基于貝葉斯決策樹(shù)的小麥鎘風(fēng)險(xiǎn)識(shí)別規(guī)則提取

仝桂杰,吳紹華,袁毓婕,顏道浩,周生路,李富富

?

基于貝葉斯決策樹(shù)的小麥鎘風(fēng)險(xiǎn)識(shí)別規(guī)則提取

仝桂杰1,吳紹華2*,袁毓婕1,顏道浩1,周生路1,李富富1

(1.南京大學(xué)地理與海洋科學(xué)學(xué)院,江蘇 南京 210023；2.浙江財(cái)經(jīng)大學(xué)土地與城鄉(xiāng)發(fā)展研究院,浙江 杭州 310018)

為揭示環(huán)境因素與小麥Cd超標(biāo)風(fēng)險(xiǎn)的關(guān)系,綜合考慮了小麥Cd富集的7個(gè)影響因素(土壤Cd濃度、污染企業(yè)、城鎮(zhèn)村及工礦用地、交通運(yùn)輸用地、土壤類型、土壤有機(jī)質(zhì)含量(SOM)和土壤pH值),采用ID3算法與樸素貝葉斯算法,建立起5棵貝葉斯決策樹(shù).提出了15條小麥Cd超標(biāo)風(fēng)險(xiǎn)的識(shí)別規(guī)則,將超標(biāo)風(fēng)險(xiǎn)分為5級(jí)并確定了小麥Cd富集的3個(gè)主控因子:污染企業(yè)、土壤pH值和土壤Cd濃度.經(jīng)檢驗(yàn),5棵決策樹(shù)風(fēng)險(xiǎn)識(shí)別的平均精度為81.14%,而使用風(fēng)險(xiǎn)識(shí)別規(guī)則和貝葉斯算法后識(shí)別精度提高為89.32%.該模型將貝葉斯算法融入到了決策樹(shù)模型,可以評(píng)估數(shù)據(jù)完整或缺失樣本的Cd污染風(fēng)險(xiǎn),確定小麥Cd富集的主控因子,同時(shí)可以基于風(fēng)險(xiǎn)識(shí)別規(guī)則判定小麥Cd風(fēng)險(xiǎn)程度和范圍,為土壤安全利用和小麥安全生產(chǎn)區(qū)的劃定提供參考.

貝葉斯決策樹(shù)；重金屬；風(fēng)險(xiǎn)；小麥

隨著城市化與工業(yè)化的快速發(fā)展,日益增多的Cd元素通過(guò)農(nóng)業(yè)、工業(yè)和交通途徑[1-3]排放到土壤中.農(nóng)藥、化肥和塑料薄膜的使用,垃圾焚燒、道路揚(yáng)塵、機(jī)動(dòng)車尾氣及非尾氣(輪胎及剎車片磨損)的排放,以及陶瓷工廠和耐火材料廠等企業(yè)的污染最終都匯聚到土壤中[4-9],并隨著作物的生長(zhǎng)發(fā)育遷移到作物體內(nèi).研究表明,與其他重金屬元素相比,Cd在作物體內(nèi)的遷移富集能力很強(qiáng)[10-11],在小麥籽粒和玉米中都呈現(xiàn)出Cd>Cu>Pb的規(guī)律[12-14],在水稻籽粒中則呈現(xiàn)出Cd>Cu、Pb、Ni的規(guī)律[15-16].據(jù)統(tǒng)計(jì),我國(guó)每年約有1′107t的糧食因重金屬污染而減產(chǎn),1.2′107t的糧食已經(jīng)被重金屬污染[17],糧食中的重金屬會(huì)隨著食物網(wǎng)遷移到人體內(nèi)(頭部、骨骼),危害健康[18].

Cd在小麥籽粒內(nèi)遷移富集的過(guò)程較為復(fù)雜,容易受到多種因素的影響[19-21].目前建立的Cd在土壤-作物系統(tǒng)中遷移的模型可以分為經(jīng)驗(yàn)?zāi)Ｐ秃蜋C(jī)理模型.在機(jī)理模型方面,主要采用質(zhì)量平衡和蒙特卡洛方法,并結(jié)合作物的生長(zhǎng)過(guò)程,建立城市土壤與農(nóng)業(yè)土壤的重金屬遷移積累模型[22-23],在經(jīng)驗(yàn)?zāi)Ｐ头矫?綜合考慮土壤重金屬遷移的主控因子(土壤的理化性質(zhì)、重金屬的形態(tài)、二硫酸鹽-檸檬酸鹽提取態(tài)鋁(CD-Al)和二硫酸鹽-檸檬酸鹽提取態(tài)鐵 (CD-Fe)等),揭示其與作物重金屬的定量關(guān)系,并建立土壤-作物的重金屬遷移模型[24-25],ID3(Iterative dichotomiser 3)決策樹(shù)模型[26]也屬于經(jīng)驗(yàn)?zāi)Ｐ?主要用來(lái)研究數(shù)據(jù)的分類和預(yù)測(cè)問(wèn)題,可以識(shí)別土壤重金屬污染的主控因子,并對(duì)污染程度進(jìn)行分級(jí)與評(píng)價(jià)[27-29].

