 50(6): 226-232. (in Chinese with English abstract)
[18]Inoue Y, Sakaiya E, Zhu Y, et al. Diagnostic mapping of canopy nitrogen content in rice based on hyperspectral measurements[J]. Remote Sensing of Environment, 2012, 126: 210-221.
[19]Yao Xia, Si Haiyang, Cheng Tao, et al. Hyperspectral estimation of canopy leaf biomass phenotype per ground area using a continuous wavelet analysis in wheat[J]. Frontiers in Plant Science, 2018, 9, 1360-1360.
[20]Cheng Tao, Rivard B, Sánchez-Azofeifa A G, et al. Deriving leaf mass per area (LMA) from foliar reflectance across a variety of plant species using continuous wavelet analysis[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2014, 87: 28-38.
[21]Gitelson A A. Remote estimation of crop fractional vegetation cover: The use of noise equivalent as an indicator of performance of vegetation indices[J]. International Journal of Remote Sensing, 2013, 34(17): 6054-6066.
[22]孟亞利,曹衛(wèi)星,柳新偉,等. 水稻莖蘗動態(tài)的模擬研究[J].南京農(nóng)業(yè)大學(xué)學(xué)報(bào),2003,26(2):1-6. Meng Yali, Cao Weixing, Liu Xinwei, et al. Simulation on the tiller dynamic in rice[J]. Journal of Nanjing Agricultural University, 2003, 26(2): 1-6. (in Chinese with English abstract)
[23]Miller B, Foin T, Hill J. CARICE: A rice model for scheduling and evaluating management actions[J]. Agronomy Journal, 1993, 85(4): 938-947.
[24]楊峰,范亞民,李建龍,等. 高光譜數(shù)據(jù)估測稻麥葉面積指數(shù)和葉綠素密度[J]. 農(nóng)業(yè)工程學(xué)報(bào),2010,26(2):237-243. Yang Feng, Fan Yamin, Li Jianlong, et al. Estimating LAI and CCD of rice and wheat using hyperspectral remote sensing data[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2010, 26(2): 237-243. (in Chinese with English abstract)
[25]楊杰,田永超,姚霞,等. 水稻上部葉片葉綠素含量的高光譜估算模型[J]. 生態(tài)學(xué)報(bào),2009,29(12):6561-6571. Yang Jie, Tian Yongchao, Yao Xia, et al. Hyperspectral estimation model for chlorophyll concentrations in top leaves of rice[J]. Acta Ecologica Sinica, 2009, 29(12): 6561-6571. (in Chinese with English abstract)
[26]王秀珍,黃敬峰,李云梅,等. 水稻葉面積指數(shù)的高光譜遙感估算模型[J]. 遙感學(xué)報(bào),2004,8(1):81-88. Wang Xiuzhen, Huang Jingfeng, Li Yunmei, et al. The study on hyperspectral remote sensing estimation models about LAI of rice[J]. Journal of Remote Sensing, 2004, 8(1): 188-191. (in Chinese with English abstract)
[27]Darvishzadeh R, Skidmore A, Atzberger C, et al. Estimation of vegetation LAI from hyperspectral reflectance data: Effects of soil type and plant architecture[J]. International Journal of Applied Earth Observation and Geoinformation, 2008, 10(3): 358-373.
[28]Zarco-Tejada P J, Hornero A, Hernández-Clemente R, et al. Understanding the temporal dimension of the red-edge spectral region for forest decline detection using high-resolution hyperspectral and Sentinel-2A imagery[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2018, 137: 134-148.
[29]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.
[30]Cao Zhongsheng, Cheng Tao, Ma Xue, et al. A new three-band spectral index for mitigating the saturation in the estimation of leaf area index in wheat[J]. International Journal of Remote Sensing, 2017, 38(23): 3865-3885.
[31]Liu Meiling, Liu Xiangnan, Ding Weicui, et al. Monitoring stress levels on rice with heavy metal pollution from hyperspectral reflectance data using wavelet-fractal analysis[J]. International Journal of Applied Earth Observation and Geoinformation, 2011, 13(2): 246-255.
