|Home | About | Journals | Submit | Contact Us | Français|
Oil palm flowering and fruit production show seasonal maxima whose causes are unknown. Drought periods confound these rhythms, making it difficult to analyse or predict dynamics of production. The present work aims to analyse phenological and growth responses of adult oil palms to seasonal and inter-annual climatic variability.
Two oil palm genotypes planted in a replicated design at two sites in Indonesia underwent monthly observations during 22 months in 2006–2008. Measurements included growth of vegetative and reproductive organs, morphology and phenology. Drought was estimated from climatic water balance (rainfall – potential evapotranspiration) and simulated fraction of transpirable soil water. Production history of the same plants for 2001–2005 was used for inter-annual analyses.
Drought was absent at the equatorial Kandista site (0°55′N) but the Batu Mulia site (3°12′S) had a dry season with variable severity. Vegetative growth and leaf appearance rate fluctuated with drought level. Yield of fruit, a function of the number of female inflorescences produced, was negatively correlated with photoperiod at Kandista. Dual annual maxima were observed supporting a recent theory of circadian control. The photoperiod-sensitive phases were estimated at 9 (or 9 + 12 × n) months before bunch maturity for a given phytomer. The main sensitive phase for drought effects was estimated at 29 months before bunch maturity, presumably associated with inflorescence sex determination.
It is assumed that seasonal peaks of flowering in oil palm are controlled even near the equator by photoperiod response within a phytomer. These patterns are confounded with drought effects that affect flowering (yield) with long time-lag. Resulting dynamics are complex, but if the present results are confirmed it will be possible to predict them with models.
Oil palm (Elaeis guineensis) is by far the highest yielding oil crop with mean fresh fruit yield equal to 13 t ha−1 year−1 or to 10 t ha−1 year−1 palm oil under favourable conditions (Esteulle and Perennes, 2007). However, oil palm is sensitive to drought (Bredas and Scuvie, 1960; Maillard et al., 1974; Nouy et al., 1999), and it is thus strongly affected by the increasing frequency of climate anomalies such as El Niño causing drought in south-east Asia. Intra- and inter-annual yield variations are reportedly increasing (Henson, 1999). This observation is difficult to interpret because even the physiological causes of general seasonal variation of yield are largely unknown (Henson, 2006). Annual production is continuous but generally shows marked seasonal peaks which can neither be explained by carbon assimilation nor by phenology alone (Henson, 2006). Overall little is known about the environmental or endogenous control of periodic events in tropical trees (Rivera and Borchert, 2001).
Oil palm is a perennial, arborescent, monocotyledonous crop. Plants have a single woody stem without secondary growth. A single apical stem meristem produces new phytomers in regular succession consisting of an internode, a leaf and a male or female inflorescence on its axil. The development duration of a phytomer beginning with leaf initiation and ending with maturity of its axillary female inflorescence is around 4 years (Corley, 1977). Vegetative growth and development are continuous and constant under favourable conditions despite strong seasonal variations of reproductive growth (Corley and Tinker, 2003; Henson, 2007). It was therefore suggested that vegetative growth and development of oil palm constitute priority sinks for assimilates (Henson, 2007).
Flowering and maturity dates show a marked yearly rhythm but the control of their periodicity is not well understood. The main yield components of oil palm are the number and the weight of harvested bunches per month (Corley, 1977; Corley and Tinker, 2003). They are determined by (in chronological order): sex ratio (female:total inflorescence number), fraction of aborted inflorescences, bunch failure (leading to absence of fruit set) and mesocarp:total bunch dry mass ratio (Corley, 1977). Given the long development duration (4 years) of female inflorescences and the resulting large number of infloresences having different developmental stage on a plant at any given time, the effects of climatic variability on yield components are complex. The effects of a drought period on yield components occur with long time-lags because drought-sensitive processes such as floral sex determination or early inflorescence abortion take place months or years before the maturity of a given bunch (Dufour et al., 1988; Caliman and Southworth, 1998). The same stress period may thus cause different responses on different phytomers (Jones, 1997) depending on their physiological ages at the time of stress. These phytomers in turn are not independent of each other because they depend on shared resources within the plant, probably causing complex feedbacks on the scale of the plant.
