ArticlePDF Available


Savannas are globally important ecosystems of great significance to human economies. In these biomes, which are characterized by the co-dominance of trees and grasses, woody cover is a chief determinant of ecosystem properties. The availability of resources (water, nutrients) and disturbance regimes (fire, herbivory) are thought to be important in regulating woody cover, but perceptions differ on which of these are the primary drivers of savanna structure. Here we show, using data from 854 sites across Africa, that maximum woody cover in savannas receiving a mean annual precipitation (MAP) of less than approximately 650 mm is constrained by, and increases linearly with, MAP. These arid and semi-arid savannas may be considered 'stable' systems in which water constrains woody cover and permits grasses to coexist, while fire, herbivory and soil properties interact to reduce woody cover below the MAP-controlled upper bound. Above a MAP of approximately 650 mm, savannas are 'unstable' systems in which MAP is sufficient for woody canopy closure, and disturbances (fire, herbivory) are required for the coexistence of trees and grass. These results provide insights into the nature of African savannas and suggest that future changes in precipitation may considerably affect their distribution and dynamics.
© 2005 Nature Publishing Group
Determinants of woody cover in African savannas
Mahesh Sankaran
, Niall P. Hanan
, Robert J. Scholes
, Jayashree Ratnam
, David J. Augustine
, Brian S. Cade
Jacques Gignoux
, Steven I. Higgins
, Xavier Le Roux
, Fulco Ludwig
, Jonas Ardo
, Feetham Banyikwa
Andries Bronn
, Gabriela Bucini
, Kelly K. Caylor
, Michael B. Coughenour
, Alioune Diouf
Wellington Ekaya
, Christie J. Feral
, Edmund C. February
, Peter G. H. Frost
, Pierre Hiernaux
Halszka Hrabar
, Kristine L. Metzger
, Herbert H. T. Prins
, Susan Ringrose
, William Sea
rg Tews
Jeff Worden
& Nick Zambatis
Savannas are globally important ecosystems of great significance
to human economies. In these biomes, which are characterized
by the co-dominance of trees and grasses, woody cover is a chief
determinant of ecosystem properties
. The availability of
resources (water, nutrients) and disturbance regimes (fire, herbi-
vory) are thought to be important in regulating woody cover
but perceptions differ on which of these are the primary drivers of
savanna str ucture. Here we show, using data from 854 sites across
Africa, that maximum woody cover in savannas receiving a mean
annual precipitation (MAP) of less than ,650 mm is constrained
by, and increases linearly with, MAP. These arid and semi-arid
savannas may be considered ‘stable’ systems in which water
constrains woody cover and permits grasses to coexis t, while
fire, herbivory and soil properties interact to reduce woody
cover below the MAP-controlled upper bound. Above a MAP of
,650 mm, savannas are ‘unstable systems in w hich MAP i s
sufficient for woody canopy closure, and disturbances (fire, herbi-
vory) are required for the coexistence of trees and grass. These
results provide insights into the nature of African savannas and
suggest that future changes in precipitation
may considerably
affect their distribution and dynamics.
Savannas occupy a fifth of the earths land surface and support a
large proportion of the world’s human population and most of its
rangeland, livestock and wild herbivore biomass
. A defining feature
of savanna ecosystems is the coexistence of trees and grasses in the
. The balance between these two life forms influences both
plant and livestock production, and has profound impacts on several
aspects of ecosystem function, including carbon, nutrient and
hydrological cycles
. The mechanisms that promote tree–grass
coexistence and the factors that determine the relative proportions of
these two life forms across different savanna types remain, however,
. Because savannas are antici pated to be among the
ecosystems that are most sensitive to future changes in land use
and climate
, a thorough understanding of factors that structure
savanna communities is urgently required to guide management
Explanations for the persistence of tree–grass mixtures in savannas
are varied and invoke such different mechanisms as competition for
water and nutrients
, demographic bottlenecks to tree recruit-
, and disturbances including fire
and large mammal
. Empirical studies provide support both for and
against each alternative mechanism and, consequently, perceptions
differ on the relative impor tance of resource limitation versus
disturbances in controlling savanna structure
. The lack of con-
sensus arises, in part, because most studies have been small-scale and
site-specific, and have often focused on a single determinant
. But
savanna sy stems are diverse and occur under a wide range of
bioclimatic conditions
, and it is likely that the importance of
different processes in regulating woody cover may vary in different
savanna regions. Thus, a comprehensive model that explains both
coexistence and the relative productivity of tree and grass com-
ponents across diverse types of savanna is unlikely to arise from
studying individual systems in isolation: it requires a synthesis of
data from savannas across broad environmental gradients
Here we use a continental scale analysis of African savannas to
investigate h ow the relative importance of resource availability
(water, nutri ents) and disturbance regimes (fire, herbivor y) in
regulating woody cover varies across broad environmental gradients.
In particular, we are interested in determining whether broad-scale
trends in savanna structure are indicative of ‘stable’ or ‘unstable’
, or whether savannas show elements of both across their
geographic range of occurrence. We use ‘stable’ in a limited sense to
mean that coexistence of trees and grasses in savannas is not
dependent on disturbances such as fire and mammalian herbivory,
while recognizing that woody community biomass and cover are
dynamic, not static, properties of the system.
Specifically, we considered that if water availability is the primary
determinant of woody cover in savannas
, then precipitation
Natural Resource Ecology Laboratory, Colorado State University, Fort Collins, Colorado 80523, USA.
Division of Forest Science and Technology, CSIR, PO Box 395, Pretoria 001,
South Africa.
USDA Forest Service, Commanche National Grassland, PO Box 12, Springfield, Colorado 81073, USA.
US Geological Survey, Fort Collins Science Center,
2150 Centre Avenue, Building C, Fort Collins, Colorado 80526-8818, USA.
Ecole Normale Superieure, Laboratoire d’ecologie, UMR 7625 CNRS, Universite
de Paris 6 ENS, 46 Rue
d’Ulm, 75230 Paris Cedex 05, France.
Umweltforschungszentrum Leipzig Halle, Sekt Okosyst Anal, Permoserstrasse 15, D-04318 Leipzig, Germany.
Microbial Ecology
Laboratory, UMR 5557 CNRS, Universite
Lyon 1 USC INRA 1193, Ba
timent G. Mendel, 43 Boulevard du 11 Novembre 1918, 69622 Villeurbanne, France.
CSIRO Centre for
Environment and Life Sciences, CSIRO Plant Industry, Private Bag 5, Wembley, Western Australia 6913, Australia.
