j ourna l homepage: ht tp : / /www.e lsev ie r .com/ locate / rama
A Modification of CIM for Prediction of Net Primary Productivity of the
Three-River Headwaters, China☆
Chong Wang, Huilong Lin ⁎, Yuting ZhaoState Key Laboratory of Grassland Agro-ecosystems, Key Laboratory of Forage and Livestock Industry Innovation of the Ministry of Agriculture and Rural Affairs, Chinese Center for StrategicResearch of Grassland Agriculture Development, College of Pastoral Agriculture Science and Technology, Lanzhou University, Lanzhou 730020, P. R. China
a b s t r a c ta r t i c l e i n f o
☆ This work was supported by the Chinese Natural(31772666) and the National Key Research and Developm⁎ Correspondence: Huilong Lin, Rm 416, Yifu Biology Bl
The Three-River Headwaters (TRH) region is covered dominantly with alpine meadow, a large part of which isconfronting severe degradation as a result of climate change and human-induced influences. The estimation ofnet primary productivity (NPP) is essential to provide support for scientific management of TRH grassland re-sources to prevent further degradation. The classification indices-basedmodel (CIM) has been applied in the es-timation of NPP and its response to global warming because of its simple structure and easily obtained indices.However, CIM is considered to estimate the potential NPP rather than the actual value. Thus, its applicationhas been restricted. In this study, the normalized difference vegetation index (NDVI) was applied to modifythe CIM. Then, CIM and modified CIM were compared with the other three models. The assessment of NPP esti-mates indicated that the modified CIM had a fair performance among the NPP models (R2 = 0.42, RMSE =178.08). All theNPP estimationmodels revealed that NPP increased from the northwest to the southeast. Accord-ing to themodified CIM, themean NPP of TRH grasslandwas 135.44 gC·m−2·yr−1 and the total NPP was 3.22 ×1013 gC·yr−1. Among the classes of the grassland of TRH in the comprehensive and sequential classification sys-tem (CSCS), the frigid perhumid rain tundra and alpinemeadow occupied most of the grassland NPP, which was3.06× 1013 gC·yr−1.With the help of theNDVI, themodifiedCIMperformed better than the CIM; however, thereis still much room for the improvement of CIM in future research.
The Three-River Headwaters (TRH) region is the source of the YangtzeRiver, Yellow River (Huang River), and Lantsang (Mekong) River. This re-gion provides 25% of thewater volume of the Yangtze River, 49% of the Yel-low River, and 15% of the Lantsang River (Mao et al., 2012). Therefore, it isconsidered “the Water Tower of China.” The alpine meadow occupies65.37% of the TRH’s total area (Liu et al., 2008). Due to global warmingand human-induced disturbance, TRH faces severe grassland degradation.Fan et al. (2010) reported that 26−46% of the grassland in TRH has beensignificantly degraded, and Liu et al. (2008) also indicated that the area ofmoderately and severely degraded grassland had reached 1.2 × 107 ha, ac-counting for 58% of the available grassland area. Furthermore, Yu et al.(2010) reported that the average aboveground biomass decreased from 3605.25 kg·ha−1 in the 1980s to 2 954.44 kg·ha−1 in 2005. Yield per unitarea declined by 30−50%, high-quality forage decreased by 20−30%, poi-sonous weeds increased by 70−80%, grassland vegetation coverage de-creased by 15−25%, and height of forage decreased by 30−50%,
nce Foundation Projectslan (2016YFC0501906).2, S Tianshui Rd, Lanzhou,
ed by Elsevier Inc. All rights res
compared with the 1950s. TRH is the most important area for livestockgrazing in Qinghai Province, especially for the Tibetan sheep and yak(Wang et al., 2009b). Since the degradation of grassland has threatenedlivestock grazing and the security of the grassland ecosystem, the Chinesegovernment took an unprecedented conservation action, the EcologicalProtection and Restoration Program (EPRP), to monitor and estimate im-pact levels in the TRH region (Wang et al., 2010). EPRP has conserved andrehabilitated the TRHecosystembyecologicalmigration, grazingbans,wet-land protection, and restoration of degraded grassland since 2005.
Net Primary Productivity (NPP) is originally defined as the amountof photosynthetically fixed carbon available to the first heterotrophiclevel in an ecosystem (Field et al., 1998). It is a crucial indicator to eval-uate the health and ecological balance of an ecosystem (Gao et al., 2009)and characterize the carbon cycle of grassland ecosystems (Yeganeh etal., 2012). As the base years of EPRP, the NPP estimation in the year of2005 and 2006 is a step toward formulating management strategy andrequires continued monitoring. Given the harsh environmental condi-tions, it is often difficult to obtain the observed NPP at TRH; therefore,NPP estimation by models is more practical. Referring to the formerstudies (Ruimy et al., 1994; Lin, 2009; Li et al., 2013; Tang et al., 2014;Sun et al., 2017), NPP estimation models can be broadly classified intothree types: statistical models (e.g., MIAMI, CIM) (Lieth, 1973; Lin,2009); light-use efficiency models (e.g., CASA, GLO-PEM) (Potter et al.,
328 C. Wang et al. / Rangeland Ecology & Management 72 (2019) 327–335
1993; Prince and Goward, 1995); and ecophysiological process-basedmodels (e.g., TEM, BIOME-BGC) (Running and Hunt, 1993; Zhuang etal., 2003). Light-use efficiency models can be driven by remote sensingdata, making them capable of predicting NPP at a large scale. Several for-mer studies have applied the Carnegie-Ames-Stanford Approach (CASA)in the spatial and temporal distribution of NPP (Piao et al., 2006; Tanget al., 2014; Zhang et al., 2014). Fan et al. (2010) applied the Global Pro-ductionEfficiencyModel (GLO-PEM) in thequantification of the contribu-tion of climate change and grazing onNPP variation. Among these studies,some crucial parameters, such as the moisture stress coefficient in CASA,weremodified due to the difficulties in data acquisition (Chen et al., 2012;Gao et al., 2013; Tang et al., 2014). The ecophysiological process-basedmodels are often driven by three types of datasets, which are site-specificdata, ecophysiological data, and other data, such as meteorological andsoil temperature data (Sun et al., 2017). NPP estimations with thesetypes of models often gain satisfactory accuracy (Zhuang et al., 2010; Yeet al., 2013). However, themodel structure is complex. In addition, multi-ple