Kaleka Agroforest in Central Kalimantan (Indonesia): Soil Quality, Hydrological Protection of Adjacent Peatlands, and Sustainability

: Increased agricultural use of tropical peatlands has negative environmental effects. Drainage leads to landscape-wide degradation and ﬁre risks. Livelihood strategies in peatland ecosystems have traditionally focused on transitions from riverbanks to peatland forests. Riparian ‘ Kaleka ’ agroforests with more than 100 years of history persist in the peatlands of Central Kalimantan (Indonesia), where large-scale open-ﬁeld agricultural projects have dramatically failed. Our ﬁeld study in a Dayak Ngaju village on the Kahayan river in the Pulang Pisau district involved characterizing land uses, surveying vegetation, measuring soil characteristics, and monitoring groundwater during a period of 16 months. We focused on how local practices and farmer knowledge compare with standard soil fertility (physical, chemical, biological) measurements to make meaningful assessments of risks and opportunities for sustainable land use within site-speciﬁc constraints. The Kaleka agroforests around a former settlement and sacred historical meaning are species-rich agroforests dominated by local fruit trees and rubber close to the riverbank. They function well with high wet-season groundwater tables (up to − 15 cm) compatible with peatland restoration targets. Existing soil quality indices rate the soils, with low soil pH and high Al exch , as having low suitability for most annual crops, but active tree regeneration in Kaleka shows sustainability.


Introduction
Soils of a wide range of physical, chemical and biological soil characteristics can be sustainably used, as long as the vegetation is adapted to the root environment, and if the long-term carbon, nutrient, and water balance is respected by matching inputs to output [1][2][3]. Soil health and soil quality constrain land use depending on the management applied. The direction of change, interpreted as degradation or restoration, is more relevant to local land users than absolute measurable quantities [4]. Impacts of plot-level change can reach beyond the plot scale of a direct land user and affect ecosystem services 'downstream' [5,6]. Long traditions in 'land evaluation' have informed the locational choice for specific development projects by assessing soil quality in relation to specified requirements of pre-selected (tree) crops [7,8]. In contrast, local people had to deal with and make the best of local conditions through their land use choices, and developed locally calibrated process-level knowledge, spatially explicit indicators, and adjustable management options, even though their site conditions would be considered to be sub-optimal if other choices were available [9,10].

1.
How does the groundwater level fluctuate in the various land uses (including Kaleka)? 2.
What are the soil's physical, chemical, and biological characteristics and the existing soil quality indices in the various land uses? 3.
How does botanical composition and vegetation structure of the land uses relate to soil quality, water tables, and internal regeneration as sustainability indicator?

Location and Period of Study
The research was carried out in the village Henda in the Kahayan watershed, administratively part of Kecamatan Jabiren Raya, Kabupaten Pulang Pisau, in Kalimantan Tengah province, Indonesia ( Figure 1).
Land 2021, 10, x FOR PEER REVIEW 3 of 20 peat domes beyond, but harvesting of forest products from these zones part of traditional livelihoods [41]. The Tumbang Anoi Peace Treaty of 1894 recorded 96 customary law provisions related to Dayak heritage, including rules about Kaleka 'orchards' [42]. Kaleka agroforests occur along the Kapuas river, both upstream and downstream of Palangka Raya [43]; they may, eventually, be abandoned and return to forest conditions [44]. Appreciation of their function as wet ecosystems that protect the peat domes will require an understanding of the socio-ecological drivers of forest degradation in historical context, but also in current policy framing [45,46]. As agroforests, the Kaleka have a balance of local fruit trees and rubber, introduced over a century ago, compare well in tree diversity and biomass with well-studied agroforest examples in Indonesia, such as the mixed fruit tree gardens ('simpukng') in East Kalimantan [47], the repong or damar agroforest of Lampung that begins as coffee gardens but becomes dominated by Shorea trees that can be tapped for resin with various industrial uses [48], and the rubber agroforests of Sumatra and W Kalimantan [49,50].
Here, we provide a more detailed analysis of soil conditions in the various land uses in relation to groundwater dynamics. Specific question for this research were: 1. How does the groundwater level fluctuate in the various land uses (including Kaleka)? 2. What are the soil's physical, chemical, and biological characteristics and the existing soil quality indices in the various land uses? 3. How does botanical composition and vegetation structure of the land uses relate to soil quality, water tables, and internal regeneration as sustainability indicator?

