基于PM2.5站點(diǎn)監(jiān)測數(shù)據(jù)的京津冀AOD補(bǔ)值研究

宋春杰,魏 強(qiáng),范麗行,王 衛(wèi)*,韓 芳,李偉妙,李夫星,成賀璽

基于PM2.5站點(diǎn)監(jiān)測數(shù)據(jù)的京津冀AOD補(bǔ)值研究

宋春杰1,2,魏 強(qiáng)1,2,范麗行1,2,王 衛(wèi)1,2*,韓 芳1,2,李偉妙1,2,李夫星1,3**,成賀璽4

(1.河北師范大學(xué)地理科學(xué)學(xué)院,河北 石家莊 050024；2.河北省環(huán)境演變與生態(tài)建設(shè)實(shí)驗(yàn)室,河北 石家莊 050024；3.河北省環(huán)境變化遙感技術(shù)識別創(chuàng)新中心,河北 石家莊 050024；4.邯鄲市城鄉(xiāng)規(guī)劃編制研究中心,河北 邯鄲 056000)

以京津冀2020年318個(gè)地面監(jiān)測站點(diǎn)的PM2.5數(shù)據(jù)為估算因子,構(gòu)建了時(shí)空線性混合效應(yīng)模型(STLME)和時(shí)空嵌套線性混合效應(yīng)模型(STNLME),為AOD數(shù)據(jù)的補(bǔ)值研究提供了一種新方法.結(jié)果表明:在有AOD -PM2.5匹配數(shù)據(jù)的日期,上述兩個(gè)模型估算精度相近,交叉驗(yàn)證后決定系數(shù)R分別為0.868和0.874,均方根誤差RMSE分別為0.112和0.109;在無AOD-PM2.5匹配數(shù)據(jù)的日期,嵌套模型估算精度明顯高于非嵌套模型,交叉驗(yàn)證后決定系數(shù)2分別為0.63和0.26.經(jīng)過模型補(bǔ)值后,研究區(qū)監(jiān)測站點(diǎn)所在網(wǎng)格AOD數(shù)據(jù)空間維有效比率從原始數(shù)據(jù)的44.35%提高到99.35%,時(shí)間維有效比率從87.94%提高到100%;同時(shí),每個(gè)站點(diǎn)的年均AOD值都有明顯提高,彌補(bǔ)了高PM2.5濃度條件下缺失的AOD數(shù)據(jù),可以減少空氣污染和健康研究中暴露評估的偏差.

MAIAC AOD；監(jiān)測站點(diǎn)AOD補(bǔ)值；時(shí)空混合效應(yīng)模型；時(shí)空嵌套混合效應(yīng)模型；京津冀

PM2.5等氣溶膠顆粒已成為以京津冀地區(qū)為代表的華北平原的主要環(huán)境空氣污染物.如何準(zhǔn)確估算PM2.5的長期和短期暴露量,是當(dāng)前迫切需要解決的問題[1-5].我國的PM2.5監(jiān)測網(wǎng)自2013年開始運(yùn)行,但這些地面監(jiān)測站仍然受到布局稀疏和分布不均的限制[6].中分辨率成像光譜儀(MODIS)等衛(wèi)星儀器觀測到的氣溶膠光學(xué)厚度(AOD)數(shù)據(jù)由于其廣泛的空間覆蓋范圍和對地球表面及大氣的重復(fù)觀測而被廣泛用于估算地面PM2.5濃度[7-10].但是,由于云/雪/水體的覆蓋、高地面反射率和極高的氣溶膠負(fù)荷,衛(wèi)星數(shù)據(jù)受到非隨機(jī)缺失的挑戰(zhàn)[11-13].研究表明,AOD數(shù)據(jù)的非隨機(jī)性缺失可能會(huì)導(dǎo)致PM2.5暴露評估的偏差.如果由于PM2.5濃度過高而導(dǎo)致AOD數(shù)據(jù)丟失,將會(huì)導(dǎo)致PM2.5平均濃度的低估[14-15].因此,有必要探索相應(yīng)的方法來填補(bǔ)AOD數(shù)據(jù)的缺失,以提高PM2.5濃度預(yù)測的覆蓋率和準(zhǔn)確性.

國內(nèi)外學(xué)術(shù)界對缺失AOD數(shù)據(jù)填補(bǔ)問題開展了廣泛研究,大體可以歸納為三個(gè)方面.一是多源遙感AOD數(shù)據(jù)的融合[16-20].常用數(shù)據(jù)源有MODIS AOD、MISR AOD、SeaWiFs AOD、Caliop AOD、Himavari-8AOD等,上述數(shù)據(jù)源大多存在同時(shí)缺失、互補(bǔ)性差的問題,只有高時(shí)間分辨率的Himavari- 8AOD產(chǎn)品能夠一定程度上改善上述問題,因此應(yīng)用案例逐漸增多[21],但到目前為止單獨(dú)使用多源遙感AOD數(shù)據(jù)融合方法仍然難以達(dá)到時(shí)空全覆蓋的目標(biāo)要求.二是基于AOD高相關(guān)因子的補(bǔ)值.與AOD高度相關(guān)的因子主要有地面站點(diǎn)監(jiān)測的PM2.5質(zhì)量濃度數(shù)據(jù)、地面氣象站點(diǎn)監(jiān)測的氣象能見度數(shù)據(jù)、空氣質(zhì)量模式模擬的PM2.5數(shù)據(jù)等[22-23].基于PM2.5因子進(jìn)行線性回歸補(bǔ)值[24]和改進(jìn)的Elterman經(jīng)驗(yàn)?zāi)Ｐ头囱軦OD的方法[25-26]存在AOD值被高估的缺點(diǎn),而衛(wèi)星遙感AOD數(shù)據(jù)與源于空氣質(zhì)量模式的AOD數(shù)據(jù)相結(jié)合能夠獲得時(shí)空全覆蓋的AOD數(shù)據(jù)集,因此近年來這類研究明顯增加[18,21,29-31],其填補(bǔ)缺失值的總體精度主要決定于空氣質(zhì)量模式AOD的模擬精度,Xiao等[27]的多種插補(bǔ)結(jié)果代表了目前的總體精度.三是基于AOD數(shù)據(jù)的插值.用于插值的AOD數(shù)據(jù)可以是單一來源的遙感數(shù)據(jù),也可以是多源遙感融合數(shù)據(jù)和基于高相關(guān)因子的補(bǔ)值數(shù)據(jù),比較常用的地統(tǒng)計(jì)插值方法有普通克里金插值[15]和時(shí)空克里金插值[32]等,時(shí)空克里金同時(shí)考慮了時(shí)間和空間自相關(guān),與普通克里金法相對比,插值后的精度更高[32].這類研究的總體精度主要取決于原始AOD數(shù)據(jù)的時(shí)空覆蓋狀況能否滿足相關(guān)插值方法的最優(yōu)采樣要求,能滿足最優(yōu)采樣要求者則插值精度高,反之則相反.

