夏 楠,塔西甫拉提.特依拜,丁建麗,依力亞斯江.努爾麥麥提,張 東,劉 芳
(1.新疆大學資源與環境科學學院,烏魯木齊 830046;2.新疆大學綠洲生態教育部重點實驗室,烏魯木齊830046)
基于多光譜數據的荒漠礦區土壤有機質估算模型
夏 楠,塔西甫拉提.特依拜※,丁建麗,依力亞斯江.努爾麥麥提,張 東,劉 芳
(1.新疆大學資源與環境科學學院,烏魯木齊 830046;2.新疆大學綠洲生態教育部重點實驗室,烏魯木齊830046)
目前運用高光譜數據估算土壤有機質的模型精度已經可以達到精準農業的要求,但其數據的整理和運算過程較為復雜且觀測尺度較小。為節省資源,提高效率并為多光譜遙感估算土壤有機質積累經驗,該文將Landsat8_OLI多光譜遙感影像各波段的反射率數據與地面土壤有機質SOM(soil organic matter)實測數據相結合,利用SPSS軟件及多元線性回歸分析方法建立基于反射率R、反射率倒數1/R、反射率倒數對數LN(1/R)、反射率一階導數FDR(first derivative reflectance)的土壤有機質定量估算模型,精度檢驗后擇取最優模型通過多光譜遙感波段運算的方式推廣至整個研究區。結果表明:FDR模型的精度更高,RMSE為0.215,F檢驗結果為4.072,預測值與實際值之間的決定系數R2為0.963。基于該模型估算研究區空間范圍的土壤有機質含量,得出土壤有機質含量在0~5 g/kg之間的面積占總研究區的84.065%,>10 g/kg的面積僅僅為0.001 5%。在4種土地類型中工礦用地SOM平均含量為最高的7.35 g/kg,受開采的煤炭中有機質影響較大。裸地面積2 674.44 km2,占研究區面積的63%,SOM平均含量6.12 g/kg;鹽漬地和荒漠林地SOM含量偏低。總之,運用多光譜遙感數據估算干旱區土壤有機質的方法可行,也為遙感估算其他地表參數提供參考。
土壤;遙感;光譜分析;荒漠;SOM;建模;多光譜;估算
夏 楠,塔西甫拉提.特依拜,丁建麗,依力亞斯江.努爾麥麥提,張 東,劉 芳.基于多光譜數據的荒漠礦區土壤有機質估算模型[J].農業工程學報,2016,32(6):263-267.doi:10.11975/j.issn.1002-6819.2016.06.036 http://www.tcsae.org
Xia Nan,Tashpolat.Tiyip,Ding Jianli,Ilyas Nurmemet,Zhang Dong,Liu Fang.Estimation model of soil organic matter in desert mining area based on multispectral image data[J].Transactions of the Chinese Society of Agricultural Engineering(Transactions of the CSAE),2016,32(6):263-267.(in Chinese with English abstract) doi:10.11975/j.issn.1002-6819.2016.06.036 http://www. tcsae.org
土壤有機質SOM(soil organic matter)是土壤的重要組成部分,它提供著植物生長所必須的碳元素,其含量的多少是衡量土壤肥力的一項重要指標[1]。而土壤肥力通過影響植物的生長從而影響物種的多樣性和生態系統的穩定。因此在農業、林業以及維持生態平衡上獲取土壤有機質含量的意義極其重要。運用傳統方法測得土壤有機質雖然具有很高的精度,但是在選點、采樣、試驗過程中都將耗費大量財力人力,這就需要一種便捷、準確的而且能宏觀運用于研究工作的技術。運用遙感手段測定土壤有機質一直倍受學術界研究者們關注,它的優勢在于依托較少的時間和物力資源而獲取大范圍的、較為精確的地物反射光譜信息,通過各種應用模型進行人們所需地理信息的表達[2]。