Location and Period of Study
The research was carried out in the village Henda in the Kahayan watershed, administratively part of Kecamatan Jabiren Raya, Kabupaten Pulang Pisau, in Kalimantan Tengah province, Indonesia ( Figure 1).  Description of the vegetation, sampling of earthworms, and collection of soil samples for physical and chemical analysis took place in September-October 2017; groundwater tubes were installed and monitored on a two-weekly schedule until mid-January 2019. Interviews with farmers on their ethnobotanical knowledge and appreciation for various land uses was reported elsewhere [38]. The mean annual rainfall in the area is 3194 mm, with seven humid, three moist, and three dry months, according to the Schmidt-Ferguson criteria used in Indonesian climate statistics [51]. The study area is part of the lower Kahayan River, at 0-5 m above sea level, with still some tidal variation in river flow but no saline intrusion. The peat domes that developed in Central Kalimantan's river deltas are now understood to be part of peat hydrology units (KHG = Kesatuan Hidrologi Gambut).
Kalimantan (the Indonesian part of Borneo Island) is geologically part of the Sunda shelf, dominated by a quaternary geology that reflects fluctuating sea levels and climates in which peat soils could be formed and maintained, alternating with drier and more seasonal climates [52]. The Pulang Pisau district is flat, with the slope on 94% of the land less than 3% and that on 81% less than 1%. As described in the semi-detailed soil map for the district [53], approximately 60% of the area is classified as peat (Histosols): 47.9% ombrotrophic peat that depends on rainfall only rather than river water, and 11.2% peat influenced by rivers on the edges of the interfluvial peat domes. Most of the human activity has historically been in the alluvial deposits (mostly Entisols) by and along a number of rivers (10.2% of the area) and in the coastal zone (13.5% plus 9.5% as fluvio-marine contact zone) with mangroves and associated pyrite concretions in the subsoil that can lead to acid sulphate soil reactions when drained. The remaining 7.7% of the area is tectonic in origin and tends to be used for settlements, such as the city of Palangka Raya; these soils are mostly Inceptisols. A large area (around 440,000 ha) located in the Pulang Pisau Regency became known as 'Block C' in the 'Mega-Rice project' started in the 1990s and abandoned after the government change of 1998 [54]. Of the total area of the Mega Rice Project, 32% consisted of fine-textured mineral soils with a peat layer (histic epipedon) less than 40 cm thick, classified as Sulfaquents (Entisols with a tendency to become acid sulphate soils after drainage), Fluvaquents, Tropaquepts, Dystropepts, and Hapludults.
Henda Village, with an area of 538 km 2 and 648 inhabitants (1.2 km −2 ) [55], is at the center of the study area (1 • 32 00-3 • 28 00 South, 113 • 30 00-120 • 00 00 East) of the earlier ethnobotanical study [38]. According to the soil map [53], the soils closest to the river in Henda are Aeric Endoaquepts, followed by Typic Haplosaprist (sapric peat) and Hemic Haplosaprists. As Henda Village is the oldest settlement, has the most extensive land area, and widest range of land uses, it was chosen for more detailed observations. The Dayak Ngaju communities along the Kahayan river interact with various groups of migrants from elsewhere in Indonesia, under the government-sponsored transmigration program as well as spontaneous migrants.
Permission for research was obtained from the village elders, and the first author, with five years' experience in the area, was welcomed in the homes of many villagers, with follow-up interviews in the field, and participation in village events. The results presented here were obtained in parallel with an ethnobotanical study reported elsewhere [38].

