《土壤学》课程教学资源（参考资料）Monitoring the Plant Density of Cotton

Monitoring the Plant Density of Cotton with Remotely Sensed Data Junhua Bai Jing Li,and Shaokun Li 1Institute of Crop Sciences,Chinese Academy of Agriculture Sciences, Beijing 100081,P.R.China Tel:+861082108891 shaokun0004@sina.com 2 State Key Laboratory of Remote Sensing Science.Jointly Sponsored by the Institute of ese emy of Sc ences and Beijing 01.P.E niversity. China Abstract.PDC(Plant Density of Cotton)was an essential parameter for esti- mating the cotton yield and developing the zone-management measurements This paper proposed a new method to retrieve PDC from the satellite remote sensing data.The thirteen fields of Xinjiang Production and Construction Corps (XPCC)(total 630 hm2)were selected as the study area,where the sowing date emergence date,and PDC were investigated.Based on the investigation data the linear models to estimate PDC are established using EVI and DEVI respec tively.The results indicated that the difference of seedling size caused by the emergence time decreased the estim tion accuracy o To improve the e uracy were estab I in te ng the 01 Tring the n fom Jun licated tha the optim ng PD 0h 48 per of XPCC It luded that the tin nd th y of PDC and the vith the e time could i h DEV could make the mont ring time fom ard,and the optimal monitoring time was from the squa ring stage to the full-flowering s .This res arch provides ar efficient,rapid and intact way to monitor PDC.and it is significant for opera- tional application at a regional scale. Keywords:Cotton,Plant Density,Remote sensing. 1 Introduction PDC (Plant Density of Cotton)is one of key factors affecting cotton vield.and is the most important index for analyzi ing cott on growth and taking the management Corresponding author. D.Li,Y.Liu,and Y.Chen (Eds.):CCTA 2010.Part II,IFIP AICT 345.pp.90-101.2011. IFIP International Federation for Information Processing 2011

D. Li, Y. Liu, and Y. Chen (Eds.): CCTA 2010, Part II, IFIP AICT 345, pp. 90–101, 2011. © IFIP International Federation for Information Processing 2011 Monitoring the Plant Density of Cotton with Remotely Sensed Data Junhua Bai1,2,3, Jing Li2 , and Shaokun Li1,* 1 Institute of Crop Sciences, Chinese Academy of Agriculture Sciences, Beijing 100081, P. R. China Tel.: +86 10 82108891 shaokun0004@sina.com 2 State Key Laboratory of Remote Sensing Science, Jointly Sponsored by the Institute of Remote Sensing Applications of Chinese Academy of Sciences and Beijing Normal University, Beijing 100101, P. R. China 3 College of Agricultural Sciences of Shihezi University, Shihezi Xinjiang, 832003, P. R. China Abstract. PDC (Plant Density of Cotton) was an essential parameter for esti￾mating the cotton yield and developing the zone-management measurements. This paper proposed a new method to retrieve PDC from the satellite remote sensing data. The thirteen fields of Xinjiang Production and Construction Corps (XPCC) (total 630 hm2 ) were selected as the study area, where the sowing date, emergence date, and PDC were investigated. Based on the investigation data the linear models to estimate PDC are established using EVI and DEVI respec￾tively. The results indicated that the difference of seedling size caused by the emergence time decreased the estimation accuracy of PDC. To improve the es￾timation accuracy the partition functions were established in terms of sowing date. DEVI is capable of reducing the influence of soil background significantly and it can bring the monitoring time forward from June 9th to May 24th in this research. The results indicated that the optimal time monitoring PDC would be from squaring to full-flowering of cotton growing period. A demonstration to monitor PDC was taken on June 9th in the 148th farm of XPCC. It can be con￾cluded that the emergence time and the non-cotton background were the main factors affecting the monitoring accuracy of PDC, and the partition function with the emergence time could improve the estimation accuracy, and DEVI could make the monitoring time forward, and the optimal monitoring time was from the squaring stage to the full-flowering stage. This research provides an efficient, rapid and intact way to monitor PDC, and it is significant for opera￾tional application at a regional scale. Keywords: Cotton, Plant Density, Remote sensing. 1 Introduction PDC (Plant Density of Cotton) is one of key factors affecting cotton yield, and is the most important index for analyzing cotton growth and taking the management * Corresponding author

