Tarım Bilimleri Dergisi Journal of Agricultural Sciences Tar. Bil. Der. Dergi web sayfası: Journal homepage: www.agri.ankara.edu.tr/dergi www.agri.ankara.edu.tr/journal

Primary Production Estimation of Çankırı Province’s Rangelands Using Light Use Efficiency (LUE) Model with Satellite Data and AgrometShell Module

Ediz ÜNALa, İlhami BAYRAMİNb aField Crops Central Research Institute, Şehit Cem Ersever Street, No: 9-11, Ankara, TURKEY b

Ankara University, Faculty of Agriculture, Department of Soil Science and Plant Nutrition, 06110, Ankara, TURKEY 22 (2016) 555-565

ARTICLE INFO Research Article Corresponding Author: Ediz ÜNAL, E-mail: [email protected], Tel: +90 (312) 343 10 50 Received: 25 February 2014, Received in Revised Form: 14 August 2015, Accepted: 14 August 2015

ABSTRACT

In this study, monthly and annual gross primary production (GPP) of rangelands in Çankırı province for the period of 2000-2009 was calculated using light use efficiency (LUE) model with the inputs of satellite data and AgrometShell module. The average production of rangelands varied seasonally and annually (from 12630 to 37701 tons) and was approximately 17800 tons for the last ten years. The amount of rainfall and changing number of animal grazing in the region probably led to the variation. Model performance was tested with integrated normalized difference vegetation index (INDVI) approach which produced a moderate significant correlation (R2= 0.69, P<0.05) between LUE model gross primary productivity (GPP) output and INDVI values. On the other hand, comparison of modelled results of annual gross primary production (GPP) with above ground measurements, indicated that correlation between the variables were insignificant (r = 0.60, P>0.05 for 2008, r= 0.41, P>0.05 for 2009) due to some factors such as sampled plant type, scale differences between satellite data and ground sample size, and subjective sampling errors. This study indicates that LUE Model together with the inputs of AgrometShell module is suitable tool for estimation of rangeland primary production. Keywords: Biomass; Çankırı; Range; Remote sensing; Vegetation JOURNAL OF AGRICULTURAL SCIENCES JOURNAL OF AGRICULTURAL

—

Uydu Verisi ve AgrometShell Modülü ile Işık Kullanım Etkinliği (LUE) Modeli Kullanarak Çankırı İli Meralarının Birincil Üretim Tahmini ESER BİLGİSİ Araştırma Makalesi Sorumlu Yazar: Ediz ÜNAL, E-posta: [email protected], Tel: +90 (312) 343 10 50 Geliş Tarihi: 25 Şubat 2014, Düzeltmelerin Gelişi: 14 Ağustos 2015, Kabul: 14 Ağustos 2015

ÖZET

Bu çalışmada, Çankırı meralarının 2000-2009 arasındaki aylık ve yıllık toplam birincil üretimleri ışık kullanım etkinliği BİLİMLERİ DERGİSİ TARIM modeli ile hesaplanmıştır. Elde edilen bulgulara göre il sınırları içinde kalan meraların son on yıllık ortalama birincil üretimi yaklaşık 17877 tondur ve bu üretim hem mevsimsel hem de yıllık olarak (12630-37701 ton arası) değişkenlik Primary Production Estimation of Çankırı Province’s Rangelands Using Light Use Efficiency (LUE) Model..., Ünal & Bayramin

göstermektedir. Bu değişkenliğin ana sebepleri içinde bölgeye düşen yağış miktarı ve otlayan hayvan sayısındaki değişimler gösterilebilir. Model performansı, toplanmış normalize edilmiş farklılık indeksi (INDVI) ile test edilmiştir. Test sonucuna göre, INDVI ve toplam birincil üretim arasında orta seviyede bir ilişki (R2= 0.69, P<0.05) bulunmuştur. Uygulanan hassaslık analizi sonuçları, orantılı fotosentetik aktif radyasyonun (FPAR) en hassas değişken olduğunu göstermiştir. Diğer taraftan, modelden hesaplanan yıllık birincil üretim (GPP) değerleri ve arazi çalışmaları ile hesaplanan biyokütle arasında önemsiz ilişki bulunmuştur (r = 0.60, P>0.05, 2008; r= 0.41, P>0.05, 2009). Örneklenen bitki türleri, kişisel örnekleme hataları ve uydu verileri ile örnekleme alanı arasındaki ölçek farklılığı ilişki çıkmamasının ana sebepleri olarak gösterilebilir. Bu çalışma, AgrometShell girdilerini kullanan LUE modelinin meralarda birincil üretim miktarının tahmin edilmesinde iyi bir araç olduğunu ortaya koymaktadır. Anahtar Kelimeler: Biyokütle; Çankırı; Mera; Uzaktan algılama; Vejetasyon

© Ankara Üniversitesi Ziraat Fakültesi

1. Introduction Where; GPP, gross primary production (g m-2) converted to dry plant matter (DM) through Rangelands are important natural resources for photosynthesis; Ɛp, LUE factor (g DM MJ-2) livestock feeding and providing habitats for expressing conversion of light energy into dry biological diversity. Being a challenging issue, mass; Ɛ, unitless environmental stress factor; PAR, assessing productivity and gross primary production photosynthetic active radiation (MJ m-2) of sun light (GPP) of rangelands is important for their in the spectral range of 400-700 nm and FPAR, efficient management. The employment of remote fraction of absorbed light by vegetation. sensing which has been the most frequently used method utilizes two approaches; a) establishing The main objective of this study was to estimate relationships between spectral reflectance and and map annual and monthly GPPs of rangelands in biomass (Tucker et al 1983) and b) modelling Çankırı province using a light use efficiency model. GPP from remotely sensed spectral reflectance to The following steps were also achieved by reaching estimate the amount of absorbed photosynthetically the main objective; 1) a validation of calculated active1. Introduction radiation (APAR) (Brogaard et al 2004). GPP by ground data and 2) performing a sensitivity analysis of the model variables. A light use efficiency (LUE) approach is widely appliedRangelands concept are forimportant modelling natural the GPPresources (Goetz for et livestockal feeding and providing habitats for biological 1999;diversity. Hilker Being et al a 2008), challenging and expresses issue, assessingthe GPP asproductivity 2. Material and grossand Methodsprimary production (GPP) of arang productelands of is theimportant APAR. forThis their approach efficient is management.the main The employment of remote sensing that could be componentthe most frequently of the current used methodstudy based utilizes on twothe approaches;idea 2.1. Studya) establishing area relationships between spectral thatreflectance biological and productionbiomass (Tucker is directly et alproportional 1983) and b)The mo studydelling area GPP is from Çankırı remotely province sensed located spectral in the toreflectance the photosynthetically to estimate the amountactive ofradiation absorbed (PAR) photosynthetically Central Anatolia, active radiationTurkey (Figure (APAR) 1). (Brogaard The landscape et al absorbed2004). A bylight the use green efficiency vegetation (LUE) (Monteith approach 1972). is widelyof applied Çankırı concept is mostlyfor modelling mountainous the GPP with(Goetz hilly et al 19The99; revisedHilker et model al 2008), of and Seaquist expresses et al the (2003) GPP as atopography product of thecovering APAR. approximately This approach 60%is the of main the presentedcomponent in of Equationthe current 1 studywas usedbased in on thisthe ideastudy, that biologicalprovince. Theproduction average is elevation directly proportionalis 723 m with to hillsthe becausephotosynthetically it includes active environmental radiation effects(PAR) absorbed(drought, by theand green plateaus vegetation interrupted (Monteith by Ilgaz 1972). Mountain ranges. temperature, pollution, nutrient deficiency, illness Continental climate dominates the region with long The revised model of Seaquist et al (2003) presented in Equation 1 was used in this study, because it etc.) as stress factors which play an important role term average rainfall of 500 mm. The land is mostly includes environmental effects (drought, temperature, pollution, nutrient deficiency, illness etc.) as stress in biological activities of the plant and hence in the bare on the hills and plateaus, while the mountains factors which play an important role in biological activities of the plant and hence in the GPP. are covered with coniferous trees. Foot lands are GPP. n generally used for grain cultivation. Soil erosion is GPP εp ε FPAR PAR (1) prevalent across the province, which explains (1)why i 1 the non-cropped lands are used mostly as rangelands.

