Quantification of carbon sequestration in soils following pasture to forest conversion in northwestern Ecuador

Authors


Abstract

[1] We studied the changes in soil carbon contents when pastures are converted to either secondary forest or plantation forest in north-western Ecuador. At 40 sites within the region, paired pasture and forest plots were compared. We related the observed soil carbon concentrations, stocks, and changes (in the 0–0.25 m and 0.25–0.5 m layers) to land use history, climate, and soil characteristics. Variation in carbon concentrations over sites in volcanic soils could be well predicted for both pastures (R2 = 0.96) and forests (R2 = 0.93) on the basis of soil mineralogy, while for sedimentary soils, clearly less variation could be explained (R2 = 0.14 for pastures and 0.39 for forests). The dominant factor explaining changes in carbon stocks following pasture to forest conversion was pasture age. Forests, paired with pastures less than 10 years old, had on average 9.3 Mg ha−1 less soil carbon than the pastures, while forests paired with pastures between 20 and 30 years old had on average 18.8 Mg ha−1 more soil carbon and forest paired with pastures older than 30 years had on average 15.8 Mg ha−1 more carbon than the pastures. In this region, reforestation of old pastures will generally lead to an increase of soil carbon stocks. These results can be used for optimal site selection for carbon sequestration projects and for including soil carbon in the estimated benefits of these projects.

1. Introduction

[2] The terrestrial biosphere plays an important role in the global carbon cycle because of its large carbon stock, which is estimated at ∼654 Gt C for the terrestrial biomass [Watson et al., 2000] and ∼2344 Gt C for organic matter in the top 3 m of soils. Soils in the tropics are estimated to contribute about 44% to this soil organic matter pool [Jobbagy and Jackson, 2000]. This large pool of soil organic carbon (soil C) can respond to environmental changes like land-use change of which deforestation is the best-known example. Deforestation is presently estimated to contribute about 23% to the human-induced CO2 emissions, from which the majority (about 75%) originates from the aboveground biomass and the remaining 25% is attributed to the decomposition of soil C [Detwiler, 1986; Melillo et al., 1996; Houghton, 1999].

[3] Various studies exist on the impact of land use conversion on soil C, especially the effect of the conversion of tropical forest to agricultural land or pastures (see reviews by Detwiler [1986]; Veldkamp [2003]; and Post and Kwon [2000]). However, the effect on soil C stocks of conversion of agricultural lands or pastures into forests is much less studied and only very few were in (sub) tropical climates [e.g., Brown and Lugo, 1990]. Soil C changes measured after deforestation are not simply reversible, because they depend on factors like primary production, partitioning of carbon over the different stable and less stable pools and availability of limiting nutrients like, for example, nitrogen [Lugo et al., 1986]. In a review of changes in soil C following afforestation, Paul et al. [2002] concluded that on average, forested soils tend to loose carbon during the first 5 years. The rate of decrease in soil C stocks became slower in time, and around 30-years age, soil C levels had recovered to original levels. In a meta-analysis of soil C stocks and land use change, no significant change in soil C was revealed when analyzing the change from pasture to secondary forest [Guo and Gifford, 2002]. However, the number of pasture to forest conversions studied was only six.

[4] Most studies compiled in the above-mentioned reviews are based on a small amount of point data and few were conducted in tropical areas. Furthermore, the majority of these studies were not executed with the main objective to address changes in soil C stocks, which is why only few studies were well designed for this purpose. Although reviews are useful to deduct general trends and effects on soil C stocks and to formulate working hypotheses, they are insufficient to make estimates of regional carbon budgets [Tate et al., 2000] or for planning carbon sequestration activities. For these applications, regional studies are necessary that determine the relation between stocks and changes in soil C and the biotic and abiotic factors that control them within the region.

[5] Reliable estimates of land use change related fluxes of carbon, including changes in soil C stocks, are essential for national greenhouse gas inventories used in the Conferences of the Parties (COP) of the United Nations Framework Convention of Climate Change (UNFCC) [Watson et al., 2000]. Equally important is the quantification of changes in soil C stocks in afforestation or reforestation projects specifically designed to mitigate carbon emissions through the sequestration of carbon. In developing countries, such projects can be designed within the Clean Development Mechanism, as being defined in the COP agreements [UNFCC, 1998, 2001] and with specific objectives to support sustainable development, or through other market initiatives. For carbon sequestration projects, it is invariably necessary to calculate base line and project additionally, perform a cost-benefit analysis, and design a monitoring, accounting, and certification scheme [Brown et al., 2000]. For these assessments, estimates of aboveground carbon sequestration in biomass are necessary, and measurement of carbon sequestration in soils is recommended because the changes in soil carbon can be significant but costs to achieve sufficient precision could be high [Brown et al., 2000].

[6] The first objective of the present study was to quantify the impact of conversions of extensively managed tropical pastures to either secondary forests or plantation forests on soil C concentrations and stocks for a region in northwest Ecuador. The second objective was to identify biophysical variables (climate, soil characteristics, biomass) and management related variables that can be used to predict changes in soil C concentrations and stocks in this region. On the basis of a review of forest to pasture conversion [Veldkamp, 2003], we formulated the following working hypotheses: (1) Soil carbon content increases after conversion of extensively managed pastures to forests. (2) The amount of change in carbon stocks follows a pattern which can be predicted based on land use history, soil characteristics (pH, texture, mineralogy), and climate. To test these hypotheses, we collected information and soil samples from 40 paired pasture/forest sites in tropical northwestern Ecuador, a region that offers ample opportunities for pasture/forest comparisons in a range of climate and soil conditions.

2. Materials and Methods

2.1. Approach

[7] Experiments that monitor changes in soil C following land conversion require years or decades of measurements to provide conclusive results and were therefore not a feasible option. We used space-for-time substitution and measured soil C stocks in paired pastures and secondary forests or plantation forests across a wide range of vegetation age, climatic, and soil conditions. We measured soil C concentrations and stocks in the paired pasture and forest sites and their difference (dependent variables) and related these to potential independent edaphic (soil pH, texture, and mineralogy) and biophysical (elevation, slope, and landscape position, average annual temperature, aboveground biomass, and age of land use) drivers or their proxies (independent variables). A similar approach was followed by Powers and Schlesinger [2002] in a landscape scale study in Costa Rica.