Monitoring the Plant Density of Cotton with Remotely Sensed Data 91 PDC weakens the land ent (W.E.Zho 2004:X.L Hou 2006.Ths vation technology for increasing cotton yield. Agriculture is one of imp rtant a and the the ar yield itorins (O.Wendroth et al.2003:P.C.Dore al.2004:X eal.2006.i cotton m onito method of P atching ha been use ract the el in th ale(J. L et al. 200),and mage textur tis on growth( Liao,et a 2006),and th rse the area ind .Bal,er a 2007.V Alchan atis (2005)used the hyper-spectrum image to distinguish the weeds from cot- ton hield and the action of water stress is compared with the root disease by infrared sensor (N.R.Falkenberg et al.,2007).And although PDC plays an important role in the farmland practices,few researches was done on PDC monitoring by satellite data. Normal Difference Vegetation Index(NDVD)is one of the most popular parameters used to evaluate the crop growth condition (J.W.Rouse,1974:F.Baret,G.Guyot. 1989.1991).And some researches show that NDVI is susceptible to the influence of the non-vegetation background.and it saturates when the vegetation cover increases to the some extent (O.Wang et al.,2005:A.E.K.Douaoui,2006).Other indices were constructed to make up for the inadequacy,and the Enhance Vegetation Index(EVI) is one of the most popular in them. At present,the field investigation isamost the ony way toobtain PC.The dis of thi way is that it ca t the point data and umes much time the p an we analyzed the vestigation data from the em mporal atellite images couplin ergence to full-fl wering stage of co otton to (I construct the linear model to estimate PDC.(2)analyze the main factors affecting the monitoring accuracy of PDC and perfect the model,(3)obtain the optimal period to monitor PDC,(4)demonstrate the PDC monitoring method in the regional scale. 2 Methods 2.1 Study Area locnthe farm of XPCC.besie the Manas river valle E). e typ e is the typical continent climate,and the sunshine rate is ing the cotton growing period.The heand ra the for over 90% in the farm early of June,flowers at early July.The row space is usually 55cm+20cm+55cm and plants space is usually 9.6cm.At the middle of July.the vegetation coverage fraction rises to over 80%.and the top-ending measurement was taken to control the unlimite character of cotton.In 2007.the sowing date is mostly trom

Monitoring the Plant Density of Cotton with Remotely Sensed Data 91 measurements. The cotton yield in Xinjiang accounts for about 30% in China and 8% in the international cotton trade. PDC is a key item in all measurements for the cotton farmland management (W. F. Zhang et al., 2004; X. L. Hou et al., 2006). The low PDC weakens the applications of cultivation technology for increasing cotton yield. Agriculture is one of important applications of satellite remote sensing technology, and the numerous researches have been done on the area estimation, yield estimation and growth monitoring (O. Wendroth et al., 2003; P. C. Doraiswamy et al., 2004; X. M. Xiao et al., 2006). In cotton monitoring, the method of spectrum matching has been used to extract the cotton pixel in the regional scale(J. Li et al., 2005), and the image texture of geo-statistical has been introduced into monitoring cotton growth(C. J. Liao, et al., 2006), and the hyper-spectrum can be used to inverse the biomass and leaf area index quantitatively (D. L. Zhang et al., 2007; J. H. Bai, et al., 2007), V. Alchanatis (2005) used the hyper-spectrum image to distinguish the weeds from cot￾ton field, and the action of water stress is compared with the root disease by infrared sensor (N. R. Falkenberg et al., 2007). And although PDC plays an important role in the farmland practices, few researches was done on PDC monitoring by satellite data. Normal Difference Vegetation Index (NDVI) is one of the most popular parameters used to evaluate the crop growth condition (J. W. Rouse, 1974; F. Baret, G. Guyot, 1989, 1991). And some researches show that NDVI is susceptible to the influence of the non-vegetation background, and it saturates when the vegetation cover increases to the some extent (Q. Wang et al., 2005; A. E. K. Douaoui, 2006). Other indices were constructed to make up for the inadequacy, and the Enhance Vegetation Index (EVI) is one of the most popular in them. At present, the field investigation is almost the only way to obtain PDC. The dis￾advantage of this way is that it can only get the point data, and consumes much time and energy. In this paper, we analyzed the multi-temporal satellite images coupling the field investigation data from the emergence to full-flowering stage of cotton to (1) construct the linear model to estimate PDC, (2) analyze the main factors affecting the monitoring accuracy of PDC and perfect the model, (3) obtain the optimal period to monitor PDC, (4) demonstrate the PDC monitoring method in the regional scale. 2 Methods 2.1 Study Area The study area is located in the148th farm of XPCC, beside the Manas river valley ((45°00′N 86°11′E; 44°46′N 86°23′E). The climate type is the typical continental-arid climate, and the sunshine rate is about 80% during the cotton growing period. The cotton planting proportion accounts for over 90% in the farm. Generally it seeds from the early April to the middle April, and Squaring at the early of June, flowers at early July. The row space is usually 55cm+20cm+55cm and plants space is usually 9.6cm. At the middle of July, the vegetation coverage fraction rises to over 80%, and the top-ending measurement was taken to control the unlimited growing character of cotton. In 2007, the sowing date is mostly from April 13th to 19th