Department of Physical Geography and Ecosystems Analysis,
Lund University, So
lvegatan 12, 223 62 Lund, Sweden.
Department of Botany, University of Dar es Salaam, PO Box 35060, Dar es Salaam, Tanzania.
Department of
Agriculture and Game Management, Private Bag X6011, Port Elizabeth Technikon, Port Elizabeth 6000, South Africa.
Department of Civil & Environmental Engineering,
Princeton University, Princeton, New Jersey 08544, USA.
Centre de Suivi Ecologique, BP 15532, Dakar, Senegal.
University of Nairobi, Department of Range Management,
PO Box 29053, Nairobi, Kenya.
Environmental Sciences Department, University of Virginia, PO Box 400123, 291 McCormick Road, Charlottesville, Virginia 22904 -4123, USA.
Department of Botany, University of Cape Town, University Private Bag, Rondebosch 7700, South Africa.
Institute of Environmental Studies, University of Zimbabwe, PO Box
MP 167, Mount Pleasant, Hara re, Zimbabwe.
CESBIO, 18 Avenue E. Belin, 31401 Toulouse Cedex 9, France.
Mammal Research Institute, University of Pretoria, Pretoria 002,
South Africa.
Department of Zoology, University of Wisconsin, 430 Lincoln Drive, Madison, Wisconsin 53706, USA.
Resource Ecology Group, Wageningen University,
Bornsesteeg 69, 6708 PD Wageningen, The Netherlands.
Harry Oppenheimer Okavango Research Centre, University of Botswana, Private Bag 285, Maun, Botswana.
University of Potsdam, Institute of Biochemistry & Biology, Plant Ecology & Nature Conservation, Maulbeerallee 2, D-14469 Potsdam, Germany.
Scientific Services, Kruger
National Park, Private Bag X402, Skukuza 1350, South Africa.
Vol 438|8 December 2005|doi:10.1038/nature04070
© 2005 Nature Publishing Group
should limit the potential tree cover that can be supported at any
given site, and maximum realizable woody cover should gradually
inc rease with MAP
. By contrast, if disturbances such as fire
and herbivory primarily maintain savannas
, then we expect an
abrupt, rather than gradual, increase in maximum realizable woody
cover with increasing MAP
: below a critical threshold of rainfall
sufficient to permit tree growth outside riparian areas or depressions,
grasslands should dominate; above this threshold, the maximum
woody cover should correspond to a closed-canopy woodland state
Depending on the level of disturbance, a particular location might
have reduced woody cover, but the upper bound would not depend
on MAP.
We evaluated relationships between woody cover and MAP, soil
characteristics (texture, percentage nitrogen, nitrogen mineraliza-
tion, total phosphorus) and disturbance regimes (fire-return inter-
vals, mammalian herbivore biomass) from 854 sites across Africa
(Supplementary Fig. S1 and Methods). Woody cover ranges from 0 to
90% across sites and tends to increase with MAP (Fig. 1). More
particularly, within a narrow range of MAP from ,100 to 650 mm,
an upper bound exists on the maximum realizable woody cover
(Fig. 1). In these arid to semi-arid sites (,650 ^ 134 mm MAP;
see Fig. 1), maximum realized woody cover increases with MAP
(Fig. 2a), but shows no relationship w ith fire-return inter vals,
herbivore bio mass or soil characteristics (Fig. 2b–f), su ggesting
that the observed upper limit on woody cover in arid and semi-
arid African savannas is primarily a consequence of m oisture
limitation. The presence of an upper bound on woody cover in
these savannas that is linked primarily to MAP is not consistent w ith
the view that savannas are inherently unstable systems maintained by
Within this MAP range (,650 ^ 134 mm MAP), our analysis
suggests that tre e–grass coexistence is stable to the extent that
disturbances such as fire and herbivory, although capable of modify-
ing tree to grass ratios, are not necessary for coexistence. In these
“climatically determined savannas”
(,650 ^ 134 mm MAP),
restrictions on maximum woody cover as a result of water limitation
permit grasses to persist in the system. By contrast, in areas that
receive a MAP in excess of 650 ^ 134 mm, water availability seems to
be sufficient to allow trees to approach canopy closure such that
grasses may be effectively excluded. These disturbance-driven
represent unstable systems in which disturbances such
as fire, grazing and browsing are required to maintain both trees
and grasses in the system by buffering against transitions to a closed-
canopy state
Whereas MAP drives the upper bound on woody cover in arid and
semi-arid savannas, disturbance regimes and soil characteristics
impose significant controls on savanna structure by influ encing
woody cover below the bound. A regression tree analysis of mean
woody cover for a restricted subset of sites for which all data were
available (Fig. 3 and Methods) further highlights the importance of
MAP as a principal driver of savanna structure and suggests that
MAP also mediates the relative importance of other savanna drivers
such as fire and soil characteristics.
Below a MAP of ,350 mm, woody cover is typically low (Fig. 3). In
these sites, soil properties and disturbances such as fire and herbivory
rarely regulate woody cover. As MAP increases above this threshold,
fire in particular becomes a common factor that reduces woody cover
Figure 1 | Change in woody cover of African savannas as a function of
Maximum tree cover is represented by using a 99th quantile piece-
wise linear reg ression. The regression analysis identifies the breakpoint (the
rainfall at which maximum tree cover is attained) in the interval
650 ^ 134 mm MAP (between 516 and 784 mm; see Methods). Trees are
typically absent below 101 mm MAP. The equation for the line quantifying
the upper bound on tree cover between 101 and 650 mm MAP is
Cover(%) ¼ 0.14(MAP) 2 14.2. Data are from 854 sites across Africa.
Figure 2 | Woody cover as a function of MAP, soil properties and
disturbance regimes in arid and semi-arid savannas.
between woody cover and MAP (a; n ¼ 529), fire-return intervals
(b; n ¼ 302), herbivore biomass (c; n ¼ 145), percentage of clay
(d; n ¼ 234), nitrogen mineralization potential (e; n ¼ 109) and soil total
phosphorus (f; n ¼ 118) for savannas receiving ,650 mm MAP. Unbroken
and broken lines represent the 99th and 90th linear quantiles, respectively.
Maximum woody cover increased with MAP, but showed no consistent
relationship with other vari ables. For MAP, both quantile slopes were
significantly different from zero. For fire-return intervals, herbivore
biomass, clay and nitrogen mineralization rates, neither regression line had
a significant non-zero slope. For total phosphorus, the 90th but not the 99th
quantile slope differed from zero.