Survey Design for Kaleka as Local Land Use
As a settlement, Henda dates back to at least 1840; it primarily relied on a swidden/fallow rice production system in the riparian zone for food security, with agroforest (farmer-modified forest, with increased presence of preferred species) as source of fruits, medicine, building material, firewood, and marketable 'forest products'. The local name for these agroforests is Kaleka, and they can vary in the relative dominance of fruit trees and rubber. When the local population density increased, these riparian, wet agroforests spread further from the riverbanks, but not into deeper peat soils. Privately owned and inherited in the local custom, all these agroforests are owned by Dayak Ngaju families. Based on the local adat history, Henda became a separate village in 1902. Around 1985, 'modern' techniques for rice production were introduced and efforts were made to intensify rubber production as the main income source. The rubber expanded into shallow (<50 cm) peat soils that were made accessible when a paved road connecting Banjarmasin and Palangka Raya (the provincial capitals of S and C Kalimantan) was constructed. Five land cover types, representing Kaleka as the major current land use, were purposely sampled (in three plot-level replicates) around the Henda village ( Figure 2). Based on the local adat history, Henda became a separate village in 1902. Around 1985, 'modern' techniques for rice production were introduced and efforts were made to intensify rubber production as the main income source. The rubber expanded into shallow (<50 cm) peat soils that were made accessible when a paved road connecting Banjarmasin and Palangka Raya (the provincial capitals of S and C Kalimantan) was constructed. Five land cover types, representing Kaleka as the major current land use, were purposely sampled (in three plot-level replicates) around the Henda village ( Figure 2). Fruit agroforest (FAF) is closest to the village and what is most likely an older location of the village that shifted slightly to the South. Some of the trees are, according to local sources, at least a hundred years old. Important local fruit trees include 'cempedak' (Artocarpus integer), manggis (Garcinia sp.), rambai (Baccaurea motleyana), langsat (Lansium domesticum), and paken (Durio kutejensis). There are also rubber (Hevea brasiliensis) trees in the plots.
The old rubber agroforest (RAFo) was planted with rubber approximately 150 years ago and relies on the natural regeneration of these trees within the plot along with local fruit trees, including cempedak, rambai, durian (Durio sp.), binjai (Mangifera caesia), and rambutan (Nephelium lappaceum).
The young rubber agroforest (RAFy) was planted with rubber approximately 30 years ago (according to local informants) and is similar in tree composition to the RAFo plots. Local trees include hampalam (Mangifera sp.), manggis, rambutan, and sentol (Sandoricum koetjape).
The rubber monoculture (RMO), also planted approximately 30 years ago, has a higher rubber density but still contains some local trees, such as durian, rambutan, and cempedak. Fruit agroforest (FAF) is closest to the village and what is most likely an older location of the village that shifted slightly to the South. Some of the trees are, according to local sources, at least a hundred years old. Important local fruit trees include 'cempedak' (Artocarpus integer), manggis (Garcinia sp.), rambai (Baccaurea motleyana), langsat (Lansium domesticum), and paken (Durio kutejensis). There are also rubber (Hevea brasiliensis) trees in the plots.
The old rubber agroforest (RAF o ) was planted with rubber approximately 150 years ago and relies on the natural regeneration of these trees within the plot along with local fruit trees, including cempedak, rambai, durian (Durio sp.), binjai (Mangifera caesia), and rambutan (Nephelium lappaceum).
The young rubber agroforest (RAF y ) was planted with rubber approximately 30 years ago (according to local informants) and is similar in tree composition to the RAF o plots. Local trees include hampalam (Mangifera sp.), manggis, rambutan, and sentol (Sandoricum koetjape).
The rubber monoculture (RMO), also planted approximately 30 years ago, has a higher rubber density but still contains some local trees, such as durian, rambutan, and cempedak.
The open-field agriculture (OFA) is located in a plot that previously had been rubber agroforest but was opened 5 years before our study. At the time of sampling, watermelon was the main crop planted for selling at markets in the nearby city of Palangka Raya.
Plot-level data were collected on (1) the fluctuations of groundwater levels; (2) soil physical, chemical, and biological characteristics; and (3) vegetation composition and other characteristics. All measurements were coordinated in the sample plot design. The starting point and direction of the plot were randomly selected, with constraints to ensure that the plot would not exit from the land cover type. In the analysis of variance of the various soil and vegetation characteristics, the five land cover types were treated as main effect and the three sample plots as (pseudo) replicates against a null-hypothesis that all data derived from a single population. Where this null hypothesis was rejected (p < 0.05) in an ANOVA in the Genstat 19 software [56], a Duncan test was used to identify significant differences in parameter values (again at p < 0.05) between land cover types. This procedure identifies differences between the sites sampled but cannot separate cause-effect relations from locational effects. Table  2

.3.1. Measurements
To measure the depth of the groundwater table below the plot surface, holes were drilled to a depth of 1.5 m and perforated PVC pipes of 2.5 cm diameter inserted [29]. Manual observations of the depth of groundwater were made at a two-weekly interval for a period of 16 months (September 2017-January 2019).

