Above-ground biomass estimation based on NPP time-series − A novel approach for biomass estimation in semi-arid Kazakhstan

Ecological Indicators 72 (2017) 13–22

Contents lists available at ScienceDirect

Ecological Indicators journal homepage: www.elsevier.com/locate/ecolind

Above-ground biomass estimation based on NPP time-series − A novel approach for biomass estimation in semi-arid Kazakhstan Christina Eisfelder a,∗ , Igor Klein a , Aruzhan Bekkuliyeva b , Claudia Kuenzer a , Manfred F. Buchroithner c , Stefan Dech a a

German Remote Sensing Data Center, DFD, German Aerospace Center, DLR, 82234 Oberpfaffenhofen, Germany Institute of Geography, Academy of Sciences, 99 Pushkin Str., 050010, Almaty, Kazakhstan c Institute for Cartography, Dresden University of Technology, 01062 Dresden, Germany b

a r t i c l e

i n f o

Article history: Received 23 February 2016 Received in revised form 2 July 2016 Accepted 25 July 2016 Keywords: Biomass Fractional cover Net primary productivity Kazakhstan Semi-arid environments Relative growth rates

a b s t r a c t Biomass is a sensitive indicator of environmental change and ecological functioning. Quantification of biomass is essential to identify and monitor those areas threatened by degradation and desertification. This is especially important in arid and semi-arid environments. However, robust techniques to monitor carbon stocks over large areas and through time are still missing. The major objective of the presented study is to develop a novel approach for biomass estimation in semi-arid environments using remotesensing based Net Primary Productivity (NPP) data. The developed methodical concept aims at derivation of above-ground grass and shrub biomass for natural environments. It is based on NPP time-series and plants’ relative growth rates. Fractional cover data provide information about grass and shrub coverage. The developed approach has been applied to three study areas in Kazakhstan, in which field data were collected for validation. Biomass maps were derived that show the spatial distribution of grass and shrub biomass. Validation revealed a moderate correlation (R = 0.68) with field data for grass biomass. For shrub biomass, a high correlation (R = 0.83) is retrieved when fractional cover information from field observations is used. The presented novel approach for biomass estimation is based on remote sensing derived NPP timeseries and is thus potentially transferable in space and time. This is a great advantage compared to commonly applied empirical relationships. The presented concept can be adapted to be applied to other vegetation communities. Providing the necessary data about fractional vegetation cover is available, the method will allow for repeated and large-area biomass estimation for natural semi-arid environments as needed for observing changes in biomass and support sustainable land management. © 2016 Published by Elsevier Ltd.

1. Introduction In the past years, rising concerns regarding the future evolution of the global climate and its impact on and interactions with the Earth’s ecosystems, have led to an increased interest in biomass estimation. Vegetation biomass is a sensitive indicator of environmental change and ecological functioning. It largely influences biodiversity and environmental processes such as the hydrological cycle, soil erosion, and degradation (e.g. Fabricius et al., 2003; Lu, 2006). It is a determining factor for ecosystem variability and

∗ Corresponding author. E-mail addresses: [email protected] (C. Eisfelder), [email protected] (I. Klein), [email protected] (A. Bekkuliyeva), [email protected] (C. Kuenzer), [email protected] (M.F. Buchroithner), [email protected] (S. Dech). http://dx.doi.org/10.1016/j.ecolind.2016.07.042 1470-160X/© 2016 Published by Elsevier Ltd.

resilience (Segoli et al., 2008; Verón et al., 2010) and provides important information for understanding the responses of vegetation to the climate system and currently observed global change. Biomass is also one of the essential climate variables (e.g. GCOS, 2005) and reporting on carbon stock levels is needed for several programmes (e.g. Costanza et al., 1997; MEAB, 2005; WRI, 2008). Standing biomass is the vegetation biomass per unit area at a definite point in time. It is usually measured in mass of carbon per unit area (e.g. in g C m−2 ). Changes in vegetation biomass can be quantified by the net primary productivity (NPP) of plants. NPP provides information about the biomass production by green vegetation, commonly in terms of carbon gain or loss. NPP is usually measured in mass of carbon per unit area per unit time (e.g. in g C m−2 year−1 ). Biomass estimation is especially important for arid and semiarid regions, which are particularly susceptible to environmental

14

C. Eisfelder et al. / Ecological Indicators 72 (2017) 13–22

degradation and desertification (UN, 2004; CDP, 2015). The quantification of biomass is essential to identify and monitor those areas under high risk of degradation and desertification (Hirata et al., 2001; Moleele et al., 2001) and to assess the degradation status of semi-arid regions (e.g. Liu et al., 2003). Large arid and semi-arid regions can be found in Kazakhstan (Lioubimtseva and Adams, 2004; Eisfelder et al., 2012), a country that has experienced varying human influences and political decisions with dramatic ecological and environmental consequences. Prominent examples are the decline of the Aral Sea and the ‘Virgin Lands Programme’ in the second half of the 20th century that led to dramatic steppe deterioration (de Beurs and Henebry, 2004). In addition to human impacts, also changing climate effects the environment. Temperatures have increased since beginning of the 20th century, and are expected to further increase 1–2 ◦ C by 2030–2050 (Lioubimtseva et al., 2005). Precipitation trends are highly variable, but indicate an overall decrease and aridity is expected to intensify (Lioubimtseva and Henebry, 2009). Due to these diverse anthropogenic and climatic influences, large areas in Kazakhstan are threatened by land degradation and desertification (ADB, 2010). Mapping of biomass distribution is essential to identify regions that are more vulnerable to changing climate, and thus support sustainable land management. Biomass quantification is obviously an important issue and robust, low-cost techniques to measure and monitor carbon stocks over large areas and through time are essential (e.g. Pfaff et al., 2000; Rosenqvist et al., 2003; Gibbs et al., 2007; Grassi et al., 2008; Mitchard et al., 2009). However, spatial biomass mapping and quantification are challenging tasks, especially for large areas. Field measurements cannot be applied large-scale as they involve destructive sampling. Satellite-based remote-sensing is able to cover large areas. Previous studies on biomass estimation, however, revealed several challenges for remote-sensing based biomass estimation in semi-arid regions, such as the influence of the soil background on the spectral response in ecosystems with scarce vegetation, efficient field sampling schemes for validation, and varying accuracy and site-specific performance of models applied (Eisfelder et al., 2012). The largest challenge is the transferability of biomass estimation approaches in time and space, which is of great importance for repeatable application and coverage of large areas. Modelling approaches are among the methods that obtained most promising results regarding transferability. Nevertheless, modelling approaches have not been extensively analysed in the context of biomass estimation yet (Eisfelder et al., 2012). NPP models are commonly applied for large areas and to obtain time-series for several years. Based on remote-sensing data, NPP models allow for NPP calculation with a reasonable resolution for regional studies (Eisfelder et al., 2014). The aim of the presented study is to develop a methodical approach for vegetation biomass estimation in semi-arid natural environments based on NPP time-series data. The innovative method is based on the fact that NPP is closely related to AGB (e.g. Titlyanova et al., 1999; Fensholt et al., 2006). The objectives of the presented study are (1) to analyse the suitability of combining NPP data with plants’ relative growth rates (RGRs) for biomass estimation, (2) to apply the developed approach for selected study areas in Kazakhstan, and (3) to validate the results with observations from field measurements.