NATURE|Vol 438|8 December 2005 LETTERS
© 2005 Nature Publishing Group
below the MAP-controlled upper bound (Fig. 3). Woody cover is
higher, on average, where fires are infrequent (fire-return interval
.10.5 yr). In sites with more frequent fires, woody cover is typically
low, except on very sandy soils (mostly concentrated on the Kalahari
sand sheets), which tend to support higher woody cover (Fig. 3). The
dependence of fire frequency on MAP presumably arises because
increased grass production in mesic sites leads to greater fuel loads
that can support more frequent fires
(Supplementary Fig. S2). Very
high sand content, which correlates with low nutrient availability
(Supplementary Table S1), may promote higher woody cover if the
positive effects of coarse-textured soils, such as lower wilting points
and greater water percolation to soil layers below grass rooting
, override the negative effects associated w ith lower
nutrient availability in these soils
Herbivore effects on woody cover are, however, less apparent.
Although we found a tendency for grazers to enhance woody cover
and browsers and mixed feeders to depress it, such effects were weak
and could not be generalized beyond our data set (see Methods;
measures of herbivore biomass were retained in the complete, but not
pruned, regression tree). The lack of consistent herbivore effects
across sites most probably reflects differences in herbivore guilds,
seasonality of herbivory, and variation in herbivore body-size distri-
butions across sites, features for which data were not available.
Larger, more detailed data sets will undoubtedly provide greater
resolution of how different driver variables interact to influence
mean woody cover.
These results have the power to inform savanna management
strategies because they bear directly on our ability to predict savanna
responses to changing environmental drivers. In particular, our data
indicate that woody encroachment, a phenomenon in which many
savannas across the world show a directional trend of increasing
woody cover
, may be a bounded process in savannas receiving a
MAP of ,650 ^ 134 mm, ultimately limited by water availability.
For sites close to or at the MAP-controlled bound (Fig. 1), changes in
precipitation regimes that lead to increased water availability
fore may be a cause for concern with respect to woody encroachment.
However, the enormous variation in woody cover, with most sites far
from the climatic bound (Fig. 1), suggests that processes other than
MAP regulate actual tree cover in many savannas of Africa. In
particular, our results suggest that if disturbances by fire, browsers
and humans were absent, then large sections of the African continent
would switch to a wooded state (hatched regions in Fig. 4).
The patterns described here for African savannas suggest that the
dominant ecological theories for tree–grass coexistence in these
systems need to be combined: it is clear that most savannas are
strongly affected by disturbances that maintain woody cover well
below the resource-limited upper bound. Disturbance-based models
do not consider and are unable to explain, however, the upper bound
to tree cover. The result s emerging from this continental scale
analysis strongly indicate that water limits the maximum cover of
woody species in many African savanna systems, but that disturbance
dynamics control savanna structure below the maximum. These
results have important implications both for our understanding of
the fundamental nature of African savanna systems and for our
ability to predict their responses to changing environmental drivers.
It remains to be established whether the patterns obser ved here for
African savannas also hold in other tropical savanna regions or in
temperate savannas where the effects of winter precipitation and
temperature on moisture distribution through the soil profile can
markedly alter water par titioning between woody and herbaceous
plants, and thus can influence maximum woody cover.
Data collection. Data on projected woody cover (the percentage of ground
surface covered when crowns are projected vertically), MAP, soil characteristics
(texture, total nitrogen and phosphorus, and nitrogen mineralization), fire and
herbivory regimes were gathered from several sources for a range of sites across
Africa. We included only sites for which vegetation was sampled over sufficiently
large spatial scales (.0.25 ha for plot measurements and .100 m for transect
sampling). Sites located in riparian or seasonally flooded areas, or in net water
run-on areas such as depressions, and sites in which trees were known to access
ground water resources (that is, sources of water not dependent on rainfall in the
immediate vicinity or in recent years) were excluded from the analysis because
MAP is not a relevant descriptor of water availability in these sites. We also
excluded sites that had been cultivated or harvested by humans ,10 yr before
sampling from the analysis.
Rainfall data included estimates from eld measurements and regional
rainfall maps (n ¼ 469) and from fitted climatic g rids (0.058 resolution,
Figure 4 | The distributions of MAP-determined (‘stable’) and disturbance-
determined (‘unstable’) savannas in Africa.
Grey areas represent the
existing distribution of savannas in Africa according to ref. 30. Vertically
hatched areas show the unstable savannas (.784 mm MAP); cross-hatched
areas show the transition between stable and unstable savannas (516–
784 mm MAP); grey areas that are not hatched show the stable savannas
(,516 mm MAP).
Figure 3 | Regression tree showing generalized relationships between
woody cover and MAP, fire-return interval and percentage of sand.
tree is pruned to four terminal nodes and is based on 161 sites for which all
data were available. No consistent herbivore effects were detected. Branches
are labelled with criteria used to segregate data. Values in terminal nodes
represent mean woody cover of sites grouped within the cluster. The pruned
tree explained ,45.2% of the variance in woody cover, which is significantly
more than a random tree (P , 0.001). Of this, 31% was accounted for by the
first split; the second split explained an additional 10% of the variance in
woody cover.
LETTERS NATURE|Vol 438|8 December 2005
© 2005 Nature Publishing Group
n ¼ 383) of monthly mean rainfall for Africa from the ANU-CRES (ref. 20; and Fire-return periods were obtained
from field records (n ¼ 182) and from burnt-area maps of Africa at 5-km
resolution (n ¼ 670) derived from AVHRR (advanced very high resolution
radiometer) images based on 8 yr of data (1981–1983 and 1985–1991; ref. 21).
Herbivore density estimates were available for 180 sites. Soils were obtained from
166 sites and analysed under standardized laboratory conditions for texture,
total nitrogen and phosphorus, and nitrogen mineralization potential (s ee
Supplementary Information). Our data set included sites encompassing a
wide range of rainfall (132–1, 185 mm MAP), fire-return intervals (1 to
.50 yr), herbivore biomass (0–8,000 kg km
), soil texture (sand, 6.7–98%;
clay, 0.6–62.8%), soil percentage nitrogen (0.013–0.31%), soil total phosphorus
) and potential nitrogen mineralization rates (222.8 to
per week; see Supplementary Fig. S3).
Data analyses. To characterize the effects of MAP on the upper limit to woody
cover across sites, we analysed data using a bent-cable form of a piece-wise linear
estimated with nonlinear quantile regression
, as implemented in the
‘quantreg’ librar y in the statistical package R ( We
used 0.90 to 0.99 conditional quantiles to obtain estimates near the upper
boundary of the percentage of woody cover as it changes with MAP, which better
reflects the process of MAP limiting maximum woody cover than does mean
(see Supplementary Information for details of this and additional
analyses). We conducted additional analyses on the subset of sites that received
,650 mm rainfall annually to investigate further how fire regimes, herbivory
and soil properties influenced the upper bound on woody cover that was evident
in these sites. We analysed these data by linear quantile regression
implemented in the ‘quantreg’ library, which permits computation of confidence
intervals for estimated parameters
and enabled us to test whether the regression
slopes were different from zero.
