Plant and Soil211: 191–205, 1999.© 1999Kluwer Academic Publishers. Printed in the Netherlands.

191

Root distributions in a Grevillea robusta-maize agroforestry system insemi-arid Kenya

D.M. Smith1,∗, N.A. Jackson1, J.M. Roberts1 and C.K. Ong21Institute of Hydrology, Crowmarsh Gifford, Wallingford, Oxfordshire OX10 8BB, UK and2International Centrefor Research in Agroforestry, P.O. Box 30677, Nairobi, Kenya

Received 29 January 1999. Accepted in revised form 27 May 1999

Key words:competition, complementarity, root length, soil water, tree pruning

Abstract

Limited knowledge of root distributions in agroforestry systems has resulted in assumptions that various treespecies are more suited to agroforestry than others, because they are presumed to have few superficial lateralroots. This assumption was tested forGrevillea robustawhen grown with maize (Zea mays) in an agroforestrysystem in a semi-arid region of Kenya. At a site with a shallow soil, root lengths of both species between the soilsurface and bedrock were quantified by soil coring, at intervals over four cropping seasons, in plots containingsole stands and mixtures of the trees and crop; the trees were 4–6 years old and they were severely pruned beforethe third season. Profiles of soil water content were measured using a neutron probe. Prior to pruning of the trees,recharge of soil water below the deepest maize roots did not occur, resulting in significant (P<0.05) suppressionof maize root lengths and downward root growth. Maximum root length densities for both species occurred at thetop of the soil profile, reaching 1.1–1.7 cm cm−3 for G. robusta, but only 0.5 cm cm−3 for maize grown withtrees. Root populations in mixed plots were dominated byG. robustaat all times, all depths and all distances fromtrees and maize and, thus, there was no spatial separation of the rooting zones of the trees and crop. CompetitionbetweenG. robustaand maize for soil water stored near the surface was unavoidable, although pruning reduced itsimpact; complementary use of water by the trees and crop would only have been possible if alternative sources ofwater were available.

Introduction

An important goal of combining trees and crops inagroforestry systems is to make better use of the en-vironmental resources required by plants for growth.One of the means of accomplishing this is to utiliseresources that would otherwise be lost from the sys-tem (Cannell et al., 1996; Sanchez, 1995). Productiveuse of a higher proportion of resources available frombelow ground can be achieved if deep networks of treeroots take up water or nutrients draining or leachingthrough the rooting zone of the crop (van Noordwijket al., 1996). Where such enhancements to resourceutilisation are realised in land-use systems, real gainsin productivity can be achieved that are sustainable in

∗ FAX No: +44 1491 692424. E-mail: m.smith@ioh.ac.uk

the long term (Cooper et al., 1996; Ong and Black,1994).

A drawback to introducing trees into systems ofcrop production, however, is that the trees and cropmay compete for resources. Competition betweentrees and crops results when exploitation of a resourceby trees, for example, reduces its availability to levelsthat limit growth and productivity of the crop (An-derson and Sinclair, 1993; Grime, 1979). If the cropis more valuable to farmers than products from thetrees, such competition will result in failure of theagroforestry system, because it will not be adoptedby farmers. Where nutrients or water is the resourcemost limiting production, below-ground competitionis most important.

In semi-arid regions, water is often the most limit-ing resource and competition for water can impair the

192

effectiveness of agroforestry systems (Onyewotu et al.,1994; Smith et al., 1998). Govindarajan et al. (1996)and McIntyre et al. (1997) identified competition forwater as the major reason for the poor performanceof hedgerow intercropping systems in semi-arid zonesof East Africa. Both found that where seasonal rainswere not sufficient to cause recharge of soil below therooting zone of the crop, uptake of water by hedgesreduced the availability of water to adjacent crops ofmaize or cowpea (Vigna unguiculata), causing lossesin yield.

A key to successfully using agroforestry to in-crease the productive use of resources is to ensure thatthe trees and crop exploit different resource pools, par-ticularly at times when the availability of a resourceis potentially limiting (Ong et al., 1996). In systemswhere such niche differentiation occurs between treesand crops, capture of a limiting resource by treesdoes not reduce its availability to the crop; product-ive utilisation of resources can then increase withoutimpaired crop performance because of competition.Where trees and crops exploit spatially or temporallydistinct resource pools, they are complementary intheir use of resources and are said to exhibit ‘com-plementarity’ in resource use (Ong et al., 1996; vanNoordwijk et al., 1996).

To achieve complementarity in use of below-ground resources, it is widely recommended that deep-rooted tree species are used in agroforestry (Andersonand Sinclair, 1993; van Noordwijk et al., 1996). How-ever, there is a lack of quantitative information on rootdistributions in agroforestry systems which has resul-ted in frequently unfounded suppositions that varioustree species are more suited to agroforestry than oth-ers, because they have few lateral roots near the soilsurface.

An example isGrevillea robusta, a tree used inagroforestry in the highlands of East and CentralAfrica, which is believed to be deep rooting, with fewsuperficial lateral roots (Howard et al., 1997). The aimof the study reported here was to test this assumptionfor G. robustagrown with maize (Zea mays) in anagroforestry system in a semi-arid region of Kenya. Toaccomplish this, the root length distributions of bothG. robustaand maize were quantified in sole standsand mixtures of the trees and crop.

