Using SPOT data and leaf area index for rice yield estimation in Egyptian Nile delta

Please download to get full document.

View again

of 9
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Information Report



Views: 11 | Pages: 9

Extension: PDF | Download: 0

Using SPOT data and leaf area index for rice yield estimation in Egyptian Nile delta
  ORIGINAL ARTICLE Using SPOT data and leaf area index for riceyield estimation in Egyptian Nile delta M. Aboelghar  a, * , S. Araf at  a , M. Abo Yousef   b , M. El-Shirbeny  a , S. Naeem  b ,A. Massoud  a , N. Saleh  a a National Authority for Remote Sensing and Space Sciences (NARSS), P.O. Box 1564, Alf Maskan, Cairo, Egypt b Rice Research Center, Agricultural Research Center (ARC),Giza, Egypt Received 20 June 2011; revised 8 September 2011; accepted 21 September 2011Available online 22 October 2011 KEYWORDS SPOT;Leaf area index;Vegetation Indices;Statistical models Abstract  The objective of the current work is to generate statistical empirical rice yield estimationmodels under the local conditions of the Egyptian Nile delta. The methodology is based on regress-ing measured yield with satellite derived spectral information or leaf area index (LAI). LAI fieldmeasurements and spectral information from SPOT data collected during two crop seasons areexamined against measured yield to generate the yield models. Near-infrared and red bands, six veg-etation indices and LAI of 100 points are used as the main inputs for the modeling process while 20points of the same are used for validation process. Nine models are generated and tested against theobserved yield. Comparing the generated models show relatively higher superiority of (LAI-yield)and (infrared-yield) models over the rest of the models with (0.061) and (0.090) as a standard errorof estimate and (0.945) and (0.883) as coefficient of determinations between modeled and observedyield. The models are applicable a month before harvest for similar regions with same conditions.   2011 National Authority for Remote Sensing and Space Sciences.Production and hosting by Elsevier B.V. All rights reserved. 1. Introduction Accurate andtimely assessment of cropyield isanessential pro-cesstoensuretheadequacyofanation’sfoodsupply.Itprovidespolicy makers, governmental agencies and commodity traderswith the necessary information to better manage harvest, stor-age, import/export, transportation and marketing activities.Thesoonerthisinformationisavailable,thelowertheeconomicrisk,thegreatertheefficiencyandtheincreasedreturnoninvest-ments Salazar et al., 2007). Synoptic observation and repetitivecoverage of the satellite remote sensing data is considered to bean effective methodology for real-time crop monitoring andcrop yield prediction on both local and regional scales. Spectralempirical modeling of such data is an important approach for *Corresponding author. Address: National Authority for RemoteSensing and Space Sciences (NARSS), 23 Joseph Tito St., P.O. Box,1564 Alf Maskan, El-Nozha El-Gedida, Cairo, Egypt.E-mail addresses:, maboelghar@ (M. Aboelghar).1110-9823   2011 National Authority for Remote Sensing and SpaceSciences. Production and hosting by Elsevier B.V. All rights reserved.Peer review under responsibility of National Authority for RemoteSensing and Space Sciences.doi:10.1016/j.ejrs.2011.09.002 Production and hosting by Elsevier The Egyptian Journal of Remote Sensing and Space Sciences (2011)  14 , 81–89 National Authority for Remote Sensing and Space Sciences The Egyptian Journal of Remote Sensing and SpaceSciences  cropyieldestimation.This approachhastheadvantageofbeingsimple with all the required data either readily available on a re-gional or global scale or easy to collect (Prasadet al., 2007).Thebasis of this modeling process is generating statistical relation-ships between spectral variables and crop yield or estimatingcrop yield through measurable bio-physical parameters thatare highly correlated with crop canopy vigor and structureand hence, correlated with the spectral characteristics of thecrops. The normalized difference vegetation index (NDVI) isknown to be able to respond to changes in the amount of greenbiomass, chlorophyll content, and canopy water stress. It iseffective in predicting surface properties when the vegetationcanopy is not too dense or too sparse (Liang, 2004). The rela-tionship between NDVI and production has been confirmedby various field experiments (Prince and Justice, 1991; Rasmus- sen, 1992). It has a direct strong correlation with leaf area index(LAI), biomass and vegetation cover (Tucker, 1979; Holben et al., 1980; (Ahlrichs and Bauer, 1983; Nemani and Running, 1989; Wiegand et al., 1990). These parameters drive the crop production,andarelargelyinfluencedbyvariationsinsoilfertil-ity (Hinzman et al., 1986), soil moisture (Daughtry et al., 1980; Teng, 1990), planting date (Crist, 1984) and crop density (Aase and Siddoway, 1981). They are also related to the crop yieldassuming the absence of significant stresses during the head-ing/filling stages (Hartfield, 1983; Wiegand and Richardson, 1990). The general drawback of most methods using statisticalrelationships between vegetation indices (VI), leaf area index(LAI)andcropyieldisthattheyhaveastrongempiricalcharac-ter (Groten, 1993; Sharma et al., 1993). No models unless devel- oped and tested locally are suitable for local use (Shresthan andNaikaset, 2003). Therefore, the main objective of the currentstudy is to use SPOT satellite imagery and LAI field measure- Figure 1  Location map of Sakha area. 82 M. Aboelghar et al.  ments to generate statistical rice yield estimation models underthe local conditions of the Nile delta, north Egypt. 2. Materials and methods Two SPOT4 (HRVIR) images with 20 m spatial resolution,26 days temporal resolution and four spectral bands: band 1 – green (0.50–0.59  l m), band 2 – red (0.61–0.68  l m), band 3 – near infrared (0.78–0.89  l m) and band 4 – shortwave infra-red (1.58–1.75  l m) acquired during the rice seasons in Augustof 2008 and 2009, are used in the current study. Vegetationindices are calculated from green, red and near-infrared bandsand used in the current observation. Observed crop yield aswell as leaf area index (LAI) field measurements have beengathered throughout the same rice seasons. Then, all the previ-ous parameters are used as inputs for the models. The follow-ing sub-sections explain the applied methods to collect theseinputs. 2.1. Study area and satellite data pre-processing ThestudytookplaceinaleadingriceproducingareainKafrEl-Sheikh Governorate (Sakha region), north Egypt. It is part of the Egyptian Nile delta that is characterized by extensive ricecultivation. It is located between 31  06 0 40 00 and 31  06 0 0 00 Northand 30  54 0 30 00 and 30  55 0 60 00 East (Fig. 1). The total area of the observation site is 24,000 m 2 (2.4 ha), 1.2 ha are cultivatedby the variety (Sakha 102) while the rest is cultivated by (Sakha104). These two varieties are the most common Egyptian ricelocal varieties in the Nile delta. Both the varieties are Egyptianshort-grain varieties with national average yield 9.1–9.6 ton/ha  1 , blast-resistant, early maturing, need 125 days from sowingtill harvest, with high milling output (72%). The highest andlowest air temperature is observed with the highest is duringthe month of August while the lowest is during the month of April. The maximum relative humidity is around 82% whilethe lowest is about 43%. Fig. 2 and Table 1 show the monthly average of air temperature and relative humidity during thetwo rice seasons of 2008 and 2009 from May to September.The source of the meteorological data is the meteorological sta-tion of Sakha experimental agricultural research station. Thesoiltypeinthestudyareaisheavyclaysoil.TwoSPOT4imageryofthetestsiteacquiredin(August24,2008andAugust23,2009)(K111/J287) during the rice seasons are used. Geometrictransformation is carried out using selected ground controlpoints (GCPs) to (Lat/Long) projection system. The root meansquare (RMS) error below half pixel (0.5) is accepted with thefirst-degree polynomial and nearest neighbor resembling algo-rithm technique. 2.2. Generation of estimator variables of crop yield  The observation points included most of variations in fieldconditions and crop production parameters, so that the modelcould be used with different crop production conditions thatare presented in the current study area. Based on a grid systemdesigned by the research team, the study area is divided into 60parcels; 30 parcels are cultivated by each variety and each par-cel (20  ·  20 m) that represents a single SPOT pixel is fixed asone plot of measurements (Fig. 3). The location of the centersquare meter of each plot is recorded using global positioningsystem (GPS). Within each parcel, five LAI measurements arecollected and the average is recorded. At the end of each riceseason, a harvester is used to measure the yield of each parceland the average of yield (kg/m 2 ) is calculated. Finally, thewhole dataset is completed as: 60 points for each rice seasonof (rice yield, LAI measurements, spectral variables includingred and near-infrared bands represented as digital numbersand six vegetation indices). The data of the two years are com-bined in one dataset and regression analysis between observedyield and each individual variable is performed. One hundredpoints from the two seasons are randomly chosen for modeling 01020304050607080May-08Jun-08Jul-08 Aug-08Sep-08Oct-08Nov-08Dec-08Jan-09Feb-09Mar-09Apr-09May-09Jun-09Jul-09 Aug-09Sep-09  Average Air temp  Average RH% Figure 2  The recorded air temperature and relative humidity of Sakha region.   