Quantitative inverse modeling of nitrogen content from hyperion data under stress of exhausted coal

AvailableMININGScience DirectSCIENCE ANDTECHNOLOGYELSEVIERMining Science and Technology 19(2009)0031-0035www.elsevier.com/locate/jcumtQuantitative inverse modeling of nitrogen content fromhyperion data under stress of exhausted coal mining sitesLU Xia, HU Zhen-qi, GUO Liartment of Marine Technology, Huaihai Institute of Technolo2 college of resources& Safery Engineering, China University of Mining& Technology, Beijing 100083, ChinaBeijing institute of Geological Sunvey, Beijing 100083, ChinaAbstract: Monitoring and evaluating the nutritional status of vegetation under stress from exhausted coal mining sites byper-spectral remote sensing is important in future ecological restoration engineering. The Wangpingcun coal mine, located in theMentougou district of Beijing, was chosen as a case study. The ecological damage was analyzed by 3S technology, field investigation and from chemical data. The derivative spectra of the diagnostic absorption bands are derived from the spectra measured in thefield and used as characteristic spectral variables. a correlation analysis was conducted for the nitrogen content of the vegetationsamples and the first derivative spectrum and the estimation model of nitrogen content established by a multiple stepwise linearsion method. The spatial distribution of nitrogen content was extracted by a parameter mapping method from the Hyperiondata which revealed the distribution of the nitrogen content. In addition, the estimation model was evaluated for two evaluationindicators which are important for the precision of the model. Experimental results indicate that by linear regression and parametermapping, the estimation model precision was very high. The coefficient of determination, R, was 0.795 and the standard deviationof residual(SDR)0. 19. The nitrogen content of most samples was about 1.03% and the nitrogen content in the study site seeminversely proportional to the distance from the piles of coal waste. Therefore, we can conclude that inversely modeling nitrogencontent by hyper-spectral remote sensing in exhausted coal mining sites is feasible and our study can be taken as reference in species selection and in subsequent management and maintenance in ecological restoration.Keywords: Hyperion; nitrogen content; estimation model; linear regression1 Introductionestimations of bio-physical and bio-chemical parameters2-12. Most of these parameters,extractedWhen coal resources are exhausted coal mines be- from hyper-spectral data, are mainly used in crop orcome abandoned, after having brought about consid- forest areas, while studies on the nutrient status of therable economic benefits. At that point, coal mines vegetation in coal mining sites are rarely seen 3jsubsidence, water, air and soil pollution, and so on. mining area as a case study and will analyze its eco-Once the environment is seriously damaged, the con- logical damage. On the basis of this analysis, we disditions of the ecosystem will have changed and ecocuss the relationship between nitrogen content anlogical service functions have degenerated. There- the spectra of vegetation samples and establish anfore, ecological rehabilitation and restoration at coal estimation model of nitrogen content by multiple,mining sites become very important in order to re- stepwise linear regression methods. Furthermore, anbuild a favorable coal mining eco-environment. Un- inverse precision of model is analyzed according toderstanding the nutritional status of the vegetation its R and SDR.around a coal mining area is critical in order to realizea rapid and ordered succession of communities. Ni- 2 Ecological damage situation of a coaltrogen elements play an important role in these eco-logical service functions from the point of nutritionalcycle of the vegetation. At present, experts both atThe study was conducted in the Wangpingcun coalhome and abroad have achieved outstanding results mining site located in Wangping town. It is one of theby using hyper-spectral remote sensing technology, stat中国煤化工 d by the Beijingfrom identification of vegetation types to quantitative M-ast to west is 5CNMHGReceived 1l May 2008: accepted 27 July 2008ngauthorTel+86-15061321290:E-mailaddressLuxial210@163.coming Science and TechnologyVol 19km and its width from north to south 2.5 km. The and potassium, as well as the alkali-hydrolyzable ni-coal deposits belong to the c-p periods of the Paleo- trogen in all samples were abundant and did not con-zoic Era. The coal type is anthracitetain a risk of potential pollutionExploitation of coal resources will inevitably causecological damage. From the point of view of2.3 Biological damageecosystem, ecological damage can be classified asLarge scale coal mining did