92 J.Bai,J.Li,and S.Li Xinjiang provi ene of investigating ng st Rive alinizati Manasi river valley The The remote sensing images consist of TM6+TM Fig.1.Location of the experiment 2.2 Investigation and Computation for PDC PDC was investigated from June 5th to 15th,in 2007 and 2008.In order to make the sample data sentative.we chose the investigation points far from the field patch on op gro byGPS.The area appro tely owner of each te.The sowing 0 M ce par 0 sample points from the middle sowing date.4 from the early sowing date,and from the late sowing date.The data consisting of three sowing date are employed evenly into the establishing and testing model.PDC could be calculated as: N.=+m,+m×375 (1) 3

92 J. Bai, J. Li, and S. Li Fig. 1. Location of the experiment 2.2 Investigation and Computation for PDC PDC was investigated from June 5th to 15th, in 2007 and 2008. In order to make the sample data representative, we chose the investigation points far from the field patch boundary to avoid the boundary effect on the crop growing. Three sampling units were investigated in every investigation points. The sampling points were positioned by GPS. The area of sampling units is approximately 26.7m2 . We visited the land￾owner of each sampling fields to record the seeding and emergence date. The sowing time was divided into three partitions including the early, middle, and late of the sow￾ing date 52 sample points are obtained through investigating thirteen fields (630 hm2 ), including 40 sample points from the middle sowing date, 4 from the early sowing date, and 8 from the late sowing date. The data consisting of three sowing date are employed evenly into the establishing and testing model. PDC could be calculated as: 123 375 3 u nn n N + + = × (1)

Monitoring the plant density of Cotton with remotely sensed data Where Na was PDC per hectare(x10'plantl hm).n (i=1,2,3)was the number of plants in the sampling unit. 2.3 Remote Sensing Images and Pre-processing LANDSAT-5/TM images in April 22.May 24h June 9,June 25 and July 11h 2007 were he geo-locat romati wave as the asic imag transferring mo del uses 6s model and the erosol ical was done u The atmospheric radia was inversed by the algorithm of dense dark vegetation.The pixel size was 28.5mx28.5m. 2.4 Vegetation Index EVI= P+C,x P-C+LXG Paur Pred (2 sw心 e:L is the C:were the canopy ba gain facto 110C60.C7 To further explore the suitable remotely sensed parameters.DEVI is the difference of EVI between the growth stage and the bared soil stage: DEVI=EVI-EVI 3 2.5 The Evaluation of Estimation Accuracy The Relative Extremenum Predication Error(REPE)is used to evaluate the estimation accuracy of PDC,REPE is calculated as RMSE REPE No max-No min g (N.p-N. RMSE (5) n-1 NN is the maximum and minimum value in the testing models,respec- tively:Np,Nt is the estimation and measured value,respectively. 3 Results 3.1 PDC Estimation with EVI PDC estimation ba

