ChapterPDF Available

Natural Landslides Which Impact Current Regulating Services: Environmental Preconditions and Modeling


Abstract and Figures

Recurrent landslide activity in the natural mountain forest is assumed to be a major factor for maintaining its high biodiversity. It is hypothesized that abiotic–biotic interactions are a prerequisite for natural landslides. A statistical model solely driven by topographic predictors can explain areas prone to landslides but also shows that other factors (e.g., geology, soil, climate, vegetation) than topography might play an important role to improve model performance. Thus, the chapter also shows approaches to derive spatial information on soil properties and wind stress as potential driving predictors for the model. Furthermore, it can be shown that even changes in the biogeochemical cycle and the regulation between nutrient input and biomass production might influence the risk of landslides.
Content may be subject to copyright.
Chapter 12
Natural Landslides Which Impact Current
Regulating Services: Environmental
Preconditions and Modeling
Jo¨rg Bendix, Claudia Dislich, Andreas Huth, Bernd Huwe, Mareike Ließ,
Boris Schro¨der, Boris Thies, Peter Vorpahl, Julia Wagemann,
and Wolfgang Wilcke
12.1 Introduction
Manifold interactions between the abiotic and the biotic environment doubtlessly
exist in the complex biodiversity hotspot of the Rio San Francisco valley. Hitherto,
it is not unveiled how the natural forest and its biodiversity which regulates
(regulating services) the local abiotic conditions (climate, water, soil) is subjected
to feedbacks regarding the preservation of species richness. Different hypotheses
how interactions and feedbacks between abiotic factors and biota contribute to
determine biodiversity are under discussion since decades. Widely accepted in the
J. Bendix (*) • B. Thies • J. Wagemann
Faculty of Geography, Laboratory for Climatology and Remote Sensing, University of
Marburg, Deutschhausstraße 10, 35032 Marburg, Germany
C. Dislich • A. Huth
Department of Ecological Modelling, Helmholtz Centre for Environmental Research – UFZ,
P.O. Box 500 136, 04301 Leipzig, Germany
B. Huwe • M. Ließ
Department of Soil Physics, University of Bayreuth, Universita
¨tsstraße 30, 95447 Bayreuth,
B. Schro
Department of Ecology and Ecosystem Management, Technical University of Munich,
Emil-Ramann-Str 6, 85354 Freising-Weihenstephan, Germany
P. Vorpahl
Department of Ecology and Ecosystem Management, Technical University of Munich,
Emil-Ramann-Str 6, 85354 Freising-Weihenstephan, Germany
Institute of Earth and Environmental Science, University of Potsdam, Karl-Liebknecht-Str.
24-25, 14476 Potsdam-Golm, Germany
W. Wilcke
Geographic Institute of the University of Bern (GIUB), University of Bern, Hallerstrasse 12,
3012 Bern, Switzerland
J. Bendix et al. (eds.), Ecosystem Services, Biodiversity and Environmental Change
in a Tropical Mountain Ecosystem of South Ecuador, Ecological Studies 221,
DOI 10.1007/978-3-642-38137-9_12, ©Springer-Verlag Berlin Heidelberg 2013
group of hypotheses regarding internal feedbacks controlling biodiversity is the
intermediate disturbance hypothesis (IDH) (Connell 1978; Molino and Sabatier
2001) which means that moderate disturbances are fostering the highest degree of
species richness. Roxburgh et al. (2004) stressed that the original specification of
the IDH requires patchy disturbances. Sheil and Burslem (2003) emphasized that
landslides are proven to be one important patchy disturbance type promoting
biodiversity above and below ground. It is meanwhile undisputed that landslides
are a major factor of natural disturbance in the mountain forest of the study area
(Wilcke et al. 2003, Chap. 1). While reducing the overall aboveground biomass,
landslides increase the spatial heterogeneity of biomass distribution and thus create
distinct habitat types (Dislich and Huth 2012). Particularly plant succession after a
landslide and the related above ground species pool of mosses, lichens, vascular
plants like orchids and pioneer tree species contribute to the high biodiversity of the
mountain rain forest and its resilience against natural disturbances (refer to Chap.
8). Below ground diversity and abundance (e.g., AM fungi) might be affected by
landslides, too (refer to Chap. 7).
Profound knowledge on physical interactions between abiotic factors and the
forest that are assumed to trigger landslides is mandatory for predicting landslide
occurrence probabilities and potential future changes.
The basic factors controlling landslide occurrence in the study area are geology
in terms of bedrock material, climate, and topography (Fig. 12.1).
While the geological substrate in the study area is nearly homogenous, the
topographic situation is highly variable. In this context, elevation, slope position,
steepness, and terrain curvature are the most important factors (Sect. 12.2.1,
Fig. 12.1a). Regarding climatic parameters, particularly the abundant rainfall
enhances the weight of vegetation and soil and reduces soil strength. Because
rainfall generally increases with terrain altitude (Chaps. 1,19, and 24) elevation
is a good proxy for rainfall. High wind speed and resulting dynamic pressure
particularly at windward sides at higher altitudes transfer the dynamic stress of
trees into the tree root layer and thus are also expected to be important predictors to
assess landslide risks (Sect. 12.3.3, Fig. 12.1c). Soil conditions (thickness of the
organic and mineral soil layers, soil water logging conditions as indicated by
stagnic horizon occurrence probability) are suspected to play a major role and
should be considered for landslide prediction (Fig. 12.1b), too. Beyond physical
interactions, also chemical interactions might influence the risk for landslides. The
role of a specific abiotic–biotic interaction — the relation between soil nutrient
availability and fine litter production as a proxy for biomass production and thus
vegetation and organic layer weight (Fig. 12.1d) — is discussed in this chapter.
Nutrient availability in the soil as an important control of biomass production
(influencing the weight of the vegetation) and organic matter degradation
(influencing the weight of organic layers) is thus assumed to be an important
predictor for landslide probability.
To disentangle the processes responsible for landslide activity, spatial explicit
models as presented in this chapter are necessary, which are currently based solely
on topographical predictor variables (Sect. 12.2.1, Fig. 12.1a). For a future
improvement of the presented model, further spatial input data of relevant climatic
154 J. Bendix et al.
and soil predictors as described above are required (Sects. Most of
these data were not available when developing the model described in Sect. 12.2.1.
Consequently, this chapter is also devoted to exemplarily present methods to
regionalize point-based soil and climate data.
It should be stressed that landslides in a protected, unused pristine mountain forest
are not a direct ecosystem service (refer to Chap. 4). However, natural landslide
dynamics cause feedbacks to other abiotic and biotic ecosystem components which
give reason to expect impacts on several service levels as, e.g., regulation services.
On the landscape scale, naturally and anthropogenically induced landslides seem to
play a major role in sediment regulation of the catchment, being claimed to be
responsible for a quasi-continuous export of sediment loads independent on precipi-
tation peaks (refer to Chap. 9). On the smaller scale, nutrient regulation is clearly
affected by landslides. Nutrients are removed with the biomass and the organic layers
from the slide area but deposited and concentrated in its foot area (refer to Chap. 11).
Regarding carbon regulation, landslides are characterized by reduced tree growth on
the slides due to the poor nutrient conditions, thus diminishing aboveground carbon
stocks considerably (refer to Chap. 24).
