4.1 Belowground C Stocks
 Several groups have reported fine root biomass values in TMFs of Hawaii [Herbert and Fownes, 1999]; Colombia [Cavalier, 1996; Sierra et al., 2007]; India [Sundarapandian and Swamy, 1996]; Costa Rica [Maycock and Congdon, 2000]; and Australia [Vance and Nadkarni, 1992]. However, only three groups have published information on belowground carbon stocks distribution along TMF elevational gradients. Studies from Borneo [Kitayama and Aiba, 2002], Southern Ecuador [Roderstein et al., 2005; Moser et al., 2008; Graefe et al., 2008a], and the present transect [Girardin et al., 2010] demonstrated that TMFs contain higher fine root biomass. In the Southern Ecuadorian Andes, fine root biomass increased from 2.68 to 11.27 Mg ha-1 from 1050 m to 3060 m elevation [Leuschner et al., 2007; Moser et al., 2008]. In two elevational transects from 700 to 3100 m on Mt. Kinabalu, Borneo, Kitayama and Aiba  also found a marked increase in fine root biomass with elevation from 5.20 to 14.40 Mg ha-1. Girardin et al.  reported an increase in fine root biomass with increasing elevation. However, unlike the previous work, these authors also reported a step change in fine root stocks occurring at the base of the cloud zone.
 Figures 2 and 3 provide interesting insights on soil and fine root C stock trends along the Kosñipata transect. Several reasons for the substantial investment of carbon in the growth of a large fine root system in TMFs have been proposed. Unfavourable soil conditions [Bruijnzeel and Veneklaas, 1998; Cavelier, 1992], and slow mineralisation and nitrification rates of plant litter in relatively young soils resulting in low availability of N at high elevations [Priess et al., 1999; Soethe et al., 2008] may contribute to higher belowground carbon stocks at high elevations. In a global review of nutrient cycling in moist tropical forests, Vitousek and Sanford, 1986 and Vitousek  found that upper montane forests cycle less N than lower montane forests. Conversely, we may expect to observe more P limitation in the younger, freshly weathered Andean soils than in lowland Amazonian soils [Quesada et al., 2011; Garzione et al., 2008; Tanner et al., 1998]. More recently, Fisher et al.  reported from leaf stochiometric data from the present elevation transect that N limitation increased and P limitation appeared to decrease with elevation, with colimitation at midelevation. Water shortage is unlikely to be a limiting factor for plant growth at tropical cloud forest elevations. However, whereas soil water content is not directly correlated to fine root carbon stocks, cooler temperatures, fog, and heavier rainfall interact to produce high soil water contents (up to 40%), inhibiting mineralisation and impeding fine root nutrient supply. These differences in soil conditions may explain the change in belowground carbon stocks observed from lowlands to midelevations and the importance of cloud cover incidence for the step change observed at the base of the cloud zone. Bloom et al.  suggest that plants respond to imbalances in resource availability (e.g., nutrients and/or water) by adjusting their carbon stocks so that the limitation for growth is equalized for all resources. Raich  further suggested that there is a belowground shift in the partitioning of carbon as materials taken up by fine roots (water and nutrients) become more limiting to growth than those taken up by leaves (sunlight and CO2). An increased investment in fine root material would facilitate nutrient foraging under low N supply, increasing plant growth and reproductive functions [Vitousek and Sanford, 1986; Cuevas and Medina, 1988; Priess et al., 1999; Aragão et al., 2009].
