Elsevier

Science of The Total Environment

Volume 684, 20 September 2019, Pages 96-112
Science of The Total Environment

Mapping precipitation-corrected NDVI trends across Namibia

https://doi.org/10.1016/j.scitotenv.2019.05.158Get rights and content

Highlights

  • The effects of precipitation were removed from a MODIS NDVI time-series.

  • Rate, intensity and spatial extent of trends were computed for each biome.

  • Significant greening covered 27.14% of Namibia.

  • Negative trends occurred mostly in the desert biome, implying land degradation.

  • High resolution imagery could identify negative trends as vegetation clearing.

Abstract

Savannas comprise a major component of the Earth system and contribute ecosystem services and functions essential to human livelihoods. Monitoring spatial and temporal trends in savanna vegetation and understanding change drivers is therefore crucial. Widespread greening has been identified across southern Africa; yet its drivers and manifestations on the ground remain ambiguous. This study removes the effects of precipitation on an NDVI time-series, thereby identifying trends not driven by rainfall. It utilizes the significant correlation between vegetation and precipitation as captured using MODIS and rainfall estimates. A linear regression between variables was used to derive its residual (corrected) time-series, and the rate and spatial extent of trends were evaluated in relation to biomes. A random sample-based qualitative interpretation of high spatial resolution imagery was then used to evaluate the nature of the trend on the ground. 23.25% of the country, including all biomes exhibited positive trends. We propose that greening may be related to a reduction in woody species richness, loss of the large trees and a shift towards drought tolerant shrub species, as has been shown in other sub-Saharan environments. 3.23% of the country exhibited negative trends, which were mostly associated with more humid (forested) regions pointing to deforestation as a cause; these manifested as vegetation clearing, identifiable using high resolution multi-temporal imagery. Greening trends could not be identified using this approach; instead, they point to the occurrence of gradual vegetation change caused by indirect drivers.

Introduction

Land cover change is increasingly impacting Earth system processes, from biodiversity, the carbon and water cycles, to the global surface energy balance and ultimately, climate (Turner et al., 2007; Foley, 2005; Le Quéré et al., 2017; Alkama and Cescatti, 2016; Duveiller et al., 2018). Recent land-use intensification and climate changes have caused widespread vegetation structural and functional shifts in savannas, which are highly sensitive to both precipitation and anthropogenic disturbances (Adeel et al., 2005). Savannas are characterized by the co-occurrence of woody (trees and shrubs) and herbaceous (grasses and forbes) vegetation and provide key ecosystem services and functions locally and globally, from forest and forage resources to the regulation of carbon and hydrological cycles, as well as the Earth's surface energy balance. For instance, savannas in the southern hemisphere are a major carbon sink (Adeel et al., 2005; Poulter et al., 2014; Scheffer et al., 2001; Liu et al., 2015), and African savannas, which cover 36% of the continent, contribute significantly to the carbon cycle (Poulter et al., 2014; Ahlström et al., 2015; Liu et al., 2015), being responsible for 14% of global Net Primary Productivity (NPP) (Grace et al., 2006). Moreover, African savannas are of great agricultural, economic and environmental importance to a large part of the continent's rural population, who comprise a substantial proportion of the continent's total (Jacquin et al., 2010; Fensholt et al., 2013; Pettorelli et al., 2017). Thus, any change in savanna vegetation is likely to have a large impact on social and ecosystem processes. Yet important uncertainties remain concerning the role of African savanna vegetation in these processes, as well as the extent and drivers of its change. For instance, recent studies find widespread greening in dryland areas, the carbon sequestration effects of which are potentially being offset by important deforestation in humid regions (Ciais et al., 2009; Hansen et al., 2013; Brandt et al., 2017; Saha et al., 2015).