目前相關(guān)學(xué)者開發(fā)的混合效應(yīng)模型和時(shí)空混合效應(yīng)模型以AOD為主要估測變量,加入氣溫、降水等時(shí)間輔助變量以及海拔高度、人口密度等空間輔助變量對近地表PM2.5濃度進(jìn)行預(yù)測,并取得了較好的預(yù)測效果[8,33-44].而本研究提供了一種AOD補(bǔ)值的新方法,以2020年京津冀區(qū)域多角度大氣校正的氣溶膠光學(xué)厚度(MAIAC AOD)數(shù)據(jù)和地面監(jiān)測網(wǎng)絡(luò)的PM2.5質(zhì)量濃度監(jiān)測數(shù)據(jù)為基礎(chǔ),首次將PM2.5站點(diǎn)數(shù)據(jù)為預(yù)報(bào)因子,建立了時(shí)空混合效應(yīng)模型以及在此基礎(chǔ)上嵌套不同時(shí)間尺度后的嵌套模型,填補(bǔ)缺失站點(diǎn)所在網(wǎng)格的AOD數(shù)據(jù).旨在提高站點(diǎn)處AOD數(shù)據(jù)的覆蓋率,為相關(guān)PM2.5質(zhì)量濃度預(yù)測模型的建立提供完整的站點(diǎn)AOD數(shù)據(jù).

1 材料與方法

1.1 研究區(qū)概況

京津冀地區(qū)位于渤海西岸,是中國北方經(jīng)濟(jì)中心地帶,面積21.8萬km2,截止到2021年總?cè)丝跒?.103億.平原主要分布在京津冀東南部,這里人口密度大、工業(yè)種類繁多,使空氣污染更加嚴(yán)重;此外,冬季以小風(fēng)為主的偏南風(fēng)被燕山和太行山阻擋,污染物很難擴(kuò)散,進(jìn)一步加劇了該地區(qū)的空氣污染.使用ArcMap10.3將整個(gè)區(qū)域劃分為1×1km2的網(wǎng)格單元,總共生成216,082個(gè)柵格單元.圖1顯示了京津冀地區(qū)的地理位置、海拔高度、主要城市以及國家級和省級空氣質(zhì)量監(jiān)測站的分布情況.

圖1 研究區(qū)地理位置及空氣質(zhì)量監(jiān)測站分布

1.2 MAIAC AOD

多角度大氣校正算法(MAIAC)是基于MODIS測量開發(fā)的算法.它使用時(shí)間序列分析和基于圖像的處理技術(shù),在空間分辨率為1km的暗色植被和較亮表面范圍內(nèi)進(jìn)行云檢測、氣溶膠檢索和大氣校正[45-46].MAIAC AOD比現(xiàn)有的10km MODIS AOD具有更好的性能,可用于估算地表大氣顆粒物質(zhì)量濃度[47-48].從LAADS DAAC (https://ladsweb.modaps. eosdis. nasa.gov/)下載2020年1月1日～2020年12月31日覆蓋整個(gè)京津冀區(qū)域AOD數(shù)據(jù).MAIAC提供指示檢索質(zhì)量的質(zhì)量保證(QA)標(biāo)志,包括陸地/水/積雪掩模、鄰接掩模和云掩模(即靠近積雪或云).本文應(yīng)用基于MAIAC指導(dǎo)的閾值,應(yīng)用質(zhì)量保證(QA)標(biāo)志和不確定性值(UN)來排除具有錯(cuò)誤AOD值的像素[49].為了提高M(jìn)AIAC檢索的覆蓋率,本文逐月建立Aqua和Terra MAIAC AOD之間的線性回歸模型,用于估計(jì)缺失的Aqua或者Terra AOD,上午星和下午星的相關(guān)系數(shù)較高,各月平均2為0.85(0.79～ 0.93).總體來看,對應(yīng)監(jiān)測站點(diǎn)的AOD數(shù)據(jù)缺失嚴(yán)重,有效數(shù)據(jù)為51615個(gè),全年大約有56%柵格單元AOD缺失,夏季缺失最為嚴(yán)重,并且存在某日站點(diǎn)AOD完全缺失的情況(例如2020年第6、129、322日等).

1.3 AERONET AOD數(shù)據(jù)

通過全球布站的氣溶膠特性地基觀測網(wǎng)(http://aeronet.gsfc.nasa.gov/)下載AERONET AOD數(shù)據(jù),用于驗(yàn)證MAIAC AOD數(shù)據(jù)融合后的精度,由于AERONET AOD處理程度不同產(chǎn)生不同水平的AOD數(shù)據(jù),分別為1級即沒有經(jīng)過云過濾和質(zhì)量檢驗(yàn),1.5級即經(jīng)過云過濾但沒有質(zhì)量檢驗(yàn),2級即有云過濾也有質(zhì)量檢驗(yàn).2020年AERONET AOD只提供了1.5級的數(shù)據(jù)產(chǎn)品.因此,本文下載2020年1月1日～2020年12月31日期間每日Beijing、Beijing- CAMS和Xianghe3個(gè)站點(diǎn)的Version3的1.5級AERONET AOD數(shù)據(jù)用于驗(yàn)證.

AERONET站點(diǎn)不包括550nm波段的AOD數(shù)據(jù),因此通過AERONET 440nm和675nm2個(gè)波段的AOD插出550nm波段的AOD值[50].本文采用融合后的AOD數(shù)據(jù)與AERONET AOD進(jìn)行了驗(yàn)證,決定系數(shù)R為0.91,平均偏差為0.07,具有較好的精度.

1.4 PM2.5數(shù)據(jù)

2020年P(guān)M2.5地面濃度基礎(chǔ)數(shù)據(jù)來源于全國城市空氣質(zhì)量實(shí)時(shí)發(fā)布平臺(http://106.37.208.233: 20035/),從中獲取時(shí)間分辨率為小時(shí)的京津冀范圍內(nèi)318個(gè)國家空氣質(zhì)量監(jiān)測站和省控空氣質(zhì)量監(jiān)測站數(shù)據(jù),其中北京市范圍內(nèi)有34個(gè)監(jiān)測點(diǎn)位,天津市范圍內(nèi)有27個(gè)監(jiān)測點(diǎn)位.河北省包括石家莊等11個(gè)地級市共257個(gè)監(jiān)測點(diǎn)位.本文刪除了至少連續(xù)3個(gè)小時(shí)的重復(fù)觀測量,因?yàn)檫@些測量可能是由于儀器故障造成的[51].每小時(shí)測量值<1μg/m3的數(shù)據(jù)也被刪除,因?yàn)樗陀趦x器的檢測極限[36].每日有效小時(shí)數(shù)小于18的數(shù)據(jù)也被剔除[37].監(jiān)測站點(diǎn)PM2.5數(shù)據(jù)完整度比較好,只有少部分?jǐn)?shù)據(jù)缺失,日均值有效數(shù)據(jù)為115049個(gè),占全部應(yīng)有數(shù)據(jù)總量的98.84%.