12.2 Methods
12.2.1 The Statistical Landslide Model
Conditions leading to slope failure in the past are likely to cause landslides in the
future as well. Thus, inventories of past landslides combined with topographic
information and thematic maps of controlling factors are used to train statistical
Fig. 12.1 Overview on factors controlling landslide susceptibility in the study area. Arrows
indicate aspects covered by this chapter: The topographic control on landsliding (a), on soil
formation (b), on the distribution of local wind fields (c), and the dependence of organic matter
decomposition, organic layer mass, and biomass dynamics (d). Future model parameters written in
gray are developed and discussed in this chapter
12 Natural Landslides Which Impact Current Regulating Services... 155
landslide models with multiple predictors. Univariate response curves of these
models can provide insights into driving factors of landslides if the following
preconditions are met: (1) The model quality (in terms of performance and calibra-
tion) is sufficient. (2) Consistency between mechanistic assumptions and training
data is maintained. (3) The chosen predictors are interpretable.
(1) Vorpahl et al. (2012) provided a unified framework to train, test and compare
different statistical methods. Applying this framework to eight different
methods from statistics and machine learning (i.e., generalized linear and
additive models, multivariate adaptive regression splines, artificial neural
networks, classification tree analysis, random forests, boosted regression
trees, and the maximum entropy method), they generated weighted model
(2) Vorpahl et al. (2012) maintained consistency between training data and mech-
anistic assumptions by using a subset of five historical landslide inventories of
the RBSF provided by Stoyan (2000) and confined their analysis to landslides
that occurred in an area free of anthropogenic interference (Fig. 12.2). Further-
more, they distinguished different functional units of landslides: i.e., initiation,
transport, and deposition zones. This distinction is of key importance for an
interpretation of univariate model response curves, since linkages between
model predictors and actual mechanisms in the distinct functional units differ.
(3) In a case study, Vorpahl et al. (2012) exclusively used terrain attributes derived
from a digital elevation model (DEM) as predictors: elevation above sea level
(ALT), slope angle, topographic wetness index (TWI), stream power index
(SPI), convergence index (CI), topographic position index (TPI) with two
different radii (100 m and 500 m), and the aspect. To model landslide initiation
as a phenomenon of abiotic–biotic interactions by assessing the importance of
abiotic and biotic predictor values in later applications of the method, spatial
parameter values as presented in the succeeding sections might be helpful.
12.2.2 Potential Model Parameter: Regionalization of Soil
The spatially explicit prediction of histic and stagnic soil horizons is necessary as a
major precondition to understand the landslide dynamics in the study area.
Soil regionalization is based on the general concept (e.g., Jenny 1941) that soil
genesis and, hence, the soils’ distribution throughout the landscape mainly depend
on topography, among other parameters. Therefore, topographic parameters can be
used as predictors to develop digital maps of various soil attributes.
Soil horizons were assessed by 56 soil profiles and 315 auger sampling points.
Key topographical parameters were calculated based on the DEM and implemented
area wide as predictors in the software SAGA GIS. To collect a representative
dataset, sampling sites were selected according to a 24 terrain classes comprising
156 J. Bendix et al.
sampling design along transects extending along side valley slopes (Liess
et al. 2009).
The regionalization as presented here is based on earlier attempts to predict these
horizons (Ließ 2011). In comparison to Ließ (2011), an improvement could be
achieved by focusing on (1) additional terrain parameter selection and by (2)
investigating the dependence on scale as well as (3) the performance of another
recursive partitioning method, Random Forest (RF) (Breiman 2001).
Fig. 12.2 (a) Landslide inventories created by evaluation of aerial photographs of five different
years (i.e., 1692, 1969, 1976, 1989, and 1998) by Brenning (2005) and landslide susceptibility
maps as produced by weighted model ensembles for (b) landslide initiation, (c) transport, and (d)
deposition zones (cf. Vorpahl et al. 2012)
12 Natural Landslides Which Impact Current Regulating Services... 157
(1) Based on the investigations by Ließ (2011), predictors representing climate
(altitude, PISR), water accumulation (curvature, convergence, KRAarea),
water discharge (slope, KRAslope), the insulating effect of the heterogeneous
geomorphology with the ridge—side valley structure in particular (TRI,
normalized height, valley depth) and the wind effect (wind effect, aspect)
were selected to model the soil pattern of this area, all calculated by using
SAGA GIS (Table 12.1,Bo
¨hner et al. 2006).
(2) According to the assumptions of Ließ (2011) that the influence of certain
predictors on soil property development is scale dependent, Brown et al.
(2004) had reported this for the influence of curvature on soil texture, terrain
parameters were calculated for three different GIS raster grid cell sizes (10, 20,
30 m).
(3) Because RF shows a strong dependence on the used dataset used for model
development (Ließ et al. 2012), i.e., the terrain parameters used as predictors
with the soil parameter as response variable, 100-fold RF calculations of the
spatial water stagnation pattern as well as organic layer and stagnic horizon
thickness were carried out. For each of the 100 model runs, the used dataset was
varied by using 9/10 random Jackknife partitions data subsets of the complete
dataset. The 100 models’ prediction results were then averaged and displayed
as two maps: the mean prediction value of the particular soil parameter and its
prediction uncertainty which is represented by the coefficient of variation.
Cross validation is applied to the remaining 1/10 of the dataset, which was
not used to develop the RF models, for model evaluation.
Table 12.1 SAGA modules to calculate terrain parameters
Terrain parameter Module library Module
Altitude Terrain analysis—preprocessing Fill sinks (Planchon/Darboux, 2001)
Slope Terrain analysis—morphometry Slope, aspect, curvature
Profile curvature
Plan curvature
Convergence index Terrain analysis—morphometry Convergence Index (search radius)
Normalized height Terrain analysis—morphometry Relative heights and slope positions
Valley depth
TRI Terrain analysis—morphometry Terrain Ruggedness Index
Wind effect Terrain analysis—morphometry Wind effect
KRAarea Terrain analysis—hydrology Catchment area (flow tracing)
SWI Terrain analysis—hydrology Saga Wetness Index
PISR Terrain analysis—lighting, visibility Potential incoming solar radiation—
direct insolation
158 J. Bendix et al.
12.2.3 Potential Model Parameter: Regionalization of Wind
Regionalization of meteorological point observations facilitates the analysis of
interactions between the abiotic environment and biosphere (e.g., Fries et al.
2009,2012). Strong wind pressure to forest trees might be one reason fostering
landslides and shaping the tree line. Therefore, digital wind speed and dynamic
pressure maps are determined using the following procedure: (1) Statistical analysis
of wind speed observations using the Weibull density function. (2) Calculation of
digital wind speed maps by applying a sheltering factor—algorithm to a DEM. (3)
Validation of calculated wind speed using model-independent meteorological
stations. (4) Calculation of dynamic pressure maps based on the tropical standard
atmosphere and the generated wind speed maps.
(1) Point measurements of hourly wind speed data and the wind direction at 2 m
above surface level for a period of 8 years (1999–2006) for five meteorological
stations (Cerro, ECSF, El Trio, Paramo, TS1, Fig. 12.3) were analyzed regard-
ing mean and maximum wind speed. According to meteorological conventions
(e.g., Weisser 2003), mean and maximum wind speed per 45-wind direction
class are derived from the Weibull density function (50 % and 95 % percentile),
where the parameters of the distribution are estimated by the maximum likeli-
hood method.