Simple Water Balance Model
To help in the interpretation of the groundwater data, a simple water balance model (Appendix A) was constructed in which inputs by precipitation (P) are balanced by drainage, evapotranspiration, and changes in groundwater level, with soil porosity modulating the relation between water volume and height change.

Soil Characteristics 2.4.1. Samples for Physical and Chemical Analysis
Soil samples for each of the three replicates of the five land cover types were composited from five locations in the 20 × 100 m 2 sample plot, for three soil layers 0-10, 10-20, and 20-30 cm depth, respectively ( Figure 3). agroforest but was opened 5 years before our study. At the time of sampling, watermelon was the main crop planted for selling at markets in the nearby city of Palangka Raya.
Plot-level data were collected on (1) the fluctuations of groundwater levels; (2) soil physical, chemical, and biological characteristics; and (3) vegetation composition and other characteristics. All measurements were coordinated in the sample plot design. The starting point and direction of the plot were randomly selected, with constraints to ensure that the plot would not exit from the land cover type.
In the analysis of variance of the various soil and vegetation characteristics, the five land cover types were treated as main effect and the three sample plots as (pseudo) replicates against a null-hypothesis that all data derived from a single population. Where this null hypothesis was rejected (p < 0.05) in an ANOVA in the Genstat 19 software [56], a Duncan test was used to identify significant differences in parameter values (again at p < 0.05) between land cover types. This procedure identifies differences between the sites sampled but cannot separate cause-effect relations from locational effects.

Measurements
To measure the depth of the groundwater table below the plot surface, holes were drilled to a depth of 1.5 m and perforated PVC pipes of 2.5 cm diameter inserted [29]. Manual observations of the depth of groundwater were made at a two-weekly interval for a period of 16 months (September 2017-January 2019).

Simple Water Balance Model
To help in the interpretation of the groundwater data, a simple water balance model (Appendix A) was constructed in which inputs by precipitation (P) are balanced by drainage, evapotranspiration, and changes in groundwater level, with soil porosity modulating the relation between water volume and height change.
After drying and sieving, the samples were taken to the laboratory, further ground (to break up aggregates) and analyzed for C org (oxidizable matter in a soil sample based on the dichromate method developed by Walkley and Black [57]), pH(H 2 O) and pH(KCl) (1:1 soil:fluid ratio with distilled water or 1 M KCl, respectively) with standard lab methods [58]. Elsewhere in the sampling plots (Figure 3), undisturbed block samples of 20 × 20 × 10 cm 3 were taken with an iron frame to measure bulk density, particle density, and texture at the same three depth intervals. Porosity was calculated from bulk and particle density.

Reference Level for Organic Carbon
To compare the measured C-org concentrations with what can be expected for a mineral soil in the humid tropics of similar texture, pH, and elevation (proxy for temperature), the pedotransfer of [59] for Sumatra as modified for effects of sampling depth in [60] was used to calculate C ref values.

Earthworm Populations
Earthworm populations were sampled in soil monoliths (50 cm × 50 cm; Figure 4) following the methods developed by the tropical soil biology and fertility (TSBF) network [58,61]. The soil, collected at three depths (0-10, 10-20, and 20-30 cm, respectively), was searched, and all earthworms (or cocoons) were collected, rinsed, and preserved in 70% alcohol for identification in the laboratory, followed by drying and weighing. Results were expressed as the number of individuals and dry weight per unit soil area as well as average weight per individual.
After drying and sieving, the samples were taken to the laboratory, further ground (to break up aggregates) and analyzed for Corg (oxidizable matter in a soil sample based on the dichromate method developed by Walkley and Black [57]), pH(H2O) and pH(KCl) (1:1 soil:fluid ratio with distilled water or 1 M KCl, respectively) with standard lab methods [58]. Elsewhere in the sampling plots (Figure 3), undisturbed block samples of 20 × 20 × 10 cm 3 were taken with an iron frame to measure bulk density, particle density, and texture at the same three depth intervals. Porosity was calculated from bulk and particle density.

Reference Level for Organic Carbon
To compare the measured C-org concentrations with what can be expected for a mineral soil in the humid tropics of similar texture, pH, and elevation (proxy for temperature), the pedotransfer of [59] for Sumatra as modified for effects of sampling depth in [60] was used to calculate Cref values.