2.2. Study Area

[8] The study area is located near the equator in the northwestern part of Ecuador, covering the whole of the province of Esmeraldas and the most northwestern part of the province of Pichincha, within the geographical coordinates of 80°05′W, 1°30′N (northwestern corner) and 78°40′W, 0°05′S (southeastern corner, Figure 1). The area is roughly delimited in the east by the Western Cordillera of the Andes, in the west by the Pacific Ocean, in the north by the country border with Colombia, and in the south by the province of Manabí. Elevation varies between sea level and 1600 m, corresponding to an average temperature between 26°C and 19°C. Yearly annual precipitation varies from 1000 mm near the city of Esmeraldas to over 5000 mm in the sub-montane area of the western Cordillera of the Andes. The study area is a continuation of the Colombian Chocó and is known as one of the world's hot spots of biodiversity [Myers et al., 2000].

Figure 1.

Study area with the sampling sites. Source of base map: Clirsen/Patra [1998].

[9] The soils can broadly be divided into two groups. The soils on the foot slopes of the Western Cordillera (most eastern part of the area) and some lower lying valleys are of volcanic origin, while the soils of the undulating coastal lowlands are developed on sediments [Ministry of Agriculture of Ecuador/Office de la Recherche Scientifique et Technique Outre Mer (MAG/ORSTOM), 1980]. The volcanic soils are relatively young soils developed on ashes with a mineralogy characterized by the presence of allophane. They generally are acid or slightly acid, and have high water retention, a low bulk density, a sandy or loamy texture, a base saturation under 35 mmol/100 g, and a medium fertility. The sedimentary soils are more developed, with a clayey or loamy clay texture. In the most humid areas, iron and aluminum oxides are present. Clay minerals are dominated by kaolinite, while in dryer areas, smectite clay minerals dominate. The following USDA Soil Taxonomy sub-orders can be found in the study area: Tropepts, Aquents, Orthents, Fluvents, Udalfs, Udolls, and Psamments [MAG/ORSTOM, 1980; Clirsen/Patra, 1998].

[10] Land use is very dynamic in the region with high deforestation rates due to timber extraction and conversion to agricultural land [Sierra and Stallings, 1998]. The estimated forest cover (primary and secondary) in the province of Esmeraldas is, depending on source, between 50% (source: Clirsen/Patra [1998]) and 73% (source: Instituto Ecuatoriano de Estadísticas y Censos (INEC) [1995]). The dominant agricultural land use is grassland for cattle grazing, followed by permanent crops such as oil palm, banana, cacao, plantain, and coffee. Small areas of annual crops exist, such as maize, rice, and cassava. Most pastures have been established after the cutting and/or burning of natural, in many cases intervened, forest. Stocking densities are low: on average around one animal per hectare. In case pastures are abandoned, rapid regrowth of secondary forest takes place. The area of forest plantations is limited (approximately between 7000 and 10000 ha) and is owned by a small number of companies.

2.3. Site Selection

[11] We selected 40 sites where pasture was found next to or very close, less than 1 km, to a secondary forest or forest plantation. Both forest and pasture plots had an area of at least 1 ha. We chose areas with soil and terrain conditions as similar as possible for the two plots within a site, allowing for pair-wise comparisons between pasture and forest. In 34 sites, a pasture plot was paired with a secondary forest plot, and in six sites a pasture plot was paired with a forest plantation plot. Sites were selected to include pastures and forests of different ages, allowing for reconstruction of carbon changes over time by means of chronosequences. All secondary forests and plantation forests were established after abandonment or conversion of former pasture. All pastures were established after cutting and/or burning of former forest. Different grass species were found over the 40 sites. Dominant grass species were Saboya (Panicum maximum), Gramelote (Axonopus scoparius), Estrella (Cynodon nlemfuensis), and Brachiaria (Brachiaria decumbens). None of the pastures had been fertilized. In many pastures, some scattered trees, remnants of the former forest, occurred.

2.4. Field Data Collection

[12] Soil samples were collected according to a stratified random sampling scheme. Two randomly positioned sampling points were selected from each of four parallel 50-m-long transects. Spacing between transects was 16.6 m. At these eight sampling points, mineral soil samples were collected with a Dutch auger (diameter 3 cm) at two depth intervals: 0–0.25 m and 0.25–0.50 m; thus we collected, in total, 16 samples per plot. Samples were air-dried, passed through a 2-mm sieve, and stored in plastic bags. At the four sampling points on the two extreme transects, bulk density samples were taken at two depth intervals. For each soil sample for bulk density, we carefully inserted a metal canister (250 cm3) into the undisturbed soil layer. As only a few of the bulk density samples had stones, we made no correction to gravel content. At the same four sampling points where bulk density samples were taken, pH was determined in the field with a portable field pH-meter.

[13] Apart from the soil samples, the following site and terrain characteristics were determined for all plots: geographical coordinates (using GPS), altitude, slope, orientation, stoniness, drainage, evidence of erosion, and visual observations of soil characteristics. Land use history, vegetation age, and actual land management was obtained through interviews with landowners. In the forest plots, tree biomass was estimated by means of non-destructive inventories of areas varying from 600 m2 to 1000 m2. Of all trees with a diameter at breast height (dbh) of at least 5 cm, dbh and tree height were measured, and tree species identified. Trunk biomass and total biomass were estimated using secondary information on species-specific wood densities, form factors, and biomass expansion factors [López et al., 2002]. No biomass was measured in pasture plots. Annual precipitation was estimated using a digital interpolated precipitation map based on 20 weather stations in the study area.

2.5. Laboratory Analyses

[14] Total carbon and nitrogen were analyzed for all individual samples (80 plots × 2 depths × 8 samples = 1280 samples), while analyses of the independent edaphic drivers were done for composite samples made for each depth (80 plots × 2 depths = 160 samples). For total carbon and nitrogen analyses, a subsample of the air-dried soil samples was ground to powder using a ball mill. Total organic C and N was determined using an automated C and N analyzer (Heraeus vario EL). Results are reported as the mean value from eight samples per depth.

[15] Particle size distribution was measured on composite samples using the pipette sedimentation method, distinguishing the fractions: clay (particle size < 0.002 mm), silt (particle size between 0.002 mm and 0.063 mm), and sand (particle size between 0.063 mm and 2 mm). Many of our soil samples are of volcanic origin. These soils are difficult to disperse, which gives problems in some analytical procedures; particularly, the particle size distribution is often highly unreliable [Mizota and van Reeuwijk, 1989].