2. Study area Three study areas were selected within Kazakhstan for validation of the developed approach. They present the most typical and representative semi-arid environments of Kazakhstan and are covered by predominantly natural vegetation. The location of the three

study areas is shown in Fig. 1. The first study area lies in Central Kazakhstan in the Karaganda oblast between the cities of Balkhash and Karaganda, between 46◦ –50◦ N and 72◦ –76◦ E. The second study area is situated in South Kazakhstan in the Zhambyl oblast. Geographically, the region is limited between 43◦ –45◦ N and 72◦ –75◦ E. A third study area is defined in West Kazakhstan. This study area spreads along the Ural River between the cities of Atyrau, at the northern shore of the Caspian Sea, and Chapaev near Kazakhstan’s northern border to Russia. This study area belongs to the oblasts West Kazakhstan and Atyrau and spreads between 47◦ –51◦ N and 50◦ –53◦ E. The northern part of the study area in Central Kazakhstan can be characterised as steppe, while the southern part is semidesert. The climate is semi-arid and high continental. Mean annual temperature generally increases from North to South from about 3.4 ◦ C (Karaganda) to 5.8 ◦ C (Balkhash). July average temperatures reach 20.8 ◦ C to 24.4 ◦ C. Mean annual precipitation shows a gradient from about 260 mm/a in the North to 150 mm/a in the South (Walter and Breckle, 1989). The Caspian Lowland in West Kazakhstan is arid with very cold winters. Aridity increases and rainfall decreases from North to South. The northernmost part belongs to the steppe zone, while the larger southern part can be characterised as semi-desert. The climate of the West Kazakh steppe is similar to Central Kazakhstan. The climate of the semi-deserts is more continental than in the steppe zones. The mean temperature in July is typically 24–26 ◦ C. Precipitation in the semi-desert reaches 160–250 mm per year (Berg, 1959). The southernmost part of the study area in West Kazakhstan shows characteristics of deserts. In the study area around Shu in South Kazakhstan also large deserts can be found, such as the arid Betpak-Dala with a spring or biseasonal rainy season and an annual precipitation of 100–150 mm, and the arid to semi-arid Mujunkum with precipitation amounts of 170–300 mm. Mean July temperatures reach 26–30 ◦ C with large daily amplitudes (Berg, 1959). The genera of characteristic plant species for the study areas in Kazakhstan are listed in Table 1. Species occurrence was derived from field observations and a detailed map of vegetation communities’ distribution (Volkova et al., 2010). Main grass and herbaceous genera are Agropyron, Artemisia, Festuca, Poa, and Stipa. Main shrub genera are Anabasis, Calligonum, Caragana, Haloxylon, Salsola, and Tamarix.

3. Data and methods 3.1. NPP time-series data NPP data for Kazakhstan were available from Eisfelder et al. (2014). They were calculated with the Biosphere Energy Transfer Hydrology (BETHY/DLR) model, which has been adapted at the German Aerospace Center (DLR) to be driven by remote sensing data (Eisfelder et al., 2013; Wißkirchen et al., 2013). The spatial resolution of the NPP output product depends on the resolution of leaf area index (LAI) and land cover input data. The available NPP data for Kazakhstan have a spatial resolution of ∼0.00833◦ (ca. 925 m × 650 m for Kazakhstan). They cover the years 2003–2011. Continuous time-series of climatic and phenological input data allow for a very high temporal resolution of NPP outputs of one day (Eisfelder et al., 2014). The calculation of NPP for Kazakhstan with BETHY/DLR was based on the Central Asia land cover and land use map (Klein et al., 2012). This classification is used to define the vegetation types. For each land cover pixel, two vegetation types can be defined. A weighting factor gives the relative spatial fraction of the primary and the secondary vegetation type. This approach allows to NPP for coverage of less than 100% and allows for separate calculation of grass NPP and shrub NPP for each pixel. For most natural vegeta-

C. Eisfelder et al. / Ecological Indicators 72 (2017) 13–22

15

Fig. 1. Map showing the location of the three study areas within Kazakhstan (left: West Kazakhstan, center: South Kazakhstan, and right: Central Kazakhstan). The individual test sites are marked with red triangle signatures. The Central Asia land cover and land use map (Klein et al., 2012) illustrates the land cover in 2009 (for interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article).