Materials and methods

Location and climate

This study formed part of the CIRUS (for Comple-mentarity in Resource Use on Sloping land) trial atthe ICRAF Research Station, Machakos, Kenya (1◦33′S, 37◦8′ E; 1560 m above mean sea level). The trialwas located on a south-facing hillside, with a slope of∼20%. The soil was an alfisol (Khandic Rhodustalf),sandy clay loam in texture, overlying hard gneiss bed-rock, which varied in depth considerably, rangingfrom 0.2 m or less in places to 2 m. A water table wasnot observed at the site and excavations revealed thatthe bedrock was free of deep cracks and that roots ofthe trees did not penetrate more than 2 to 3 mm intothe weathered surface of the rock. Soil fertility washigh and soil analyses carried out in May 1996 showedthat none of the concentrations of major nutrients werelimiting for plant growth (Lott, 1998).

The climate of the surrounding region is semi-arid,with a bimodal distribution of rainfall; rainy seasonsextend from March to June (the ‘long rains’) and Octo-ber to December (the ‘short rains’). The mean annualtotal is 782 mm, with 345 and 265 mm in the longand short rains, respectively, and the remainder fallingoutside the growing seasons.

Experimental design and site management

The CIRUS trial was initiated in October 1991, whenthree-month old seedlings ofG. robusta(local Embuprovenance) were transplanted into plots measuring20×20 m. A variety of treatments was established,but sampling of roots was confined to the crop-only(designated Cg) treatment and the tree-only (Td) andtree+crop (CTd) treatments; these contained treesplanted in a grid pattern at a spacing of 3×4 m (Figure1), to create a dispersed configuration of trees with adensity of 833 ha−1. Three replications of each treat-ment were used, with the plots arranged in randomisedcomplete blocks.

The trees were pruned annually by loppingbranches from the lowest 1 m of the crowns. In Octo-ber 1996, prior to the short rains, however, the treeswere severely pruned, leaving only the uppermost∼15% of the volume of each tree crown. Interfer-ence by tree roots with the growth of monocropswas minimised by trenching around the boundariesof each crop-only plot to a depth of 1 m at thebeginning of each season. A strip of vetiver grass (Vet-iveria zizanoides), planted along the contour across the

193

Figure 1. Locations of soil cores extracted from plots with: (a)maize only (Cg treatment); trees only (Td); and (c) trees mixed withmaize (CTd). Position numbers are marked for each core. Positions1,2 and 3 were 0.025, 0.26 and, nominally, 0.52 m from the closestmaize plant in the Cg plots, and 0.5, 1.25 and 2.5 m from the base ofthe tree in the Td plots. The coring configuration in the Cg plots wasrepeated for two plants in the CTd plots which lay, approximately,along the transect between the tree and mid-point of the plot.

centre of each plot as an erosion-control measure, wasclipped at 7–14 day intervals to minimise competitionwith adjacent trees or crops.

If not dry-sown prior to the first rain of each grow-ing season, maize (var. Katumani composite) wasplanted in relevant plots after at least 20 mm of rainhad fallen within a period of 7–10 days. Rows of maizespaced 1 m apart were planted along the contours ofthe slope. Approximately 12 days after emergence, themaize was thinned to leave a spacing between plantsin the same row of 0.3 m, giving a density of 33 333plants ha−1.

Collection of root samples

Soil coring was used to assess the vertical and hori-zontal distribution of tree and crop roots. Cores werecollected twice in each of the four growing seasonsbetween the short rains of 1995 and the 1997 longrains, when the trees were 4–6 years of age and 10–12 m in height. Samples were collected after the firstweeks of maize growth and then at anthesis; detailsof the timing of coring in each season are provided inTable 1.

Soil cores were collected using a corer (48 mm in-ternal diameter) that was driven into the ground witha hydraulically-powered hammer and extracted usinga ball clamp and jack. Cores were always taken to thedepth of the bedrock and, after extraction, they wereseparated into increments 0.2 m in length.

Three sets of cores were collected from each of theplots sampled. Each set of cores was taken along atransect between a maize plant or tree and the mid-point between rows (Figure 1). The locations of eachtransect within the plots were selected randomly, al-though the area within 2 m of the vetiver grass stripwas excluded. Cores were collected only once fromany location.

There were three cores per set in the Cg andTd plots and six cores in the CTd treatment (Fig-ure 1); thus, at each sampling, a total of 54 cores(6 positions×3 sets×3 plots) were collected from theCTd plots and 27 (3 positions×3 sets×3 plots) fromthe Cg and Td treatments. A modified sampling regimewas used only in the 1996 long rains, when, in theTd plots, each set contained only two cores and, inthe CTd plots, tree roots were collected from only twocores per set.