[16] Because of the volcanic origin of some of our soils, we included selective dissolution techniques for poorly ordered or “amorphous” materials. These components (including allophane, imogolite, ferrihydrite, and Al/Fe-humus complexes) were extracted from the composite samples with 0.2 M ammonium oxalate solution in the dark. Fe (Feo), Al (Alo), and Si (Sio) were analyzed in the transparent solution by inductively coupled plasma emission spectrometry (ICP) [Buurman et al., 1996]. Pyrophosphate extractions were used to quantify Al, Fe and C present in humus complexes (Alp, Fep and Cp). Pyrophosphate extractions were conducted with a 0.1 M sodium pyrophosphate solution at pH 10, using a 1:100 soil:solution ratio. The amount of Al and Fe was analyzed by ICP. For the analysis of Cp, 10 mL of the transparent solution was centrifuged for 30 min at 4000 rpm. Subsequently, dissolved carbon was determined using a Dorrman Autoanalyser. The ratio of Alp/Alo is indicative of the contents of allophane versus al-humus complexes in volcanic soils. Alp/Alo values near 0 would suggest that allophane is dominant while Alp/Alo values near 1 indicate the predominance of Al-humus complexes [Mizota and van Reeuwijk, 1989]. Similarly, Alo minus Alp is an indication for non-crystalline minerals, with high values indicating high contents of these minerals. Oxalate extracted iron (Feo) minus pyrophosphate extracted iron (Fep) is an indication of the amount of ferrihydrite.

2.6. Calculations and Statistical Analyses

[17] Site 40, a site with volcanic soils, was excluded from analysis because it was an obvious outlier, which was probably caused by frequent inundations.

[18] Carbon stocks in 0- to 0.25-m and 0.25- to 0.50-m layers were calculated with the bulk density data for these layers and expressed as total carbon stock for the 0- to 0.50-m layer. Pasture soils are generally more compacted. In order to be able to compare the same soil mass for pastures and forests, and avoid concluding carbon accumulation as result of compaction in pastures [Veldkamp, 1994], for each site the bulk density data for forest were used for both forests and pastures. An alternative method used in the literature is to adjust the mass of the more compacted soil in order to compare the same soil mass for different land uses, but our approach accomplishes the same objective.

[19] Preliminary analysis of our data revealed that the two soil groups (volcanic soils and sedimentary soils) showed different correlations between soil C stocks and their drivers. Consequently statistical analyses were done for all sites together (N = 39) and for sedimentary (N = 27) and volcanic soils (N = 12) separately. The same preliminary analysis did not show differences between secondary forests and plantation forests. Therefore the two forest types were not analyzed separately. Spearman rank correlation coefficients were computed for the dependent variables: soil C concentration (%) in the 0- to 0.25-m layer in forest and in pasture, soil C concentration (%) in the 0.25- to 0.5-m layer in forest and in pasture, and the difference in total soil C stock (Mg ha−1) in the 0- to 0.5-m layer between pastures and forests and the independent drivers. Non-parametric correlation was chosen, as not all variables were normally distributed. We used multiple regressions with stepwise variable selection to relate dependent variables to driver variables. Separate multiple regression analyses were run for each of the dependent variables for pastures and forests. Three sites with sedimentary soils were excluded from multiple regression analysis, because they presented problems in accurately determining Al-contents. We did t-tests to calculate differences in soil C concentration and stocks between forest and pasture at each site. Paired t-tests were done to test for differences in dependent variables and independent drivers between forest and pasture.

3. Results

3.1. Soil Carbon and C/N Ratios in Pastures and Forests

[20] Table 1 shows precipitation per site, vegetation age, altitude, soil texture, and soil carbon concentrations and stocks for all sites. Soil C concentrations in the 0- to 0.25-m layer ranged between 1.2% and 6.9% in pastures and 1.6% and 6.5% in forests (Table 1). In the 0.25- to 0.50-m layer, carbon contents were clearly lower, ranging between 0.3% and 3.4% in pastures and between 0.3% and 3.5% in forests. In most sites carbon content in the deeper layer was 2 to 3 times lower than in the top layer. Average coefficient of variation (CV) per layer was between 22% and 27%, but large differences exist among sites, especially for the 0.25- to 0.50-m layer. Average C/N ratios in the pastures were 9.95 in the 0- to 0.25-m layer and 9.35 in the 0.25- to 0.5-m layer. In the forests, C/N ratios were 10.18 in the 0- to 0.25-m layer and 9.46 in the 0.25- to 0.5-m layer. Carbon stocks in the top 0.50-m soil layer varied between 59.2 Mg ha−1 and 195.2 Mg ha−1 in pastures and varied between 66.2 Mg ha−1 and 170.9 Mg ha−1 in forests (Table 1).

Table 1. Site Characteristics (Precipitation, Altitude, Texture, Bulk Density) and Soil C Concentrations (%), C/N Ratios, Soil C Stocks in Layers 0–0.25 m and 0.25–0.50 m in Pasture and Forest Plots, and Differences Between Pasture and Forest in Soil C Concentration (%) in the 0- to 0.25-m Layer and 0.25- to 0.50-m Layer and Differences Between Pasture and Forest in Total Soil C Stocks (Mg ha−1) in the Top 0.50 ma
SiteSoil TypePrecip., mm yr−1PasturesForestsDifferences
Alt., maslAge, yearsClay, %Silt. %Sand, %Bd 0–0.25 m, g dm−3Bd 0.25–0.5 m, g dm−3Soil c 0–0.25 m, %Soil c 0.25–0.5 m, %Soil cb 0–0.5 m, Mg ha−1Alt., maslAge, yearsClay, %Silt, %Sand, %Bd 0–0.25 m, g dm−3Bd 0.25–0.5 m, g dm−3Soil c 0–0.25 m, %Soil c 0.25–0.5 m, %Soil c 0–0.5 m, Mg ha−1Soil c 0–0.25 m, %Soil c 0.25–0.50 m, %Soil c 0–0.50 m, Mg ha−1
MeanCVMeanCVMeanCVMeanCV
  • a

    Precip, precipitation; Alt, altitude; bd, bulk density; CV, coefficient of variation; Soil type S, sedimentary soils; Soil type V, volcanic soils. Differences are value of forest minus value of pasture.

  • b

    Soil C stocks in pastures were calculated with bd data of forest (see text).

  • c

    Significance levels of t-test for differences in soil C concentration (%), p-value < 0.01.

  • d

    Significance levels of t-test for differences in soil C concentration (%), p-value < 0.05.