Against this background, vegetation change in southern African savannas takes two principal forms. Firstly, the persistent disappearance of all vegetation (i.e. deforestation and desertification), and secondly, a shift towards a woodier condition where herbaceous vegetation layers are replaced with hardy, often unpalatable shrub species (i.e. shrub encroachment) (Hudak and Wessman, 1998; Ward et al., 2000; Erkkilä, 2001; Mendelsohn and el Obeid, 2005; Ward, 2005; Mitchard and Flintrop, 2013; Tian et al., 2016; Wingate et al., 2016). Both these processes are widespread and result in a new equilibrium state with a modification of ecosystem service provision, processes, functions and biodiversity (Anyamba and Tucker, 2005; Olsson et al., 2005; Saha et al., 2015). Although findings are often nuanced for shrub encroachment, it has generally been observed that both processes exacerbate soil erosion, reduce land productivity, animal and plant diversity including valuable forage species (Briggs et al., 2005; Scholes and Archer, 1997).

Change processes are generally attributed either to direct human activities or indirect drivers, such as climate change (Song et al., 2018). For instance, deforestation is most often the result of direct human activities, including urbanization and land-use intensification. Others drivers remain poorly understood and can often be attributed to indirect causes, for example, those responsible for shrub encroachment (Archer et al., 1995; Hudak and Wessman, 1998; Fensham et al., 2005; Ward, 2005; Mitchard and Flintrop, 2013; Caviezel et al., 2014; O'Connor et al., 2014; Adeel et al., 2005). They include, shifts in fire activity (Bond and Midgley, 2000; Bowman et al., 2010), overstocking and the removal of browsers and the herbaceous layer (Asner et al., 2004; Ward, 2005), long-term rainfall changes and climate changes (Fensham et al., 2005), and the interaction and synergy between these (Ward, 2005). Lastly, rising atmospheric carbon dioxide concentrations have been shown to cause a shift away from the predominance of recently evolved C4 grasses, adapted to low CO2 and found extensively in savannas, to the older C3 shrubs and trees (Bond and Midgley, 2000; Donohue et al., 2013).

Satellite remote sensing has proved invaluable for quantifying environmental change and identifying its drivers, over the last five decades. In particular, the available moderate to high resolution sensors, including MODIS, Landsat and Sentinel-2, continuously provides synoptic measurements at spatial and temporal scales which allow the quantitative detection of land cover changes processes (Coppin et al., 2004; Singh, 1989; Drusch et al., 2012). Satellite-derived spectral vegetation “greenness” indices such as the Normalized Difference Vegetation Index (NDVI), are fundamental tools for monitoring trends in global vegetation cover and condition (Myneni et al., 1995; Olsson et al., 2005). In addition, a number of rainfall data sets have been providing consistent daily precipitation estimates over the past three decades, including the Tropical Applications of Meteorology using SATellite data and ground-based observations Research Group (TAMSAT). It is based on Meteosat thermal-infrared (TIR) observations across Africa, and yields estimates comparable to analogous remotely sensed precipitation datasets (Maidment et al., 2014). Studies integrating multi-data sources, for instance, NDVI and rainfall among others, provide key insights into regional vegetation shifts resulting from climatic and human drivers, and are fundamental tools for studying their temporal and spatial relationship (Herrmann et al., 2005; Nicholson et al., 1998). Specifically, they have contributed towards addressing questions pertaining to NPP, carbon sequestration, and ecosystem responses to climatic shifts and anthropogenic disturbances (Tucker et al., 1983; Anyamba and Tucker, 2005; Archibald and Scholes, 2007; Donohue et al., 2009; Andela et al., 2013; Fensholt et al., 2006, Fensholt et al., 2012).