Within a few hours of collection, roots in thesamples were carefully washed out of the soil overa 0.5 mm sieve. Root samples were then placed in a

194

Table 1. Growth stage of maize and timing of root sampling for each season, with the number of root samples (n) collected ineach season

Year Seasona Coring Growth stageb Dates of coring Timing of coring (DAEc) n

95 SR A 5–7 leaves 13–17 Nov. 15–19 980

B tasselling/anthesis 13–18 Dec. 45–50 1464

96 LR A 7–8 leaves 23–27 Apr. 20–24 841

B anthesis 29 May–4 Jun. 56–62 864

SR A 7–9 leaves 9–13 Dec. 22–26 1334

B anthesis 13–15 Jan. 57–59 1159

97 LR A 7–9 leaves 28 Apr.–2 May 23–27 1188

B anthesis 27–30 May 52–55 2020

aSR=‘short rains’; LR=‘long rains’.bgrowth stage for maize in Cg plots.cDays after emergence.

small amount of 9% (v/v) vinegar solution and storedin a refrigerator until manually sorted by a team ofworkers into maize andG. robustacomponents, whichwere readily distinguished by colour and morphology;other organic debris and dead roots were discarded.G. robusta is a member of the Proteaceae familywith proteoid roots, which are characterised by seg-ments of roots with tightly-packed clusters of rootlets(Dinkelaker et al., 1995). These were removed andkept separately from the main sample. Before sortingof each sample was completed, it was checked by asingle supervisor to ensure that roots were sorted asconsistently as possible.

Measurement of root length

The length of roots in each sample was determinedby digital image analysis. Prior to analysis, eachsample was stained with 0.01% (w/w) methyl violet.Samples were spread randomly over a transparentsheet and then scanned using a flat-bed scanner (Scan-Jet IIcx, Hewlett Packard Co., Palo Alto CA, USA.)and desktop computer. Samples of proteoid rootletswere prepared in the same way, but were scannedafter using a scalpel blade to shave off the rootletsinto water and then collecting them by filtration. Thelength of roots in each digital image obtained wasdetermined using image-analysis software (DT-Scan,Delta-T Ltd., Cambridge, UK) which utilised the in-tercept method (Newman, 1966) to estimate lengthand stochastic relationships to correct for overlapsbetween roots (Kirchhof, 1992).

Root lengths were divided by the volume of theoriginal soil sample to give root length densities (Ld)for maize and both ordinary, non-proteoid roots and

proteoid rootlets ofG. robusta. Root lengths per unitground area (La) were calculated by dividing the totalroot length for each core by the cross-sectional areaof the core. Dry mass of all root samples was sub-sequently measured, after drying the samples in anoven at 70◦C.

Measurement of soil water content

Profiles of volumetric soil water content were meas-ured using a neutron probe (model IH II, DidcotInstruments Co., Abingdon, UK). Weekly measure-ments were made in arrays of aluminium access tubes(44.5 mm external diameter) installed in 1993. Eacharray in the CTd and Td plots comprised tubes at sixdistances from a tree; in the Cg plots, there were fourtubes per plot. Measurements were replicated in threeplots of each treatment. Details of the configuration ofaccess tubes and calibration of the neutron probe aregiven by Jackson et al. (1999).

Data analysis

To compare quantities of roots among treatments, rootlengths per tree and per maize plant (Lr) were calcu-lated from the profiles of root length density. This wasaccomplished by integrating root length densities overthe depth of each core and the area of soil for whicheach core was representative (Figure 2). An alternativewould have been to calculate mean root length dens-ities weighted by the volume of soil represented byeach sample, but interpretation of the results was lesscomplex if data were expressed as root lengths per treeor plant. Root lengths were normalised to 12 m2 ofland-surface area for each tree and to 0.3 m2 for eachmaize plant.

195

Figure 2. To estimate root length per tree or per plant, root lengthsmeasured in the cores marked 1,2 and 3 were integrated over theareas marked A, B and C, respectively, where the stem of the treeor maize plant was at the centre of area A. Root lengths per treein CTd plots were estimated by integration of root lengths from sixcores over six areas, rather than three.

To compare vertical root distributions among treat-ments, the depth of 50% cumulative root length (d50)was determined for each soil core. Non-linear regres-sion was used to fit the function (Gale and Grigal,1987)

fc = 1− βd (1)

to the profile of cumulative root fraction (fc), from thesoil surface downwards, for each soil core, whered isdepth (in cm) andβ is a regression coefficient. Valuesof d50 for each core were then calculated from

d50= ln(0.5)

ln(β). (2)

Differences among treatments inLr andd50 wereassessed by analysis of variance, with sub-samplesfrom within plots treated as split plots. Effects of lat-eral distance from the plant or tree were tested bytreating core positions as split-split plots. Core depth,or the minimum core depth in a set of cores, was usedas a covariate if its inclusion substantially reducedthe size of the error term in the model. Comparisonsamong individual means were made using contrasts.

Results and discussion

Rainfall

Rainfall at the Machakos Research Station betweenthe beginning and end of this study is plotted in Fig-ure 3. Rainfall was close to the seasonal average inthe 1995 short rains and the 1997 long rains, butwell-below average in the two rainy seasons of 1996.However, in all seasons except the 1997 long rains,substantial dry spells began within 12 days or lessof the emergence of the maize crop. Thus, the dis-tribution of rainfall in the first three seasons of theproject was poor, causing the young maize plantsto be exposed to dry conditions. In contrast, dur-ing the 1997 long rains, conditions for crop growthwere good, as there was frequent and abundant rainfallfor approximately 30 days after the emergence of themaize.