Table 1 Main plant genera that are characteristic for the three study areas in Central, South, and West Kazakhstan; plant life-form (G: Grass, H: Herb, S: Shrub) and occurrence (X: Genus is typical for study area). Genus

Plant life-form (grass, herb or shrub)

Agropyron Anabasis Artemisia Calligonum Caragana Festuca Haloxylon Poa Salsola Stipa Tamarix

G S H S S G S G S G S

Central Kazakhstan

South Kazakhstan

West Kazakhstan

X

X X X

X X X X

X X

X X

tion classes (sparse vegetation, grassland, closed shrubland, open shrubland, and bare areas) the types C3 grass and deciduous shrubs were applied for NPP calculation. 3.2. RGR literature data Relative growth rates of plants have been widely studied in plant physiology. They provide information on the speed of plant growth. The RGR of a plant is defined as the mass increase per standing biomass per unit of time, for example as g g−1 d−1 . Some studies consider total plant biomass (TPB), while others relate the increase only to the above-ground biomass (AGB) present, as shown in equation (1) (Levang-Brilz and Biondini, 2002). RGRAGB =

1 dTPB AGB dt

(1)

TPB and AGB are commonly defined as dry weight biomass of an individual species at the time of investigation (Shipley and Keddy, 1988). Maximum growth rates are often derived (e.g. Shipley and Keddy, 1988; Levang-Brilz and Biondini, 2002; Vile et al., 2006),

X X X X X

X X X

which define the dry weight increase per unit of biomass and per unit of time under optimum conditions (Poorter and Remkes, 1990). For the main plant genera listed in Table 1, maximum RGR values were selected from the literature. RGRs for species that grow in semi-arid regions, especially from Central Asia and similar environments, were preferred. RGRs for relevant grass/herbaceous species were available from Wilhelm and Nelson (1978), Eissenstat and Caldwell (1987), Oesterheld (1992), Reich et al. (2003), and Zheng et al. (2008). RGRs for shrub species were available from Song et al. (2006), Zheng et al. (2008), and Hayes et al. (2009). The available data from Eissenstat and Caldwell (1987) cover both seedling experiments and data for established plants for Agropyron species and can thus also provide information about how strong RGRs recess and which fraction of seedling RGRs can be applied for older plants. Additional data were available from Levang-Brilz and Biondini (2002), who experimentally investigated maximum RGR of both TPB and AGB for 55 plant species (including 38 shrubs and 9C3 grasses), of which several genera are also present in semi-arid Kazakhstan (e.g. Artemisia, Agropyron, Poa, and Stipa).

16

C. Eisfelder et al. / Ecological Indicators 72 (2017) 13–22

dry weight above-ground biomass, BF: fresh weight above-ground biomass), as shown in equations (3) and (4).

Fig. 2. Schematic visualisation of the sampling design. For each test site the vegetation is stratified according to apparent vegetation bulks in four strata: 0 = bare, 1 = low, 2 = medium, 3 = high. A minimum of 12 sample plots (three for strata 1 and 3, six for strata 2) are randomly selected along the 1000 m transect (adapted from Hiernaux et al., 2009).

BDgrass/herbs = 0.546 · BFgrass/herbs

(3)

BDshrubs = 0.618 · BFshrubs

(4)

Finally, for comparison of field data and biomass estimation results, the carbon content of the above-ground biomass field data was calculated using conversion factors published by the IPCC (2006) (herbaceous biomass: 0.47 t C (t DM)−1 , woody biomass: 0.50 t C (t DM)−1 ). Moreover, based on ground cover recordings, mean fractional vegetation cover (F¯ i ) for grass and shrubs was calculated for each test site according to equation (5), with k number of strata, pk the absolute frequency of strata k recorded along the transect, ¯ i,k the arithmetic mean of ground cover of plant life-form and m i (grasses/herbs or shrubs) in strata k. Additionally, relative cover (r F¯ i ) of grass and shrubs was derived with F¯ i mean fractional cover of the vegetation type i and F¯ j mean fractional cover of the other plant life-form. F¯ i =

 k   ¯ pk m  i,k pk

1

F¯ · 100 r F¯ i = i F¯ i + F¯ j

3.3. Biomass field data For validation of the biomass estimates, field data were collected in the three study areas in Kazakhstan in June 2011. Thirty test sites were sampled that span a wide range of typical biomass amounts. The test sites were located in areas of relatively homogeneous vegetation cover. In total, 364 destructive sample plots were collected within the 30 test sites in June 2011. Biomass field data collection for each test site followed a stratified random sampling design (Hiernaux et al., 2009). The applied sampling approach combines non-destructive stratification along transects with destructive measurements of 1 m2 sample plots (see Fig. 2). This approach is especially designed for obtaining measurements that are suitable for comparison at approximately 1 km2 spatial resolution. The sample plots represented the three strata of low, medium, and high biomass loads. The frequency of strata was recorded along a 1 km transect for each test site. The biomass ¯ was then calculated from the frequency of the of each test site (M) vegetation strata k (pk ) and the arithmetic mean of sample items ¯ k ). in strata k (m

¯ = M

 k   ¯k pk m  1

pk

(2)

After recording of attributes, the aboveground vegetation within the 1 m2 sample plots was harvested to ground level. For the collected field data in Central Kazakhstan and South Kazakhstan, dry weight was obtained after oven drying for 48 h at 60 ◦ C at an institute in Almaty. For each sample, the oven-drying was done separately for grass/herbs and shrubs. The samples collected in West Kazakhstan could not be oven-dried. Therefore, mean conversion factors from fresh to dry biomass were derived based on fresh and dry weights measured in the study areas Central and South Kazakhstan in June 2011. For grass/herbs 192 individual samples were considered and a conversion factor of 54.6% (standard deviation: 18% absolute) derived. Shrub biomass was recorded at 85 sample plots; the dry weight was at average 61.8% (standard deviation: 16% absolute) of fresh weight. These conversion factors were applied to derive dry weight biomass for the test sites in West Kazakhstan (BD:

