gross rainfall and its partitioning into throughfall...

Gross rainfall and its partitioning into throughfall, stemflow andevaporation of intercepted water in four forest ecosystems in

western Amazonia

C. Tobon Marina,b,* , W. Boutena, J. Sevinka

aFysisch Geografisch Bodemkundig Laboratorium, University of Amsterdam, Nieuwe Achtergracht 166, 1018 WV Amsterdam, The NetherlandsbThe Tropenbos Foundation, Colombia

Received 29 July 1999; revised 4 May 2000; accepted 4 July 2000

Abstract

The partitioning of gross rainfall into throughfall, stemflow and evaporation of intercepted rainfall was studied in four forestecosystems in the Middle Caqueta´, Colombian Amazonia. Data on climate was collected automatically on an hourly basisduring a five-year period. Weekly measurements of rainfall, throughfall and stemflow were carried out during a period of twoyears, while daily measurements, on an event basis, were carried out during two subsequent years. Throughfall, stemflow andevaporation in each forest were checked for correlations with gross rainfall characteristics, canopy gap fraction, tree crown areaand bark texture. Canopy gap fraction differed between forests, ranging from 9% on the flood plain to 17% on the Tertiarysedimentary plain. Rainfall was rather evenly distributed over the year, with one dry period from December to February. 92% ofthe rain fell in single showers of less than 30 mm and most of the storms (56%) fell in less than one hour, during the afternoon orearly night. Throughfall ranged from 82 to 87% of gross rainfall in the forests studied and varied with gross rainfall in allforests. It depended on the amounts and characteristics of rainfall, but differences in throughfall among forests, when comparingsimilar rainfall events, clearly indicated that throughfall also depends on forest structure. Stemflow contributed little to netprecipitation (on average 1.1% of gross rainfall in all forests) and showed a power relation with gross rainfall. Correlationsbetween stemflow per tree, projected crown area and bark texture were very poor as indicated by the low coefficients ofdetermination. Evaporation during rainfall events exhibited a linear relation with rainfall duration and the ratio of evaporationover gross rainfall increased with forest cover (1-gap fraction) in the forests studied. The structure of the forests seemed to varyconsiderably and given its influence on rainfall partitioning it may explain both differences and similarities between resultsfrom this study and those from most other studies within Amazonia.q 2000 Elsevier Science B.V. All rights reserved.

Keywords: Amazonia; Tropical rain forest; Throughfall; Stemflow; Evaporation; Forest structure

1. Introduction

The Amazonian rain forest seems to play an impor-tant role in the regulation of regional and global

climate (Salati and Vose, 1984). Using a modellingapproach, Eltahir and Bras (1993) concluded that theatmosphere in the Amazon basin is an open systemand that the net input of atmospheric moisture into thebasin is about 32%, about 68% of the gross inputleaving the basin. They also found that the recyclingratio of the Amazon basin is about 25–35%. This ratiodiffers from those found by others, which were based

Journal of Hydrology 237 (2000) 40–57www.elsevier.com/locate/jhydrol

0022-1694/00/$ - see front matterq 2000 Elsevier Science B.V. All rights reserved.PII: S0022-1694(00)00301-2

* Corresponding author. Tel.:131-205257442; fax: 131-205257431.

E-mail address:[email protected] (C. Tobo´n Marin).

on the erroneous hypothesis that the atmosphere overthe Amazon basin is a closed system (Molion, 1975;Lettau et al., 1979; Salati et al., 1979). Eltahir andBras (1993) furthermore concluded that deforestationof Amazonia would increase the surface temperatureand decrease the heating of the upper troposphere,which would result in a reduction of precipitation.The foregoing illustrates that recent concern abouttropical rain forest deforestation focuses on its impacton climate at regional and global scale (e.g. TheAnglo Brazilian Amazonia Climate observationstudy ABRACOS, Gash et al., 1996; the LargeScale Biosphere–Atmosphere experiment in Amazo-nia, Nobre et al., 1996; Eltahir and Bras, 1993).Current research concentrates on global circulationmodels and on those parameters and fluxes that playa role in the global climate (e.g. Gash et al., 1996).Nevertheless, it is often argued that additional work isneeded to explain and improve model predictions forimpacts of deforestation (Bruijnzeel, 1990; Nobre etal., 1991). Evidently, local hydrological studies onundisturbed mature rain forests would provide baselevel information on initial conditions that mightallow for an evaluation of the presumed influence ofdeforestation on regional and global climate. Thispaper concerns such local study.

In the Middle Caqueta´ (Colombian Amazonia), thestructure and species composition of the forest varyconsiderably between the different landscape units(Duivenvoorden and Lips, 1995; London˜o, 1993).This causes equally large variations in temporal andspatial patterns of water fluxes in these units, leadingto local differences in water and nutrient stocks in thevarious forest compartments (Vitousek and Denslow,1986; Tanner, 1985). Thus, the partitioning of rainfallinto throughfall and stemflow leads not only to a morediffuse input of water into the forest floor, but also tolocal concentration around the base of tree stems,which is known to induce spatial variability in soilproperties and soil moisture conditions (Waidi et al.,1992). In tropical forests, the abundance of epiphytes,climbers and aerial roots renders this partitioningmuch more complicated than in temperate forests(Longman and Jenı´k, 1990).

It is well understood that in forested areas generallytotal evaporation is larger than in areas with shortervegetation (e.g. grass) mainly due to the larger inter-ception by the forest canopy (Bosch and Hewlett,

1982), which has been related to the large aerody-namic conductance of forest (Stewart, 1977). Addi-tionally, there has been an increasing awareness thatevaporation of intercepted rainfall has to be investi-gated separate from transpiration, especially in veryhumid areas (Hutjes et al., 1990; Shuttleworth andCalder, 1979). Furthermore, most studies on forestinterception showed that this interception is closelyrelated to gross rainfall amounts and characteristics.However, the influence of forest structure oninterception is poorly known.

Most rainfall interception and water balance studiesin Amazonia were executed in Brazil (Ubarana, 1996;Leopoldo et al., 1995; Lesak, 1993; Lloyd andMarques, 1988; Shuttleworth, 1988) and only a fewin other parts of Amazonia (Ho¨lscher et al., 1997;Jetten, 1996; Wright et al., 1992; Jordan, 1978).Thus, in Colombian Amazonia, which representsone of the most humid areas within the basin, verylittle attention has been paid to the hydrology of forestecosystems and to the effects of forest structure onwater dynamics.