如今遙感技術已經成為一種重要的手段被應用于測定土壤有機質含量,國內外學者運用不同遙感數據進行SOM反演都取得了一定效果[3-8]。Al-Abbas等[9]得出了土壤的有機質含量與其光譜反射率之間存在顯著的負相關關系。侯艷軍等[10]提出在土壤有機質的高光譜建模上多元線性回歸模型精度高于一元線性回歸模型。田永超等[11]通過應用熱紅外光聲光譜技術估測土壤有機質含量得到經過一階導數濾波平滑后的光譜建模精度較高。這些學者普遍基于高光譜數據進行研究,也有一些學者運用多光譜數據建模。劉煥軍等[12]通過對實測數據與多光譜影像數據建立相關性提取出相關波段進行建模,其建模后的絕對系數R2為0.665,均方根誤差RMSE為0.553,并且得出土壤含水量的變化會影響估算結果;張法升等[13]發現TM(thematic mapper)影像中TM3、TM5波段的DN(digital number)值與SOM含量之間滿足二次多項式回歸關系。其中一些學者[7,10-11,14]運用偏最小二乘法建模,雖然可以達到較好的模型精度要求,但計算量大,非數學類和軟件類專業的學者運用比較吃力,而運用多光譜影像的光譜信息數據不僅可以建立精度較高的多元線性回歸模型,而且能將模型通過影像可視化地表達,從而進行宏觀的SOM含量時空分布規律的分析。因此,多光譜數據建模估算地表的SOM含量是更加普適、便捷、高效的手段。
綜上所述,本文中作者運用Landsat8_OLI多光譜遙感影像數據和實地土壤有機質含量測定數據建立多元線性回歸模型,并進行精度驗證和模型的預測,從而得到能夠較為精確反演研究區土壤有機質含量的數學模型。基于建立的模型進行研究區土壤有機質的空間分布特征分析,為五彩灣礦區及其周邊生態環境修復提供數據支撐以及為整個準噶爾東部經濟開發區的生態環境規劃建設提供參考。
五彩灣礦區位于新疆準噶爾盆地東部,吉木薩爾縣境內,喀拉麥里山的山前地帶,煤田面積901.05 km2。地貌為戈壁灘平原,地形平坦開闊,其工業基地范圍內平均海拔在500~700 m之間,總體地勢北高南地,如圖1所示。該地區屬于大陸暖溫帶干旱氣候,年平均蒸發量2 090.4 mm,年平均降水量159.1 mm,年平均相對濕度為57%,年平均日照時間為2 861.1 h。主要以荒漠堿土、石膏棕模土和荒漠風沙土為主的土壤類型,表層SOM比不足2%[15]。植被類型主要是琵琶柴、蛇麻黃、白刺、駱駝刺等耐旱植被。

圖1 研究區地理位置圖及采樣點分布圖Fig.1 Geographical position map of study area and distribution of sampling points
2.1 影像數據
研究所用的 Landsat8_OLI數據免費下載于 http:// glovis.usgs.gov/,選取2014年5月份的影像,其分辨率為30 m,云量0,地圖投影WGS84坐標投影,衛星軌道號141-29。在ENVI5.1軟件下進行圖像的裁剪,輻射定標以及大氣校正。在獲取影像的光譜信息時會受到大氣中的水汽、分子和氣溶膠影響產生波段噪聲和信息模糊,使用FLAASH大氣校正,校正由于漫反射引起的連帶效應,很好地消除這些噪聲使信息清晰,降低鄰近像元之間的輻射干擾,也可調整由于人為抑止而導致的波譜平滑[16]。
2.2 土壤數據
2014年5月,在五彩灣礦區周邊選取45個采樣點收集土壤樣本并用GPS記錄其坐標,按照0~10、10~20、20~30 cm共3個土層采樣并用事先稱重過的鋁盒在各層取一定的土樣。將收集的土壤樣本帶回實驗室自然風干,磨碎過20目篩后,采用重鉻酸鉀容量法[17]對其進行SOM含量測定;將鋁盒帶土一并稱重后,放入烘干箱烘干24 h,再次稱取重量,通過計算烘干前后的重量差得到土壤含水率數據。整理數據并計算出每個樣點3個土層的土壤有機質平均值,與經預處理后的遙感影像一同導入ArcGIS軟件,運用軟件的Extraction工具得到每個采樣點所對應各個波段的DN值。
3.1 試驗數據整理
經由大氣校正后的遙感影像的像元DN值為反射率值,范圍0~1。將各建模樣點(30個)的SOM含量實測值、土壤含水率與影像各個波段一一對應,再按SOM含量的大小升序排列得到表1。表中除26~30號點外,其余各點SOM含量均不足5 g/kg,1號點更是不足1 g/kg。土壤含水率平均值3%,最大值在20號點的14.04%,最小值在28號點為0.44%。在極度干旱的荒漠地區,土壤含水率極低,由含水率導致的光譜信息差異比較小,相比濕潤地區,用遙感多光譜信息反演地表SOM可信度更高[12]。

表1 30個樣點的SOM含量和土壤含水率對應遙感影像各波段的反射率Table 1 SOM and soil moisture content of 30 samples corresponding with reflectance of each band on image
3.2 建立模型及驗證