In addition to analysing patterns in maximum woody cover, we use d
regression tree analysis
, as implemented in the ‘rpart’ library in R, to determine
how resource availability and disturbance regimes influenced mean realized
woody cover in sites (see Supplementary Information). After tree construction,
cross-validation procedures were used to prune trees to a size that best
represented relationships that could be generalized outside the sample to the
rest of the continent
. Woody cover values were log-transformed to stabilize
. To avoid problems arising from collinearity among soil variables,
only sand content was retained for the analysis as it was the variable that was
most strongly correlated to other soil variables (Supplementary Table S1). The
results of the analysis were unchanged if grazer biomass and mixed feeder plus
browser biomass were retained as two separate variables, or if total herbivore
biomass was used as the predictor variable. The analysis was based on 161 sites
for which data on MAP, fire-return intervals, herbivore biomass density and soil
sand content were available. To determine whether the pruned tree explained
more variance than a random tree of equal complexity, the square of the
correlation coefficient (r
) of the pruned tree was compared with r
values of
similar sized trees generated from 2,000 random associations between predictor
variables and woody cover
. Further details on the methodology and results
from additional analyses are provided in the Supplementary Information.
Received 26 April; accepted 22 July 2005.
1. Scholes, R. J. & Archer, S. R. Tree–-grass interactions in savannas. Annu. Rev.
Ecol. Syst. 28, 517–-544 (1997).
2. House, J. et al. Conundrums in mixed woody–-herbaceous plant systems.
J. Biogeog. 30, 1763–-1777 (2003).
3. Jackson, R. B., Banner, J. L., Jobaggy, E. G., Pockman, W. T. & Wall, D. H.
Ecosystem carbon loss with woody plant invasion of grasslands. Nature 418,
623–-626 (2002).
4. Sankaran, M., Ratnam, J. & Hanan, N. P. Tree grass coexistence in savannas
insights from an examination of assumptions and mechanisms
invoked in existing models. Ecol. Letters 7, 480–-490 (2004).
5. Jeltsch, F., Weber, G. E. & Grimm, V. Ecological buffering mechanisms in
savannas: a unifying theory of long-term tree–-grass coexistence. Plant Ecol.
150, 161–-171 (2000).
6. Intergovernmental Panel on Climate Change. Climate Change 2001: Synthesis
Report. A Contribution of Working Groups I, II, and III to the Third Assessment
Report of the Intergovernmental Panel on Climate Change (eds Watson, R. T.,
Core Writing Team) (Cambridge Univ. Press, Cambridge, UK, 2001).
7. Simioni, G., Gignoux, J. & Le Roux, X. How does the spatial structure of the tree
layer influence water balance and primary production in savannas? Results of a
3D modeling approach. Ecology 84, 1879–-1894 (2003).
8. Sala, O. E. et al. Biodiversity
global biodiversity scenarios for the year 2100.
Science 287, 1770–-1774 (2000).
9. Bond, W. J., Midgley, G. F. & Woodward, W. I. The importance of low
atmospheric CO
and fire in promoting the spread of grasslands and savannas.
Glob. Change Biol. 9, 973–-982 (2003).
10. Weltzin, J. F. & McPherson, G. R. (eds) Changing Precipitation Regimes and
Terrestrial Ecosystems: A North American Perspective (Univ. Arizona Press,
Tucson, 2003).
11. Walter, H. Ecology of Tropical and Subtropical Vegetation (Oliver and Boyd,
Edinburgh, 1971).
12. Walker, B. H. & Noy-Meir, I. in Ecology of Tropical Savannas (eds Huntley, B. J.
& Walker, B. H.) 556–-590 (Springer, Berlin, 1982).
13. Walker, B. H., Ludwig, D., Holling, C. S. & Peterman, R. M. Stability of semi-arid
savanna grazing systems. J. Ecol. 69, 473–-498 (1981).
14. Higgins, S. I., Bond, W. J. & Trollope, W. S. W. Fire, resprouting and variability:
a recipe for grass–-tree coexistence in savanna. J. Ecol. 88, 213–-229 (2000).
15. Frost, P. G. et al. (eds) Response of Savannas to Stress and Disturbance (IUBS,
Paris, 1986).
16. van Langevelde, F. et al. Effects of fire and herbivory on the stability of savanna
ecosystems. Ecology 84, 337–-350 (2003).
17. Bond, W. J., Midgley, G. F. & Woodwar d, F. I. What controls South African
climate or fire? S. Afr. J. Bot. 69, 79–-91 (2003).
18. McNaughton, S. J. The propagation of disturbance in savannas through food
webs. J. Veg. Sci. 3, 301–-314 (1992).
19. Walker, B. H. & Langridge, J. L. Predicting savanna vegetation structure on the
basis of plant available moisture (PAM) and plant available nutrients (PAN): a
case study from Australia. J. Biogeog. 24, 813–-825 (1997).
20. Hutchinson, M. F., Nix, H. A., McMahon, J. P. & Ord, K. D. Proceedings of the
Third International Conference/Workshop on Integrating GIS and Environmental
Modeling (National Center for Geographic Information and Analysis, Santa
Barbara, 1996).
21. Barbosa, P. M., Stroppiana, D., Gregoire, J. M. & Pereira, J. M. C. An
assessment of vegetation fire in Africa (1981–-1991): burned areas, burned
biomass, and atmospheric emissions. Glob. Biogeochem. Cycles 13, 933–-950
22. Chiu, G. S. Bent-cable Regression for Assessing Abruptness of Change. Thesis,
Simon Fraser Univ. (2002).
23. Koenker, R. & Park, B. J. An interior point algorithm for nonlinear quantile
regression. J. Econometrics 71, 265–-283 (1994).
24. Cade, B. S. & Noon, B. R. A gentle introduction to quantile regression for
ecologists. Front. Ecol. Environ. 1, 412–-420 (2003).
25. Koenker, R. W. & d’Orey, V. Computing regression quantiles. Appl. Stat. 36,
383–-393 (1987).
26. Koenker, R. W. in Asymptotic Statistics (eds Mandl, P. & Huskova, M.) 349–-359
(Springer, New York, 1994).