Monitoring the Plant Density of Cotton with Remotely Sensed Data 93 Where Nu was PDC per hectare(×105 plant hm-2), i n (i = 1, 2, 3) was the number of plants in the sampling unit. 2.3 Remote Sensing Images and Pre-processing LANDSAT-5/TM images in April 22th, May 24th, June 9th, June 25th, and July 11th, 2007 were used. The geo-location was done by ENVI4.2 with the geo-referenced panchromatic waveband image from LANDSAT-7 as the basic image. The atmos￾pheric correction was done using the AgRSIS software. The atmospheric radiation transferring model uses 6S model and the aerosol optical thickness was inversed by the algorithm of dense dark vegetation. The pixel size was 28.5m×28.5m. 2.4 Vegetation Index 1 2 nir red nir red blue EVI G CC L ρ ρ ρρ ρ − = × +× − × + (2) Where ρnir, ρred, ρbule was reflectance in near-infrared (NIR), red and blue waveband respectively. C1 and C2 were the correction coefficients of atmospheric resistance in red and blue; L is the canopy background brightness correction factor and G is the gain factor. The coefficients adopted in the EVI algorithm are, L=1.0, C1=6.0, C2=7.5 and G (gain factor) =2.5 (Huete et al., 1994; Huete et al., 1997). To further explore the suitable remotely sensed parameters, DEVI is the difference of EVI between the growth stage and the bared soil stage: DEVI EVI EVI = −i e (3) 2.5 The Evaluation of Estimation Accuracy The Relative Extremenum Predication Error (REPE) is used to evaluate the estimation accuracy of PDC, REPE is calculated as max min U U RMSE REPE N N = − (4) 2 1 ( ) 1 n u u i N p Nt RMSE n = − = − ∑ " (5) Nu max, Nu min is the maximum and minimum value in the testing models, respec￾tively; Nup, Nut is the estimation and measured value, respectively. 3 Results 3.1 PDC Estimation with EVI PDC estimation based on the data from three emergence stages. As shown in Fig.2, from May 24th to July 11th, EVI rose with the cotton growing. The average EVI

94 J.Bai,J.Li,and S.Li for each growing stage is 0.11,0.27.0.53 and 0.82.On May 24,the range of EVI distribution is narrow and it shows insignificant linear relationship between EVI and PDC.However,EVI distribute more widely on June 9 and June 25,and it shows the obvious linear relationship.EVI on June 25th is higher than on June 9,the correlation degree isn't kept increasing tendency as time goes on,EVI difference on July 11 is less than on June 25th and the linear isn't significant.From the above analysis,we could deduce the emergence time caused the difference of seedling size,which affects the linear between PDC and EVI.Furthermore,the mixed pixels with non-cotton information decrease the relation on May 24%h and July 11th,and the relations aren't nearly interrupted on July 9th and July 25 coefficients bety .the estima June9:N=2.61EV+1.22 R2=0.23 (6) June25:N=2.41EV+0.65 R2=0.30 。 · 12 A. 26 26 24 。 14 2 g。 04045 Fig.2.Relationship between PDC and EVI

94 J. Bai, J. Li, and S. Li for each growing stage is 0.11, 0.27, 0.53 and 0.82. On May 24th, the range of EVI distribution is narrow and it shows insignificant linear relationship between EVI and PDC. However, EVI distribute more widely on June 9th and June 25th, and it shows the obvious linear relationship. EVI on June 25th is higher than on June 9th, the correlation degree isn’t kept increasing tendency as time goes on, EVI difference on July 11th is less than on June 25th and the linear isn’t significant. From the above analysis, we could deduce the emergence time caused the difference of seedling size, which affects the linear between PDC and EVI. Furthermore, the mixed pixels with non-cotton information decrease the relation on May 24th and July 11th, and the relations aren’t nearly interrupted on July 9th and July 25th. The correlation coefficients between PDC and EVI haven’t reached significant level on May 24th and July 11th and reached significant level on June 9th and June 25th, and correspondingly R2 is 0.23 on June 9 and 0.30 on June 25. Therefore, the estima￾tion models on PDC with EVI as below: June 9th : Nu=2.61EVI+1.22 R2 =0.23** (6) June 25th : Nu =2.41EVI+0.65 R2 =0.30** (7) R2 1, R2 2 showed the coefficient of determination from three sowing dates and middle sowing date. △ the late sowing; × the early sowing; ■ the middle sowing; ◇ at the edge of farmland, and the same as below. Fig. 2. Relationship between PDC and EVI

