An important deficiency of the tropical forest data set on above-ground net primary productivity (ANPP) is the paucity of studies where requisite components of forest productivity have been measured at the same location. Missing data on above-ground biomass increment (ABI, which refers to the incremental growth of trees) and fine-litter production (leaves, fruit, flowers, small twigs, but excluding coarse woody debris) is particularly problematic as these are the two major components of ANPP. The fragmentary nature of the data is reflected by the fact that only 13 of 39 (33%) plots reviewed by Clark et al. (Reference CLARK, BROWN, KICKLIGHTER, CHAMBERS, THOMLINSON, NI and HOLLAND2001) and 8 of 104 (8%) plots reviewed by Malhi et al. (Reference MALHI, BAKER, PHILLIPS, ALMEIDA, ALVAREZ, ARROYO, CHAVE, CZIMCZIK, FIORE, HIGUCHI, KILLEEN, LAURANCE, LAURANCE, LEWIS, MARÍA MERCADO MONTOYA, MONTEAGUDO, NEILL, NÚÑEZ VARGAS, PATIÑO, PITMAN, ALBERTO QUESADA, SALOMÃO, NATALINO, SILVA, TORRES LEZAMA, VÁSQUEZ MARTÍNEZ, TERBORGH, VINCETI and LLOYD2004) had data on both major components of productivity. In an attempt to retain the geographic coverage and replication of data in analyses, researchers have proposed ways to infer missing data. Typically ratios or (more recently) fitted relationships between ABI and litter production have been used for this purpose (Bray & Gorham Reference BRAY and GORHAM1964, Clark et al. Reference CLARK, BROWN, KICKLIGHTER, CHAMBERS, THOMLINSON, NI and HOLLAND2001, Murphy Reference MURPHY, Lieth and Whittaker1975).

Although simple relationships have been forwarded to relate the two components of productivity, some researchers have been cautious in interpreting the generality of the relationships (Brown & Lugo Reference BROWN and LUGO1982, Clark et al. Reference CLARK, BROWN, KICKLIGHTER, CHAMBERS, THOMLINSON, NI and HOLLAND2001, Malhi et al. Reference MALHI, BAKER, PHILLIPS, ALMEIDA, ALVAREZ, ARROYO, CHAVE, CZIMCZIK, FIORE, HIGUCHI, KILLEEN, LAURANCE, LAURANCE, LEWIS, MARÍA MERCADO MONTOYA, MONTEAGUDO, NEILL, NÚÑEZ VARGAS, PATIÑO, PITMAN, ALBERTO QUESADA, SALOMÃO, NATALINO, SILVA, TORRES LEZAMA, VÁSQUEZ MARTÍNEZ, TERBORGH, VINCETI and LLOYD2004), largely because they have been established on a much smaller complete data set which may not be representative of tropical forests as a whole. Further, there has been scant discussion as to why such a simple relationship should actually exist. Nevertheless, such estimates of ANPP continue to be used in the ecological literature, in some instances without explicit mention of these limitations nor the procedures used to infer missing data.

Substantial emerging data on tropical forest productivity offer a valuable opportunity to re-examine the generality of the relationship between ABI and litter production. This has important implications particularly for studies that set out to quantify ANPP over large areas, for example, where data on components of productivity are missing or where intensive monitoring of components such as litter production is not feasible (Paoli & Curran Reference PAOLI and CURRAN2007). Here, we ask whether there really is a simple relationship between the two major components of ANPP that would justify indirect estimates of missing data (ABI or litter production).

Data were compiled for old-growth tropical forests (excluding flooded forests) from the synthesis of Clark et al. (Reference CLARK, BROWN, KICKLIGHTER, CHAMBERS, THOMLINSON, NI and HOLLAND2001) and updated with subsequent studies that adhered to methodological requirements for estimating ABI and litter production described therein. A ‘global’ relationship with poor predictive ability does not preclude the possibility that locally derived relationships may still allow reasonable estimates of missing data within more narrowly defined geographic or environmental regions. To explore this further, we also conducted an analysis with available data partitioned into subunits. Subunits were delineated geographically by landmass (continents and islands) and environmentally by elevation (i.e. above or below 500 m asl) to reflect differences in species composition between landmasses and structural characteristics and growth form between upland and lowland tropical forests.