1.5 異常值處理

庫克距離(Cook's Distance)描述了單個(gè)樣本對整個(gè)回歸模型的影響程度.庫克距離越大,說明影響越大,在最理想的情況下,每個(gè)樣本對模型的影響是相等的.如某個(gè)樣本的庫克距離非常大,便可以視為這個(gè)樣本是異常值[52].為了降低異常值對模型的影響,本文計(jì)算了模型估算AOD值與MAIAC AOD之間的庫克距離.庫克距離計(jì)算公式如下:

式中:D表示第個(gè)記錄的庫克距離; y表示第個(gè)記錄的模型擬合值;y表示去掉第個(gè)記錄后重新擬合得到的第個(gè)記錄的模型擬合值;為回歸模型系數(shù)的個(gè)數(shù);MSE為均方誤差.本研究刪除0.75%對模型結(jié)果影響較大的記錄(共343條數(shù)據(jù)記錄)后重新進(jìn)行建模.

1.6 數(shù)據(jù)整合

在AOD和PM2.5數(shù)據(jù)匹配的過程中,剔除了一日中不滿足4個(gè)分區(qū)中至少有1個(gè)AOD-PM2.5匹配數(shù)據(jù)的情況,并且剔除異常值后最終得到有效日數(shù)321d,AOD與PM2.5數(shù)據(jù)匹配后的有效數(shù)據(jù)為44902條.缺失數(shù)據(jù)日數(shù)45d,未出現(xiàn)周數(shù)據(jù)完全缺失的情況.

1.7 模型建立

時(shí)空異質(zhì)性是地表PM2.5濃度的重要特征,早期的線性混合效應(yīng)模型僅僅考慮了AOD-PM2.5關(guān)系的時(shí)間隨機(jī)效應(yīng).而在較大的地理區(qū)域內(nèi),由于顆粒物組成、邊界層高度、相對濕度、PM2.5濃度垂直廓線等因素的不同, AOD-PM2.5關(guān)系也存在空間隨機(jī)效應(yīng)[53].為此,本文在早期的時(shí)間混合效應(yīng)模型基礎(chǔ)上,加入空間隨機(jī)效應(yīng),構(gòu)建了時(shí)空線性混合效應(yīng)模型 (STLME) 模擬AOD -PM2.5關(guān)系的時(shí)空變化.根據(jù)研究區(qū)地理?xiàng)l件和發(fā)展水平綜合差異將其劃分成京津(北京、天津)、河北山地(承德、張家口)、河北內(nèi)陸(石家莊、保定、衡水、廊坊、邢臺、邯鄲)和河北沿海(秦皇島、唐山、滄州)等4個(gè)次區(qū)域[33].在模型的時(shí)間隨機(jī)效應(yīng)中嵌套了次區(qū)域的空間隨機(jī)效應(yīng),公式如下:

式(2)在AOD數(shù)據(jù)全部缺失的日期,只能通過固定效應(yīng)參數(shù)估算AOD值,預(yù)期精度將會(huì)較差.為了提高AOD數(shù)據(jù)全部缺失情況下模型估算AOD值的精度、進(jìn)而提高AOD補(bǔ)值的時(shí)空覆蓋率,在公式(2)基礎(chǔ)上,本文將月、周、日3種不同時(shí)間尺度加入到時(shí)空混合效應(yīng)模型中,構(gòu)建了時(shí)空嵌套線性混合效應(yīng)模型(STNLME),該模型不僅反映了AOD-PM2.5關(guān)系的每日、每周、每月的時(shí)間隨機(jī)效應(yīng),而且反映了相應(yīng)時(shí)間尺度下的空間隨機(jī)效應(yīng).與公式(2)相比,在AOD數(shù)據(jù)全部缺失的某日可以通過固定效應(yīng)參數(shù)、周尺度隨機(jī)效應(yīng)參數(shù)和對應(yīng)的空間隨機(jī)效應(yīng)參數(shù)估算AOD值;同理,當(dāng)某周AOD數(shù)據(jù)全部缺失時(shí),通過月尺度隨機(jī)效應(yīng)等估算每日的AOD值.模型如下:

式中:Month,Week,Day分別為模型月、周、日的隨機(jī)截距,包括建模年份中參與數(shù)據(jù)匹配的每月、每周、每日的隨機(jī)截距;Month,Week,Day分別為模型的月、周、日的隨機(jī)斜率;Month(reg),Week(reg),Day(reg)分別為各個(gè)分區(qū)每月、每周、每日的隨機(jī)截距;Month(reg),Month(reg),Month(reg)分別為各個(gè)分區(qū)每月、每周、每日的隨機(jī)斜率;1,2,3為日、周、月隨機(jī)效應(yīng)的方差-協(xié)方差矩陣,1REG,2REG,3REG分別為嵌套在日、周、月中的空間隨機(jī)效應(yīng)的方差-協(xié)方差矩陣.

為了比較模型的估算效果,本文將對時(shí)空混合效應(yīng)模型擬合的結(jié)果與時(shí)間線性混合效應(yīng)模型和線性回歸模型相應(yīng)結(jié)果進(jìn)行比較分析.

1.8 模型合理性驗(yàn)證

首先使用基于樣本十折交叉驗(yàn)證(Sample based CV)的方法分別對所建立模型的過擬合程度進(jìn)行檢測.該方法的原理是將建模數(shù)據(jù)集整體隨機(jī)分成10份,其中每一份數(shù)據(jù)大約包含整體數(shù)據(jù)集的1/10.在交叉驗(yàn)證中,選擇其中的1份作為測試集,而剩余的9份則作為訓(xùn)練集為測試集提供模型的參數(shù).以上方法重復(fù)10以確保每1份數(shù)據(jù)都參與了模型驗(yàn)證.

基于樣本的十折交叉驗(yàn)證方法常常出現(xiàn)參與建模的子集和進(jìn)行驗(yàn)證的子集具有相同的日期,這樣難以驗(yàn)證沒有AOD-PM2.5匹配情況下模型模擬精度.所以本文繼而使用了基于日序的十折交叉驗(yàn)證方法(Day-of-Year based CV),原理是將建模數(shù)據(jù)所有日期即321d隨機(jī)分為10個(gè)子集,建模和驗(yàn)證子集將不出現(xiàn)相同日期的情況,基于日序的十折交叉驗(yàn)證的方法可以用于評估那些沒有AOD-PM2.5匹配日期模型的估算性能.