Where; GPP, gross primary production (g m-2) converted to dry plant matter (DM) through -2 556photosynthesis; Ɛp, LUETarım factor Bilimleri (g DM Dergisi MJ ) expressing – Journal conversionof Agricultural of light Sciences energy into 22 (2016) dry mass; 555-565 Ɛ, unitless environmental stress factor; PAR, photosynthetic active radiation (MJ m-2) of sun light in the spectral range of 400-700 nm and FPAR, fraction of absorbed light by vegetation.

The main objective of this study was to estimate and map annual and monthly GPPs of rangelands in Çankırı province using a light use efficiency model. The following steps were also achieved by reaching the main objective; 1) a validation of calculated GPP by ground data and 2) performing a sensitivity analysis of the model variables.

2. Material and Methods

2.1. Study area

The study area is Çankırı province located in the Central Anatolia, Turkey (Figure 1). The landscape of Çankırı is mostly mountainous with hilly topography covering approximately 60% of the province. The average elevation is 723 m with hills and plateaus interrupted by Ilgaz Mountain ranges. Continental climate dominates the region with long term average rainfall of 500 mm. The land is mostly bare on the hills and plateaus, while the mountains are covered with coniferous trees. Foot lands are generally used for grain cultivation. Soil erosion is prevalent across the province, which explains why the non-cropped lands are used mostly as rangelands. These rangelands have some characteristics of desert plant species, resulting from low rainfall of 300-500 mm and over grazing (Ketenoğlu et al 1983). The botanical composition of rangelands generally consists of short grass (Festuca sp, Poa sp.), broad leaves (Medicago sp.) and various thorny species (Astragalus sp.) (Kurt et al 2006).

In this study, the GPP was calculated both monthly and annually for 10 years (2000-2009) for the active vegetation period of a growing season that starts with first leaf appearance and ends with senescence. The growing season was divided into 10-day periods (dekad) totaling to 14 dekads for each year. The first dekad starts on March 20th and ends on March 31st. The last dekad spans from August 1st to August 10th.

2 Uydu Verisi ve AgrometShell Modülü ile Işık Kullanım Etkinliği (LUE) Modeli Kullanarak Çankırı İli Meralarının..., Ünal & Bayramin

These rangelands have some characteristics of desert 2004) and includes a database of weather data as plant species, resulting from low rainfall of 300- 10-day average of temperature (°C), solar radiation 500 mm and over grazing (Ketenoğlu et al 1983). (Cal m-2), wind speed (m s-1) and 10-day sum of The botanical composition of rangelands generally rainfall (mm) as well as crop specific information consists of short grass (Festuca sp, Poa sp.), broad such as crop type, crop cycle length, irrigation, leaves (Medicago sp.) and various thorny species etc. AgrometShell module runs a water balance (Astragalus sp.) (Kurt et al 2006). model to produce actual evapotranspiration (AET) In this study, the GPP was calculated both and potential evapotranspiration (PET) which monthly and annually for 10 years (2000-2009) for were used later for water stress calculation in LUE the active vegetation period of a growing season model. Calculated PET and AET values were then that starts with first leaf appearance and ends with converted into grid format by the inverse distance senescence. The growing season was divided into weighting (IDW) kriging interpolation method (Ha 10-day periods (dekad) totaling to 14 dekads for et al 2011) to generate surfaces with same cell size each year. The first dekad starts on March 20th and of 1 km of NDVI data. ends on March 31st. The last dekad spans from August 1st to August 10th.

Figure 2- Meteorological stations and survey points Şekil 2- Meteoroloji istasyonları ve sörvey noktaları Figure 1- Location of study area and land cover classes NDVI images were used as satellite data, Şekil 1- Çalışma alanı ve arazi örtüsü sınıfları which is widely regarded as measurements of surface vegetation condition and dynamics and indicates the greenness of live vegetation (Huete 2.2. Used data et al 2002). NDVI is numerically calculated from The meteorological data stored in AgrometShell (NIR-RED/NIR+RED). RED and NIR stand for module retrieved from 36 automatic weather the spectral reflectance measurements acquired stations (AWS) distributed over the province of in red and near-infrared regions of the spectrum, Çankırı and neighboring provinces (Figure 2). The respectively. SPOT-Vegetation (SPOT-Veg) 10-days AWSs measured many weather parameters some of maximum value composite (MVC) NDVI images which were the inputs in database of AgrometShell (S-10 product) were used in the model. S10 data module developed by FAO Environment and were derived from physical products, which were Natural Resources Service (SDRN). The module surface reflectance’s corrected for molecular and provides a toolbox for agro-meteorological crop aerosol scattering, water vapor, ozone and other gas monitoring and forecasting (Mukhala & Hoefsloot absorption (Holben 1986). A total of 140 composited

Tarım Bilimleri Dergisi – Journal of Agricultural Sciences 22 (2016) 555-565 557 Primary Production Estimation of Çankırı Province’s Rangelands Using Light Use Efficiency (LUE) Model..., Ünal & Bayramin

