The study was conducted in three ecosystems of the EAR: (a) Evergreen Lower Montane Forest (ELMF), northeastern Cordillera of the Andes; (b) Evergreen Piedmont Forest (EPF), northeastern Cordillera of the Andes; and (c) Evergreen Lowland Forest (ELF), Napo–Curaray, according to the Ecosystem Classification of Continental Ecuador (Ministerio de Ambiente del Ecuador (MAE), 2013). The study sites were located within and in the buffer zone of the Colonso–Chalupas Biological Reserve and the Lowland Forest in the Protective Forest of the Runashito Community, respectively (Figure 1). Mean annual temperatures and mean annual rainfall range from 17.9 to 24.2°C and from 3,229 to 3,986 mm, respectively, and elevation varies between 373 and 1,826 m.a.s.l. (Table 1). The ELF plots are classified as megathermal and hyperhumid with no water deficit, whereas the EPF and ELMF sites are classified as mesothermal and humid with little water deficit (Thornthwaite, 1948; Table 1).

A cluster of three 60 × 60-m permanent plots was installed in each type of forest according to the National Forest Inventory methodology (MAE and FAO, 2015). The three plots in the ELF were installed in an “L” shape. EPF and the Evergreen Lower Montane Forest plots were installed equidistant along the altitudinal gradient because in mountainous areas, with steep slopes, it is not achievable to install plots in an “L” shape (Casanoves et al., 2014). Each plot was divided into a “grid” of 20 × 60 m delimited by plastic tubes and nylon rope. All trees ≥ 10-cm diameter at breast height (DBH) were labeled (numbered and barcoded). The variables DBH and spatial location of each individual were recorded. Tree species were identified by a dendrologist either in the field or by comparing collected samples in Ecuador’s National Herbarium (QCNE). The project data will eventually be part of the “Red de Bosques Andinos” network.¹

A scientific name verification process was carried out using the fdiversity software (Casanoves et al., 2011) and the Plant List database version 1.1 (The Plant List, 2013). In addition, with the generation of a Python Script, the scientific names of the project database were verified with the trees registered in the Ecuadorian Amazon (Andino et al., 2019). The number of stems, basal area, and AGB (Eq. 1, 2 in Table 2) were then calculated per species and forest type.

We used allometric equations of Chave et al. (2005) to calculate AGB (Table 2). Equations were used for the Moist Forest (Eq. 1 in Table 2) and Wet Forest (Eq. 2 in Table 2); these equations require DBH and wood specific gravity (ρ) values for each tree.

Specific gravity values (ρ in g cm^–3), required for allometric equations, were obtained from Baker et al. (2004). When ρ was not available, the mean of 0.623 g cm^–3 (SD = 0.178) for tropical South America was used (Chave et al., 2009); this value is obtained from the Global Wood Density database (Zanne et al., 2009). A biomass importance value (BIV) was calculated as the average of the relative density, relative basal area, and relative AGB values (Torres et al., 2020).

As indicators of biodiversity, the total richness, species richness per unit area, Fisher α, Shannon, and Simpson indices were calculated. Total richness (S) was taken as the count of total species per unit area. Species richness was calculated through the Margalef Index, MI = (NS - 1)/ln (N), where MI = Margalef Index, NS = number of species, and N = total number of individuals (Margalef, 1957).

All measurements by ecosystem were calculated as the average of the data obtained by the three established plots. Standard deviations are also provided to understand variations among plots of the same forest type (Table 3).

To analyze the differences in height, DBH, and AGB between forest types and plots, we used non-parametric test Kruskal–Wallis as our data were not normal. In order to observe differences between forest types we applied Mann–Whitney post-hoc test. To assess differences in species composition, we calculated Bray–Curtis dissimilarity Index between all studied plots. A Pearson correlation matrix between environmental variables was developed to remove statistically redundant variables. All correlation values > 0.8 were considered strongly related (e.g., r > ± 0.8). Based on the selected variables, we tested mixed models [AGB = scale (var₁) +. + scale (var_n) + random effect (site)] and simple linear regressions to assess the relationship between AGB and abiotic factors, such as altitude, temperature, and precipitation. All statistical analyses were performed using R 4.0.5 (R Core Team, 2021).