研究表明小麥籽粒中的Cd含量與土壤Cd含量并不是線性關(guān)系,即有些土壤Cd未超標(biāo)而小麥超標(biāo),有些土壤Cd超標(biāo)而小麥不超標(biāo),但目前還沒(méi)有較好的方法精確識(shí)別小麥Cd的超標(biāo)風(fēng)險(xiǎn),本文采用的貝葉斯決策樹(shù)是一種較好的風(fēng)險(xiǎn)識(shí)別方法,將在綜合考慮影響研究區(qū)內(nèi)小麥籽粒中Cd的遷移積累因子基礎(chǔ)上,采用ID3算法與貝葉斯算法建立決策樹(shù)模型并提取風(fēng)險(xiǎn)識(shí)別規(guī)則.該模型既可以依據(jù)風(fēng)險(xiǎn)識(shí)別規(guī)則來(lái)評(píng)估屬性數(shù)據(jù)完整的樣本Cd污染風(fēng)險(xiǎn),而且可以采用貝葉斯算法來(lái)預(yù)測(cè)缺失某些屬性數(shù)據(jù)的樣本Cd污染風(fēng)險(xiǎn);不僅可以確定小麥Cd富集的主控因子,同時(shí)可以判定小麥Cd風(fēng)險(xiǎn)的程度和范圍,為土壤的安全利用和小麥Cd超標(biāo)風(fēng)險(xiǎn)區(qū)的劃定提供參考.

1 材料與方法

1.1 樣品的采集與測(cè)試

本研究選取中國(guó)東部某鄉(xiāng)鎮(zhèn),鎮(zhèn)域內(nèi)化工企業(yè)分布密集,土壤及小麥籽粒中的Cd超標(biāo)情況較為普遍.采用網(wǎng)格布點(diǎn)法(500m′500m),在研究區(qū)采集成熟小麥籽粒樣品178處,并在小麥樣品處采集土壤樣品(圖3).將樣品烘干稱重并粉碎,采用混酸(HNO3-HCLO4)消煮植物樣本,并使用ICP-MS (Agilent 7700X,安捷倫科技有限公司)測(cè)定Cd元素的全量[23].采用三酸消化法 (HF-HNO3-HCLO4)消煮土壤樣品,并使用石墨爐原子吸收法測(cè)定Cd元素的全量[23].

圖2 采樣點(diǎn)分布 Fig.2 Distribution map of sampling points

1.2 貝葉斯決策樹(shù)

1.2.1 基本思想與優(yōu)點(diǎn) 基于ID3算法的貝葉斯決策樹(shù)是一種釆用自上而下的遞歸方式來(lái)進(jìn)行歸納學(xué)習(xí)的算法,在小規(guī)模樣本中的分類精度較高.它的基本思想是在數(shù)據(jù)的預(yù)處理階段,根據(jù)最大后驗(yàn)假定補(bǔ)齊缺失的數(shù)據(jù)屬性,通過(guò)比較各屬性信息增益的方法建立決策樹(shù)并添加貝葉斯節(jié)點(diǎn).相對(duì)于其他分類方法,貝葉斯決策樹(shù)算法的優(yōu)點(diǎn)包括:數(shù)據(jù)預(yù)處理階段補(bǔ)齊了某些樣本缺失的屬性,保證了樣本總體的完整性;構(gòu)建原理較為簡(jiǎn)單,構(gòu)建者不需要了解很多樣本數(shù)據(jù)的相關(guān)知識(shí);訓(xùn)練時(shí)間較少,樹(shù)狀結(jié)構(gòu)容易理解;針對(duì)缺失了某些屬性的樣本,可以運(yùn)用貝葉斯方法較為準(zhǔn)確地預(yù)測(cè)其分類.

1.2.2 決策樹(shù)的構(gòu)建 (1)數(shù)據(jù)預(yù)處理.首先篩選出構(gòu)建決策樹(shù)所需的樣本屬性,并將數(shù)據(jù)分為2部分:訓(xùn)練樣本和檢驗(yàn)樣本.使用貝葉斯方法補(bǔ)齊缺失的訓(xùn)練樣本數(shù)據(jù),并對(duì)數(shù)據(jù)進(jìn)行離散化處理.樸素的貝葉斯方法是使用已知的樣本屬性,將具有最高后驗(yàn)概率的類C賦給未知的樣本屬性[30].假定共有個(gè)數(shù)據(jù)且屬性相互獨(dú)立,其中數(shù)據(jù)已知個(gè)屬性1、2、…、x但屬性未知,則未知屬性屬于C類的概率為:

(2)構(gòu)建決策樹(shù).計(jì)算決策樹(shù)根節(jié)點(diǎn)各屬性的信息增益,選擇最大值所對(duì)應(yīng)的屬性作為根節(jié)點(diǎn)的屬性,再自頂向下遞歸計(jì)算其他子節(jié)點(diǎn)的屬性.設(shè)樣本有個(gè)數(shù)據(jù),葉子節(jié)點(diǎn)有個(gè)不同的分類C(=1, 2,…,);s為C中的樣本數(shù),p是樣本屬于C的概率.設(shè)樣本的某個(gè)屬性具有個(gè)不同的類,根據(jù)屬性可以將樣本劃分為個(gè)子集,s為子集s中類C的樣本數(shù),p=s/s是s中樣本屬于C的概率.則樣本分類所需的期望、根據(jù)屬性劃分子集的熵()、子集s的期望及屬性對(duì)應(yīng)的信息增益()分別為:

在父節(jié)點(diǎn)和子節(jié)點(diǎn)之間添加貝葉斯節(jié)點(diǎn),檢驗(yàn)父節(jié)點(diǎn)的屬性是否可知(Y:已知,N:未知).并在構(gòu)建完決策樹(shù)后,對(duì)其進(jìn)行后期剪枝.