SPOT 10-day MVC NDVI images, hereafter called revised for potential evapotranspiration (PET) and NDVI, covering a 10 year (2000-2009) time period actual evapotranspiration (AET) calculations. For was obtained from ARTEMIS Project of Food and the PET calculations, the FAO Penman-Monteith Agricultural Organization (FAO) of United Nations. (Allen et al 1998) method was used, while AET The NDVI images coincided with growing season was calculated by FAO AgrometShell water balance dekads. model. 10-day average PAR, FPAR and stress factors Vector dataset (shapefile polygon) of rangeland rasterized as grid format with the same spatial size of 2 borders produced earlier throughout the National NDVI data (1 km ) were used in the model, while LUE Rangeland Project (Mermer et al 2012) was used as factor was used as constant. Graphical representation a mask file to exclude the areas other than rangelands of the LUE model is shown in Figure 3. for which the GPP was calculated. Reference data were obtained from the field surveys in both July of 2008 and 2009. Stratified random sampling method was applied to determine field visit locations. 1/25000 scaled topographic map grids were used as sampling frame laid over the rangeland polygons as upper layer in a geographic information software (GIS). Automatically generated random points intersecting both rangeland polygons and map grids were selected as sampling points. The number of sampling points was determined by the approach that each map grid had at least two survey locations representing rangelands in the grid (Anonymous 2012). A total Figure 3- Graphical representation of LUE model of 41 points were identified (Figure 2) for field Şekil 3- LUE Modelinin grafiksel gösterimi measurements which included registering botanical composition of 1 m2 quadrats and cutting the live vegetation in each quadrat. Clipped vegetation was 2.3.1. Photosynthetic active radiation (PAR) dried at sun for 7-10 days and then weighed as a dry PAR is a part of total incoming solar radiation in -2 mass (g m ). the visible spectrum (400-700 nm), which shows 2.3. Model application distinct temporal and spatial patterns due to varying atmospheric conditions (Uzun & Demir 2012). Primary production considers biomass accumulation Only the PAR of total solar radiation can be used by in vegetation as the results of succession stages green vegetation to produce organic matter through during which sun light energy is intercepted photosynthesis. The common and simple method to (Monteith 1972). Primary production is deduced as calculate the PAR is to proportionate solar radiation the product of radiation energy (PAR), fraction of by total radiation received at the surface. The PAR PAR by the plant (FPAR), conversion efficiency of was calculated from the solar radiation (MJ m-2) absorbed radiation into biomass (LUE factor) and which was measured at meteorological station. The environmental factor (stress factor). Therefore, the solar radiation incident on canopies tends to contain primary production is assumed to be proportional to a relatively constant fraction of PAR varying these variables (Equation 1). from 45% to 50% depending on location and sky Methodology used similar to the one that Seaquist conditions (Le Roux et al 1997). On average, 48% et al (2003) and Brogaard et al (2004) employed was of PAR ratio was used in calculations.

558 Tarım Bilimleri Dergisi – Journal of Agricultural Sciences 22 (2016) 555-565 Uydu Verisi ve AgrometShell Modülü ile Işık Kullanım Etkinliği (LUE) Modeli Kullanarak Çankırı İli Meralarının..., Ünal & Bayramin

2.3.2. Fractioned photosynthetic active radiation for short grass species ranged between 0.42 and (FPAR) 0.62. Therefore a higher LUE value of 0.62 for early FPAR is the fraction of the absorbed PAR by the stage corresponding the period of March 20-May 10 plant canopy and can generally be calculated by was used, while the lower LUE value of 0.42 was either physical models (Los et al 2005) or empirical used for the remaining stages of May 10-August methods (Huete et al 2002) taking into account 10. These values (Table 1) were used in the model, of spectral vegetation indexes. The most known because there were not any measured LUE values index is NDVI which relates to the FPAR given in for our study area where the botanical compositions Equation 2 (Goetz et al 1999). consisted of mostly short grass steppe.

FPAR = a ± b× NDVI (2) Table 1- Measured mean LUE values (Sims & Singh Where; a and b, correlation coefficients. For 1978) the FPAR calculation, empirical method was used Çizelge 1- Ölçülmüş ortalama LUE değerleri in this study as used by Seaquist et al (2003) and Species Location Mean LUE Brogaard et al (2004). The only difference was Desert New Mexico 0.10 the use of additional coefficients. Background and Desert New Mexico 0.07 dead materials play a substantial role on the FPAR, Short Texas 0.51 especially during senescence period of the plant. Short Texas 0.62 The FPAR absorbed by dead material reaches by Short Colorado 0.51 20% (Le Roux et al 1997) and thus, the FPAR value Short Colorado 0.42 decreases gradually after development stage. We Herbaceous Washington 0.06 therefore added the coefficient of 0.80 to the FPAR Herbaceous Washington 0.04 equation to compensate dead material effects for the Mixed South Dakota 0.84 period of senescence which corresponds to between Mixed South Dakota 0.79 st th July 1 and August 10 (Equation 3). During the Mixed North Dakota 1.74 th initial and development stage (March 20 -July Mixed North Dakota 2.00 st 31 ), no additional coefficient was added to FPAR Mixed Kansas 1.02 equation (Equation 4). Mixed Kansas 0.93 FPAR = (1.67x(NDVI) - 0.07) x 0.80 (3) Tall grass Oklahoma 0.88 Tall grass Oklahoma 1.15 FPAR = 1.67x(NDVI) - 0.07 (4)

2.3.3. LUE factor (ɛp) 2.3.4. Stress factor (ɛ) LUE factor, regarded as empirical constant, The stress component of the model is presented represents the actual efficiency of a absorbed by lack of adequate soil moisture which plays an important role in biological activities of the plant. A radiation energy used by plants to produce biomass. 2.3.4. Stress factor (ɛ) Seasonal changes in LUE factor are closely related stress factor was used in the LUE models as a part of a scalar environment factor (ε) in which drought, to soil water content and phenological stage of the The stress component of the model is presented by lack of adequate soil moisture which plays an important temperature, pollution, nutrient deficiency, illness plant (Prince 1991). In the case that available water role in biological activities of the plant. A stress factor was used in the LUE models as a part of a scalar and other elements were covered (Prince 1991). The in the soil is adequate, the value of LUE factor in environment factor (ε) in which drought, temperature, pollution, nutrient deficiency, illness and other early stage is higher than the one in development elementsstress factor were was covered calculated (Prince by 1991). the ratio The ofstress actual factor was calculated by the ratio of actual transpiration stage of the plant (Le Roux et al 1997). The available (AT)transpiration to potential (AT) transpiration to potential (PT) transpiration (Equation 5).(PT) water is assumed to be adequate in the early stages (Equation 5). of rangeland plants in the study area. According to AT (5) (5) Sims & Singh (1978), mean LUE values measured PT

Where; epsilon, ɛ, stress factor in the model; AT (mm), term emphasizes evaporated water during Tarım Bilimleri Dergisi – Journal of Agricultural Sciencesrespiration 22process (2016) of 555-565 plant, and PT, transpiration559 (mm) when there is no soil water deficit. The AT was calculated by multiplying actual evapotranspiration (AET) with fraction of vegetation cover (FVC) (Equation 6).