(5)

(6)

3.4. Methodical approach for derivation of biomass estimates from NPP time-series A plant’s absolute growth rate depends on both the amount of plant biomass and its relative growth rate (RGR) (Oesterheld, 1992). Since NPP describes the absolute growth rate (e.g. in g d−1 ), knowledge about the plant’s relative growth rate can be used to derive the plant’s standing biomass from NPP. As Titlyanova et al. (1999), for example, formulated, NPP can be described as the product of the relative rate of dry matter production per day RGR (e.g. in g g−1 d−1 ), and the standing dry matter biomass B (e.g. in g), as presented in equation (5). NPP = RGR · B

(5)

The most relevant time for biomass estimation is at peak biomass or at maximum productivity (e.g. Hobbs, 1995; Mangiarotti et al., 2008). For Kazakhstan, the period of maximum vegetation productivity is in June. The available field data for validation were also collected in June 2011. Maximum RGRs are available for several species of genera typical for the study areas in Kazakhstan. Thus, the developed approach aims at estimating standing biomass for the period of maximum vegetation growth. Equation (5) establishes the relationship between NPP, biomass, and RGR. This equation can be transposed, so that it defines how biomass can be derived from NPP and RGR. As the aim of this study is to estimate the above-ground biomass for the period of maximum vegetation growth, i.e. June for Kazakhstan, maximum daily relative growth rates are combined with the maximum daily NPP for this period (equation (6)). AGB =

NPPmax RGRAGB,max

(6)

NPP for grass and shrubs was modelled separately for each pixel with BETHY/DLR. Thus, it is possible to apply RGR values separately for these two vegetation types to derive grass and shrub biomass. Distinction between grass and shrub RGRs is necessary, because several studies have reported that RGR is generally lower for woody species than for herbaceous species (e.g. Grime and Hunt, 1975; Hunt and Cornelissen, 1997).

C. Eisfelder et al. / Ecological Indicators 72 (2017) 13–22

17

NPP me-series modelled with BETHY/DLR

Daily NPP for grass

Daily NPP for shrubs

Maximum RGRs for individual species

Available in the literature

Defined in land cover map / from field observaons

Maximum daily NPP for grass

Maximum daily NPP for shrubs

Adjusted maximum RGRs in terms of AGB for individual species

Conversion factors for RGRs

Relave fraconal cover for grass/shrubs

Derived from literature data Maximum daily NPP for grass fracon

AGB of grass

Maximum daily NPP for shrub fracon

Average maximum RGRAGB for grasses

Average maximum RGRAGB for shrubs

AGB of shrubs

Total AGB of grass and shrubs Fig. 3. Flowchart showing the approach for derivation of above-ground biomass estimates from NPP data and RGR values for grass and shrub biomass.

Fig. 3 shows a flowchart of the developed approach for estimation of AGB from NPP. The upper left part of the diagram shows the derivation of the required NPP values. NPP input data comprise daily time-series of grass NPP and shrub NPP for each pixel. From the time-series, the maximum daily NPP values for grass and shrubs for the period of maximum vegetation productivity (e.g. June 2011 in Kazakhstan) are extracted. The maximum daily NPP values are then combined with information about relative fractional cover of grass and shrubs. This information is used to scale the NPP of grass and shrubs according to their relative fractional cover. This results in maximum daily NPP values for the grass fraction and the shrub fraction within a pixel. The right part of the diagram illustrates the preparation steps for RGR values. The main input parameters are maximum RGRs for individual species. These were selected from the literature. Some RGR values need to be adjusted to be applicable to established plants. For this, an adjustment factor has been derived based on the experiment data from Eissenstat and Caldwell (1987). Their data indicated a reduction of RGR to about 57% for established plants compared to seedlings. A second conversion of the RGR values is necessary for those studies that aimed at deriving RGRTPB instead of RGRAGB . The needed conversion factor regarding the relation between RGRTPB and RGRAGB was calculated based on the data from Levang-Brilz and Biondini (2002), separately for C3 grass and shrub species. This is important, because the two plant life-forms feature significantly different root-shoot ratios, which influences the ratio between TPB and AGB. The obtained ratios indicate that RGRAGB was at average 150.2% of RGRTPB for shrub species. For grass species, RGRAGB was at average 166.1% of RGRTPB . These percentages are applied to the available RGRs of grasses and shrubs to derive RGRs in terms of AGB. Finally, average RGRAGB values for the plant life-forms of grasses and shrubs are needed. Average RGRAGB values for grass were derived separately for the three study areas based on mean RGRAGB of respective characteristic genera (Table 2). Derived RGRAGB values are 44 mg g−1 d−1 for Central Kazakhstan, 46.5 mg g−1 d−1 for South Kazakhstan, and 53 mg g−1 d−1 for West Kazakhstan. RGRAGB values for shrubs were defined by the characteristic shrub genera for the study area and are 11.1 mg g−1 d−1 for Central Kazakhstan