Large fluctuations in whole-plant assimilate sink–source relationships can be expected because of climate and seasonality in particular drought periods. While some adjustment processes may be rapid, particularly changes in assimilate source strength, some responses in phenology, organ growth rate and final organ size have long time-lag periods and may thus not be able to compensate for sink–source discrepancies in the short term (Jones, 1997). For coconut, biologically and structurally close to oil palm, such transitory imbalances are, in small part, buffered by carbohydrate reserves and, to a greater extent, by fluctuations in radiation use efficiency (Mialet-Serra et al., 2008). In this case, the magnitude of plastic responses differed among compartments: low for above-ground vegetative compartments, high for the reproductive compartment.
The present study attempted to relate seasonal and inter-annual variations in vegetative and reproductive growth, phenology and yield components to environmental variables while estimating the lag periods at which they respond to situations of sink–source imbalances. Adult populations of two genotypes reputed to differ in productivity and drought response were observed during 2 years at two sites in Indonesia, one of them prone to seasonal drought periods. The overall aim is to provide a conceptual basis to build a crop model for oil palm.
The study was conducted on two experimental plantations of the SMART Research Institute (SMARTRI, SMART Tbk.) located in Kandista Estate (Riau province, Sumatra island, Indonesia, 0°55′0″N, 101°21′0″E, 100 m a.s.l.) and Batu Mulia Estate (South Kalimantan province, Borneo island, Indonesia, 3°12′15·4′'S, 116°01′46·9″E, 15 m a.s.l.). Both sites have a tropical humid climate. In Kandista, the rainy season occurs between November to January, in Batu Mulia between March and May with a monthly mean rainfall of 240 mm month−1 (averages for 1993–2005 in Kandista and 2001–2005 in Batu Mulia). A drier season usually occurs from June to August in Kandista and from August to October in Batu Mulia. The dry season in Batu Mulia (67 mm month−1) is more pronounced than in Kandista. Mean daily global radiation (Rg) during the dry season (wet season) in Kandista and in Batu Mulia is, respectively, 18·9 MJ m−2 d−1 and 17·2 MJ m−2 d−1 (16·9 MJ m−2 d−1 and 16·1 MJ m−2 d−1); mean air temperature is 27·4 °C and 27·6 °C (27·0 °C and 27·8 °C), mean relative humidity is 79·0 % and 75·8 % (81·1 % and 81·5 %), mean vapour pressure deficit is 0·98 kPa and 1·25 kPa (0·84 kPa and 0·93 kPa) and mean evapotranspiration (Penman-Monteith) 4·5 mm and 4·1 mm (3·9 mm and 3·6 mm). Kandista is considered as a favourable (drought free) site for oil palm with regular but short water-deficit periods and a climatic water balance (CWB) not falling below 150 mm month−1 (see CWB in Fig. 1). Batu Mulia is characterized by a longer and sometimes severe dry season.
Two genotypes of Elaeis guineensis Jacq.were studied (hybrids 63 and 83). Both are tenera high-yielding hybrids and results of crosses between dura deli origin (fruits with thin mesocarp, thick endocarp and large kernel) and pisifera avros origin (fruits with thick mesocarp, small endocarp and small kernel) varieties. Genotype 63 has higher productivity and a supposedly greater drought resistance than 83 according to practioners. The genotypes were also chosen because of different crown characteristics (more profuse for 83).
The experimental plots were part of two larger, long-term genetic trials covering 30 ha in Kandista and 9 ha in Batu Mulia. Populations were 136 plants ha−1 in a 9·5-m equilateral, triangular pattern. Plants were 13 years old at the onset of the study (July 2006). The present experiment took advantage of the replicated block design already in place. Six plants (or replicates) per genotype were selected randomly in plots, located sufficiently far from the plot border to avoid border effects.
In Kandista the soil is a sandy loam (78·0 % sand, 11·1 % silt and 10·8 % clay) of homogeneous texture with >3 m depth. The volumetric soil water content at field capacity and at permanent wilting point is 0·17 and 0·06 m3 m−3, respectively. The soil in Batu Mulia is a silty clay loam soil (25·1 % sand, 44·1 % silt and 30·8 % clay) with 2 m depth. The volumetric soil water content is 0·48 m3 m−3 at field capacity and 0·29 m3 m−3 at wilting point.