(2) Wind speed maps are calculated in three steps: First, data of the station Zamora
and the highest meteorological station (Paramo) are used to calculate a linear
decrease of average and maximum background wind speed with decreasing
terrain altitude. Second, the approach of Winstral and Marks (2002) is used to
derive the maximum upwind slope parameter which is a measure of topo-
graphic shelter or exposure relative to a particular wind direction. The finally
determined shelter factor is multiplied with the background wind speed for
every pixel, providing the digital wind speed maps for every wind direction
(3) Wind speed is extracted for the grid points of the meteorological stations not
used for the regionalization and compared to the modeled data. For the most
stations (e.g., ECSF, Cerro), the correlation is significant and well-suited,
except for the station El Tiro which is known to be strongly influenced by
topographic venturi effects not considered by the regionalization method.
(4) Finally maps of average and maximum dynamic pressure are calculated from
the wind speed maps and average air density where the latter is derived by
blending the tropical standard atmosphere with the DEM.
12 Natural Landslides Which Impact Current Regulating Services... 159
12.2.4 Soil Properties and Litterfall
Between 1998 and 2010 we collected data from 12 sites in the study area (one in
each of the microcatchments (MC) 1 and 3, three in each of MCs 3 and 5 and the
four control sites of Nutrient Manipulation Experiment (NUMEX, for locations see
Chap. 1, NUMEX is explained in Chap. 23). Monitoring in MC2 lasted for 12 years,
in MCs 1, 3, and 5 for 5 years and in NUMEX for 1 year. At each site, mass of
organic layer was determined once by measuring depth and densities of the organic
horizons (Oi, Oe, and Oa) and mass of fine litterfall was determined with three- to
sixfold replicated 0.3 0.3 m
to 0.6 0.6 m
large litter traps in at least monthly
resolution. Furthermore, free-draining litter lysimeters just below the organic layer
were used to collect litter leachate in weekly to fortnightly resolution in which
mineral N (NH
–N + NO
–N) concentrations were determined with a Continu-
ous Flow Analyzer and K, Na, Ca, and Mg concentrations with flame Atomic
Absorption Spectrometry.
12.3 Results and Discussion
12.3.1 Statistical Landslide Modeling
With the exception of classification tree analysis all techniques performed compar-
atively well while being outperformed by weighted model ensembles (refer to
Vorpahl et al. 2012 for details). As expected, models trained on different functional
units of landslides led to different model outcomes (Fig. 12.2).
Fig. 12.3 Univariate response curves (black lines) and predictor importance scores of weighted
ensembles of statistical models. Response quartile ranges are shaded in gray. The curves in each
column show the probability of observing a landslide initiation, transport or deposition zone as a
function of a single predictor variable, i.e., elevation above sea level (ALT), the convergence index
(CI), indicating small scale concavities (CI <0) or convexities (CI >0), the topographic wetness
index (TWI), the slope angle (Slope), the stream power index (SPI), the topographic position index
(TPI), describing the difference between local elevation and the mean elevation within two
different radii of 100 m (TPI.100) and 500 m (TPI.500), respectively, and the direction of the
steepest slope angle (Aspect) (Vorpahl et al. 2012)
160 J. Bendix et al.
Univariate model response curves to changes in predictor values—also called
partial dependency plots (Fig. 12.3)—show that landslide deposition zones tend to
be located at valley bottoms, indicated by high values of SPI and TWI as well as by
negative values of CI and TPI.
Landslides follow the local topography by sliding along shallow ducts in the
slope as indicated by a maximum susceptibility for transport zones at slightly
negative values of CI and TPI. The response curve for initiation zones to changes
in slope indicated an increasing contribution up to ~52. Even steeper slopes lead to
a decrease of landslide susceptibility. This can be attributed to the fact that on
extremely steep slopes the soil layer is usually thinner and hence insufficient for
landslide initiation.
Model response to elevation above sea level exposed an increasing landslide
initiation probability with elevation up to 2,400 m a.s.l. At higher elevations,
landslide initiation probability decreases. Rollenbeck (2006) reported an altitudinal
increase of average precipitation (from about 2,050 mm a
at 1,960 m a.s.l. up to
4,400 mm a
at 3,200 m a.s.l.) in the research area. If rainfall is an important
factor, this should hint towards a positive correlation between precipitation and
landslide susceptibility which contradicts the above finding. As additional factors at
the intersection from dense forest into the Pa
´ramo, lower standing biomass and
lower inclination may strongly reduce landslide formation. Furthermore, Bussmann
et al. (2008) gave a possible explanation for the decreasing landslide susceptibility
at higher elevations by a change in soil substrate from slightly metamorphosed
clayey/sandy sediments, originating from phyllites, at the lower and intermediate
elevations to a more quartzite rich substrate at higher elevations.
Other altitudinal gradients reported for the research area are related to vegeta-
tion. The decrease of average tree heights with higher elevations (Bra
¨uning et al.
2008), for example, may cause a reduced contribution of plant biomass to slope
instability. Smaller trees are less capable of transferring wind forces into the ground
via a turning moment. Soethe et al. (2006a,b) as well as Leuschner et al. (2007)
reported an altitudinal change in tree root structure and in the ratio of aboveground
to belowground biomass. Thus an increase of root contribution to slope stability at
higher elevations can be additionally suspected.
12.3.2 Digital Soil Maps
To predict organic layer thickness, the models based on 20 or 30 m DEM resolution
performed better than those using 10 m. Regarding the prediction of the occurrence
of a stagnic color pattern, all models using 10 m resolution performed better or
equally well than those of lower resolution. Chaplot et al. (2000) found prediction
accuracy to be highly dependent on DEM resolution: Regarding the prediction of
hydromorphic features 10 m DEM resolution outperformed lower resolutions.
Compared to the median r
resulting from CART methodology and a smaller
number of prediction parameters (Ließ 2011), model performance was now
12 Natural Landslides Which Impact Current Regulating Services... 161
improved for all three predicted soil parameters: regarding stagnic horizon occur-
rence probability it was improved by 0.1 (0.6), regarding horizon thickness it was
doubled (0.36), and regarding organic layer thickness it was more than doubled
The digital soil map of the stagnic horizon occurrence probability is shown in
Fig. 12.4a, b. A low coefficient of variation (10 % for >80 % of the area, see
Fig. 12.4b) shows that the dataset is well suited to model the stagnic properties
pattern within this area. The influence of the relative slope position on the occur-
rence probability is clearly visible: The exposed mountain ridges between 2,100
and 2,650 m a.s.l. display a very high probability of stagnic soil properties, >0.8,
which is decreasing down the side valley slopes to a probability of 0.4 (minimum
¼0.2). The flat platform-like areas on top of the ridges, display a particularly high
probability of >0.9. The areas below 2,100 and above 2,650 m a.s.l. are predicted
with an overall lower probability. Below 2,100 m a.s.l. the lower bulk density (Ließ
et al. 2011) and above 2,650 m a.s.l. the coarser soil texture (Ließ et al. 2012)
leads to a higher saturated hydraulic conductivity and therefore less chance for
the development of stagnic soil properties. For the development of the model to
predict stagnic horizon occurrence probability, all terrain parameters were included.
Fig. 12.4 Mean stagnic horizon occurrence probability (a) and thickness (c) with coefficient of
variation (b,d) and mean organic layer thickness (e) with coefficient of variation (f) (Overlaid hill
shading with light source from north)
162 J. Bendix et al.
This confirms the assumption that it is the complex pattern of climate (altitude, PISR),
water accumulation (curvature, convergence, KRAarea), water discharge (slope,
KRAslope), the insulating effect of the heterogeneous geomorphology with the
ridge—side valley structure in particular (TRI, normalized height, valley depth) as
well as the wind effect (wind effect, aspect) which lead to the distribution pattern of
stagnic soil properties within the investigation area.