(Caragana), 19.6 mg g−1 d−1 for South Kazakhstan (Haloxylon), and 28.3 mg g−1 d−1 for West Kazakhstan (Tamarix). Finally, the maximum NPP values and the maximum RGRAGB values are combined to estimate AGB. This results in estimates of grass and shrub AGB. These two values can finally be summed to obtain the total standing AGB of grass and shrubs. 4. Results 4.1. Biomass estimation results for the three study areas The developed method is applied to the NPP data derived with BETHY/DLR for the three study areas in Central, South, and West Kazakhstan. For demonstration and validation of the approach, two biomass estimates are derived based on different fractional cover definitions. For application to the entire study areas, the coverage definitions from the Central Asia land cover and land use map (Klein et al., 2012) are used. For the test sites, additionally, biomass estimates based on fractional cover information from the field are obtained. The spatial resolution of the resulting biomass estimate is ∼0.00833◦ (ca. 925 m × 650 m for Kazakhstan), the same as for NPP data. For pixels that are classified as agriculture in the land cover and land use classification no biomass is calculated. For forest areas only biomass of the secondary class ‘C3 short grass’ is obtained. Finally, estimated biomass amounts for grass and shrubs are summed to obtain a map of total AGB for the three study areas. The results of the above-ground biomass estimation for grass and shrub biomass are presented in Figs. 4 and 5 respectively. Additionally, Fig. 6 shows the result of the sum of grass and shrub biomass. The obtained above-ground grass biomass for Central Kazakhstan shows higher biomass in the North of the study area. South of Akshatau grass AGB is typically between 5 and 30 g C m−2 , north of Akshatau between 30 and 100 g C m−2 . In West Kazakhstan also high grass biomass can be observed for large areas in the North, where grass AGB is typically 40–80 g C m−2 . In the southern part grass biomass is mainly between 10 and 20 g C m−2 . Above-ground grass biomass estimates for the study area in South Kazakhstan show high biomass values along the valley of the Shu River. Highest grass biomass with up to >200 g C m−2

18

C. Eisfelder et al. / Ecological Indicators 72 (2017) 13–22

Table 2 Mean RGRAGB values for grass genera and resulting average RGRAGB for the three study areas in Central, South, and West Kazakhstan. X: Genus is typical for study area. Genus

Mean RGRAGB [mg g−1 d−1 ]

Agropyron Artemisia Festuca Stipa Average RGRAGB [mg g−1 d−1 ]

79.7 26.5 39.0 66.5

Central Kazakhstan

South Kazakhstan

X X X 44.0

X X 46.5

West Kazakhstan X X X X 53.0

Fig. 4. Result of above-ground grass biomass for June 2011 derived from the developed approach for the three study areas in Central, South, and West Kazakhstan.

Fig. 5. Result of above-ground shrub biomass for June 2011 derived from the developed approach for the three study areas in Central, South, and West Kazakhstan.

C. Eisfelder et al. / Ecological Indicators 72 (2017) 13–22

19

Fig. 6. Result of above-ground grass and shrub biomass for June 2011 derived from the developed approach for the three study areas in Central, South, and West Kazakhstan. This map shows the sum of the above-ground grass and shrub biomass.

can be observed in the South-East of the study area. This is a mountainous grassland region with Kastanozem soil and relatively high precipitation. For other parts of the study area in South Kazakhstan only low grass biomass, mainly between 10 and 20 g C m−2 is obtained. The results of above-ground shrub biomass show a different pattern. In Central Kazakhstan high shrub biomass is estimated in the South, within a strip north of Lake Balkhash. Above-ground shrub biomass in this zone is between 30 and 60 g C m−2 . For most other parts of Central Kazakhstan above-ground shrub biomass between 5 and 25 g C m−2 is obtained. In West Kazakhstan low shrub biomass with typically 15–30 g C m−2 covers the southern part of the study area. Highest shrub biomass occurs in the very northern part with up to 100–150 g C m−2 . Between these two regions a strip with low shrub biomass can be observed especially in the eastern part. In the study area in South Kazakhstan large areas show above-ground shrub biomass between 20 and 40 g C m−2 . The grassland region in the South-East shows very low shrub biomass. Along the Shu River and south-west of the Shu River valley some areas with shrub biomass of up to 80–150 g C m−2 can be found.

4.2. Validation of above-ground biomass estimates with field data Two biomass estimates were obtained for the test sites in Kazakhstan. The first estimate is calculated using fractional cover values based on the land cover and land use map. The second estimate is calculated using fractional cover estimates derived from field observations. The estimated above-ground biomass of grass and shrubs are compared to field-measured above-ground biomass. For the validation of grass biomass, 28 test sites are available; for shrub biomass, 20 test sites. The number for shrubs is lower than for grass, because shrubs were present at fewer test sites. Results of comparison for grass biomass are presented in Fig. 7 and results for shrub biomass are presented in Fig. 8. As the results of the comparison show, for grass, ground-based biomass observations are higher than biomass estimates from NPP

data for most test sites (cf. Fig. 7). The underestimation is especially strong for the biomass estimates derived from fractional cover definitions based on the land cover map. The overall RMSE is 28.5 g C m−2 and the correlation coefficient R is 0.68. The biomass results obtained with fractional cover information from the field show better results (Fig. 7, right). The estimated biomass for the individual test sites becomes closer to the field observations. This results in an overall RMSE of about 25.4 g C m−2 . The correlation coefficient R is 0.64. Above-ground shrub biomass estimates calculated from NPP with fractional cover definitions based on the land cover map show a low correlation to the field-based biomass data with R of 0.33 (cf. Fig. 8, left). The RMSE of 56 g C m2 is also high. The relationship is significantly better when fractional cover information from field observations is used (cf. Fig. 8, right). Here, the correlation coefficient R reaches 0.83. The slope of the regression line is 0.51. This slope is strongly influenced by two sites with very high field-based biomass. For these two sites the results indicate an underestimation of 50%. For all other test sites the above-ground shrub biomass estimates based on field fractional cover ratios are closer to the field observations. The overall RMSE is 24.4 g C m−2 and only three test sites show an error higher than ±20 g C m−2 compared to field biomass observations.

5. Discussion In this study, an approach for derivation of AGB estimates for natural environments based on NPP data has been developed. The methodical concept has been presented and the application has been demonstrated for three study areas in semi-arid Kazakhstan. The biomass estimation approach is based on NPP data, RGRs, and fractional cover information. It is designed to derive the standing biomass for the period of maximum vegetation growth. The validation with field data showed that NPP-based aboveground grass biomass estimates are lower than field-observed grass biomass for most test sites for both fractional cover bases. This indi-

20

C. Eisfelder et al. / Ecological Indicators 72 (2017) 13–22

Fig. 7. Correlation between grass biomass derived from NPP data and grass biomass from field observation. Left: Biomass from NPP data derived with fractional cover based on land cover classes. Right: Biomass from NPP data derived with fractional cover from field observation.