parameters are required to be calibratedwith site-specific data, whichis time-consuming and costly. The statistical models can be applied to es-timate not only the current NPP but also the NPP under future climatechange scenarios, and its variables are easy to obtain. However, the accu-racy may not be as good as the other two types, and these statisticalmodels lack interpretation on the ecophysiological process of vegetation.There is a trade-off among the data acquisition, complexity, and perfor-mances of NPP models. So far, none of the reported NPP models can beperfectly applied anywhere. Therefore, full consideration should begiven to the characteristics of a study area, the data acquisition, and thespatial and temporal scale. Considering the data scarcity at TRH, modelswith a simple structure and easily obtained indices are preferable. There-fore, among the three types of NPP models, the statistical models are stillcompetitive. To our best knowledge, the classification indices-basedmodel (CIM) is the only NPP model laying its foundation on a classifica-tion system, which is the comprehensive sequential classification system(CSCS) (Ren et al., 2008; Lin, 2009). The NPP predictions by the CIM atvarious spatial scales in former studies have shown satisfactory results(Lin et al., 2012, 2013; Lin and Zhang, 2013; Li et al., 2014b). However,CIM is only driven by meteorological data, which makes it consideredfor predicting the potential NPP rather than the actual NPP (Yang et al.,2016). Since the defect of predicting potential NPP restricts the applica-tion of CIM, we seek to find amodification index capable of predicting ac-tual NPP of grasslands. The normalized difference vegetation index(NDVI) is reported to reflect the coverage and intensity of photosynthesis.The NDVI has also been applied to estimate NPP directly or indirectly(Potter et al., 1993; Fang et al., 2003; Buono et al., 2010; An et al., 2013).Therefore, in this study NDVI was applied to modify CIM. The objectivesof this study were to 1) modify CIM to estimate the grassland NPP atTRH and 2) evaluate the estimates of NPP generated fromvariousmodels.
Materials and Methods
Study Area
The TRH region is located in the southern Qinghai Province, China,between 31°36'N−36°16'N and 89°24'E−102°15'E. It covers an areaof 363,000 km2, which is 50.3% of the total area of Qinghai Province(Fan et al., 2010). The altitude ranges from 2 800 m to 6 564 m, andthe average altitude is 4 585 m with the majority area being in therange of 4 000 m−5 800 m. TRH has a plateau continental climate.The population is 55 720, most of whom are of Tibetan origin and relyon traditional agricultural practices (Fan et al., 2010).
Data Collection
NPP ObservationsThe biomass samples were collected from the surveys conducted by
the National Animal Husbandry and Veterinary Service of the Ministry
of Agriculture of the People’s Republic of China in July and August of2005 and 2006, approximately the time of peak biomass production.The sampling sites were randomly selected considering altitude, slope,vegetation coverage, dominant vegetation species, and grasslandtypes. At each site the vegetation was harvested at the ground surfaceusing the quadrat method in a 1 m × 1 m plot (Ren, 1998). Overall, atotal of 450 samples were collected, compiled, and analyzed (Fig. 1).The biomass of collected samples varied from 10 g·m−2 to 561.33g·m−2, and the average was 136.37 g·m−2. It was converted to above-ground NPP using a conversion factor of 0.45 (Gill et al., 2002; Ji et al.,2016) according to the standard methods (Long et al., 1989; Scurlocket al., 2002; Lin et al., 2012) and then converted to the total NPP accord-ing to the root-shoot ratio (Piao et al., 2004). A total of 314 samples(70%) were randomly selected as the calibration dataset for the modifi-cation of CIM. The remaining samples constituted the validation datasetand were implemented in the assessments of modified CIM and otherNPP estimates.
Climate DataThe daily precipitation, daily maximum, and minimum temperature
in 2005 and 2006 of 72 meteorological stations in the TRH and its sur-rounding areawere obtained from the ChinaMeteorological Data SharingService System. The meteorological records were organized by monthlyaverage precipitation, total annual precipitation (TAP), monthly averagetemperature, mean annual temperature (MAT), and annual cumulativetemperature above 0°C (ACT). All the climate data were interpolated byANUSPLIN (version 4.3, Centre for Research in Engineering Science, Aus-tralian National University, Canberra) (McKenney et al., 2006). ANUSPLINwas developed on the theory of partial thin-plate smoothing splines,which allows the assistance of ancillary variables in the interpolationand improves the performance of interpolation (McKenney et al., 2006).In this study, the aspect and elevation were applied as ancillary variablesin the interpolation by ANUSPLIN.
Terrain and Remote Sensing DataThe elevation data were downloaded from the SRTM 90mDigital El-
evation Database (version 4.1) of the Consortium for Spatial Informa-tion website (CGIAR-CSI, 2006). The aspect was generated fromelevation by ArcGIS 9.3. The Moderate Resolution ImagingSpectroradiometer (MODIS) images were downloaded from the LandProcess Distributed Active Archive Center of the US Geological Survey(NASA, 2015). TheMOD09GA of MODIS product provides an estimationof daily surface reflectance information with a spatial resolution of 500m. The monthly NDVI was derived from the MOD13A3 with a spatialresolution of 1 km. The annual NDVI was calculated via the maxi-mum-value composite procedure. The MOD15A2 provides the fractionof photosynthetically active radiation (FPAR) with an 8-d interval and1-kmspatial resolution. TheMOD17 product is thefirst annually contin-uous satellite-driven dataset monitoring global vegetation productivityat 1-kmspatial resolution (Zhao et al., 2006). TheMODIS images of 2005and 2006 mentioned earlier were compiled.
Legacy DataThree base maps were compiled in this study. The grassland class
map (Fig. 2) and the county-level administrative divisions of QinghaiProvince were provided by the State Key Laboratory of GrasslandAgro-ecosystems, College of Pastoral Agriculture Science and Technol-ogy, Lanzhou University. The land use types of TRH in 2005 were pro-vided by the Data Center for Resources and Environmental Sciences,Chinese Academy of Sciences.
