You are on page 1of 48

Spatially-distributed tracer-aided modelling to explore water and isotope

transport, storage and mixing in a pristine, humid tropical catchment

Joni Dehaspe1,2, Christian Birkel1,3, Doerthe Tetzlaff3,4,5, Ricardo Sanchez-Murillo6, Ana María
Duran-Quesada7 and Chris Soulsby3

1Department of Geography, University of Costa Rica, Costa Rica


2Institute of Environmental Engineering, ETH Zurich, Switzerland
3Northern Rivers Institute, University of Aberdeen, Scotland

4Department of Geography, Humboldt University Berlin, Germany

5Leibniz Institute of Freshwater Ecology and Inland Fisheries, Berlin, Germany

6Stable Isotope Research Laboratory, National University of Costa Rica, Costa Rica
7Center for Geophysical Research, University of Costa Rica, Costa Rica
Corresponding author: christian.birkel@ucr.ac.cr

This article has been accepted for publication and undergone full peer review but has not been
through the copyediting, typesetting, pagination and proofreading process which may lead to
differences between this version and the Version of Record. Please cite this article as doi:
10.1002/hyp.13258

This article is protected by copyright. All rights reserved.


Abstract

Rapidly transforming headwater catchments in the humid tropics provide important


resources for drinking water, irrigation, hydropower and ecosystem connectivity. However,
such resources for downstream use remain unstudied. To improve understanding of the
behaviour and influence of pristine rainforests on water and tracer fluxes, we adapted the
relatively parsimonious, Spatially-distributed Tracer-Aided Rainfall Runoff (STARR) model
using event-based stable isotope data for the 3.2 km2 San Lorencito catchment in Costa
Rica. STARR was used to simulate rainforest interception of water and stable isotopes,
which showed a significant isotopic enrichment in throughfall compared to gross rainfall.
Acceptable concurrent simulations of discharge (Kling Gupta Efficiency, KGE~0.8) and stable
isotopes in stream water (KGE~0.6) at high spatial (10m) and temporal (hourly) resolution
indicated a rapidly responding system. Around 90% of average annual streamflow
(2099mm) was composed of quick, near-surface runoff components, while only ~10%
originated from groundwater in deeper layers. Simulated actual evapotranspiration (ET)
from interception and soil storage were low (~420mm/yr) due to high relative humidity
(average 96%) and cloud cover limiting radiation inputs. Modelling suggested a highly
variable groundwater storage (~10 to 500mm) in this steep, fractured volcanic catchment
that sustains dry season baseflows. This groundwater is concentrated in riparian areas as an
alluvial-colluvial aquifer connected to the stream. This was supported by rainfall-runoff
isotope simulations, showing a "flashy" stream response to rainfall with only a moderate
damping effect and a constant isotope signature from deeper groundwater (~400mm
additional mixing volume) during baseflow. The work serves as a first attempt to apply a
spatially distributed tracer-aided model to a tropical rainforest environment exploring the
hydrological functioning of a steep, fractured-volcanic catchment. We also highlight
limitations and propose a roadmap for future data collection and spatially distributed
tracer-aided model development in tropical headwater catchments.

Key words: humid tropics, Costa Rica, ReBAMB, tracers, tracer-aided modelling, stable
isotopes.

This article is protected by copyright. All rights reserved.


1. Introduction

Headwater catchments provide important water resources for downstream uses such as
irrigation, drinking water supply and hydropower. Protecting such ecosystem services is a
key priority in sustainable land and water resource management. In tropical headwater
catchments, intense precipitation events often cause flood events, landslides and other soil
mass movements like debris flows (Capra et al., 2003). Tropical catchments are therefore
characterized by high erosion rates and rapid landscape evolution and water quality issues
are often a concern. Anthropogenic land use transformations from forests to pasture and
urbanization often exacerbate these landscape-forming effects (Wohl et al., 2012), making
the tropics one of the fastest changing environments on earth (Gibbs et al., 2010).
Moreover, Giorgi (2006) identified Central America as the principal tropical climate change
hotspot due to the narrow land mass sitting between two major water bodies (Pacific Ocean
and Caribbean Sea). Increasing temperature extremes have been identified for Central
America and are linked to drying trends resulting in a risk for tropical ecosystems (Aguilar et
al., 2005).