For each site, fertilization was managed according to leaflet nutrient contents which is general practice in oil palm cultivation; other field management practices were uniform between both sites.
Photoperiod (PP) or day length was defined and calculated as the duration between sunrise and sunset including civil twilight, the time from when the centre of the sun is 6° below the horizon before sunrise until the centre of the sun is 6° below the horizon after sunset (Forsythe et al., 1995; Almorox et al., 2005).
To evaluate drought extent, two integrative variables were calculated: the CWB calculated as monthly rainfall minus potential evapotranspiration is the core indicator for the length of the growing period in agro-ecological zoning for rainfed systems (FAO, 1996). Potential evapotranspiration (mm month−1) was calculated according to FAO guidelines (Allen et al., 1998). The second indicator of drought was the FTSW simulated at a daily time step. It estimates the soil water available to the plant and, unlike CWB, takes into account cumulative effects of continuous drought such as depletion of the soil water reserve.
The model used to calculate FTSW was ‘EcoPalm 2008’ (http://ecotrop.cirad.fr) and will not be presented here in full because only its water balance was used. Only essentials of this routine are described in the following because it is an adaptation of a generic FAO model (Allen et al., 1998) described in more detail in Sultan et al. (2005) and in Mishra et al. (2008).
Two layers of soil were defined: a top layer (with a zS depth or thickness) and a deep layer (with a zD thickness) with the root front located at zS + zD. Based on observations in several trenches dug on each plot (data not shown) about 80 % of oil palm roots were present within 3 m depth in Kandista and 2 m in Batu Mulia.
FTSW was calculated by using the following equation defined by Sinclair and Ludlow (1986):
where TAWS and TAWD are the maximum available soil water in the surface and the deep layers, respectively. AWS(t) is defined as the daily water storage in the top layer of soil, AWD(t) as the daily water storage in the deep layer of soil.
In the top layer, AWS(t) is calculated as:
where AWS(t – 1) is the water storage in the top layer of soil calculated during the previous day; R(t), daily rainfall (mm d−1); Rf(t), run-off (mm d−1); ES(t), soil evaporation (mm d−1) calculated as described by Allen et al. (1998); and TrP is plant transpiration (mm d−1).
In the deep layer, AWD(t) is calculated as:
where AWD(t – 1) is the water storage in the deep layer of soil calculated for the previous day; Dr(t), the drainage (mm d−1) after run-off, soil evaporation and plant transpiration; and TrP is plant transpiration (mm d−1). Deep drainage was considered equal to the fraction of precipitation not allocated to runoff, soil water replenishment in the root zone and transpiration. This water balance does not take into account the water intercepted and stored by the canopy. Rainfall interception of oil palm canopies is <5 % for a daily precipitation of 30 mm and even smaller at higher rainfall intensity (Dufrêne, 1989).
Oil palm stems have no secondary growth and therefore have roughly constant diameter but variable height due to production of new phytomers. Stem structural dry mass growth rate (kgDM month−1) was assumed to consist of two components: (1) height increase due to addition of stem segments associated with apical production of new phytomers; and (2) density increase due to lignification as stem tissues age.
Apical stem mass growth was estimated from the stem height increment per month (number of new phytomers × internode length per phytomer), stem diameter and the empirical density of newly formed ‘wood’ (dry mass:wet volume ratio equal to 0·10 g cm−3). The dry mass increment of existing stem parts were assumed to undergo a gradual weight increase due to lignification equal to 0·00074 g cm−3 month−1 (Corley and Tinker, 2003).
The total number of fully expanded leaves, newly appeared leaves and pruned or broken leaves were recorded twice a month from July 2006 onwards. When leaves were pruned to facilitate bunch harvest (which is the general practice in oil palm cultivation) leaves were ranked up to 33 (in ascending rank from the crown top) and their morphology (i.e. their length and leaflet number) described. Dry mass of the petiole, rachis and the leaflets and leaflet number were measured. Individual new leaf growth rate (kgDM month−1) was assumed to be linear (function forced with its final dry mass) over 3 months – time of rapid growth after a leaf's appearance (on rank 1). To estimate the aggregate dry mass growth rates for new leaves growing during a given month, the monthly growth rate was calculated for all such leaves on the plant.