AT AET FVC (6)

Where; AET (actual evapotranspiration, mm), total water loss from both soil and plant during respiration. It was calculated based on the water balance model that FAO AgrometShell module performs (Mukhala & Hoefsloot 2004). The FVC was calculated from NDVI data (Equation 7).

2 NDVI - NDVI FVC s (7) NDVI v - NDVIs

Where; FVC, values range between 0 and 1 corresponding non vegetation and full vegetation cover, respectively; NDVIs, pixel value of NDVI corresponding non vegetation; NDVIv, full vegetation cover of the NDVI and NDVI, cell-specific NDVI value in the vegetation map.

The PT is the rate of transpiration that occurs in a large area completely covered with vegetation with access to unlimited water supply and calculated by multiplying the FVC with the potential evapotranspiration (PET) and crop coefficient (Kc) (Equation 8).

PT Kc PET FVC (8)

Kc incorporates crop characteristics and effects of evaporation from the soil. A value of 0.85 was used for the development stage of vegetation, during which the Kc corresponds to amount of ground cover and plant development (Wight & Hanks 1981). For the calculation of PET, FAO’s Penman-Monteith method was used, which considers many parameters related to the evapotranspiration process such as solar radiation, air temperature, vapor pressure deficit and wind speed (Allen et al 1998). All these parameters were composited as 10-day average to coincidence with NDVI data.

3. Results and Discussion

3.1. Monthly and annual production

Monthly and annual primary production calculations with LUE model were presented in Table 2. The rangelands were estimated to produce 17877 tons of 10 year mean annual production in approximately 301939 hectares of range area.

6 2.3.4. Stress factor (ɛ)