使用驗(yàn)證得到的估算的AOD值與衛(wèi)星反演的AOD值之間線性回歸決定系數(shù)(R)、均方根誤差(RMSE)和相對預(yù)測誤差(RPE)等指標(biāo)來檢測模型的估算精度.其中,RMSE和RPE計(jì)算公式如下:

2 結(jié)果分析

2.1 描述性統(tǒng)計(jì)

如圖2所示,2020年的PM2.5數(shù)據(jù)和AOD數(shù)據(jù)均成正態(tài)分布,符合建模要求.表1為2020年參與建模數(shù)據(jù)PM2.5和AOD的均值、極值、標(biāo)準(zhǔn)差等統(tǒng)計(jì)量描述.其中AOD的年均值為0.34,PM2.5的均值為39.73μg/m3,接近國家環(huán)境空氣質(zhì)量(GB 3095-2012)二級標(biāo)準(zhǔn)的限值(35μg/m3).此外,參與建模的數(shù)據(jù)集的變量表現(xiàn)出顯著的空間差異,如保定、石家莊、邢臺、邯鄲等中南部平原PM2.5濃度較高,張家口、承德等北部山區(qū)PM2.5濃度較低,標(biāo)準(zhǔn)差范圍為18.21～ 37.54μg/m3.AOD表現(xiàn)出和PM2.5類似的分布特征,石家莊、邯鄲最高,保定市次之,承德、張家口最低.

圖2 2020年參與建模數(shù)據(jù)

表1 參與建模數(shù)據(jù)描述性統(tǒng)計(jì)

表2 模型的固定效應(yīng)

表2總結(jié)了時(shí)空線性混合效應(yīng)模型(STLME)和時(shí)空嵌套線性混合效應(yīng)模型(STNLME)的固定效應(yīng)參數(shù),模型的固定效應(yīng)包括固定截距和固定斜率,表示自變量PM2.5對AOD的固定影響.其中兩個(gè)模型的固定斜率分別為0.007和0.007,表明PM2.5和AOD之間存在正相關(guān)關(guān)系,這與相關(guān)學(xué)者的研究相一致[41,43,54-57].兩個(gè)模型截距和斜率檢驗(yàn)的值均<0.0001,具有統(tǒng)計(jì)學(xué)上的顯著性.

2.2 模型擬合與驗(yàn)證

2.2.1 模型擬合與驗(yàn)證結(jié)果分析 模型對AOD的估算精度將直接影響建模的合理性,如圖3所示,STLME擬合后的決定系數(shù)2為0.884,RMSE為0.105;STNLME擬合后的2為0.887,RMSE為0.103;均表現(xiàn)出良好的估算性能.STLME和STNLME基于樣本的CV2值分別為0.868和0.874,CV RMSE分別為0.112和0.109,STNLME模型比STLME模型的精度略有改善,并且兩模型均無過擬合現(xiàn)象.由于用于驗(yàn)證樣本的數(shù)據(jù)和在訓(xùn)練樣本中數(shù)據(jù)可能存在相同的日數(shù),所以基于樣本的十折交叉驗(yàn)證方法只能反映當(dāng)日存在AOD-PM2.5匹配數(shù)據(jù)情況下的模型性能.也就是說,與非嵌套模型相比,在存在AOD-PM2.5匹配數(shù)據(jù)的日期,嵌套模型并沒有改善性能.然而,從基于日的十折交叉驗(yàn)證的結(jié)果看,嵌套模型能顯著提高在沒有AOD-PM2.5匹配數(shù)據(jù)時(shí)模型的性能,STNLME基于日的CV2為0.630,明顯高于非嵌套的STLME(CV2=0.263).可見,嵌套模型通過引入周和月隨機(jī)效應(yīng),明顯提高了在日AOD數(shù)據(jù)完全缺失情況下對AOD值的估算精度,并提高了AOD補(bǔ)值的時(shí)空覆蓋率.

圖3 2020年STLME和STNLME擬合效果和交叉驗(yàn)證結(jié)果對比

圖4 線性回歸估算AOD擬合結(jié)果

圖5 2020年LME和NLME擬合效果和交叉驗(yàn)證結(jié)果對比

2.2.2 與線性回歸模型、時(shí)間線性混合效應(yīng)模型的比較 遵循Lü等[24]提出的方法,用PM2.5監(jiān)測站點(diǎn)數(shù)據(jù)在暖季和冷季分別對每個(gè)城市進(jìn)行線性回歸擬合來填補(bǔ)網(wǎng)格單元中的AOD缺失.擬合結(jié)果如圖4所示,模型擬合2為0.383,斜率為0.383.并且在十折交叉驗(yàn)證后,2為0.381.線性回歸模型(LR)具有較高的均方誤差(0.241),模型的估算精度較低.如圖5所示,時(shí)空混合效應(yīng)模型(STLME)與時(shí)間混合效應(yīng)模型(LME)相比、嵌套時(shí)空混合效應(yīng)模型(STNLME)與嵌套時(shí)間混合效應(yīng)模型(NLME)相比,前者的估算精度均高于后者.同時(shí)也可以看出,時(shí)間隨機(jī)效應(yīng)對模型估算精度的改善明顯高于空間隨機(jī)效應(yīng).

2.3 AOD補(bǔ)值后數(shù)據(jù)覆蓋率分析

為了定量描述AOD在補(bǔ)值前后的時(shí)間和空間分布特征,引入空間維有效比率(SVVR)和時(shí)間維有效比率(TVVR)的概念.SVVR定義為每日監(jiān)測站點(diǎn)AOD有效柵格數(shù)與同日該區(qū)域監(jiān)測站點(diǎn)全部柵格數(shù)之比,TVVR定義為監(jiān)測站點(diǎn)AOD有效天數(shù)與全部天數(shù)之比.圖6(a)是2020年站點(diǎn)補(bǔ)值前每個(gè)日期的空間覆蓋率散點(diǎn)圖,據(jù)統(tǒng)計(jì)全年有209d站點(diǎn)空間覆蓋率不足50%,夏季數(shù)據(jù)缺失最為嚴(yán)重.圖6(b)為模型補(bǔ)值后的每個(gè)日期的空間覆蓋率散點(diǎn)圖,空間覆蓋率除了一月份有部分日期沒有超過90%之外,其余日期的空間覆蓋率全部在90%以上.