Monitoring the plant density of cotton with remotely sensed data 多o”based jon the sats from ent of the r J bu June9:N=4.364EVI+0.686 R2-0.75 8 June25h:N.=3.185EV7+0.148 R2-0.63 9 i .15 2 : w Fig.3.Relationship between PDC and DEVI 3.2 PDC Estimation with DEVI get DEVI on the corresponding time.It is shown in Fig.3.the determination coefficients of the linear equation between PDC and DEVI are 0.34 reaching the great significant level on May 24,and on the contrary the linear disappears when EVI estimated PDC on May 24.In addition,the determination coefficients rise from 0.23 to 0.27 on June 9 and from 0.3 to 0.34 on June 25.And the linear still disappears on July 11.The estimation models on PDC with DEVI as below:

Monitoring the Plant Density of Cotton with Remotely Sensed Data 95 PDC estimation based on the data from middle emergence stage. As shown in Fig.2, coefficient of the linear equation between PDC and EVI rise to 0.75 for June 9th and 0.63 for June 25th when the data from middle emergence stage is only used, but unfortunately, the determination coefficient hasn’t still reach the significant level on May 24th and July 11th. The estimation models on PDC with EVI as below: June 9th : Nu = 4.364EVI+0.686 R2 =0.75** (8) June 25th : Nu = 3.185EVI+0.148 R2 =0.63** (9) Fig. 3. Relationship between PDC and DEVI 3.2 PDC Estimation with DEVI PDC estimation based on the data from three emergence stages. EVI on May 24th, June 9th, June 25th and the July 11th, respectively minus EVI on April 22th get DEVI on the corresponding time. It is shown in Fig.3, the determination coefficients of the linear equation between PDC and DEVI are 0.34 reaching the great significant level on May 24th, and on the contrary the linear disappears when EVI estimated PDC on May 24th. In addition, the determination coefficients rise from 0.23 to 0.27 on June 9 and from 0.3 to 0.34 on June 25th. And the linear still disappears on July 11th. The estimation models on PDC with DEVI as below:

96 J.Bai.J.Li,and S.Li May24:N=12.13DEVM+1.13 R2=0.34 (10) June 9th:N =2.82DEV1+1.31 R2=0.27 (11) June 25th N.2.65DEV/+0.64 R2=0.34 (12) PDC estimation based on the data from middle c n PDC and DEvI from .From Fig,the mized on May Juneand and the dete th extremel leve co046 0.76 Unfortu esn't reach the significan May24:N,=10.95DEV+l.147 R2=0.46* (13) June9h:N=4.114DEV7+0.980 R2=0.76* (14 June25:N.=3.174DEV7+0.329 R2=0.67 (15) Table 1.Testing about the different models of estimating PDC Time The estimating The regress equations between measured RMSE methods REPE and estimated values y=0.184x+1.507R2=0.181 0.301 27.1 May 24 y=0.454x+0.939(R2=0.418 0.248 22.7 I y=0.252x+1.393(R2=0.293" 0.280 25.7 白 y=0.711x+0.446(R2=0.656” 0.205 18.7 Ⅲ y=0.282x+1.342(R2=0.274) 0.282 25.4 N y=0.727x+0.408(R2=0.660” 0.207 18.7 y=0.322x+1.279(R2=0.466 0.253 23.2 y=0.559x+0.718(R2=0.664” 0.208 19.1 June 25 y=0.307x+1.317R2=0.431 0.259 23.3 y=0.599x+0.636(R2=0.647” 0.213 19.2 Noting:/shows the testing result based on EVI from the early,middling and late emergence stage: shows the testing result based on EVI form the middling emergence stage:I shows the testing result base .midling and late emrgence stage:shows the testing based DEV