1S293821152760130,770,742.94212.5432111.41715326260,820,803.12132.0723105.70.18−0.47−5.7
2S28694841666181,001,003.5791.7212130.648252152260,981,002.19321.022179.3−1.38c−0.7c−51.3
3S26252992760131,131,192.16131.051892.218252166131,161,132150.773379.8−0.15−0.28d−12.4
4V3459164920024761,051,082.42272.1910114.2165020019801,020,963.86302.3917155.81.44d0.241.6
5V3485141018221771,041,033.69272.5624125.9139016124750,800,825.21373.18371691.52d0.6243.1
6S200096152553221,051,232.65311.1514107.411892960111,111,192.14290.852384.5−0.51−0.3c−22.9
7V486060013049510,730,694.99352.4327131.460021255440,710,704.65102.5412127.1−0.350.11−4.3
8V403885015038621,080,973.09171.8211121.485015131681,040,903.65261.8624136.80.560.0415.3
9V231012513637571,041,333.08431.5958136.3124915542531,171,163.96221.4657158.60.89−0.1322.2
10V27981986154450,660,656.91153.4416195.219415443530,750,775.77223.2917170.9−1.14−0.15−24.3
11V303020420347500,740,765.18113.3317134.320417250480,620,656.02163.2316145.70.84d−0.111.5
12S117020252351261,301,332.0470.951491.520103553111,191,292.3151.165106.40.270.2114.9
13S9781525386021,071,041.9290.89482.725183155141,161,222.02120.852484.40.09−0.031.7
14S2383140202352251,121,192.17350.89317817083048221,040,972.1191.001679−0.060.111.1
15V315016630012881,301,422.3450.591887.214011011891,141,392.48170.753097.10.140.179.9
16S18235635464861,191,202.15210.992588.15615455051,101,182.31140.981392.30.17−0.014.2
17S132714732603830,970,972.62121.231696.214714554140,981,053.17231.3731113.40.550.1417.3
18S315713254343140,981,042.55151.421898.688303548170,991,012.33290.912780.4−0.22−0.51c−18.1
19S28086751764201,341,231.62150.621061.445253549161,091,112.08261.092786.90.46d0.47c25.4
20V308017016032681,041,062.86190.683891.717010336610,981,252.98190.417186.10.11−0.27−5.6
21V34022817050500,640,754.9152.8619130.824711648460,670,694.1482.1817106.4−0.76−0.68−24.4
22S140154302244331,461,271.2390.744659.266103251171,241,191.78340.782378.50.580.0419.2
23S176246302948231,271,271.34340.682860.838112846261,191,221.55140.661766.20.2−0.025.4
24S24961720375581,091,071.46110.621863.851302145341,261,141.8270.713677.20.340.0913.4
25S173446453346221,261,252.6281.082086.746154346110,960,894.18500.8121118.21.57−0.27d31.5
26S2421136152841311,231,291.75481.334591.4140102051291,191,172.55200.8622101.50.8d−0.4710
27S193113710613460,840,922.82111.613583.313715464860,740,763.59471.463094.60.76−0.1511.4
28S2370116383054160,981,002.56181.131779.5116163240280,820,951.97341.494275.9−0.590.36−3.6
29S2021221202856161,061,091.93310.695264.7221153256120,990,982.6412.0935115.50.671.4c50.9
30S2561768385290,950,943.11291.1639110.1508385391,041,022.29341.252591.4−0.820.1−18.7
31S20238822854181,021,052.1440.714272.613083154151,001,142.06300.993579.8−0.040.287.2
32S1705147303652121,021,062.16421.435891.112710385580,991,053.22251.5526120.71.06d0.1229.5
33S250953253636271,321,391.85171.06248558173644201,181,162.54171.128106.40.69c0.0421.4
34S26832742856171,081,012.61191.3532123.627252143361,231,292.1250.842291.2−0.51−0.51c−32.4
35S235632354039211,141,212.75111.7322101.132203845170,841,013.78311.6426120.61.03d−0.0819.5
36V3114163301856261,021,042.2271.473394.416372251261,011,052.07332.5638119.7−0.131.09d25.4
37V346729420554410,650,665.69482.9448133.929410352450,610,646.5183.4634154.40.810.5120.5
38S37306852039421,011,172.6670.982381.567132257220,900,893.29241.4623106.50.63d0.48c25
39S24743224817761,161,371.62180.251961.221794881,311,341.88470.332572.50.270.0711.2
Mean 2576239192346321,051,082.78221.432798.7237162346311,001,033.03251.4726106.10.260.047.4

3.2. Correlations Among Altitude and Edaphic Variables

[21] Complex relationships between altitude and soil characteristics of the 0- to 0.25-m layer were revealed by the Spearman rank correlation coefficients (Table 2). Altitude is positively correlated with precipitation. Higher percentages of clay and silt were found at lower elevations and high fractions of sand in the soils at higher elevations. There was also a clear correlation between altitude and soil mineralogy. At higher elevations, soils show increasingly volcanic properties with increasing amounts of extractable aluminum, iron, and silica. These results demonstrate that soils of volcanic origin were situated at higher elevations and soils of sedimentary origin were found at lower elevations.

Table 2. Spearman Rank Correlation Coefficients Between Altitude and Site Characteristicsa
 PasturesForests
  • a

    Soil variables refer to the 0- to 0.25-m layer. N = 39. Alo, oxalate extracted Al; Feo, oxalate extracted Fe; Sio, oxalate extracted Si; Alp, pyrophosphate extracted Al; Fep, pyrophosphate extracted Fe.

  • b

    A p-value < 0.01.

  • c

    A p-value < 0.05.

Precipitation0.48b0.43b
Clay−0.56b−0.52b
Sand0.55b0.49b
Silt−0.38c−0.29
Clay + silt−0.55b−0.49b
Alo0.62b0.60b
Feo0.52b0.50b
Sio0.59b0.57b
Alp0.64b0.65b
Fep0.250.37
Alp/Alo−0.26−0.19
Alo-Alp0.57b0.49b

3.3. Correlations Between Soil Carbon Concentrations and Independent Variables

[22] In the sedimentary and volcanic soils, soil C showed completely different relationships with the independent variables, which became apparent especially in the forest sites (Table 3). In contrast to the volcanic soils, sedimentary soils under forest showed a strong positive correlation with clay content (and a negative correlation with sand content). This correlation was not found in the 0- to 0.25-m layer of the pasture soils, but did appear in the 0.25- to 0.5-m layer of the pasture soils. In the volcanic soils, a strong negative correlation was found with Alp/Alo and a strong positive correlation was found with Alo-Alp as shown for the 0- to 0.25-m layer in Figure 2. A negative relationship was found between pasture age and soil C concentration. The opposite was found for forest age and soil C concentration in volcanic soils, although this correlation was not significant. Furthermore, in the volcanic soils a strong positive correlation was found with the standing above ground biomass. Together, these results suggest that soil C concentration in sedimentary soils is related to the amount of clay, while the soil C concentration in volcanic soils is related to the amount of non-crystalline minerals (allophane, imogolite, etc.).