NPP Models
MIAMI ModelTheMIAMImodel (Lieth, 1973; Zaks et al., 2007) was one of the first
statistical models on the global scale for NPP prediction. It is usually
Figure 1. Locations of net primary productivity observations (black dots).
329C. Wang et al. / Rangeland Ecology & Management 72 (2019) 327–335
employed as the baseline for model comparison in NPP estimation(Cramer et al., 1999). Established in the 1970s, it usesmean annual tem-perature and total annual precipitation separately in two functions. Theestimated NPP is theminimum value of the two functions. NPP is calcu-lated by Equations [1−3].
NPPM ¼ min NPPt;NPPp� �
½1�
with
NPPt ¼ 3000= 1þ e1:315−0:1196t� �
½2�
NPPp ¼ 3000 1−e−0:000664p� �
½3�
where NPPM is the NPP estimated by the MIAMI model, NPPt is the NPPestimated by the MAT, and t is the MAT (°C). NPPp is the NPP estimatedby the TAP, and p is TAP (mm).
CASACASA is one of the most well-known light use efficiency models
(Potter et al., 1993; Field et al., 1995). It is driven by remote-sensing-based data and climate data (Equations [4] to [8]).
NPPC ¼ IPAR� ε ½4�
where NPPc is the NPP estimated by CASA, IPAR is the intercepted pho-tosynthetically active radiation, and ε is the light utilization efficiency.IPAR (MJ) is given by
IPAR ¼ SOL � FPAR� 0:5 ½5�
and ε (gC MJ−1 PAR) is calculated by
ε ¼ Tε1 � Tε2 �Wε � ε� ½6�
where SOL (MJ·mo−1) is the total solar radiation. FPAR is the fraction ofthe incoming photosynthetically active radiation (PAR, MJ·mo−1)intercepted by green vegetation, and the factor 0.5 accounts for thefact that approximately half of the incoming solar radiation is in thePAR waveband. FPAR can be derived from MOD15A2. Tε1 and Tε2 aretemperature stress factors, Wε is the water stress factor, and ε⁎ is themaximum possible efficiency (gC·MJ−1·PAR). Wε is initially calculatedby estimated evapotranspiration and potential evapotranspiration(Potter et al., 1993). Limited by the data acquisition in the TRH, a mod-ified calculation of Wε by the Land SurfaceWater Index (LSWI) (Xiao etal., 2004; Wang et al., 2011; Bao et al., 2016) was adopted in this study.Wε and LSWI are given by the Equations [7] and [8].
Wε ¼1þ LSWI
1þ LSWImax½7�
LSWI ¼ ρnir−ρswir
ρnir−ρswir½8�
where LSWI stands for the land surface water index and LSWImax repre-sents themaximum annual LSWI. ρnir and ρswir stand for the near-infra-red band and the shortwave infrared band, respectively, correspondingto band 2 and band 6 of MOD09GA.
MOD17A3MOD17A3 is a MODIS product, which can be applied for a regular
global estimate of gross primary productivity (GPP) and NPP of the en-tire terrestrial earth surface at 1-km spatial resolution (Running et al.,2000; Thornton et al., 2002; Heinsch et al., 2003). It has been used inthe evaluation of NPP, global change, etc. (Fensholt et al., 2006; Zomeret al., 2008; Huang et al., 2014).The GPP of MOD17A3 is calculated onthe basis of the light efficiency algorithm (Heinsch et al., 2003), andthe photosynthetically active radiation (PAR) conversion efficiency iscomputed by the BIOME-BGC model (Running and Hunt, 1993). NPP
Figure 2.Grassland types of Three-River Headwaters according to the comprehensive and sequential classification system, overlappedwith the landuse types in 2005.White areas are notcovered by grassland.
Figure 3. Distribution of net primary productivity observations and the correspondingnormalized difference vegetation index. The solid black line stands for the fitted linearregression line.
330 C. Wang et al. / Rangeland Ecology & Management 72 (2019) 327–335
is the difference of GPP, maintenance, and growth respiration. The logicto calculate NPP in MOD17A3 is shown in Equations [9−12], and thedetailed description of the algorithm to develop MOD17A3 is given byHeinsch et al. (2003).
NPP ¼ GPP–MR–GR ½9�
where MR represents the maintenance respiration of livewood and GRrepresents the growth respiration of deadwood, leaf, and fine root.
GPP ¼ ε � APAR ½10�
where APAR is the absorbed PAR and ε is the PAR conversion efficiency.
APAR ¼ IPAR � FPAR ½11�
where IPAR is the PAR incident on the vegetative surface and FPAR is thefraction of incident PAR absorbed by the surface.
ε ¼ εmaxTMINscalar � VPDscalar ½12�
whereVPD is the daylight average vapor pressure deficit and TMIN is thedaily minimum temperature at which ε = 0 (at any VPD). TMINscalar
and VPDscalar are attenuation scalars of TMIN and VPD. εmax is the max-imum of conversion efficiency.
CIMReferring to the theory of CSCS, Lin (2009) replaced the biological
temperature in the synthesized model (Zhang et al., 1996; Zhou andZhang, 1996) with ACT to calculate radiative dryness index and pro-posed the CIM. CIM adopted the same indices as the “class” of CSCS,which were ACT and the moisture index. It was validated to be reliableand accurate at regional and global scales (Lin et al., 2012; Li et al.,2014a, b). NPP is calculated by the Equations [13−15].
L Kð Þ ¼ 0:58802K3 þ 0:50698K2−0:0257081Kþ 0:0005163874 ½14�
K ¼ r0:1Σθ
½15�
where NPPL is the estimated NPP by CIM. Σθ is the annual cumulativetemperature above 0°C, and r is the mean annual precipitation (mm).K is the moisture index.