Root lengths and biomass

A summary of the determinations of root length madebetween 1995 and 1997, from a total of 9850 rootsamples, is shown in Figure 4. Mean root lengthsper tree, estimated by extrapolating from root lengthsmeasured in soil cores, are shown for each of the eighttimes cores were collected. The population of rootswas dominated consistently by tree roots, which had amean root length per tree of 90±2.6 km for the Td andCTd plots over all seasons. When mixed with trees,the mean length of maize roots per tree for all seasonswas much less, at just 8±0.8 km; this was also lessthan the mean maize root length for a comparable areaof the maize-only crop, which was 20±1.7 km for anarea of 12 m2, the area occupied by a single tree in thetree-crop mixture.

Root lengths per maize plant were not affected bythe proximity of maize to the trees, as differences inroot lengths between maize plants sampled in the firstand second crop rows away from trees in the CTd plotswere almost always non-significant (P<0.05) (Figure5(a)). When root lengths for maize in the monocropand the tree-crop mixture were compared, differenceswere not significant (P<0.05) in the early vegetat-ive growth stages (Figure 5(b)), although root lengthstended to be higher for plants in the monocrop. At an-thesis, maize root lengths were significantly (P<0.05)or close-to-significantly lower in the mixture in all sea-sons except the 1997 long rains (Figure 5(b)). Hence,when rains were poor and dry spells occurred priorto anthesis, the trees suppressed maize root growth

196

Figure 3. Rainfall at Machakos Research Station, Kenya, between September 1995 and July 1997. The shaded areas mark the average timingof the short rains (SR) and long rains (LR) and the arrows along the x-axis mark the date of emergence in each season. Total rainfall for eachseason is given at the top of each shaded bar.

Figure 4. Root lengths normalised to 12 m2 of ground area, the area occupied by a single tree. The lines show ordinary (solid lines) andproteoid (dotted lines) root lengths forG. robustain tree-only (open symbols) and mixed (closed symbols) plots; the bars show maize rootlengths for the equivalent area in crop-only (open bars) and mixed (closed bars) plots. The labels for each coring are defined in Table 1. Errorbars show±1 s.e. (symbols) or +1 s.e. (bars).

197

Figure 5. (a) Maize root lengths (Lr) in the first and second crop row relative to the trees in mixed plots (see Figure 1(c)); (b) maizeLrin crop-only and mixed plots; (c)Lr for G. robustain tree-only and mixed plots; and (d)Lr for G. robustaand maize in mixed plots. Thesignificance of comparisons within each coring are shown along the ‘coring’ axis; ns=not significant; (∗)=P<0.10; ∗=P<0.05; ∗∗=P<0.01.The labels for each coring are defined in Table 1.

after development of the crop had progressed past theseventh or eighth leaf, about 21 days after emergence.With abundant and well-distributed rains, as occurredin the 1997 long rains, and after the leaf area of thetrees was severely reduced by pruning, the trees didnot affect the growth of maize roots.

Comparison of results in tree-only and mixed plotsshowed that maize did not significantly (P<0.05) af-fect root lengths forG. robusta(Figure 5(c)). Rootlengths in the mixed plots were significantly (P<0.01)less for maize than the trees on all occasions (Figure5(d)); thus, tree roots dominated the root populationin the tree-crop mixture at all times, regardless of

whether rainfall was abundant or the trees had beenpruned.

Mean root biomass ofG. robustaand maize atmaize anthesis is given for each treatment in Figure6. The values forG. robustawere similar to values oftotal root biomass estimated by Dhyani et al. (1990)and Toky and Bisht (1992) for trees commonly usedin agroforestry in drylands; they were considerablyhigher, however, than values of fine root biomass pub-lished by Jonsson et al. (1988), because coarse rootswere retained in the samples. Root biomass for themaize monocrop and the tree-crop mixture are similar

198

Figure 6. Biomass forG. robustaroots, proteoid rootlets and maize roots in crop-only, tree-only and mixed plots.

to values given by Jackson et al. (1996) as typical of,respectively, crops and tropical savannah.

Lengths and biomass of proteoid rootlets

Proteoid rootlets increased the root length per treeby an average of 35±1.8 km (Figure 4), but accoun-ted for less than 10% of the below-ground biomassof G. robusta(Figure 6). The main function of pro-teoid rootlets is to enhance the uptake of nutrientswith low solubility, for example phosphorus and man-ganese, by increasing the surface area of roots in small,nutrient-rich patches and by releasing chemical exud-ates into soil to mobilise these nutrients (Dinkelaker etal., 1995; Skene et al., 1996). Their role in the uptakeof water is unclear, but because water is highly mobilein soil, they are unlikely to have substantially alteredthe ability of the trees to take up water. Lengths anddistributions of proteoid rootlets did not differ betweenthe tree-only and mixed plots (data not shown).

Root length densities

Profiles of root length density forG. robustaand maizeare shown in Figure 7. Such profiles were determined

for each coring in all seasons of the study, but onlythose from the time of anthesis in the maize monocropduring the 1996 short rains, when rainfall was belowaverage and poorly distributed, and 1997 long rains,when rainfall was well distributed, are shown.

In all seasons, root length densities (Ld) for rootsof G. robustadeclined approximately exponentiallyfrom the surface downwards (Figure 7), with meanvalues for roots ofG. robustaat the top of the pro-file for each time of sampling ranging from 1.1 to1.7 cm cm−3; profiles for the density of proteoidrootlets had a similar shape, with maximum valuesbetween 0.5 and 1.1 cm cm−3 (data not shown). Val-ues ofLd measured by Jones et al. (1998) forAcacianilotica andProsopis julifloratrees in a semi-arid re-gion of Nigeria were similar, but Govindarajan et al.(1996) found maximum values of only 0.4 cm cm−3

for pruned hedgerows ofLeucaena leucocephalaatMachakos.