2.3.4The stress. Stress component factor (ɛ) of the model is presented by lack of adequate soil moisture which plays an important role in biological activities of the plant. A stress factor was used in the LUE models as a part of a scalar Theenvironment stress component factor (ε) of inthe which model drought, is presented temperature, by lack of pollution,adequate soilnutrient moisture deficiency, which plays illness an importantand other roleelements in biological were covered activities (Prince of the 1991). plant. The A stress factor was usedcalculated in the by LUE the modelsratio of asactual a part transpiration of a scalar environment(AT)Primary to Production potential factor Estimation transpiration (ε) inof Çankırı which (PT) Province’s drought, (Equation Rangelands temperature, 5). Using Light pollution, Use Efficiency nutrient (LUE) Model...,deficiency, Ünal & illnessBayramin and other 2.3.4elements. Stress were factor covered (ɛ) (Prince 1991). The stress factor was calculated by the ratio of actual transpiration (AT) toAT potential transpiration (PT) (Equation 5). Where; epsilon, ɛ, stress factor in the model; AT Kc incorporates crop characteristics and effects (5) The(mm), stressPT term component emphasizes of the evaporated model is presentedwater during by lack ofof adequateevaporation soil frommoisture the whichsoil. A plays value an of important 0.85 was AT rolerespiration in biological process activities of plant, of the and plant. PT, A stresspotential factor used was forused the in development the LUE models stage as of a vegetation, part of a scalar during (5) environmenttranspirationWhere;PT epsilon, factor(mm) (ε) ɛ,when stress in whichthere factor drought,is inno the soil model;temperature, water AT which (mm), pollution, the term Kc nutrient emphasizescorresponds deficiency, evaporatedto amount illness of water ground and during other cover respiration process of plant, and PT, transpiration (mm) when there is no soil water deficit. The AT was elementsdeficit. were The covered AT was (Prince calculated 1991). The stressby multiplying factor andwas plant calculated development by the ratio(Wight of actual& Hanks transpiration 1981). For calculated by multiplying actual evapotranspiration (AET) with fraction of vegetation cover (FVC) (AT)actualWhere; to evapotranspirationpotential epsilon, transpiration ɛ, stress (AET) factor (PT) with (Equation in thefraction model; 5). of AT the (mm), calculation term emphasizes of PET, evaporated FAO’s Penman-Monteith water during respiration(Equation 6). process of plant, and PT, transpiration (mm) when there is no soil water deficit. The AT was vegetation cover (FVC) (Equation 6). method was used, which considers many parameters calculatedAT by multiplying actual evapotranspiration (AET) with fraction of vegetation cover (FVC) (EquationAT AET 6). FVC (6) related to the evapotranspiration process such (5)(6) as PT solar radiation, air temperature, vapor pressure Where; AET (actual evapotranspiration, mm), Where; AET (actual evapotranspiration, mm), totaldeficit water and loss wind from speed both (Allen soil etand al 1998).plant duringAll these totalATWhere; waterAET epsilon,lossFVC from ɛ, stress both soil factor and in plantthe model; during AT (mm), term emphasizes evaporated water during (6) respiration. It was calculated based on the water balanceparameters model that were FAO composited AgrometShell as 10-daymodule averageperforms to respirationrespiration. process It was ofcalculated plant, and based PT, transpirationon the water (mm) when there is no soil water deficit. The AT was (MukhalaWhere; & AET Hoefsloot (actual 2004). evapotranspiration, The FVC was calculated mm), total coincidencefrom water NDVI loss datawith from (EquationNDVI both data. 7).soil and plant during calculatedbalance model by multiplying that FAO actual AgrometShell evapotranspiration module (AET) with fraction of vegetation cover (FVC) respiration.(Equation 6). It was calculated based on the water balance model that FAO AgrometShell module performs (Mukhalaperforms &(Mukhala Hoefsloot & 2004).Hoefsloot2 The FVC2004). was The calculated FVC from NDVI data (Equation 7). NDVI - NDVI s 3. Results and Discussion wasFVC calculated from NDVI data (Equation 7). (7) AT AETNDVIFVCv - NDVI s (6) 2 3.1. Monthly and annual production NDVI - NDVI s (7) (7) FVCWhere; AET (actual evapotranspiration, mm), total water loss from both soil and plant during Where;NDVI FVC,v - NDVIvaluess range between 0 and 1 correspondingMonthly nonand annualvegetation primary and productionfull vegetation calculations cover, respiration. It was calculated based on the water balance model that FAO AgrometShell module performs respectively; NDVIs, pixel value of NDVI correspondingwith non LUE vegetation; model NDVI were presentedv, full vegetation in Table cover 2. Theof Where; FVC, values range between 0 and 1 (Mukhalathe Where;NDVI &and FVC, Hoefsloot NDVI, values cell 2004). range-specific The between FVCNDVI was0 valueand calculated 1 in corresponding the vegetation fromrangelands NDVI non map. were datavegetation estimated(Equation and to7). full produce vegetation 17877 cover, tons of corresponding non vegetation and full vegetation respectively; NDVIs, pixel value of NDVI corresponding non vegetation; NDVIv, full vegetation cover of 2 10 year mean annual production in approximately cover, respectively; NDVIs, pixel value of NDVI the TheNDVI PT NDVIand is theNDVI, - NDVrateI cellsof transpiration-specific NDVI that value occurs in inthe a vegetationlarge301939 area hectares completelymap. of range covered area. with vegetation with correspondingFVC non vegetation; NDVI , full (7) access toNDVI unlimited - NDVI water supply and vcalculated by multiplying the FVC with the potential vegetation coverv of thes NDVI and NDVI, cell- The 10 year figures illustrated the least production evapotranspiration The PT is the rate (PET) of transpiration and crop coefficient that occurs (Kc) in (Equation a large area 8). completely covered with vegetation with specific NDVI value in the vegetation map. was observed in both central and southern region of accessWhere; to FVC,unlimited values water range supplybetween and 0 and calc 1 ulatedcorresponding by multiplying non vegetation the FVCand fullwith vegetation the potential cover, evapotranspirationPT Kc PET (PET) FVC and crop coefficient (Kc) (Equation the province 8). (Figure 4a). Field surveys showed (8)that respectively;The PT isNDVI the rates, pixel of transpirationvalue of NDVI that corresponding occurs non vegetation; NDVIv, full vegetation cover of the rangelands were mostly covered by stones and the thein aNDVI large andarea NDVI, completely cell-specific covered NDVI with vegetationvalue in the vegetation map. withPT Kc access Kcincorporates toPET unlimited cropFVC watercharacteristics supply and and calculated effects of evaporation plants were from of a steppe the soil. character A value resulting of 0.85 in was very used (8) low forby Thethemultiplying development PT is the ratethe stage ofFVC transpiration of vegetation,with thethat during occurspotential which in a largethecanopy Kc area corresponds cover. completely On the to coveredother amount hand, withof groundin vegetationthe northern cover with and part plant development (Wight & Hanks 1981). For the calculation of PET, FAO’s Penman-Monteith method accessevapotranspirationKc incorporatesto unlimited (PET)crop water characteristics and supplycrop coefficient and and effectscalc (Kc)ulated of evaporation andby northwestmultiplying from regions the the soil. ofFVC AÇerkeş value with andof 0.85 theIlgaz waspotential counties, used forwas the used, development which considers stage of vegetation,many parameters during whichrelated the to Kcthe corresponds evapotranspiration to amount process of ground such cover as solar and evapotranspiration(Equation 8). (PET) and crop coefficient (Kc) (Equationthe rangelands 8). produced greater biomasses than the plantradiation, development air temperature, (Wight vapor& Hanks pressure 1981). deficit For the and calculation windrest speedof the of region (AllenPET, FAO’s(Figureet al 1998). Penman4a). Ranges All- Monteiththese in theseparameters method regions waswerePT used,compositedKc whichPET asconsiders 10FVC-day average many parameters to coincidence related (8) with were toNDVI the calculated data.evapotranspiration to have higher process canopy such covers as solar (8)and radiation, air temperature, vapor pressure deficit and wind speed (Allen et al 1998). All these parameters were3. Results composited and asDiscussion 10-day average to coincidence with NDVI data. TableKc 2- incorporates Modelled monthly crop characteristics and annual summed and effects GPP of evaporation from the soil. A value of 0.85 was used for the development stage of vegetation, during which the Kc corresponds to amount of ground cover and plant3.3.1.Çizelge Results Monthly development 2- Aylık and and ve annualDiscussion yıllık(Wight toplam production & Hanks GPP 1981). For the calculation of PET, FAO’s Penman-Monteith method was used, which considers many parameters relatedGPP to (ton)the evapotranspiration process such as solar 3.1.Monthly Monthly and andannual annual primary production production calculations with LUE model were presented in Table 2. The radiation,Months air temperature,2000 2001 vapor pressure2002 deficit2003 and 2004wind speed2005 (Allen 2006et al 1998).2007 All these2008 parameters2009 rangelands were estimated to produce 17877 tons of 10 year mean annual production in approximately wereMarch composited156 as 10-day223 average to96 coincidence228 with261 NDVI data.300 303 54 113 142 Monthly301939 hectares and annual of range primary area. production calculations with LUE model were presented in Table 2. The April 2484 2347 2044 2090 2038 1840 3342 2992 1839 1525 3.rangelands Results were and estimatedDiscussion to produce 17877 tons of 10 year mean annual production in approximately 301939 May hectares15439 of range4204 area. 7765 7478 9873 7528 7038 5720 5782 6681 June 17688 4076 6492 6987 7322 8027 5871 3912 6227 3659 3.1. Monthly and annual production July 1665 249 572 502 655 789 92 146 375 529 August 269 62 121 95 151 137 3 22 52 100 Monthly and annual primary production calculations with LUE model were presented in Table 2. The TOTAL 37701 111618 170906 173801 203004 186214 166490 128463 14388 12636 rangelands were estimated to produce 17877 tons of 10 year mean annual production in approximately 301939 hectares of range area.