Former studies have reported the saturation of NDVI happens whenvegetation coverage is dense (Huete et al., 1997). When NDVI is satu-rated, the correlation betweenNDVI and vegetation productivity is non-linear and NDVI is not sensitive to the variation of some vegetationfeatures, such as vegetation productivity, leaf area index, and plant func-tional types (Luo et al., 2004; Jin et al., 2013; Cai et al., 2014). The satu-ration can happen when NDVI is N 0.78 (Gu et al., 2013) or the leaf areaindex is above 2mm−2 (Cao et al., 2017) or 4 mm−2 (Duchemin et al.,2006). Considering the saturation of NDVI, a piecewise function wasused in CIM modification. The interval of NDVI refers to former studiesand the correlation of NDVI and NPP observations in this study.
Figure 4. Correlation between predicted net primary productivity (NPP) by the modifiedclassification indices-based model (MCIM) and NPP observations. The solid black linestands for the fitted linear regression line. The red dashed line is the 1:1 line.
331C. Wang et al. / Rangeland Ecology & Management 72 (2019) 327–335
All estimates of NPP in this study were the average value of the an-nual NPP in 2005 and 2006. The coefficient of determination (R2) androot mean square error (RMSE) were employed to assess the NPP
Figure 5. Comparison of net primary productivity estimates byMIAMI, CIM, CASA, andMOD17Athe 1:1 lines.
where yo is the observed value and ys is the estimated value.
Results
Modification of CIM
The correlation between NDVI and NPP observations indicates thatthey are not linearly correlated when NDVI is N 0.7 (Fig. 3). Thus, weused 0.7 as the segmentation point of NDVI, and the modified CIM isexpressed as Equation [17]. NPP is underestimated when it is N 500 gCm−2 yr−1, and the parameter CIM × NDVI can explain 42% of the NPPvariation (Fig. 4).
Figure 6. Spatial distributions of grassland net primary productivity (NPP) at Three-River Headwaters estimated by MIAMI (a), classification indices-based model (CIM) (b), Carnegie-Ames-Stanford Approach (c), MOD17A3 (d), and modified CIM (e). The values are the average NPP of 2005 and 2006. White areas are not covered by grassland.
332 C. Wang et al. / Rangeland Ecology & Management 72 (2019) 327–335
Comparison of Estimates of NPP
The modified CIM performed better than CIM (Fig. 5), which indi-cated that the introduction of NDVI improved the performance of CIM.The estimated NPP of MIAMI, CIM, CASA, andMOD17A3were validatedwith the observed NPP of validation set, and the results indicated thatthe MOD17A3 had the highest performance (R2 = 0.50, RMSE =173.90). Overall, NPP estimates of five models shared a similar spatialdistribution trend, increasing from the northwest to the southeast(Fig. 6). The NPP derived fromMOD17A3 had the largest range. Regard-ing themean value, CASA estimated the lowest valuewhile the CIM andMIAMI models generated higher mean NPP than the others (Table 1).
Table 1Comparison of estimates by net primary productivity models.
According to the estimated NPP by modified CIM, the grassland NPPincreased from the northwest to the southeast and varied from 0gC·m−2·yr−1 to 465.73 gC·m−2·yr−1 with a mean NPP of 135.44gC·m−2·yr−1. 3.22 × 1013 gC·yr−1 was stored by the TRH grassland.Table 2 presented the NPP of each CSCS class of grassland at TRH. Theclass of frigid perhumid rain tundra and alpine meadow dominatedthe grassland of TRH (Fig. 6) and therefore had the vast majority of NPP.
Montane meadow steppe is short for cold temperate−subhumidmontanemeadow steppe;meadow steppe is short for cool temperate−subhumid meadow steppe; tundra, alpine meadow is short for frigid-humid tundra, alpine meadow; montane meadow is short for coldtemperate−humid montane meadow; rain tundra, alpine meadow isshort for frigid perhumid rain tundra, alpine meadow.
Discussion
Owing to the development of remote sensing techniques, a geo-graphic information system (GIS), and computer science, the accuracyof NPP estimation has been improved remarkably. However, there is atrade-off among data acquisition, complexity of model structure, andprecision of model estimation. Given the credit to the simple structureand easily obtained indices, statistical models have advantages,
Table 2Estimates of net primary productivity (NPP) by modified classification indices-based model of comprehensive and sequential classification system classes of Three-River Headwatersgrassland.
NPP model Minimum NPP (gC m−2 yr−1) Maximum NPP (gC m−2 yr−1) Mean NPP (gC m−2 yr−1) Total NPP (×1010 gC yr−1)
333C. Wang et al. / Rangeland Ecology & Management 72 (2019) 327–335
especially in data-poor regions. CIM has been compared with other sta-tisticalmodels in previous studies (Lin and Zhang, 2013), but to our bestknowledge, few comparisons have been made between CIM and lightuse efficiency models or the process-based models. In this study, weproposed a modification of CIM with NDVI, aiming to estimate the ac-tual NPP in data-poor regions such as TRH. The comparison amongNPP estimates indicated that the MOD17A3 performed best while theMIAMI performed worst. Besides, the performance of CIM was en-hanced by NDVI, even competitive with CASA andMOD17A3. However,only 42% of NPP variation can be explained by the modified CIM, whichis not satisfactory. We analyzed the correlation between the NPP obser-vations and NDVI, MAT, and CIM parameters (Table 3). NPP is signifi-cantly correlated with NDVI and TAP at the 0.01 level and significantlycorrelated with MAT at the 0.05 level, although it is not correlatedwith ACT and K. NDVI is significantly correlated with other variablesat the 0.01 level except K. More than half (58%) of NPP variation cannotbe explained by parameters ofmodified CIM, andwe think theremay betwo reasons. Firstly, even though the piecewise function has been ap-plied, the saturation of NDVI may not be sufficiently fixed; thus, CIM ×NDVI is not sensitive enough to reflect the NPP variation of high-cover-age grassland. Furthermore, NDVI saturation can cause the underesti-mation of vegetation productivity (Gu et al., 2013), which may explainthe underestimation when NPP is N 500 gCm−2 yr−1 in this study. Sec-ondly, other factors need to be explored for a better explanation of NPPvariation. Former studies reported that although RVI (the ratio betweenthe near infrared reflectance and the visible red reflectance) is not assensitive as NDVI when vegetation coverage is sparse, it is sensitivewhen vegetation coverage is dense, which is almost opposite withNDVI (Huete et al., 1997; Gu et al., 2013). Therefore, some studieshave coupled NDVI and RVI to cope with the NDVI saturation problem(Huete et al., 1997; Gu et al., 2013; Cai et al., 2014; Li et al., 2014a, b). Be-sides, the enhanced vegetation index (EVI) was developed against thesaturation of NDVI (Huete et al., 1997). Given all that, the CIM modelstill has much room for improvement. The application of other parame-ters to modify CIM requires further analysis in future research.