選取以上30個土壤有機質實測值為建模樣本,其余15個作為驗證樣本。SOM實測值Y為因變量,各波段像元反射率值(X1、X2、X3、X4、X5、X7)為自變量X建立多元線性回歸模型。為了得到最為精確的回歸模型,分別對R,1/R,LN(1/R),FDR進行建模。通過判定系數R2、F檢驗、顯著性檢驗Sig.、均方根誤差RMSE進行模型的精度檢驗。其中R2和F值越大,RMSE越小說明該模型具有較高的精度[9];Sig.小于0.05則說明該模型具有較高的顯著性。
由上述方法得到不同反射率指標的土壤有機質反演模型,從表2中可以看出由FDR(first derivative reflectance)建模的效果最好,其中R2為0.964。F檢驗值為所有模型中最大并且通過了顯著性P<0.05的檢驗,RMSE也為最低的0.215。其他3種模型的R2、F檢驗、RMSE都偏小或者偏大。其中R中的R2是0.4,F是2.223是4組模型中最小的,而RMSE值3.616和Sig.的檢驗結果0.083為最大,也就是說通過R建模的效果最差。1/R和LN(1/R)模型的RMSE都一樣偏大,也并未通過顯著性檢驗。所以前3組模型的建模效果偏差,FDR建模的精度最佳。

表2 土壤有機質遙感模型及精度驗證Table 2 Remote sensing models of SOM and precision validation
由其余15個土壤樣本對FDR反演的模型進行驗證,通過計算得到SOM預測值并與實測值進行比較(如圖2)。圖中由實測值與預測值擬合形成趨勢線y=2.983x-1.273,R2為0.963 3,一能說明通過FDR建模反演土壤有機質可行,二能說明其預測效果理想,能較好地表達研究區不同空間的土壤有機質含量。

圖2 預測值與實際值的關系Fig.2 Relationships between actual and estimated values
3.3 土壤有機質空間分析
運用3.2得到的FDR有機質反演模型,通過ENVI軟件來實現整個研究區的土壤有機質含量預測,得到圖3。圖中土壤有機質含量的最大值為13.065 g/kg,最小值為0.355 g/kg。其中有機質含量在0~5 g/kg之間的面積占總研究區的84.065%,在5~10 g/kg之間的面積占研究區的15.933%,>10 g/kg的面積僅僅為0.001 5%。相比其附近奇臺縣農田的土壤有機質含量[18],五彩灣地區的土壤有機質含量極少,大多數地區土壤有機質質量分數不足1%。

圖3 SOM含量的空間分布Fig.3 SOM content spatial distribution
對遙感影像進行圖像分類,分為4種地類:工礦用地、裸地、荒漠林帶、鹽漬地。再對不同地類進行土壤有機質的空間分析得到表3。表中顯示,工礦用地面積為339.618 km2,占研究區面積的8%,SOM平均含量為最高的7.35 g/kg。礦區有機質含量高主要是由于煤炭作為有機物被開采而露出地表;在煤炭運輸、粉碎、存儲過程中會散落到地表;煤炭的不充分燃燒使得未被燃燒盡的煤粉進入大氣后沉降到地表。裸地面積2 674.44 km2,占研究區面積的63%,其SOM含量均值較高,一方面受上述煤炭開采影響,另一方面戈壁灘氣候相對惡劣,植被覆蓋度極低,增加了土壤有機質的流失。而不同的是鹽漬地和荒漠林地植被覆蓋度較高,SOM含量本應該很高,但是相反,由于這片區域鹽漬化和荒漠化的加劇發展,這2個區域的土壤沙化嚴重,地表有機質流失也很嚴重。荒漠植物都具有很強的耐旱性,即使在SOM含量和降水極低的情況下也能生長。

表3 不同地類的SOM含量Table 3 SOM content of different land types
本文通過FDR建立模型的R2為0.963。運用該模型預測研究區空間范圍的SOM,得出土壤有機質SOM含量>10 g/kg的面積僅僅為0.001 5%,土壤有機質含量整體匱乏。對SOM含量數據進行空間分析得出工礦用地SOM平均含量為最高的7.35 g/kg,受開采的煤炭中有機質影響較大。裸地SOM平均含量為6.12 g/kg,鹽漬地和荒漠林地SOM含量均不高。總之,運用多光譜遙感數據和實測數據相結合建模的方法在干旱區適用。
該地區土壤有機質含量平均值偏低,個別地區極低,加之該地區降水量稀少,土地荒漠化程度加劇,進行生態修復很有必要。選取合適的植被在荒漠戈壁灘和沙地中種植將成為生態修復成敗的關鍵。此外通過合理的放牧和卡拉麥里保護區的嚴格監管有助于減緩該地區土地荒漠化的速度,從而避免人為造成土壤有機質的流失。
[1]Sparling G P,Wheeler D,Vesely E T,et al.What is soil organic matter worth?[J].Journal of Environmental Quality,2006,35(2): 548-557.