The model to regionalize stagnic horizon thickness is less stable than the model
to predict the horizon’s occurrence probability. This is indicated by the higher
values of the coefficient of variation in Fig. 12.4d. Ließ (2011) describes similar
results. According to Park and Vlek (2002), soil attributes of which the vertical
distribution is strongly determined by pedogenesis or unknown factors are poorly
modeled by environmental variables. Accordingly, the frequent change of parent
material within one soil profile (Ließ et al. 2012) might be the reason why stagnic
horizon thickness cannot be explained by geomorphology alone. The thickest
stagnic layers >40 or even >60 cm are found along the mountain ridges, with
decreasing thickness while proceeding down side valley slopes.
The low uncertainty of the digital soil map of the organic layer (Fig. 12.4e, f)
indicates a stable model. The thickest organic layers are found on mid-slope
positions, decreasing towards the creeks and towards the crests. Furthermore, altitude
is not among the five most influential predictors of organic layer thickness and there is
no correlation between the occurrence of stagnic horizons and organic layer thick-
ness. This is unexpected because in previous work it was shown that the crests had
usually thicker organic layers than the valley bottom positions in the study area
(Wilcke et al. 2010) in line with reports from a similar forest in Puerto Rico (Silver
1994). Furthermore, studies in Costa Rica (Marrs et al. 1988; Grieve et al. 1990)and
at our study site in Ecuador (Schrumpf et al. 2001; Wilcke et al. 2008a,b)haveshown
that organic layer thickness usually increases with increasing altitude because of
decreasing microbial turnover of organic matter with increasing altitude (Benner
et al. 2010). Table 10.1 shows a general trend towards increasing organic layer
thickness with altitude. However, the transect that was investigated covers a much
larger distance (30 km compared to c. 4 km), and spatial data coverage is therefore
limited. Taking a closer look, the results of Chap. 10 also do not describe any positive
correlation between organic layer thickness and altitude for the altitudinal range
between 1,890 and 3,060 m a.s.l. studied here. Finally, it is assumed that soil
waterlogging limits organic matter turnover (Schuur and Matson 2001;Roman
et al. 2010) which results in the expectation of a positive correlation between the
occurrence of waterlogging (as indicated by stagnic horizons) and organic layer
thickness. However, there is a considerable variation in organic layer thickness at
small scale (Wilcke et al. 2002,2008b) illustrating that none of altitude, topographic
position, and waterlogging alone can explain the entire variability in organic layer
A possible explanation for the seeming contradictions might be that our dataset
is representative for the whole study area and therefore also includes landslide sites
with incomplete organic layers which form an important part of the studied forest
area (Bussmann et al. 2008). Wilcke et al. (2003) have shown that the full
12 Natural Landslides Which Impact Current Regulating Services... 163
regeneration of the organic layer only occurs at the time scale of a few decades. It
also seems likely that waterlogging favors the initiation of landslides because of the
associated high soil weight (Ließ et al. 2011). The results in the literature, in
contrast, usually refer to undisturbed old-growth forest sites. An alternative expla-
nation might be that litterfall rates are lower on crest sites than at lower topographic
positions associated with a smaller accumulation of organic matter on top of the
mineral soil. However, in Sect. 12.3.4 we show for a limited dataset of 12 study
sites that the decrease in litterfall rates is overcompensated by the decrease in
degradation rates resulting in even higher organic layer thickness at low litterfall
rates. We conclude that the relationships of altitude, topographic position, and
waterlogging with organic layer thickness might have to consider the state of
succession after landslide to explain and predict the spatial distribution of organic
layer thickness in the study area.
12.3.3 Digital Wind Maps
Figure 12.6a shows the calculated digital map of maximum wind speed which
reveals spatial structures comparable to the map of mean wind speed (not shown
here). Maximum wind speed increases with altitude but is locally modified by
topographic shelter effects towards the predominant wind direction. Obviously,
steep and narrow valleys and ravines breaching the Cordillera exhibit the lowest
wind speeds (partly close to calm) on a specific altitudinal level. It is striking that
especially the east-facing slopes without any protection by upstream topographic
structures exhibit severe wind speeds up to 17 m s
. The reason is the all-year
dominating circulation from the east (Rollenbeck and Bendix 2011) impinging
particularly the eastern slopes of the Cordillera. The high wind speeds at relatively
low altitudes are a result of the Andean depression (Chap. 1) which allows the
easterlies to affect the upper mountain areas nearly unbridled. By blending the land
use classification of Go
¨ttlicher et al. (2009) with the digital maps of mean and
maximum dynamic pressure, the interaction of wind pressure and trees can be
assessed, e.g., for the tree line ecotone (Fig. 12.5b). The statistical evaluation
clearly reveals that the trees at the treeline of the eastern escarpment exhibit clearly
stronger mechanical exposure than on the western slopes where in the most
situations, wind dynamic pressure falls into the lowest category (mean <5Nm
maximum <20 N m
12.3.4 Chemical Interactions: Soil Nutrients and Litter
There were close positive correlations between nutrient concentrations in soil
solution and annual fine litterfall as proxy of biomass productivity and close
negative correlations between nutrient concentrations in soil solution and mass of
164 J. Bendix et al.
Fig. 12.5 (a) Digital map of maximum wind speed [95 % percentile] determined as an
occurrence-weighted average of eight wind direction classes. (b) Mean and maximum dynamic
pressure depending on aspect along the tree line ecotone
12 Natural Landslides Which Impact Current Regulating Services... 165
organic layer (Fig. 12.6). The effect of bases (K, Na, Ca, and Mg) is more
pronounced than that of N. Consequently, increased nutrient availability resulted
in increased fine litterfall production and—if litterfall is proportional to standing
biomass—increased weight of vegetation. The increased weight of vegetation is
counteracted by decreased weight of organic layer because of a faster turnover at
higher nutrient availability. Expressed in percent of the intercept the response of
litterfall is more pronounced (35 % for N and 139 % for bases) than that of mass of
organic layer (11 and 49 %) suggesting that the standing biomass will more
strongly increase than the mass of the organic layer decrease if nutrient availability
improves. The latter can be expected for the near future because of increased
dryness enhancing release of nutrients from the organic matter by mineralization
and because of rising deposition of reactive nitrogen and possibly also of base
metals because of the shortening of the El Nin
˜o Southern Oscillation (ENSO) cycle
(see Chap. 11). This might imply that the total weight of vegetation plus organic
layer will increase in the near future in response to environmental change thereby
enhancing the risk of landslides.
Nutrient availability in the study area generally decreases with increasing
altitude and at the same altitude is different between valley bottom and ridge top
position (Wilcke et al. 2008b,2010). Furthermore, the frequently occurring shallow
landslides in the study area remove the vegetation and the organic layer resulting in
nutrient loss which is only replenished during a few decades (Wilcke et al. 2003).
Fig. 12.6 Relationship between mean mineral N concentrations (NH
–N + NO
–N) in litter
leachate and (a) mean annual litterfall and (b) mass of the organic layer and between mean sum of
base metal concentrations (charge equivalents of K, Na, Ca, and Mg) in litter leachate and (c) mean
annual litterfall and (d) mass of the organic layer. Mean nutrient concentrations in litter leachate
and mean annual litterfall was determined during 1–10 years depending on the specific site
166 J. Bendix et al.
The latter effect is not included in the relationship between nutrient availability and
litterfall/mass of organic layer in Fig. 12.6 because all our measurements were
taken at old-growth forest sites which were not impacted by landslides in the last
decades. The high variability of altitude and topography in our study area results in
a high spatial variability of nutrient availability and thus also a high ability of
organic layer mass and standing forest biomass together determining the weight on
top of the mineral soil (Wilcke et al. 2002; Moser et al. 2008).