TAP indicates total annual precipitation; MAP, mean annual precipi-tation; ACT, annual cumulative temperature above 0°C; K, moistureindex; NDVI, normalized difference vegetation index.
Previous studies have shown the discrepancy of estimated NPPcaused by the algorithmof different NPPmodels, time spans, and spatialscales. Zhang et al. (2014) used the GLO-PEMmodel and concluded thatthe mean NPP of TRH from 2005 to 2012 was 694 kg·ha−1·yr−1
, while
Table 3Correlations between observed net primary productivity (NPP) and parameters of classi-fication indices-based model.
1 Correlation is significant at the 0.01 level (2-tailed).2 Correlation is significant at the 0.05 level (2-tailed).
Fan et al. (2010) used the samemodel and concluded that themeanNPPof TRH from 1988 to 2005 was 422.29 kg·dry matter ha−1·yr−1. Ac-cording to Wang et al. (2009a, b) the mean NPP was 160.90 gC·m−2·yr−1. Wo et al. (2014) estimated the NPP with CASA and concludedthat the total NPP of TRH in 2010 was 52.15 × 1012 gC·yr−1, comparedwith the total NPP of 30.63×1012 gC·yr−1 in this study. Besides, the av-erage NPP of CASA was 65.93 gC·m−2·yr−1, which was less than theaverage NPP reported by Wang et al. (2009a, b) and Wo et al. (2014).The differences of estimates by CASA can be explained in two ways.Firstly, the CASA applied in this study was modified because of the lim-itation of input variables, which can introduce variance to the estimatedNPP. Secondly, it was reported that the grassland of TRH recovered fol-lowing the implementation of EPRP in 2005 (Fan et al., 2010). Thus, itwas tolerable for the increase of NPP after 2005.
With the help of NDVI, the performance of CIM was greatly im-proved. In terms of spatial distribution, the estimations by all themodels used indicated that NPP at TRH increased from the northwestto southeast. The difference of minimum values estimated by differentmodels was noteworthy. The minimum values of MIAMI and CIMwere N0, while the minimum of other estimates was 0. According tothe functions generating NPP, the estimates are always N0. However,because NDVI was applied in CIM modification, the estimate can be 0where NDVI equals 0. Overall, NPP estimation by models in this studyis not satisfactory. The sampling sites are not evenly distributed due tothe limitation of the harsh natural climate condition and poor road ac-cessibility at TRH. Themodel performances can be limited by spatial res-olution of remote sensing images. Most of the remote sensing data usedin this study are developed with a 1-km spatial resolution. Therefore,when we extracted the values from them at sample sites, two differentsites may share the same value. Thus, to improve model performances,remote sensing data with finer spatial resolution should be used.
Implications
Grassland classes and their distribution patterns correspond to cer-tain climatic types in a series of mathematical forms. Thus, climate var-iables can be used to represent grassland classes and their distributionand vice versa (Ren et al., 2008). In this study, NDVI was applied inCIM modification to reflect the influences of human activities on NPP.Considering the saturation of NDVI, a piecewise function was used andthe assessment of NPP estimates showed that NDVI has remarkably im-proved the performance of CIM. It indicated that for data-scarce regionsit is possible to estimate NPP bymodels with easily obtained data. How-ever, the modified CIM cannot optimally explain the spatial variation ofNPP, which implies that other parameters and data with finer spatialresolution need to be supplemented for the further improvement ofCIM. TRH plays an irreplaceable role in water and biological resourcesin China and even in Southeast Asia. A robust estimation of NPP contrib-utes significantly to the understanding of the pattern, process, and dy-namics of grassland ecosystems of this region and provides basic datasupport for the policymaking and policy appraisal at TRH. Furthermore,the results presented in this study are not only specific to TRHbut, moreimportantly, to the given grassland classes according to the CSCS ap-proach, which can be scaled up from plots to estimate landscape-scaleeffects.
334 C. Wang et al. / Rangeland Ecology & Management 72 (2019) 327–335
Acknowledgments
The authors thank Joshua Philp for the revision of this manuscriptand Dr. Qisheng Feng for his support in data processing.
References
An, N., Price, K.P., Blair, J.M., 2013. Estimating above-ground net primary productivity ofthe tallgrass prairie ecosystem of the Central Great Plains using AVHRR NDVI. Inter-national Journal of Remote Sensing 34, 3717–3735.
Bao, G., Bao, Y., Qin, Z., Xin, X., Bao, Y., Bayarsaikan, S., Zhou, Y., Chuntai, B., 2016.Modelingnet primary productivity of terrestrial ecosystems in the semi-arid climate of theMongolian Plateau using LSWI-based CASA ecosystem model. International Journalof Applied Earth Observation and Geoinformation 46, 84–93.
Buono, G., Oesterheld, M., Nakamatsu, V., Paruelo, J.M., 2010. Spatial and temporal varia-tion of primary production of Patagonian wet meadows. Journal of Arid Environ-ments 74, 1257–1261.
Cai, D., Guan, Y., Guo, S., Zhang, C., Fraedrich, K., 2014. Mapping plant functional typesover broad mountainous regions: a hierarchical soft time-space classification appliedto the tibetan plateau. Remote Sensing 6, 3511–3532.
Cao, Z., Cheng, T., Ma, X., Tian, Y., Zhu, Y., Yao, X., Chen, Q., Liu, S., Guo, Z., Zhen, Q., Li, X.,2017. A new three-band spectral index for mitigating the saturation in the estimationof leaf area index in wheat. International Journal of Remote Sensing 38, 3865–3885.