Observed climatic changes such as increased air and sea surface temperatures and lower
gross precipitation are expected to intensify in the near future throughout the region
(González et al., 2017). Nonetheless, addressing the projected impacts of climate change on
tropical catchment hydrology with adaptation strategies is hindered by limited quantitative
information on flow pathways, sediment sources and transport dynamics linked to water
quality. This knowledge gap is mostly related to the lack of high spatio-temporal resolution
hydrometric and tracer datasets and appropriate modelling tools to assess and predict the
spatio-temporal variability in hydrological processes that generate water and solute fluxes in
the tropics.

Tropical headwater catchments often have volcanic origins and exhibit stratified volcanic
rocks and soils overlying steep terrain, hence they are subjected to a complex suite of non-
linear and preferential hydrological processes. Generic conceptual models used to predict
streamflow in humid tropical catchments include, among others HBV-light (Birkel et al.,
2012; Beck et al., 2013) and TOPMODEL (Campling et al., 2002; Moličová et al., 1997), both
showing good prediction performance but a limited capability to assess streamflow sources.

This article is protected by copyright. All rights reserved.


More physically-based and complex models such as SWAT were also widely applied for
simulating hydrologic behaviour of tropical catchments (Pereira et al., 2016 for an example
from Brazil). However, Kirchner (2006) pointed out that successful hydrograph simulations
for smaller catchments can be misleading and often result in over-fitted complex models. In
contrast, the development of parsimonious rainfall-runoff models that give "the right
answers for the right reasons" with a minimal parameterization remains a challenge
(Kirchner, 2006). When addressing the quest for internally consistent and parsimonious
models, it has become increasingly important in catchment hydrology research to look
beyond the hydrograph and identify contributing water sources, flow paths and transit
times (McDonnell et al., 2007; Kuppel et al., 2018).

Tracer data has proven to be particularly useful for informing hydrological models (e.g.
Tetzlaff et al., 2007), mainly when assessing the temporal dynamics of water and tracer
transport and storage (Hrachowitz et al., 2013). More recently, conservative tracer data has
been incorporated into tracer-aided rainfall-runoff models (TAM) to simultaneously
calculate the hydrological response time and transit time of catchments (Birkel and Soulsby,
2015; Soulsby et al., 2006). The latter requires differentiating water celerity from water
velocity (McDonnell and Beven, 2014). The celerity controls the direct influence of
precipitation on the hydrograph and is also known as the catchment response time, while
water velocity is related to the much slower transport of water particles which equates to
the transit time (McGuire and McDonnell, 2006). Although adding complexity and
computation time, tracers such as stable water isotopes can be used to test model
structures (McMillan et al., 2012) and constrain rainfall-runoff model parameters thus
facilitating model evaluation as shown by Birkel et al. (2014). Also, coupled, dynamic water
and tracer transport models can be used to bridge the gap towards water quality
assessments (Hrachowitz et al., 2015). Moreover, TAMs have demonstrated added value in
simulating ecohydrological water partitioning (Knighton et al., 2017). However, Kirchner
(2016) showed a potential bias in transit times calculated with lumped conceptual TAMs in
heterogeneous catchments. This further emphasizes the need for spatially distributed
model development and more cautious process aggregation over spatial scales.
Furthermore, Birkel and Soulsby (2016) have highlighted that TAMs were mostly applied in
rather well-monitored northern latitude catchments and stressed the need for hydrological

This article is protected by copyright. All rights reserved.


studies using tracer data in combination with rainfall-runoff models carried out on a larger
set of more geographically and hydroclimatically diverse catchments.

To further advance tracer hydrology research in terms of using TAMs in the humid tropics
(Birkel and Soulsby, 2016), in this study we present an investigation of the Reserva Biológica
Alberto Manuel Brenes (ReBAMB), located in North-Eastern Costa Rica. The ReBAMB, a
biodiversity hotspot, is only accessible for research and education and has undergone a
minimal anthropogenic influence. The small 3.2 km2 headwater catchment of the San
Lorencito river was equipped with a climate station registering high resolution
meteorological data since 2008. From around 2013, the availability of suitable measurement
equipment and funding increased, allowing for improved isotopic and streamflow data
collection. Together with routine hydrometric measurements (water level and discharge),
these tracer data provide a unique opportunity for enhancing tracer-aided hydrological
modelling of a pristine (close to 100% primary rainforest cover) humid tropical catchment.
Nonetheless, ReBAMB's dynamic landscape, with high intensity rainfall inputs and landslide-
prone hillslopes, dictates a spatial heterogeneity in potentially landscape-forming
hydrological processes. Thus, a spatially-distributed approach, in contrast to lumped
models, accounting for the dense vegetation that partitions and re-distributes water in
terms of interception, throughfall and evapotranspiration is deemed essential for assessing
hydrological dynamics in this area (Kumagai et al., 2016).