(3)決策樹(shù)的精度檢驗(yàn)

將檢驗(yàn)樣本的數(shù)據(jù)由上至下遍歷決策樹(shù),得到最終的分類結(jié)果.當(dāng)檢驗(yàn)樣本的某個(gè)屬性缺失時(shí),通過(guò)貝葉斯算法選取具有最大后驗(yàn)概率的類作為樣本的分類.將最終的分類與檢驗(yàn)樣本的真實(shí)分類相比較,檢驗(yàn)決策樹(shù)的分類精度.

2 結(jié)果與分析

2.1 小麥籽粒Cd含量影響因子識(shí)別

影響小麥籽粒中Cd含量的因素較多,在Cd的來(lái)源方面,土壤中的Cd濃度對(duì)小麥籽粒中Cd積累的影響最為直接,而陶瓷工廠和耐火材料廠等污染企業(yè),以及城鎮(zhèn)村和工礦用地會(huì)排放大量的廢渣、廢水和廢氣,同時(shí)道路揚(yáng)塵及機(jī)動(dòng)車尾氣中含有豐富的Cd元素[31],對(duì)周邊的土壤和大氣帶來(lái)嚴(yán)重的Cd污染.在Cd的遷移積累方面,不同的土壤類型、有機(jī)質(zhì)及土壤pH值都會(huì)影響土壤中Cd的遷移[32-34].因此本文選取了土壤Cd濃度、距污染企業(yè)的距離、距城鎮(zhèn)村及工礦用地的距離、距交通運(yùn)輸用地的距離、土壤類型、土壤有機(jī)質(zhì)含量和pH值為建樹(shù)的屬性,并將小麥籽粒Cd濃度作為最終的分類屬性.

表1 建樹(shù)屬性的分類及分類精度的評(píng)估 Table 1 Classification of tree attributes and evaluation of classification accuracy

依據(jù)《食品安全國(guó)家標(biāo)準(zhǔn)食品中污染物限量》(GB 2762-2017)[35]及《土壤環(huán)境質(zhì)量農(nóng)用地土壤污染風(fēng)險(xiǎn)管控標(biāo)準(zhǔn)(試行)》(GB 15618-2018)[36],并結(jié)合樣本各屬性數(shù)值的分布特征將樣本各屬性分為有風(fēng)險(xiǎn)和無(wú)風(fēng)險(xiǎn)2類,盡量使得2類的數(shù)量差別不大,避免因某一分支的樣本過(guò)少而被剪枝.統(tǒng)計(jì)每個(gè)屬性的每個(gè)分類的樣本數(shù)量,并統(tǒng)計(jì)各個(gè)屬性有風(fēng)險(xiǎn)的樣本中的籽粒超標(biāo)率,評(píng)估分類的準(zhǔn)確性(表1).經(jīng)統(tǒng)計(jì),有風(fēng)險(xiǎn)樣本中籽粒的平均超標(biāo)率為91%,分類可以較為準(zhǔn)確對(duì)應(yīng)籽粒的超標(biāo)情況.

2.2 貝葉斯決策樹(shù)的構(gòu)建

2.2.1 樣本的分組及預(yù)處理 將178個(gè)樣本按照4:1的數(shù)量分配原則分為2組:訓(xùn)練樣本(143)與檢驗(yàn)樣本(35).由于178個(gè)樣本在采樣時(shí)是從研究區(qū)西北到東南順序編號(hào),為了保證檢驗(yàn)樣本在研究區(qū)內(nèi)均勻分布,降低決策樹(shù)模型的不確定性,因此檢驗(yàn)樣本采用等間隔(5)取樣法.將178個(gè)樣本分別分為5組訓(xùn)練樣本和檢驗(yàn)樣本(表2).依次檢查訓(xùn)練樣本的數(shù)據(jù)屬性完整性,使用貝葉斯算法將具有二義性或缺失的屬性數(shù)據(jù)補(bǔ)齊.

2.2.2 決策樹(shù)的建立 根據(jù)5組訓(xùn)練樣本可以建立5棵決策樹(shù),依次編號(hào)為決策樹(shù)1~5.以第2組訓(xùn)練樣本為例,計(jì)算其7個(gè)屬性的信息增益,選取具有最大信息增益的屬性作為決策樹(shù)的根節(jié)點(diǎn).經(jīng)計(jì)算,屬性“距污染企業(yè)距離”的信息增益最高,因此將其作為決策樹(shù)2的根節(jié)點(diǎn),并在根節(jié)點(diǎn)下添加貝葉斯節(jié)點(diǎn).依次向下計(jì)算,構(gòu)建出整棵決策樹(shù).