Earthworm Populations
Earthworm populations were sampled in soil monoliths (50 cm × 50 cm; Figure 4) following the methods developed by the tropical soil biology and fertility (TSBF) network [58,61]. The soil, collected at three depths (0-10, 10-20, and 20-30 cm, respectively), was searched, and all earthworms (or cocoons) were collected, rinsed, and preserved in 70% alcohol for identification in the laboratory, followed by drying and weighing. Results were expressed as the number of individuals and dry weight per unit soil area as well as average weight per individual. The various measured soil properties were combined into a 'soil quality index' based on a minimum data defined by [62,63] through a Principal Component Analysis (PCA) on datasets for various crops, soil types, and climates. The scores were combined to form a weighted average according to [64][65][66]: where, SQI = soil quality index; Wi = factor weight in the PCA; Si = (normalized) score (0-1) for each indicator i.

Soil Quality Index
The various measured soil properties were combined into a 'soil quality index' based on a minimum data defined by [62,63] through a Principal Component Analysis (PCA) on datasets for various crops, soil types, and climates. The scores were combined to form a weighted average according to [64][65][66]: where, SQI = soil quality index; W i = factor weight in the PCA; S i = (normalized) score (0-1) for each indicator i.

Groundwater Levels
The groundwater levels differed clearly and significantly (p < 0.001) between the land cover types (locations), with the highest average (−33 cm) in FAF and the lowest (−99 cm) in OFA ( Figure 5). The percentage of observations that met the mandated −40 cm groundwater level varied between land uses: 64.7% for FAF, 29.4% for both RAFo and RAFy, 14.7% for RMO, and only 2.9% for AFO.
Across all land cover types, the groundwater depths followed the decreasing rainfall in the February-August period and, with some delay, increased with rainfall in the October-December period ( Figure 6). Five parameters for the model were adjusted to match the pattern averaged across the land uses, with three describing an annual pattern in evapotranspirational demand (Epot) (Epotmax: 7 mm day −1 ; AmplitEpotVar: 0.26; Toffset: −10 day), and two the overall drainage patterns (InterceptEt 0.3; GWdrainfrac: 0.3 per two weeks).
Acceptable fits for this model (given the two-weekly intervals of data collection), were obtained with drain-depths that varied from −40 cm for FAF, RAFo and RAFy, via −60 cm for RMO to −110 cm for OFA ( Table 1). The groundwater depth below which tree water uptake will fall short of demand varied less: from 14 cm for FAF to −57 cm for RMO, suggesting that the drain depth at OFA is deeper than what is desirable for plant growth. In OFA, only 22% of rainfall was used for evapotranspiration according to the water balance model, while at the wetter sites, this was 30-51%.

Soil Physical and Chemical Properties
Bulk density, particle density, porosity, texture, C org , and exchangeable Al all differed significantly (p < 0.01) between the land covers (locations) ( Table 2). Most properties also had significant trends with depth within the 0-30 cm top layer. Nearly all bulk densities were below 1 g cm −3 and porosities relatively high at 55-60%, showing no physical limitations to root development and aeration. The soils contained hardly any sand and the clay-to-silt ratio varied from 3:1 in FAF to 1:1 in RMO and OFA. The C org concentrations of RAF y , RMO, and OFA were more than twice what can be expected as C ref for mineral soils (based on texture, pH, elevation) suggesting a possible history as shallow peat soils. All pH values were in the 3.5-4 range, corresponding with high exchangeable Al concentrations, and a small difference between pH(H 2 O) and pH(KCl) suggesting a low buffering of acidity

Earthworms
All earthworms that could be identified (as they had a developed clitellum) belonged to Pontoscolex corethrurus. The highest population densities (around 60 m −2 ) and biomass (around 35 g m −2 ) and the largest worms (around 0.5 g per individual) were found in the FAF and RAF o plots, and the lowest and smallest in the OFA plots (Table 3). Differences between land covers (locations) and depths were highly significant (p < 0.001), statistically. The largest differences between the soil layers in earthworm presence was found in the wettest (FAF) site.

Soil Quality Indicators
Combining nine measured soil attributes into a single 'soil quality index' (Table 4) showed low values of 0.21-0.25 on a 0-1 scale. Soil quality at the FAF, RAF o , and RAF y sites is considered to be higher than that at the RMO and OFA sites.