This paper aims to compile, integrate and process hydrometric and isotopic data from this
small (3.2 km2) humid tropical San Lorencito catchment, located inside the ReBAMB, to
serve as input for a Spatially-distributed Tracer-Aided Rainfall Runoff model, STARR (van
Huijgevoort et al., 2016a and b). The model will be adapted according to the existing
preliminary perception of the catchments hydrological functioning for this tropical
environment and used to gain insights into how water is potentially transported, stored and
mixed.

The specific objectives of the study were:

i) To compile a consistent biogeophysical, hydrometeorological and isotopic data set


useful for application in a tracer-aided rainfall-runoff model.

This article is protected by copyright. All rights reserved.


ii) To develop a perceptual model of the hydrological functioning based on the
compiled data set of the San Lorencito study catchment and to inform model
adjustments particularly for evapotranspiration, interception and runoff generation
of this humid tropical rainforest environment.
iii) To adapt and apply the STARR tracer-aided rainfall-runoff model as a preliminary
test to the study site exploring hydrological and isotopic dynamics through tracing
their fluxes, storage, mixing and flow pathways.

2. Study catchment
The Reserva Biologica Alberto Manuel Brenes (ReBAMB) is an 80 km2 protected biological
reserve of primary humid tropical rainforest located in central Northern Costa Rica (Figure
1A). It includes the 3.2 km2 densely vegetated San Lorencito study catchment which drains
towards the Caribbean Sea (to the East). Although situated East of the continental divide,
both the Northern Pacific Ocean and the Caribbean Sea climatically affect this catchment
(Sanchez-Murillo et al., 2013). The elevation in the catchment ranges from 874 m to 1472
masl with steep slopes of 22%, on average (Figure 1 and Table 1). Based on the range of
elevation, the study site classifies as a transitional tropical rainforest situated between the
higher elevation cloud forests (>1800 masl) and lower pre-montane and lowland rainforests
below 800 masl.

The humid tropical climate of San Lorencito is characterized by high annual rainfall and a
constant temperature; these yield on average 2589 mm (2013-2016) and almost 20°C,
respectively. Annual discharge averages 2099 mm and potential evapotranspiration is
estimated at around 500 mm per year (Table 1). The relative humidity is constantly high at
around 96%. The monthly precipitation regime shown in Figure 1C indicates a wet season
(May-November) and a moderate dry season (December-April). Despite moderate
seasonality, convective rainfall events are common throughout the year. The climate is
further influenced by the El Niño Southern Oscillation (ENSO) with the ENSO warm phase (El
Niño) producing drier conditions on the Pacific slope and at the same time wetter conditions
in the Caribbean (Duran-Quesada et al., 2010; Waylen et al., 1996). The ReBAMB receives
seasonally orographic rainfall, but the local surface conditions (e.g. vegetation) dominate
the local rainfall input.

This article is protected by copyright. All rights reserved.


The study site is part of the Cordillera Tilaran which was formed through volcanic activity
during the upper Pleistocene and lower Tertiary (Bergoeing, 2007). However, the catchment
geomorphology is the product of erosion (Denyer and Kussmaul, 2000) and the underlying
geology is composed of igneous basalts, andesites and pyroclastic deposits. The mobile San
Lorencito riverbed contains variably sized sediments, including many large clasts and a high
bed roughness. Furthermore, at the monitoring station indicated in Figure 1B, a narrow
floodplain is visible (~30 m wide), directly connected to steep (> 30%) and densely vegetated
hillslopes (Figure 1D).

The soil types range from alluvial, less developed sandy Entisols close to the river bed to
deeper (>1 m) volcanic soils (Andisols) on the hillslopes that are rich in organic matter (~
10% to 20%). Additionally, volcanic ash layers are often found in the upper soil profiles (0-
0.5 m) (Solano et al., in review). Soil mass movement is common in the San Lorencito
catchment and initiated by the intense rains that eventually saturate the steep hillslopes,
likely producing near-surface runoff. The dominant rainforest trees (up to 40m high), giant
palm trees (Lauraceae, Ticodendraceae ~20m high) and bushes make up about half of the
vegetation, with the other half consisting of lianas and epiphytes (Orchidaceae,
Bromeliaceae, Araceae) (Salazar-Rodríguez, 2003). Hoelscher et al. (2004) stress the
importance of epiphytes, a major vegetation characteristic form of montane cloud forests,
for the interception component of the hydrological cycle of rain forests. Lastly, it has been
generally assumed that the vegetation in this catchment is uniformly distributed in space
and is of a pristine nature. Little is yet known about the San Lorencito streamflow dynamics,
sources and composition. Since the start of continuous measurements in June 2015, the
perennial stream maintains a constant baseflow during drier periods (little or no rain) at the
gauged cross section of the catchment outlet, presumably due to constant groundwater
supply. Precipitation events cause a rapid response in the stream with an overall catchment
response time of around 40 minutes.