This paper concerns a study designed to addressthis lack of knowledge by measuring rainfall and itspartitioning after entering the canopy in four undis-turbed rain forests in the Middle Caqueta´, ColombianAmazonia. It focuses on the analysis of long-termhydrological measurements of rainfall, throughfall,stemflow, the resultant evaporation and the relatedstructure of these forests.

2. The study area

The study area is in Pen˜a Roja (Nonuya Indiancommunity) near Araracuara, Middle Caqueta´,Colombia, (08 370 and 18 240 S, 728 230 and 708 430

W; Fig. 1). Climate is classified as equatorial super-humid Afi (Koppen, 1936). The research sites arepermanent undisturbed forest plots, used by theTropenbos Foundation for its research. They lieapproximately 200–250 metres above sea level andform a sequence from the lower terrace of the RiverCaqueta´ to the Tertiary sedimentary plain. Based ondata from the manual meteorological station at Arara-cuara (IDEAM), average annual rainfall in the area isabout 3100, April being the wettest month andJanuary the driest.

C. Tobon Marin et al. / Journal of Hydrology 237 (2000) 40–57 41

Colombian Amazonia comprises 403,000 km2 andthe major part of this area is covered by mature rainforests, classified as ombrophilous tropical forest(Duivenvoorden, 1995). The research plots are

located in the four main land units in the area: theTertiary sedimentary plain, the upland terraces ofthe River Caqueta´ (high and low terraces) and theflood plain. The vegetation is very rich in species

C. Tobon Marin et al. / Journal of Hydrology 237 (2000) 40–5742

Fig. 1. Location of the research sites in the various land units in the Middle Caqueta´, Colombian Amazonia.

(Duivenvoorden and Lips, 1995; Londonõ, 1993) andis typical for undisturbed mature forests in the westernpart of the Amazon basin. The canopy reaches about25–30 m above the forest floor with some emergenttrees reaching up to 45 m in the rarely inundated floodplain. There are three to four canopy layers, but largetree crowns in the upper canopy form the bulk of thevegetation. Lower canopies contribute far less to theforest cover. Small palms reaching a height of 2–4 mconstitute the lowest layer. The land units differ in thetotal standing biomass, species diversity and treedensity (Duivenvoorden and Lips, 1995; Londonõ,1993). Other differences between plots pertain to thestructure of the forest canopy (canopy cover) and thecontribution of epiphytes, climbers and aerial roots. Amore detailed description and vegetation classifica-tion of the research sites is given by Duivenvoordenand Lips (1995); and Londonõ (1993).

3. Materials and methods

The areas for the present study were selected asbeing representative for the natural vegetation inthe main land units from this part of the Amazonbasin. Three subplots were selected in the Tertiarysedimentary plain (SP) and two subplots in thehigh terrace (HT), the low terrace (LT) and theflood plain (FP), respectively, to measure grossrainfall above the forest canopy, throughfall andstemflow (Fig. 1). In 1992, approximately 3 kmfrom the plots, in an open area of about 20 hectare(within an Indian community village), anautomatic weather station (AWS) was installedto measure gross rainfall, temperature, airhumidity, incoming radiation, wind speed, winddirection and Class A pan evaporation.Parameters were measured and recorded every30 s using a datalogger (CR10 Campbell ScientificInstruments), which additionally recorded meansor totals every 20 min.

Rainfall in the open area was measured by a tippingbucket raingauge with a resolution of 0.2 mm provid-ing information on the number and duration ofshowers and on total rainfall. Gross rainfall in eachplot was measured in two ways: (1) automaticmeasurements with one tipping bucket installed inthe top of an emergent tree crown after clearing all

branches; (2) manual measurements with two rain-gauges per subplot suspended from cords attached totwo emergent trees in small gaps within the forest.Throughfall was measured in the same rainfallsubplots using 20 collectors per subplot, randomlylocated in an area of 50 by 20 m (1000 m2). Evapora-tion from the collectors was avoided by using aninternal plastic tube running from the funnels to thebottom of the collectors. To allow for direct correla-tion, all funnels for gross rainfall above the forestcanopy and for throughfall had an orifice of298.6 cm2. Throughfall and forest rainfall collectorswere calibrated against standard raingauges in theopen.

Because of the large variability in throughfall dueto the forest structure (Jetten, 1996; Ford and Deans,1978) many readings are needed to study forest inter-ception. However, when using average values,moving the collectors has a positive effect by reducingthe standard error of estimations (Lloyd and Marques,1988). Therefore, each month (after five measure-ments) collectors were randomly relocated withinall subplots during the period of ‘single eventmeasurements’, i.e. between 1996–1997.

Stemflow was measured for 15 randomly selectedtrees in each subplot. Collars, constructed from 8 mmthick black polyethylene plastic, were sealed to thestems in an upward spiral pattern and the waterdiverted into bottle gauges on the forest floor. Theopening of each collar extended only about 2–3 cmfrom the trunk to avoid drips from the branches orleaves being collected by the collars. The amount ofwater, which drops in a diffuse pattern around rough-barked trees rather than adhering to and flowing downthe trunks, was considered throughfall. For practicalreasons, only trees with diameter larger than 10 cmwere selected for stemflow measurements. Wherepalm trees were present in the subplots, dependingon their frequency of occurrence one or two ofpalms per subplot were randomly selected. In the SPand HT plots, two palms were selected, in the LT fourpalms in total and in the FP three palms. Stemflowmeasurements were rotated once by installing collarsaround stems of new trees within the same subplots.

Horizontally or downwards inclined branches oftrees may not direct intercepted rainfall to the centreof the tree to be drained as stemflow. Therefore, theflat area of the tree crown was mapped by means of


vertical projections from the edge of outstandingupward branches to the forest floor. At least sixprojection lines were drawn for each tree. Thecrown area was found by integrating measured areasof each triangle. Stemflow is expressed as millilitre ofwater coming in over the horizontal area of the treecrown and flowing down along the tree trunk. Thus,each tree was considered as a single catchment area.Estimates for stemflow at subplot level were obtainedby multiplying the average stemflow per tree of aspecific diameter by the total number of trees withapproximately the same diameter within the subplot.Data on sampled trees with a diameter of about 10 cmwere used to extrapolate the information to trees withsmaller diameter. Average stemflow per subplot wasrelated to gross rainfall measured in the same subplotto determine the percentage of rainfall coming to theforest floor as stemflow.