NDVI consists of the spectral reflectances from the normalized ratio of the near-infrared (NIR) and red bands. It is used as a proxy for vegetation “greenness”, cover, biomass and gross primary productivity (Pettorelli et al., 2005). However, several constraints are associated with NDVI for vegetation mapping generally, including: illumination and observation geometry, soil background effects in areas with exposed soil, atmospheric noise, a non-linear response at low and high vegetation density levels. Further issues arise in multi-layer canopies due to the different phenologies of understory vegetation strata, for instance grasses (herbaceous) and canopy trees (woody), and are of particular concern in savannas (Chidumayo, 2001; Soudani et al., 2012). These limitations become apparent when attempting to quantify vegetation changes in savannas, and arise primarily as a result of the spatially and temporally patchy nature of tree, shrub and herbaceous vegetation together with their stratification. Indeed, savanna vegetation composition often ranges from open grassland and shrub land to woodland and forest (Scholes and Archer, 1997; Asner and Lobell, 2000; Herold et al., 2008); moreover, it typically exhibits high inter- and intra-annual variability in response to climate, fire and anthropogenic land-use - herbaceous vegetation in particular. Collectively, these factors render change detection using NDVI in savannas more complex than for forests with closed canopies (Donohue et al., 2009; Zhu et al., 2016). For instance, NDVI measures vegetation “greenness”, but does not distinguish between vegetation functional types; hence, shifts from herbaceous to woody vegetation, such as would be expected from shrub encroachment, may not be directly measured using NDVI. In other words, changes in vegetation composition, where the “greenness” signal characteristics do not change, may be concealed (Brandt et al., 2016; Horion et al., 2014). Thus, a degradation process which is not characterized by decreases, but instead by increases in “greenness”, such as shrub encroachment, may exhibit positive trends in NDVI. Lastly, several methods have been proposed to distinguish between different vegetation functional types and density characteristics, including NDVI signal amplitude (Wagenseil and Samimi, 2006; Olsson et al., 2005).

The availability of water and precipitation, often highly variable both seasonally and inter-annually, are the main controls on vegetation structure, composition and distribution in savannas; thus, precipitation greatly impacts the amount of vegetation present in any given year. In effect, the dynamics of savanna vegetation in relation to precipitation have been monitored at continental scales, and were found to be sensitive to climatic fluctuations within relatively short time scales (i.e. months) (Scanlon et al., 2002). Across most of the African continent, inter-annual fluctuations in rainfall were found to not often trigger much variability in NDVI (Goward and Prince, 1995; Fuller and Prince, 1996). Yet, in some regions a strong relationship between rainfall and NDVI has been identified. In particular, over much of southern Africa and in so called “marginal zones” found between areas of high and low annual rainfall, several authors identify a strong response of NDVI to inter-annual precipitation fluctuations (Goward and Prince, 1995). This response is presumed to result from the relative proportions of woody and herbaceous vegetation in these zones; these areas encompass the bulk of the study area, hence the NDVI signal is expected to show a strong correlation with rainfall (Goward and Prince, 1995; Herrmann et al., 2005).

In the Sahel region in particular, the correlation between precipitation and vegetation has been extensively studied (Nicholson, 2001; Fensholt et al., 2012; Brandt et al., 2015). For example, NDVI was found to be increasing in line with rainfall for part of the Sahel between 1982 and 1999 (Eklundh and Olsson, 2003). Several studies have used models to investigate the effects of precipitation on vegetation proxies (Andela et al., 2013; Tian et al., 2016). For instance, a linear relationship was identified between NDVI and precipitation in the Sahel in areas where rainfall reached 1000 mm per year; this relationship was log-linear for large parts of eastern Africa with similar annual rainfall regimes (Nicholson et al., 1990). Further, the observed relationship was often significant enough to use NDVI as a proxy for variations in precipitation (Tucker and Nicholson, 1999). In Botswana, which has a similar climate and vegetation composition to Namibia, a linear relationship between NDVI and precipitation for regions with <500 mm of rainfall per year was identified. Above this threshold, a “saturation” response could be established with a much reduced rate of NDVI increase (Nicholson and Farrar, 1994). In southern Africa, several studies have focused on the interconnections between NDVI and rainfall estimates, to investigate NPP (Zhu and Southworth, 2013), phenology (Chidumayo, 2001; Azzali and Menenti, 2000), the relationship between NDVI and precipitation (Richard and Poccard, 1998; Chamaille-Jammes et al., 2006; Nicholson and Farrar, 1994), land degradation (Wessels et al., 2007) and vegetation trends in relation to climate and human drivers (Herrmann et al., 2005; Evans and Geerken, 2004).