表2 訓(xùn)練樣本與檢驗(yàn)樣本的劃分 Table 2 Division of training samples and test samples

表3 決策樹(shù)2根節(jié)點(diǎn)各個(gè)屬性的信息增益 Table 3 Information gain of each attribute in decision tree 2root node

圖4 貝葉斯決策樹(shù)2 Fig.4 Bayesian decision tree 2

如圖5所示,貝葉斯節(jié)點(diǎn)上的Y代表樣本在該節(jié)點(diǎn)的屬性數(shù)據(jù)已知,N代表未知.葉子節(jié)點(diǎn)上的F表示小麥籽粒有風(fēng)險(xiǎn),T表示無(wú)風(fēng)險(xiǎn).另外4棵決策樹(shù)的簡(jiǎn)圖如圖5所示,簡(jiǎn)圖省略了貝葉斯節(jié)點(diǎn)與左右分支上的的屬性數(shù)值,各節(jié)點(diǎn)字母的含義分別為:(距污染企業(yè))、Cd(土壤Cd濃度)、pH值(土壤pH值)、SOM、(距城鎮(zhèn)村及工礦用地)、(距交通運(yùn)輸用地)、(土壤類型).

圖5 貝葉斯決策樹(shù) Fig.5 Simple graph of Bayesian decision tree

2.3 Cd風(fēng)險(xiǎn)識(shí)別規(guī)則提取

在貝葉斯決策樹(shù)中,從根節(jié)點(diǎn)到葉子節(jié)點(diǎn)的每一條路徑都對(duì)應(yīng)著一條分類規(guī)則,5棵決策樹(shù)模型中共有86條分類規(guī)則.為了提高風(fēng)險(xiǎn)區(qū)識(shí)別結(jié)果的可靠性,將5棵決策樹(shù)中出現(xiàn)次數(shù)較多的分類規(guī)則(32次)提取出來(lái),并將識(shí)別范圍重復(fù)的規(guī)則去掉,最終得到15條小麥Cd超標(biāo)風(fēng)險(xiǎn)識(shí)別規(guī)則,將小麥Cd超標(biāo)風(fēng)險(xiǎn)分為了5級(jí)(表4),五級(jí)的風(fēng)險(xiǎn)最高,一級(jí)的風(fēng)險(xiǎn)最低,其中三、四、五級(jí)表示小麥籽粒Cd超標(biāo),一級(jí)和二級(jí)風(fēng)險(xiǎn)表示小麥籽粒Cd未超標(biāo),可以進(jìn)行小麥的安全生產(chǎn).

計(jì)算樣本總體的各個(gè)屬性信息增益,可以得到研究區(qū)小麥Cd超標(biāo)的主控因子.其中“距污染企業(yè)”、“土壤pH值”和“土壤Cd濃度”的信息增益明顯高于其他屬性,即這3個(gè)屬性對(duì)該研究區(qū)的小麥Cd超標(biāo)風(fēng)險(xiǎn)識(shí)別的貢獻(xiàn)最大,為主控因子,在進(jìn)行風(fēng)險(xiǎn)識(shí)別時(shí)應(yīng)優(yōu)先考慮.

表4 小麥Cd超標(biāo)風(fēng)險(xiǎn)識(shí)別規(guī)則表 Table 4 Risk identification rules table for excessive Cd in wheat

續(xù)表4

序號(hào)距污染企業(yè)(m)土壤pH值土壤Cd濃度(mg/kg)距城鎮(zhèn)村及工礦用地(m)距交通運(yùn)輸用地(m)土壤類型(是否水稻土)SOM(%)次數(shù)等級(jí) <8003800<636>0.3￡0.3<80380<40340否是<2.532.5 5√√√√2三 6√√√√√2三 7√√√√√√√2三 8√√√√√√2三 9√2三 10√√√√√√√2二 11√√√√√√√2二 12√√√√√√2二 13√√√√√2二 14√√√√2二 15√√√√√√√3一

表5 樣本總體各屬性的信息增益

15條識(shí)別規(guī)則并不是從同一棵決策樹(shù)中提取,因此必然存在規(guī)則的遺漏,即有些樣點(diǎn)并不能根據(jù)這些規(guī)則進(jìn)行識(shí)別,或者樣點(diǎn)存在屬性數(shù)據(jù)的缺失,無(wú)法適用該規(guī)則.因此針對(duì)某一未知風(fēng)險(xiǎn)的樣點(diǎn),首先遍歷該規(guī)則,若不可以判斷其污染風(fēng)險(xiǎn),則使用貝葉斯方法分別計(jì)算該樣點(diǎn)有風(fēng)險(xiǎn)和無(wú)風(fēng)險(xiǎn)的最大后驗(yàn)概率,選取概率最大的選項(xiàng)作為最終的風(fēng)險(xiǎn)分類.

2.4 精度檢驗(yàn)與不確定性分析

比較決策樹(shù)35個(gè)檢驗(yàn)樣本的預(yù)測(cè)值與實(shí)際值,5棵決策樹(shù)的準(zhǔn)確度均高于70%,平均精度為81.14%,精度標(biāo)準(zhǔn)差為4.33%,表明該貝葉斯決策樹(shù)模型較為穩(wěn)定,可以比較準(zhǔn)確的預(yù)測(cè)小麥Cd超標(biāo)風(fēng)險(xiǎn).

表6 決策樹(shù)預(yù)測(cè)精度

從5棵貝葉斯決策樹(shù)提取出現(xiàn)次數(shù)較多的15條風(fēng)險(xiǎn)識(shí)別規(guī)則,作為研究區(qū)最終的小麥Cd超標(biāo)風(fēng)險(xiǎn)識(shí)別規(guī)則.使用該風(fēng)險(xiǎn)識(shí)別規(guī)則判斷所有樣本的Cd超標(biāo)情況,規(guī)則未覆蓋到的樣本則采用貝葉斯算法.經(jīng)檢驗(yàn),86.51%的樣本都可以使用風(fēng)險(xiǎn)識(shí)別規(guī)則,識(shí)別精確度為91.56%,樣本總體的識(shí)別精度為89.32%,明顯高于僅使用單棵決策樹(shù)的預(yù)測(cè)精度,表明綜合了5棵決策樹(shù)的風(fēng)險(xiǎn)識(shí)別規(guī)則可以更精確地識(shí)別小麥Cd超標(biāo)風(fēng)險(xiǎn).