The consolidated data set brought together information from subunits comprising 37 forest plots. These were: Islands (1) Upland Hawaii, 12 plots from a precipitation gradient on Maui (Schuur & Matson Reference SCHUUR and MATSON2001), an elevational gradient on Mauna Loa (sites 5 and 6 in Raich et al. Reference RAICH, RUSSELL and VITOUSEK1997) and a chronosequence of older soils from Laupahoehoe, Kohala, Kolekole, Kokee (Herbert & Fownes Reference HERBERT and FOWNES1999); (2) Upland Borneo, 12 plots from an elevational gradient established on Mount Kinabalu, Malaysia (Kitayama & Aiba Reference KITAYAMA and AIBA2002; plus Q, T, U ridge and U lower slope in Takyu et al. Reference TAKYU, AIBA and KITAYAMA2003); (3) Lowland Borneo, three habitats on different parent material (each an average across plots) west Kalimantan, Indonesia (Paoli & Curran Reference PAOLI and CURRAN2007); Continents (4) Lowland South America, six plots from a collection of neotropical studies undertaken at San Carlos terra firme and San Carlos caatinga (Cuevas & Medina Reference CUEVAS and MEDINA1986, Malhi et al. Reference MALHI, BAKER, PHILLIPS, ALMEIDA, ALVAREZ, ARROYO, CHAVE, CZIMCZIK, FIORE, HIGUCHI, KILLEEN, LAURANCE, LAURANCE, LEWIS, MARÍA MERCADO MONTOYA, MONTEAGUDO, NEILL, NÚÑEZ VARGAS, PATIÑO, PITMAN, ALBERTO QUESADA, SALOMÃO, NATALINO, SILVA, TORRES LEZAMA, VÁSQUEZ MARTÍNEZ, TERBORGH, VINCETI and LLOYD2004), BDFFP Fazenda Dimona (Malhi et al. Reference MALHI, BAKER, PHILLIPS, ALMEIDA, ALVAREZ, ARROYO, CHAVE, CZIMCZIK, FIORE, HIGUCHI, KILLEEN, LAURANCE, LAURANCE, LEWIS, MARÍA MERCADO MONTOYA, MONTEAGUDO, NEILL, NÚÑEZ VARGAS, PATIÑO, PITMAN, ALBERTO QUESADA, SALOMÃO, NATALINO, SILVA, TORRES LEZAMA, VÁSQUEZ MARTÍNEZ, TERBORGH, VINCETI and LLOYD2004, interior plots in Sizer et al. Reference SIZER, TANNER and KOSSMANN FERRAZ2000), Tapajós Brazil (control plot in Nepstad et al. Reference NEPSTAD, MOUTINHO, DIAS-FILHO, DAVIDSON, CARDINOT, MARKEWITZ, FIGUEIREDO, VIANNA, CHAMBERS, RAY, GUERREIROS, LEFEBVRE, STERNBERG, MOREIRA, BARROS, ISHIDA, TOHLVER, BELK, KALIF and SCHWALBE2002), Paragominas Brazil (J. Chambers and D. Nepstad pers. comm. in Clark et al. Reference CLARK, BROWN, KICKLIGHTER, CHAMBERS, THOMLINSON, NI and HOLLAND2001); (5) Upland South America, one habitat (average across plots) from Porce Colombia (Sierra et al. Reference SIERRA, HARMON, MORENO, ORREGO and DEL VALLE2007); (6) Lowland North America, three plots from Chamela, Mexico (Martinez-Yrizar et al. Reference MARTINEZ-YRIZAR, MAASS, PÉREZ-JIMÉNEZ and 1996); and Lowland Asia, one plot from Pasoh Malaysia (Kira Reference KIRA, Tomlinson and Zimmerman1978). For consistency, it was assumed that dry mass of vegetation was on average 50% C allowing productivity to be reported in the units Mg C ha⁻¹ y⁻¹.