 Improved nutrient availability [Soethe et al., 2006], C availability, aeration and more penetrable soil structures [Metcalfe et al., 2002] concur to create a favourable environment for fine root growth in soil organic layers. Hence, fine root carbon stocks are greater in the soil organic matter. Further, soil organic matter depth increased significantly with elevation, with a step change at the base of the cloud zone (Figure 2b), contributing to the increase in fine root biomass. Recent data on fine root NPP and carbon stocks from a 1750 m plot in the present elevation transect corroborated with the data we present in this paper and confirmed the occurrence of a step change at the base of the cloud immersion zone [Huaraca Huasco, 2013], emphasizing the important role of cloud cover at that elevation. Ultimately, a decrease in temperature resulting in slower decomposition [Takyu et al., 2003] rates and increased root longevity [Graefe et al., 2008b] may explain the increase in belowground biomass observed above the cloud line. Nonetheless, we identified large spatial variation within the region. For instance, the replicated plots at approximately 3000 m indicated the spatial variation that we can expect to find at each elevation: soil organic layer depth on the mountain side (3025 m, 20.20±1.20 cm) was almost half of that found on the ridge top, (3020 m, 42.20±2.70 cm). The 3025 m stand was, on average, drier, less cloud-prone and received higher annual solar radiation than the 3020 m stand.
4.2 Fine Root Production
4.2.1 NPPFineRoot Along the Elevational Gradient
 To date, only a few studies have been conducted on production and turnover of fine roots in tropical forests [Cuevas and Medina, 1988; Ostertag, 2001], and only two studies have measured NPPFineRoot in a TMF [Roderstein et al., 2005; Moser et al., 2011; Girardin et al., 2010]. This lack of information reflects a serious limitation of previous studies on tropical montane forest productivity, which have assumed that aboveground NPP is a reasonable proxy of total productivity. Roderstein et al.  challenged this assumption by reporting a threefold increase in fine root productivity coupled with a decline in aboveground productivity along a transect of three plots in the Ecuadorian Andes (1050–3060 m), revealing a clear belowground shift in carbon allocation with increasing elevation.
 In terms of fine root productivity, our results do not corroborate with those from the South Ecuadorian transect [Roderstein et al., 2005; Soethe et al., 2007], highlighting the danger of extrapolating findings from individual transects to the Andes or tropical montane forests in general. The present study found that NPPFineRoot of all premontane and montane forest sites was comparable to that found in the least fertile soils of lowland Amazonia (Tapajós and Manaus in Brazil, Amacayacu and Zafire in Colombia) [Aragão et al., 2009], which may be explained by the low N uptake rates systematically found in TMF soils [Vitousek et al., 1983; Tanner, 1985; Kitayama et al., 2000; Soethe \et al., 2008; Fisher et al., 2013]. We observed a step change in productivity between most lowland and premontane plots at the base of the mountain (~1000 m) and no significant change in NPPFineRoot with elevation above 1000 m (ANOVA: F = 0.14, P = 0.99). If we concentrate our analysis on the plots located along our elevational gradient, from the Tambopata biosphere reserve to the Kosñipata valley, soil and air temperatures explained 66% and 77%, respectively, of the decline in NPPFineRoot in the TMF sites. However, the variation between our sites along the transect is no greater than the variation found between sites in lowland Amazonia on contrasting soils; soil properties are clearly an additional strong determinant of NPPFineRoot [Malhi et al., 2009; Aragão et al., 2009]. Finally, the lack of step-change in fine root production suggests that the step-change in carbon stocks is a direct result of the observed step-change in residence time, indicating that fine root carbon stocks increase in tropical montane forests as a result of an increase in residence time.
 In a review of methods used to estimate fine root production in forests, Hertel and Leuschner  highlighted that no consensus existed on how best to measure fine root productivity in forests [Majdi, 1996; Vogt et al., 1998]. Nonetheless, our estimates of NPP from data gathered using ingrowth cores and rhizotrons showed consistency between the two methods. This indicates the robustness of our findings, despite the use of different methods. Unlike rhizotrons, ingrowth cores entail substantial and continual disturbance of the soil as well as infrequent measurements. Conversely, rhizotrons are likely to result in an underestimation of fine root production after a few months as the rhizotron observation panel becomes saturated with fine roots. Some NPP components and processes are difficult to measure directly and were not included in our estimates. These include C in exudates from roots and carbohydrates transferred to symbionts (e.g., mycorrhizae) and parasites. Clark et al.  estimated that these elements can easily amount up to 20% of total NPP in tropical forests. For practical reasons, these terms are often incorporated into “root and rhizosphere” respiration, as they consist of labile carbon that is rapidly metabolised.