表7 基于風(fēng)險(xiǎn)識(shí)別規(guī)則的決策樹(shù)預(yù)測(cè)精度

3 討論

本文建立了5棵貝葉斯決策樹(shù)模型,提取出15條識(shí)別小麥籽粒Cd超標(biāo)風(fēng)險(xiǎn)的規(guī)則,將研究區(qū)內(nèi)的小麥Cd超標(biāo)風(fēng)險(xiǎn)劃分為五級(jí),并確定了Cd超標(biāo)的主控環(huán)境因子.模型在數(shù)據(jù)的預(yù)處理階段采用貝葉斯算法保證了數(shù)據(jù)的完整性,在建樹(shù)階段增加貝葉斯節(jié)點(diǎn)解決了樣本屬性數(shù)據(jù)缺失時(shí)的風(fēng)險(xiǎn)識(shí)別問(wèn)題.總體來(lái)說(shuō),貝葉斯決策樹(shù)模型較為穩(wěn)定,提取的識(shí)別規(guī)則簡(jiǎn)潔易懂,可以較為準(zhǔn)確地識(shí)別研究區(qū)小麥Cd的超標(biāo)風(fēng)險(xiǎn).

此前的研究在建立決策樹(shù)前的數(shù)據(jù)預(yù)處理中,會(huì)將部分指標(biāo)缺失(>30%)的數(shù)據(jù)剔除[37],而本研究則采用樸素貝葉斯算法補(bǔ)齊缺失的數(shù)據(jù),保證了數(shù)據(jù)的完整性,在對(duì)土壤重金屬污染的程度進(jìn)行識(shí)別時(shí),許多研究使用單一決策樹(shù)進(jìn)行判別[28,38],這存在著訓(xùn)練樣本選取的偶然性,而本研究則將5棵決策樹(shù)融合,提高了決策樹(shù)模型在預(yù)測(cè)時(shí)的穩(wěn)定性,在研究土壤-作物系統(tǒng)中重金屬的遷移時(shí),許多機(jī)理模型和經(jīng)驗(yàn)?zāi)Ｐ投加休^好的精度,但它們都是針對(duì)連續(xù)型的土壤屬性數(shù)據(jù),例如土壤中重金屬含量、有機(jī)質(zhì)含量和土壤pH值等[11,22-25],而決策樹(shù)模型則可以兼顧連續(xù)型數(shù)據(jù)和離散型數(shù)據(jù),將更多影響因素考慮在內(nèi)(土壤類型、土壤質(zhì)地、地質(zhì)類型等),許多研究在研究土壤和作物重金屬污染的主控因子時(shí),大多歸結(jié)于土壤重金屬含量、土壤pH值和土壤電導(dǎo)率[11,39],但實(shí)際上人類活動(dòng)也會(huì)影響重金屬的富集[40],本研究區(qū)83.6%的土壤Cd含量未超標(biāo)但小麥籽粒存在Cd超標(biāo)的風(fēng)險(xiǎn),這表明有其他的因素影響小麥Cd的富集,研究結(jié)果表明研究區(qū)內(nèi)污染企業(yè)是作物Cd污染的主導(dǎo)因子,“土壤pH值”和“土壤Cd濃度”次之,人類活動(dòng)是導(dǎo)致該地區(qū)小麥Cd超標(biāo)的主要因素.

總體來(lái)說(shuō),本研究綜合考慮了自然因素與人類活動(dòng),并基于貝葉斯決策樹(shù)提取了識(shí)別規(guī)則,揭示了研究區(qū)內(nèi)小麥Cd超標(biāo)的主控因子,較為精確的識(shí)別小麥Cd超標(biāo)風(fēng)險(xiǎn)的程度,為小麥安全生產(chǎn)區(qū)的劃定提供理論依據(jù)與科學(xué)參考.

4 結(jié)論

4.4 基于ID3算法與貝葉斯算法,建立了多棵貝葉斯決策樹(shù),并整合提取出15條風(fēng)險(xiǎn)識(shí)別規(guī)則,風(fēng)險(xiǎn)識(shí)別精度為89.32%,明顯高于單棵決策樹(shù)模型的預(yù)測(cè)精度.

4.2 “距污染企業(yè)”、“土壤pH”和“土壤Cd濃度”為研究區(qū)小麥Cd超標(biāo)的主控因子.土壤Cd濃度并不是識(shí)別小麥Cd超標(biāo)的決定性因素,人類活動(dòng)對(duì)小麥Cd富集的貢獻(xiàn)不可忽視,是造成小麥在Cd的低地質(zhì)背景區(qū)中超標(biāo)而在高地質(zhì)背景區(qū)不超標(biāo)的現(xiàn)象的重要原因.

[1] Hong Y, Bonhomme C, Van den Bout B, et al. Integrating atmospheric deposition, soil erosion and sewer transport models to assess the transfer of traffic-related pollutants in urban areas [J]. Environmental Modelling & Software, 2017,96:158-171.