560 Tarım Bilimleri Dergisi – Journal of Agricultural Sciences 22 (2016) 555-565

6

6

6 Uydu Verisi ve AgrometShell Modülü ile Işık Kullanım Etkinliği (LUE) Modeli Kullanarak Çankırı İli Meralarının..., Ünal & Bayramin dry yields which was confirmed with the field survey results. The differences in 10 year average production may be resulted from the changes in the amount of rainfall and the frequency of animal grazing. The average rainfall during growing season (throughout 2000-2009) was 156 mm, which explains the low yield in the southern and eastern regions. However in the northern parts of the province, where the yield was relatively higher, average rainfall was 200 mm. This situation supported the general conclusion that the higher rainfall is, the more biomass the plant produce (Figure 4a and 4b). The number of grazing animals can be another factor effecting variation in biomass production. There was a gradual decrease in the animal number for 9 year-period of 2000-2008 (Figure 5). Unfortunately, there weren’t any statistical records of number of grazing animals for the 2009. In the 9 years, the total number of ruminants has been decreased by 21.347% totaling to 90539 in 2008 (Table 3). Figure 4c shows percentage change of the animal number with respect to the province’s counties. Positive variations in the period explained the increases in animal number, while negatives indicated the decreases. The animal numbers decreased in 9 out of 12 counties, but increased in 3 counties (Figure 4c). It was seen that average GPP was higher in the counties which had negative variations in the number of animal than those which had positive variations. Besides, the northern counties (Çerkes, Bayramören, Ilgaz, Kurşunlu and Atkaracalar) had higher GPP values as a result of higher amount of precipitation (Figure 4b) compared to the southern counties and lower number of grazing animals resulting in lower grazing pressures on the live vegetation. The results Figure 4- a, average GPP; b, average rainfall; c, showed both rainfall and animal number exhibited animal number variation for 10 years (2000-2009) an interactive role in rangelands’ production. by counties Şekil 4- a, 10 yıllık ortalama GPP üretimi; b, ortalama 3.2. Evaluation of model performance yağış; c, ilçelere göre hayvan sayısındaki 10 yıllık Model performance was tested to determine varyasyon how well the LUE model estimated the GPP normalized difference vegetation index (INDVI) of rangelands. Two approaches were used for performance evaluation; a) comparison of field and LUE model’s GPPs. biomass measurements with the LUE model’s Relationships between field measurements GPPs and b) regression analysis between integrated of biomass and LUE model GPPs were tested by

Tarım Bilimleri Dergisi – Journal of Agricultural Sciences 22 (2016) 555-565 561 Primary Production Estimation of Çankırı Province’s Rangelands Using Light Use Efficiency (LUE) Model..., Ünal & Bayramin

120000 and pixel size of satellite image, and measurement

115000 errors. The sampled plant type was the most important

110000 factor making the biomass measurements inconsistent.

105000 For instance, herbaceous plants (Festuca sp., Poa sp., Bromus sp.) had high ground coverage due to their 100000 extensive spread habit causing high NDVI values 95000 which drove the model to calculate higher GPPs than 90000

The Number of Animal the that of woody plants such as thyme (Thymus sp.), 85000 The Number of Animal and astragalus (Astragalus sp.). Scale differences 80000 caused location errors between ground observations 2000 2001 2002 2003 2004 2005 2006 2007 2008 and satellite data (NDVI) and thus low correlations, Years because biomass measurements were taken from 2 FigureFigure 5- The number5- The of animalnumber between of 2000 animal-2008 in betweenÇankırı province 2000-2008 (TUIK 2009) very small site (1 m ) which was approximately Şekil 5- 2000-2008 yılları arası Çankırı ilindeki hayvan sayısı (TUIK 2009) 1/1.000.000 of one pixel size of NDVI data. in Çankırı province (TUIK 2009) Table 3- The number of animals based on counties in 2000 and 2008 (TUIK 2009) ÇizelgeŞekil 3- 2000 5- ve 2008 2000-2008 yıllarındaki ilçelere yılları göre hayvan arası sayısı Çankırı (TUIK 2009) ilindeki The second method for model evaluation was the use of INDVI accounting for the summed NDVI hayvan sayısı The number (TUIK of animal 2009) Difference Counties 2000 2008 (%)* of during growing stage. The INDVI was regressed Atkaracalar 4438 3011 -32.15 with the GPPs calculated from the LUE model BayramörenTable 3- The 3609 number 2092 of animals -42.03 by counties in 2000 Çerkeş 24808 17463 -29.60 (LUE-GPPs). The correlation was significantly high Eldivanand 2008 3218(TUIK 2009) 3224 0.18 Ilgaz 11572 7095 -38.68 (r= 0.83) by giving the regression coefficient of Kızılırmak 9759 7350 -24.68 KorgunÇizelge 3- 5265 2000 ve 2008 4335 yıllarında -17.66 ilçelere göre hayvan 0.69 which means that approximately 69% of total Kurşunlusayısı (TUIK 9425 2009) 7986 -15.26 Merkez 14337 16360 14.11 variation in the LUE-GPPs can be explained by the Orta 9439 9766 3.46 linear relationship between these variables (Figure Şabanözü 9900 The 6324 number -of36.12 animal Difference YapraklıCounties 9340 5533 -40.76 6). Significance test of regression explained that Total 115110 200090539 2008-21.34 (%)* *, positiveAtkaracalar variations in the period explained4438 the increases3011 in animal number, while-32.15 negatives indicated the decreases.there was a relationship between the variables in the linear regression model (t= 7.4, P<0.05). According 3.2. BayramörenEvaluation of model performance3609 2092 -42.03 to ANOVA test, the INDVI and the LUE-GPP were Çerkeş 24808 17463 -29.60 Model performance was tested to determine how well the LUE model estimated the GPP statisticallyof rangelands. significant (Table 4). TwoEldivan approaches were used for3218 performance evaluation;3224 a) comparison0.18 of field biomass measurements with the LUE model’s GPPs and b) regression analysis between integrated normalized difference vegetation indexIlgaz (INDVI) and LUE model’s11572 GPPs. 7095 -38.68

RelationshipsKızılırmak between field9759 measurements7350 of biomass and LUE-24.68 model GPPs were tested by correlation analysis for both years. There were moderate correlations between variables (r= 0.60 and r= 0.41 for 2008 and Korgun2009 respectively), the5265 relationships were4335 insignificant for-17.66 both years (P>0.05). The correlations weren’tKurşunlu very high because of9425 the various factors7986 such as sampled-15.26 plant types, scale differences between )

sampling area and pixel size of satellite image, and measurement errors. The sampled plant -2 type was the mostMerkez important factor making14337 the biomass measurements16360 inconsistent.14.11 For instance, herbaceous plants (Festuca sp., Poa sp., Bromus sp.) had high ground coverage due to their extensive spread habit causing highOrta NDVI values which drove9439 the model to calculate9766 higher GPPs3.46 than the that of woody plants such as thymeŞabanözü (Thymus sp.), and astragalus9900 (Astragalus6324 sp.). Scale differences-36.12 caused location errors between Yapraklı 9340 5533 -40.76 Total 115110 90539 -21.34 *, positive variations in the period explained the increases in (g m model GPPs LUE 9 animal number, while negatives indicated the decreases.

correlation analysis for both years. There were moderate correlations between variables (r= 0.60 and r= 0.41 for 2008 and 2009, respectively). The INDVI relationships were insignificant for both years Figure 6- Correlation between INDVI and LUE (P>0.05). Unfortunately the correlations were not very model GPPs high because of the various factors such as sampled Şekil 6- INDVI ve LUE modeli GPP arasındaki plant types, scale differences between sampling area korelasyon

562 Tarım Bilimleri Dergisi – Journal of Agricultural Sciences 22 (2016) 555-565 ground observations and satellite data (NDVI) and thus low correlations, because biomass measurements were taken from very small site (1 m2) which was approximately 1/1.000.000 of one pixel size of NDVI data.