Tree Populations, Basal Area, and Tree Biomass
All agroforest plots showed active natural regeneration with an abundance of saplings and poles. Across all trees, the basal area was 25-35 m 2 ha −1 for the FAF and RAF o plots and around 20 m 2 ha −1 for RAF y and RMO (Table 5), with the difference mostly caused by larger populations of trees with DBH > 30 cm (around 100 trees ha −1 ). Aboveground tree biomass was approximately 300 Mg ha −1 in FAF and RAF o plots, and approximately 145 Mg ha −1 in RAF y and RMO. Aboveground biomass is related to basal area, but also depends on wood density and the share of large trees, as the allometric relation involves powers for the stem diameter above 2.0. Table 4. Soil quality index (SQI) rating (0-1) for the various parameters in the land use systems and classification (L = low, VL = very low, M = medium) of the various metrics based on [71]. FAF had a higher calculated biomass and lower basal area than RAF O , for example, as rubber has a medium wood density (0.48 g cm −3 ), and many of the fruit trees have higher values, such as durian (0.61 g cm −3 ), manggis (0.79 g cm −3 ), and rambutan (0.80 g cm −3 ). Table 5. Population densities of seedlings, saplings, poles, and small and large trees in the various land cover types (s.e.d. = standard error of differences).

Species Composition
Ranked by importance index, rubber (Hevea brasiliensis) dominated among the large trees, with cempedak locally known as mangkahai (Artocarpus integer) and kecapi (Sandoricum koetjape) leading among local fruit trees (Table 6).

Diversity Indices
Across the various stages (from seedlings to trees), the highest species richness (32 woody species) was found in RAFo, followed by FAF and RAFy (both 25 species), with the RMO at 10 species but still meeting the 'monoculture' definition of >85% of basal area in the dominant species (Figure 7). The Shannon-Wiener (H') diversity indices were highest for the seedlings and saplings in the understory vegetation (0.45-1.70), followed by the small tree stage (0.0-1.51).

Groundwater Level Fluctuations
Probably the most salient result for peatland rewetting and restoration is the confirmation that the Kaleka agroforest, in either its fruit tree or rubber dominated forms is compatible with a wet peatland hinterland, as it functions at high groundwater tables itself. The fruit agroforest functioned at an average groundwater level of −33 cm, and measurements indicated that, 55 percent of the time, it exceeded the −40 cm level that is mandated for peatlands. The old rubber agroforest had an average groundwater depth of −50 cm and demonstrated that rubber trees can function well under wet conditions. This form of land management through agroforest, developed as part of Dayak Ngaju traditions, is thus compatible with the hydrological goals of rewetting peatland landscapes, understood as hydrological units and including non-peat soils. When peatlands are to be successfully restored, rewetting the peat domes will need to be accompanied by land uses around the peat dome that maintain high groundwater levels.
The estimates of the plot-level water balance based on groundwater fluctuations suggested that drainage dominated over evapotranspiration for most land covers; however, in the May-October period, evapotranspiration exceeded rainfall and groundwater levels dropped. In this period, however, lateral interactions in the landscape are reduced to aboveground (mesoclimatic) effects via air humidity, and, if they occur, fires spreading through dry ground vegetation. Hydrologically, the high groundwater levels in the wet season are directly relevant to the drainage of the hinterland.

Soil Quality
The low pH values (3.5-4) and high exchangeable Al concentrations indicate that the soil, in its current state, would be suitable only for crops with high Al tolerance, or that a high dosage of lime would be needed to allow open-field agriculture to thrive.
The relatively high C-org concentrations for a mineral soil could point at a history as shallow peat soils. Interestingly, the C org /C ref ratio was above two for the three land covers with a shorter use history (RAFy, RMO, and FAF) and at or below one for the oldest Kaleka (FAF and RAFo) plots, suggesting that, over time, there may be a convergence of conditions in mineral soils elsewhere.
The association of earthworms with the most forest-like land covers, rather than with the highest C org concentrations, may be due to the continued and diverse litter layer in these systems that provides food for earthworms [72,73]. Earthworm diversity in these agroforests is, however, lower than that reported elsewhere [74]. Pontoscolex corethrurus is a species with a wide distribution, often associated with agriculture and replacing a more diverse forest fauna [75].
The low soil quality in the peat zones and surrounding mineral soils was known at the time the Mega Rice Project was planned [76], and has subsequently contributed to the failure of large-scale rice production. Tree crops tolerant of acid-soil and wet conditions, such as rubber, are to be preferred on the alluvial soils.