表3 2020年站點(diǎn)AOD數(shù)據(jù)補(bǔ)值前后時(shí)空維有效比率

模型補(bǔ)值后的空間有效覆蓋率如表3所示, 2020年原始數(shù)據(jù)的空間覆蓋率很低,年均SVVR低于45%,經(jīng)過時(shí)空混合效應(yīng)模型補(bǔ)值后,年均覆蓋率從原始數(shù)據(jù)的44.35%提升到87.05%,經(jīng)時(shí)空嵌套混合效應(yīng)模型補(bǔ)值后提高到99.35%,只有極少站點(diǎn)存在數(shù)據(jù)缺失,這是因站點(diǎn)PM2.5數(shù)據(jù)缺失造成的.據(jù)統(tǒng)計(jì)在366d所有站點(diǎn)對應(yīng)的PM2.5數(shù)據(jù)有1339條是無效數(shù)據(jù),占數(shù)據(jù)總條數(shù)的1.14%.再看AOD補(bǔ)值前后時(shí)間維有效比率,2020年存在AOD數(shù)值的有效天數(shù)為321d,經(jīng)過時(shí)空嵌套混合效應(yīng)模型補(bǔ)值后均提升到100%.時(shí)空混合效應(yīng)模型并沒有提高時(shí)間覆蓋率,而時(shí)空嵌套混合效應(yīng)模型補(bǔ)值顯著提高了時(shí)間維度的覆蓋率.

圖6 2020年站點(diǎn)處AOD補(bǔ)值前后空間有效覆蓋率

2.4 京津冀監(jiān)測站點(diǎn)AOD補(bǔ)值前后年均值分析

如圖7所示,AOD補(bǔ)值前(圖7(a))和AOD補(bǔ)值后(圖7(b))有著類似的空間分布,但補(bǔ)值后的AOD年均值明顯高于補(bǔ)值前AOD年均值.以石家莊的監(jiān)測站點(diǎn)為例,在補(bǔ)值前AOD均值范圍大約在0.35～ 0.58,填補(bǔ)后AOD值上升到0.58～0.67.由圖8可知,補(bǔ)值每個(gè)站點(diǎn)年均AOD的值明顯提高,補(bǔ)值后的數(shù)據(jù)可以有效彌補(bǔ)AOD估算PM2.5時(shí)出現(xiàn)的高值低估問題的不足.2020年年均AOD低值區(qū)主要分布在北部燕山山區(qū)和西部太行山地區(qū),主要包括承德和張家口地區(qū);年均AOD高值區(qū)則分布在京津冀南部內(nèi)陸平原地區(qū),主要以邢臺、邯鄲、石家莊等監(jiān)測站點(diǎn)為主,并與京津冀PM2.5濃度的空間分布狀況基本一致.

圖7 AOD補(bǔ)值前后年均值空間分布

圖8 AOD補(bǔ)值前后各站點(diǎn)年均值大小對比

3 討論

3.1 AOD缺失的影響

本文分別求出當(dāng)AOD缺失時(shí)對應(yīng)站點(diǎn)的PM2.5濃度年均值(圖9短點(diǎn)劃線),沒有AOD數(shù)據(jù)缺失時(shí)對應(yīng)站點(diǎn)的PM2.5年平均值(圖9短劃線)以及每一個(gè)站點(diǎn)的總平均值(圖9實(shí)線),對2020年中共318個(gè)站點(diǎn)進(jìn)行編號,以總平均值的升序繪制了每個(gè)站點(diǎn)對應(yīng)的PM2.5均值,顯然當(dāng)AOD數(shù)據(jù)缺失時(shí)對應(yīng)的站點(diǎn)處PM2.5濃度整體上高于AOD數(shù)據(jù)有效時(shí)對應(yīng)的站點(diǎn)處的PM2.5濃度.所以由于缺失的AOD值往往與更高的污染水平相關(guān),不考慮缺失的AOD數(shù)據(jù)就去估算PM2.5的濃度往往會(huì)出現(xiàn)被低估的情況.

圖9 AOD缺失與否與站點(diǎn)PM2.5濃度的關(guān)系

3.2 本模型對AOD-PM2.5關(guān)系復(fù)雜性的處理

氣溶膠的吸濕性增長特性、各類顆粒質(zhì)譜特征、粒徑分布差異、邊界層的高低等因素會(huì)對AOD值的大小產(chǎn)生影響,濕度、溫度、風(fēng)速等氣象要素也會(huì)影響氣溶膠聚集、傳輸和擴(kuò)散,因此AOD和PM2.5之間存在著復(fù)雜的相關(guān)關(guān)系,而并非簡單的線性關(guān)系.有關(guān)研究表明在氣溶膠集中、云量稀少、濕度低的情況下, AOD- PM2.5線性相關(guān)性較為顯著,而在其他條件下AOD和PM2.5也會(huì)呈現(xiàn)出非線性關(guān)系[58].本文的模型通過隨機(jī)效應(yīng)項(xiàng)表征AOD-PM2.5關(guān)系的復(fù)雜性,對兩者的非線性關(guān)系起到了校正作用.圖10為時(shí)空混合效應(yīng)模型擬合的每日隨機(jī)斜率變化圖,可以看出AOD-PM2.5關(guān)系存在著從日到季節(jié)不同時(shí)間尺度的非線性變化.隨機(jī)斜率整體上呈現(xiàn)出先升高后下降的變化趨勢,春季和秋季的斜率變化比較復(fù)雜,斜率有正有負(fù).夏季的隨機(jī)斜率絕大多數(shù)為正數(shù),AOD的估算值要高于總體平均值,這與夏季AOD值較高相對應(yīng);而冬季隨機(jī)斜率大多數(shù)為負(fù)值,AOD隨PM2.5濃度的增加而減小,AOD的估算值要低于總體平均值,這與冬季AOD的低值相對應(yīng).模型通過每日的隨機(jī)效應(yīng)有效避免了夏季AOD估算值被低估,冬季AOD估算值被高估的問題.

圖10 日隨機(jī)斜率的變化

3.3 與其他AOD補(bǔ)值方法的比較

圖11 基于時(shí)空混合效應(yīng)模型估算的AOD與MAIAC AOD月均值擬合

本研究建立了STLME和STNLME模型來估算有監(jiān)測PM2.5數(shù)據(jù)但沒有AOD有效數(shù)據(jù)的網(wǎng)格單元的AOD值,基于AOD值與PM2.5觀測值呈線性關(guān)系的假設(shè),即考慮了日、周和月的時(shí)間尺度變化效應(yīng)也考慮了空間變化效應(yīng),獲得了較好的AOD估算性能.表4顯示了本文的補(bǔ)值方法與國內(nèi)外其他補(bǔ)值方法的比較,與Lv等[24]建立的華北地區(qū)冷季節(jié)和暖季節(jié)的線性回歸模型相比,本模型具有更高的模型擬合2和更低的均方根誤差.與Xiao等[27]估算長江三角洲地區(qū)缺失的MAIAC AOD使用的多重插補(bǔ)(MI)方法相比,2也有明顯的提高.Zhang等[26]基于氣象能見度應(yīng)用KM-Elterman模型反演近地面AOD,月反演AOD與MODIS測量AOD擬合的相關(guān)系數(shù)為0.71、RMSE為0.207.而本模型擬合達(dá)到了0.93,RMSE僅有0.054,模型估算精度高(圖11).當(dāng)然,模型雖然精度和穩(wěn)健性都較好但也存在局限性,首先,對缺失AOD的估算在很大程度上依賴于PM2.5的測量,當(dāng)PM2.5測量值稀疏或不存在時(shí),模型的應(yīng)用將受到限制.另外,本模型可以提高站點(diǎn)AOD數(shù)據(jù)的覆蓋率,但不能覆蓋整個(gè)區(qū)域,需要進(jìn)一步采用空間插值等其他方法,這也是下一步要解決的關(guān)鍵問題.