Stemflow amounts from each rainfall event (ml)and from individual trees were correlated with themeasured projected area of the central crown of thetree concerned and with some tree characteristics suchas trunk surface area and bark texture. Consequently,the periphery of the trunks or surface area was deter-mined by measuring the tree trunk circumference atbreast height and the height of the tree trunk to the firstset of divergent branches. Trunk area was determinedby considering the tree trunk as being rectangular andapplying a correction factor to account for the conicalform of tree. Bark texture was classified according tothe roughness of the bark in a range from smooth tovery fibrous. Per plot, 40 hemispherical photographswere taken to estimate the gap fraction of the forestcanopy in the subplots where throughfall collectorswere installed. The photographic camera was installedhorizontally at the same height as the throughfallcollectors. The black and white photographs weredigitised with a scanner and analysed withthe Hemiphot program (Ter Stegee, 1994), whichcalculates the gap fraction of a forest and roughlyestimates the LAI from the forest cover. Althoughthe photographs were taken under covered sky withdiffuse sun light, they were corrected for the reflectedlight from leaves, branches and trunks. It is assumedthat the mean fraction of white pixels in thephotographs gives the best estimate of the gapfraction.

Most methods for the determination of the water

storage capacity of the forest canopy (C), whichhave been presented thusfar (Klaassen et al., 1998),mainly differ in the way of accounting for the drainagebefore the forest is completely saturated and thegradual saturation of forest layers with continuouslyproceeding evaporation. In the present study,C is theintercept of the regression of the estimated evapora-tion versus gross rainfall. Therefore, only single, high-intensity rainfall events of short duration (less thanone hour) were used. Late afternoon and night rain-falls were preferred, representing rainfalls underconditions of a low moisture deficit. In total, 30–40events were selected for each ecosystem. Freethroughfall is considered as an important parameter,among others, in studies on nutrient cycling, since itmay represent the fraction of throughfall which is notinvolved in washing out dry deposition, exudates andreleased nutrients from leaves, branches and trunks.Free throughfall in the forests may be taken as the gapfraction (ps). Therefore, it was estimated for eachforest from the set of digitised and scanned blackand white photographs taken in each subplot andfrom the regression coefficient of throughfall versusgross rainfall (pt), using data from small storms,which were insufficient to saturate the forest canopy(Gash and Morton, 1978). About 14 small storms wereselected for each forest to determine the value of pt.

Manual measurements of rainfall, throughfall andstemflow were carried out on weekly basis fromDecember 1993 to February 1996, without movingcollectors. During these years, some measurementswere carried out on event basis during periods withlimited or no rainfall (mostly in the dry season). Dailymeasurements were performed from February 1996 toAugust 1997. Readings were made early in the morn-ing (around 08:00 local time) or after the storm. Perso-nal observation in the plots showed that the uppercanopy dried out within 6 h after rainfall, if thisevent occurred during the daytime or part of it.Therefore, the criteria to separate the single eventswas that only those events where no rainfall occurredin the previous 6 h were considered. In those caseswhere two or more events occurred during the nighttime, since evaporation during the night is expected tobe low, they were considered as a single event. Mostanalysed single events in this study are events,preceded by a dry period of more than 10 h.

Evaporation (E), which is taken here as the total


amount of water intercepted and evaporated from theforest canopy, is calculated on event basis from thedifference between gross rainfall (Pg) and net rainfall(throughfallTh plus stemflowSf) of the single rainfallevents. Cumulative evaporation during the rainfallevents (Ew) is calculated from the difference betweengross rainfall and net precipitation plus the waterstored in the forest canopy after rainfall ceased (S),deduced from the estimated evaporation of singleevents minus the storage capacity (C)

Ew � Pg 2 �Th 1 Sf 1 S� �1�To avoid negative values in Eq. 1, although

they may reflect those events where the forestcanopy does not reach saturation, the value forSwas inferred from the single rainfall events whereE was larger thanC.

Statistical analyses were made for the entirecollected data. However, for the assessment of theeffect of rainfall sizes on net rainfall and resultingevaporation, measurements of single rainfall eventsare required. Therefore, throughfall percentages rela-tive to gross rainfall and evaporation were calculatedfrom these single events. Moreover, to determine theeffect of forest structure on throughfall percentage andevaporation, we used only data from single rainfallevents collected during the period of weeklymeasurements when collectors were not relocated.

4. Results

4.1. Forest structure

Average values of measured variables of foreststructure in each forest ecosystem studied arepresented in Table 1. The largest mapped crown

area was 62.2 m2 and the smallest 2.4 m2, both inthe high-terrace plot. For the trunk surface area, thecorrection factor of 0.5 was applied to trees withdiameter larger than 0.1 m. The apparent inconsis-tency in the flood plain data, of a large tree crownarea in combination with the lowest tree trunk surfacearea, is mainly due to the abundance of small treeswith a diameter of about 0.1 m. The data on the esti-mated gap fraction of the canopy and the free through-fall coefficient show that considerable differencesexist between forests, the largest values beingobserved in the sedimentary plain. Moreover, theleaf area index appeared to decrease from the sedi-mentary plain to the flood plain of the River Caqueta´.The estimated LAI values are within the range ofvalues found for similar forests in Brazilian Amazonia(Roberts et al., 1996; Klinge et al., 1975). Thedecrease in gap fraction and increase in LAI towardsthe flood plain is in line with the results from Duiven-voorden and Lips (1995), who observed that litter fallin the flood plain of the River Caqueta´ is highest whencompared to the other ecosystems in the area.

4.2. Rainfall characteristics

Gross rainfall above the forest canopy did not varyconsiderably within plots. On average, the differencesin the amounts of gross rainfall between subplots inthe SP were 5.5% (5.6) with n� 3; whereas in theHT these were 3.7% (3.5), in the LT gauges 3.7%(^3.0) and in the FP 3.1% (2.7), all with n� 2:Rainfall distribution differs between plots whenexamining separate storms, although annual totalsare rather similar. During the measurement period(1992–1997), the mean annual rainfall at the AWSwas about 3400 mm y21 and the average period withrainfall was 616 h y21. In total, 1584 rainfall events


Table 1Forest structure characteristics of four forest ecosystems in the Middle Caqueta´, Colombian Amazonia