Anthesis (defined as spathe opening date of female inflorescences) and maturity (defined as harvest date for mature bunch) were monitored twice monthly. At maturity the following parameters were measured: the numbers of fertilized and unfertilized fruits per harvested bunch, the dry masses of the harvested bunch stalk, the spikelets, and the two major fruit compartments (mesocarp and nut composed of endocarp and kernel) and oil:mesocarp dry mass ratio. Growth rate (kgDM month−1) of each harvested bunch was assumed to follow a linear function (forced with the final mass of the harvested bunch) over 10 months, time of rapid growth before maturity (i.e. harvest). To estimate the aggregate dry mass growth rates for all bunches growing during a given month, the monthly growth increments was calculated for each of them.
Partitioning of above-ground dry matter production to fruits was assessed as harvest index (entire fruit dry mass production divided by total above-ground dry mass production).
Two-way analysis of variance (ANOVA using Statistix, version 8·1 software, Tallahassee, FL, USA) was performed to evaluate the effect of genotype (genotypes 63 and 83) and site (Kandista and Batu Mulia) on yield and growth parameters.
Given the fact that environmental effects on production involve long lag periods in oil palm due to the long development duration of a phytomer (Corley, 1977; Adam et al., 2005), the 22 months of detailed observations on plants were augmented with monthly records on bunch production for the same populations (plots) since 2001, obtained from the plantation's management. The resulting continuous data for 8 years were used to test hypotheses on effects of PP and FTSW on bunch production using linear regression analyses (Statbox, version 6·5, Grimmersoft) for these two environmental variables versus monthly yield. The duration of the phenological lag of the environmental effects on monthly number of harvested bunches was estimated by searching for the temporal shift between predictor and predicted variables giving the strongest effect in terms of slope coefficient.
Mean annual production statistics are shown in Table 1. Dry fruit yield was significantly higher in genotype 63 than in 83 and higher in Kandista than in Batu Mulia. Differences in yield were mainly due to bunch number and not to mean bunch weight which was similar for both genotypes.
After the 10th year after planting, corresponding to late 2003, regular seasonal peaks of production occurred in Kandista, resulting from seasonal concentration of flowering events (Fig. 1). At the same time production dynamics became synchronized between the two cultivars. In Batu Mulia, production dynamics of the cultivars were synchronized from early 2003 onwards, but no clear seasonal recurrent pattern was observed.
In Kandista, seasonal production peaks were observed early in the 2nd half of the year (August to October). This pattern was associated with the absence of marked dry and rainy seasons as shown by the CWB in Fig. 1; rarely, was the CWB negative in two consecutive months and when this did occur it was not during a recurrent time of year. In Batu Mulia marked minima of production occurred for both cultivars around August 2003, March 2004, March 2005 and November 2006 (Fig. 1). These temporal dynamics were associated with the occurrence of a marked dry season early in the second half of the year (i.e. August to October), as indicated by clusters of months having negative CWB. However, this dry season varied among years in intensity and duration and was completely absent in 2003. Neither the peaks nor the minima of production coincided consistently with dry or humid periods.
In Kandista, climatic conditions during the experimental period (from July 2006 to May 2008; Fig. 2) were similar to the average recorded for the previous 12 years. The beginning of this period was characterized by a 2-month drought spell (July and August 2006) followed by mostly abundant rainfall. In Batu Mulia, however, 2006 and 2007 were unusual years; 2006 having a severe dry season with only 51 mm from August to October and 2007 being a wet year with heavy rainfall in June and July (750 mm) followed by a mild dry season. Mean monthly vapour pressure deficit to which oil palm is known to be sensitive (Dufrêne and Saugier, 1993), fluctuated between 0·7 and 1·0 kPa in Kandista and 0·8 and 1·2 kPa in Batu Mulia except during the severe dry season of 2006 when it attained 1·7 kPa. In Batu Mulia, maximum temperature was around 33·2 in 2006, i.e. 1 °C higher than in 2007 whereas minimum temperatures were similar among years (22·5 °C). In Kandista, mean monthly FTSW dropped to 0·4 in August 2006 constituting a mild stress. In Batu Mulia, during the 2006 dry season, mean monthly FTSW dropped to 0·25 and remained for 3 months below 0·5 whereas it barely dropped to 0·5 during the 2007 dry season.