27. Breiman, L., Friedman, J. H., Olshen, R. A. & Stone, C. G. Classification and
Regression Trees (Wadsworth International Group, Belmont, 1984).
28. De’ath, G. & Fabricus, K. E. Classification and regression trees: a powerful yet
simple technique for ecological data analysis. Ecology 81, 3178–-3192 (2000).
29. Rejwan, C., Collins, N. C., Brunner, L. J., Shuter, B. J. & Ridgway, M. S. Tree
regression analysis on the nesting habitat of smallmouth bass. Ecology 80,
341–-348 (1999).
30. White, F. The Vegetation of Africa: A Descriptive Memoir to Accompany the
UNESCO/AETFAT/UNSO Vegetation Map of Africa (UNESCO, Paris, 1983).
Supplementary Information is linked to the online version of the paper at
Acknowledgements This paper arose from a workshop on savanna complexity
funded by the NSF. We thank R. Boone, I. McHugh, R. Grant, H. Biggs,
W. T. Starmer, P. M. Barbosa, D. Ruess, J. Rettenmayer, C. Williams, J. Klein,
M. T. Anderson, W. J. Parton, J. C. Neff, N. Govender and the Kruger Park
Scientific Services for comments, help with data collection and analysis, and for
providing access to otherwise unpublished data.
Author Contributions All authors contributed data or intellectual input to the
Author Information Reprints and permissions information is available at The authors declare no competing
financial interests. Correspondence and requests for materials should be
addressed to M.S. (
NATURE|Vol 438|8 December 2005 LETTERS
... Les savanes peuvent être classées en fonction du pourcentage de la composition arbreherbe allant des savanes herbeuses aux savanes arborées ( Figure 1). Du fait de l'importance des savanes mentionnée précédemment, de nombreuses études ont été menées sur leur fonctionnement (Ekblom & Gillson, 2010), leur dynamique (Gignoux et al., 2017;N'Dri et al., 2014) et leur stabilité (Sankaran et al., 2005). ...
... La disponibilité de l'eau est considérée comme un facteur essentiel de la structure et du fonctionnement des savanes (Archibald & Bond, 2003;Breshears & Barnes, 1999;Fensham et al., 2005;Sankaran et al., 2005). En effet, c'est elle qui fixe les limites à la quantité de couverture ligneuse qui peut être supportée sur un site. ...
... En effet, c'est elle qui fixe les limites à la quantité de couverture ligneuse qui peut être supportée sur un site. Selon Sankaran et al., (2005) (Figure 2), en dessous de 650 mm de pluie par an, la proportion maximale de ligneux augmentent linéairement avec les précipitations. Au-dessus de ce seuil, les arbres peuvent atteindre une densité maximale élevée (80 % de couvert, équivalent à un couvert forestier) indépendante de la pluviosité, mais d'autres facteurs peuvent empêcher d'atteindre cette densité. ...
Les savanes occupent 12% de la couverture terrestre. Le feu y est le principal facteur perturbant le cycle de vie des végétaux. Après le feu, la plante alloue sa biomasse carbonée disponible dans ses racines à toutes ses parties pour sa régénération. Des modèles ont été construit sur les rejets en savane sans prendre en compte formellement la dynamique de l’allocation des réserves permettant à la plante de survivre au feu. Le but de ma thèse est d’étudier par modélisation les stratégies d’allocation permettant aux arbres en savane de repousser après le feu. J’ai proposé un modèle général d’allocation des ressources carbonées lors de la régénération des arbres après le passage du feu. Il s’agit d’un modèle avec un rejet avec une ou plusieurs tiges. Les stratégies qui permettent aux plantes de survivre au feu (grandir au-delà de la hauteur des flammes) sont celles où l’allocation de biomasse à la partie aérienne est supérieure à 60 %. L’architecture de la plante a une influence sur ces stratégies d’allocation. Concernant les stratégies d’allocation, l’action du feu apparait comme une saison sèche renforcée, comparée à la saison des pluies. Le nombre de tiges change l’allocation feuilles/tiges et permet de produire plus de biomasse foliaire par rapport à la biomasse de tige. Couplé à un modèle démographique, ces modèles vont nous permettre de prendre en compte le rôle de la stratégie d’allocation de biomasse des plantes, dans les zones soumises au feu où les pressions anthropiques et les changements globaux menacent la survie de plusieurs espèces.
... Vegetation cover is often maintained by a complex interplay of fire, herbivores and dispersal [49][50][51]. In this study, we aimed to construct a state diagram-as it is ideal for understanding the stability structures of ecosystems. ...
... Therefore, a further decrease of MAP (less than 1000 mm), an increase in the incidence of forest fires, and deforestation can lead to shifts in vegetation (such as forest to savannah) or change in the plant community structure. Such changes have indeed been observed in other tropical regions, for instance, in the savannah of Africa [49], and Amazonian rain forests [65]. Therefore, further studies exploring the impact of fires, changing rainfall patterns and deforestation on the NEI forests are needed. ...
Full-text available
Globally, forests and savannah are shown to be alternative stable states for intermediate rainfall regimes. This has implications for how these ecosystems respond to changing rainfall conditions. However, we know little about the occurrence of alternative stable states in forest ecosystems in India. In this study, we investigate the possibility of alternative stable states in the vegetation cover of northeastern India, which is a part of the Eastern Himalaya and the Indo-Burma biodiversity hotspots. To do so, we construct the so-called state diagram, by plotting frequency distributions of vegetation cover as a function of mean annual precipitation (MAP). We use remotely sensed satellite data of the enhanced vegetation index (EVI) as a proxy for vegetation cover (at 1 km resolution). We find that EVI exhibits unimodal distribution across a wide range of MAP. Specifically, EVI increases monotonically in the range 1000–2000 mm of MAP, after which it plateaus. This range of MAP corresponds to the vegetation transitional zone (1200–3700 m), whereas MAP greater than 2000 mm covers the larger extent of the tropical forest (less than or equal to 1200 m) of northeast India. In other words, we find no evidence for alternative stable states in vegetation cover or forest states at coarser scales in northeast India.
... Shrub encroachment appears to result from a number of multiple interacting factors such as overgrazing, fire suppression and climate change (Archer et al., 2017;Dechoum et al., 2018;Devine et al., 2017). Given that water availability is an important factor controlling plant growth in arid and semi-arid grasslands (Knapp et al., 2002;Sankaran et al., 2005), it is assumed that precipitation changes would have a critical impact on shrub encroachment under future climate change (Sankaran et al., 2005;Staver et al., 2011). ...