表4 不同補(bǔ)值方法的性能國內(nèi)外比較

注:“—”表示文章中沒有相關(guān)數(shù)據(jù).

4 結(jié)論

4.1 由于同時(shí)考慮了AOD-PM2.5關(guān)系的時(shí)空異質(zhì)性,時(shí)空混合效應(yīng)模型(STLME)比早期的時(shí)間混合效應(yīng)模型(LME)估算精度高,也明顯高于相關(guān)學(xué)者使用的線性回歸模型(LR)和非線性回歸模型,為AOD補(bǔ)值研究提供了一種新方法.通過該模型補(bǔ)值,使研究區(qū)監(jiān)測站點(diǎn)年均AOD數(shù)據(jù)空間維有效比率從44.35%提高到87.05%.

4.2 在有AOD -PM2.5匹配數(shù)據(jù)的日期,STLME模型和STNLME模型的估算精度相近;在無AOD -PM2.5匹配數(shù)據(jù)的日期,STLME模型的AOD估算值精度較低,而STNLME模型通過引入周隨機(jī)效應(yīng)和月隨機(jī)效應(yīng),大大提高了AOD值的估算精度.通過該模型補(bǔ)值,使得研究區(qū)監(jiān)測站點(diǎn)年均AOD數(shù)據(jù)空間維有效比率從87.05%提高到99.35%、時(shí)間維有效比率從87.94%提高到100%.

4.3 經(jīng)嵌套模型補(bǔ)值后,每個(gè)站點(diǎn)的年均AOD值都得到明顯提高,表明缺失的高值A(chǔ)OD被估算出來.這一結(jié)果可以有效糾正AOD估算PM2.5時(shí)出現(xiàn)的高值低估問題,對降低其在健康影響評價(jià)中的偏差具有重要意義.

[1] Kaufman Y J, Tanre D, Boucher O. A satellite view of aerosols in the climate system [J]. Nature, 2002,419(6903):215-223.

[2] Zhang M, Ma Y, Gong W, et al. Aerosol optical properties and radiative effects: Assessment of urban aerosols in central China using 10-year observations [J]. Atmospheric Environment, 2018,182:275- 285.

[3] Brauer M, Freedman G, Frostad J, et al. Ambient air pollution exposure estimation for the global burden of disease 2013 [J]. Environmental Science & Technology, 2016,50(1):79-88.

[4] Chow J C, Watson J G, Mauderly J L, et al. Health effects of fine particulate air pollution: lines that connect [J]. Journal of the Air & Waste Management Association, 2006,56(10):1368-1380.

[5] Liao Q, Jin W, Tao Y, et al. Health and economic loss assessment of PM2.5pollution during 2015～2017 in Gansu Province, China [J]. International Journal of Environmental Research and Public Health, 2020,17(9):3253.

[6] Chen G, Li S, Knibbs L D, et al. A machine learning method to estimate PM2.5concentrations across China with remote sensing, meteorological and land use information [J]. Science of the Total Environment, 2018,636:52-60.

[7] Kloog I, Nordio F, Coull B A, et al. Incorporating local land use regression and satellite aerosol optical depth in a hybrid model of spatiotemporal PM2.5exposures in the Mid-Atlantic states [J]. Environmental Science & Technology, 2012,46(21):11913-11921.

[8] Ma Z, Liu R, Liu Y, et al. Effects of air pollution control policies on PM2.5pollution improvement in China from 2005 to 2017: a satellite-based perspective [J]. Atmospheric Chemistry and Physics, 2019,19(10):6861-6877.

[9] van Donkelaar A, Martin R V, Levy R C, et al. Satellite-based estimates of ground-level fine particulate matter during extreme events: A case study of the Moscow fires in 2010 [J]. Atmospheric Environment, 2011,45(34):6225-6232.

[10] Zhang Y, Li Z. Remote sensing of atmospheric fine particulate matter PM2.5mass concentration near the ground from satellite observation [J]. Remote Sensing of Environment, 2015,160:252-262.

[11] Levy R C, Remer L A, Kleidman R G, et al. Global evaluation of the Collection 5MODIS dark-target aerosol products over land [J]. Atmospheric Chemistry and Physics, 2010,10(21):10399-10420.

[12] Tao M, Chen L, Su L, et al. Satellite observation of regional haze pollution over the North China Plain [J]. Journal of Geophysical Research: Atmospheres, 2012,117.

[13] Superczynski S D, Kondragunta S, Lyapustin A I. Evaluation of the multi-angle implementation of atmospheric correction (MAIAC) aerosol algorithm through intercomparison with VIIRS aerosol products and AERONET [J]. Journal of Geophysical Research- Atmospheres, 2017,122(5):3005-3022.

[14] Li X, Zhang X. Predicting ground-level PM2.5concentrations in the Beijing-Tianjin-Hebei region: A hybrid remote sensing and machine learning approach [J]. Environmental Pollution, 2019,249:735-749.

[15] Lv B, Hu Y, Chang H H, et al. Improving the accuracy of daily PM2.5distributions derived from the fusion of ground-level measurements with aerosol optical depth observations, a case study in North China [J]. Environmental Science & Technology, 2016,50(9):4752-4759.

[16] Chatterjee A, Michalak A M, Kahn R A, et al. A geostatistical data fusion technique for merging remote sensing and ground-based observations of aerosol optical thickness [J]. Journal of Geophysical Research-Atmospheres, 2010,doi:10.1029/2009JD013765.

[17] Zubko V, Leptoukh G G, Gopalan A. Study of data-merging and interpolation methods for use in an interactive online analysis system: MODIS Terra and Aqua daily aerosol case [J]. Ieee Transactions on Geoscience and Remote Sensing, 2010,48(12):4219-4235.

[18] van Donkelaar A, Martin R V, Brauer M, et al. Global estimates of fine particulate matter using a combined geophysical-statistical method with information from satellites, models, and monitors [J]. Environmental Science & Technology, 2016,50(7):3762-3772.

[19] Xu H, Guang J, Xue Y, et al. A consistent aerosol optical depth (AOD) dataset over mainland China by integration of several AOD products [J]. Atmospheric Environment, 2015,114:48-56.