4.2.2 Seasonality of NPPFineRoot
 We report significant seasonality of fine root productivity at all elevations. With the exception of the highest sites (~3000 m), root production increased during the austral winter (dry season) at all sites. Huaraca Huasco et al.  and Girardin et al.  observed a seasonal shift in allocation between wood and fine roots, indicating an optimal investment according to seasonal changes in resource availability over the annual cycle. Those authors identified solar radiation as a key factor controlling the seasonal above to belowground shift in allocation. We concur with those findings by reporting an increase in NPPFineRoot when solar radiation is at its lowest (dry season) at all elevations, except at the highest elevation plots where below ground-productivity is prioritised at the height of the austral summer.
4.2.3 NPPFineRoot Per Diameter Class
 When plotting NPPFineRoot against diameter class, we expected to see a decrease in NPP with increasing diameter class. Here, fine roots <0.6 mm displayed small NPPFineRoot values, peaking at 0.6 mm and decreasing linearly thereafter (Figure 5). The finest roots maximise surface area and are the most efficient roots for extracting water and nutrients from the soil, providing a greater uptake potential. Nonetheless, soil penetrability is often a decisive factor controlling root production and it is possible that a root diameter of 0.6 mm represents the optimum balance between root strength and uptake potential. This result could simply reflect a real pattern, demonstrating that fine roots in the 0.5–0.6 mm diameter class are the most productive. However, we suggest that the observed peak is in fact a result of a sampling bias. As there is no fundamental physiological constraint on producing minute fine roots as small as 0.025 mm in diameter [Pallant et al., 1993], the most likely explanation is that we are observing a sampling bias, as sorting roots finer than 0.6 mm in the field is impractical. Further, it is possible that the rapid turnover rates of the finest roots [Gill and Jackson, 2000] lead us to underestimate their NPPFineRoot with three monthly sampling frequencies. Hence, we argue that the decline in NPPFineRoot recorded in roots <0.6 mm is likely to reflect mainly (1) an under sampling of finest roots and (2) the rapid turnover rates of roots finer than 0.6 mm. With this in mind, we used the relationship between root diameter class and NPPFineRoot to extrapolate the expected NPPFineRoot values for roots smaller than 0.6 mm and established that our estimates of fine root productivity were underestimated by an average of 31%. Fine roots (<2 mm) represent on average 87% of all roots collected from ingrowth cores (<5 mm). Thus, we estimated that fine roots (< 2 mm) are underestimated by approximately 31% if we do not account for undersampling of finest roots <0.6 mm diameter. By applying this correction, we obtain NPP values varying between 4.27±0.56 Mg C ha-1 yr-1 (1855 m) and 1.72±0.87 Mg C ha-1 yr-1 (3020 m). We suggest that this is an additional methodological consideration to take into account when estimating fine root production from field sorting of root samples.