Parameter Sedimentary plain High terrace Low terrace Flood plain

Crown area (m2) 9.5 ( 7.3) 17.8 ( 14.0) 9.8 ( 7.2) 12.1 ( 8.3)Stem trunk area (m2) 7.2 ( 4.6) 8.2 ( 5.2) 8.4 ( 5.4) 5.4 ( 4.6)Gap fraction (%) 16.8 ( 2.4) 15.4 ( 3.4) 11.7 ( 1.5) 8.2 ( 1.5)Free throughfall coefficient, pt 0.59 0.52 0.49 0.27LAI 4.4 (^0.7) 4.9 ( 0.8) 5.6 ( 0.6) 6.6 ( 0.4)Forest storage capacity (mm) 1.16 1.28 1.32 1.55


Fig. 2. Rainfall characteristics at the research site in the Middle Caqueta´, Colombian Amazonia.(August 1992 to August 1997)


Fig. 2. (continued)

were recorded at the AWS in Pen˜a Roja, with stormsranging from 0.2 to 161.6 mm and lasting between20 min and 13 h. During the total period, 37% of theincident rain fell in single showers of less than 2 mmand 92.3% of these showers contributed with less than30 mm. Rainfall intensity, calculated for the totalperiod with some rainfall, averaged 5.46 mm h21

with a maximum of 78.16 mm h21 (Fig. 2a). Mostshowers (63%) fell during the afternoon and at night(Fig. 2b) and 56% of these storms fell in less than 1 h(Fig. 2c). Monthly rainfall distribution during the five-year period shows that there was a slightly drierperiod from December to February (Fig. 2d). Compar-ing our data on five years rainfall with data fromearlier years in the Middle Caqueta´ (Duivenvoordenand Lips, 1995), rainfall characteristics appear to besimilar to the long-term average.

4.3. Throughfall

The variability of throughfall within a subplot waslarge, with the smallest variation in the FP forest,although differences in average values betweensubplots were small. The average coefficient ofvariation (CV) of individual gauges in each plot was0.285 ( 0.10) in the SP, 0.306 (0.07) in the HT,0.279 ( 0.09) in the LT and 0.225 (0.08) in theFP forest. The CV of the mean throughfall in eachsubplot was 0.062 (0.058) in the SP, 0.043(^0.05) in the HT, 0.046 ( 0.04) in the LT and0.047 ( 0.04) in the FP forest. As a general trend,for small rainfall events the value of the standarddeviation (std) of throughfall (expressed as a percen-

tage of mean throughfall) over gross rainfall variedmore than for major events. Furthermore, for someindividual throughfall gauges values exceeded grossrainfall (e.g. 29% of the individual gauge values in theSP were larger than gross rainfall, whereas in the HTthis was 30%, in the LT 27% and in the FP forest21%), but the average (of 60 and 40 gauges) wasalways lower than gross rainfall.

Throughfall was calculated as a percentage of grossrainfall for five different rainfall sizes and from thetotals of the measured daily gross rainfall andthroughfall during the study. Throughfall rangedfrom zero, with events below 2 mm, to 95% in stormslarger than 100 mm, but mean throughfall varied from50 to 93% depending on gross rainfall amounts andthe type of forest (Table 2). The calculated value oftotal throughfall relative to total gross rainfall rangedfrom 82 to 87% in the four forests.

Although empirical regression equations provideonly site-specific information, they may indicate atrend especially if explained variance is high. Therefore,


Table 2Throughfall percentages of daily gross rainfall for 5 storm classes, in four forests in the Middle Caqueta´, Colombian Amazonia. (SP)sedimentary plain, (HT) high terrace, (LT) low terrace, (FP) flood plain, (std) standard deviation of the means and (n) number of events

Rainfall ranges Throughfall %

SP HT LT FP

(mm) % std n % std n % std n % std n

, 5 58.7 11.4 41 56.2 12.1 34 52.3 9.6 32 47.4 13.2 275–20 81.4 6.3 78 80.5 5.6 68 79.8 6.8 71 74.5 6.9 5720–40 88.9 2.8 39 87.9 2.6 36 87.7 3.4 41 83.0 2.5 3240–80 90.6 2.1 19 90.0 1.9 19 88.8 2.5 17 84.6 3.1 19. 80 92.8 – 1 92.2 0.8 3 92.4 1.0 2 88.5 1.2 5Total 87.2 2.4 178 86.7 2.4 160 85.8 1.2 163 81.9 1.0 140

Table 3Regression parameters of throughfall versus gross rainfall in fourdifferent forest ecosystems in Colombian Amazonia. (se) standarderror of regression coefficient (Note: The equation for linear form isT � a 1 bPg; whereT is throughfall amount andPg is gross rainfall(mm))

Landscape unit a b se R2 n

Sedimentary plain 21.02 0.926 0.003 0.99 102High terrace 21.02 0.918 0.003 0.99 97Low terrace 21.07 0.906 0.004 0.99 97Flood plain 21.48 0.887 0.003 0.99 84

regressions of throughfall versus gross rainfallwere computed from single storms for each forest(Table 3). Average throughfall per plot was highlycorrelated with gross rainfall in all forests (Fig. 3).ANOVA analysis showed that the ratio of meanthroughfall over gross rainfall in the FP forest wassignificantly different from the other forests (at 95%level).

4.4. Stemflow

Large differences were observed in the amount ofstemflow of individual trees and among subplots. Ingeneral, however, the contribution of stemflow to netrainfall was very low. The average CV in each plotwas 0.295 ( 0.12) in the SP, 0.207 (0.12) in the HT,0.323 ( 0.20) in the LT and 0.303 (0.26) in the FP


Fig. 3. Average throughfall and its standard deviation (std) against gross rainfall in a forest ecosystem (Sedimentary plain) in the MiddleCaqueta´, Colombian Amazonia.

Fig. 4. Average stemflow and its standard deviation (std) against gross rainfall in a forest ecosystem (Sedimentary plain) in the Middle Caqueta´,Colombian Amazonia.

forest. The percentage of stemflow in all plots variedfrom 0.2 to 3.2% of gross rainfall. The total averagepercentage of stemflow relative to gross rainfall was0.85% ( 0.46) in the SP, 0.94 (0.51) in the HT, 1.45(^0.88) in the LT and 1.12 (0.56) in the FP forest.Differences are mainly due to the higher contributionof tree palms to the total stemflow per plot. Forpalms, high-capacity collectors (more than 35 l)were required to measure the incoming water. Insubplots with abundant palms, these palms producedabout 43% of total stemflow.