In this study, the total species richness per 0.1 hectare was recorded in the Evergreen Lower Montane Forest (21.11 ± 4.15), in the EPF (27.22 ± 0.96), and in the ELF (22.69 ± 2.36). In relation to the biodiversity indices calculated, the highest values for all of them were found in EPF (892–1,195 m.a.s.l). The lowest values were found in ELMF (1,421–1,826 m.a.s.l.) for total richness (21.11) and Margalef Index (13.59) and in ELF for Fisher α (37.60) and Shannon–Wiener Index (3.78).

Trees are concentrated in the first diameter class (10–20 cm), with few trees with large diameters. The density of trees per hectare in the Evergreen Lower Montane Forest was 687, 773 in the EPF, and 672 in ELF (Table 3 and Figure 2).

Families with most species per hectare are Fabaceae (17), Rubiaceae (17), Moraceae (15), and Lauraceae (11) in the ELMF; Fabaceae (16), Moraceae (14), Lauraceae (13), and Rubiaceae (13) in the EPF; Fabaceae (16), Meliaceae (13), Moraceae (13), and Sapotaceae (10) in the ELF. The most abundant species in the ELM forest are Mabea standleyi (7%), Otoba glycycarpa (4.7%), Miconia sp. (4.6%), and Eschweilera coriacea (3.6%); in the EPF Pouteria sp. (3.7%), Pseudolmedia laevigata (3.6%), in the ELF Iriartea deltoidea (10.9%), Otoba glycycarpa (4.4%), and Grias neuberthii (4.1%) (Table 4).

Bray–Curtis dissimilarity Index maximum value can be found between ELF-1 and ELMF-3 (0.97) and minimum value between ELF-2 and ELF-3 (0.55). Close value to 1 means that the two sites hardly share species and close to 0 that the two plots have very similar composition. In general, lower values can be found between plots of the same site (Table 5).

We found a higher tree density in moderate altitudes (892–1,195 m.a.s.l.), EPF (773), and lower in the Evergreen Lower Montane Forest (687) and ELF (671), but no significant differences were found according to Kruskal–Wallis (p = 0.547). Between forest types, significant differences according to Kruskal–Wallis test were found in all the metrics studied: tree height, dbh, and AGB (Table 6 and Figure 2).

Within forest types, plots are more homogeneous in ELF (where only significant differences in height were found p < 0.001), followed by EPF (ns for dbh, p < 0.001 for height and p = 0.042 for AGB). In ELMF is where we found more differences among plots (p = 0.002 for dbh, p < 0.001 for height, and p = 0.0052 for AGB; Table 7).

The 10 most common species concentrate 51, 47, and 41% of AGB, respectively (Table 4). Otoba glycycarpa and Alchornea latifolia were the species with the highest BIV in the Evergreen Lower Montane Forest (1,421–1,826 m.a.s.l.); Spirotheca rosea and Pouteria glomerata in the EPF (892–1,195 m.a.s.l.); and Otoba glycycarpa and Iriartea deltoidea in the ELF (373–394 m.a.s.l.) (Table 4). The highest BIV is for Otoba glycycarpa in the Evergreen Lower Montane Forest (9.21%) and ELF (8.13%) and Spirotheca rosea (8.15%) in the EPF. Also, least abundant tree species contributed with a high percentage of the AGB, that is, Alchornea latifolia 11.64% (1,421–1,826 m.a.s.l.), Vochysia braceliniae 8.15% (892–1,195 m.a.s.l.), Guarea kunthiana 7.84% (373–394 m.a.s.l.).

As a result of the correlation matrix, we found that temperatures T (min, max, and mean) are highly correlated with altitude (> | 0.9|) and annual precipitation (P) with the wettest and driest month. From seven variables, we finally selected two: elevation and annual P. We scaled these independent variables and tested mixed models. We use scaled P, elevation, and richness independent variables as fixed effects and forest type as random effect [AGB = scale (elevation) + scale (annualP) + scale (richness) + random effect (site)], but none of them had a significant effect.

In order to show the positive relationship (although not significant) between elevation (or mean temperature as they are highly correlated) and biomass, in the next figure a linear regression is presented with total AGB per plot by site and elevation.