Building upon these studies, we explore vegetation trends using an approach to remove the precipitation signal from the NDVI time-series (Evans and Geerken, 2004; Herrmann et al., 2005; Wessels et al., 2007; Li et al., 2012; Andela et al., 2013a; Fensholt et al., 2013; Saha et al., 2015). The theoretical basis for this centres on the observed linear relationship between precipitation and NDVI, identified throughout large parts of sub-Saharan Africa with a mean annual precipitation of up to 600 mm. This linear relationship implies that rainfall is the main driver of vegetation photosynthesis and hence NDVI signal (Tucker and Nicholson, 1999). Such an equation computes predicted NDVI from observed precipitation estimates and observed NDVI, and hence the residuals. The approach theoretically permits the precipitation signal to be removed from the NDVI time-series, enabling the main vegetation changes not associated with precipitation to be detected. Using this approach, the following model assumptions are made: if there is a significant anthropogenic effect, as opposed to precipitation effect, on the NDVI signal, it may be identified in the residuals. More specifically, removing the effects of precipitation from NDVI time-series (i.e. corrected NDVI) allows the remaining signal changes to be attributed to anthropogenic disturbances. For instance, areas exhibiting negative trends in corrected NDVI may be undergoing vegetation losses due to anthropogenic disturbances rather than drought. Conversely, areas experiencing positive trends may be undergoing vegetation gains due to factors other than precipitation increases, for example, land management. However, such an approach is dependent on a strong linear relationship between NDVI and precipitation, and is complicated by factors which reduce the strength of this relationship. They include gradients in precipitation and vegetation density, differing responses of eco-floristic regions to rainfall patterns, and distinct ratios of herbaceous and woody vegetation together with their respective phenologies. In particular, these factors are expected to affect the corrected time-series by causing marked and irregular seasonality and thus constituting an important limitation in terms of model accuracy.

Considering the widespread vegetation changes identified across southern Africa, together with the uncertainty surrounding their cause, extent and rate of change, there is a pressing need for quantitative and reproducible assessments of long-term (>15 years) land cover change, as well as an identification of their drivers (Strohbach, 2001). Such an assessment will permit the intensity change processes, including shrub encroachment, land degradation and deforestation to be quantified, and hence serve for developing, implementing and revising sustainable land management, policy, practices and conservation objectives (Wessels et al., 2004; Jacquin et al., 2010). To address these gaps, this study maps trends in corrected NDVI, with the purpose of measuring the extent and rate of vegetation changes un-related to precipitation across Namibia. The main hypothesis is that precipitation-corrected NDVI can be used to map vegetation changes resulting from anthropogenic disturbances. The aims of the study are to: i) quantify the rate, trajectory and spatial extent of trends in corrected NDVI, ii) evaluate these results in relation to the main biomes, and iii) attribute observed change to direct and indirect drivers. These aims are addressed through the following objectives: a) model the relationship between precipitation and NDVI using linear regression; this allows predicted NDVI and hence residual NDVI (corrected NDVI) to be computed; b) compare trends in the corrected NDVI time-series with those of the original or raw (un-corrected) time-series; and finally, c) qualitatively evaluate a random sample of plots exhibiting marked changes with high resolution multi-temporal imagery (Saha et al., 2015).

Section snippets

Study area

We investigate the country of Namibia, which lies in southern Africa, and its constituent Food and Agricultural Organization (FAO) biomes (Fig. 1). The climate ranges from sub-humid to hyper-arid, with rainfall extending from an annual average of 650 mm in the northeast, to 50 mm in the southwest. Precipitation is highly variable over space and time, both intra- and inter-annually, with frequent drought events occurring for any given period. It is concentrated within the five summer months from

Linear regression

The R image derived from the linear regression between mean monthly NDVI and concurrent cumulative monthly TAMSAT precipitation (2000–2017) is plotted in Fig. 3. Rainfall and NDVI show a strong correlation (R > 0.5) across 64.08% of the country (528,903.44 km2). Higher coefficients are evident across the northeast, with western and southern-most regions showing the lowest values. Low R values were found in hyper-arid areas (little or no precipitation or NDVI signal), and in regions where the

Trend extent and intensity

Considering the spatial extent and intensity of the observed trends, important changes in vegetation structure and function are likely to be taking place across the country. These are likely to affect key ecosystem services and functions, which in turn will have important consequences for human livelihoods, biodiversity, the carbon and water cycles, as well as the global surface energy balance; hence, further research into these topics is required.