Sanchez-Murillo and Birkel (2016) and Rhodes et al. (2006) showed that Costa Rica receives
moisture inputs from the Caribbean Sea and the Pacific Ocean with a distinct isotopic
signature. The Caribbean coast exhibits a more enriched rainfall of around -10‰ for
deuterium, while the moisture input on the Pacific slope is usually less enriched in

This article is protected by copyright. All rights reserved.


deuterium (up to -50‰). Furthermore, isotopic depletion is related to high rainfall amounts
and topographic Rayleigh-distillation processes. Based on moisture inputs from both the
Caribbean and the Pacific Ocean, the San Lorencito exhibits a mean precipitation isotope
signature of around -20‰ to -30‰ (Table 1).

3. Field work and monitoring


3.1 Data compilation
A Davis Vantage Pro climate station (N10°13'12.0", W84°35'13.3") located 1 km outside the
San Lorencito catchment at 967 masl registered 30-minute wind speed, humidity, incident
solar radiation, temperature and precipitation. The station complied with the World
Meteorological Organisation (WMO) standards stipulating sensor placement 2 m above the
ground with the exception of wind speed, which was measured at 10 m. Moreover,
clearance of vegetation within 25 meters of the station ensured minimal disturbance. The
meteorological data were available from 2008 until 2016 (Table 1). A sensor box installed at
the catchment outlet indicated in Figure 1B measured water level using a Global Water GL-
500 data logger and pressure transducer sensor (among other water quality parameters) at
5-minute intervals since June 2015. Manual streamflow gaugings were carried out roughly
once per month since 2013 to construct a preliminary rating curve (R2 =0.98) for discharge
calculations. Here, average stream velocities were measured at 50 cm intervals over the
same river cross section using a current meter (Global Water). Event (15-minute intervals)
sampling campaigns of streamflow at the gauged cross section, above canopy (gross)
precipitation, throughfall and sporadic stream tributaries at baseflow (from 2013)
supplement the basic hydrometric data. The streamflow samples were analysed for
deuterium (δ2H) and oxygen-18 (δ18O) at the Chemistry School of the National University in
Heredia, Costa Rica using a Cavity Ring Down Spectroscopy (CRDS) water isotope analyser
L2120-I (Picarro, CA) and at the Northern Rivers Institute at the University of Aberdeen with
a Los Gatos DLT-100 isotope analyser. The analytic precision of the instruments was 0.1‰
for oxygen-18 and 0.6‰ for deuterium.

3.2 Data quality

This article is protected by copyright. All rights reserved.


An extreme event (~ 50-year return period) produced water levels up to almost 3 m in
August 2015 resulting in an uncertain rating curve. Despite this, the stage during this
extreme event could be inferred accurately from flood marks (Solano et al., in review).
However, a rating curve and model evaluation that uses such a single extreme value is likely
to produce biased results towards the outlier while neglecting most of the other observed
hydrodynamics. We therefore used the 65 available manual gaugings ranging from the
lowest flows (0.09 m3/s) to medium sized flows (up to 3.5 m3/s) for model evaluation
directly.

In general, climate data capture only resulted in <10% data gaps, which, in the case of
temperature, were infilled using a sine wave algorithm with distinct frequencies that can
approximate diurnal and hourly temperature fluctuations (Equation 1). The parameters a, b
and c were calculated using non-linear least squares fitting in R (R core team, 2016) and
equal to 24, 2 and 20, respectively. For wind speed, solar radiation, humidity and
precipitation gaps, an "average year" of hourly measurements was created by averaging
each available measurement over the corresponding years. This ‘average year’ was then
used to fill in data gaps.