Upon rainfall, in all forests stemflow increasedvery gradually until a threshold of about 25 mmgross rainfall is reached (Fig. 4). However, valuestend to scatter with increasing rainfall. The rela-tionship between measured stemflow and grossrainfall could be described with a power function(Table 4). Some rainfall events smaller than 3 mmdid not produce stemflow in most plots, whichexplains the lower number of events (n) reportedfor stemflow regressions. For rainfall events withan intensity.5 mm h21, stemflow showed no clearrelationship with tree trunk area or bark texture�R2 � 0:3�: Nevertheless, there seems to be aninverse relationship between crown area and theamount of collected stemflow for each tree. Wealso observed that lower parts of tree trunks withfibrous bark texture were slowly wetted duringlong storm events, which points to high waterstorage.

4.5. Evaporation

Evaporation of intercepted water by the forest


Table 4Summary statistics for regressions of daily stemflow against grossrainfall, in four forest ecosystems in the Colombian Amazonia(Note: The power form isPs � c�Pd

g� wherec andd are the regres-sion coefficients for stemflow (mm))

Landscape unit Regression coefficient se R2 nc d

Sedimentary plain 0.0015 1.53 0.049 0.92 86High terrace 0.0020 1.467 0.038 0.94 92Low terrace 0.0029 1.423 0.035 0.95 87Flood plain 0.0031 1.325 0.050 0.91 73

Fig. 5. Evaporation against gross rainfall in the sedimentary plain forest ecosystem, Middle Caqueta´, Colombian Amazonia.

canopy is calculated by subtracting the measureddaily throughfall and stemflow from gross rainfall.Furthermore, it is related to gross rainfall character-istics and forest system parameters. Following netthroughfall trends, the percentage of evaporation rela-tive to gross rainfall varied from 6 to 100% in allforests, depending on rainfall size. Mean evaporation,expressed as percentage of total gross rainfall, alsodiffered between the forests: 11.84 (^2.4) in the SP,12.24 ( 1.2) in the HT, 12.92 ( 1.1) in the LT and17.15 ( 0.96) in the FP forest. For small storms (lessthan 2 mm), evaporation values were very close tothose of gross rainfall. For heavy showers, however,the relative value of evaporation became smaller(Fig. 5).

An assessment of the cumulative evaporationduring rainfall (Ew) was calculated with Eq. 1 forthe single rainfall events in each forest. Additionally,Ew was related to rainfall duration. The averageevaporation rate during the rainfall varied from 0.34to 0.68 mm h21 among the forests (Table 5) andEw

exhibited a linear relation with rainfall duration for allforests (Fig. 6). Negative values correspond to rainfallevents of short duration and low intensity, indicatingthat the forest canopy did not reach saturated condi-tions during such events.

Though climatic conditions are similar in theforests studied, there is a clear difference in amountsof evaporation when comparing similar rainfallevents. This implies that amounts of evaporationfrom the wet forest canopy did not only dependupon gross rainfall and climate conditions. Differ-ences in evaporation between these close-by forestsmay be related to differences in their structure. For

that reason, the ratio of evaporation over gross rainfallwas plotted against the mean forest cover (1-gapfraction) established for each forest. Fig. 7 indicatesthat there is an increase of evaporation from the wetforest canopy with increasing canopy cover. Thisfigure also shows an inflexion in the curve, i.e. a steep-ing from the low terrace to the flood plain, whichindicates that also other forest structural parameters(e.g. leaf surface characteristics) affected canopyinterception.

5. Discussion

Storage capacity values of the forests studied on thewhole resemble the values found by Ubarana (1996)in the reserves Vale do Rio Doce and Jaru Duke inBrazil, which were based on linear regressions ofthroughfall against gross rainfall. However, Ubaranaconcluded that this method results in an overestima-tion of evaporative losses. Since our estimates of thecanopy storage capacity were based on specific eventsfor which it was assumed that evaporation was negli-gible, it might be that we somewhat overestimated thestorage capacity by neglecting evaporation during theselected events. The storage capacity of the floodplain forest is higher than the values commonlyreported in Amazonian rain forest studies. This canbe explained by the fact that most of the latterstudies were executed in so called “terra firme” forests(non-flooded ecosystem). As we studied a broaderrange of ecosystems, differences between parametersvalues reported here and those reported in otherstudies therefore should be interpreted in terms of


Table 5Statistics of evaporation from wet forest canopy in single rainfall events in four forest ecosystems in the Middle Caqueta´, Colombian Amazonia.Equations are of the formEw � e1 ft; (e in mm andf in mm h21) (Note: e and f are the regression coefficients for the linear function ofevaporation loss during rainfall events)

Forest Totalrainfall(mm)

Totalthroughfall(mm)

Totalstemf.(mm)

time(h)

Evap. duringrainfall (Ew)mm

Evap. ratefrom wetcanopymm h21

Ew versus rainfallduration. Regressioncoeffficient

R2 n

e f 178

SP 3273.8 2853.7 32.4 557.0 190.1 0.342 20.424 0.46 0.86 178HT 3293.0 2854.8 36.2 464.2 207.4 0.447 20.263 0.52 0.75 160LT 3158.4 2711.7 38.6 487.8 201.1 0.412 20.351 0.52 0.82 163FP 3120.9 2555.0 30.5 472.3 320.0 0.677 20.366 0.78 0.88 140

differences in forest structure between the ecosystemsstudied.

The free throughfall coefficients of our forests,derived from the data on selected small storms (pt),conform to the values found by Jetten (1996).However, these values which range from 0.27 to0.59 clearly differ from those derived from thescanned photographs (ps) and their applicationresulted in an overestimation of net rainfall rates inall forests. According to Ubarana (1996) this may beexplained by the waxy nature of the tree leaf surfacescausing the drops to splash off, which thus contributeto throughfall before the canopy is saturated. Addi-tionally, our values were based on the results forsmall showers producing throughfall and this mayhave influenced the estimation of the free throughfallbecause of the low frequency of such events (for theresearch sites only 7–10 events were registeredduring the total period). Photographs taken fromthe forest canopy under non-direct sunlight andwith covered sky can easily be analysed, the whitepixels reflecting the non-covered part. This estimatemay provide better results, especially if large

numbers of photographs have been taken andanalysed.

The range in our values for throughfall and stem-flow, expressed as percentage of gross rainfall, issimilar to the range in values reported in earlierstudies on rainfall partitioning in Amazonian rainforests (Table 6). This most probably also explainswhy such variability exist in the latter values, i.e. itis probably largely due to differences in rainfall char-acteristics and forest structure between the forestsstudied. That coefficients of determination for ourregressions are significantly higher than most valuespresented in the literature can be explained by the sizeof our data set, which is much larger than in earlierstudies. Our results also indicate that the partitioningof rainfall depends, among others, on the size of therainfall event. Moreover, it is clear from the relationbetween throughfall and storm size (Table 2) that thehigh CV of throughfall is the result of the large varia-bility in rainfall classes.