Plant and canopy developments were generally inferior at Batu Mulia, the site affected by a marked dry season, than at the more favourable site Kandista (Table 2). In July 2006 (the beginning of the experimental period), stem height was similar for both genotypes but slightly lower in Batu Mulia than in Kandista particularly for genotype 63 (8·2 m in Batu Mulia vs. 9·7 m in Kandista). Both genotypes had the same LAI in Kandista (around 5·1) but genotype 63 had slightly lower LAI than 83 in Batu Mulia (3·66 vs. 3·93, P < 0·05). The morphological components of canopy (leaf number, specific leaf area) were different among the genotypes, genotype 63 having a smaller number of leaves in the crown and higher specific leaf area but larger and heavier leaves and leaflets than genotype 83.
Phyllochron was not different between genotypes but significantly longer (P < 0·01) at Batu Mulia (20 d) than at Kandista (17 d; Table 3). Mean stem growth rate for the experimental period was identical for both genotypes and sites.
Monthly dynamics of the growth rate of bulk above-ground vegetative organs (stem + leaves) were near-identical for the two genotypes at both sites (Fig. 3C, D). In Kandista, only minor fluctuations over time were observed, among them a significant (P < 0·05) depression in September to October 2006 that coincided with the dry spell as indicated by CWB and simulated FTSW (Fig. 3A). A smaller depression was observed around December 2007, a wet period. The magnitude of these fluctuations did not exceed ±10 % of the mean which was 9 kg plant−1 month−1 for both genotypes.
At Batu Mulia, variations in vegetative growth rate were much greater (Fig. 3D). Rates dropped to 3–4 kg plant−1 month−1 for both genotypes in October and November 2006, associated with a severe dry season as indicated by CWB and simulated FTSW (Fig. 3B). During the subsequent dry season in August to October 2007 which was much milder, a slight depression in growth rates was also observed.
Overall temporal dynamics of CWB were more synchronized than FTSW with vegetative growth rates because changes in FTSW occurred 1 month later than changes in CWB, a result of the buffer effect of the soil water reservoir. Consequently, vegetative growth responses to drought were rapid, probably either related to the more rapid changes in the top soil layer or to atmospheric drought. The fall in growth rates in June 2007 was probably not related to drought (which set on only 2 months later) but to excess rainfall resulting in waterlogging (observed visually in the field; no data available).
During the experimental period of July 2006 to April 2008, yield components (Table 4) showed the same trends as observed for records between years 2001–2005 (Table 1) with inferior yield and bunch number for genotype 83 in Batu Mulia. Harvest index was similar for both genotypes – between 44 % and 48 %. Lower yields observed in Batu Mulia, the climatically less favourable site, were in part due to a higher rate of unfertilized fruits per bunch, in particular for genotype 83, whereas mean bunch dry mass was not significantly different between sites. Mean mature fertilized fruit dry mass was higher in Kandista for genotype 63 and higher in Batu Mulia for genotype 83 associated with a strong (P < 0·01) genotype × site interaction.
Monthly growth rates of reproductive organ dry matter (Fig. 4C, D) varied strongly at both sites and for both genotypes. In Kandista growth rates varied between 3 and 15 kg plant−1 month−1 for genotype 63 (mean: 8·5 kg plant−1 month−1) and between 4 and 13 kg plant−1 month−1 for genotype 83 (mean: 7·9 kg plant−1 month−1; Fig. 4C). Temporal dynamics of the genotypes had some similarities, e.g. a maximum at the beginning of the observation period and a minimum in mid-2007.
In Batu Mulia, reproductive growth rate (Fig. 4D) showed a similar magnitude of monthly variations but at a slightly lower level (means: 7·1 kg plant−1 month−1 for genotype 63 and 6·5 kg plant−1 month−1 for genotype 83). Again, temporal dynamics were similar for the two genotypes with maxima in mid-2007 and late 2007 and pronounced minima in late 2006 and early 2008. The number of female inflorescences was the main determinant of yield fluctuations at both sites and for both genotypes (data not shown). In contrast to vegetative growth rates, peaks or depressions of reproductive growth rate did not coincide consistently with drought level (FTSW) or PP minima (Fig. 4A, B).