2𝜋𝑥 2𝜋𝑥
𝑦 = 𝑠𝑖𝑛 ( ) + 𝑠𝑖𝑛 ( )+𝑐 Eq. 1
𝑏 𝑎

4. Model development
4.1 Spatial input data
For spatially-distributed modelling, ArcMap 10.4.1 was used together with the PCRaster
Python environment for map calculations (Karssenberg et al., 2010). The difference between
a Digital Elevation Model (DEM) and Digital Surface Model (DSM) represents the vegetation
height. The DEM and DSM were calculated from a multi-spectral WorldView 2 and SPOT 7
satellite image from GeoSpatial Innovations, CR and were available at a spatial resolution of
5 m and 12 m, respectively. To link the catchment topography to the hydrological dynamics,
the Topographical Wetness Index (TWI) on a 10-m grid (Beven and Kirkby, 1979) was used to
distribute runoff generation mechanisms in the newly modified STARR model. The TWI was

This article is protected by copyright. All rights reserved.


found to correlate well with hydrological variables such as depth to groundwater and field
capacity in steep catchments (Sørensen et al., 2006). The spatial distribution of TWI was
used to distinguish the water holding capacity of the riparian zone near the floodplain from
the hillslopes supported by soil sampling and measurements (30 soil samples indicated
shallower and coarse textured soils with less water holding capacity closer to the riparian
zone compared to deeper and more porous volcanic soils on the hillslopes; Solano et al., in
Review). The TWI ranged from 2 to 16 (Figure 2A). Lower values were mainly found on the
hillslopes while higher values were detected near the river network. For the present model,
the median TWI of 6.5 was selected to separate the hillslope and valley landscape units to
conceptualize the different runoff generation mechanisms of those areas in the model.

To account for the role of vegetation on interception processes in the catchment, we used
the Leaf Area Index (LAI). This dimensionless biophysical variable is defined as leaf area per
unit ground surface (Zheng and Moskal, 2009) and was extracted from the SPOT 7 image (12
m grid) based on the Normalized Difference Vegetation Index (NDVI). After
orthorectification and atmospheric correction of the satellite image, the LAI was calculated
using Equation 2 by Saito et al. (2001). In absence of ground measurements and a similar
range of LAI, such a LAI – NDVI correlation, originally derived for deciduous broadleaf forest,
was used as a first approximation.

𝐿𝐴𝐼 = 0.57𝑒 2.33𝑁𝐷𝑉𝐼 Eq. 2

Figure 2C shows that the Leaf Area Index (LAI) calculated for each 10 m2 grid cell of the San
Lorencito catchment ranged between 1.3 and 5.8 which is line with values of old-growth
forest (5.6) at La Selva research station in Costa Rica measured with a terrestrial LiDAR by
Tang et al. (2012). As values below 1 were not detected, this would indicate all precipitation
is initially intercepted. However, field observations showed canopy gaps and for modelling
purposes it was therefore assumed that 5% of incoming rain falls directly on the ground.
From Figure 2C it is visible that the northern hillslope had an overall higher vegetation

This article is protected by copyright. All rights reserved.


density likely related to the lower mean slope compared to the southern hillslope. A local
slope map is included in Figure 2B for comparison. Here, the river forms a clear boundary
between the north and south hillslopes of the catchment or right and left hillslopes,
respectively. The LAI influences the maximum canopy storage capacity, S (Table 2, Figure S3)
which tends to increase on lower slope areas. In general, the vegetation height (Figure 2D)
seems uniformly distributed with values ranging from close to 0 m at the river network to 25
m in the centre of the catchment. This parameter mainly plays a role in the calculation of
roughness that is used for estimating evapotranspiration potential in the catchment. All
maps were aggregated and re-sampled to a consistent 10 m grid size for modelling purposes
(Figure 2).

4.2 Hydrometric and isotopic input data


The combination of meteorological, hydrological and isotope data rendered an integrated
dataset (2008 until 2017). The data was uniformly aggregated to hourly time steps.
Precipitation, streamflow, throughfall, baseflow and tributary isotope signatures (δ18O and
δ2H) were compared against the Global Meteoric Water Line (GMWL) (δ2H =8 δ18O +10) as
proposed by Craig (1961). The isotope deuterium was preferred over oxygen-18 for
modelling and visualization due to its higher analytical precision and wider range, but both
isotopes showed similar behaviour (R2 = 0.92).

The TAM was applied at hourly timesteps and thus required a continuous hourly input, both
for incoming precipitation and precipitation isotope (deuterium) signatures. A multiple
Linear Regression (MLR) model was developed for interpolation of such a continuous hourly
precipitation deuterium input series based on available event measurements. The MLR
revealed an empirical relation between the deuterium signatures in gross precipitation
(Pgross) and the San Lorencito climatic variables. Selection of the appropriate climatic
predictors (while avoiding redundancy) was achieved via a forward selection procedure
implemented in the R package packfor (Dray et al., 2009). The forward algorithm penalizes a
higher number of predictors using the adjusted coefficient of determination adjR 2 (Equation
3). Here, the r2 is the coefficient of determination and n and p, respectively, the number of
observations and parameters. The input isotope signature was assumed uniform in absence

This article is protected by copyright. All rights reserved.


of more detailed measurements due to the small catchment size (3.2 km2) and only a
moderate elevation gradient (~400 m).