12.4 Conclusion
The presented statistical model ensembles revealed that the occurrence of
landslides is mainly controlled by factors related to the general position along a
slope (i.e., ridge, open slope, or valley). However, there is a clear contradiction
between the altitudinal gradient of rainfall (increasing with altitude) as an assumed
major trigger and landslide probability (decreasing with altitude above 2,400 m a.s.
l.). This indicates that more complex interactions control landslide activity in the
study area which can be explained with a model ensemble purely forced with
DEM-derived proxy predictors. Digital soil maps show a sandier soil texture and
lower soil water logging probability above 2,400 m a.s.l and hence provide a good
explanation. We further assume that variation in above and belowground biomass
mitigating dynamic wind pressure to the forest in the higher parts are major factors
causing these contradictory findings. Thus, it is necessary to provide further spatial
predictor maps related to geology, vegetation biomass, and climate. By additionally
considering predictors related to vegetation, soil and climate, statistical models will
allow for predicting potential future changes in landslide probability patterns.
Dynamic forest models like FORMIND can be used to further quantify effects on
the aboveground biomass production (Chap. 24).
Regarding maps of soil conditions, statistical models based on comprehensive
soil field surveys are applied to spatially predict organic layer and stagnic horizon
thickness as well as stagnic horizon occurrence probability. Forcing parameters are
solely derived from topographical analyses of the DEM. Even if the main influence
of the relative slope position as exposed mountain ridges and flat platform-like
areas on top of the ridges are the best predictors for the occurrence probability of
stagnic horizons, the results point out complex interactions of different factors
behind this. Particularly, the determination of the stagnic horizon thickness is less
stable, most likely due to unconsidered, non geomorphologic factors. For prediction
of organic layer thickness, the degree of succession after landslide might also play
an important role and should be considered besides the well established relationship
of waterlogging, topographic position, and altitude with organic layer thickness.
Digital maps of mean and maximum wind speeds as well as dynamic wind
pressures as additional potential forcing parameters were derived by means of field
observations of wind speed, data on air density, and a DEM by introducing a terrain
shelter factor. It could be shown that dynamic pressure to the forest generally
12 Natural Landslides Which Impact Current Regulating Services... 167
increases with altitude but differs with exposition to the main wind direction.
Because easterly wind directions are predominant, the tree line ecotone on the
eastern slopes is affected by clearly higher wind stress.
Finally, it could be shown that interactions in the biogeochemical cycles might
be relevant for the risk of landslides. Nutrient availability in soil influenced litterfall
production positively and organic layer thickness negatively. An increased nutrient
availability in the future will most likely result in an increase of standing biomass,
thus, enhancing the risk of landslides in response to future environmental change.
Regarding ecosystem services, landslide dynamics will influence different ser-
vice levels. As emphasized in the introduction, landslides are most likely a precon-
dition for the high biological diversity in the mountain forest and thus, directly
related to the cultural services of the forest (Chap. 4). Because the forest structure
characterized by its high species richness properly regulates abiotic processes,
landslides indirectly contribute to the regulating services of the forest (Chap. 4)
too. On the scale of a single landslide, regulation of abiotic parameters changes
significantly. For instance, microclimate (temperature, humidity) regulation is
reduced in comparison to areas sheltered by tree canopies (Fries et al. 2009,
2012; Chap. 9). On this scale also sediment and nutrient regulation are affected.
While sediment and its nutrient is accumulated at the forest edges on the foot of the
slide, also slope wash of matter is higher in young landslides than under natural
forest (e.g., Larsen et al. 1999). On the scale of the forest as a mosaic of trees and
gap areas originating, e.g., from landslides, mass and energy fluxes to the atmo-
sphere are different than those of closed canopies (e.g., Zhang et al. 2007) which
means that landslides maintain the specific regulation of forest–atmosphere
interactions. Also the carbon regulation function of the mountain ecosystem is
determined by the landslide occurrence. Landslides increase carbon turnover and
change the forest composition towards a higher fraction of pioneer species—
however overall forest productivity may be reduced compared to old growth forest
without landslide disturbances due to the unfavourable environmental conditions on
landslide sites (Chap. 24).
Benner J, Vitousek PM, Ostertag R (2010) Nutrient cycling and nutrient limitation in torpical
montane cloud forests. In: Bruijnzeel LA, Scatena FN, Hamilton LS (eds) Tropical montane
cloud forests. International hydrology series. Cambridge University Press, Cambridge, pp 90–100
¨hner J, McCloy KR, Strobl J (2006) SAGA – analysis and modelling application. Go
Geographische Abhandlungen, vol 115. Geographisches Institut der Universita
¨uning A, Homeier J, Cueva E, Beck E, Gu
¨nter S (2008) Growth dynamics of trees in tropical
mountain ecosystems. In: Beck E, Bendix J, Kottke I, Makeschin F, Mosandl R (eds) Gradients
in a tropical mountain ecosystem of Ecuador (Ecological Studies 198). Springer, Berlin,
pp 291–302
Breiman L (2001) Random forests. Machine Learning 45:5–32
Brenning A (2005) Spatial prediction models for landslide hazards: review, comparison and
evaluation. Nat Hazards Earth Syst Sci 5:853–862
168 J. Bendix et al.
Brown DJ, Clayton MK, McSweeney K (2004) Potential terrain controls on soil color, texture
contrast and grain-size deposition for the original catena landscape in Uganda. Geoderma
Bussmann RW, Wilcke W, Richter M (2008) Landslides as important disturbance regimes causes
and regeneration. In: Beck E, Bendix J, Kottke I, Makeschin F, Mosandl R (eds) Gradients in
a tropical mountain ecosystem of Ecuador. Ecological studies, vol 198. Springer, Berlin,
pp 319–330
Chaplot V, Walter C, Curmi P (2000) Improving soil hydromorphy prediction according to DEM
resolution and available pedological data. Geoderma 97:405–422
Connell JH (1978) Diversity in tropical rain forests and coral reefs. Science 199:1302–1310
Dislich C, Huth A (2012) Modelling the impact of shallow landslides on forest structure in tropical
montane forests. Ecol Model 239:40–53
Fries A, Rollenbeck R, Go
¨ttlicher D, Nauß T, Homeier J, Peters T, Bendix J (2009) Thermal
structure of a megadiverse Andean mountain ecosystem in southern Ecuador, and its regional-
ization. Erdkunde 63:321–335
Fries A, Rollenbeck R, Nauß T, Peters T, Bendix J (2012) Near surface air humidity in a
megadiverse Andean mountain ecosystem of southern Ecuador and its regionalization. Agric
For Meteorol 152:17–30
¨ttlicher D, Obrego
´n A, Homeier J, Rollenbeck R, Nauß T, Bendix J (2009) Land cover
classification in the Andes of southern Ecuador using Landsat ETM+ data as a basis for
SVAT modelling. Int J Remote Sens 30:1867–1886
Grieve IC, Proctor J, Cousins SA (1990) Soil variation with altitude on volcan Barva, Costa Rica.