considerable harm tolandscape damage, environmental damage and bio- the health of the native habitats and gradually reducedlogical damagevegetation communities, resulting in a degraded suc-2.1 Landscape damagecession of vegetation. At the same time, the numberand types of wild species decreased. In addition, theDamage to the landscape in our study site mainly rate of incidence of disease in the population of theincludes piles of coal waste, abandoned construction coal mining site rose gentlysites and houses, mining subsidence and crackingIn brief, the ecosystem in this coal mining area wasThe piles of coal waste occupy an area of almost severely damaged and ecological service functions215000 m. An additional problem is that, because decreased. Under such stress, future land reclamationthe piles of coal waste accumulate in low areas, they and ecological restoration in this coal mining site hasgreatly affect flood drainage and threaten the safety become an urgent matter. We need to understand theof buildings. The land used for construction sites has nutrient status of the vegetation in order to contributechanged greatly because of mining activities, which to benign ecological and environment rahabilitationhas caused the amount of farmland to decreasegradually with an attendant increase in uneven and 3 Data acquisition and image processingbarren grassland. A large area of exhausted drifts under the surface is caused by coal mining activities andCloud-free imaging spectrometer data were ac-resulted in surface subsidence( cracks in the surface) quired on June 15, 2006 with Hyperion instrumenta-leading to large areas of abandoned houses(17013 m) tion. The Hyperion level iB data were provided asin Lvjiapo village In our study, 1l surface subsidence calibrated radiance in 220 contiguous channels fromsites were investigated with an area of 2149.095 m". 0.4-2.5 m with a spectral resolution of 0.01 m and aCracks in the study area are classified into three types: spatial resolution of 30 m. The level 1B Hyperionground fissures, building cracks and mountain cracks. radiance data were first corrected for"streaking"or2.2 Environmental damagevariation in balance among vertical columns in thetrack direction of the image data, a product of theMany years of coal mining activities may pollute "pushbroom"design of the sensor, most evident inand impoverish the soil. If the amount of heavy met- the shortwave infrared channels (1.0-2.5 m). Imageals in the soil is relatively high, it will restrain the processing includes the choice of useful bands, thegrowth of and do harm to the vegetation. Therefore, restoration of bad lines, the removal of vertical barsthe vegetation restoration in an exhausted coal mining apparent surface reflectance rebuilding, image resiz-area will become difficult. Fourteen soil and waste ing and geometric correction. Among these pretreatsamples were collected within the area of a circle, ments, radiance data were transformed to apparentwith its center at the Jidou pile of coal waste and a surface reflectance, using the FLAASH atmosphericradius of 1500 m. Heavy metals in the test experi- correction program. Field spectrometer data andment included five elements: Cr, Cu, Pb, Zn and Cd, vegetation samples in 24 homogeneous plots(30 m,respectively. The parameters of the nutrient compo- 30 m)were collected when the Earth Observing Onenents of our soil samples include pH, organicsatellite passed over the study site in the Wangpingtotal nitrogen, rapidly available phosphorus,cun coal mining area. Canopy-based spectra wereavailable potassium and alkali-hydrolyzable nitro- obtained from a portable ASD spectro-radiometer.Each vegetation sample consisted of a composite ofb The change in the amount of heavy metals of each leaves collected from several heights in the canopy. Asample as a function of the distance between the semi-micro Kjeldahl approach, with an auto analyzersample collection position and the coal pile is dis- determination method, was used to extract nitrogencussed, which indicate that Cu and Cr pollution in the (N). Table I provides for the presentation of vegetastudy area is not a problem, but Cd pollution is seri- tion types and nitrogen contents.Comparing the nutrients in each sample with the 4 Estimation model of nitrogen contentading standards of soil nutrient content, we foundthat the amounts of organic matter and total nitrogen 4.1中国煤化工ristie variablesin all soil samples are deficient, while the total nitro-gen in the coal waste samples was relatively higherCNMH Ga measure of thethan the standard value. Rapidly available phosphorus slope of the spectral curve at every point and resultsLU Xia et alin a spectrum which allows baseline offsets and tion domain of nitrogen elements in green vegeta-low-frequency variation to be removed or