Correcting for the effects of precipitation

Conclusions

This study presents country-level results of a MODIS precipitation-corrected NDVI trend analyses. It confirms extensive greening in a southern African country previously reported with satellite-derived vegetation proxies. Specifically, significant greening trends cover 23.25% of Namibia, which is greening at an average annual rate of 0.06 over the 16 year study period (equal to a net change of 0.10 NDVI units). Our results stand in contrast to one recent study which finds, on an average, a

References (103)

  • Stefanie M. Herrmann et al.

    Vegetation impoverishment despite greening: a case study from Central Senegal

    J. Arid Environ.

    (2013)
  • Stefanie M. Herrmann et al.

    Recent trends in vegetation dynamics in the African Sahel and their relationship to climate

    Glob. Environ. Chang.

    (2005)
  • Andrew T. Hudak et al.

    Textural analysis of historical aerial photography to characterize woody plant encroachment in South African savanna

    Remote Sens. Environ.

    (1998)
  • Cheikh Mbow et al.

    Can vegetation productivity be derived from greenness in a semi-arid environment? Evidence from ground-based measurements

    J. Arid Environ.

    (2013)
  • S.E. Nicholson et al.

    The influence of soil type on the relationships between NDVI, rainfall, and soil moisture in semiarid Botswana. I. NDVI response to rainfall

    Remote Sens. Environ.

    (1994)
  • L. Olsson et al.

    A recent greening of the Sahel—trends, patterns and potential causes

    J. Arid Environ.

    (2005)
  • Nathalie Pettorelli et al.

    Using the satellite-derived NDVI to assess ecological responses to environmental change

    Trends Ecol. Evol.

    (2005)
  • Richard F. Rohde et al.

    The historical ecology of Namibian rangelands: vegetation change since 1876 in response to local and global drivers

    Sci. Total Environ.

    (2012)
  • Todd M. Scanlon et al.

    Determining land surface fractional cover from NDVI and rainfall time series for a savanna ecosystem

    Remote Sens. Environ.

    (2002)
  • Heike Schmidt et al.

    Remote sensing of the seasonal variability of vegetation in a semi-arid environment

    J. Arid Environ.

    (2000)
  • Kamel Soudani et al.

    Ground-based network of NDVI measurements for tracking temporal dynamics of canopy structure and vegetation phenology in different biomes

    Remote Sens. Environ.

    (2012)
  • C. Tucker et al.

    Satellite remote sensing of total dry matter production in the Senegalese Sahel

    Remote Sens. Environ.

    (1983)
  • David Ward et al.

    Perceptions and realities of land degradation in arid Otjimbingwe, Namibia

    J. Arid Environ.

    (2000)
  • K.J Wessels et al.

    Assessing the Effects of Human-Induced Land Degradation in the Former Homelands of Northern South Africa with a 1 Km AVHRR NDVI Time-Series

    Remote Sens. Environ.

    (2004)
  • K.J. Wessels et al.

    Can human-induced land degradation be distinguished from the effects of rainfall variability? A case study in South Africa

    J. Arid Environ.

    (2007)
  • Zafar Adeel et al.

    Ecosystems and Human Well-Being: Desertification Synthesis

    (2005)
  • Anders Ahlström et al.

    The dominant role of semi-arid ecosystems in the trend and variability of the land CO2 sink

    Science

    (2015)
  • Ramdane Alkama et al.

    Biophysical climate impacts of recent changes in global forest cover

    Science

    (2016)
  • N. Andela et al.

    Global changes in dryland vegetation dynamics (1988–2008) assessed by satellite remote sensing: comparing a new passive microwave vegetation density record with reflective greenness data

    Biogeosciences

    (2013)
  • Steve Archer et al.

    Mechanisms of shrubland expansion: land use, climate or CO2?

    Clim. Chang.

    (1995)
  • S. Archibald et al.

    Leaf green-up in a semi-arid African savanna–separating tree and grass responses to environmental cues

    J. Veg. Sci.

    (2007)
  • Gregory P. Asner et al.

    Grazing systems, ecosystem responses, and global change

    Annu. Rev. Environ. Resour.