... Shrub encroachment appears to result from a number of multiple interacting factors such as overgrazing, fire suppression and climate change (Archer et al., 2017;Dechoum et al., 2018;Devine et al., 2017). Given that water availability is an important factor controlling plant growth in arid and semi-arid grasslands (Knapp et al., 2002;Sankaran et al., 2005), it is assumed that precipitation changes would have a critical impact on shrub encroachment under future climate change (Sankaran et al., 2005;Staver et al., 2011). ...
Full-text available
Widespread shrub encroachment is profoundly impacting the structures and functions of global drylands, and precipitation change is assumed to be one of the most critical factors affecting this phenomenon. However, there is little evidence to show how precipitation changes will affect the process. In this study, we conducted a 6‐year precipitation manipulation experiment (‐30%, ambient, +30%, and +50%) to investigate the effects of precipitation changes on the growth of shrubs and herbaceous plants in a shrub‐encroached grassland in Inner Mongolia. We found that the increasing precipitation significantly increased the mean height, coverage, and aboveground biomass of herbaceous species, while the growth of shrub species did not exhibit a significant response to precipitation changes. With increasing precipitation, the relative coverage of shrubs decreased, while that of herbs increased. The native dominant herbaceous plant (Leymus chinensis) with more sensitive maximum photosynthetic rate to the precipitation change, showed higher photosynthetic nitrogen use efficiency and water use efficiency than those of the encroached shrub species (Caragana microphylla) at high soil moisture contents, reflecting that the ecophysiological characteristics of L. chinensis might provide it a competitive advantage under increased precipitation. Our findings suggest that increasing precipitation may slow down shrub encroachment by facilitating herbaceous growth in Mongolian grasslands, and consequently affect the forage value and carbon budget in these ecosystems.
... Woody encroachment into grasslands can potentially be reversed by a combination of management (frequent fires) and climatic events (drought; Roques et al., 2001). In these areas using fire as a management strategy can decrease shrub and invasive species, and has been successfully employed throughout the continent (Sankaran et al., 2005;Venter et al., 2018). Additionally, reducing grazing pressure by decreasing livestock numbers can positively affect grassland areas (Archer et al., 2017). ...
Full-text available
The Ngorongoro Conservation Area (NCA) of Tanzania, is globally significant for biodiversity conservation due to the presence of iconic fauna, and, since 1959 has been managed as a unique multiple land‐use areas to mutually benefit wildlife and indigenous residents. Understating vegetation dynamics and ongoing land cover change processes in protected areas is important to protect biodiversity and ensure sustainable development. However, land cover changes in savannahs are especially difficult, as changes are often long‐term and subtle. Here, we demonstrate a Landsat‐based monitoring strategy incorporating (i) regression‐based unmixing for the accurate mapping of the fraction of the different land cover types, and (ii) a combination of linear regression and the BFAST trend break analysis technique for mapping and quantifying land cover changes. Using Google Earth Pro and the EnMap‐Box software, the fractional cover of the main land cover types of the NCA were accurately mapped for the first time, namely bareland, bushland, cropland, forest, grassland, montane heath, shrubland, water and woodland. Our results show that the main changes occurring in the NCA are the degradation of upland forests into bushland: we exemplify this with a case study in the Lerai Forest; and found declines in grassland and co‐incident increases in shrubland in the Serengeti Plains, suggesting woody encroachment. These changes threaten the wellbeing of livestock, the livelihoods of resident pastoralists and of the wildlife dependent on these grazing areas. Some of the land cover changes may be occurring naturally and caused by herbivory, rainfall patterns and vegetation succession, but many are linked to human activity, specifically, management policies, tourism development and the increase in human population and livestock. Our study provides for the first time much needed and highly accurate information on long‐term land cover changes in the NCA that can support the sustainable management and conservation of this unique UNESCO World Heritage Site. The Ngorongoro Conservation Area (NCA), a UNESCO World Heritage Site, is globally important for biodiversity conservation due to the presence of iconic megafauna. For decades now, the NCA is experiencing a number of notoriously difficult to address challenges; understanding its land cover dynamics is therefore increasingly important to improve habitat monitoring, preserve biodiversity and ensure sustainable development. We used multi‐temporal Landsat data spanning 35 years and a combination of regression‐based unmixing, linear regression and a trend break analysis to map and quantify the land cover dynamics in the area. We found a decrease in forest and grassland cover as well as a significant amount of woody encroachment which is often linked to land degradation in African savannahs. These changes are consistent with other savannah ecosystems and pose a threat to the wellbeing of livestock, the livelihoods of the pastoralist communities, and the wildlife of the NCA.
... Derner et al. (2005) suggested that under high-pCO 2 conditions, woody plants have higher net photosynthetic efficiency than herbaceous plants, resulting in a higher growth rate of woody plants than herbaceous plants, which would promote the process of grassland scrubbing. High precipitation also leads to an increase in woody plant coverage (Sankaran et al., 2005) and enhanced carbon sequestration capacity. In the context of global warming, the middle and high latitudes of terrestrial systems become wet, precipitation occurs (Yarincik, 2000), and the burial of silicates during the cooling process leads to the transfer of carbon from surface systems to sediments (Shields and Mills, 2021), which in turn neutralises the high concentrations of CO 2 in the atmosphere. ...
Full-text available
Eocene hyperthermal events are thought to be closely linked to rapid increases in atmospheric CO2. It is important to better understand the atmospheric triggers of these extreme events; however, reconstruction of pCO2 from sedimentary records remains challenging. Eocene hyperthermal events (Eocene thermal maximum, ETM; Early Eocene Climatic Optimum, EECO; Mid-Eocene Climatic Optimum, MECO) are well documented in the Fushun Basin of northeast China, but the source of organic matter (OM) used as a pCO2 archive in previous studies includes land plants, algae, mixed land and aquatic organisms in the Yprian, Lutetian and Bartonian intervals, respectively. In this paper, we calculate pCO2 concentrations using δ¹³CTOC and δ¹³C27–31 and discuss the factors influencing discrepancies based on various calculations. Results show that both δ¹³CTOC and δ¹³C27–31 from single OM sources are reliable proxies for pCO2. During the ETM2, ETM3, EECO and MECO intervals, pCO2 was 638–2277 ppm, and palaeotemperature was approximately 2.6–8.9 °C higher than present. The background and peak values of pCO2 during hyperthermal events were compared with those related to methane hydrate, OM oxidation and volcanic sources of pCO2 carbon. The comparison revealed that pCO2 values associated with the ETM2, ETM3, EECO and MECO events were more consistent with those calculated by OM oxidation and locally overlapped with volcanic sources. These results indicate that these pCO2 estimates of hyperthermal events are predominantly derived from terrestrial OM oxidation and were influenced by volcanism.