Catena 17:525–534
Jenny H (1941) Factors of soil formation. A system of quantitative pedology. Dover, New York
Larsen MC, Torres-Sa
´nchez AJ, Concepcio
´n IM (1999) Slopewash, surface runoff and fine-litter
transport in forest and landslide scars in humid-tropical steeplands, Luquillo Experimental
Forest, Puerto Rico. Earth Surf Process Landforms 24:481–502
Leuschner C, Moser G, Bertsch C, Ro
¨derstein M, Hertel D (2007) Large altitudinal increase in tree
root/shoot ratio in tropical mountain forests of Ecuador. Basic Appl Ecol 219–230
Ließ M (2011) Soil-landscape modelling in an Andean mountain forest region in southern
Ecuador. PhD Thesis, University of Bayreuth, Bayreuth
Liess M, Glaser B, Huwe B (2009) Digital soil mapping in southern Ecuador. Erdkunde
Ließ M, Glaser B, Huwe B (2011) Functional soil-landscape modelling to estimate slope stability
in a steep Andean mountain forest region. Geomorphology 132(3–4):287–299
Ließ M, Glaser B, Huwe B (2012) Uncertainty in the spatial prediction of soil texture – comparison
of regression tree and random forest models. Geoderma 170:70–79
Marrs RH, Proctor J, Heaney A, Mountford MD (1988) Changes in soil nitrogen-mineralization
and nitrification along an altitudinal transect in tropical rain forest in Costa Rica. J Ecol
Molino JF, Sabatier D (2001) Tree diversity in tropical rain forests: a validation of the intermediate
disturbance hypothesis. Science 294:1702–1704
Moser G, Ro
¨derstein M, Soethe N, Hertel D, Leuschner C (2008) Altitudinal changes in stand
structure and biomass allocaton of tropical mountain forests in relation to microclimate and soil
chemistry. In: Beck E, Bendix J, Kottke I, Makeschin F, Mosandl R (eds) Gradients in
a tropical mountain ecosystem of Ecuador. Ecological studies, vol 198. Springer, Berlin,
pp 229–242
Park SJ, Vlek PLG (2002) Environmental correlation of three-dimensional soil spatial variability:
a comparison of three adaptive techniques. Geoderma 109:117–140
Rollenbeck R (2006) Variability of precipitation in the Reserva Biologica San Francisco/Southern
Ecuador. Lyonia 9:43–51
12 Natural Landslides Which Impact Current Regulating Services... 169
Rollenbeck R, Bendix J (2011) Rainfall distribution in the Andes of southern Ecuador derived
from blending weather radar data and meteorological field observations. Atmos Res
Roman L, Scatena FN, Bruijnzeel LA (2010) Global and local variations in tropical montane cloud
forest soils. In: Bruijnzeel LA, Scatena FN, Hamilton LS (eds) Tropical montane cloud forests.
International hydrology series. Cambridge University Press, Cambridge, pp 77–89
Roxburgh SH, Shea K, Wilson B (2004) The intermediate disturbance hypothesis: patch dynamics
and mechanisms of species coexistence. Ecology 85:359–371
Schrumpf M, Guggenberger G, Schubert C, Valarezo C, Zech W (2001) Tropical montane rain
forest soils: development and nutrient status along an altitudinal gradient in the south
Ecuadorian Andes. Die Erde 132:43–59
Schuur EAG, Matson PA (2001) Net primary productivity and nutrient cycling across a mesic to
wet precipitation gradient in Hawaiian montane forest. Oecologia 128:431–442
Sheil D, Burslem DFRP (2003) Disturbing hypotheses in tropical forests. Trends Ecol Evol
Silver WL (1994) Is nutrient availability related to plant nutrient use in humid tropical forests?
Oecologia 98:336–343
Soethe N, Lehmann J, Engels C (2006a) Root morphology and anchorage of six native tree species
from a tropical montane forest and an elefin forest in Ecuador. Plant Soil 279:173–185
Soethe N, Lehmann J, Engels C (2006b) The vertical pattern of rooting and nutrient uptake at
different altitudes of a south Ecuadorian montane forest. Plant Soil 286:287–299
Stoyan R (2000) Aktivita
¨t, Ursachen und Klassifikation der Rutschungen in San Francisco/
¨decuador. Diploma Thesis, Friedrich-Alexander-Unversita
¨t Erlangen-Nu
Vorpahl P, Elsenbeer H, Ma
¨rker M, Schro
¨der B (2012) How can statistical models help to
determine driving factors of landslides? Ecol Model 239:27–39. doi:10.1016/j.
Weisser D (2003) A wind energy analysis of Grenada: an estimation using the ‘Weibull’ density
function. Renew Energy 28:1803–1812
Wilcke W, Yasin S, Abramowski U, Valarezo C, Zech W (2002) Nutrient storage and turnover in
organic layers under tropical montane rainforest in Ecuador. Eur J Soil Sci 53:15–27
Wilcke W, Valladarez H, Stoyan R, Yasin S, Valarezo C, Zech W (2003) Soil properties on a
chronosequence of landslides in montane rain forest, Ecuador. Catena 53:79–95
Wilcke W, Oelmann Y, Schmitt A, Valarezo C, Zech W, Homeier J (2008a) Soil properties and
tree growth along an altitudinal transect in Ecuadorian tropical montane forest. J Plant Nutr
Soil Sci 171:220–230
Wilcke W, Yasin S, Schmitt A, Valarezo C, Zech W (2008b) Soils along the altitudinal transect
and in catchments. In: Beck E, Bendix J, Kottke I, Makeschin F, Mosandl R (eds) Gradients
in a tropical mountain ecosystem of Ecuador. Ecological studies, vol 198. Springer, Berlin,
pp 75–85
Wilcke W, Boy J, Goller R, Fleischbein K, Valarezo C, Zech W (2010) Effect of topography on
soil fertility and water flow in an Ecaudorian lower montane forest. In: Bruijnzeel LA, Scatena
FN, Hamilton LS (eds) Tropical montane cloud Forests, International hydrology series.
Cambridge Academic Press, Cambridge, pp 402–409
Winstral A, Marks D (2002) Simulating wind fields and snow redistribution using terrain based
parameters to model snow accumulation and melt over a semi-arid mountain catchment.
Hydrol Process 16:3585–3603
Zhang G, Thomas C, Leclerc MY, Karipot A, Gholz HL, Binford M, Foken T (2007) On the effect
of clearcuts on turbulence structure above a forest canopy. Theor Appl Climatol 88:133–137
170 J. Bendix et al.
Full-text available
Weather radar networks are indispensable tools for forecasting and disaster prevention in industrialized countries. However, they are far less common in countries of South America, which frequently suffer from an underdeveloped network of meteorological stations. To address this problem in southern Ecuador, this article presents a novel radar network using cost-effective single-polarization X-band technology, the RadarNet-Sur. The RadarNet-Sur network is based on three scanning X-band weather radar units that cover approx. 87000km² of southern Ecuador. Several instruments, including 5 optical disdrometers and 2 vertically aligned K-band Doppler radar profilers, are used to properly (inter-) calibrate the radars. Radar signal processing is a major issue in the high mountains of Ecuador because cost-effective radar technologies typically lack of Doppler capabilities. Thus, special procedures were developed for clutter detection and beam blockage correction by integrating ground-based and satellite-borne measurements. To demonstrate practical applications, a map of areas frequently affected by intense rainfall is presented, based on a time series of one radar which has been in operation since 2002. Such information is of vital importance for e.g. infrastructure management because rain-driven landslides are a major issue for road maintenance and safety throughout Ecuador. The presented case study of exceptionally strong rain events during the recent El Niño in March 2015 highlights the system’s practicality in weather forecasting related to disaster management. For the first time, RadarNet-Sur warrants a spatial-explicit observation of El Niño-related heavy precipitation in a transect from the coast to the highlands in a spatial resolution of 500 m.