    (2004)
  • S. Azzali et al.

    Mapping vegetation-soil-climate complexes in Southern Africa using temporal Fourier analysis of NOAA-AVHRR NDVI data

    Int. J. Remote Sens.

    (2000)
  • Y. de Beer et al.

    Elephants and low rainfall alter woody vegetation in Etosha National Park, Namibia

    J. Arid Environ.

    (2006)
  • William J. Bond et al.

    A proposed CO2-controlled mechanism of woody plant invasion in grasslands and savannas

    Glob. Chang. Biol.

    (2000)
  • William J. Bond et al.

    Carbon dioxide and the uneasy interactions of trees and Savannah grasses

    Phil. Trans. R. Soc. B

    (2012)
  • David M.J.S. Bowman et al.

    Has global environmental change caused monsoon rainforests to expand in the Australian monsoon tropics?

    Landsc. Ecol.

    (2010)
  • Martin Brandt et al.

    Ground- and satellite-based evidence of the biophysical mechanisms behind the greening Sahel

    Glob. Chang. Biol.

    (2015)
  • Martin Brandt et al.

    Human population growth offsets climate-driven increase in woody vegetation in Sub-Saharan Africa

    Nat. Ecol. Evol.

    (2017)
  • Martin Brandt et al.

    Reduction of tree cover in West African woodlands and promotion in semi-arid farmlands

    Nat. Geosci.

    (2018)
  • John M. Briggs et al.

    An ecosystem in transition: causes and consequences of the conversion of Mesic grassland to shrubland

    BioScience

    (2005)
  • M. Broich et al.

    Land surface phenological response to decadal climate variability across Australia using satellite remote sensing

    Biogeosciences

    (2014)
  • R. Buitenwerf et al.

    Increased tree densities in South African savannas: >50 years of data suggests CO2 as a driver

    Glob. Chang. Biol.

    (2012)
  • Chatrina Caviezel et al.

    Soil–vegetation interaction on slopes with bush encroachment in the Central Alps–adapting slope stability measurements to shifting process domains

    Earth Surf. Process. Landf.

    (2014)
  • Simon Chamaille-Jammes et al.

    Spatial patterns of the NDVI–rainfall relationship at the seasonal and interannual time scales in an African savanna

    Int. J. Remote Sens.

    (2006)
  • Peter Y. Chen et al.

    Correlation: Parametric and Nonparametric Measures

    (2002)
  • E.N. Chidumayo

    Climate and phenology of savanna vegetation in Southern Africa

    J. Veg. Sci.

    (2001)
  • P. Ciais et al.

    Variability and recent trends in the African terrestrial carbon balance

    Biogeosciences

    (2009)
  • P. Coppin et al.

    Review article digital change detection methods in ecosystem monitoring: a review

    Int. J. Remote Sens.

    (2004)
  • J.N. De Klerk

    Bush Encroachment in Namibia: Report on Phase 1 of the Bush Encroachment Research, Monitoring, and Management Project

    (2004)
  • Cited by (30)

    • Modelling soil organic carbon using vegetation indices across large catchments in eastern Australia

      2022, Science of the Total Environment
      Citation Excerpt :

      This study examines Vegetation Indices (VIs) derived from satellite-borne sensors to quantify and predict SOC in pasture dominated landscapes. The assumption that links VIs and SOC concentration is that vegetation ecological productivity (which can be quantified by RS) is an important driver of organic matter inputs into the soil (via above ground litter and belowground rhizodeposition) from which SOC concentration is principally derived Cao et al., 2019; Wingate et al., 2019). In turn, SOC concentration levels help determine vegetation growth in the mid- and long-term.

    • Evaluation of spatial-temporal distribution of precipitation in mainland China by statistic and clustering methods

      2021, Atmospheric Research
      Citation Excerpt :

      The amount of precipitation will also affect the spatial distribution of plants (Maneja et al., 2020). Generally, the more precipitation, the denser the vegetation will cover (Wingate et al., 2019). As one of the most common meteorological elements, the characteristics of precipitation have been extensively studied.

    View all citing articles on Scopus
    1

    These authors contributed equally to this work.

    View full text