Possible effects of PP on frequency of fruit bunch production were analysed for the Kandista site (years 2001–2005) where confounding effects of drought stress are unlikely with simulated FTSW never decreasing below 0·5. Regression slope coefficient of PP from the linear regression between mature bunch number and PP, determined for hypothetical lags between 1 and 36 months, gave an oscillating pattern that was nearly identical for both genotypes (Fig. 5A). Best correlations (r2 = 0·28 to 0·35, with P < 0·0001) were observed for lags of 9–10, 21–22 and 33–34 months. In all cases, the best correlations were negative as can be expected for a tropical and therefore short-day plant (r2 for the positive maxima of the slope coefficient varied between 0·15 and 0·20). The best correlations between mature bunch number (i.e. maturity) and PP were thus associated with a lag of 9–10 months plus 0, 1 or 2 years (corresponding to a lag for flowering of 3 months plus 0, 1 or 2 years). In the following analyses, for convenience it is assumed that the photoperiodic lag for flowering is 3 months, without discarding the possibility that it is actually 15 or 27 months
As can be expected from the variable intensity of dry seasons and comparatively irregular seasonal peaks of production observed at Batu Mulia (Fig. 1), the same analysis conducted for that site produced much poorer correlations between bunch number and PP (data not shown).
Regressing mature bunch number frequency against FTSW (Fig. 5B), however, resulted in significant correlations for a lag of 29 months (P < 0·05) and 35 months (P < 0·05). As may be expected, the positive correlations are physiologically meaningful, the negative ones being their mirror images at a shift of ±6 months because drought is a yield-reducing factor in oil palm (Hemptinne and Ferwerda, 1961; Caliman and Southworth, 1998).
The correlation of maturing bunch number with the FTSW observed 29 months earlier at Batu Mulia improved further when PP at a 9-month lag (the lag estimated for Kandista) was used as co-variable in a multiple linear regression analysis (Fig. 5C). The correlations combining a 9-month lag for PP and a 29-month lag for FTSW effects were highly significant (P < 0·01 for genotype 63 and P < 0·0001 for genotype 83), although they explained only a fraction of the observed variability (only 16 % for genotype 63 and 35 % for genotype 83).
This model was graphically applied to the available bunch production data from 2001 to 2008 (Fig. 6). For Kandista, where drought was considered to be a negligible factor, it was observed that each of the two annual minima of PP (a phenomenon that is specific to locations near the equator) was each associated with a peak of production. The peaks thus occurred in pairs except for 2005 where a single peak was observed. Coincidentally or not, the missing second peak in 2005 was associated with a lag of 29 months with the most pronounced among the many short dry spells observed at the site.
In Batu Mulia, each dry season was associated with a depression in mature bunch number 29 months later. The greatest depression (early 2005) was associated at a lag of 29 months with the lowest FTSW simulated for the time series. Conversely, only a very small dip in bunch production was observed in early 2006 associated with an exceptionally mild dry season 29 months earlier. Several peaks of bunch production coincided at a 9-month lag with PP minima but this association was less regular than in Kandista probably because of the dominance of minima presumably associated with drought.
This study used two climatic environments, characterized by different severity of dry seasons, to explain the variable patterns of phenology, growth and fruit bunch production of oil palm. The choice of observation periods (22 months, augmented with 5 years of historical production records on the same populations) and sites provided sufficient natural variability to develop causal hypotheses derived from parameter correlations. The choice of the two genotypes recommended by experts for their different productivity and assumed sensitivity to drought was less meaningful because the genotypes, despite different absolute levels of productivity and some morphological differences, responded similarly to the environment. Their similar growth and phenological dynamics therefore strengthen the results in terms of repeatability but provided little insight into genotypic differences.
Observations confirmed numerous previous reports (Bredas and Scuvie, 1960; Corley, 1977; Henson and Mohd Tayeb, 2004) indicating pronounced, regular, seasonal baseline rhythms of production dynamics in the absence of environmental stresses and the modification of these baseline rhythms when plants are affected by drought periods.