[2]蒙繼華,吳炳方,杜鑫,等.遙感在精準農業中的應用進展及展望[J].國土資源遙感,2011,23(3):1-7.Meng Jihua,Wu Bingfang,Du Xin,et al.A review and outlook of applying remote sensing to precision agriculture[J].Remote Sensing for Land and Resources,2011,23(3):1-7.(in Chinese with English abstract)
[3]Naveen J P A,Abd-Elrahman A H,Lewis D B,et al.Modeling soilparametersusinghyperspectralimagereflectancein subtropical coastal wetlands[J].International Journal of Applied Earth Observation and Geoinformation,2014,33(11):47-56.
[4]王祥峰,蒙繼華.基于HJ-1衛星的農田土壤有機質含量監測[J].農業工程學報,2014,30(8):101-108.Wang Xiangfeng,Meng Jihua.Mapping soil organic matter content in field using HJ-1 satellite image[J].Transactions of the Chinese Society of Agricultural Engineering(Transactions of the CSAE),2014,30(8):101-108.(in Chinese with English abstract)
[5]Liu Shiliang,An Nannan,Yang Juejie,et al.Prediction of soil organic matter variability associated with different land use types in mountainous landscape in southwestern Yunnan province, China[J].Catena,2015,133(10):137-144.
[6]Liu Feng,Rossiter D G,Song Xiaodong,et al.A similarity-based method for three-dimensional prediction of soil organic matter concentration[J].Geoderma,2015,263(1):254-263.
[7]李超,劉兆剛,岳樹峰,等.基于激光雷達數據的森林表層土壤有機質空間格局反演[J].應用生態學報,2012,23(9):2451-2458. Li Chao,Liu Zhaogang,Yue Shufeng,et al.Retrieval of forest topsoil organic matter’s spatial pattern based on LiDAR data[J]. Chinese Journal of Applied Ecology,2012,23(9):2451-2458. (in Chinese with English abstract)
[8]袁征,李希燦,于濤.高光譜土壤有機質估測模型對比研究[J].測繪科學,2014,39(5):160-164.Yuan Zheng,Li Xican,Yu Tao.Contrast research on soil organic matter estimation model using hyper-spectral data[J].Science of Surveying and Mapping,2014,39(5):160-164.(in Chinese with English abstract)
[9]Al-Abbas A H,Swain P H,Baumgarder M F,et al.Relating organic matter and clay content to the multi-spectral radiance of soils[J].Soil Science,1972,114(6):477-485.