𝑛−1
𝑎𝑑𝑗𝑅 2 = 1 − (𝑛−𝑝−1) (1 − 𝑟 2 ) Eq. 3

4.3 STARR model for the humid tropics


The original Spatially-distributed Tracer-Aided Rainfall Runoff (STARR) model (van
Huijgevoort et al., 2016a) incorporated stable isotope tracers within the structure of a
conceptual HBV-light type rainfall-runoff model (Bergström and Forsman, 1973; Seibert and
Vis, 2012). The PCRaster Python software was adopted for its development and model input
and output was analysed in R. STARR was originally developed for a Scottish catchment on a
rather coarse 100 m x 100 m grid and daily timescales. A recent adaptation of STARR for
smaller catchments has adopted finer grids (25 m), though the timestep was again daily
(Ala-aho et al., 2017b). For application in the topographically complex San Lorencito
catchment a much finer 10 m x 10 m grid and higher resolution hourly time steps were
needed, which required modifications to the model structure. The conceptual flowchart of
the modified model is presented in Figure 3, the algorithms used are summarized in Table 2
and the model parameters in Table 3. The subsequent paragraphs discuss each of the
modules of the STARR model and highlight deviations from the original model structure
used by van Huijgevoort et al. (2016a).

Interception and evapotranspiration:

The evapotranspiration for each gridcell of the catchment was calculated according to the
FAO standard approach (Allen et al., 1998) with the Penman-Monteith equation shown in
Table 2. Penman-Monteith was evaluated with and without the surface resistance and
coupled to the interception and soil storage module. The computed evaporation represents
the potential loss for a certain atmospheric demand subject to water availability.
Furthermore, it was assumed that ground radiation loss cannot be neglected for an hourly
time step and is equal to a fraction of the incoming net radiation. Moreover, net longwave

This article is protected by copyright. All rights reserved.


radiation flux during night-time was set equal to the net longwave radiation two hours
before sunset on that day.

The presence of the dense and often epiphytic vegetation in the San Lorencito catchment
indicated that interception is likely to be an important component of the local hydrological
system (Muzylo et al., 2009; Loescher et al., 2002). The Rutter model (Rutter et al., 1971 and
1975) (Table 2) was considered the preferred option over the Gash interception model
originally used in in STARR (Gash et al., 1995) as it has a running water balance. Also, in
contrast to the Gash model, Rutter does not assume the canopy to fully dry between events.
According to Hoelscher et al. (2004) this is a reasonable assumption for the study
catchment. The rate of evaporation and drainage of intercepted water additionally depends
on the availability in the interception store and is reduced when the maximum canopy
storage capacity S is not reached. Table 2 shows how the LAI allows deduction of the
maximum canopy storage capacity. If the interception store is sufficiently full Dunn and
MacKay (1995) and Shuttleworth (1978) suggested that the surface resistance can be
neglected in evapotranspiration calculations. Finally, the stemflow component proposed by
Rutter was not explicitly implemented. It was postulated that the water stored on the stems
is part of the total interception storage with evaporation and drainage fluxes happening at
the same rate. The drainage parameters of the Rutter model Ds and b were arbitrarily
initialized and subsequently calibrated. Their units and initial range can be found in Table 2.

The above procedure only applies to the incoming rain at canopy level. Figure 3 shows that
a certain fraction of the rain will hit the ground directly. This fraction is determined by the
Canopy Gap Fraction and was set to 5% of the incoming precipitation based on empirical
estimation of canopy gaps.

Soil component:

The soil store describes the re-distribution of water from the upper soil layer. The depth or
water holding capacity of this is labelled FC (calibrated parameter) and depends on the
location of the grid cell considered; near the stream or on the hillslopes. This distinction is
based on the TWI as described above. Gross rainfall and throughfall potentially constitute an
almost instantaneous surface runoff Qs when saturation of the soil store occurs after a high

This article is protected by copyright. All rights reserved.


intensity storm event. The remaining water entering the soil is subject to evaporation Sevap
and Seepage to the groundwater store. Discharge QSTO can be generated as formulated in
Table 2 and a capillary flux that flows from the groundwater store back to the soil store
according to a moisture gradient. Evaporation from the soil store Sevap was limited according
to water availability using the threshold LP (Table 2). Furthermore, Sevap, does not include
canopy resistance, rs ,and the total evaporation in the interception and soil store does not
exceed the potential evaporation.