Figure 2.

Relation between aluminum indicators and carbon content in the 0- to 0.25-m layer for pastures and forests in volcanic soils.

Table 3. Spearman Rank Correlation Coefficients Between Soil C Concentration and Site Characteristicsa
 PasturesForests
SedimentaryVolcanicAllSedimentaryVolcanicAll
  • a

    Here n = 12 for volcanic soils, and n = 27 for sedimentary soils. Alo, oxalate extracted Al; Feo, oxalate extracted Fe; Sio, oxalate extracted Si; Alp, pyrophosphate extracted Al; Fep, pyrophosphate extracted Fe; Cp, pyrophosphate extracted C. Oxalate and pyrophosphate extractions were only done for the 0- to 0.25-cm layer.

  • b

    A p-value < 0.01.

  • c

    A p-value < 0.05.

C% in 0- to 0.25-m Soil Layer
Altitude0.150.150.52b0.260.310.60b
Slope−0.20−0.20−0.11−0.040.160.03
Precipitation0.290.060.54b−0.0040.060.40c
Clay0.16−0.01−0.410.57b0.04−0.27
Sand−0.27−0.450.28−0.52b−0.280.22
Silt0.070.43−0.070.150.43−0.13
Clay + silt0.270.45−0.280.52b0.28−0.22
pH−0.09−0.07−0.300.03−0.13−0.31
Bulk density−0.70b−0.69c−0.76b−0.65b−0.76b−0.71b
Age−0.23−0.46−0.35c−0.020.380.11
Biomass   −0.090.78b0.08
Alo0.41c0.88b0.70b0.40c0.85b0.69b
Feo0.37−0.380.54b−0.02−0.270.39c
Sio0.260.77b0.62b0.160.81b0.58b
Alp0.380.93b0.69b0.160.97b0.61b
Fep0.270.430.52b0.120.390.40c
Alp/Alo−0.05−0.78b−0.41b0.04−0.77b−0.35c
Alo-Alp0.320.85b0.68b0.140.83b0.56b
Feo-Fep0.38−0.370.52b−0.05−0.350.33c
Cp0.78b0.89b0.85b0.74b0.88b0.84b
 
C% 0.25- to 0.50-m Soil Layer
Altitude0.160.220.48b0.300.150.58b
Slope0.06−0.050.160.210.040.19
Precipitation0.180.020.43b0.070.090.41b
Clay0.45c−0.14−0.160.080.09−0.32c
Sand−0.31−0.380.14−0.36−0.570.13
Silt0.060.530.290.41c0.520.48b
Clay + silt0.310.38−0.140.360.57−0.13
pH0.02−0.18−0.14−0.050.04−0.26
Bulk density−0.50c−0.83b0.68b−0.63b−0.77b−0.78b
Age−0.07−0.30−0.22−0.040.160.07
Biomass   −0.040.58b0.12

3.4. Influences of Independent Variables on Soil C Concentration and Stocks

[23] Multiple regression analysis was done with the following dependent variables: soil C concentration (%) in the top 0.25 m and cumulative soil C stocks (Mg ha−1) in the top 0.5 m. Independent drivers tested were aboveground biomass (Mg ha−1), elevation (masl), slope (degrees), precipitation (mm yr−1), clay (%), sand (%), silt (%), clay + silt (%), Alp(%), Fep(%), Alp/Alo, Alo-Alp, and Feo-Fep. Both edaphic and biophysical variables were selected with the stepwise variable selection procedure (Table 4). In general the models for soil C concentration or cumulative soil C stocks explained more of the variance in volcanic soils (R2 between 0.64 and 0.96) than in sedimentary soils (R2 between 0.13 and 0.39), with lower R2 for stocks than for concentrations. The models including all soils explained between 56% and 71% of the variance. Variables explaining soil C concentration and cumulative C content in volcanic soils included: Alp, Fep, and altitude. Variables explaining soil C concentration and cumulative C content in sedimentary soils included clay content, Alp, and (Alo-Alp). These results strongly suggest that different mechanisms of carbon stabilization exist in the two soil groups.

Table 4. Multiple Regression Models for the Prediction of Soil C Concentration (%) in the Top 0.25 m and Soil C Stock (Mg ha−1) in the Top 0.50 m Under Pastures and Forestsa
Soil GroupModelp-Value ModelR2
  • a

    Here n = 12 for volcanic soils, and n = 24 for sedimentary soils. Alo, oxalate extracted Al; Alp, pyrophosphate extracted Al; Fep, pyrophosphate extracted Fe.

Soil C Concentration (%) Top 0.25 m
Pastures
   All soilsLog (C%) = 0.103 + 0.74 × Alp + 0.0019 × (clay + silt)0.000.71
   VolcanicC% = −0.17 + 6.9 × Alp + 0.38 × (Alo-Alp)0.0000.96
   SedimentaryC% = 1.82 + 3.97 × Alp0.0000.14
Forests
   All soilsLog (C%) = 0.31 + 0.56 × Alp0.0230.60
   VolcanicC% = 0.45 + 9.34 × Alp − 4.58 Fep0.0000.93
   SedimentaryC% = 0.45 + 0.049 × clay + 5.55 × (Alo-Alp)0.0020.39
 
Soil C Stock (Mg ha−1) Top 0.50 m
Pastures
   All soilsLog C (Mg ha−1) = 1.88 + 0.41 × Alp0.0000.56
   VolcanicC (Mg ha−1) = 46.19 + 149.4 × Alp0.0000.75
   SedimentaryC (Mg ha−1) = 71.87 + 132.06 × Alp0.0470.13
Forests
   All soilsLog C (Mg ha−1) = 1.87 + 0.37 × Alp + 0.00014 × Altitude + 0.0067 × clay − 0.15 × Alp/Alo0.0000.68
   VolcanicC (Mg ha−1) = 60.94 + 106.7 × Alp + 0.026 × Altitude0.0040.64
   SedimentaryC (Mg ha−1) = 65.40 + 0.87 × clay0.0100.22