CGIAR-CSI, 2006. SRTM 90m Digital Elevation Database v4.1. Available at:. http://srtm.csi.cgiar.org/SELECTION/inputCoord.asp, Accessed date: 13 July 2018.
Chen, Z., Shao, Q., Liu, J., Wang, J., 2012. Analysis of net primary productivity of terrestrialvegetation on the Qinghai-Tibet Plateau, based on MODIS remote sensing data. Sci-ence China Earth Sciences 55, 1306–1312.
Cramer, W., Kicklighter, D.W., Bondeau, A., Iii, B.M., Churkina, G., Nemry, B., Ruimy, A.,Schloss, A.L., Intercomparison, T. P. O. F. T. P. N. M, 1999. Comparing global modelsof terrestrial net primary productivity (NPP): overview and key results. GlobalChange Biology 5, 1–15.
Duchemin, B., Hadria, R., Erraki, S., Boulet, G., Maisongrande, P., Chehbouni, A., Escadafal,R., Ezzahar, J., Hoedjes, J.C.B., Kharrou, M.H., Khabba, S., Mougenot, B., Olioso, A.,Rodriguez, J.C., Simonneaux, V., 2006. Monitoring wheat phenology and irrigationin Central Morocco: on the use of relationships between evapotranspiration, cropscoefficients, leaf area index and remotely-sensed vegetation indices. AgricultureWater Management 79, 1–27.
Fan, J., Shao, Q., Liu, J., Wang, J., Harris, W., Chen, Z., Zhong, H., Xu, X., Liu, R., 2010. Assess-ment of effects of climate change and grazing activity on grassland yield in the ThreeRivers Headwaters Region of Qinghai-Tibet Plateau, China. Environmental Monitor-ing and Assessment 170, 571–584.
Fang, J., Piao, S., Field, C.B., Pan, Y., Guo, Q., Zhou, L., Peng, C., Tao, S., 2003. Increasing netprimary production in China from 1982 to 1999. Frontiers in Ecology and the Envi-ronment 1, 293–297.
Fensholt, R., Sandholt, I., Rasmussen, M.S., Stisen, S., Diouf, A., 2006. Evaluation of satellitebased primary production modelling in the semi-arid Sahel. Remote Sensing of Envi-ronment 105, 173–188.
Field, C.B., Behrenfeld, M.J., Randerson, J.T., Falkowski, P., 1998. Primary production of thebiosphere: integrating terrestrial and oceanic components. Science 281, 237–240.
Field, C.B., Randerson, J.T., Malmström, C.M., 1995. Global net primary production: com-bining ecology and remote sensing. Remote Sensing of Environment, Remote Sensingof Land Surface for Studies of Global Change 51, 74–88.
Gao, Q., Li, Y., Wan, Y., Qin, X., Jiangcun, W., Liu, Y., 2009. Dynamics of alpine grasslandNPP and its response to climate change in Northern Tibet. Climate Change 97,515–528.
Gao, Y., Zhou, X., Wang, Q., Wang, C., Zhan, Z., Chen, L., Yan, J., Qu, R., 2013. Vegetation netprimary productivity and its response to climate change during 2001-2008 in the Ti-betan Plateau. Science of the Total Environment 444, 356–362.
Gill, R.A., Kelly, R.H., Parton, W.J., Day, K.A., Jackson, R.B., Morgan, J.A., Scurlock, J.M.O.,Tieszen, L.L., Castle, J.V., Ojima, D.S., Zhang, X.S., 2002. Using simple environmentalvariables to estimate below-ground productivity in grasslands. Global Ecology andBiogeography 11, 79–86.
Gu, Y., Wylie, B.K., Howard, D.M., Phuyal, K.P., Ji, L., 2013. NDVI saturation adjustment: anew approach for improving cropland performance estimates in the Greater PlatteRiver Basin, USA. Ecology Indices 30, 1–6.
Heinsch, F.A., Reeves, M., Votava, P., Kang, S., Milesi, C., Zhao, M., Glassy, J., Jolly, W.M.,Loehman, R., Bowker, C.F., Kimball, J.S., Nemani, R.R., Running, S.W., 2003. User’sguide, GPP and NPP (MOD 17A2/A3) products. NASA MODIS land algorithm. Version2.0, December 2, 2003.
Huang, C., Li, Y., Liu, G., Zhang, H., Liu, Q., 2014. Recent climate variability and its impact onprecipitation, temperature, and vegetation dynamics in the Lancang River headwaterarea of China. International Journal of Remote Sensing 35, 2822–2834.
Huete, A.R., Liu, HuiQing, van Leeuwen, W.J.D., 1997. The use of vegetation indices in for-ested regions: issues of linearity and saturation. IGARSS’97. 1997 IEEE InternationalGeoscience and Remote Sensing Symposium Proceedings. Remote Sensing - A Scien-tific Vision for Sustainable Development 4, pp. 1966–1968.
Ji, W., Li, S., Chen, S., Shi, Z., Viscarra Rossel, R.A., Mouazen, A.M., 2016. Prediction of soilattributes using the Chinese soil spectral library and standardized spectra recordedat field conditions. Soil and Tillage Research 155, 492–500.
Jin, C., Xiao, X., Merbold, L., Arneth, A., Veenendaal, E., Kutsch, W.L., 2013. Phenology andgross primary production of two dominant savanna woodland ecosystems in South-ern Africa. Remote Sensing of Environment 135, 189–201.
Li, F., Chen, W., Zeng, Y., Zhao, Q., Wu, B., 2014a. Improving estimates of grassland frac-tional vegetation cover based on a pixel dichotomymodel: a case study in InnerMon-golia, China. Remote Sensing 6, 4705–4722.
Li, J., Cui, Y., Liu, J., Shi, W., Qin, Y., 2013. Estimation and analysis of net primary produc-tivity by integrating MODIS remote sensing data with a light use efficiency model.Ecological Modelling: Ecological Modelling for Global Change and Coupled Humanand Natural Systems 252, 3–10.
Li, Q., Tuo, D., Zhang, L., Wei, X., Wei, Y., Yang, N., Xu, Y., Anten, N.P.R., Pan, X., 2014b. Im-pacts of climate change on net primary productivity of grasslands in Inner Mongolia.Rangeland Journal 36, 493–503.