The second method for model evaluation was the use of INDVI accounting for the summed NDVI of during growing stage. The INDVI was regressed with the GPPs calculated from the LUE model (LUE- GPPs). The correlation was significantly high (r= 0.83) by giving the regression coefficient of 0.69 which means that approximately 69% of total variation in the LUE-GPPs can be explained by the linear relationship between these variables (Figure 6). Significance test of regression explained that there was a relationship between the variables in the linear regression model (t= 7.4, P<0.05). According to ANOVA test, the INDVI and the LUE-GPP were statistically significant (Table 4).

Uydu Verisi ve AgrometShell Modülü ile Işık Kullanım Etkinliği (LUE) Modeli Kullanarak Çankırı İli Meralarının..., Ünal & Bayramin

Table 4- Significance test statistics for linear regression of INDVI and model GPP Çizelge 4- INDVI ve modelin ürettiği GPP arasındaki doğrusal regresyonun anlamlılık testi df SS MS F Significance F Regression 1 371.75 371.75 55.40 0.00 FigureResidual 6- Correlation25 between INDVI167.73 and LUE model 6.70 GPPs ŞekilTotal 6- INDVI ve LUE26 modeli GPP539.49 arasındaki korelasyon Coefficients Standard error t Stat P-value TableIntercept 4- Significance-2.39 test statistics 1.57for linear regression-1.52 of INDVI0.14 and model GPP ÇizelgeINDVI 4- INDVI ve 3.74 model GPP arasındaki0.50 doğrusal 7.44 regresyonun0.00 anlamlılık testi

df SS MS F Significance F 3.3.Regression Sensitivity analysis 1 of the model371.75 371.75 minimum55.40 and maximum0.00 values for the “worst” and SensitivityResidual analysis25 was performed167.73 to determine 6.70 “best” case scenarios, respectively; y1, represents Total 26 539.49 to what extent the choice of variables affected the the average GPP calculated from variables’ means; Coefficients Standard error t Stat P-value model output. There is a large body of scientific y2, represents the GPP calculated from the variables Intercept -2.39 1.57 -1.52 one of0.14 whose value is changed in accordance with literatureINDVI on various 3.74 methods of sensitivity0.50 analysis 7.44 0.00 scenarios. The minimum and maximum values of the (Morgan & Henrion 1990; Shevenell & Hoffman stated variables were used to build “Best Case” and 3.3.1993). Sensitivity Of the analysisapproaches of the for model sensitivity analysis, no single method serves as the best analysis for all “Worst Case” scenarios for calculation of sensitivity Sensitivitymodelling efforts.analysis The was sensitivity performed ratio, to determine also known to what ratios extent (Table the choice 5). Bestof variables and worst affected case the scenarios model output.as elasticity There equation, is a large is body used offor scientific the analysis literature in this on variousrespectively methods denote of sensitivity minimum analysisand maximum (Morgan GPP & Henrionstudy (Anonymous 1990; Shevenell 2001). & The Hoffman sensitivity 1993). ratio O (SR)f the approachesto be produced, for sensitivity which analysis,is directly no proportionalsingle method to servesshown as in the Equation best analysis 9 is thefor allpercentage modelling change efforts. in The degreesensitivity of variables ratio, also in knownGPP equation. as elasticity The equation,minimum isoutput used dividedfor the analysisby the percentage in this study change (Anonymous in input for 2001). and The maximum sensitivity values ratio of(SR) FPAR shown, PAR in and Equation ɛ variables 9 is thea specific percentage input change variable. in outputWhen thedivided sensitivity by the ratiopercentage were changederived in from input 10 foryear a gridspecific data input used, variable.while the Whenis higher, the thesensitivity more sensitive ratio is higher,the model the outputmore sensitive is. theminimum model outputand maximum is. LUE values were directly taken from empirical measurements given in the y2 y1 x2 x1 SR (9) Table 1 (Sims & Singh 1978). (9) y1 x1 For the worst case scenario, all variables have the Where; SR, sensitivity ratio; x1, average value SR value of around 1.00, which means that biomass calculatedWhere; fromSR, sensitivitythe 10 year ratio; data ofx1 model, average variables value calculatedproduced from is thedominated 10 year bydata all of variablesmodel variables almost (FPAR, PAR and ɛ); x1 value for the LUE variable was taken as 0.500 approximating the average of highest (FPAR, PAR and ɛ); x1 value for the LUE variable equally (Table 5). As for the best case scenario, and lowest values for the short grass steppe (Table 1); x2, can take values of x1 variable’s minimum and was taken as 0.500 approximating the average of similar results were obtained that SR values are maximum values for the “worst” and “best” case scenarios, respectively; y1, represents the average GPP highest and lowest values for the short grass steppe smaller than 1.00 but around it. The most sensitive (Table 1); x2, can take values of x1 variable’s variable is FPAR (SR= 0.99). It is because the FPAR

Table 5- Sensitivity ratios (SR) for LUE model variables for the best case and the worst case scenarios 10 Çizelge 5- En iyi ve en kötü senaryolara göre LUE modeli değişkenlerinin hassasiyet oranı (SR) Worst case scenario Best case scenario Variables X1 X2 Y1 Y2 SR X1 X2 Y1 Y2 SR FPAR 0.261 0.003 0.592 0.0043 1.00 0.261 0.588 0.592 1.330 0.99 PAR 8.660 6.050 0.592 0.4101 1.02 8.660 10.915 0.592 0.741 0.96 ɛ 0.529 0.061 0.592 0.0666 1.00 0.529 0.979 0.592 1.082 0.97 LUE 0.500 0.420 0.592 0.4883 1.01 0.505 0.620 0.592 0.720 0.94