[10]侯艷軍,塔西甫拉提.特依拜,買買提.沙吾提,等.荒漠土壤有機質含量高光譜估算模型[J].農業工程學報,2014,30(16):113-120.Hou Yanj un,Tiyip T,Sawut M.Estimation model of desert soil organic matter content using hyperspectral data[J].Transactions of the Chinese Society of Agricultural Engineering(Transactions of the CSAE),2014,30(16):113-120.(in Chinese with English abstract)
[11]田永超,張娟娟,姚霞,等.基于近紅外光聲光譜的土壤有機質含量定量建模方法 [J].農業工程學報,2012,28(1):145-152.Tian Yongchao,Zhang Juanjuan,Yao Xia,et al.Quantitative modeling method of soil organic matter content based on nearinfrared photoacoustic spectroscopy[J].Transactions of the Chinese Society of Agricultural Engineering(Transactions of the CSAE),2012,28(1):145-152.(in Chinese with English abstract)
[12]劉煥軍,趙春江,王紀華,等.黑土典型區土壤有機質遙感反演[J].農業工程學報,2011,27(8):211-215.Liu Huanjun,Zhang Bai,Zhao Jun,et al.Spectral models for prediction of organic matter in black soil[J].Transactions of the Chinese Society of Agricultural Engineering(Transactions of the CSAE),2007,44(1):27-32.(in Chinese with English abstract)
[13]張法升,曲威,尹光華,等.基于多光譜遙感影像的表層土壤有機質空間格局反演[J].應用生態學報,2010,21(4):883-888.Zhang Fasheng,Qu Wei,Yin Guanghua,et al.Spatial pattern of surface soil organic matter based on remotely sensed multispectral imagery[J].Chinese Journal of Applied Ecology,2010, 21(4):883-888.(in Chinese with English abstract)
[14]于雷,洪永勝,耿雷,等.基于偏最小二乘回歸的土壤有機質含量高光譜估算[J].農業工程學報,2015,31(14):103-109.Yu Lei,Hong Yongsheng,Geng Lei,et al.Hyperspectral estimation of soil organic matter content based on partial least squares regression[J].Transactions of the Chinese Society of Agricultural Engineering(Transactions of the CSAE),2015,31 (14):103-109.(in Chinese with English abstract)
[15]崔文采.新疆土壤[M].北京:科學出版社,1996:182-186.
[16]姚峰,古麗.加帕爾,包安明,等.基于遙感技術的干旱荒漠區露天煤礦植被群落受損評估[J].中國環境科學,2013,33(4):707-713. Yao Feng,Jiapaer G,Bao Anming,et al.Damage assessment of the vegetable types based on remote sensing in the open coalmine of arid desert area[J].China Environmental Science, 2013,33(4):707-713.(in Chinese with English abstract)
[17]鮑士旦.土壤農化分析[M].北京:科學出版社,1996:182-186.