Lieth, H., 1973. Primary production: terrestrial ecosystems. Human Ecology 1, 303–332.Lin, H., 2009. A new model of grassland net primary productivity (NPP) based on the in-
tegrated orderly classification system of grassland. 2009 Sixth International Confer-ence on Fuzzy Systems and Knowledge Discovery. IEEE, Tianjin, China, pp. 52–56.
Lin, H., Feng, Q., Liang, T., Ren, J., 2012. Modelling global-scale potential grassland changesin spatio-temporal patterns to global climate change. International Journal of Sustain-ability in the Developing World Ecology 20, 83–96.
Lin, H., Wang, X., Zhang, Y., Liang, T., Feng, Q., Ren, J., 2013. Spatio-temporal dynamics onthe distribution, extent, and net primary productivity of potential grassland in re-sponse to climate changes in China. Rangeland Journal 35, 409–425.
Lin, H., Zhang, Y., 2013. Evaluation of six methods to predict grassland net primary pro-ductivity along an altitudinal gradient in the Alxa Rangeland, Western Inner Mongo-lia, China. Grassland Science 59, 100–110.
Lin, H., Zhao, J., Liang, T., Bogaert, J., Li, Z., 2012. A classification indices-based model fornet primary productivity (npp) and potential productivity of vegetation in china. In-ternational Journal of Biomathematics 5, 1260009.
Liu, J., Xu, X., Shao, Q., 2008. Grassland degradation in the “Three-River Headwaters” re-gion, Qinghai Province. Journal of Geographical Sciences 18, 259–273. https://doi.org/10.1007/s11442-008-0259-2.
Long, S.P., Garcia Moya, E., Imbamba, S.K., Kamnalrut, A., Piedade, M.T.F., Scurlock, J.M.O.,Shen, Y.K., Hall, D.O., 1989. Primary productivity of natural grass ecosystems of thetropics: a reappraisal. Plant and Soil 115, 155–166.
Luo, T., Pan, Y., Ouyang, H., Shi, P., Luo, J., Yu, Z., Lu, Q., 2004. Leaf area index and net pri-mary productivity along subtropical to alpine gradients in the Tibetan Plateau. GlobalEcology and Biogeography 13, 345–358.
Mao, X.F., Wei, X.Y., Xia, J.X., 2012. Evaluation of ecological migrants’ adaptation to theirnew living area in Three-River Headwater wetlands, China. Procedia EnvironmentalScience, 18th Biennial ISEM Conference on Ecological Modelling for Global Changeand Coupled Human and Natural System. 13. Procedia Environmental Sciences, Bei-jing, China, pp. 1346–1353.
McKenney, D.W., Pedlar, J.H., Papadopol, P., Hutchinson, M.F., 2006. The development of1901–2000 historical monthly climate models for Canada and the United States. Ag-riculture and Forestry Meteorology 138, 69–81.
National Aeronautics and Space Administration (NASA), 2015. the MODIS Land Products[WWW Document]. National Aeronaut. Sp. Adm. L. Process. Distribution Act. Arch.Center Land Processes Distributed Active Archive Center (LP DAAC). Available at:.https://lpdaac.usgs.gov/dataset_discovery/modis/modis_products_table, Accesseddate: 13 July 2018.
Piao, S., Fang, J., He, J., 2006. Variations in vegetation net primary production in the Qing-hai-Xizang Plateau, China, from 1982 to 1999. Climate Change 74, 253–267.
Piao, S., Fang, J., He, J., Xiao, Y., 2004. Spatial distribution of grassland biomass in China.Acta Phytoecology Sinica 28, 491–498.
Potter, C.S., Randerson, J.T., Field, C.B., Matson, P.A., Vitousek, P.M., Mooney, H.A., Klooster,S.A., 1993. Terrestrial ecosystemproduction: a processmodel based on global satelliteand surface data. Global Biogeochemical Cycles 7, 811–841.
Prince, S.D., Goward, S.N., 1995. Global primary production: a remote sensing approach.Journal of Biogeography 22, 815–835.
Ren, J., 1998. Research methods of grassland science (in Chinese). China AgriculturalPress, Beijing, China.
Ren, J.Z., Hu, Z.Z., Zhao, J., Zhang, D.G., Hou, F.J., Lin, H.L., Mu, X.D., 2008. A grassland clas-sification system and its application in China. Rangeland Journal 30, 199–209.
Ruimy, A., Saugier, B., Dedieu, G., 1994.Methodology for the estimationof terrestrial net primaryproduction from remotely sensed data. Journal of Geophysical Research 99, 5263–5283.
Running, S.W., Hunt Jr., E.R., 1993. Generalization of a forest ecosystem process model forother biomes, BIOME-BGC, and an application for global-scale models. In: Ehleringer,J.R., Field, C.B. (Eds.), Scaling physiological processes, physiological ecology. AcademicPress, San Diego, CA, USA, pp. 141–158.
Running, S.W., Thornton, P.E., Nemani, R., Glassy, J.M., 2000. Global terrestrial gross andnet primary productivity from the earth observing system. Methods in ecosystem sci-ence. Springer New York, New York, NY, USA, pp. 44–57.
Scurlock, J.M.O., Johnson, K., Olson, R.J., 2002. Estimating net primary productivity fromgrassland biomass dynamics measurements. Global Changes in Biology 8, 736–753.
Sun, Q., Li, B., Zhang, T., Yuan, Y., Gao, X., Ge, J., Li, F., Zhang, Z., 2017. An improved Biome-BGC model for estimating net primary productivity of alpine meadow on the Qing-hai-Tibet Plateau. Ecology Modelling 350, 55–68.
Tang, C., Fu, X., Jiang, D., Fu, J., Zhang, X., Zhou, S., 2014. Simulating spatiotemporal dy-namics of sichuan grassland net primary productivity using the CASA model and insitu observations. ScienceWorld Journal https://doi.org/10.1155/2014/956963 Avail-able at:, Accessed date: 8 November 2018.