The seasonal rhythms of bunch production at the favourable site, Kandista, showed no clear pattern initially but became regular in 2003 when the plants were 9 years old (Fig. 6). Oil palm usually shows such rhythms from about 10 years after planting onwards (Corley and Gray, 1976). The production rhythms observed were mainly due to the number of fruit bunches harvested per month and thus a function of flowering events on female inflorescences. Such phenological seasonal rhythms in tropical perennial plants can be either related to recurrent climatic patterns or to photoperiodism. For tropical dicotyledon trees, Singh and Krushwaha (2006) distinguished among five flowering types: summer flowering (on foliated shoots), rainy-season flowering (on foliated shoots following significant rains), autumn flowering (on shoots with mature leaves), winter flowering (on shoots undergoing leaf fall), and dry-season flowering (on leafless shoots). Seasonal climatic patterns are sometimes responsible for these patterns, e.g. in the rubber tree where flowering seems to be synchronized by periods of high solar radiation intensity observed near the equator (Yeang, 2007). This author argues that photoperiodism is unlikely to be involved in such cases because day length varies little or not at all, although Rivera and Borchert (2001) reported that several tropical trees are sensitive to annual variations in PP of l <30 min. More recently, Borchert et al. (2005) postulated that, even at the equator, plants are able to sense the small seasonal changes in day and night durations by means of a mechanism called the circadian clock, an endogenous timekeeping mechanism based upon several interconnected negative feedback loops (Jackson, 2009).
Photoperiodic control of flowering at equatorial latitudes is thus physiologically possible and should result in two flowering peaks per year (Borchert et al., 2005) because of the peculiar biphasic pattern of day length prevailing there. This is what was observed at Kandista, the equatorial (0°55·0′N) and climatically favourable site. The present analysis and resulting hypothesis on the photoperiodic nature of the dual annual maxima of flowering in Kandista, however, represent simplifications of a complex system and require further confirmation. The linear-regression-based correlation of monthly bunch number with PP which indicated that the photoperiodic stimulus for flowering happens about 9, 21 or 33 months before bunch harvest did not take into account the non-linear response commonly observed. It was also assumed that oil palm is a short-day plant like most tropical plants (Rivera and Borchert, 2001; Kouressy et al., 2008) but would also have obtained significant (although less good) correlations with the opposite assumption which would result in a stimulus occurring 6 months earlier. The present model also assumes that the stimulus is effective at the phytomer scale and not transferable to other phytomers (that are necessarily n plastochrons earlier or later in development) which is an unproven hypothesis. Lastly, it is not necessarily the minima of day length that constitute the effective stimulus but any parameter of the day-length dynamics that the plant can theoretically perceive. Clerget et al. (2004) and Borchert et al. (2005) elaborated several alternative models that can be considered. The present results render photoperiodic explanations of flowering control in oil palm likely and should motivate further investigation into the question.
The observations at the drought-prone site of Batu Mulia suggested the presence of flowering peaks associated with minima of PP in accordance with the model developed for Kandista but this pattern was confounded by other more dominant sources of variation. Annual drought periods recurring with variable intensity were the most likely explanation. Drought intensity was estimated by CWB and simulated FTSW. The former is an established agro-climatic term (Benoit, 1977; Kar and Verma, 2005) commonly used in agro-ecological zoning (FAO, 1996) but has the disadvantage of not taking into account the cumulative effects of continuous drought such as depletion of the soil water reserve. FTSW considers this effect but on a hypothetical basis because profiles of water infiltration, deep drainage and extraction by roots were not measured and the model thus not validated in this respect. Despite this uncertainty we preferred using simulated FTSW as a reference for soil water resources because oil palm root systems are quite dense in the 0–1·5 m horizon (Jourdan and Rey, 1997a) resulting in water depletion in this horizon during drought. It is known, however, that not only soil water resources but also atmospheric drought affects gas exchange (Dufrêne, 1989; Cornaire et al., 1994) and growth of oil palm. Within the soil profile having roots, their main activity is located in the upper 80 cm of the soil and particularly in the surface horizon (Jourdan and Rey, 1997b). Water extraction occurs preferentially and rapidly in this zone (Nelson et al., 2006). Consequently, both the instantaneous parameter CWB and the cumulative parameter FTSW were considered in the analysis.