[2] Wei B G, Yang L S. A review of heavy metal contaminations in urban soils, urban road dusts and agricultural soils from China [J]. Microchemical Journal, 2010,94(2):99-107.

[3] Duruibe J O, Ogwuegbu M O C. Heavy metal pollution and human biotoxic effects [J]. International Journal of Physical Sciences, 2007,2(5):112-118.

[4] Chen L, Zhou S, Shi Y, et al. Heavy metals in food crops, soil, and water in the Lihe River Watershed of the Taihu Region and their potential health risks when ingested [J]. Science of the Total Environment, 2018,615:141-149.

[5] 武媛媛,李如梅,彭 林,等.運(yùn)城市道路揚(yáng)塵化學(xué)組成特征及來(lái)源分析[J]. 環(huán)境科學(xué), 2017,38(5):1799-1806.Wu Yuanyuan, Li Rumei, Peng Lin, et al. Chemical Compositions and Source Apportionment of Road Dust in Yuncheng [J]. Environmental Science, 2017,38(5):1799-1806.

[6] Liu G, Li J H, Wu D, et al. Chemical composition and source apportionment of the ambient PM2.5in Hangzhou, China [J]. Particuology, 2015,18:135-143.

[7] Grigoratos T, Martini G. Brake wear particle emissions: a review [J]. Environmental Science and Pollution Research, 2015,22(4):2491- 2504.

[8] 樊 霆,葉文玲,陳海燕,等.農(nóng)田土壤重金屬污染狀況及修復(fù)技術(shù)研究[J]. 生態(tài)環(huán)境學(xué)報(bào), 2013,(10):1727-1736.Fan Ting, Ye Wenling, Chen Haiyan, et al. Review on contamination and remediation technology of heavy metal in agricultural soil [J]. Ecology and Environment Sciences, 2013,(10):1727-1736.

[9] 許 悅,王 可,劉雪梅,等.室內(nèi)外PM2.5中金屬元素的污染特征及來(lái)源[J]. 中國(guó)環(huán)境科學(xué), 2018,38(4):1257-1264.Xu Yue, Wang Ke, Liu Xuemei, et al. Pollution characteristics and sources of metal elements in indoor and outdoor PM2.5[J]. China Environmental Science, 2018,38(4):1257-1264.

[10] 楊 賀,劉 杰.礦業(yè)生產(chǎn)影響區(qū)水稻田系統(tǒng)砷、鉛、鎘的污染特征及風(fēng)險(xiǎn)評(píng)價(jià)——以西江流域大環(huán)江下游為例[J]. 江蘇農(nóng)業(yè)科學(xué), 2018,46(15):205-209.Yang He, Liu Jie. Pollution characteristics and risk assessment of arsenic, lead and cadmium in paddy field systems in mining production affected areas: A case study of the lower reaches of the greater Huanjiang River in Xijiang River Basin [J]. Jiangsu Agricultural Sciences, 2018,46(15):205-209.

[11] 王世玉,吳文勇,劉 菲,等.典型污灌區(qū)土壤與作物中重金屬健康風(fēng)險(xiǎn)評(píng)估[J]. 中國(guó)環(huán)境科學(xué), 2018,38(4):1550-1560.Wang Shiyu, Wu Wenyong, Liu Fei, et al. Assessment of human health risks of heavy metals in the typical sewage irrigation areas [J]. China Environmental Science, 2018,38(4):1550-1560.

[12] Zhang Y C, Yin S, Cao L, et al. Heavy metal accumulation and health risk assessment in soil-wheat system under different nitrogen levels[J]. Science of the Total Environment, 2018,622-623:1499.

[13] Wang S, Wu W, Liu F, et al. Accumulation of heavy metals in soil-crop systems: a review for wheat and corn [J]. Environmental Science & Pollution Research International, 2017,24(2):1-17.

[14] Ali A M, Ibrahim S M, El-Hady Y A A, et al. Assessment of bioavailability of some heavy metals to wheat and faba bean in Sahl El-Tina, Egypt [J]. Agricultural Research, 2018,7(1):72-82.

[15] Xie L, Hao P, Cheng Y, et al. Effect of combined application of lead, cadmium, chromium and copper on grain, leaf and stem heavy metal contents at different growth stages in rice [J]. Ecotoxicology & Environmental Safety, 2018,162:71-76.

[16] Ogunkunle C O, Varun M, Jimoh M A, et al. Evaluating the trace metal pollution of an urban paddy soil and bioaccumulation in rice (Oryza sativ a L.) with the associated dietary risks to local population: a case study of Ilorin, north-central Nigeria [J]. Environmental Earth Sciences, 2016,75(20):1383-1394.

[17] 羅錫文.對(duì)加速我國(guó)農(nóng)業(yè)機(jī)械化發(fā)展的思考[J]. 農(nóng)業(yè)工程, 2011,1(4):1-8,56.Luo Xiwen. Thoughts on accelerating the development of agricultural mechanization in China [J]. Agricultural Engineering, 2011,1(4):1-8, 56.

[18] Gall J E, Boyd R S, Rajakaruna N. Transfer of heavy metals through terrestrial food webs: a review [J]. Environmental Monitoring & Assessment, 2015,187(4):201.