Our synthesis demonstrates that the practice of using fitted relationships to estimate missing components of ANPP is unsound (Figure 1). While the global relationship between the two major components of ANPP (i.e. ABI and litter production) shows a general positive trend for tropical forest, it is not at all clear what the fitted relationship should be, and it is, at best, one with considerable variation and low explanatory power (Table 1). This is in contrast to findings of previous studies. Clark et al. (Reference CLARK, BROWN, KICKLIGHTER, CHAMBERS, THOMLINSON, NI and HOLLAND2001) reported a logarithmic relationship between fine-litter production and ABI (r² = 0.69, n = 13). Malhi et al. (Reference MALHI, BAKER, PHILLIPS, ALMEIDA, ALVAREZ, ARROYO, CHAVE, CZIMCZIK, FIORE, HIGUCHI, KILLEEN, LAURANCE, LAURANCE, LEWIS, MARÍA MERCADO MONTOYA, MONTEAGUDO, NEILL, NÚÑEZ VARGAS, PATIÑO, PITMAN, ALBERTO QUESADA, SALOMÃO, NATALINO, SILVA, TORRES LEZAMA, VÁSQUEZ MARTÍNEZ, TERBORGH, VINCETI and LLOYD2004) reported a linear relationship both for their data (r² = 0.79, n = 8) and the data set of Clark et al. (Reference CLARK, BROWN, KICKLIGHTER, CHAMBERS, THOMLINSON, NI and HOLLAND2001) (r² = 0.57) when the relationship was constrained to pass through the origin.

Figure 1. Relationship between fine-litter production (leaves, fruit, flowers, small twigs, but excluding coarse woody debris) and annual above-ground biomass increment (ABI) (forest wood production) in the units Mg C ha⁻¹ y⁻¹ for the complete tropical forest data set (n = 37 forest plots). Labels refer to subunits: lsa = lowland South America; usa = upland South America; lna = lowland North America; la = lowland Asia; uh = upland Hawaii; lb = lowland Borneo; ub = upland Borneo. Forest plots from ‘upland Borneo’ are highlighted by open circles.

Table 1. Explanatory power of the relationship between fine-litter production (leaves, fruit, flowers, small twigs, but excluding coarse woody debris) and annual above-ground biomass increment (ABI) (forest wood production) for predicting missing data on components of above-ground tropical forest productivity using a range of curve-fitting algorithms. Numbers refer to r² (P-value) and are only shown where n ≥ 5 plots.

Paoli & Curran (Reference PAOLI and CURRAN2007) reported a weaker relationship between the two major components of ANPP for a mature lowland forest of south-western Borneo (r² = 0.176, n = 30). Interestingly, they also found a much shallower slope for the relationship and hypothesized that the higher biomass increment in their study, relative to other tropical forests, may have been responsible for the discrepancy. They proposed that the empirical relationship between components of ANPP may be better described as a saturating function when data from high productivity forests are included. This could result from more efficient conversion of leaf production to tree biomass gain on more productive sites. However, the generality of this saturating function was not reinforced when we added extensive new data from upland forests of Borneo (Kitayama & Aiba Reference KITAYAMA and AIBA2002, Takyu et al. Reference TAKYU, AIBA and KITAYAMA2003) and Hawaii (Schuur & Matson Reference SCHUUR and MATSON2001). Curvilinear methods such as power functions did not explain much more of the variance in the data than a simple linear function (r² = 0.18 versus 0.17, respectively). Limiting the analysis to within subunits did not generally improve the explanatory power of the analysis, but there was one notable exception (Table 1). That is, there was a strong relationship between ABI and litter production for upland forest of Borneo (linear relationship r² = 0.749, Figure 1).

The use of fitted relationships to infer missing data represents a trade-off between increasing the replication of ANPP data across tropical forests and generating unrealistic estimates of productivity based on untested assumptions. Our updated analysis indicates that the trade-off is not justified except maybe for upland forests of Borneo where estimation of one component from the other might be possible. We therefore recommend that data sets only be used to address ecological questions (e.g. links between local climate and forest productivity) where litter production and ABI have been measured directly. It is worth emphasizing that the consolidated data set has a somewhat limited geographic coverage and is dominated by data from a small collection of tropical forests – one large group of plots from the species-poor oceanic islands of Hawaii, a single mountain and lowland forest from Borneo and a scattering of lowland sites in the neotropics. The poor geographic coverage of the reduced (but methodologically complete) data set considerably decreases analytical power for examining trends across tropical forests.