3.5. Differences in Soil Carbon Concentrations and Stocks Between Pastures and Forests

[24] Differences of soil C concentration were tested with a t-test (Table 1), but differences in soil C stocks were not tested with a t-test as they were based on average bulk density per plot. Positive differences indicate higher carbon concentration in the forest than in the pasture. The average difference in soil C contents in the 0- to 0.25-m layer was 0.26%. When expressing this difference as percentage of the soil C content in pastures, the average soil C content in forests was 9.2% higher than the average soil C content in pastures. In the 0.25- to 0.50-m layer the average difference was 0.04%, a difference which, when expressed as percentage of the average soil C content in pastures, represents an average soil C content of 2.7% higher in forests than in pastures. While the average difference in soil C content in the top layer was 0.26%, the average difference in pyrophosphate-extracted carbon content (Cp) in this layer was only 0.046%. This means that 18% of the differences are related to the humus-associated fraction. This fraction constitutes on average about 31% of soil C in both pastures and forests.

[25] In the 0- to 0.25-m layer, 27 sites had a positive difference of soil C contents, of which nine were significant (t-test, p < 0.05). Of the 13 negative differences, only one was significant (t-test, p < 0.05). In the 0.25- to 0.50-m layer, 21 sites had a positive difference of soil C contents, of which four were significant (t-test, p < 0.05), and 19 sites had a negative difference of which six were significant (t-test, p < 0.05). In terms of soil C stocks for the top 0.50 m, differences between pastures and forests fluctuated between −51.3 and 50.9 Mg ha−1 (Table 1). The average soil C stock in pastures was 98.7 Mg ha−1 and in forests 106.1 Mg ha−1, an average difference of 7.4 Mg ha−1, equivalent to a soil C stock in forests on average 7.5% higher than the soil C stock in pastures. Absolute differences are bigger in volcanic soils (10.9 Mg ha−1) than in sedimentary soils (5.7 Mg ha−1) (Figure 3). Relative differences, expressed as percentage of pasture soil C stock, are 8.7% for volcanic soils and 6.6% for sedimentary soils.

Figure 3.

Average soil C stock (Mg ha−1) in layers 0–0.25 m and 0.25–0.50 m of pasture and forest, for sedimentary soils and volcanic soils. Error bars indicate 95% confidence intervals.

[26] A paired t-test was used to test the effect of land use on soil C and bulk density when data of all sites are included (Table 5). Differences in soil C between pasture and forest were significant in the 0- to 0.25-m layer but not for the 0.25- to 0.50-m layer. Cumulative soil C in the top 0.50 m showed significant differences as well. Pyrophosphate-extracted carbon content (Cp) in the 0- to 0.25-m layer was not different. However, total soil C content minus Cp was different. Significant differences in bulk density indicate compaction in pastures due to cattle grazing. Soil texture, pH, and mineralogy variables were not different between pasture and forest for all sites, except for Feo.

Table 5. Comparison Between Pasture and Forest of Soil Characteristics Over All Sites: Paired t-test Significance Levelsa
VariableSignificance Level
  • a

    Alo, oxalate extracted Al; Feo, oxalate extracted Fe; Sio, oxalate extracted Si; Alp, pyrophosphate extracted Al; Fep, pyrophosphate extracted Fe; Cp, pyrophosphate extracted C.

  • b

    A p-value < 0.05.

C (%) 0–0.25 m0.027b
C (%) 0.25–0.50 m0.577
C (Mg ha−1) 0–0.25 m0.018b
C (Mg ha−1) 0.25–0.50 m0.593
C (Mg ha−1) 0–0.50 m0.042b
Cp (%) 0–0.25 m0.226
C (%) minus Cp (%) 0–0.25 m0.023b
Bulk density 0–0.25 m0.024b
Bulk density 0.25–0.50 m0.031b
pH layer 0–0.25 m0.960
pH layer 0.25–0.50 m0.214
Clay 0–0.25 m0.612
Clay 0.25–0.50 m0.790
Sand 0–0.25 m0.797
Sand 0.25–0.50 m0.837
Silt 0–0.25 m0.891
Silt 0.25–0.50 m0.957
Altitude0.487
Alo 0–0.25 m0.880
Feo 0–0.25 m0.047b
Sio 0–0.25 m0.924
Alp 0–0.25 m0.376
Fep 0–0.25 m0.849

3.6. Correlations Between Changes in Soil C Following Forest Regrowth and Independent Variables

[27] Pasture age explained much of the variation in the difference in soil C stocks between pasture and forest for all sites (Table 6). In addition to pasture age, significant negative correlations were found for Alo and Feo and Feo-Fep in sedimentary soils. In volcanic soils, significant positive correlation was found only for slope. For forest biomass or forest age, no significant correlations were found.

Table 6. Spearman Rank Correlation Coefficients Between Site Characteristics and the Difference in Soil C Stocks (Mg ha−1) in the Top 0.50 m Between Pastures and Forests (Forest Soil C Stock Minus Pasture Soil C Stock)a
 SedimentaryVolcanicAll
  • a

    For texture and mineralogy variables, averages per site for the 0- to 0.25-m layer are used. Alo, oxalate extracted Al; Feo, oxalate extracted Fe; Sio, oxalate extracted Si; Alp, pyrophosphate extracted Al; Fep, pyrophosphate extracted Fe.

  • b

    A p-value < 0.05.

  • c

    A p-value < 0.01.

Altitude0.190.530.26
Slope0.010.59b0.24
Precipitation−0.340.20−0.08
Age pasture0.42b0.420.38b
Age forest−0.090.15−0.05
Forest biomass−0.270.01−0.21
Clay0.210.040.03
Sand0.040.200.11
Silt−0.33−0.17−0.28
Alo−0.35−0.37−0.11
Feo−0.58c0.24−0.23
Sio0.06−0.290.08
Alp−0.14−0,25−0.01
Fep−0.37−0.06−0.23
Alp/Alo0.260.480.28
Alo-Alp−0.48b−0.35−0.17
Feo-Fep−0.57c0.06−0.30

[28] Independent variables included for stepwise variable selection in the multiple regression analyses for soil C differences were the same as the variables used for the soil C concentrations and stocks. The stepwise regression resulted in the selection of the variables pasture age and altitude for all models for lumped soils and volcanic soils (Table 7). For the sedimentary soils, pasture age was selected for differences in soil C concentration and the non-humus related iron fraction was selected for differences in soil C stocks (Table 7). The explained variance was highest for the volcanic soils, lowest for the sedimentary soils and intermediate if all soils were lumped together. The models for differences in soil C stocks had higher R2 than the models for differences in soil C concentrations.