Table 1  The average air temperature and relative humidityduring the two rice seasons of 2008 and 2009. Date Average air temp. Average RH%May-08 20 67Jun-08 24 66Jul-08 24 68Aug-08 25 70Sep-08 24 63May-09 21 59Jun-09 26 63Jul-09 27 65Aug-09 26 66Sep-09 26 62 Using SPOT data and leaf area index for rice yield estimation in Egyptian Nile delta 83  process while 20 points from the two seasons are used to val-idate the models.LAI-2000 plant canopy analyzer is used to measure LAI foreach point during the two seasons. This device calculates LAIand other canopy structure attributes from radiation measure-ments made with a (fish-eye) optical sensor (148   field-of-view). Measurements made above the canopy and belowthe canopy are used to determine canopy light interception atfiveangles,fromwhichLAIiscomputedusingamodelofradio-activetransferinvegetativecanopy.Measurementsaremadebypositioning the optical sensor and pressing a button; data areautomatically logged into the control unit for storage and LAIcalculations. Multiple below-canopy readings and the fish-eyefield-of-view assure that LAI calculations are based on a largesample of the foliage canopy. After collecting above and be-low-canopy measurements, the control unit performs all calcu-lations and the results are available for immediate on-siteinspection.Six vegetation indices are tested in the current study; nor-malized difference vegetation index (NDVI), soil adjusted veg-etation index (SAVI), green vegetation index (GVI), infraredpercentage vegetation Index (IPVI), ratio vegetation index(RVI) and difference vegetation Index (DVI). NDVI is deter-mined using the red (R) and near-infrared (NIR) bands of agiven image (Rouse et al., 1973) and is expressed as follows: NDVI   ¼  q ir    q r q ir  þ  q r ð 1 Þ where  q r  and  q ir  are spectral reflectance from the red and NIR-band images, respectively. The green vegetation index (GVI) isdetermined using GVI   ¼ q ir    q  g q ir  þ  q  g ð 2 Þ where  q  g  and  q ir  are spectral reflectance from the Green andNIR-band images, respectively (Panda et al., 2010). Soiladjusted vegetation index (SAVI) is determined as SAVI   ¼  q ir    q r q ir  þ  q r  þ L     ð 1  þ L Þ ð 3 Þ where  q r  and  q ir  are spectral reflectance from the red andNIR-band images, respectively and L is an optimal adjustmentfactor. Huete (1988) defined the optimal adjustment factor of  L = 0.25 to be considered for higher vegetation density inthe field,  L = 0.5 for intermediate vegetation density, and L = 1 for the low vegetation density. He suggested that SAVI( L = 0.5) successfully minimized the effect of soil variations ingreen vegetation compared to NDVI. Based on our observa-tions, we considered canopy cover of the rice crop in the fieldas high dense during the satellite images acquisition time in2008 and 2009. Thus, 0.25 is used as the (L) factor using theHuete strategy of selecting the (L) factor, which is also sup-ported by Thiam and Eastmen (1999). IPVI is the infrared per-centage vegetation index which was first described by Crippen(1990). He found that the subtraction of the red in the numer-ator as is done with SAVI to be irrelevant, and proposed thisindex as a way of improving calculation speed. It is restrictedto values between 0 and 1 and eliminates the conceptualstrangeness of negative values for vegetation indices. It is cal-culated as explained in the following equation: IPVI   ¼  NIRNIR þ R ð 4 Þ Figure 3  Gridding system for field measurements and data collection. 84 M. Aboelghar et al.  RVI is the ratio vegetation index which is first described byJordan (1969). It is used to eliminate various albedo effects andit is calculated as shown in the following equation: RVI  ¼ NIR Re d   ð 5 Þ DVI is the difference vegetation index, which is describedby Richardson and Everitt (1992) as follows: DVI  ¼ NIR   Re d   ð 6 Þ All VI values and LAI are considered for the regressionanalysis and integrated yield prediction models of the two riceseasons are produced. The explanatory power of the indepen-dent variables in the model and eventual prediction accuracyof the generated models can be assessed with statistical param-eters such as standard error of estimate (SEE), ( t ) test, andcoefficient of determination ( R 2 ). 