Our AGB values (ELMF: 310.26 ± 81.59 SD Mg ha^–1, EPF: 347.73 ± 90.38 SD Mg ha^–1, ELF: 377.39 ± 42.73 SD Mg ha^–1) (Table 3) are within the ranges of studies carried out in similar ecosystems. Biomass records for Evergreen Lower Montane Forest have been reported between 216.09 and 359.95 Mg ha^–1 in an altitude range from 1,421 to 1,826 m.a.s.l. (Nadkarni et al., 2004; Sierra et al., 2007; Unger et al., 2012; Girardin et al., 2014). Higher biomass values (270.37–447.08 Mg ha^–1) were reported for EPF values in altitude ranges between 892 and 1,195 (Moser et al., 2011; Unger et al., 2012; Torres et al., 2020; García et al., 2021). Biomass records for ELF values between 349.39 and 426.57 Mg ha^–1 in altitude range between 373 and 426 (Girardin et al., 2014; Cabrera Quezada et al., 2019; Tierres et al., 2020).

Significant differences were found in AGB between ELMF and the other two forest types. No significant differences were found between ELF and EPF; however, a trend can be observed of lower biomass as elevation increases (Figure 2 and Table 6). Future studies should focus on investigating specific gravity values for all species and developing local allometric models that include the height variable for accurate biomass calculations (Báez et al., 2015; Phillips et al., 2016).

Forest biomass variability could be linked to the interaction of abiotic factors, such as temperature, precipitation, and nutrient availability (Rutishauser et al., 2015); or factors that influence plant regeneration, such as landslides (Myster, 2020); or species-inherent factors, such as wood density (Keeling and Phillips, 2007). In the mixed models, none of these variables were significant, which casts doubt on the effect of these abiotic factors on biomass in our study plots. However, this may be due to the sample size. The observed trend of biomass increasing with decreasing altitude, or with increasing temperatures, is confirmed in other studies (González-Caro et al., 2020). This also is in agreement with the results obtained for differences between plots (Table 7). Those plots of the same forest type in which there is a greater difference in elevation (ELMF) are those where significant differences were found for all variables (DBH, height, and AGB) (Table 7). Furthermore, in the linear regression between aboveground elevation and altitude, although the relationship is not significant, there is a clear trend (Figure 3). The lowest significance values were found for precipitation, practically ruling out that this is a factor influencing the amount of biomass. This is because the study area does not have significant water deficits.

Most biomass (Figure 2) belongs to individuals that are between 10 and 20 cm in diameter. This corresponds to a negative exponential diametric distribution, commonly known as reverse J–shaped curve. This type of curve is typical of mature stages in low-disturbed forest (Imaña et al., 2011).

The use of mixed models for the present study is a good choice as our data are nested (plots are nested within forest types). The fact that none of the variables are significant may be due to the level of resolution of environmental layers. As we do not have specific meteorological data for the plots studied, we used climatic information from WorldClim, which offers these data at a resolution of 1 km², which may be too low for our study plots. In addition, WorldClim data were calculated from interpolation, including elevation (Fick and Hijmans, 2017), which makes the independence of the variables studied even less likely.

Ecuador has relatively low greenhouse gas emissions (Ritchie and Roser, 2020). However, the country made a strong commitment to reduce these within the framework of Nationally Determined Contributions (Samaniego et al., 2019). Two national strategies stand out to meet the objectives. The first strategy is focused on avoiding forest degradation through the national program “Socio Bosque,” where the state pays an economic incentive to forest owners for their conservation (Jones et al., 2020). The second strategy is focused on increasing carbon stocks through the national restoration program (Wiegant et al., 2020). Forest restoration is the mechanism to increase carbon stocks and counteract greenhouse gas emissions from deforestation and forest degradation (UNFCCC, 2008). The forest restoration is one of the priorities of the world environmental agenda because of its potential for reducing greenhouse gases emissions (Silver et al., 2000; Hawes, 2018). Active restoration helps accelerate the reestablishment of the original biomass (Wheeler et al., 2016; Hietz et al., 2019). For a successful restoration, it is necessary to consider native species with genetic variability (Di Sacco et al., 2021) and its “ecological niche” (Silvertown, 2004; Poorter, 2007). The restoration approach changes, depending on the ecosystem service to be restored. The objectives of restoration could be carbon storage, forests products, biodiversity, and hydrologic ecosystem services (Di Sacco et al., 2021); if the objective of the restoration program is to recover the original carbon stocks, species with potential for carbon sequestration should be used (Cerullo and Edwards, 2019) that maximizes the highest amount of biomass per type of forest and be based on a reference ecosystem that supports the restoration (Gann et al., 2019). This consideration is important for future restoration programs in Ecuador, and this study identified the species with the highest biomass by forest type. Forest restoration in ELM, EPF, and ELMF with a focus on carbon sequestration should prioritize planting of the 10 most significant native species.