substan-tionthese are listed in Table 2. Extracting thetially minimized. Earlier experts listed corresponding first derivative spectra within the characteristic ababsorption bands of 42 biochemical constituents in sorption bands(DSAB)are chosen as the spectralthe reflectance spectra. Investigators, such as Zhi et characteristic variablesal., have summarized the characteristic band absorp-Table 1 Vegetation types and corresponding nitrogen content in coal mining areaTree of heavenVitexussopetia8SpeciesVitexApricotVitexVitexApricot141516SpeciesVitexCane4scent wikstroemia Oriental arborvitaeN(%)21222324Tree of heavenTree of heaven1.642.182Table 2 Absorption bands and usable band range for4.3 Quantitative estimation model of nitrogenvegetation nitrogen contentcontentCharacteristic absorption bandIn order to develop relationships between the spec-910-1081content,we have examined an empirical method ofmultiple linear regression(MLiR) of the first deriva1796-22141980,2060,2130-2180tive spectra of characteristic nitrogen absorption2230-24152240,2300,2350bands. From our many experiments, the final inputvariables have resulted in our selection of the first4.2 Correlation analysis of DSAB and nitrogen derivative spectra of 917, 2306 and 678 nm, respectively. The order of stepwise input was 917, 2306 andThe reflectance spectra for all 24 green vegetation678 nm. All of theiedsamples were first arranged according to the absorp-SPSS software. The estimation model of nitrogention band domain, listed in Table 2. Secondly, the firstderivative spectra were extracted based on the corre-y=2077-439397Dsponding bands. The correlation analysis of the de-+382328D20-743.077D678(1)rivative spectra and the nitrogen content has beersuccessfully conducted. The analytical results, listed where y is nitrogen content, a wavelength and Da917,in Table 3, indicate that nitrogen content is signifi- D 306, and da678 are the derivative spectra in the 917,cantly correlated with 12 derivative spectra at the 0.01 2306 and 678 nm bandsThe matrix of correlation coefficients between the583 to 750 nm, three from 910 to 1081 nm, one from input variables is shown in Table 4. Ensuring high1270 to 1666 nm. two from 1796 to 2214 nm and precision of the model, we have relatively low correfour from 2230 to 2415 nm. among these 12 highly lation between these input variablescorrelated derivative spectra, the spectra at 678, 1522Table 4 Correlation matrix of input bands917 and 2019 nm are negatively correlated withSpectral bands(nm)2306trogen content.0.364Table 3 High correlation bands and coefficients forvegetation nitrogen content0.3640.175Band(mm)61667891291710571522coefticient()0.5260.6190.5490.6710.530-0.56444H中国煤化工 contentBand(m)2019202022702303230623CN MHGamLir method,Ic Iust actIvative spectra ofcoefficien0.6020.5830.5190.5360.6570.519spectral field data measurements. One of the reasonsMining Science and Technologyfor this selection is that spectral resolution is higher given an rof 0.795 and a SDR of 0.19than that of Hyperion data; the second reason is thatthe collection position of the 24 vegetation samples 6 Conclusionsas homogeneous at the 30 m, 30 m plots. In order toobtain the spatial prediction of the nitrogen content ofI)The ecological damage in the exhausted coalgreen vegetation in the study area, the estimation mining site is serious. the piles of coal waste occupymodel was applied to the Hyperion image by pa- an area of 215000 m, abandoned houses 17013 mrameter mapping. The Hyperion bands are 681. 21, and the area of subsidence is 2149.095 m Severe Cd915.24 and 2304.72 nm, which are adjacent to the pollution exists in the study area and the levels ofinput variables in the model. Because there are some organic matter and total nitrogen content are deficient,landscapes such as water, buildings, coal waste pilewhile the nitrogen content in the coal waste is highand so on, other than green vegetation, the processing The vegetation has been reduced and the health of theof non-green cover regions is performed by mask population of the area has deterioratedtechnology the Hyperion image. Fig. 1 shows a spa-2) The first derivative spectra, limited to the n ab-tial prediction map of nitrogen content, which is clas- sorption bands, was selected as the characteristicsified into four classes over the area. The nitrogen spectral variable. The estimation model of N was es-content over the study area varied from 0% to ap- tablished by MLIR. The spatial prediction of nitrogenproximately 3. 