4.2.4 Fine Root Growth Characteristics
 The present study provides the first observations of fine root area growth, fine root length growth, specific fine root length (SFRL) and specific fine root area (SFRA) production in tropical montane forests. For most of the fine root characteristics described in this study, we
 found evidence of a step change occurring in the premontane forest. Fine root area growth, fine root length growth, SFRL and SFRA showed evidence of a change in regime, increasing substantially from lowland Amazonia (e.g., fine root area, 1.20–2.40 km m-2 yr-1, Metcalfe et al. ) to premontane and montane stands (4.06–9.58 km m-2 yr-1) (Figure 6). Fine roots are longer and thinner in the TMF. Altering fine root morphology may provide an additional approach for plants to increase their nutrient uptake by ensuring close proximity between the root surface and low mobility nutrients (water is more readily available through osmosis than nutrients through diffusion and active transport) and increasing the volume of soil exploited per unit of biomass produced. These findings are consistent with the theory of resource allocation controlled by local soil conditions [Bengough et al., 2006]. The change in root morphology from lowland to montane forests may also be explained by soil structure: cortical cells of roots grown in dense soil are generally shorter and fatter than those grown in loose soil [Clark et al., 2003]. Finally, the effects of herbivory pressure on fine root growth characteristics remain largely unexplored as it is difficult to differentiate root mortality by herbivory and senescence [Hunter, 2008]. Studies carried out in controlled conditions on temperate species suggest a change in root anatomy with increasing herbivory pressure [Huber et al., 2005], and studies on insects feeding on aboveground plant parts often demonstrate a decline in herbivory pressure with elevation [Hodkinson, 2005] but very little is known about equivalent patterns for soil herbivores. Thus, longer and thinner roots in the montane plots may also reflect a decrease in root herbivory along the elevation gradient, although this remains to be investigated.
4.2.5 Soil Profile of NPPFineRoot and C Residence Time
 Soil profiles of NPPFineRoots (Figure 7) demonstrated a decrease in NPPFineRoots with increasing depth. Belowground biomass decreased with depth at lower elevations and remained constant at mid to high elevations due to the increased depth of the soil organic layer (Figure 2b). A deeper organic layer at high elevations and increased fine root longevity result in increased fine root biomass values within the cloud zone, estimated at 1500 to 3500 m (Figure 2a). Nevertheless, the depth distribution of fine root biomass has been reported as an exponential reduction with increasing soil depth, and most of the variation of fine roots can be explained by the concentration of nitrogen in the soil [Cavelier, 1992; Soethe et al., 2006]. We did not record this decrease, as the organic layer depth at high elevations reached up to 44 cm at high elevations and our 30 cm deep soil cores did not record fine root biomass of the mineral layer in most TMF sites. The observed decrease in NPPFineRoots with depth, combined with little change in biomass with soil depth at mid to high elevations suggested an increase in fine root residence time (the ratio of fine roots biomass to NPPFineRoots) with increasing soil depth. It is likely that cooler temperatures and lower nutrient availability of deeper soil layers contribute to increase fine root longevity [Gill and Jackson, 2000; Graefe et al., 2008a].
4.3 Fine Root C Residence Time
 Data on patterns of fine root residence time in TMFs are limited to the South Ecuadorian transect [Graefe et al., 2008a; Graefe et al., 2008b] and the present site [Girardin et al., 2010; this study]. These studies recorded an increase in fine root C residence time with increasing elevations, up to 3000 m. The present study's observations corroborate the findings of most studies comparing forest fine root residence time across latitudinal gradients: shorter fine root residence times were found at warmer mean annual temperatures [Vogt et al., 1986; Gill and Jackson, 2000]. From our results, we estimate that soil temperature is the key driver of residence time, with a decline in mean residence time of 0.17 years (about 3 months) for each degree Celsius rise in air temperature (slope = -0.17, r2 = 0.64, P < 0.05, n = 7). However, other factors that covary with temperature along this gradient may influence this relationship. A number of studies have found an increase in fine root residence time with decreasing nutrient (in particular N) availability [Nadelhoffer et al., 1985; Pregitzer et al., 1995; Aragão et al., 2009]. Here, we assent that adverse soil mineral conditions result in a longer residence time of fine roots C. Further, Godbold et al.  found that root longevity significantly decreased in highly acidic soils compared to less acidic soils in a temperate forest. As soil pH decreases significantly with increasing elevation along the elevational gradient, acidic soils may further explain an increase in fine root residence time. Thus, low temperatures, acidic soils, and low nutrient availability at high elevations are likely to be the dominant factors controlling fine root C residence time in TMF sites.