Tarım Bilimleri Dergisi – Journal of Agricultural Sciences 22 (2016) 555-565 563 Primary Production Estimation of Çankırı Province’s Rangelands Using Light Use Efficiency (LUE) Model..., Ünal & Bayramin is positively related to the vegetation index which is The results are encouraging that the LUE model an estimate of how much photosynthetically active could be applied in any rangeland region where vegetation is present. As a general conclusion for conventional weather and satellite NDVI data exist. sensitivity analysis, the variables are each important On the other hand, accurate estimation of GPP over factors affecting biomass production and this large areas as rangelands, pastures, and forest does conclusion is supported by high SR values. still have some drawbacks such as the effects of varying environmental conditions on vegetation 4. Conclusions canopy reflectance issues and the estimation of empirical LUE factor and other stress factors. It is This study combines the outputs of AgrometShell apparent that GPP modelling studies will be active module and satellite NDVI data using the LUE research area in future to overcome those handicaps. model approach to estimate annual GPP of Çankırı rangelands. The 10-year average modelled GPPs point out that annual production was approximately Acknowledgements 17880 tons which corresponds to 11.4 g m-2 of dry This research study was supported by the project (No: yield. The GPP production of the region varied 106G017) of “National Rangeland Management” seasonally and annually. The main reasons of this funds of the Scientific and Technological Research variability were rainfall and the number of animals Council of Turkey (TUBITAK). We express our grazed. gratitude to all staff for helping during the field visit and office work. Comparison of model results of annual primary production (GPP) with ground truthing showed that the model unfortunately did not produce References significant correlations (P>0.05) due to several Allen R G, Pereira L S, Raes D & Smith M (1998). sources of uncertainty and inconsistencies such as Crop Evapotranspiration: Guidelines for Computing sampled plant types, erratic LUE values, residues’ Crop Water Requirements. United Nations Food and Agriculture Organization (FAO), Irrigation and background effect, estimation of incident PAR and Drainage Paper 56, Rome, Italy scale differences between satellite data and ground Anonymous (2001). Risk assessment guidance for sample size. On the other hand, integrated NDVI superfund (RAGS). In: Volume III-Part A. Process approach produced higher correlation (r= 0.83, for Conducting Probabilistic Risk Assessment. U.S. P<0.05) between the LUE model GPP output and Environmental Protection Agency Washington, DC, the INDV values. According to sensitivity analysis, pp. 385-387 the model variables almost equally affect the GPP Anonymous (2012). Ulusal Mera Kullanım ve Yönetim production in the worst case scenario. On the other Projesi sonuç raporu. Proje No: 106G017. TÜBİTAK hand the FPAR variable was the most significant Kamu Kurumları Araştırma ve Geliştirme Projelerini factor affecting the biomass production in the best Destekleme Programı (1007 Programı) case. Brogaard S, Runnström M & Seaquist J A (2004). Primary The main advantage of the model used in this production of inner Mongolia, China between 1982 and 1999 estimated by a satellite-driven light use study was that it simulates primary production efficiency model. Global and Planetary Change 45: by considering the water used by plants (actual 313-332 transpiration), which was difficult parameter Goetz S J, Prince S D, Goward S N, Thawley M M & to measure needing extensive ground based Small J (1999). Satellite remote sensing of primary measurement equipment. However, AgrometShell production: an improved production efficiency module used here easily calculated this parameter modelling approach. Ecological Modelling 122: 239- through the weather data. 255

564 Tarım Bilimleri Dergisi – Journal of Agricultural Sciences 22 (2016) 555-565 Uydu Verisi ve AgrometShell Modülü ile Işık Kullanım Etkinliği (LUE) Modeli Kullanarak Çankırı İli Meralarının..., Ünal & Bayramin

Ha W, Gowda P, Oommen T, Marek T, Porter D & Howell Monteith J L (1972). Solar radiation and productivity in T (2011). Spatial interpolation of daily reference tropical ecosystems. Journal of Applied Ecology 9: evapotranspiration in the Texas High Plains. In: World 747-766 Environmental and Water Resources Congress, May Morgan G M & Henrion M (1990). Uncertainty: A Guide 22-26, Palm Springs, CA, pp. 2796-2804 to Dealing with Uncertainty in Quantitative Risk and Hilker T, Nicholas C C, Wulder M A, Black T A & Guy Policy Analysis. Cambridge University Press, NY R D (2008). The use of remote sensing in light use Mukhala E & Hoefsloot P (2004). AgrometShell efficiency based models of gross primary production: Manual. Agrometeorology Group, Environment and A review of current status and future requirements. Natural Resources Service, Food and Agricultural Science of the Total Environment 404: 411-423 Organization Rome, Italy Holben B N (1986). Characteristics of maximum-value Prince S D (1991). Satellite remote sensing of primary composite images from temporal AVHRR data. production: comparison of results for Sahelian International Journal of Remote Sensing 7: 1417- grasslands 1981-1988. International Journal of 1434 Remote Sensing 12: 1301-1311 Huete A, Didan K, Miura T & Rodriguez E (2002). Seaquist J W, Olsson L & Ardö J (2003). A remote sensing Overview of the radiometric and biophysical based primary production model for grassland biomes. performance of the MODIS vegetation indices. Ecological Modelling 169: 131-155 Remote Sensing of Environment 83: 195-213 Shevenell L & Hoffman F O (1993). Necessity of Ketenoğlu O, Quezel P, Akman Y & Aydoğdu M (1983). uncertainty analyses in risk assessment. Journal of New syntaxa on the gypsaceous formation in the Hazardous Materials 35: 369-385 Central Anatolia. Ecologia Mediterranea 9(3-4): 211- 221 Sims P L & Singh J S (1978). The structure and function of ten western North America grasslands. II. Intra- Kurt L, Tuğ G N & Ketenoğlu O (2006). Synoptic view seasonal dynamics in primary producer compartments. of the steppe vegetation of Central Anatolia (Turkey). Journal of Ecology 66: 251-285 Asian Journal of Plant Sciences 5(4): 733-739 Tucker C J, Vanpraet C, Boerwinkel E & Gaston A Le Roux H X, Gauthier A, Begue H & Sinoquet H (1997). (1983). Satellite remote sensing of total dry matter Radiation absorption and use by humid savannah grassland: assessment using remote sensing and production in the Senegalese Sahel. Remote Sensing of Environment : 461-474 modelling. Agricultural and Forest Meteorology 85: 13 117-132 TUIK (2009). Livestock statistics. Retrieved in July, 18 Los S O, North P R J, Grey W M F & Barnsley M J 2009 from http://www.tuik.gov.tr (2005). A method to convert AVHRR normalized Uzun B & Demir V (2012). Fotosentetik aktif radyasyon difference vegetation index time series to a standard (FAR) ölçümlerinde LED ve foto diyotların hassas viewing and illumination geometry. Remote Sensing tarım açısından kullanılabilirliği üzerine bir araştırma. of Environment 99(4): 400-411 Tarım Bilimleri Dergisi-Journal of Agricultural Mermer A, Ünal E, Aydoğdu M, Urla Ö, Yıldız H & Sciences 18 (3): 214-225 Torunlar H (2012). Determining rangeland areas Wight J R & Hanks R J (1981). A water-balance, climate by satellite images. TABAD-Research Journal of model for range herbage production. Journal of Range Agricultural Sciences 5: 107-110 Management 4: 307-311

Tarım Bilimleri Dergisi – Journal of Agricultural Sciences 22 (2016) 555-565 565