Fig. 8. Correlation between shrub biomass derived from NPP data and shrub biomass from field observation. Left: Biomass from NPP data derived with fractional cover based on land cover classes. Right: Biomass from NPP data derived with fractional cover from field observation.

cates that the NPP input is too low and/or that the applied RGRs for grass are too high. The validation of the grass NPP input data, which were calculated with BETHY/DLR, indicated that the modelled NPP results underestimate the NPP (Eisfelder et al., 2014). To reduce the possible error associated with the RGR values, more experimental studies with plants typical for the study areas in Kazakhstan are needed. Particularly, measurements of maximum RGRs of mature plants in terms of AGB would be desirable. The moderate overall correlation between modelled grass biomass and field-observed grass biomass (R = 0.64) might be caused by a high variation of herbaceous species present within the test sites. The applied RGR values within this study did not differentiate between different herbaceous vegetation types within a study area and for NPP calculation with BETHY/DLR only two grass types were distinguished. Information about the spatial distribution of different vegetation types and the separate modelling of additional vegetation types with BETHY/DLR could improve the grass biomass results. The validation of the shrub biomass estimates showed a large difference between the results obtained with the two different fractional cover bases. This reveals the high importance of accurate fractional cover information for shrub biomass estimation. As comparison to field data shows, the fractional cover definitions based on the land cover map have a mean absolute error of 25%. The accuracy for individual sites seems not sufficient for

reliable shrub biomass estimation. Unfortunately, a suitable fractional cover dataset currently does not exists for Kazakhstan. Such information can, however, be derived from remote sensing data (e.g. Gessner et al., 2013; Shoshany and Svoray, 2002) and improve remote sensing based biomass retrieval (Svoray and Shoshany, 2003). Provided that accurate information on the fractional cover of shrubs is available, the developed approach yields good estimates of above-ground shrub biomass (R = 0.83). The focus of this study was on biomass estimation for natural vegetation in semi-arid environments in Kazakhstan. The methodical development therefore focused on derivation of grass and shrub biomass, which are the two main vegetation types in these areas. The developed approach could also be applied to other natural grasslands and shrublands, but the RGR values would have to be adapted. The developed approach can also be applied to other NPP data, providing that these are available with high temporal resolution. This is necessary for the combination with maximum RGRs. The availability of separate NPP data for different vegetation types within a pixel is also essential. As a precondition for the applicability of constant RGR values, the individual NPP values have to represent a characteristic mixture of species and plants in order to achieve a natural age distribution of plants within the site. Thus, the NPP data should be based on spatially explicit remote sensing

C. Eisfelder et al. / Ecological Indicators 72 (2017) 13–22

datasets with a medium to large pixel size to ensure sufficient plant and species variability within each pixel. The presented approach for biomass estimation is novel, as no research combining NPP and RGR to estimate standing biomass in semi-arid environments has been reported previously. RGR is the most useful single comparator of innate growth potential (Hunt and Cornelissen, 1997), but has not yet been applied in the context of estimation of biomass from NPP data. Because NPP is usually modelled for several years and large areas, the developed approach has the advantage of better transferability than the commonly applied regression approaches (Eisfelder et al., 2012). If the presented approach will be refined and successfully applied over large areas in future, this would be a big step forward for several applications that need large-scale biomass estimates on a regular basis. 6. Conclusion In this study we have presented a novel approach for biomass estimation in natural semi-arid environments based on NPP data. The developed methodology is new, as NPP data and RGR values have not been combined previously for biomass mapping. The approach aims at derivation of the AGB of grass and shrubs for the period of maximum vegetation growth. Validation with field data from three study areas in Kazakhstan showed that biomass information can successfully be estimated based on NPP and RGRs. However, accurate fractional cover information is needed, especially for shrub biomass derivation. Compared to the commonly applied empirical relationships between remote-sensing-derived indices and biomass field data, this approach has the advantage of a potentially better transferability. Providing the necessary data regarding vegetation cover is available, this method will allow for repeated and large-area biomass estimations of natural vegetation in Kazakhstan and other semi-arid environments. The combination of annual estimations of biomass with continuous NPP time-series might allow for the setup of an operational monitoring system in the future. This would be a big step forward for observing and understanding the changes in biomass in semi-arid areas. It would also provide essential base information for sustainable land management, the derivation of possible climate change related trends or value-added carbon storage products. Acknowledgements We thank Prof. Dr. Farida Akiyanova from the Institute of Geography, Almaty, for providing support for the field campaign and Dr. Kazbek Toleubayev from the Kazakh Research Institute for Plant Protection and Quarantine, Almaty, and his laboratory staff for oven drying of biomass samples. The study by C. Eisfelder was funded by DLR-DFD in the frame of the Network EOS PhD Program. The authors are grateful to the anonymous reviewers for their constructive comments. References ADB Asian Development Bank, 2010. Central Asia Atlas of Natural Resources, Central Asian Countries Initiative for Land Management. Asian Development Bank, Manila, Philippines. Berg, L.S., 1959. Die geographischen Zonen der Sowjetunion Band 2. B.G. Teubner, Leipzig [in German]. CDP, 2015. CDP Global Climate Change Report 2015, Available at: https://www. cdp.net/CDPResults/CDP-global-climate-change-report-2015.pdf (accessed 20 January 2016). Costanza, R., d’Arge, R., de Groot, R., Farber, S., Grasso, M., Hannon, B., Limburg, K., Naeem, S., O’Neill, R.V., Paruelo, J., Raskin, R.G., Sutton, P., van den Belt, M., 1997. The value of the world’s ecosystem services and natural capital. Nature 387, 253–260.