Groundwater:

The groundwater storage is unconfined as shown in Figure 3. It exchanges seepage and


capillary fluxes with the soil store, and behaves like a linear reservoir generating discharge
QGW (Table 2). The steep terrain in the catchment required a lateral flow component.
Therefore, an approximation of the horizontal component of the Darcy equation (Darcy,
1856) for saturated porous media was applied (Table 2). Moreover, the hydraulic head is
estimated by the slope of the sum of the DEM and the amount of water in the groundwater
store (m). The saturated flow Ksat was determined through calibration (Table 2). The water
volume available for lateral flow of each grid cell is routed to a neighbouring cell in
accordance with the time-constant topography-based local drain direction (ldd). The total
discharge at the catchment outlet Qtot was simply calculated as the sum of all stream
velocities in the channel network per timestep. The water celerity was fixed at 0.7 m/s or
252 grid cells per hour travel time based on measured stream velocities at the catchment
outlet. This is considerably higher than the hourly model time step avoiding numerical
instability and a reasonable estimate according to van Huijgevoort et al. (2016a).

Coupled isotope calculations:

STARR is a TAM and thus tracks tracer concentrations in addition to water fluxes. Isotope
signatures i are calculated through mass balance equations and retained at the end of each
time step (Figure 3). An overview of the isotope calculations can be found in Table 2.
Isotope fractionation from the interception storage is possible due to the preferential

This article is protected by copyright. All rights reserved.


evaporation of the "lighter" isotopes. Fractionation in the interception module was
conceptualized as an empirical relationship based on a simple linear regression of deuterium
signatures in gross rain and in throughfall (Table 2). Furthermore, the calibrated additional
mixing volumes or passive storages STOpas and GWpas (Birkel et al., 2011) were distributed
using the TWI differentiating hillslopes and valley bottom to avoid additional model
parameters (Figure 3).

4.4 Model evaluation


The initial parameter ranges used for Monte Carlo (MC) sampling are shown in Table 3. The
uniform MC sampling approach generated 1319 model runs in total for a four-year period
(2013 until 2017). To rule out numerical instabilities and initialise the water and isotope
storages, we looped the year 2013 twice for a model warm-up period. The MC analysis was
performed using a cluster computing facility with 10 nodes, each controlling 32 CPU units at
the Center for Geophysical Research (CIGEFI) at the University of Costa Rica. For model
evaluation, the MC output was analysed in three stages:

Firstly, unacceptable Monte Carlo runs were rejected. A run was rejected when it did not
fulfil one of the following hydrological criteria:

- The annual streamflow should not deviate more than 10% of the incoming
precipitation minus the total outgoing evapotranspiration for any simulated year.
Because the model structure and fluxes between the different modules are based on
water balances, no large deviations on an annual scale are expected.
- The simulated discharge cannot fall below 0.02m3/s, which is the minimum observed
discharge in the San Lorencito catchment.
- The Nash-Sutcliffe Efficiency, NSE, (Nash and Sutcliffe, 1970) (maximum at unity) for
discharge at the catchment outlet cannot be zero or negative.
Secondly, two additional efficiencies were computed, the modified Kling Gupta Efficiency
(KGE) (Gupta et al., 2009; Kling et al., 2012) and the logarithmic Nash-Sutcliffe (lnNSE). The
KGE has its maximum at unity. The NSE, KGE and lnNSE were summed up for an overall
efficiency criterion E for streamflow simulations. To compare simulations with the fewer
observed isotope values, an alternative criterion had to be adopted (Equation 4), comparing

This article is protected by copyright. All rights reserved.


the mean and standard deviation of observations (meanobs, sdobs) and simulations (meansim,
sdsim). The criterion has its optimum value at unity:

|𝑚𝑒𝑎𝑛 |+𝑠𝑑
𝐼𝑠𝑜𝐶𝑟𝑖𝑡 = |𝑚𝑒𝑎𝑛 𝑜𝑏𝑠 |+𝑠𝑑𝑜𝑏𝑠 Eq. 4
𝑠𝑖𝑚 𝑠𝑖𝑚

Thirdly, the acceptable parameter sets were selected using those models whose parameters
perform above the 90th percentile of the retained parameter sets from step 1 and 2. These
parameter sets were used to simulate the hydrological and isotope dynamics of the
catchment. All retained parameter sets were visualized as an indication of model
uncertainty. The best-fit parameter set was further used to calculate the simulated time
series and spatial maps of all variables.