Savannas cover a wide climatic gradient across large portions of the Earth’s land surface and are an important component of the terrestrial biosphere. Savannas have been undergoing changes that alter the composition and structure of their vegetation such as the encroachment of woody vegetation and increasing land-use intensity. Monitoring the spatial and temporal dynamics of savanna ecosystem structure (e.g., partitioning woody and herbaceous vegetation) and function (e.g., aboveground biomass) is of high importance. Major challenges include misclassification of savannas as forests at the mesic end of their range, disentangling the contribution of woody and herbaceous vegetation to aboveground biomass, and quantifying and mapping fuel loads. Here, we review current (2010–present) research in the application of satellite remote sensing in savannas at regional and global scales. We identify emerging opportunities in satellite remote sensing that can help overcome existing challenges. We provide recommendations on how these opportunities can be leveraged, specifically (1) the development of a conceptual framework that leads to a consistent definition of savannas in remote sensing; (2) improving mapping of savannas to include ecologically relevant information such as soil properties and fire activity; (3) exploiting high-resolution imagery provided by nanosatellites to better understand the role of landscape structure in ecosystem functioning; and (4) using novel approaches from artificial intelligence and machine learning in combination with multisource satellite observations, e.g., multi-/hyperspectral, synthetic aperture radar (SAR), and light detection and ranging (lidar), and data on plant traits to infer potentially new relationships between biotic and abiotic components of savannas that can be either proven or disproven with targeted field experiments.
North Africa features some of the most frequently burnt biomes on Earth, including the semi-arid grasslands of the Sahel and wetter savannas immediately to the south. Natural fires are fuelled by rapid biomass production during the wet season, its desiccation during the dry season and ignition by frequent dry lightning strikes. Today, fire activity decreases markedly both to the north of the Sahel, where rainfall is extremely low, almost eliminating biomass over the Sahara, and to the south where forest biomes are too wet to burn. Over the last glacial cycle, rainfall and vegetation cover over northern Africa varied dramatically in response to gradual astronomically-forced insolation change, changes in atmospheric carbon dioxide levels, and abrupt cooling events over the North Atlantic Ocean associated with the reorganisation of Meridional Overturning Circulation (MOC). Here we report the results of a study into the impact of these climate changes on fire activity in northern African over the last 50,000 years (50 kyr). Our reconstructions come from marine sediments with strong age control that provide an uninterrupted record of charcoal particles exported from the African continent. We studied three sites on a latitudinal transect along the northwest African margin between 21 and 9°N. Our sites exhibit a distinct latitudinal relationship between past changes in rainfall and fire activity. At the southernmost site (GeoB9528-3, 9°N), fire activity decreased during intervals of increasing humidity, while our northernmost site (ODP Site 658, 21°N) clearly demonstrates the opposite relationship. The site in the middle of our transect, offshore of the present day southern Sahel today (GeoB9508-5, 15°N), exhibits a “Goldilocks” relationship between fire activity and hydroclimate, wherein charcoal fluxes peak under intermediate rainfall climate conditions and are supressed by transition to more arid or more humid conditions. Our results are remarkably consistent with the predictions of the intermediate fire-productivity hypothesis developed in conceptual macroecological models and supported by empirical evidence of modern day fire activity. Feedback processes operating between fire, climate and vegetation are undoubtedly complex but temperature is suggested to be the main driver of temporal change in fire activity globally, with the precipitation-evaporation balance perhaps a secondary influence in the Holocene tropics. However, there is only sparse coverage of Africa in the composite records upon which those interpretations are based. We conclude that hydroclimate (not temperature) exerted the dominant control on burning in the tropics of northern Africa well before the Holocene (from at least 50 ka).
Forest–savanna mosaics exist across all major tropical regions. Yet, the influence of environmental factors on the distribution of these mosaics is not well explored, limiting our understanding of the environmental constraints on savannas especially in Southeast Asia, where most savannas exist in mosaics. Despite clear structural and functional characteristics indicative of savannas, most SE Asian savannas continue to be classified as forest. This designation is problematic because SE Asian savannas are threatened by both fragmentation and forest‐centric management practices. By studying forest–savanna mosaics across SE Asia, we aimed to parse out how landscape mosaics of forest and savanna may be constrained by fire, climate and soil characteristics. We used remotely sensed data to characterize the distribution of tree cover and forest–savanna mosaics. Using regression models, we quantified the relative effects of precipitation, fire frequency, seasonality and soil characteristics on average tree cover and landscape patchiness. We found that low tree cover, indicative of savannas, occurs in drier, seasonal subregions that experience frequent fire. Further, our results demonstrate that fire and precipitation strongly shape landscape patchiness. Landscapes were patchiest in subregions with low precipitation and intermediate fire frequency. These results demonstrate that the environmental factors important in delineating the distribution of savannas globally shape the distribution of tree cover and landscape patchiness across SE Asia. Fire especially drives patterns of tree cover across scales. In a region where fire suppression is a common management strategy, our results suggest that further research studying vegetation response to fire and fire suppression is needed to improve management and conservation of these mosaic landscapes. More broadly, this work demonstrates a useful approach for studying the environmental drivers that influence the distribution of forest–savanna mosaics.
We co-designed an agent-based model of an Afroalpine grassland in Ethiopia that is experiencing unwanted shrub encroachment. The goal was to enable managers of a community conservation area to better understand the drivers of shrub encroachment and to test possible management actions for controlling shrubs. Due to limited site-specific data, we parameterized this model using insights from published literature, remote sensing, and expert opinion from scientists and local managers. We therefore sought to explore potential future scenarios rather than make highly accurate predictions, focusing on facilitating discussions and learning among the diverse co-management team. We evaluated three social-ecological scenarios with our model, examining: (1) the impact of changing precipitation regimes on vegetation, (2) whether changing the frequency of guassa grass harvests would improve the long-term sustainability of the grassland, and (3) whether the combination of grass harvest and shrub removal would affect shrub encroachment. We found that the model was highly sensitive to the amount of grass harvested each year for local use. Our results indicate that the guassa grass was more resilient than shrubs during persistent dry climatic conditions, whereas a reduction in only the early spring rains (known as the “belg”) resulted in considerable loss of grass biomass. While our modeling results lacked the quantitative specificity desired by managers, participants in the collaborative modeling process learned new approaches to planning and management of the conservation area and expanded their knowledge of the ecological complexity of the system. Several participants used the model as a boundary object, interpreting it in ways that reinforced their cultural values and goals for the conservation area. Our work highlights the lack of detailed scientific knowledge of Afroalpine ecosystems, and urges managers to reconnect with traditional ecological management of the conservation area in their pursuit of shrub encroachment solutions. The decline or absence of the belg rains is becoming increasingly common in the Ethiopian highlands, and our results underscore the need for more widespread understanding of how this changing climatic regime impacts local environmental management. This work lays a foundation for social-ecological research to improve both understanding and management of these highly threatened ecosystems.