To define the total error (t.e.) of the mean through-fall as a percentage of gross rainfall, we applied theproposed formula for random relocation of n gauges


Fig. 6. Evaporation from the wet forest canopy (Ew) in relation to rainfall duration, in a forest ecosystem (sedimentary plain) in the MiddleCaqueta´, Colombian Amazonia.

by Lloyd and Marques (1988), although largerdiameter funnels were used in the current study

t:e: � s:e:�1 1��N=nmp � �2�

where s.e. is the standard error resulting from therandom relocations of collectors, expressed as thebest estimate of the standard error of mean throughfallin each collector, under the assumption that the speci-fic canopy structure is properly described byN(number of grid points) andm (the relocation ofcollectors). Based on the formula of Lloyd andMarques (1988), the arrangement of 60 funnels with23 relocations in the SP forest results in a total error inmeasured throughfall of 3.5% of gross rainfall,whereas in the other ecosystems with 40 gauges and23 relocations the error is 3.8% due to variation incanopy structure. These figures are lower than thosefound by other authors, which can be explained by thecontinuous relocation of our collectors. Nevertheless,our values for t.e. are larger than those found by Lloydand Marques (1988).

Whether throughfall percentage depends on stormsize remains to be established, as clearly stated byLloyd and Marques (1988). Accordingly, we investi-gated the effect of storm size on the variability ofthroughfall percentage by using only those singleevents that were measured during the 20-monthsperiod in which weekly data of throughfall andgross rainfall were collected. During that period, thesame methodologies were used with the exceptionthat gauges were not relocated. Fig. 8, given as anexample, shows that the variation in the ratio ofthroughfall from a single funnel over the average of20 funnels tends to decrease as storm size increases.This is a trend observed for most non-moving collec-tors but also for the relocated collectors, as statedearlier in this paper, which suggests that storm sizealso affect throughfall variability in our ecosystems.

We did not fully investigate the relation betweenstorm size and throughfall (expressed as percentage ofgross rainfall). Nevertheless, we conclude from ourresults that when the method of relocation of collec-tors is used to estimate this throughfall, it is essential


Fig. 7. Fraction of evaporation from gross rainfall as a function of forest cover fraction (FC� 1-gap fraction) in the four forest ecosystems.Middle Caqueta´, Colombian Amazonia.

that this relocation is preceded by sampling of a widerange of storms sizes with a fixed set of collectors, inorder to assess the combined effect of site foreststructure and rainfall characteristics on throughfall.Lastly, although litter fall in the ecosystems studiedexhibits some temporal dynamics (Duivenvoordenand Lips, 1995), no relation was observed betweenthroughfall percentage and litter fall. Such relationwas reported for a Bornean rain forest by Burghoutset al. (1988).

The variability of stemflow in mature tropical rainforest has been attributed to the high species diversity(Hutjes et al., 1990; Hertwitz, 1985) and this variabil-ity certainly is larger in tropical forests than in tempe-rate forest plantations (Lloyd and Marques, 1988). Inthe present study, this parameter was estimated fordifferent tree species with different diameter. Stem-flow values from this study ranged from 0.9 to 1.5% ofgross rainfall, which is within the range of valuespresented in other studies on similar forest types(Table 6). Although the contribution of stemflow tonet rainfall was very low, it probably causes an impor-tant input of solutes to the forest floor, concentratedaround the base of trees. Results suggest that littlewater was stored in excess of the storage capacity ofthe stem elements, as indicated by the very smallstemflow quantities collected once rainfall has ceasedor during small storms. This can be explained by thepresence of some tree trunks with hydrophobic bark(personal observations 1992–1997) and of bark withfibrous texture. Upon rainfall, tree species with thesecharacteristics exhibited significant stemflow, evenwithout being completely wet. However, once rainfallstopped, there was a sharp decline in stemflow.

We found static models to be capable of describingrainfall partitioning for the forests studied. The applic-ability of these models is most probably restricted tothe area and conditions during the period of research.

Though the observed relationships may contributelittle to the explanation of the hydrological processesat canopy level, the models nevertheless provide clearindications for the extent to which this partitioning iscontrolled by the parameters used. While linear func-tions produce better fits for correlation betweenthroughfall and gross rainfall, power functionsproduce better fits for such correlation with stemflow,in terms of the significance levels and standard devia-tion of residuals. The linear regression equations ofthroughfall versus gross rainfall fit most points andhave a high coefficient of determination in all ecosys-tems. Nevertheless, their application to very smallstorms (lower than 2 mm) results in negative valuesfor throughfall and they slightly underestimatethroughfall for very high-rainfall events, which illus-trates the limitation of regressions, which fit a curve toa set of data.

Throughfall and gross rainfall were highlycorrelated in all forests, which is probably due tothe similarity of our forests with regard to relevantsystem parameters. However, the correlation betweeninterception values and gross rainfall is less promi-nent, the coefficient of determination of the regressionbeing distinctly lower (R2 between 0.66 and 0.83). Inother words, throughfall percentages can be predictedwith a high accuracy based on data on rainfallamounts and characteristics, whereas for the predic-tion of interception other parameters, such as foreststructure, must be included.

Although our values are within the range of inter-ception values reported in other studies from theAmazon basin (Table 6), the values found for the FPand LT forests are rather high compared to thosereported in earlier Amazonian studies (Lloyd andMarques, 1988). This is in line with the higher canopystorage capacity of our forests as compared to thosedescribed in these other studies (e.g. Lloyd and


Table 6Partitioning of gross rainfall (percentages) in Amazonian rain forests

Location Forest type Throughfall % Stemflow % Evaporative loss % Reference

Venezuela Catinga 91 0.8–14 – Herrera (1979)Brazil Rain forest 80.2 – 19.8 Franken et al. (1992)Brazil Rain forest 87–91 1.8 8.9(3.6) Lloyd and Marques (1988)Brazil Rain forest 86–87 0.8–1.4 11.6–12.9(^5.9) Ubarana (1996)Colombia Rain forest 82–87 0.9–1.5 12–17 This study

Marques, 1988). Thus, even though climatic condi-tions for our forests were similar, forest interceptionvaried, due to differences in amounts and character-istics of the rainfall and in forest structure. The rela-tive proportion of evaporation from our forests wasalso higher than the values reported in these earlierstudies. This can be attributed to the higher gross rain-fall in our forests, at least when compared with themean annual value of 2500 mm reported from centralAmazonia (Leopoldo et al., 1987).