21

de Beurs, K.M., Henebry, G.M., 2004. Land surface phenology, climatic variation, and institutional change: analyzing agricultural land cover change in Kazakhstan. Remote Sens. Environ. 89 (4), 423–433. Eisfelder, C., Kuenzer, C., Dech, S., 2012. Derivation of biomass information for semi-arid areas using remote sensing data. Int. J. Remote Sens. 33 (9), 2937–2984. Eisfelder, C., Kuenzer, C., Dech, S., Buchroithner, M.F., 2013. Comparison of two remote sensing based models for regional net primary productivity estimation −a case study in semi-arid Central Kazakhstan. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 6 (4), 1843–1856. Eisfelder, C., Klein, I., Niklaus, M., Kuenzer, C., 2014. Net primary productivity in Kazakhstan, its spatio-temporal patterns and relation to meteorological variables. J. Arid Environ. 103, 17–30. Eissenstat, D.M., Caldwell, M.M., 1987. Characteristics of successful competitors: an evaluation of potential growth rate in two cold desert tussock grasses. Oecologia 71 (2), 167–173. Fabricius, C., Burger, M., Hockey, P.A.R., 2003. Comparing biodiversity between protected areas and adjacent rangeland in xeric succulent thicket, South Africa: arthropods and reptiles. J. Appl. Ecol. 40 (2), 392–403. Fensholt, R., Sandholt, I., Schultz Rasmussen, M., Stisen, S., Diouf, A., 2006. Evaluation of satellite based primary production modelling in the semi-arid Sahel. Remote Sens. Environ. 105 (3), 173–188. GCOS, 2005. GCOS Regional Action Plan for Eastern and Central Europe, December 2005. Available at: http://www.wmo.int/pages/prog/gcos/documents/GCOS ECE RAP Dec05.pdf (accessed 16 March 2010). Gessner, U., Machwitz, M., Conrad, C., Dech, S., 2013. Estimating the fractional cover of growth forms and bare surface in savannas: a multi-resolution approach based on regression tree ensembles. Remote Sens. Environ. 129, 90–102. Gibbs, H.K., Brown, S., Niles, J.O., Foley, J.A., 2007. Monitoring and estimating tropical forest carbon stocks: making REDD a reality. Environ. Res. Lett. 2 (4), 045023 (13 pp.). Grassi, G., Monni, S., Federici, S., Achard, F., Mollicone, D., 2008. Applying the conservativeness principle to REDD to deal with the uncertainties of the estimates. Environ. Res. Lett. 3 (3), 035005 (12 pp.). Grime, J.P., Hunt, R., 1975. Relative growth-rate: its range and adaptive significance in a local flora. J. Ecol. 63 (2), 393–422. Hayes II, W.E., Walker, L.R., Powell, E.A., 2009. Competitive abilities of Tamarix aphylla in southern Nevada. Plant Ecol. 202 (1), 159–167. Hiernaux, P., Mougin, E., Diarra, L., Soumaguel, N., Lavenu, F., Tracol, Y., Diawara, M., 2009. Sahelian rangeland response to changes in rainfall over two decades in the Grouma region, Mali. J. Hydrol. 375 (1–2), 114–127. Hirata, M., Kogab, N., Shinjo, H., Fujita, H., Gintzburger, G., Miyazaki, A., 2001. Vegetation classification by satellite image processing in a dry area of northeastern Syria. Int. J. Remote Sens. 22 (4), 507–516. Hobbs, T.J., 1995. The use of NOAA-AVHRR NDVI data to assess herbage production in the arid rangelands of Central Australia. Int. J. Remote Sens. 16 (7), 1289–1302. Hunt, R., Cornelissen, J.H.C., 1997. Components of relative growth rate and their interrelations in 59 temperate plant species. New Phytol. 135 (3), 395–417. IPCC, 2006. 2006 IPCC Guidelines for National Greenhouse Gas Inventories. Prepared by the National Greenhouse Gas Inventories Programme. In: Eggleston, H.S., Buendia, L., Miwa, K., Ngara, T., Tanabe, K. (Eds.). IGES, Japan, Available at: http://www.ipcc-nggip.iges.or.jp/public/2006gl/index.html (accessed 10 May 2012). Klein, I., Gessner, U., Kuenzer, C., 2012. Regional land cover mapping and change detection in Central Asia using MODIS time series. Appl. Geogr. 35 (1–2), 219–234. Levang-Brilz, N., Biondini, M.E., 2002. Growth rate, root development and nutrient uptake of 55 plant species from the Great Plains Grasslands, USA. Plant Ecol. 165 (1), 117–144. Lioubimtseva, E., Adams, J.M., 2004. Possible implications of increased carbon dioxide levels and climate change for desert ecosystems. Environ. Manage. 33 (1), 388–404. Lioubimtseva, E., Henebry, G., 2009. Climate and environmental change in arid Central Asia: impacts, vulnerability, and adaptations. J. Arid Environ. 73 (11), 963–977. Lioubimtseva, E., Cole, R., Adams, J.M., Kapustin, G., 2005. Impacts of climate and land-cover changes in arid lands of Central Asia. J. Arid Environ. 62 (2), 285–308. Liu, A.X., Liu, Z.J., Wang, C.Y., Niu, Z., 2003. Monitoring of desertification in central Asia and western China using long term NOAA-AVHRR NDVI time-series data. 21–25 July 2003, Proceedings, IEEE International In: Proceedings of International Geoscience and Remote Sensing Symposium 2003, (IGARSS 2003), 4, pp. 2278–2280. Lu, D., 2006. The potential and challenge of remote sensing-based biomass estimation. Int. J. Remote Sens. 27 (7), 1297–1328. MEAB Millennium Ecosystem Assessment Board, 2005. Living Beyond Our Means. Natural Assets and Human Well-Being. Statement from the Board. Millenium Ecosystem Assessment Board, Available at: http://www.millenniumassessment.org/documents/document.429.aspx.pdf (accessed 16 March 2010). Mangiarotti, S., Mazzega, P., Jarlan, L., Mougin, E., Baup, F., Demarty, J., 2008. Evolutionary bi-objective optimization of a semi-arid vegetation dynamics model with NDVI and ␴0 satellite data. Remote Sens. Environ. 112 (4), 1365–1380.