[20] Wei X, Bai K, Chang N-B, et al. Multi-source hierarchical data fusion for high-resolution AOD mapping in a forest fire event [J]. International Journal of Applied Earth Observation and Geoinformation, 2021,102:102366.

[21] Jiang T, Chen B, Nie Z, et al. Estimation of hourly full-coverage PM2.5concentrations at 1-km resolution in China using a two-stage random forest model [J]. Atmospheric Research, 2021,248.

[22] 董 焱,許 丹,鮑艷松,等.基于AGRI顆粒物濃度遙感反演及季節(jié)變化分析 [J]. 中國環(huán)境科學(xué), 2021,41(2):633-642.

Dong Y, Xu D, Bao Y S, et al. Remote sensing retrieval and seasonal variation analysis of particulate matter concentration based on AGRI [J]. China Environmental Science, 2021,41(2):633-642.

[23] 楊 旭,唐穎瀟,蔡子穎,等.基于氣溶膠三維變分同化天津PM2.5數(shù)值預(yù)報(bào)研究 [J]. 中國環(huán)境科學(xué), 2021,41(12):5476-5484.

Yang X, Tang Y X, Cai Z Y, et al. Study on numerical prediction of Tianjin PM2.5based on three-dimensional variational assimilation of Aerosol [J]. China Environmental Science, 2021,41(12):5476-5484.

[24] Lv B, Hu Y, Chang H H, et al. Daily estimation of ground-level PM2.5concentrations at 4km resolution over Beijing-Tianjin-Hebei by fusing MODIS AOD and ground observations [J]. Science of the Total Environment, 2017,580:235-244.

[25] Wu J, Luo J, Zhang L, et al. Improvement of aerosol optical depth retrieval using visibility data in China during the past 50years [J]. Journal of Geophysical Research: Atmospheres, 2014,119(23):13,370- 313,387.

[26] Zhang Z, Wu W, Wei J, et al. Aerosol optical depth retrieval from visibility in China during 1973～2014 [J]. Atmospheric Environment, 2017,171:38-48.

[27] Xiao Q, Wang Y, Chang H H, et al. Full-coverage high-resolution daily PM2.5estimation using MAIAC AOD in the Yangtze River Delta of China [J]. Remote Sensing of Environment, 2017,199:437-446.

[28] Chameides W L, Luo C, Saylor R, et al. Correlation between model-calculated anthropogenic aerosols and satellite-derived cloud optical depths: Indication of indirect effect? [J]. Journal of Geophysical Research: Atmospheres, 2002,107(D10):AAC 2-1-AAC 2-15.

[29] Li L, Franklin M, Girguis M, et al. Spatiotemporal Imputation of MAIAC AOD Using Deep Learning with Downscaling [J]. Remote Sensing of Environment, 2020,237.

[30] 李建新,劉小生,劉 靜,等.基于MRMR-HK-SVM模型的PM2.5濃度預(yù)測 [J]. 中國環(huán)境科學(xué), 2019,39(6):2304-2310.

Li J X, Liu X S, Liu J, et al. PM2.5concentration prediction based on MRMR-HK-SVM model [J]. China Environmental Science, 2019, 39(6):2304-2310.

[31] 嚴(yán)瑩婷,陸小曼,王嘉佳,等.基于GF-4衛(wèi)星的長三角城市群PM2.5遙感反演 [J]. 中國環(huán)境科學(xué), 2022,42(3):1005-1012.

Yan Y T, Lu X M, Wang J J, et al. Remote sensing retrieval of Yangtze River Delta Economic Zone PM2.5based on GF-4satellite [J]. China Environmental Science, 2022,42(3):1005-1012.

[32] Yang J, Hu M. Filling the missing data gaps of daily MODIS AOD using spatiotemporal interpolation [J]. Science of the Total Environment, 2018,633:677-683.

[33] 郝 靜.京津冀地區(qū)PM2.5濃度時(shí)空變化定量模擬 [D]. 石家莊:河北師范大學(xué), 2018.

Hao J. Simulation of the spatio-temporally resolved PM2.5aerosol mass concentration of the Beijing-Tianjin-Hebei Region [D]. Shijiazhuang: Hebei Normal University, 2018.

[34] Lee H J, Liu Y, Coull B A, et al. A novel calibration approach of MODIS AOD data to predict PM2.5concentrations [J]. Atmospheric Chemistry and Physics, 2011,11(15):7991-8002.

[35] Hu X, Waller L A, Al-Hamdan M Z, et al. Estimating ground-level PM2.5concentrations in the southeastern U.S. using geographically weighted regression [J]. Environmental Research, 2013,121:1-10.

[36] Hu X, Waller L A, Lyapustin A, et al. Estimating ground-level PM2.5concentrations in the Southeastern United States using MAIAC AOD retrievals and a two-stage model [J]. Remote Sensing of Environment, 2014,140:220-232.

[37] Ma Z, Hu X, Huang L, et al. Estimating ground-level PM2.5in China using satellite remote sensing [J]. Environmental Science & Technology, 2014,48(13):7436-7444.

[38] Ma Z, Liu Y, Zhao Q, et al. Satellite-derived high resolution PM2.5concentrations in Yangtze River Delta Region of China using improved linear mixed effects model [J]. Atmospheric Environment, 2016,133:156-164.

[39] Huang K, Xiao Q, Meng X, et al. Predicting monthly high-resolution PM2.5concentrations with random forest model in the North China Plain [J]. Environmental Pollution, 2018,242(Pt A):675-683.

[40] Liang F, Xiao Q, Wang Y, et al. MAIAC-based long-term spatiotemporal trends of PM2.5in Beijing, China [J]. Science of the Total Environment, 2018,616-617:1589-1598.

[41] Zhang Y, Wang W, Ma Y, et al. Improvement in hourly PM2.5estimations for the Beijing-Tianjin-Hebei region by introducing an aerosol modeling product from MASINGAR [J]. Environmental Pollution, 2020,264:114691.

[42] Wang W, He J, Miao Z, et al. Space–time linear mixed-effects (STLME) model for mapping hourly fine particulate loadings in the Beijing–Tianjin–Hebei region, China [J]. Journal of Cleaner Production, 2021,292.

[43] Xue W, Zhang J, Zhong C, et al. Spatiotemporal PM2.5variations and its response to the industrial structure from 2000 to 2018 in the Beijing-Tianjin-Hebei region [J]. Journal of Cleaner Production, 2021,27.

[44] 周 爽,王春林,孫 睿,等.基于LME/BME的珠江三角洲PM2.5星地融合技術(shù)研究 [J]. 中國環(huán)境科學(xué), 2019,39(5):1869-1878.

Zhou S, Wang C L, Sun R, et al. PM2.5satellite-earth fusion technology in Pearl River Delta based on LME/BME [J]. China Environmental Science, 2019,39(5):1869-1878.