Full-text available
Classification and regression trees are ideally suited for the analysis of complex ecological data. For such data, we require flexible and robust analytical methods, which can deal with nonlinear relationships, high-order interactions, and missing values. Despite such difficulties, the methods should be simple to understand and give easily interpretable results. Trees explain variation of a single response variable by repeatedly splitting the data into more homogeneous groups, using combinations of explanatory variables that may be categorical and/or numeric. Each group is characterized by a typical value of the response variable, the number of observations in the group, and the values of the explanatory variables that define it. The tree is represented graphically, and this aids exploration and understanding. Trees can be used for interactive exploration and for description and prediction of patterns and processes. Advantages of trees include: (1) the flexibility to handle a broad range of response types, including numeric, categorical, ratings, and survival data; (2) invariance to monotonic transformations of the explanatory variables; (3) ease and robustness of construction; (4) case of interpretation; and (5) the ability to handle missing values in both response and explanatory variables. Thus, trees complement or represent an alternative to many traditional statistical techniques, including multiple regression, analysis of variance, logistic regression, log-linear models, linear discriminant analysis, and survival models. We use classification and regression trees to analyze survey data from the Australian central Great Barrier Reef, comprising abundances of soft coral taxa (Cnidaria: Octocorallia) and physical and spatial environmental information. Regression tree analyses showed that dense aggregations, typically formed by three taxa, were restricted to distinct habitat types, each of which was defined by combinations of 3-4 environmental variables. The habitat definitions were consistent with known experimental findings on the nutrition of these taxa. When used separately, physical and spatial variables were similarly strong predictors of abundances and lost little in comparison with their joint use. The spatial variables are thus effective surrogates for the physical variables in this extensive reef complex, where information on the physical environment is often not available. Finally, we compare the use of regression trees and linear models for the analysis of these data and show how linear models fail to find patterns uncovered by the trees.
Savannas are among the most variable of terrestrial ecosystems. They undergo large and frequent changes in production, composition and structure and they contain some of the worst examples of degradation by man. There are, accordingly, many references to them as “fragile” and “brittle” ecosystems, with frequent predictions of imminent “collapse”. But these terms have been used loosely, as jargon, and we need now to progress beyond this vague terminology. What, precisely, do we mean by these terms? Is it possible to be more precise? More specifically, can we achieve an understanding of the equilibrium behaviour of savannas? For this deals with how they change, and how much they can change before the change is irrevocable.
The role of fire in determining biome distribution in South Africa has long been debated. Acocks labelled veld types that he thought were 'fire climax' as 'false'. He hypothesised that their current extent was due to extensive forest clearance by Iron Age farmers. We tested the relative importance of fire and climate in determining ecosystem characteristics by simulating potential vegetation of South Africa with and without fire using a Dynamic Global Vegetation Model (DGVM). The simulations suggest that most of the eastern half of the country could support much higher stem biomass without fire and that the vegetation would be dominated by trees instead of grasses. Fynbos regions in mesic winter rainfall areas would also become tree dominated. We collated results of long term fire exclusion studies to further test the relative importance of fire and climate. These show that grassy ecosystems with rainfall >650mm tend towards fire-sensitive forests with fire excluded. Areas below 650mm showed changes in tree density and size but no trend of changing composition to forest. We discuss recent evidence that C4 grasslands first appeared between 6 and 8M years BP, long before the appearance of modern humans. However these grassy ecosystems are among the most recently developed biomes on the planet. We briefly discuss the importance of fire in promoting their spread in the late Tertiary.
(1) Semi-arid savannas, wherever they occur, have generally been overgrazed and encroached on by bush. A model is developed which accounts for the growth of woody vegetation and of grasses, and analyses the competition between them for available soil water. (2) The model is based on Walter's two layer hypothesis. Woody vegetation and grasses compete for water in the surface layers of the soil, but woody vegetation has exclusive access to a source of water relatively deep underground. Where there is only a small biomass of grass the soil surface tends to become impermeable and, in these conditions, the model shows that two different steady states may develop: with a lot of woody vegetation alone, or with a relatively large biomass of grass and rather little woody vegetation. (3) The results are discussed in terms of the concept of resilience. The continued existence of both stable states under ranching conditions seems to depend on periodic heavy, or over-, grazing which allows for the maintenance of unpalatable or unstable grass species, which thus set a minimum to grass biomass--a minimum which cannot be reduced by herbivores. (4) Comparison of the dynamics of various savanna and other natural systems leads to the conclusion that the resilience of the systems decreases as their stability (usually induced) increases.
The use of the plant available moisture (PAM)/plant available nutrients (PAN) concept to compare savanna structure was examined using data from twenty Australian sites. Above-ground biomass was regressed on various combinations of seventeen different estimates of PAM (plant available moisture) and two estimates of PAN (plant available nutrients). The ratios of actual transpirational loss from the subsoil to potential evapotranspiration (PET), and total annual rainfall to PET, were most highly correlated with total biomass. Grass biomass is poorly predicted by PAM on its own, and requires inclusion of woody leaf biomass in the regression. PAN had little effect on total biomass, although it is likely to be important for other, functional aspects of vegetation. The woody : grass ratio is best predicted by an index involving the ratio of subsoil : topsoil moisture. For biomass comparisons the use of a detailed water-balance model to estimate PAM is not warranted.
Several methods to construct confidence intervals for regression quan-tile estimators (Koenker and Bassett (1978)) are reviewed. Direct estimation of the asymptotic covariance matrix requires an estimate of the reciprocal of the error density (sparsity function) at the quantite of interest; some recent work on bandwidth selection for this problem will be discussed. Several versions of the bootstrap for quantile regression will be described as well as a recent proposal by Parzen, Wei, and Ying (1992) for resampling from the (approximately pivotal) estimating equation. Finally, we will describe a new approach based on inversion of a rank test suggested by Gutenbrunner, Jurečková, Koenker, and Portnoy (1993) and introduced in Hušková(1994). The latter approach has several advantages: it may be computed relatively efficiently, it is consistent under certain heteroskedastic conditions and it circumvents any explicit estimation of the sparsity function. A small monte-carlo experiment is employed to compare the competing methods.