22

C. Eisfelder et al. / Ecological Indicators 72 (2017) 13–22

Mitchard, E.T.A., Saatchi, S.S., Woodhouse, I.H., Nangendo, G., Ribeiro, N.S., Williams, M., Ryan, C.M., Lewis, S.L., Feldpausch, T.R., Meir, P., 2009. Using satellite radar backscatter to predict above-ground woody biomass: a consistent relationship across four different African landscapes. Geophys. Res. Lett. 36, L23401 (6 pp.). Moleele, N., Ringrose, S., Arnberg, W., Lunden, B., Vanderpost, C., 2001. Assessment of vegetation indexes useful for browse (forage) prediction in semi-arid rangelands. Int. J. Remote Sens. 22 (5), 741–756. Oesterheld, M., 1992. Effect of defoliation intensity on aboveground and belowground relative growth rates. Oecologia 92 (3), 313–316. Pfaff, A.S.P., Kerr, S., Hughes, R.F., Liu, S., Sanchez-Azofeifa, G.A., Schimel, D., Tosi, J., Watson, V., 2000. The Kyoto protocol and payments for tropical forest: an interdisciplinary method for estimating carbon-offset supply and increasing the feasibility of a carbon market under the CDM. Ecol. Econ. 35 (2), 203–221. Poorter, H., Remkes, C., 1990. Leaf area ratio and net assimilation rate of 24 wild species differing in relative growth rate. Oecologia 83 (4), 553–559. Reich, P.B., Buschena, C., Tjoelker, M.G., Wrage, K., Knops, J., Tilman, D., Machado, J.L., 2003. Variation in growth rate and ecophysiology among 34 grassland and savanna species under contrasting N supply: a test of functional group differences. New Phytol. 157 (3), 617–631. Rosenqvist, Å., Milne, A., Lucas, R., Imhoff, M., Dobson, C., 2003. A review of remote sensing technology in support of the Kyoto Protocol. Environ. Sci. Policy 6 (5), 441–455. Segoli, M., Ungar, E.D., Shachak, M., 2008. Shrubs enhance resilience of a semi-arid ecosystem by engineering and regrowth. Ecohydrology 1 (4), 330–339. Shipley, B., Keddy, P.A., 1988. The relationship between relative growth rate and sensitivity to nutrient stress in twenty-eight species of emergent macrophytes. J. Ecol. 76 (4), 1101–1110. Shoshany, M., Svoray, T., 2002. Multidate adaptive unmixing and its application to analysis of ecosystem transitions along a climatic gradient. Remote Sens. Environ. 82, 5–20. Song, T., Ding, X., Feng, G., Zhang, F., 2006. Nutritional and osmotic roles of nitrate in a euhalophyte and a xerophyte in saline conditions. New Phytol. 171 (2), 357–366. Svoray, T., Shoshany, M., 2003. Herbaceous biomass retrieval in habitats of complex composition: a model merging SAR images with unmixed Landsat TM data. IEEE Trans. Geosci. Remote Sens. 41 (7), 1592–1601.

Titlyanova, A.A., Romanova, I.P., Kosykh, N.P., Mironycheva-Tokareva, N.P., 1999. Pattern and process in above-ground and below-ground components of grassland ecosystems. J. Veg. Sci. 10 (3), 307–320. UN, 2004. A More Secure World: Our Shared Responsibility. Report of the Secretary-General’s High-level Panel on Threats, Challenges and Change. United Nations Department of Public Information, DPI/2367, December 2004. Available at: http://www.un.org/secureworld/report2.pdf (accessed 17 March 2010). Verón, S.R., Parualo, J.M., Oesterheld, M., 2010. Grazing-induced losses of biodiversity affect the transpiration of an arid ecosystem. Oecologia, 10, http:// dx.doi.org/10.1007/s00442-010-1780-4. Vile, D., Shipley, B., Garnier, E., 2006. Ecosystem productivity can be predicted from potential relative growth rate and species abundance. Ecol. Lett. 9 (9), 1061–1067. Volkova, E.A., Ogar, N.P., Rachkovskaya, E.I., Sadvokasov, P.E., Hramtsov, V.N., 2010. The national atlas of the republic of Kazakhstan, map: vegetation, volume 1. Almaty 2010, 110–113. WRI World Resources Institute, 2008. Ecosystem Services. A Guide for Decision Makers, Available at: http://pdf.wri.org/ecosystem services guide for decisionmakers.pdf (accessed 17 March 2010). Walter, H., Breckle, S.-W., 1989. Ecological Systems of the Geobiosphere, Volume 3, Temperate and Polar Zonobiomes of Northern Eurasia. Springer-Verlag, Berlin, Heidelberg, New York. Wißkirchen, K., Tum, M., Günther, K.P., Niklaus, M., Eisfelder, C., Knorr, W., 2013. Quantifying the carbon uptake by vegetation for Europe on a 1 km2 resolution using a remote sensing driven vegetation model. Geosci. Model Dev. 6, 1623–1640. Wilhelm, W.W., Nelson, C.J., 1978. Growth analysis of tall fescue genotypes differing in yield and leaf photosynthesis. Crop Sci. 18, 951–954. Zheng, Y., Rimmington, G.M., Xie, Z., Zhang, L., An, P., Zhou, G., Li, X., Yu, Y., Chen, L., Shimizu, H., 2008. Responses to air temperature and soil moisture of growth of four dominant species on sand dunes of central Inner Mongolia. J. Plant Res. 121 (5), 473–482.