Table 7. Multiple Regression Models for the Prediction of Differences (Dif C) in Soil C Concentrations (%) in the Top 0.25 m and Differences in the Soil C Stocks (Mg ha−1) in the Top 0.50 m Between Pastures and Forestsa
Soil GroupModelp-Value ModelR2
  • a

    Here n = 12 for volcanic soils, and n = 24 for sedimentary soils. Dif C is soil C of forest minus soil C of pasture. For texture and mineralogy variables, averages per site for the 0- to 0.25-m layer are used. Alo, oxalate extracted Al; Alp, pyrophosphate extracted Al.

Differences in Soil C Concentration (%) Top 0.25 m
All soilsDif C (%) = −0.49 + 0.029 × (pasture age) + 0.00085 × altitude0.0000.34
VolcanicDif C (%) = −1.09 + 0.042 × (pasture age) + 0.0012 × altitude0.0070.60
SedimentaryDif C (%) = −0.28 + 0.026 × (pasture age)0.0100.21
 
Differences in Soil C Stock (Mg ha−1) Top 0.50 m
All soilsDif C (Mg ha−1) = −17.15 + 0.98 × (pasture age) + 0.028 × altitude0.0000.36
VolcanicDif C (Mg ha−1) = −36.37 + 1.68 × (pasture age) + 0.033 × altitude0.0000.82
SedimentaryDif C (Mg ha−1) = 22.50 − 3.28 (Feo-Fep)0.0050.19

4. Discussion

4.1. Comparability of Forest and Pasture Plots at Each Site

[29] A critical assumption in the present study (and all studies involving space-for-time substitution) is that pasture and forest sites were comparable before land use changes took place. As we did a spatial comparison of sites, land-use- and site-inherent factors of variation are entirely confounded, making it important to justify this assumption. There are several ways in which we dealt with this problem. First, we did a careful site selection. As described earlier, we took care that the distance between forest and pasture sites was as short as possible. Furthermore, we made sure that soil and terrain conditions between the forest and pasture sites were as similar as possible. Second, we included a considerable number of replicates. The study involved 40 sites where forest and pastures were compared. This large number of observations compared to most previous studies involving space-for-time substitutions greatly reduces the risk of finding differences between forests and pastures that are not related to land use. Third, we investigated some land use-independent parameters. The quality of site selection can be tested using variables measured in the paired forest and pasture plots that are independent of land use (e.g., soil texture and soil mineralogy). A paired t-test showed no difference of soil texture between the forest and pasture plots (Table 5). Also, pyrophosphate and oxalate extracts, which can be related to soil mineralogy, showed no differences. The only exception was Feo, which was significantly higher for pastures than for secondary forests (p = 0.047). This confirms findings by Powers [2001], who also found that Feo depended on land use for a regional study in the humid tropics of Costa Rica. We conclude from this analysis that in general our sites were comparable and differences in soil C measured between pastures and forests were caused by land use and not by inherent site variability.

4.2. Landscape Variation of Soil Carbon Storage

[30] In the volcanic soils, soil C showed a strong negative correlation with Alp/Alo ratios smaller than 0.5 (Figure 2). The ratio of Alp/Alo has been used as an index for volcanic soils: Values below 0.5 are indicative of allophanic mineralogy, while a ratio above 0.5 indicates that the active Al is found in Al-humus complexes [Mizota and van Reeuwijk, 1989]. This indicates that a large part of soil C was stabilized by the presence of non-crystalline minerals such as allophane or imogolite. A similar mechanism for stabilization of soil C was described for volcanic soils in a landscape scale study in Costa Rica [Powers and Schlesinger, 2002]. In the sedimentary soils, a strong positive correlation was found with clay content in the topsoil of the forest sites (not in the pasture sites). This suggests that in the sedimentary soils, physical protection by the clay fraction is responsible for the stabilization of soil C. However, a positive correlation with Alo indicates that also in these soils, non-crystalline minerals may play a role in the stabilization of soil C. In addition to soil mineralogy and texture, other (mainly management related) factors explained some of the variation in soil C across the landscape. Pasture age was negatively correlated to soil C in the topsoil of all soils. This showed that pastures progressively lost soil C as they get older. In the volcanic soils, forest biomass displayed a strong positive correlation with soil C in the top 0.25 m, which is a first indication that soil C can increase under secondary forests in this region.

[31] The multiple regression equation for surface soil C concentration in the volcanic soils was best explained with mineralogy variables both in pastures and in forests. In the sedimentary soils, a combination of mineralogy and texture variables best explained the variation in soil C. In all cases the R2 of the regression equations of the sedimentary soils were quite low, which may reflect the complex parent material of these soils, differences in drainage conditions, or other factors influencing soil C that were not accounted for.

4.3. Landscape Variation in the Response of Soil C Following Forest Regrowth

[32] The few studies that exist on how soil C is affected if pasture is converted into forests have shown variable results [Guo and Gifford, 2002]. Also in this study, there was a considerable variation in the direction and magnitude of the changes in soil C following forest regrowth on pastures, which took place mainly in the topsoil. Topographic and management factors explained much of the variation, suggesting that careful site selection will determine whether a secondary forest will sequester soil C or not. A positive correlation in volcanic soils between the difference in soil C stock between forest and pasture and slope may indicate that slope processes like landslides can influence C sequestration in soils. Although the large variation in soil and environmental variables makes it impossible to reconstruct a chronosequence, grouping the sites according to pasture age revealed an interesting pattern (Figure 4a). Secondary forests that were paired with pastures less than 10 years old had on average 9.3 Mg ha−1 (7.9%) less soil carbon than the pastures. When paired with pastures between 10 and 20 years old, secondary forests had on average 5.2 Mg ha−1 (4.8%) more soil carbon than pastures. This difference increased to 18.8 Mg ha−1 (20.7%) for pastures between 20 and 30 years old and 15.8 Mg ha−1 (18.7%) for pastures older than 30 years. The two oldest pasture classes were significantly different from the youngest pasture age class, but not significantly different from each other or pasture age class 10 to 20 years. We measured a carbon sequestration in aboveground biomass of regrowing secondary forest at our sites of about 100 Mg of C after 30 years [de Koning et al., 2002]. Compared to this aboveground carbon sequestration, the difference of soil C in forests compared to pastures of more than 20 years age represents between 16% and 19%.