[45] Lyapustin A, Martonchik J, Wang Y, et al. Multiangle implementation of atmospheric correction (MAIAC): 1. Radiative transfer basis and look-up tables [J]. Journal of Geophysical Research, 2011,116: D03211.

[46] Lyapustin A , Wang Y, Laszlo I, et al. Multi-angle implementation of atmospheric correction for MODIS (MAIAC): 3. Atmospheric correction [J]. Remote Sensing of Environment, 2012,127:385-393.

[47] Zhang Z, Wu W, Fan M, et al. Evaluation of MAIAC aerosol retrievals over China [J]. Atmospheric Environment, 2019,202:8-16.

[48] Superczynski S D, Kondragunta S, Lyapustin A I. Evaluation of the multi-angle implementation of atmospheric correction (MAIAC) aerosol algorithm through intercomparison with VIIRS aerosol products and AERONET [J]. Journal of Geophysical Research: Atmospheres, 2017,122(5):3005-3022.

[49] Kloog I, Sorek-Hamer M, Lyapustin A, et al. Estimating daily PM2.5and PM10across the complex geo-climate region of Israel using MAIAC satellite-based AOD data [J]. Atmospheric Environment, 2015,122:409-416.

[50] Liu Y, Sarnat J A, Coull B A, et al. Validation of multiangle imaging spectroradiometer (MISR) aerosol optical thickness measurements using aerosol robotic network (AERONET) observations over the contiguous United States [J]. Journal of Geophysical Research- Atmospheres, 2004,109:D06205.

[51] Rohde R A, Muller R A. Air pollution in China: mapping of concentrations and sources [J]. PLoS One, 2015,10(8):e0135749.

[52] 馬宗偉.基于衛(wèi)星遙感的我國PM2.5時(shí)空分布研究 [D]. 南京:南京大學(xué), 2015.

Ma Z W. Study on spatial and temporal distribution of PM2.5in China based on satellite remote sensing [D]. Nanjing: Nanjing University, 2015.

[53] Kloog I, Chudnovsky A A, Just A C, et al. A new hybrid spatio- temporal model for estimating daily multi-year PM2.5concentrations across Northeastern USA using high resolution aerosol optical depth data [J]. Atmospheric Environment, 2014,95:581-590.

[54] 景 悅,孫艷玲,徐 昊,等.基于混合效應(yīng)模型的京津冀地區(qū)PM2.5日濃度估算 [J]. 中國環(huán)境科學(xué), 2018,38(8):2890-2897.

Jing Y, Sun Y L, Xu H, et al. Estimation of daily PM2.5concentrations in Beijing-Tianjin-Hebei region based on mixed effect model [J]. China Environmental Science, 2018,38(8):2890-2897.

[55] 孫 成,王 衛(wèi),劉方田,等.基于線性混合效應(yīng)模型的河北省PM2.5濃度時(shí)空變化模型研究 [J]. 環(huán)境科學(xué)研究, 2019,32(9):1500-1509.

Sun C, Wang W, Liu F T,et al. Spatial-temporal simulation of PM2.5concentrations in Hebei Province based on linear mixed effects model [J]. Research of Environmental Sciences, 2019,32(9):1500-1509.

[56] Guo W, Zhang B, Wei Q, et al. Estimating ground-level PM2.5concentrations using two-stage model in Beijing-Tianjin-Hebei, China [J]. Atmospheric Pollution Research, 2021,12(9):101154.

[57] 王家成,朱成杰,朱 勇,等.北京地區(qū)多氣溶膠遙感參量與PM2.5相關(guān)性研究 [J]. 中國環(huán)境科學(xué), 2015,35(7):1947-1956.

Wang J C, Zhu C J, Zhu Y, et al. The correlation between multiple aerosol remote sensing parameters and PM2.5in Beijing area [J]. China Environmental Science, 2015,35(7):1947-1956.

[58] 許悅蕾,劉延安,施潤和,等.氣象要素對氣溶膠光學(xué)厚度估算PM2.5的影響 [J]. 環(huán)境科學(xué)學(xué)報(bào), 2018,38(10):3868-3876.

Xu Y L, Liu Y A, Shi R H, et al. Influence of meteorological factors on Aerosol Optical Depth(AOD) estimation of PM2.5[J]. Acta Scientiae Circumstantiae, 2018,38(10):3868-3876.

Filling the missing data of AOD using the situ PM2.5monitoring measurements in the Beijing-Tianjin-Hebei region.

SONG Chun-jie1,2, WEI Qiang1,2, FAN Li-hang1,2, WANG Wei1,2*, HAN Fang1,2, LI Wei-miao1,2, LI Fu-xing1,3**, CHENG He-xi4

(1.School of Geographical Sciences, Hebei Normal University, Shijiazhuang 050024, China；2.Hebei Key Laboratory of Environmental Change and Ecological Construction, Shijiazhuang 050024, China；3.Hebei Technology Innovation Center for Remote Sensing Identification of Environmental Change, Shijiazhuang 050024, China；4.Handan Urban and Rural Planning Research Center, Handan 056000, China)., 2022,42(7)：3000～3012

A spatiotemporal linear mixed effect model (STLME) and a spatiotemporal nested linear mixed effect model (STNLME) were presented using the PM2.5measurements of 318 ground monitoring stations in Beijing-Tianjin-Hebei (BTH) in 2020 to fill the missing data of AOD. The results indicated that the STLME and STNLME models in the days with AOD-PM2.5matchups showed similar performance with the cross-validation (CV)2valued at 0.868 and 0.874, and the root mean square error (RMSE) valued at 0.112 and 0.109, respectively. However, the STNLME model with the CV2valued at 0.63 outperforms STLME with the CV2of 0.26 in the days without PM2.5-AOD matchups. After models filling, the spatial valid value ratio of AOD data in the grid where the monitoring stations are located was increased from 44.35% to 99.35%, and the temporal valid value ratio was increased from 87.94% to 100%. Meanwhile, the annual mean AOD value of each station had increased significantly, and the missing AOD were filled under the condition of high PM2.5level, which could reduce the biases of exposure assessment in air pollution and health studies.

MAIAC AOD；AOD filling of monitoring stations；spatiotemporal linear mixed effects model；spatiotemporal nested linear mixed effect model；Beijing-Tianjin-Hebei

X513

A

1000-6923(2022)07-3000-13

宋春杰(1996-),男,山東濟(jì)南人,主要研究方向?yàn)榇髿馕廴緯r(shí)空變化模擬.發(fā)表論文2篇.

2021-12-27

國家自然科學(xué)基金資助項(xiàng)目(41471091);河北省自然科學(xué)基金青年基金資助項(xiàng)目(D2019205027);河北省教育廳青年基金資助項(xiàng)目(QN2018035)

* 責(zé)任作者, 教授, wangwei@hebtu.edu.cn; ** 講師, lifuxing6042@163.com