The lack of a simple relationship between litter production and ABI could either stem from genuine mechanistic processes that generate complexity in the real world or be an artifact of methodology used to estimate production. We discuss each of these in turn. Differences in species composition can affect the ratio between litter production and ABI. For example, lianas in some forest communities can inflate litter production with only minimal effect on basal area estimates (Gentry Reference GENTRY1983, Hegarty Reference HEGARTY1991). Phenotypic or interspecific traits such as growth form or allocation of growth to foliage and wood can also vary spatially presumably in response to selective pressures imposed by different climate conditions, soil types (Brown & Lugo Reference BROWN and LUGO1982, Harrington et al. Reference HARRINGTON, FOWNES and VITOUSEK2001, Herbert & Fownes Reference HERBERT and FOWNES1999, Paoli et al. Reference PAOLI, CURRAN and ZAK2005) or a combination of both (Kitayama & Aiba Reference KITAYAMA and AIBA2002). Leaf mass per area, leaf life span and photosynthetic capacity are modulated by climate (Wright et al. Reference WRIGHT, REICH, CORNELISSEN, FALSTER, GROOM, HIKOSAKA, LEE, LUSK, NIINEMETS, OLEKSYN, OSADA, POORTER, WARTON and WESTOBY2005) and the efficiency of converting leaf production into tree growth may vary in response to nutrient content of soils (Paoli et al. Reference PAOLI, CURRAN and ZAK2005). These are just some of the reasons why a simple, generalizable relationship between litter production and ABI might not be expected.

Alternatively, a number of methodological issues may obscure a simple relationship. In some instances in our consolidated data set components of productivity were derived from different years (Kitayama & Aiba Reference KITAYAMA and AIBA2002) or short-term estimates of litterfall production were reconciled against integrated estimates of ABI derived from decadal census periods (Malhi et al. Reference MALHI, BAKER, PHILLIPS, ALMEIDA, ALVAREZ, ARROYO, CHAVE, CZIMCZIK, FIORE, HIGUCHI, KILLEEN, LAURANCE, LAURANCE, LEWIS, MARÍA MERCADO MONTOYA, MONTEAGUDO, NEILL, NÚÑEZ VARGAS, PATIÑO, PITMAN, ALBERTO QUESADA, SALOMÃO, NATALINO, SILVA, TORRES LEZAMA, VÁSQUEZ MARTÍNEZ, TERBORGH, VINCETI and LLOYD2004). The lack of temporally concurrent sampling is an important shortcoming given that tropical forest productivity is known to vary between years in relation to temperature and drought (Clark et al. Reference CLARK, PIPER, KEELING and CLARK2003, Takyu et al. Reference TAKYU, AIBA and KITAYAMA2003). Leaf herbivory and seed and fruit predation can also reduce the biomass of these materials before they reach litter traps leading to underestimates of total litter production in some forests (Filip et al. Reference FILIP, DIRZO, MAASS and SARUKHAN1995, Janzen & Vázquez-Yanes Reference JANZEN, VÁZQUEZ-YANES, Gomez-Pompa, Whitmore and Hadley1991). Also in our data, locally relevant allometric relationships were not always applied to infer local forest tree biomass. Different allometric equations are known to produce biomass values that can differ greatly (up to three times) from each other and from real measured biomass (Araújo et al. Reference ARAÚJO, HIGUCHI and DE CARVALHO1999). The use of inappropriate allometries is therefore an additional factor potentially contributing to variability in our data used for analysis.

The results from the consolidated data set reported here do not bear good news for researchers interested in scaling up the findings from the small number of complete, empirically measured forests to estimate regional or global patterns of tropical forest productivity. It seems that there may be no shortcut that will allow ANPP to be estimated over large areas using only partial data with the possible exception of northern upland forests of Borneo. As such, we may still be some way off from generating necessary data from representative tropical forests to quantify productivity adequately and develop models that enable observed variability between locations to be explained.