Figure 4.

(a) Differences in cumulative soil C stock (Mg ha−1) in the 0- to 0.50-m layer between pastures and forests according to pasture age class. (b) Differences in cumulative soil C stock (Mg ha−1) in the 0- to 0.50-m layer between pastures and forests according to pasture age class, divided by forest age. Positive differences indicate higher content in forests than in pastures. Errors bars indicate 95% confidence interval. Number of observations: age class <10: n = 11; age class 10–20: n = 8; age class 20–30: n = 10; age class >30: n = 10. Means sharing the same letter (a, b) were not discriminated by t-test (p < 0.05).

[33] When we divide the soil C differences of Figure 4a by forest age (in order to get the average change in soil C stock per year), we get an average loss of soil C in forests of −0.47 Mg ha−1 yr−1 if pastures younger than 10 years are converted into forests (Figure 4b). For pastures between 10 and 20 years old, the average yearly increase of soil C in forests is 0.22 Mg ha−1 yr−1, while for pastures between 20 and 30 years old this increase is 1.32 Mg ha−1 yr−1 and for pastures older than 30 years, 1.42 Mg ha−1 yr−1 (Figure 4b). Together, these results suggest that the amount of soil C in young pastures is higher than under secondary forests, leading to a decrease in soil C if they are converted into forest. In contrast, old pastures have soil C levels that are considerably lower than secondary forests, leading to C sequestration if reforested. Similar increases in soil C below secondary forest were reported by Rhoades et al. [2000], whose study sites were in the higher part of our study region. The high levels of soil C in young pastures may originate from the original forest. Several authors have reported an increase in soil C directly following forest clearing due to the decomposition of remaining litter and root material [Garcia-Oliva et al., 1994; McAllister et al., 1998]. After several years under pasture, decomposition of soil C and a low input of litter result in decreasing carbon levels, which are lower than the original level at pasture establishment. It must be noticed, however, that for situations where pasture management is good and/or climate and soils are favorable, forest to pasture conversion may also lead to increases of soil C because of inputs from growing grass roots [Neill et al., 1997].

[34] Our results confirm our first hypothesis that soil C increases after conversion of extensively managed pastures to forest ecosystems. Our second hypothesis that the amount of change in carbon stocks follows a pattern which can be predicted based on land use history, soil characteristics, and climate was only partly confirmed. We did find that land use history affected the soil C sequestration potential, but a direct influence of climate was not found and a relation with soil characteristics could only be partially confirmed. The paired t-test indicated that differences in Cp in the 0–0.25 m between pastures and forests were not significant. Cp represents the amount of carbon in organo-metal complexes. Changes in total carbon after pasture to forest conversion seem therefore mainly associated with the non-organo-metal carbon fraction (indicated with the significant differences in C minus Cp), which is probably carbon associated with clay and silt minerals and in the case of volcanic soils also carbon associated with allophanes.

4.4. C/N Ratio of Soil Organic Matter

[35] The C/N ratio of soil organic matter changed depending on land use (Figure 5). Under pasture the C/N ratio had the tendency to decrease in time, while under secondary forest the C/N ratio increased in time. This was especially clear for forests on volcanic soils. The decrease in C/N ratio under pastures may be partly caused by high decomposition rates, resulting in a lower C/N ratio [Balesdent et al., 1998], and possibly also by differences in litter quality between pasture and forest. Woody debris has generally a higher C/N ratio than grass residues in this area. This increase in C/N affects the C-storage capacity of the soil if nitrogen is a limiting factor in these ecosystems. A higher C/N ratio means that more carbon can be stored per unit of nitrogen. The importance of the availability of nitrogen for C sequestration was also demonstrated by the larger soil C sequestration under N-fixing trees compared with Eucalyptus trees [Resh et al., 2002]. This is probably one of the factors why the soils below secondary forest sequester carbon.

Figure 5.

C/N ratio in the 0- to 0.25-m layer in pastures and forests as function of vegetation age.

4.5. Consequences for Baseline Studies and C Sequestration in Reforestation Projects

[36] For carbon sequestration projects, it is necessary to have reliable baseline studies. In countries like Ecuador where CDM projects can take place, this information is normally not available. The present study used grassland as baseline situation for possible reforestation projects. It was shown that with relatively simple methods it is possible to establish quantitative relations between soil C stocks and concentrations and topographic and biophysical variables. This information can be used to extrapolate to the whole region if the necessary regional data exist for the dependent variables. Many of the correlations we found in our volcanic soils were also found by Powers and Schlesinger [2002]. This may open the possibility to develop predictive relations which may also be applied outside the area where they were created. Our results have also shown that the predictive relations were much stronger for volcanic soils than for sedimentary soils, underlining that the error related with a baseline study depends on the strength of the correlations [McKenzie and Ryan, 1999]. Our results also showed that past land use strongly affected the capacity of a soil to sequester carbon: If old pastures were converted into forests, considerable soil C sequestration can take place, but if young pastures were converted into forests, soil C losses can even take place. For C-sequestration projects it is therefore necessary to make an analysis of past land use before areas are selected for forest regrowth. This could be done using past and present satellite images. The results also offer scope for linkage with dynamic land use change models [de Koning et al., 1999].

5. Conclusions

[37] This study has shown that over a relatively large region in northwestern Ecuador, a large part of the variability in soil C concentrations and stocks can be predicted using topographical and biophysical variables. The direction and magnitude of changes in soil C following forest regrowth on pastures were extremely variable, but a considerable part of the variation could be explained with former land use. This information, which was obtained in a relatively simple way, should be collected before site selection is done for C-sequestration projects by forest regrowth. Furthermore, the prediction of changes in soil C opens the possibility to include C sequestration in soils in afforestation and reforestation projects, which until now was not the case.

Acknowledgments

[38] The Tropical Ecology Support Program (TOEB) of the German Technical Cooperation (GTZ) financed this study. We thank the TOEB staff: Elisabeth Mausolf, Michaela Hammer, Rudiger Wehr, Dorothé Otto, Claus Bätke, and Michael Tampe for their continuous support. We thank Klaus Werner for assistance during soil analyses and Marife Corre who reviewed an earlier version of this manuscript. We acknowledge the many landowners who allowed us to collect field data, and we thank Wolfgang Lutz of GTZ-Ecuador for his institutional support.

Ancillary