Root length densities for maize also appeared todecline approximately exponentially with depth. Atthe early vegetative growth stage, maximumLd formaize was similar in the monocrop and mixture, withmean values at the top of the profile ranging between0.25 and 0.5 cm cm−3 (data not shown). In seasons

199

Figure 7. Profiles of root length densities for roots ofG. robusta(solid lines) and maize (dotted lines) at maize anthesis in (a) the 1996 shortrains and (b) the 1997 long rains. Values are means over soil cores from all lateral positions and plots. Differences in maximum depths of treeroots among treatments resulted from variation among plots in the depth of bedrock. Error bars show +1 s.e.

when rainfall was poor,Ld for maize in the tree-cropmixture was similar at anthesis, indicating that therewas little further growth of maize roots (Figure 7(a)).In the monocrop, however, meanLd at the top ofthe profile increased to between 0.8 to 1.0 cm cm−3

by anthesis, even in seasons with low rainfall (Figure7(a)). When rainfall was good, in the 1997 long rains,such a contrast in root growth of maize between themonocrop and mixture did not occur (Figure 7(b)).

Root length densities for maize monocrops can bemuch higher than observed when hybrid varieties aregrown under high-input management regimes (Mengeland Barber, 1974; Nakamoto et al., 1992). The valuesof maizeLd measured were more typical, however,of crops grown using low-input management in EastAfrica (Govindarajan et al., 1996; Mekonnen et al.,1997).

200

Figure 8. (a) Fitting of Equation 1 to a profile of cumulative frac-tional root length (fc); the dotted lines mark the depth of 50% rootlength (d50). (b) Mean profiles offc, over all seasons, for roots ofG.robusta(–), proteoid rootlets ofG. robusta(- - -), maize at anthesisin the monocrop (-·-·) and in the tree-crop mixture (· · ·).

Vertical root distributions

To compare root distributions among treatments,depths of 50% root length (d50) were determined byfitting Equation 2 to data from each soil core, asshown, for example, in Figure 8(a). Curves fitted todata from all seasons show thatG. robustahad thedeepest root distribution and maize in mixed plots hadthe shallowest (Figure 8(b)). Proteoid rootlets weremost abundant near the soil surface, as others havefound (Dinkelaker et al., 1995) (Figure 8(b)). Differ-ences ind50 for G. robustaroots between mixed andtree-only stands were non-significant (P<0.05) (Fig-ure 9(c)) and thus maize had no effect on the verticaldistribution of tree roots. In contrast, however, thetrees affected the distribution of maize roots. Regard-less of which row maize plants were in (Figure 9(a)),their roots were generally more shallowly distributedin the mixture than the monocrop (Figure 9(b)). Val-ues of d50 for maize at anthesis were significantly(P<0.05) lower in the mixture during seasons whenrainfall was low or poorly distributed, suggesting thatthe presence of trees suppressed the downward pen-etration of roots into the soil. In the 1997 long rains,when rainfall was more abundant, this effect was notsignificant.

In the tree-crop mixture, roots ofG. robustaweresignificantly (P<0.01) more deeply distributed thanmaize roots (Figure 9(d)) at all times but one (whenmaize root samples may have been contaminated byweed roots at depth). This could give the impressionthat the trees and crop exploited separate soil nichesand suggest that they were complementary in their useof below-ground resources; however, this is a false im-

pression, as tree roots dominated the root population inthe mixture at all times and at all depths (Figure 7).

Lateral root distributions

Root lengths for maize tended to be highest immedi-ately adjacent to the stem (Figure 10). This tendencywas strongest in the tree-crop mixture, with the linearcontrast across core positions of root length per unitground area (La) significant (P<0.05) in five of theeight corings.

Roots ofG. robustadominated the population ofroots at all distances from the trees (Figure 11), butthey did not appear to be uniformly distributed withdistance from the stem. Rather than being most abund-ant closest to the tree, there was a tendency forG.robustaroot lengths to increase with distance (Figure12). The linear contrast inLa across the core positionssampled in the CTd plots was significant (P<0.05) inthree of the six corings prior to the 1997 long rains,with root lengths lower near the trees than furtheraway. This tendency was not apparent in the 1997 longrains, perhaps as a result of greater root die-back at theextremities of the root systems of the trees after thecrowns were severely pruned in October 1996.

In general, the vertical distribution of tree or croproots did not vary laterally, as values ofd50 for G.robustaroots, proteoid rootlets or maize were not sig-nificantly (P<0.05) affected by proximity to a tree ormaize plant (data not shown). There was some tend-ency for a more shallow distribution of roots adjacentto the stems of maize plants in the monocrop, but thiswas not found consistently.

Root distributions and competition for water

In the below-ground environment of the CIRUS site,maximum root length densities forG. robustaandmaize coincided at the top of the soil profile, confirm-ing an inference from root counts on trench walls byHuxley et al. (1994). Thus, spatial separation of therooting zones ofG. robustaand maize did not occurand, as an alternative source of water such as ground-water did not exist, competition for water stored in thesurface layers of the soil could not be avoided.