In many studies on rainfall interception, it isconcluded that leaf surfaces determine the intercep-tion storage capacity of woody plants (Hertwitz, 1985;Gash, 1979; Singh, 1977; Rutter et al., 1975). Studiesin the research plots, using destructive methods andderived regression equations for leaf biomass estima-tion (Overman et al., 1990; Alvarez, 1993), showedthat leaf biomass is higher in the flood plain forest(9.5 tonnes/ha) than in the other forests from whicha higher leaf surface area can be inferred. Therefore,the relatively high interception by the flood plainforest may be explained by its higher leaf biomass.

Cumulative evaporation during rainfall events (Ew)from our forests strongly depended on rainfall dura-

tion. Although not evaluated in this study, it mightalso depend on specific climatic conditions (e.g.wind speed). Additionally, a distinct relationshipseemed to exist with the forest cover fraction(Fig. 7). Although it should be realised that thenumber of forests studied is small and the relationshipis rather uncertain, it may serve for the estimationof evaporation by a forest for which measurementsare not available. Provided that climatic conditionsare similar, such estimations mainly rely upon anadequate estimation of the gap fraction or LAI.

6. Conclusions

Of the gross rainfall of about 3400 mm y21, mostfell in small showers during the afternoon and at night.The overall average rainfall intensity was about5 mm h21. These rainfall characteristics largelyexplain the partitioning of rainfall into throughfall,stemflow and ensuing evaporation in the forestsstudied.

Water fluxes in the forest canopy of four forestecosystems in western Amazonia have been quantified


Fig. 8. The ratio of single gauge throughfall over average throughfall from 20 gauges, as a function of gross rainfall size.

as a percentage of gross rainfall. Amounts of netprecipitation reaching the forest floor and evaporationfrom the wet forest canopy varied for the forestsstudied: the SP forest had the highest percentage ofthroughfall relative to gross rainfall and the FP forestthe lowest. The observed differences in throughfall,stemflow and evaporation can partly be attributed todifferences in forest structure (gap fraction). Theirrange is similar to the overall range in these para-meters as published in earlier studies from theAmazon basin, implying that the latter variabilitymay very well be connected with differences in foreststructure.

Results from the forests studied provide someinsight into the rates of evaporation from a wet forestcanopy and strengthen the understanding of thecontribution of forests to atmospheric moisture. Themean evaporation rate from a wet forest canopyduring rainfall events in the Middle Caqueta´, (Colom-bian Amazonia) was estimated at 0.47 mm h21 and itincreased with increasing forest cover. Moreover, thisstudy of throughfall, stemflow and evaporation in arange of forests demonstrates the relevance of foreststructure for the evaporation of rainfall intercepted bythe forest canopy and for the net precipitation reach-ing the forest floor. The results show that within thescope of this research forest structure can beadequately characterised by the gap fraction andLAI. These structural characteristics together withthe rainfall amount and rainfall duration are themain parameters determining rainfall partitioning inthe western Amazonian rain forests.

Acknowledgements

We are grateful to Dr John Gash from the UK Insti-tute of Hydrology and to Dr Sampurno Bruijnzeelfrom the Free University, Amsterdam for theirsuggestions and corrections of earlier drafts of thispaper. This work, forming part of a larger researchproject on water and nutrient cycling in ColombianAmazonia, was supported by the Tropenbos Founda-tion in Colombia and the Netherlands and by theColombian Institute for Science and Technology“Colciencias”. Collected data (four years data ongross rainfall, throughfall, stemflow and evaporativeloss from western Amazonia) is available on request

to the Tropenbos Foundation, E-mail: [email protected]

References

Alvarez, D.E., 1993. Composicio´n florıstica, diversidad, estructuray biomasa de un bosque inundable, en la Amazonia Colombi-ana. Ir dissertation, Universidad de Antioquia, Colombia.

Bosch, J.M., Hewlett, J.D., 1982. A review of catchment experi-ments to determine the effect of vegetation changes on wateryield and evapotranspiration. J. Hydrol. 55, 3–23.

Bruijnzeel, L.A., 1990. Hydrology of moist tropical forests andeffects of conversion: a state of knowledge review. UNESCO,Paris.

Burghouts, T.B.A., van Straalen, M.N., Bruijnzeel, L.A., 1998.Spatial heterogeneity of elements and litter turnover in aborneam rain forest. J. Trop. Ecol. 14, 477–506.

Duivenvoorden, J.F., 1995. Tree species composition and rainforest-environment relationships in the Middle Caqueta´ area,Colombia, NW Amazonia. Vegetatio 120, 1–113.

Duivenvoorden, J.M., Lips, J.M., 1995. A land-ecological study ofsoils, vegetation and plant diversity in Colombian Amazonia.PhD thesis, Landscape and Environmental Research Group,Faculty of Environmental sciences, University of AmsterdamTropenbos Foundation, Wageningen.

Eltahir, E.A., Bras, R.L., 1993. On the response of the tropicalatmosphere to large-scale deforestation. Q.J.R. Meteorol. Soc.119, 779–793.

Ford, E.D., Deans, J.D., 1978. The effects of canopy structure onstemflow, throughfall and interception loss in a young Sitkaspruce plantation. J. Appl. Ecol. 15, 905–917.

Franken, W., Leopoldo, P.R., Matsui, E., Ribeiro, M. 1992. Estudoda Interceptac¸ao da agua de chuva em cobertura florestalamazonica do tipo terra firme. Acta Amazonica, 12, 327–331.

Gash, J.H.C., Nobre, C.A., Roberts, J.M., Victoria, R.L. (Eds.),1996. Amazonian Deforestation and Climate Wiley, Chichester,UK (611pp.).

Gash, J.H.C., 1979. An analytical model of rainfall interception byforests. Q.J.R. Meteorol. Soc. 105, 43–55.

Gash, J.H.C., Morton, A.J., 1978. An application of the Ruttermodel to the estimation of the interception loss from Thetfordforest. J. Hydrol. 38, 49–58.

Herrera, R.A., 1979. Nutrient distribution and cycling in an Amazoncaatinga forest on Spodosols in southern Venezuela,Dissertation, University of Reading.

Hertwitz, S.R., 1985. Interception storage capacities of tropical rainforest canopy trees. J. Hydrol. 77, 237–252.