Botanical Composition and Vegetation Structure
The most salient observations here are that the Kaleka agroforests are rejuvenating through natural regeneration, giving the farmers a choice of local fruit (and/or timber) trees that they want to retain and let grow to maturity. The tree species selected for retention and/or cultivation are generally locally adapted, as is the case with studies [77,78] in agroforestry of local communities in Sumatra and Central Kalimantan or the Moluccas [79].
The measured tree diversity indices in the Kaleka plots were comparable to those measured in agroforests in Tumbang Nusa (C. Kalimantan) (1.64; [80]) but much lower than measurements in natural forest elsewhere in Kalimantan: 3.54 in W. Kalimantan [81], 4.17 in Barito Hulu [82], and 6.05 in the primary forest in Besiq Bermai [83]. The data fit the pattern described by [84], that, in agroforestry, tree species richness is approximately proportional to tree biomass (or carbon stocks).
The active regeneration of a wide range of species within the Kaleka agroforests is in contrast with current efforts for peatland rewetting in burnt peatlands where initial species richness remains low [80]. Beyond rewetting the peat domes and blocking existing canals [85], vegetation tolerating high groundwater levels is needed surrounding the peat area as such. As the fire ban applied to both peat and mineral soils has de facto forced an 'outsourcing' of rice as staple food [38], local livelihoods have little choice but to follow a pattern that emerged voluntarily elsewhere [86].

Sustainability
As absolute interpretation of soil quality can be done with reference to specific crops but not across the full range of land use options, considerations of sustainability need to include a time dimension. The Kaleka agroforest may have started when there was still a shallow peat layer on top of the mineral soil, as the soil map for the surrounding landscape indicate, but this cannot be ascertained at the level of sample points. In the transition from sapric peat to an organic matter-rich mineral soil, the C org /C ref ratio was lower in the older than in the younger agroforest sites. So far, however, losses of soil carbon relative to a natural vegetation reference state are not affecting productive use of the land under the agroforest management regime. Most importantly, the mineral soil shows little signs of subsidence and can drain excess water at high groundwater levels without excessive flooding, beyond the tolerance levels of the tree species used. The ability of tree root systems to survive fluctuating and high groundwater levels, probably based on internal aerenchyma, is an important functional trait [87] not shared by species currently promoted by the forestry industry in the area [31].
As indicated in an overview of agroforestry-based ecosystem services [88], the benefits that wider society derives from a well-functioning agroecosystem (the definition of ecosystem services) need to be understood at the level of interacting zones and land uses rather than as standard value per unit area multiplied by the area involved. Maintaining viable and culturally embedded wet land use in the riparian transition zone of peat domes makes it possible to hydrologically restore the peat domes that are threatened by networks of drainage canals constructed with unrealistic expectations of agricultural intensification of the area. By maintaining a diverse and healthy tree population at plot-level, the Kaleka agroforests contribute to a healthy peat hydrological unit, as target of current management. More explicit recognition of the current relevance of these agroforests is needed.

Conclusions
Data on groundwater dynamics and soil quality corroborate results of a recent ethnobotanical study on the indicator and use values of plants in Kaleka agroforests that grow on the riverbank with peat domes as their hinterland. Especially in the oldest fruit agroforest and the old rubber agroforests that may be a century old, internal rejuvenation allows a permanent tree cover and uneven age distribution of trees that mimics a natural forest and contrasts with even-aged plantations. The fruit agroforest functioned at an average groundwater level of −33 cm and measurements indicated that 55 percent of the time it exceeded the −40 cm level that is mandated for peatlands. The old rubber agroforest had an average groundwater depth of −50 cm and demonstrated that rubber trees can function well under wet conditions. This form of land management through agroforest, developed as part of Dayak Ngaju traditions, is thus compatible with the hydrological goals of rewetting peatland landscapes, understood as hydrological units and including non-peat soils. The low pH values (3.5-4) and high exchangeable Al concentrations indicate major constraints to common agricultural crops and suggest that major interventions such as liming would be needed to allow open-field agriculture to thrive. When peatlands are to be successfully restored, rewetting the peat domes will need to be accompanied by land uses around the peat dome that maintain high groundwater levels. The Kaleka agroforests described here have played that role successfully for at least a century. They deserve further attention and recognition.