The impact of water uptake by the trees on the dis-tribution of soil water is evident from records of soilwater content, which show that recharge of lower soillayers was severely reduced in the mixed plots relativeto the maize monocrop (Figure 13). The dominant rootsystem of the trees and the high density of their rootsat the top of the profile meant the trees were capable

201

Figure 9. (a) Depth of 50% root length (d50) for maize in the first and second crop row from the trees in mixed plots; (b)d50 for maize incrop-only and mixed plots; (c)d50 for G. robustaroots in tree-only and mixed plots; and (d)d50 for G. robustaroots and maize in mixed plots.The significance of comparisons within each coring are shown along the ‘coring’ axis; ns=not significant; (∗)=P<0.10;∗=P<0.05;∗∗=P<0.01.The labels for each coring are defined in Table 1.

of capturing more of the water available from rainfall.Prior to pruning of the trees in October 1996, when theleaf area and thus the water requirements of the treeswere high, uptake by the trees would have limited thesupply of water to maize and drainage into deeper soilin the mixed plots. As a result, the growth of maizeroots was suppressed (Figure 5(b)) and constrained tothe upper region of the soil profile (Figure 9(b)); belowthis, reduced maize root growth relative to the mono-crop probably resulted from low soil water availability(Figure 13).

After October 1996, with their leaf area initiallyreduced by about 85%, the water requirements of the

trees were greatly reduced. Rainfall in the short rainsof 1996 was far below average, however, and therewas less residual soil water in the mixed plots than themonocrop. As a result, there was little improvement inmaize root growth in the tree-crop mixture. In the fol-lowing season, when rainfall was close to average, theleaf area of the trees remained much less than it wasbefore pruning and their reduced water requirementsenabled substantial recharge of deeper soil layers (Fig-ure 13(b)), despite the continued dominance of the treeroot systems. This contributed to a considerable im-provement in the growth of maize roots in the mixture(Figure 5(b) and Figure 9(b)).

202

Figure 10. Root lengths per unit ground area (La) at three lateraldistances (see Figure 1) from maize plants in (a) crop-only and (b)mixed plots. Values for lateral positions 1 to 3 are shown in orderfrom left to right in each group of vertical bars. The significance oflinear contrasts across the three core positions are marked for eachcoring; ns = not significant;∗=P<0.05. The labels for each coringare defined in Table 1.

Figure 11. Root lengths per unit ground area (La) at six lateraldistances (see Figure 1) from trees forG. robustaroots (solid line)and maize (dashed line) in mixed plots at the second coring in the1996 short rains. Differences in root lengths between species weresignificant (P<0.01) at each position, in all corings. Error bars show+1 s.e.

Root distributions and complementarity

To reduce competition with crops for below-groundresources, the ideal tree for agroforestry should have

Figure 12. Root lengths per unit ground area (La) at six lateral dis-tances (see Figure 1) fromG. robustatrees in mixed plots. Values forlateral positions 1 to 6 are shown in order from left to right in eachgroup of vertical bars. The significance of linear contrasts across thesix core positions are marked for each coring; ns=not significant;(∗)=P<0.10; ∗=P<0.05. The labels for each coring are defined inTable 1.

a deep root system and little root proliferation nearthe top of the profile, thereby enabling the crop toutilise resources from near the soil surface, while thetrees have sole access to deeper layers (Schroth, 1995).However, as observed in this study forG. robustaandby others for a variety of trees used in agroforestry(Dhyani et al., 1990; Jonsson et al., 1988; Toky andBisht, 1992), most trees do not conform to this idealand have their maximum root length density at thetop of the profile (Schroth, 1995). Hence, the spa-tial differentiation of the root systems of trees andannual crops in agroforestry is most likely to be un-obtainable in reality and complementarity in use ofbelow-ground resources is never likely to be achievedby selecting one tree species over another on the basisof root distributions. In general, therefore, the extentof below-ground complementarity in agroforestry isdetermined by the distribution of resources, with treesand crops only complementary in their use of wa-ter if there is an alternative source of water uniquelyavailable to the trees.

Smith et al. (1997) showed by isotopic means thattrees and adjacent crops can be complementary in theiruse of water if groundwater is accessible to tree butnot crop roots. Failing that, spatial complementaritywould be possible if there was sufficient drainage forsubstantial quantities of water to be stored beyond therooting zone of the crop (van Noordwijk et al., 1996).However, McIntyre et al. (1997) contended that this isunlikely in semi-arid regions, because of low rainfalland uptake of water by the trees and crop from shal-low soil. Soil water records from this study confirm

203

Figure 13. Volumetric soil water content (θv) in crop-only (closed symbols) and mixed plots (open symbols) at depths of (a) 0.2 m, (b) 0.6 mand (c) 1.2 m, from October 1995 to June 1997. The shaded areas mark the average timing of the short rains (SR) and long rains (LR).

that their view was correct for the combination ofG.robustaand maize used, particularly before the treeswere pruned.

Strategies to reduce competition for water

When trees are well-established and they dominate theroot population, as in the mixed plots of the CIRUStrial, trees are capable of capturing more soil wa-ter than the crop and reducing the water availableto the crop to levels that limit growth. Root prun-ing could be used to reduce the dominance of treeroots near the soil surface, but it is usually not feas-ible in non-mechanised farming systems and it mayhave undesirable effects on carbon and nutrient cyc-ling (Schroth, 1995). Its effectiveness over time is alsouncertain, as there is little information available onre-growth after root pruning.