[18]欒福明,張小雷,熊黑鋼,等.基于影像的荒漠綠洲交錯帶土壤有機質含量反演模型[J].中國沙漠,2014,34(4):1080-1086.Luan Fuming,Zhang Xiaolei,Xiong Heigang,et al.Inversion models of soil organic matter in oasis-desert ecotone based on TM Image reflectance[J].Journal of Desert Research,2014,34 (4):1080-1086.(in Chinese with English abstract)
Estimation model of soil organic matter in desert mining area based on multispectral image data
Xia Nan,Tashpolat.Tiyip※,Ding Jianli,Ilyas Nurmemet,Zhang Dong,Liu Fang
(1.College of Resources and Environmental Sciences,Xinjiang University,Urumqi 830046,China; 2.Key Laboratory of Oasis Ecology(Xinjiang University)Ministry of Education,Urumqi 830046,China)
Soil is related closely to human living and vegetation grow.The quality of soil organic matter(SOM)influences plant development.Scientists take a variety of researcheson soil.Many findings focus on the estimation of SOM using remote sensing data,which are usually hyperspectral and multispectral.The former has a detailed result of band information,while the latter provides a macroscopical and convenient way to get in whole area.In addition,processing hyperspectral data needs a strong mathematical background and software technology,while processing multispectraldata needs less.To apply the multispectral method to make decisions on buildings and planning is of great significance.In order to save resources,increase efficiency and accuracy,in May 2014,we collected soil samples in the various layers of 0~10, 10~20 and 20~30 cm,and there were totally 45 points marked by GPS(global positioning system)on Google Earth.The weighed aluminum box was used to hold some soil in each layer.The collections were taken back and dried for 24 h.Then the dried soil was weighed and the soil moisture was calculated.Meanwhile,the image needed pretreatment.The atmospheric correction should be taken to remove bands′noises to get clear data.Then the pixels of the image for each sample point were used to establish models.And at the same time,other soil was crushed and sieved in 2 mm,and the SOM was measured by the potassium dichromate volumetric method.The final work was to combine the reflectance data of multispectral image and the measured SOM data.We used the reflectance(R),the reflectance reciprocal(1/R),the reflectance reciprocal′s logarithm(ln(1/R)),the reflectance′s first derivative(FDR)and the measured SOM to build multiple linear regression models,and then,it was found that the FDR model had a better precision with the R2of 0.963 between the predicted and the measured.This meant that the more effective approach could be applied to express the regional SOM if needed.By the FDR model,we predicted the SOM content in study area.It showed that the area with SOM content of 0~5 g/kg was 84.065%of the whole area and that with SOM content of greater than 10 g/kg was 0.001 5%.The greatest SOM value was 13.065 g/kg,and the inferior was closed to 0.355 g/kg,which was very low.The SOM content in the Wu caiwan area was lower than that in Qitai County,for the former′s SOM was less than 1%in the most area.It also indicated that the highest average SOM content in the mining area was 7.35 g/kg,which was influenced by the organic matter in coal.The bare land's area was 2 674.44 km2,accounting for 63%of all area,and the mean SOM content was 6.12 g/kg.The saline land and desert woodland had lower SOM content because of the development of water-soil loss,salinization and desertification.The low SOM content and less precipitation made the area a desert increasingly.Further more,we found that in the arid area,the soil moisture content was extremely low,so it was not only influenced weakly by moisture to using remote sensing means to estimate SOM,but also formed an advantageous method which provided a higher simulation precision.All in all,it is imperative to restore the ecologic environment in the study area.Measures should be taken immediately.Choosing appropriate vegetation to plant in desert will be the key to the restoring works,while enhancing supervision of the Kalamaili Nature Reserve and controlling grazing will contribute to slow down those negative phenomena above.
soils;remote sensing;spectrum analysis;desert;SOM;modeling;multispectral;estimation
10.11975/j.issn.1002-6819.2016.06.036
TP79;S127
A
1002-6819(2016)-06-0263-05
2015-10-20 修改日期:2016-01-23
國家科技支撐計劃項目資助(2014BAC15B01);國家自然科學基金項目資助(41130531,41561089)
夏 楠,男,新疆昌吉人,博士生,主要從事干旱區生態定量遙感方面的研究。烏魯木齊 新疆大學資源與環境科學學院、新疆大學綠洲生態教育部重點實驗室,830046。Email:xianan113693615@163.com
※通信作者:塔西甫拉提.特依拜,男,維吾爾族,新疆伊寧人,教授,博士生導師,主要從事干旱區資源環境與遙感應用研究。烏魯木齊 新疆大學資源與環境科學學院、新疆大學綠洲生態教育部重點實驗室,830046。Email:tash@xju.edu.cn