96 J. Bai, J. Li, and S. Li May 24th : Nu = 12.13DEVI+1.13 R2 = 0.34 ** (10) June 9th : Nu = 2.82DEVI+1.31 R2 = 0.27** (11) June 25th : Nu = 2.65DEVI+0.64 R2 = 0.34** (12) PDC estimation based on the data from middle emergence stage. From Fig.3, the linear correlations between PDC and DEVI from middle emergence stage are opti￾mized on May 24th, June 9th and June 25th, and the determination coefficients reach the extremely significant level and rise to 0.46, 0.76 and 0.67 distinctly. Unfortu￾nately, the coefficient on July 11th doesn’t reach the significant level. The estimation models on PDC with DEVI as below: May 24th : Nu = 10.95 DEVI+1.147 R2 = 0.46** (13) June 9th : Nu = 4.114 DEVI+0.980 R2 = 0.76** (14) June 25th : Nu = 3.174 DEVI+0.329 R2 = 0.67** (15) Table 1. Testing about the different models of estimating PDC Time The estimating methods The regress equations between measured and estimated values RMSE REPE Ⅲ y = 0.184x+1.507 (R2 = 0.181* ) 0.301 27.1 May 24 Ⅳ y = 0.454x+0.939 (R2 = 0.418**) 0.248 22.7 Ⅰ y = 0.252x+1.393 (R2 = 0.293**) 0.280 25.7 Ⅱ y = 0.711x+0.446 (R2 = 0.656**) 0.205 18.7 Ⅲ y = 0.282x+1.342 (R2 = 0.274**) 0.282 25.4 June 9 Ⅳ y = 0.727x+0.408 (R2 = 0.660**) 0.207 18.7 Ⅰ y = 0.322x+1.279 (R2 = 0.466**) 0.253 23.2 Ⅱ y = 0.559x+0.718 (R2 = 0.664**) 0.208 19.1 Ⅲ y = 0.307x+1.317 (R2 = 0.431**) 0.259 23.3 June 25 Ⅳ y = 0.599x+0.636 (R2 = 0.647**) 0.213 19.2 Noting: Ⅰ shows the testing result based on EVI from the early, middling and late emergence stage; Ⅱ shows the testing result based on EVI form the middling emergence stage; Ⅲ shows the testing result based on DEVI from the early, middling and late emergence stage; Ⅳ shows the testing result based on DEVI from the middling emergence stage

Monitoring the Plant Density of Cotton with Remotely Sensed Data 3.3 The Validation of Estimated Models and Its Application The validation of estimation models.As Table 1 shows.the veracity of regression models from(6)to(14)is tested,all determination coefficients between the estimated and measured value are validated,but it is noticeable that the determination coeffi cients of I and IV from DEVI on May 24 also pas s through e significant testing The RMSE on PDC is 0.301x10 plant.hm and REPE is 27.1%,which indicates e mae D m RMsg0n plant.hm2,respectively,and REPE is 25.7%and 23.2%,respectively.However,the estimation accuracy of models is greatly improved on May 24th June h and June 25 when the data of middle emer nce stage is used to estimate PDC.Among them RMSE and REPE of /on June s the lowest value (x1000P90n Hectare s120 2 Fig.4.Monitoring PDC with remotely sensed data from LANDSAT-5 in the 148h farm of Xinjiang.china

Monitoring the Plant Density of Cotton with Remotely Sensed Data 97 3.3 The Validation of Estimated Models and Its Application The validation of estimation models. As Table 1 shows, the veracity of regression models from (6) to (14) is tested, all determination coefficients between the estimated and measured value are validated, but it is noticeable that the determination coeffi￾cients of Ⅲ and Ⅳ from DEVI on May 24th also pass through the significant testing. The RMSE on PDC is 0.301×105 plant﹒hm-2 and REPE is 27.1%, which indicates that DEVI would bring the monitoring time forward, on the contrary, when EVI is used to estimate PDC on June 9th and June 25th, RMSE is 0.280 and 0.253×105 plant﹒hm-2, respectively, and REPE is 25.7% and 23.2%, respectively. However, the estimation accuracy of models is greatly improved on May 24th, June 9th and June 25th when the data of middle emergence stage is used to estimate PDC. Among them, RMSE and REPE of Ⅰon June 9th is the lowest value. Fig. 4. Monitoring PDC with remotely sensed data from LANDSAT-5 in the 148th farm of Xinjiang, china