Another option is to use pruning of tree crowns toreduce demand for water by the trees and, therefore,the quantity of water they extract from the soil (Jack-son et al., 1999). In addition, pruning of crowns maycause some die back of roots, as a result of reductionsin the supply of photosynthate (Fownes and Anderson,1991; van Noordwijk et al., 1996). In the case ofG.robustain the CIRUS trial, there appeared to be somedecline in root lengths in the months after pruning(Figure 4), particularly at the extremities of the rootsystem (Figure 12), but this could have resulted frompruning or the effects of, for example, cyclical rootturnover which may occur annually.

Whatever the cause of the apparent decline inG.robustaroot lengths after pruning, it was insufficientto affect the dominant position of tree roots in the rootpopulation. The trees therefore maintained a muchhigher capability to capture water resources than the

204

crop. As a consequence, pruning of crowns can onlybe effective in reducing the impact on the crop ofcompetition for water if rainfall exceeds the demandsof the trees. The probability of this condition beingfulfilled is higher after pruning. Thus, while pruningof crowns where tree roots are dominant is unlikely tochange the relative capacities of the tree and crop rootsystems to capture water resources, it will enhance thelikelihood that rain in a season will be sufficient tosupport the growth of both trees and a crop.

At the field scale, increasing the spacing betweentrees will result in a reduction in their demand forwater and, if the distance between them is suffi-cient, a reduction in the mean length density of treeroots. However, lower numbers of trees in fields maydiminish their benefits for soil conservation, crop mi-croclimate and nutrient cycling, as well as their socialand economic value to farmers. There may be optimaldensities of trees in cropped fields where the beneficialeffects of trees exceed the costs of tree-crop compet-ition by the highest amount possible. Optimal treedensities are likely to depend strongly on climatic andedaphic conditions and on the spatial configurationof trees, as some configurations may minimise com-petition while maintaining or enhancing the positiveeffects of trees.

As the root distributions of most trees and annualcrops appear to coincide (Schroth, 1995), there doesnot appear to be much prospect of finding tree spe-cies with root systems ideally suited to agroforestry.Thus, when selecting trees for use in agroforestry inregions where competition with crops for water is amajor concern, the amount of water required by dif-ferent tree species or provenances should be a moreimportant criterion than any perceived differences inroot distribution.

Conclusions

The assumption held previously, thatG. robusta isdeep rooted with few superficial lateral roots, wasincorrect for well-established trees. Under the below-ground conditions at the site studied, where the soilwas shallow and there was no water table, roots ofG.robustawere present below the deepest maize roots,but the maximum root length densities for both thetrees and crop coincided at the top of the profile. Withthe trees between four and six years old,G. robustadominated the root population at all times, all depthsand all distances from tree or maize stems. Hence,

there was no spatial separation of the rooting zonesof the trees and crop.

Unless the trees were heavily pruned and rainfallwas abundant, recharge of soil water below the maizeroots in the tree-crop mixture did not occur. As therewere no alternative sources of water available to eitherthe trees or crop, competition for rainwater storedclose to the soil surface could not be avoided. Thedominance of theG. robustaroot system ensured thatthe trees had a much higher capacity to capture waterresources than the maize. Competition for water thusresulted in severe suppression of maize root growth,which was relieved only when demand for water bythe trees was reduced by pruning.

This study has provided further evidence that, par-ticularly in semi-arid environments, the spatial differ-entiation of the root systems of trees and annual cropsin agroforestry may be unobtainable. It is not reason-able to expect trees and crops to be complementary intheir use of water just because maximum root depthsare greater for trees. Such a distribution of roots inmixtures can only result in spatial complementarityfor water if there is an alternative source of wateruniquely available to the trees. Thus, when planningstrategies for managing or implementing agroforestry,complementarity between trees and annual crops inuse of water, and below-ground resources in general,should be expected only where suitable distributionsof resources are likely to be found.

Acknowledgements

This publication is an output from a research projectfunded by the Department for International Devel-opment of the United Kingdom. However, the De-partment for International Development can accept noresponsibility for any information provided or viewsexpressed. This work was funded from project R6363of the Forestry Research Programme.

We are very grateful to Mr Elijah Kamalu andMr Patrick Angala, technicians at Machakos ResearchStation, and their team of assistants for the care andpatience they brought to the long-running programmeof soil coring undertaken. We also acknowledge withgratitude the invaluable advice and guidance given byProfessors Peter Gregory and Roger Mead of ReadingUniversity on the measurement of root length, experi-mental design and the statistical analysis of root lengthdata.

205

References

Anderson L S and Sinclair F L 1993 Ecological interactions inagroforestry systems. Agrofor. Abstr. 6, 57–91.

Cannell M G R, van Noordwijk M and Ong C K 1996 The centralagroforestry hypothesis: the trees must acquire resources that thecrop would not otherwise acquire. Agrofor. Syst. 34, 27–31.

Cooper P J M, Leakey R R B, Rao M R and Reynolds L 1996 Agro-forestry and the mitigation of land degradation in the humid andsub-humid tropics of Africa. Exper. Agric. 32, 235–290.