Model for monitoring tiller number of double cropping rice based on hyperspectral reflectance
Cao Zhongsheng, Li Yanda※, Ye Chun, Shu Shifu, Sun Binfeng, Huang Junbao, Wu Luofa
(,,,,330200,)
The fast, real-time and non-destructive monitoring of double-cropping rice tiller number has important practical significance for growth diagnosis and yield prediction. Hyperspectral sensing has been proved effective to estimate the rice growth parameters, such as the chlorophyll content, leaf area index and biomass, yet few investigations pay attention to the tiller number. The objective of this study was to establish a regulation model for estimating double-cropping rice tiller number based on the hyperspectral reflectance across a wide range of growth stages (tillering stage, jointing stage, and booting stage). In the presented study, the tiller number and hyperspectral reflectance data were firstly obtained from two double-cropping rice field experiments, which encompassed variations in two years, four cultivars and five nitrogen application rates. Then the sensitive spectral indices and wavelet features were extracted from the hyperspectral reflectance data through spectral indices approach and continuous wavelet analysis, respectively. Finally, the regression models for tiller number estimation based on sensitive spectral indices and wavelet features were developed and validated using independent field experiment datasets. The results suggested that the newly developed spectral indices and sensitive wavelet features with red-edge bands performed better than the published vegetation indices and ‘three edge’ parameters. The normalized different spectral index named NDSI (975,714) was strongly related to the early rice tiller number. It had a determination coefficient (2) of 0.724 in calibration and relative root mean square error (RRMSE) of 0.151 in validation. The ratio spectral index RSI (788,738) strongly related to the late rice tiller number with2of 0.792 and RRMSE of 0.142 in calibration and validation, respectively. Compared with the published vegetation indices, ‘three edge’ parameters and newly developed spectral indices, the sensitive wavelet features observed in the red-edge region with high scales (29and 26) performed best in the double-cropping rice tiller number estimation. The wavelet feature named db7 (9,735) was strongest related to the early rice tiller number. It had2of 0.754 in calibration and RRMSE of 0.128 in validation. The wavelet feature named mexh (6,714) was strongest related to the late rice tiller number. It had2of 0.837 in calibration and RRMSE of 0.112 in validation. Additionally, the sensitive spectral indices and wavelet features also could reduce the saturation effect with low noise equivalent (NE). It meant that in the condition the optical sensors equip few bands, the spectral indices NDSI (975,714) and RSI (788,738) could be used to monitor the early rice and late rice tiller number. Furthermore, the wavelet features db7 (9,735) and (6,714) could improve the accuracy for monitoring double-cropping rice tiller number based on the hyperspectral reflectance data with monitoring models of TNearly=3.632×db7 (9,735)+7.318 and TNlate=-15.351×mexh (6,714)+8.173, respectively.
double-cropping rice; tiller number; hyperspectral; wavelet feature; model
曹中盛,李艷大,葉 春,舒時(shí)富,孫濱峰,黃俊寶,吳羅發(fā). 基于高光譜的雙季稻分蘗數(shù)監(jiān)測模型[J]. 農(nóng)業(yè)工程學(xué)報(bào),2020,36(4):185-192. doi:10.11975/j.issn.1002-6819.2020.04.022 http://www.tcsae.org
Cao Zhongsheng, Li Yanda, Ye Chun, Shu Shifu, Sun Binfeng, Huang Junbao, Wu Luofa. Model for monitoring tiller number of double cropping rice based on hyperspectral reflectance[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2020, 36(4): 185-192. (in Chinese with English abstract) doi:10.11975/j.issn.1002-6819.2020.04.022 http://www.tcsae.org
2019-12-11
2020-01-19
國家重點(diǎn)研發(fā)計(jì)劃項(xiàng)目(2016YFD0300608);國家青年拔尖人才支持計(jì)劃項(xiàng)目;江西省科技計(jì)劃項(xiàng)目(20182BCB22015,20161BBI90012,20192ACB80005,20192BBF60050);江西省“雙千計(jì)劃”項(xiàng)目和江西省農(nóng)業(yè)科學(xué)院創(chuàng)新基金博士啟動項(xiàng)目(20182CBS001)聯(lián)合資助
曹中盛,助理研究員,博士,主要從事農(nóng)業(yè)信息技術(shù)研究。Email:czsheng2015@outlook.com
李艷大,研究員,博士,主要從事信息農(nóng)學(xué)與農(nóng)機(jī)化技術(shù)研究。Email:liyanda2008@126.com
10.11975/j.issn.1002-6819.2020.04.022
S31
A
1002-6819(2020)-04-0185-08