6%concentration. The highest N con- content in the study area was obtained by a parametertent occurred around the coal waste piles with lower mapping method. The distribution of N is inverselyvalues at greater distances from the piles, i.e, the proportional to the distance from the coal waste pilesspatial distribution of nitrogen content is inversely3)We applied the statistics R and SDr to evaluateproportional to the distance from the coal piles. This the precision of the model; R was 0.795 and SDRresult seems incompatible with the actual coal mining only 0.190site. However, detailed analysis shows that the result4)The application of imaging spectrometry andis entirely correct because large amounts of nitrogen adding ground-based spectra of green vegetation,in coal waste are gradually transferred to the sur- collected by an ASD spectrometer, indicates that es-rounding soil, from the diffusion by the monsoon and timation of canopy nitrogen content over the studysoil erosion. The nitrogen becomes absorbed by the area is entirely possiblegreen vegetation and therefore increases its nitrogencontent. This result provides the local government Acknowledgementsnd the department concemed with a very importantreference for the ecological restoration of the vegetaThis work was undertaken as part of the ecologicaltion. Meanwhile, soil fertility will be enhanced by the restoration in the Mentougou district program.Thenitrogen content in the coal waste and will improve authors thank Professor Hu Zhengi, whose instructiongrowth of the vegetationmade this program possible. Zhang Hongguang whoworks in the geological institute of the nuclear indus-based spectra. As well, the authors thank Kang Jintao,Li Haixia, Chen Tao, He Fengqin and Xu Xianlei whoprovided valuable help and advice in the collection ofsamples and chemical testingReferences[1] Song S J. The ecological damage analysis causedig. 1 Spatial distribution map of nitrogen contentexploitation and controlling strategy. Coal Procin the coal mining siteand Comprehensive Utilization, 2007(4): 44-4Chinese5 Precision evaluation[2] Sims D A, Gamon J A Relationships between leaf piment content and spectral reflectance across a wideTwo disadvantages of multiple linear regressSIonrangeanalysis are that MLIR models are likely to result inges. Remote Sensing of Environment, 2002(81): 331an over- fit and some bands are ignored in the model. [3] Tian Q3, Gong P, Zhao C, Guo X W. The analysis ofB this case, the precision evaluation of the mlirmodel becomes very important. The evaluation indi-L中国煤化工x5scators selected by us consist primarily of the multiplecoefficient of determination(R) and the standard (4)CNMHGE Zhang HC, Jiangdeviation residual (SDR). The model is quite accurate,H R. The analysis of agricultural parameters of thLU Xia et alQuantitative inverse modeling of nitrogen content from hyperion data under stress ofsummer com by red edge parameters. The Journal of theRemote Sensing of Evironment, 2001(75): 153-161University of Fujian Agriculture and Forest, 2006, 35(2): [11] Kokaly R F. Investigating a physical basis for spectro-[5 Chen Y H, Li X B, Li J, Du P J. The adjustable vensing of Environment, 2001(75): 153-161tion index study based on the aerial remote sensin[12] Curran P J, Dungan J L, Peterson D L. Estimatingplication. Journal of China University of minifoliar biochemical concentration of leaves withTechnology, 2004, 33(4): 438442(In Chinese)spectrometry. Remote Sensing of Emv[6 Blackburn G A. Spectral indices for estimating photo-synthetic pigment concentrations: a test using senescent [13] Lu X, Hu Z Q, Liu w J. Dynamic vegetation growthtree leaves, International Journal of Remote Sensing,monitor under coal exploitation stress by remote sensing1998(19)}657-675n bulianta coal mine. Journal of China University[7 Liu W D, Xiang Y Q, Zheng E F, Tong x, WuCSMining and Technology, 2007, 17(4): 479-483The correlation analysis between the hyperspectral data [14] Yao B, Shang H, Han JJ, Liang JS. The transgene vegeand the LaI and chlorophyll density. The Journal oftation restoration for heavy metal and organic pollutionRemote Sensing, 2000, 4(4): 279-283. (In Chinese)soil. The Forest Science, 2005, 41(4): 162-167. (in Chi-[8] Wu C S, Xiang Y Q, Zheng L F, Tong Qxnescphyll concentration estimation of the crop by hy[15] Bao S D. The Agrochemistry Analysis of the Soil. Beijing:tral remote sensing data. The Journal of RemotThe Press of China Agriculture, 2000.(In Chinese)20004(3):2285-2302.( n Chinese)[16] Pu R L, Gong P. Hyperspectral Remote Sensing and its[9] Niu Z, Chen Y H, Sui H Z, Zhang Y, Zhao C J. Theapplications. Beijing: Higher Education Press, 2000. (Indetection mechanism analysis of biochemical componentectral data. The[17] Zhi H, Brian J T, Stephen J D, lan R W, Foley WJ. Es-Sensing, 2000, 4(2): 125-130.(In Chinese)timating foliage nitrogen concentration from HYMAP[10] Kokaly R F, Clark R N. Spectroscopic determination ofdata using continum removal analysis. Remote Sensingleaf biochemistry using band-depth analysis of absorp-of Environment, 2004(93): 18-29tion features and stepwise multiple linear regression中国煤化工CNMHG