Thornton, P.E., Law, B.E., Gholz, H.L., Clark, K.L., Falge, E., Ellsworth, D.S., Goldstein, A.H.,Monson, R.K., Hollinger, D., Falk, M., Chen, J., Sparks, J.P., 2002. Modeling andmeasur-ing the effects of disturbance history and climate on carbon and water budgets in ev-ergreen needleleaf forests. Agriculture Forest Meteorology 113, 185–222.
Wang, C., Long, R., Wang, Q., Jing, Z., Shi, J., 2009a. Changes in plant diversity, biomass andsoil C, in alpine meadows at different degradation stages in the headwater region ofthree rivers, China. Land Degradation and Development 20, 187–198.
335C. Wang et al. / Rangeland Ecology & Management 72 (2019) 327–335
Wang, H., Ni, J., Prentice, I.C., 2011. Sensitivity of potential natural vegetation in China toprojected changes in temperature, precipitation and atmospheric CO2. Regional Envi-ronmental Change 11, 715–727.
Wang, J., Liu, J., Shao, Q., Liu, R., Fan, J., Chen, Z., 2009b. Spatial-temporal patterns of netprimary productivity for 1988-2004 based on GLOPEM-CEVSA model in the “Three-River Headwaters” region of Qinghai Province, China. Journal of Plant Ecology (Chi-nese version) 33, 254–269.
Wang, Z., Song, K., Hu, L., 2010. China’s largest scale ecological migration in the Three-River Headwater region. Ambio 39, 443–446.
Wo, X., Wu, L., Zhang, J., Zhang, L., Liu, W., 2014. Estimation of net primary production inthe Three-River headwater region using CASA model. Journal of Arid Land ResourcesEnvironment 28, 45–50.
Xiao, X., Hollinger, D., Aber, J., Goltz, M., Davidson, E.A., Zhang, Q., Moore III, B., 2004. Sat-ellite-based modeling of gross primary production in an evergreen needleleaf forest.Remote Sensing of Environment 89, 519–534.
Yang, Y., Wang, Z., Li, J., Gang, C., Zhang, Y., Zhang, Y., Odeh, I., Qi, J., 2016. Comparativeassessment of grassland degradation dynamics in response to climate variation andhuman activities in China, Mongolia, Pakistan and Uzbekistan from 2000 to 2013.Journal of Arid Environments 135, 164–172.
Ye, J., Reynolds, J., Sun, G., Li, F., 2013. Impacts of increased variability in precipitation andair temperature on net primary productivity of the Tibetan Plateau: a modeling anal-ysis. Climate Change 119, 321–332.
Yeganeh, H., Khajedein, S. Jamale, Amiri, F., Shariff, A.R.B.M., 2012. Monitoring rangelandground cover vegetation using multitemporal MODIS data. Arabian Journal of Geosci-ence 7, 287–298.
Yu, L., Tang, L., Ding, W., Ma, M., Zhou, H., 2010. Characteristics and causes of changes ofalpine grassland productivity in the source region of Yellow River. 2010 18th Interna-
tional Conference on Geoinformatics. IEEE , pp. 1–6 Available at, Accessed date: 8 No-vember 2018 https://doi.org/10.1109/GEOINFORMATICS.2010.5567879.
Zaks, D.P.M., Ramankutty, N., Barford, C.C., Foley, J.A., 2007. From Miami to Madison: in-vestigating the relationship between climate and terrestrial net primary production.Global Biogeochemical Cycles 21, GB3004.
Zhang, X., Yang, D., Zhou, G., Liu, C., Zhang, J., 1996. Model Expectation of Impacts ofGlobal Climate Change on Biomes of the Tibetan Plateau. In: Omasa, K., Kai, K.,Taoda, H., Uchijima, Z., Yoshino, M. (Eds.), Climate Change and Plants in East Asia.Springer, Tokyo, Japan.
Zhang, Y., Qi, W., Zhou, C., Ding, M., Liu, L., Gao, J., Bai, W., Wang, Z., Zheng, D., 2014. Spa-tial and temporal variability in the net primary production of alpine grassland on theTibetan Plateau since 1982. Journal of Geographic Science 24, 269–287.
Zhao, M., Running, S.W., Nemani, R.R., 2006. Sensitivity of moderate resolution imagingspectroradiometer (MODIS) terrestrial primary production to the accuracy of meteo-rological reanalyses. Journal of Geophysical Research Biogeosciences 111, G01002.
Zhou, G., Zhang, X., 1996. Study on NPP of natural vegetation in China under global cli-mate change. Acta Phytoecology Sinica 20, 11–19.
Zhuang, Q., He, J., Lu, Y., Ji, L., Xiao, J., Luo, T., 2010. Carbon dynamics of terrestrial ecosys-tems on the Tibetan Plateau during the 20th century: an analysis with a process-based biogeochemical model. Global Ecology and Biogeography 19, 649–662.
Zhuang, Q., McGuire, A.D., Melillo, J.M., Clein, J.S., Dargaville, R.J., Kicklighter, D.W., Myneni, R.B., Dong, J., Romanovsky,V.E., Harden, J., Hobbie, J.E., 2003. Carbon cycling inextratropicalterrestrial ecosystems of the Northern Hemisphere during the 20th century: a modelinganalysis of the influences of soil thermal dynamics. Tellus B 55, 751–776.
Zomer, R.J., Trabucco, A., Bossio, D.A., Verchot, L.V., 2008. Climate change mitigation: aspatial analysis of global land suitability for clean development mechanism afforesta-tion and reforestation. Agriculture, Ecosystems, and the Environment 126, 67–80.
![Is NetTraveler APT managed by PLA Military Camp in Lanzhou [China] ???.](https://static.documents.pub/doc/80x56/54b731334a795912438b4583/is-nettraveler-apt-managed-by-pla-military-camp-in-lanzhou-china-.jpg)
![ARTICLE IN PRESS - · PDF fileaDepartment of Physics, Lanzhou University, 730000 Lanzhou, ... [1–3]. They have been used in solar cells, ... in this paper we tried to perform the](https://static.documents.pub/doc/80x56/5ab1e20a7f8b9a6b468d0f9f/article-in-press-of-physics-lanzhou-university-730000-lanzhou-13.jpg)