Full-text available
Tropical ecosystems are among the most vulnerable to climate change. Understanding climate impacts on these ecosystems is a primary challenge for policy makers, ecologists, and conservationists today. We analyzed the vulnerability of ecosystems in a very heterogeneous tropical region in southern Ecuador, selected because of its exceptional biodiversity and its ecosystem services provided to people of southern Ecuador and northern Peru. The vulnerability assessment focused on three components: exposure, sensitivity, and adaptive capacity. For the first two components, we identified stressors or drivers of change that negatively influence ecosystems. For the third component, we identified existing and potential buffers that reduce impacts. This process was developed in workshops and by expert elicitation. Representative Concentration Pathway (RCP) scenarios were used, considering RCP 2.6 and RCP 8.5 for a time horizon to 2050. Under the RCP 2.6 scenario, the components of overall vulnerability in the southern region of Ecuador showed very low to moderate vulnerability for most areas, particularly in semi-deciduous forest ecosystems, Amazon semi-deciduous forest, Amazon rainforest, and mangrove forests. These areas had high vulnerability under the RCP 8.5 scenario. A variety of conservation strategies (e.g., protected areas) were shown to increase the adaptive capacity of ecosystems and reduce their vulnerability. We therefore recommend improving these conservation initiatives in ecosystems like dry forests, where the greatest vulnerability is evident.
Full-text available
Free Open and Source Software (FOSS) has sometimes been accused of being viral, lanced in helpless campaigns by few software companies making users wary of choosing free licensed products. The truth is, however, not that the software products are viral but that a worldwide increasing number of programmers, software developers, and, last but not least, scientists are obviously infected by this exciting FOSS virus. The generic and attractive idea beyond proper and clearly defined rights and duties of free software licensed products is the advantage of utilizing and communicating the development community's knowledge base. Doing this, the design of community based software goes back to the core of what science is for, the extension and spreading of knowledge, probably best covered by the German word for science:'Wissenschaft'(ie making or creating knowledge). Although the track record of open source software roots in the'hacker'culture of US computer science laboratories of the early 1970ies, FOSS rapidly evolved and today is a well established component of scientific IT-environments, seldom stand alone but almost everywhere merged with proprietary software. Particularly in those disciplines, which face the challenging task to handle and analyse huge amounts of spatial data, the development and increasing availability of powerful Free and Open GIS fostered and affected many fields of geoscientific endeavour and education. The System for Automated Geoscientific Analyses (SAGA) is a still young but fast growing child of the Free GIS family. SAGA was created and developed by my working-group Geosystem Analysis.
Although soil resources are widely considered as a major factor that reduces the productivity, stature, and diversity of tropical montane cloud forests (TMCF), systematic comparisons of soil resources within and between TMCF are lacking. This study combines published reports on TMCF soils with new data on the soils and forest structure of the Luquillo Mountains in Puerto Rico to assess the current state of knowledge regarding global and local-scale variation in TMCF soils. At the global scale, soils from 33 TMCF sites and over 150 pedons are reviewed. Compared to soils in humid lowland tropical forests, TMCF soils are relatively acidic, have higher organic matter content, and are relatively high in total nitrogen and extractable phosphorus. Across all sites, significant correlations also exist between mean annual precipitation and soil pH and base saturation, but not between any soil chemical factor and canopy height, site elevation, or air temperature. Although comparisons between TMCF are limited by inconsistent sampling protocols, analysis of available data does indicates that lower montane cloud forests (LMCF) have taller canopies, higher soil pH, lower soil nitrogen, and higher C/N ratios than upper montane cloud forests (UMCF). Within an UMCF in NE Puerto Rico, the abundance of soil nitrogen, carbon, and potassium accounted for 25% to 54% of the variation in canopy height. However, as much as 68% of the variation in stand height could be accounted for when site exposure, slope gradient, and the percent coverage of surface roots were also included in the analysis. […]
Tropical montane forests are frequently located on steep slopes with pronounced differences in topographic exposure, related microclimatic conditions and hence in composition and structure of the vegetation over small distances. The objective of this work was to test the hypothesis that topographic position significantly influences soil fertility and water flow in these forests. Soil properties were determined at various topographic positions and water samples of selected ecosystem fluxes analyzed over a 1-year period for oxygen isotopes in three small, steep watersheds under lower montane forest in the Eastern Cordillera of the Andes in southern Ecuador. The soils are subject to lateral material movement (landsliding and solifluction). This, together with the pronounced variation in climatic conditions and vegetation over small distances, resulted in high heterogeneity of soil properties. The pH of the A-horizon ranged between 3.7 and 6.4; concentrations of base metals (calcium, magnesium), sulfur and phosphorus, and trace metals (manganese, zinc) showed enormous spatial variation (coefficient of variation: 358–680% over a surface area of <30 ha). The steepness of the study area and the large contrast in hydraulic conductivities of the organic layer and the mineral soil resulted in a hillslope flow regime dominated by fast lateral flow. During baseflow conditions, δ18O values were similar to that of the sub-soil solution, but rapidly became similar to values in the top-soil solution during rain storms. The chemical composition of stormflows resembled that of the litter leachate. Stormflow had lower pH and higher organic carbon and metal concentrations than did baseflow. […].
In this chapter, the role of nutrient supply and cycling with respect to the characteristically low productivity of tropical montane cloud forests is investigated. Studies of nutrient stocks, turnover rates, and foliar nutrients all suggest that nitrogen supply to vegetation is lower in montane tropical forests than in lowland forests, whereas forest fertilization studies indicate that nitrogen and often phosphorus consistently limit above-ground productivity. Slow rates of nitrogen cycling, rather than low nitrogen inputs, appear to be responsible for the depressed nitrogen supply, and the high soil water content of many cloud-immersed montane forests is likely to be an important ultimate cause of the decreased rates of nitrogen cycling. Hydrological losses of biologically unavailable forms of nitrogen (such as dissolved organic nitrogen) may sustain nitrogen limitation over longer timescales. INTRODUCTION Regardless of location, tropical montane cloud forests (TMCF) worldwide share the same basic differences from lowland tropical forests and montane forests not affected by regular fog: lower productivity and diversity, lower canopy heights, thicker leaves with lower nutrient concentrations (especially of nitrogen), and higher soil organic matter and water content. In this chapter, the importance of nutrient availability in controlling this suite of traits is revisited, summarizing recent information on nutrient distribution, availability, and limitation in TMCF. To the extent that low nutrient availability contributes to the TMCF “syndrome, ” this chapter discusses whether it is an independent factor or a consequence of other factors that ultimately control the productivity of TMCF.