3. Results and discussion Nine statistical yield prediction models are produced. Figs. 4– 12 show the trendline and the correlation coefficients for allgenerated models. Basically, linear regression is the best modelthat represents the relation between observed yield and eachindividual estimator. It is found that the coefficient of determi-nation ( R 2 ) of all models is around (0.8) except for (GVI-yield)model that showed relatively low ones. The models are exam-ined and the relative superiority over the generated models isdecided through computing Standard Error of Estimate(SEE) and (t) test and the squared coefficient of determinationsbetween modeled and observed yield ( R 2 ). Table 2 shows thegenerated models and the validation results and Figs. 13–21show the trendline and the squared coefficient of determina-tion between observed and modeled yield.The main objective of the current study is to generate riceyield prediction models for the most common rice local varie-ties in the Egyptian Nile delta (Sakha 102) and (Sakha 104).The data of the two seasons were combined in one dataset,part of this dataset was used to generate the models whilethe other part was used to validate these models. The idea of combining the two rice seasons of the two varieties in one data-set is to generate applicable rice yield prediction models thatcould be applied through satellite remote sensing data withacceptable accuracy. The first step of yield prediction is to iso-late the investigate crop from the other land cover types. Clas-sifying different crop types in Egyptian Nile delta is stillproblematic and classifying different crop varieties within thesame crop type is almost not possible under the current agri-cultural conditions in Egyptian Nile delta using available satel-lite imagery. The models were generated using two seasons andtwo varieties to cover all possible minimal variations and to benot limited to the conditions of one season. It is expected thatthese models could be applicable in future through any type of high resolution satellite data. The generated models could beapplied to predict the yield of these two common rice varietiesunder normal agricultural practices and common environmen-tal conditions in Egyptian Nile delta. Concerning the results of the models, it is found that the statistical relationship betweenmiddle infrared, green bands and yield showed low accuracy,so, these two variables are excluded from further analysis.Among the rest of the examined variables, LAI is the best esti-mator that gave the highest ( R 2 ) and the lowest (SEE). The cal-culated yields from all variables are comparable except for(GVI) that showed lowest validation result. This may be be-cause of the calculation of (GVI) as a ratio between green bandand (IR) and as mentioned above that green band showed rel-atively low correlation with yield. Among spectral variables,(IR) band is the best estimator with the highest validation re-sult. The calculated ( t ) is less than the tabeled ( t ) with all vari-ables that reflects insignificant difference between modeled andobserved yields. (Table 1). Among the three statistical methodthat are used to validate the generated models. SEE and R 2 aremore powerful than ( t ) test method for showing the differencein prediction ability among the generated models.The generated models are site specific and limited to thearea and environment as well as to the date of the experiment.Following the rice growing circle, milking stage (when waterycontent of the grain turn to thick milky ones) is selected as thebest crop growing stage that is closely related to LAI and riceyield. This assumption is proved by the author after one-sea-son experiment during rice season of 2007 using LAI measure-ments that were gathered using (LAI-2000 plant canopyanalyzer device) and NDVI measurements that were gatheredfrom SPOT data and gathered also using (PlantPen NDVI300 m) device (Aboelghar et al., 2010). NDVI-meter measuresNDVI comparing the reflected light at two distinct wave-lengths, 660 and 740 nm. NDVI data collected by NDVI-meterduring different growing stages of rice were examined againstthe yield and against LAI. It is found that the NDVI of themilking stage of rice is the most correlated growing stage toyield and LAI. This is consistent with other results from the lit-erature (Murthy et al., 1996; Labus et al., 2002; Royo et al., 2003). The models are produced under the common soil, cli-matic and crop conditions of Egyptian Nile delta and could Table 2  The generated models and the validation results. Variable Regression equation  R 2 SEE  t  valuesNIR Yield = 0.0129IR  0.236 0.883 0.090 0.454Red Yield =   0.0923R + 3.5717 0.720 0.123   0.094NDVI Yield = 2.3606NDVI  0.2713 0.850 0.099 0.381RVI Yield = 0.2522RVI + 0.1131 0.823 0.098 0.307IPVI Yield = 4.721IPVI  2.632 0.850 0.099 0.372DVI Yield = 0.0116DVI + 0.21 0.874 0.090 0.506GVI Yield = 1.3675GVI + 0.468 0.786 0.210 0.037SAVI Yield = 1.8899SAVI  0.2697 0.850 0.099 0.369LAI Yield = 0.2846LAI  0.0764 0.945 0.061   0.419 * Yield kg/m 2 . Using SPOT data and leaf area index for rice yield estimation in Egyptian Nile delta 85
View more...
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!