The present results suggest that drought affected vegetative growth and development very differently from that of the reproductive system. Leaf and stem mass gain, as well as leaf expansion rates, were affected in real time by drought (within the temporal resolution of sampling which was monthly). Phyllochron of oil palm has been reported to be very sensitive to water deficit followed by rapid appearance and expansion of leaves as the stress subsides (Corley, 1977; Nouy et al., 1999), a behaviour that has been termed developmental plasticity (Tesfaye et al., 2006).
In contrast, reproductive organ growth and its main developmental component, the frequency of appearance of fertile, female inflorescences, was affected by drought with an estimated lag of 29 months before bunch harvest. The seemingly irregular fluctuations of reproductive growth (Corley, 1977; Dufour et al., 1988) have been attributed to several drought-sensitive phases during the approx. 4 years of development from phytomer initiation to bunch maturity, each sensitive phase corresponding to a different yield component (Corley, 1977; Dufour et al., 1988; Adam et al., 2005). The main yield components of oil palm are bunch number per palm, bunch weight and oil content. Bunch number depends on sex differentiation (determined 30–32 months before bunch harvest) and rate of inflorescence abortion (9–11 months before harvest; Corley, 1976, 1977; Dufour et al., 1988). Bunch weight is determined by fruit number and finally by oil synthesis (3–4 months before harvest). Environment thus affects yield by modifying developmental events largely before the main reproductive growth processes take place. The present conclusion that strong drought effects on bunch number happen 29 months before harvest are supported by Dufour et al. (1988) who reported one of several sensitive phases to occur 30 months before harvest. Dufour et al. (1988) and Caliman and Southworth (1998) found a sensitive period around 10 months before harvest in Ivory Cost and the Lampung province in Indonesia. The two sensitive phases probably correspond to sexual differentiation (approx. 29 or 30 months before harvest) and abortion (approx. 10 months before harvest; Corley, 1976, 1977; Dufour et al., 1988). The second drought-sensitive phase may coincide with the PP-sensitive phase as estimated here, making their distinction difficult.
The long lag periods of drought stress effects on reproductive growth and development, combined with the apparent absence of such lag periods for vegetative growth, are bound to cause sink–source imbalances in the plant. Mialet-Serra et al. (2008) recently addressed this problem for coconut, hypothesizing that sugar reserves may buffer such imbalances. They found, however, that the main adjustment at the whole-plant level was variable total growth and, in particular, reproductive growth. The most marked period of high growth rate (high assimilate demand for structural vegetative and fruit growths) followed a severe dry season and was associated with humid conditions when photosynthetically active radiation is low. Remarkably, during this period of high demand for assimilates and comparatively low photosynthetically active radiation, storage of starch was high (data not shown), possibly indicating that assimilates were not limiting, associated with high rates of photosynthesis (data not shown), conflicting with the view that oil palm yields are mostly source limited (Henson, 2007).
The present results provide crucial elements for the development of a model for oil-palm development and growth. No model currently exists that captures the specificities of this plant shared in part with other members of the palm family such as coconut (Mialet-Serra, 2005; Mialet-Serra et al., 2008): regular successions of phytomers on a single axis that each potentially produces an inflorescence, a range of phenological adjustment processes to environmental events during specific sensitive stages of phytomer development along its multi-annual phenology, sink–source imbalances, to some extent, stored reserves and, to be confirmed by further studies, photoperiodic control of flowering resulting in latitude-specific, seasonal maxima of flowering and fruiting. Such a model would not only be of interest to oil palm agronomists and breeders but also encapsulate the physiological and developmental organization of a plant type that differs markedly from other cultivated plants such as cereals, legumes and dicotyledonous trees.
The authors would like to thank SMARTRI. We gratefully acknowledge the estate managers of Kandista and Batu Mulia for their logistic and technical support in the field without which this study could not have been undertaken and in particular, Ms Reni, Ms Rosna, Mr Rinto, Mr Pujianto and Mr Agus for their precious help with data collecting during field experiments. Lastly, we wish to thank PT SMART Tbk. for partial funding of this study.