Soil-landscapes are diverse and complex due to the interaction of pedogenetic, geo-morphological and hydrological processes. The resulting soil profile reflects the balance of these processes in its properties. Early conceptual models have by now resulted into quantitative soil-landscape models including soil variation and its unpredictability as a key soil attribute. Soils in the Andean mountain rainforest area of southern Ecuador are influenced by hillslope processes and landslides in particular. The lack of knowledge on the distribution of soils and especially physical soil properties to understand slope failure, resulted in the study of this particular soil-landscape by means of statistical models relating soil to terrain attributes, i.e. predictive soil mapping. A 24 terrain classes comprising sampling design for soil investigation in mountainous areas was developed to obtain a representative dataset for statistical modelling. The soils were investigated by 56 profiles and 315 auger points. The Reference Soil Groups (RSGs) Histosol, Stagnosol, Umbrisol, Cambisol, Leptosol and Regosol were identified according to the World Reference Base for Soil Resources (WRB). While soil profiles and auger points were described in their horizon composition, thickness, soil cohesion, bulk density and texture were analysed in soil profiles only. The prediction of soil parameters was carried out with Classification and regression tree (CART) and Random Forest (RF) method. At this, prediction uncertainty was addressed with hundredfold model runs based on different random Jackknife partitions. Problems with the prediction of the RSGs, likely caused by inconsequence within the WRB (absolute and relative values as decision criteria), resulted in the proposal of “incomplete soil classification”, which relates the thickness of the diagnostic WRB horizons to the upper 100 soil centimetres. Histosol and Stagnosol have been distinguished as dominant RSGs within the inves-tigation area. While Histosol probability depended on hydrological parameters; highest Stagnosol probability was predicted on slopes < 40° and above 2146 m a.s.l. Whether the first mineral soil horizon displays stagnic properties or not, likely depends on physical soil properties in addition to terrain parameters. Incomplete soil classification resulted in histic and stagnic soil parts dominating the first 100 cm of the soil volume for most of the research area. Comparing CART and Random Forest (RF) in their model performance to predict topsoil texture and bulk density as well as mineral soil thickness by hundredfold model runs with random Jackknife partitions, RF predictions resulted more powerful. Altitude a.s.l. was the most important predictor for all three soil parameters. Increasing sand/ clay ratios with increasing altitude, on steep slopes and with overland flow distance to the channel network are caused by shallow subsurface flow removing clay particles downslope. Deeper soil layers are not influenced by the same process and therefore showed different texture properties. Terrain parameters could only explain the spatial distribution of topsoil properties to a limited extent, subsoil properties could not be predicted at all. Other parameters that likely influence soil properties within the investigation area are parent material and landslides. Strong evidence was found that topsoil horizons did not form from the bedrock underlying the soil profile. Parent material changes within short distance and often within one soil profile. Landslides have a strong influence on soil-landscape formation in shifting soil and rock material. Soil mechanical and hydrological properties in addition to terrain steepness were hypothesized to be the major factors in causing soil slides. Thus, the factor of safety (FS) was calculated as the soil shear ratio that is necessary to maintain the critical state equilibrium on a potential sliding surface. The depth of the failure plane was assumed at the lower boundary of the stagnic soil layer or complete soil depth, depending on soils being stagnic or non-stagnic. The FS was determined in dependence of soil wetness referring to 0.001, 0.01, 0.1 and 3 mm/h net rainfall rate. Sites with a FS ≥ 1 at 3 mm/h (complete saturation) were classified as unconditionally stable, sites with a FS < 1 at 0.001 mm/h as unconditionally unstable. The latter coincided quite well with landslide scars from a recent aerial photograph.
Random forests are a combination of tree predictors such that each tree depends on the values of a random vector sampled independently and with the same distribution for all trees in the forest. The generalization error for forests converges a.s. to a limit as the number of trees in the forest becomes large. The generalization error of a forest of tree classifiers depends on the strength of the individual trees in the forest and the correlation between them. Using a random selection of features to split each node yields error rates that compare favorably to Adaboost (Y. Freund & R. Schapire, Machine Learning: Proceedings of the Thirteenth International conference, ***, 148-156), but are more robust with respect to noise. Internal estimates monitor error, strength, and correlation and these are used to show the response to increasing the number of features used in the splitting. Internal estimates are also used to measure variable importance. These ideas are also applicable to regression.
(1) Surface soils (0-15 cm) were collected from six sites in mature undisturbed tropical rain forest along an altitudinal transect (100 m-2600 m) on Volcan Barva, Costa Rica. The rates of nitrogen-mineralization and nitrification were measured under (a) field conditions, where the soils were incubated under the litter layer, and (b) laboratory conditions, where the incubations were done at higher temperatures, but at field moisture content. In addition, a measure of nitrogen-mineralization was made under `improved' conditions, where the incubation temperature was increased, and aeration and moisture content improved. The moisture content, pH and the concentrations of a range of soil nutrients were also measured for each soil. A nutrient amendment experiment was also done, where selected nutrients were added to soils from three altitudes (100 m, 1000 m and 2600 m), and their effect on nitrogen mineralization and nitrification assessed. (2) The moisture content, total nitrogen and carbon, extractable potassium, calcium, magnesium and ammonium-nitrogen increased, and both extractable copper and nitrate-nitrogen decreased, with altitude. (3) Soil nitrogen-mineralization and nitrification both decreased with increasing altitude under field conditions, and both rates were not increased significantly by laboratory incubation, where the temperature alone was alleviated. However, where both temperature and aeration/moisture content were improved, then nitrogen-mineralization rate increased by between one and three orders of magnitude, and a positive relationship was found with altitude of origin. These combined results suggest that nitrogen-mineralization may be limited under field conditions by the high moisture contents found in the montane soils. (4) Partial correlation analysis of these nitrogen turnover rates with other measures of soil chemistry showed that (a) under `improved' conditions the mineralization rate was positively correlated with resource quality and negatively correlated with pH and perhaps reducing conditions, (b) under field and laboratory conditions the mineralization rate was correlated with the reaction product, and (c) nitrification was correlated with the reaction substrate, implying a resource limitation. (5) In the nutrient-addition experiment, ammonium-nitrogen increased nitrogen-mineralization rates at low and high altitudes, but decreased rates at the intermediate altitude. Calcium carbonate addition increased mineralization rates at the low and intermediate altitude, whereas calcium sulphate increased these rates only at the highest altitude. Phosphorus addition had no effect on nitrogen-mineralization at any altitude. (6) Nitrification rates were increased by substrate addition, confirming that the reaction was substrate limited. However, the stimulation was much less than the amount of added substrate, and other limiting factors are clearly important. Calcium sulphate also increased nitrification, but only at the lowest altitude. (7) These findings agree with the general hypothesis that nitrogen-mineralization and nitrification rates are reduced in montane tropical rain forest, and may be a contributory factor controlling growth, species composition and stand structure. The significance of these results for future management of montane tropical rain forest is briefly discussed.
Random forests are a combination of tree predictors such that each tree depends on the values of a random vector sampled independently and with the same distribution for all trees in the forest. The generalization error for forests converges a.s. to a limit as the number of trees in the forest becomes large. The generalization error of a forest of tree classifiers depends on the strength of the individual trees in the forest and the correlation between them. Using a random selection of features to split each node yields error rates that compare favorably to Adaboost (Y. Freund & R. Schapire, Machine Learning: Proceedings of the Thirteenth International conference, ***, 148–156), but are more robust with respect to noise. Internal estimates monitor error, strength, and correlation and these are used to show the response to increasing the number of features used in the splitting. Internal estimates are also used to measure variable importance. These ideas are also applicable to regression.
Shallow landslides are an important type of natural ecosystem disturbance in tropical montane forests. Due to landslides, vegetation and often also the upper soil layer are removed, and space for primary succession under altered environmental conditions is created. Little is known about how these altered conditions affect important aspects of forest recovery such as the establishment of new tree biomass and species composition. To address these questions we utilize a process-based forest simulation model and develop potential forest regrowth scenarios. We investigate how changes in different trees species characteristics influence forest recovery on landslide sites. The applied regrowth scenarios are: undisturbed regrowth (all tree species characteristics remain like in the undisturbed forest), reduced tree growth (induced by nutrient limitation), reduced tree establishment (due to thicket-forming vegetation and dispersal limitation) and increased tree mortality (due to post-landslide erosion and increased susceptibility). We then apply these scenarios to an evergreen tropical montane forest in southern Ecuador where landslides constitute a major source of natural disturbance. Our most important findings are