98 J.Bai,J.Li,and S.Li PDC:.x10plant.hm Fig.5.Scene from cotton farmland with the PDC difference The demonstration for montoring PDC with LANDSAT-5.It is conedom malysis abo ove that devi can e the pDC n ng time bring rd t late May, and th n on June and May 24 (Table D)have the double propertieswth the time and the higher veracity,and has the advantage of only using the single-time whch will enhance the feasibility of estimating PDC.Therefore. an be used as the optimal plan estimate PDC.Monitoring PDC in the 14 o n te tuue nly he amle farm in 2007 is approximate 15811.7 hectare,and 17.1%for s1.20x10'plant.hm 10.2%for1.20-1.35x105 plant.hm2,13.1%for1.35-1.50x105 plant.hm2,13.6% for 1.50-1.65x105 plant. hm2,12.6%for1.65-1.80x103 plant.hm2,10.9%fo 1.80-1.95x10 plant.hm2,8.7%for1.95-2.10x105 plant.hm2,6.09%for 2.10-2.25x105 plant.hm2,3.8%for2.25-2.40x10 plant.hm2,4.0%for≥40x105 plant.hmFarmlands with the different PDCcan be distinguished (Fig.5) 4 Discussion The above Firstly a e afet the gh,the reflecta vegetation wa the s rties dom fference of pixel reflectance and it was difficult to retriev PDC accurately. Secondly,the cotton co erage m crea sed from zero to almost 100 percent when the growth stage developed from the seedling stage to full-flowering period.Under the condition of the high cotton cover age,the contribution of soil to the pixel reflectance decreased,and the saturation of vegetation index happened. As shown in the results on May 24EVI is not sensitive to the PDC difference sig- nificantly.DEVI could remove the noise of the soil background and highlight the

98 J. Bai, J. Li, and S. Li Fig. 5. Scene from cotton farmland with the PDC difference The demonstration for monitoring PDC with LANDSAT-5. It is concluded from analysis above that DEVI can make the PDC monitoring time bring forward to the late May, and the accuracy on June 25th is higher than on June 9th and May 24th. However, Ⅰand Ⅲ on June 9th (Table 1) have the double properties with the earlier time and the higher veracity, andⅠ has the advantage of only using the single-time image, which will enhance the feasibility of estimating PDC. Therefore, Ⅰ on June 9th can be used as the optimal plan to estimate PDC. Monitoring PDC in the 148th farm by Ⅰ , the results from Fig.4 indicate that PDC has the strongly spatial correlation and the distributed continuity. The farmland area for cotton in the 148th farm in 2007 is approximate 15811.7 hectare, and 17.1% for ≤1.20×105 plant﹒hm-2, 10.2% for 1.20~1.35×105 plant﹒hm-2, 13.1% for 1.35~1.50×105 plant﹒hm-2, 13.6% for 1.50~1.65×105 plant﹒hm-2, 12.6% for 1.65~1.80×105 plant﹒hm-2, 10.9% for 1.80~1.95×105 plant ﹒ hm-2, 8.7% for 1.95~2.10×105 plant ﹒ hm-2, 6.0% for 2.10~2.25×105 plant﹒hm-2, 3.8% for 2.25~2.40×10 plant﹒hm-2, 4.0% for ≥40×105 plant﹒hm-2. Farmlands with the different PDC can be distinguished (Fig.5). 4 Discussion The above analysis shows that two situations may affect the monitoring accuracy. Firstly, when the coverage fraction of bared soil was high, the reflectance signal of vegetation was weak, the soil properties dominated the difference of pixel reflectance, and it was difficult to retrieve PDC accurately. Secondly, the cotton coverage in￾creased from zero to almost 100 percent when the growth stage developed from the seedling stage to full-flowering period. Under the condition of the high cotton cover￾age, the contribution of soil to the pixel reflectance decreased, and the saturation of vegetation index happened. As shown in the results on May 24th EVI is not sensitive to the PDC difference sig￾nificantly. DEVI could remove the noise of the soil background and highlight the