[19] 趙瑞瑞,石福臣,周美利,等.雪松和圓柏對(duì)城鄉(xiāng)大氣環(huán)境中重金屬的積累特征[J]. 生態(tài)學(xué)雜志, 2015,34(12):3368-3373.Zhao Ruirui, Shi Fuchen, Zhou Meili, et al. Effects of urban-rural atmospheric environment on heavy metal accumulation ofand[J]. Chinese Journal of Ecology, 2015, 34(12):3368-3373.

[20] Colpaert J V, Wevers J H L, Krznaric E, et al. How metal-tolerant ecotypes of ectomycorrhizal fungi protect plants from heavy metal pollution [J]. Annals of Forest Science, 2011,68(1):17-24.

[21] Sardans J, Pe?uelas J, Estiarte M. Warming and drought change trace element bioaccumulation patterns in a Mediterranean shrubland [J]. Chemosphere, 2008,70(5):874-885.

[22] 彭 馳.城市土壤重金屬與多環(huán)芳烴累積過(guò)程模擬[D]. 北京:中國(guó)科學(xué)院生態(tài)環(huán)境研究中心, 2013.Peng Chi. Modeling heavy metals and polycyclic aromatic hydrocarbons accumulation in urban soil [D]. Beijing: Research Center for Eco-Environmental Sciences, Chinese Academy of Sciences, 2013.

[23] 施亞星,吳紹華,周生路,等.土壤-作物系統(tǒng)中重金屬元素吸收、遷移和積累過(guò)程模擬[J]. 環(huán)境科學(xué), 2016,37(10):3996-4003.Shi Yaxing, Wu Shaohua, Zhou Shenglu, et al. Simulation of the absorption, migration and accumulation process of heavy metal elements in soil-crop system [J]. Environmental Science, 2016, 37(10):3996-4003.

[24] 丁 瓊.土壤性質(zhì)及鈍化劑對(duì)鎘在土壤—植物系統(tǒng)轉(zhuǎn)移的影響[D]. 北京:首都師范大學(xué), 2012.Ding Qiong. The effects of soil properties and soil amendments on the transfer of Cadmium in soil-plant system [D]. Beijing: Capital Normal University, 2012

[25] 趙科理.土壤-水稻系統(tǒng)重金屬空間對(duì)應(yīng)關(guān)系和定量模型研究[D]. 杭州:浙江大學(xué), 2010.Zhao Keli. Spatial relationships of heacy metals in soil-rice system and the quantitative model [D]. Hangzhou: Zhejiang University, 2010.

[26] JR Q. Introduction of decision trees [J]. Machine Learning, 1986, (1):81-107.

[27] 成 偉.數(shù)據(jù)挖掘技術(shù)支持下的土壤重金屬污染評(píng)價(jià)系統(tǒng)的研究[D]. 杭州:浙江大學(xué), 2009.Cheng Wei. A study of soil heavy metal pollution assessment system supported by data mining technology [D]. Hangzhou: Zhejiang University, 2009.

[28] 王冠星,閆學(xué)東,張 凡,等.青藏高原路側(cè)土壤重金屬含量分布規(guī)律及影響因素研究[J]. 環(huán)境科學(xué)學(xué)報(bào), 2014,34(2):431-438.Wang Guanxing, Yan Xuedong, Zhang Fan, et al. Influencing factors of heavy metal concentration in roadside-soil of Qinghai-Tibet Plateau [J]. Acta Scientiae Circumstantiae, 2014,34(2):431-438.

[29] 邱樂(lè)豐.富陽(yáng)市土地生產(chǎn)力時(shí)空變化研究[D]. 杭州:浙江大學(xué), 2012.Qiu Lefeng. Spatial and temporal variation of land productivity in Fuyang County [D]. Hangzhou: Zhejiang University, 2012.

[30] 朱 軍,胡文波.貝葉斯機(jī)器學(xué)習(xí)前沿進(jìn)展綜述[J]. 計(jì)算機(jī)研究與發(fā)展, 2015,52(1):16-26.Zhu Jun, Hu Wenbo. Recent sdvances in bayesian machine learning [J]. Journal of Computer Research and Development, 2015,52(1):16-26.

[31] 張 靜,張衍杰,方小珍,等.道路揚(yáng)塵PM2.5中金屬元素污染特征及健康風(fēng)險(xiǎn)評(píng)價(jià)[J]. 環(huán)境科學(xué), 2017,38(10):4071-4076.Zhang Jing, Zhan Yanjie, Fang Xiaozhen, et al. Characteristics and health risk assessment of metallic elements in PM2.5fraction of road dust [J]. Environmental Science, 2017,38(10):4071-4076.

[32] 陳 珊,劉 妍,陳 齊,等.添加造紙污泥對(duì)土壤-白蘿卜系統(tǒng)中鎘遷移的影響[J]. 環(huán)境工程學(xué)報(bào), 2017,11(3):1906-1912.Chen Shan, Liu Yan, Chen Qi, et al. Effect of applied paper mill sludge on migration of Cd in soil-white radish system [J]. Chinese Journal of Environmental Engineering, 2017,11(3):1906-1912.

[33] 胡 鑫,羅真華,謝會(huì)雅,等.土壤性質(zhì)對(duì)煙葉中鉛、鎘含量的影響及預(yù)測(cè)模型研究[J]. 農(nóng)業(yè)環(huán)境科學(xué)學(xué)報(bào), 2016,35(3):449-454.Hu Xin, Luo Zhenhua, Xie Huiya, et al. Soil property effects on and prediction models of lead and cadmium concentrations in tobacco leaves [J]. Journal of Agro-Environmental Science, 2016,35(3):449- 454.