Dhyani S K, Narain P and Singh R K 1990 Studies on root distri-bution of five multipurpose tree speceies in Doon Valley, India.Agrofor. Syst. 12, 149–161.

Dinkelaker B, Hengeler C and Marschner H 1995 Distribution andfunction of proteoid roots and other root clusters. Bot. Acta 108,183–200.

Fownes J H and Anderson D G 1991 Changes in nodule androot biomass ofSesbania sesbanand Leucaena leucocephalafollowing coppicing. Plant Soil 138, 9–16.

Gale M R and Grigal D F 1987 Vertical root distributions of northerntree species in relation to successional status. Can. J. For. Res. 17,829–834.

Govindarajan M, Rao M R, Mathuva M N and Nair P K R 1996Soil-water and root dynamics under hedgerow intercropping insemiarid Kenya. Agron. J. 88, 513–520.

Grime J P 1979 Plant Strategies and Vegetation Processes. JohnWiley & Sons, London. 222 p.

Howard S B, Ong C K, Black C R and Khan A A H 1997 Using sapflow gauges to quantify water uptake by tree roots from beneaththe crop rooting zone in agroforestry systems. Agrofor. Syst. 35,15–29.

Huxley P A, Pinney A, Akunda E and Muraya P 1994 A tree/crop in-terface orientation experiment with aGrevillea robustahedgerowand maize. Agrofor. Syst. 26, 23–45.

Jackson R B, Canadell J, Ehleringer J R, Mooney H A, Sala O Eand Schulze E D 1996 A global analysis of root distributions forterrestrial biomes. Oecologia 108, 389–411.

Jackson N A, Wallace J S and Ong C K 1999 Tree pruning asa means of controlling water use in an agroforestry system inKenya. For. Ecol. Manage. (In press).

Jones M, Sinclair F L and Grime V L 1998 Effect of tree species andcrown pruning on root length and soil water content in semi-aridagroforestry. Plant Soil 201, 197–207.

Jonsson K, Fidjeland L, Maghembe J A and Högberg P 1988 Thevertical distribution of fine roots of five tree species and maize inMorogoro, Tanzania. Agrofor. Syst. 26, 23–45.

Kirchhof G 1992 Measurement of root length and thickness using ahand-held computer scanner. Field Crops Res. 29, 79–88.

Lott J E 1998 Resource Capture and Use in Semi-Arid OverstoreyAgroforestry Systems. Ph.D. Thesis. University of Nottingham,Nottingham, UK. 259 p.

McIntyre B D, Riha S J and Ong C K 1997 Competition for waterin a hedge-intercrop system. Field Crops Res. 52, 151–160.

Mekonnen K, Buresh R J and Jama B 1997 Root and inorganic ni-trogen distributions in sesbania fallow, natural fallow and maizefields. Plant Soil 188, 319–327.

Mengel D B and Barber S A 1974 Development and distribution ofthe corn root system under field conditions. Agron. J. 66, 341–344.

Nakamoto T, Matsuzaki A and Shimoda K 1992 Root spatial distri-bution of field-grown maize and millets. Japan. J. Crop Sci. 61,304–309.

Newman E I 1966 A method of estimating the total length of root ina sample. J. Appl. Ecol. 3, 139–145.

Ong C K and Black C R 1994 Complementarity of resource usein intercropping and agroforestry systems.In Resource Captureby Crops. Eds J L Monteith, R K Scott and M H Unsworth. pp255–278. Nottingham University Press, Nottingham, UK.

Ong C K, Black C R, Marshall F M and Corlett J E 1996 Principlesof resource capture and utilisation of light and water.In Tree-Crop Interactions: A Physiological Approach. Eds C K Ong andP Huxley. pp 73–158. CAB International, Wallingford, Oxon,UK.

Onyewotu L O Z, Ogigirigi M A and Stigter C J 1994 A study ofcompetitive effects between aEucalyptus camaldulensisshelter-belt and an adjacent millet (Pennisetum typhoides) crop. Agric.Ecosys. Environ. 51, 281–286.

Sanchez P A 1995 Science in agroforestry. Agrofor. Syst. 30, 5–55.Schroth G 1995 Tree root characteristics as criteria for species se-

lection and systems design in agroforestry. Agrofor. Syst. 30,125–143.

Skene K R, Kierans M, Sprent J I and Raven J A 1996 Structural as-pects of cluster root development and their possible significancefor nutrient acquisition inGrevillea robusta(Proteaceae). Ann.Bot. 77, 443–451.

Smith D M, Jarvis P G and Odongo J C W 1997 Sources of wa-ter used by trees and millet in Sahelian windbreak systems. J.Hydrol. 198, 140–153.

Smith D M, Jarvis P G and Odongo J C W 1998 Management ofwindbreaks in the Sahel: the strategic implications of tree wateruse. Agrofor. Syst. 40, 83–96.

Toky O P and Bisht R P 1992 Observations on the rooting patterns ofsome agroforestry trees in an arid region of north-western India.Agrofor. Syst. 18, 245–263.

van Noordwijk M, Lawson G, Soumaré A, Groot J J R and HairiahK 1996 Root distributions of trees and crops: competition and/orcomplementarity.In Tree-Crop Interactions: A PhysiologicalApproach. Eds C K Ong and P Huxley. pp 319–364. CABInternational, Wallingford, Oxon, UK.

Section editor: B E Clothier