Monitoring the Plant Density of Cotton with Remotely Sensed Data 99 vege 24 itoring time. on May tton coverage proportio the green index of cotton canopy population The analysis also indicated that it was impossible to estimate PDC effectively in any way when the cotton canopy population in the full-blooming stage had the highest land cover,and especially the cotton coverage rose to over 80%,and the canopy population was fuzzy between pixels because of the saturation phenomenon.More over,it was difficult to monitor PDC by remote sensing because of another reason The canopy population was controlled by PDC before the flowering stage,and yet after the cotton grown into the full-flowering stage,the compensation effect of cotton plants appeared by a series of management measures of top-ending,fertilizer-water and chemical controlling,and so the population with the low PDC might develop into the big one,that meant the population with the low PDC had the higher leaf area index and biomass than the r pulation with the high PDC.As a result,the cotton canopy population was simultaneously controlled by the plant size and PDC.There fore PDC couldn't monitored successfully after the full-flo ering stage (at July 11) which was decided by the defect of remotely sensed technolo and the characteristic of cotton growth. The of which ce e the diff of e plant si t information of sowing date should monitoring accuracy. 5 Conclusions This pape ren The ed that be use images scale altho wing date and th ses can regio m satu on th m The optimal growing stage accuracy【O early-f ering stage especially on acc time required by the farmlan simultaneously. The method to monitor PDC by satellite images could be a potential way to investigate PDC at the regional scale. Acknowledgments 6AA Technology Innovation Foundation of Shihezi university(ZRKX200748)

Monitoring the Plant Density of Cotton with Remotely Sensed Data 99 vegetation information. We construct the relationship between PDC and DEVI to monitor PDC, which brings up the PDC monitoring time. Compared with on May 24th, the cotton coverage on June 9th and June 25th was increased remarkably, the proportion of cotton signal rose gradually, and the PDC could be monitored through the green index of cotton canopy population. The analysis also indicated that it was impossible to estimate PDC effectively in any way when the cotton canopy population in the full-blooming stage had the highest land cover, and especially the cotton coverage rose to over 80%, and the canopy population was fuzzy between pixels because of the saturation phenomenon. More￾over, it was difficult to monitor PDC by remote sensing because of another reason. The canopy population was controlled by PDC before the flowering stage, and yet after the cotton grown into the full-flowering stage, the compensation effect of cotton plants appeared by a series of management measures of top-ending, fertilizer-water and chemical controlling, and so the population with the low PDC might develop into the big one, that meant the population with the low PDC had the higher leaf area index and biomass than the population with the high PDC. As a result, the cotton canopy population was simultaneously controlled by the plant size and PDC. There￾fore, PDC couldn’t monitored successfully after the full-flowering stage (at July 11th), which was decided by the defect of remotely sensed technology and the characteristic of cotton growth. The cotton farmers couldn’t seed at the same time because of the limitations of human and mechanism, which could cause the difference of cotton plant size in the early stage of cotton growth, the result added the difficulty of estimating PDC with the high accuracy, the information of sowing date should be known to increase the monitoring accuracy. 5 Conclusions This paper proposed a method to monitor PDC by satellite remote sensing images. The results showed that satellite images can be used to monitor PDC at the regional scale although the sowing date and the spectrum saturation have an influence on the accuracy of PDC monitoring. The analysis indicates that the optimal PDC monitoring period is late May by DEVI. The optimal growing stage with the highest estimation accuracy to monitor PDC is from the squaring stage to the early-flowering stage, especially the late-squaring stage could be satisfied with the estimation accuracy and time required by the farmland management measurements simultaneously. The method to monitor PDC by satellite images could be a potential way to investigate PDC at the regional scale. Acknowledgments The research was supported by the National High Technology Research and Devel￾opment Program of China (2006AAl0302, 2006AA102207, and 2006AA102272), National Nature Science Foundation of China (40801143), and Nature Science and Technology Innovation Foundation of Shihezi university (ZRKX200748)