Holscher, D., de A Sa, T., Bastos, T., Denich, M., Fo¨lster, H., 1997.Evaporation from young secondary vegetation in easternAmazonia. J. Hydrol. 193, 293–305.

Hutjes, R.W.A., Wierda, A., Veen, A.W.L., 1990. Rainfallinterception in the Tai forest, Ivory Coast: application of twosimulation models to a humid tropical system. J. Hydrol. 114,259–275.

Jetten, V.G., 1996. Interception of tropical rain forest: performanceof a canopy water balance model. Hydrol. Process 10, 671–685.


Jordan, C.F., 1978. Stemflow and nutrient transfer in a tropical rainforest. Oikos 31, 257–263.

Klaassen, W., Bosveld, F.C., de Water, E., 1998. Water storage andevaporation as constituents of rainfall interception. J. Hydrol.212–213, 35–50.

Klinge, H., Rodriguez, W.A, Brunig, E.F., Fittkau, E.J., 1975.Biomass and structure in a central Amazonian Rain Forest. In:Golley, F.B., Medina, E. (Eds.). Tropical Ecological Systems.Trends in Terrestrial and Aquatic Research, Springer, Berlin,pp. 115–122.

Koppen, W., 1936. Das geographische System der Klimate.Handbuch der Klimatologie. Verlag Gebru¨der Borntraeger,Berlin (44pp.).

Leopoldo, P.R., Franken, W., Villa Nova, N., 1995. Real evapora-tion and transpiration through a tropical rain forest in centralAmazonia as estimated by the water balance method. For. Ecol.Manag. 73, 185–195.

Leopoldo, P.R., Franken, W., Salati, E., Ribeiro, M.N., 1987.Towards a water balance in central Amazonia region.Experentia 43 Birkha¨user Verlag, CH-4010 Basel/Switzerland.

Lesak, L.F.W., 1993. Water balance and hydrologic characteristicsof a rain forest catchment in the Central Amazon Basin. WaterResour. Res. 29 (3), 759–777.

Lettau, H., Lettau, K., Molion, L.C.B., 1979. Amazonian’s Hydrol-ogy cycle and the role of atmospheric recycling in assessingdeforestation effects. Mon. Weather Rev. 107, 227–238.

Londono, V.C., 1993. Ana´lisis estructural de dos bosques asociadosa unidades fisiogra´ficas contrastantes, en la region deAraracuara (Amazonia Colombiana). Ir Dissertation,Universidad Nacional de Colombia.

Longman, K.A., Jenı´k, J., 1990. Tropical Forest and itsEnvironment. Longman, New York.

Lloyd, C.R., Marques, F.A.de.O., 1988. Spatial variability ofthroughfall and stemflow measurements in Amazonian rainforests. Agri. For. Meteorol. 42, 63–73.

Molion, L.C.B., 1975. A climatonomic study of the energy andmoisture fluxes of the Amazon’s basin with considerations ofdeforestation effects. PhD thesis, University of Wisconsin,Madison.

Nobre, C.A., Sellers, P., Shukla, J., 1991. Amazonia deforestationand regional climate change. J. Climate 4, 957–988.

Nobre, C.A., Coauthors, 1996. The Large-scale Biosphere-Atmo-sphere Experiment in Amazonia (LBA). Concise experimentplan; LBA Science Planning Group, Staring Center-DLO, TheNetherlands.

Overman, J.P.M., Saldarriaga, J.G., Duivenvoorden, J.F., 1990.

Estimacion de la biomasa ae´rea en el bosque del medio Caqueta´,Colombia. Colombia Amazonica 4 (2), 135–147.

Roberts, J.M., Cabral, O.M.R., da Costa, J.P., McWilliam, A.L.C.,de A. SA, T.D., 1996. An overview of the leaf area index andphysiological measurements during ABRACOS. AmazonianDeforestation and Climate. Wiley, Chichester, UK (611pp.).

Rutter, A.J., Morton, A.J., Robins, P.C., 1975. A predictive model ofrainfall interception in forests. II Generalisation of the modeland comparison with observations in some coniferous andhardwood stands. J. Appl. Ecol. 12, 367–380.

Salati, E., Vose, P.B., 1984. Amazon Basin: a system inequilibrium. Science 225, 129–138.

Salati, E., Dall’olio, A., Matsui, E., Gat, J.R., 1979. Recycling ofwater in the Amazon Basin: an isotopic study. Water ResourcesRes. 15, 67–71.

Shuttleworth, W.J., Calder, I.R., 1979. Has the Priestley–Taylorequation any relevance to forest evaporation? J. Appl. Meteorol.18, 639–646.

Shuttleworth, W.J. 1988. Evaporation from Amazonian rain forest.Proc. R. Soc., London, B233, 321–346.

Singh, B., 1977. The effect of rainfall characteristics and post-wetting synoptic characteristics on evaporation rates from awetted hardwood canopy. Climatol. Bull. 21, 12–33.

Stewart, J.B., 1977. Evaporation from the wet canopy of a pineforest. Water Resour. Res. 13, 915–921.

Tanner, E.V.J., 1985. Jamaican Montane forests: Nutrient capitaland coast growth. J. Ecol. 75, 553–568.

Ter Stegee, H., (1994). HEMIPHOT: a programme to analyse vege-tation indices, light and light quality from hemispherical photo-graphs. Tropenbos Documents 3. The Tropenbos Foundation,Wageningen, The Netherlands.

Ubarana, V.N., 1996. Observations and modelling of rainfallinterception at two experimental sites in Amazonia. In: Gash,J.H.C., Nobre, C.A., Roberts, J.M., Victoria, R.L. (Eds.).Amazonian Deforestation and Climate, Wiley, Chichester, UK.

Vitousek, P.M., Denslow, J.S., 1986. Nitrogen and phosphorousavailability in treefall gaps of a lowland tropical rain forest.J. Ecol. 74, 1157–1178.

Waidi, S., Wong, W.M., Douglas, I., 1992. Throughfall, stemflow,overland flow and throughflow in the Ulu Segama rain forest,Sabah Malaysia. Phil. Trans. R. Soc. London. B. 335, 389–395.

Wright, I.R., Gash, J.H., da Rocha, H.R., Shuttleworth, W.J., Nobre,C.A., Maitelli, G.T., Zamparoni, C.A., Carvalho, P., 1992. Dryseason micrometeorology of central Amazonian ranchland.Q.J.R. Meteorol. Soc. 118, 1083–1099.