[34] 何 俊,王學(xué)東,陳世寶,等.典型污灌區(qū)土壤中Cd的形態(tài)、有效性及其影響因子[J]. 中國(guó)環(huán)境科學(xué), 2016,36(10):3056-3063.He Jun, Wang Xuedong, Chen Shibao, et al. The forms, bioavailability of Cd in soils of typical sewage irrigation fields in northern China and its control factors [J]. China Environmental Science, 2016,36(10): 3056-3063.

[35] GB2762－2017食品安全國(guó)家標(biāo)準(zhǔn)食品中污染物限量標(biāo)準(zhǔn) [S].GB 2762－2017 National food safety standard Contaminant limit standard in food [S].

[36] 生態(tài)環(huán)境部,國(guó)家市場(chǎng)監(jiān)督管理總局.GB 15618－2018土壤環(huán)境質(zhì)量農(nóng)用地土壤污染風(fēng)險(xiǎn)管控標(biāo)準(zhǔn)(試行) [S]. 北京:中國(guó)環(huán)境出版社, 2018. Ministry of Ecological Environment of the People's Republic of China, State Administration for Market regulation. GB 15618－2018Soil environment quality Risk control standard for soil contamination of agricultural land [S]. Beijing: China Environmental Science Press, 2018.

[37] 史 琦,陳建新,趙慧輝,等.冠心病患者痰熱互結(jié)證CHAID決策樹(shù)識(shí)別模式的研究[J]. 世界中醫(yī)藥, 2018,13(9):2095-2101.Shi Qi, Chen Jianxin, Zhao Huihui, et al. The forms, bioavailability of Cd in soils of typical sewage irrigation fields in northern China and its control factors [J]. China Environmental Science, 2018,13(9):2095- 2101.

[38] 焦夢(mèng)姝,張 麒,彭佳紅.基于決策樹(shù)技術(shù)的土壤重金屬污染預(yù)警分析研究[J]. 電腦知識(shí)與技術(shù), 2014,10(26):6210-6213.Jiao Mengshu, Zhang Qi, Peng Jiahong. Early warning analysis of heavy metal pollution in soil based on decision tree technology [J]. Computer Knowledge and Technology, 2014,10(26):6210-6213.

[39] 高智群,張美劍,趙科理,等.土壤—水稻系統(tǒng)重金屬空間異質(zhì)性研究——以浙江省嵊州市為例[J]. 中國(guó)環(huán)境科學(xué), 2016,36(1):215-224.Gao Zhiqun, Zhang Meijian, Zhao Keli, et al. Heavy metal contamination in soil-rice system and its spatial variation in Shengzhou City [J]. China Environmental Science, 2016,36(1):215- 224.

[40] 姜鳳成,李義連,楊 森,等.秦王川盆地土壤重金屬來(lái)源、分布及污染評(píng)價(jià)[J]. 中國(guó)環(huán)境科學(xué), 2018,38(6):2243-2252.Jiang Fengchen, Li Yilian, Yang Sen, et al. Source analysis, distribution and pollution assessment of the soil heavy metals in the Qinwangchuan Basin [J]. China Environmental Science, 2018,38(6): 2243-2252.

致謝：本實(shí)驗(yàn)的現(xiàn)場(chǎng)采集工作由陳蓮和施亞星等協(xié)助完成,在此表示感謝.

Identification rules of wheat Cd risk based on Bayesian decision tree.

TONG Gui-jie1, WU Shao-hua2*, YUAN Yu-jie1, YAN Dao-hao1, ZHOU Sheng-lu1, LI Fu-fu1

(1.School of Geographic and Oceanographic Sciences, Nanjing University, Nanjing 210023, China；2.Institute of Land and Urban-rural Development, Zhejiang University of Finance & Economics, Hangzhou 310018, China).2019,39(3)：1336~1344

In order to reveal the relationship between the environmental factors and the risk of wheat excessive Cd, seven factors (concentration of Cd in soil, polluting enterprises, the town and industrial land, transportation land, soil type, SOM and soil pH) are considered, and five Bayesian decision trees were established based on ID3algorithm and Naive Bayesian algorithm. 15 identification rules of wheat Cd pollution risk were extracted, and the risk was divided into five levels. Polluting enterprises, soil pH and concentration of Cd in soil were the three dominant factors of wheat Cd enrichment. According to the validation, the average prediction accuracy was 81.14%, and the overall recognition accuracy was improved to 89.32% after using the Bayesian algorithm and risk identification rules. The model integrated the Bayesian algorithm into the decision tree model, which could assess the Cd pollution risk in samples with complete or missing data, determine the dominant factors of wheat Cd enrichment, and identify the degree and region of wheat Cd pollution based on the risk identification rules. This approach could provide a scientific tool for soil safety use and the delimitation of wheat safety production area.

Bayesian decision tree；heavy metals；risk；crop

X826

A

1000-6923(2019)03-1336-09

仝桂杰(1995-),男,山東日照人,南京大學(xué)碩士研究生,主要從事土壤-作物系統(tǒng)中重金屬的遷移積累模擬的研究.發(fā)表論文1篇.

2018-08-23

國(guó)家重點(diǎn)研發(fā)計(jì)劃(2017YFD0800305)

* 責(zé)任作者, 教授, wsh@nju.edu.cn