5. Results
5.1 Hydroclimatic conditions

Average climatic variables air temperature, humidity, solar radiation and precipitation are
shown as monthly average boxplots in Figure 4. A distinct seasonality of wet (May-
November) and dry (December-April) months is clearly visible (Figure 4A). The length of the
dry season, however, varied from a maximum of four months to a minimum of only two
months with a consistently high probability of rainfall throughout the year. March was on
average the driest month and September the wettest, with monthly rainfall ranging from
only 2 mm to almost 600 mm. No mid-summer drought in July, typical for the Pacific slope,
could be identified (Figure 4A). Temperatures in the catchment remained constant
throughout the year with an average around 20°C. May and June were the hottest months
and January and December the coldest, with temperatures dropping to a monthly average
of 16°C (Figure 4D). The relative humidity was constantly high (96%) and close to saturation
with a moderate peak in June (Figure 4B). Monthly solar radiation varied between 30 W/m 2
and 140 W/m2 and spiked in September 2011 of one year with the incoming monthly
average solar radiation reaching almost 300 W/m2 (Figure 4C).

This article is protected by copyright. All rights reserved.


The climatic variability was influenced by ENSO during the ENSO warm phase and
predominantly resulted in recorded drier conditions and higher temperatures. In Figure S1A
(Appendix) the observed monthly climatic variability from January 2008 until January 2017
was plotted against the drier El Niño periods in grey (2009-2010 and 2014-2016). Monthly
wind speeds decreased during El Niño with a change of the predominant influence of north-
easterly trade winds into west and easterly winds originating from the Pacific slope. This
implied that the ReBAMB was influenced by the Pacific slope region during El Niño and
presented an atypical rainfall pattern during the June-August period.

5.2 Isotope inputs

Precipitation isotope signatures reflected a mixture of the moisture origin of both the
Northern Pacific Ocean (depleted) and the Caribbean Sea (more enriched) with mean values
of -26.7‰ for deuterium (Table 1). Gross event rainfall isotopes were close to the Global
Meteoric Water Line (GMWL) with an overall measured range from slightly positive values
to around -90‰. The Local Meteoric Water Line (LMWL) regression equation (y = 7.2x + 8.6)
resulted in a R2 = 0.92 and was close to the GMWL (Figure 5A). However, event throughfall
samples deviated from the GMWL causing a significant change of slope (Figure 5A). This
offset was likely caused by fractionation of a vegetation interception storage. Furthermore,
the throughfall and streamflow isotopes showed less variability compared to the more
variable incoming rainfall isotopes (Table 1). Although streamflow isotopes fell slightly off
the LMWL, we need to interpret the implications of evaporative fractionation affecting
streamflow isotopes cautiously mainly due to the limited number of samples. The most
prominent damping effect was found in the baseflow isotope samples which range from -
30‰ to -25‰ and clustered within the intermediate range of the full distribution of
streamflow samples (Figure 5A). The relationship between gross rainfall and throughfall was
plotted to quantify fractionation in throughfall (Figure 5B). The deuterium signatures in
throughfall were consistently more enriched compared to simultaneously collected gross
rainfall and plotted above the 1:1 line. This empirical relationship (y = 0.74x - 8.8) shown in
Table 2 and Figure 5B was used for modelling isotope fluxes.

This article is protected by copyright. All rights reserved.


A MLR (y = -222+3.5x1+1.9x2; adjR2 = 0.33) calculated continuous hourly gross precipitation
isotope time series for model input from event-based measurements of selected wind
speed (x1) and relative humidity (x2) as the main controlling meteorological variables.
Although the adjusted R2 was relatively low, the p-values for windspeed and relative
humidity were significant (0.023 and 0.064 respectively). The MLR result was further
supported by the high F-statistic indicating a suitable MLR model structure and
parameterization. The low adjusted R2 likely reflected the limited amount of high-temporal
resolution event samples used for the analysis. The simulated hourly deuterium
precipitation input time series for the period 2013-2016 ranged from slightly positive values
to around -100‰ and compare reasonably well with event measurements. To provide an
additional qualitative comparison, deuterium signatures from the nearby measurement
station "Monteverde" (30 km from study area) were used for cross-checking.

5.3 Simulating stream flow and stream isotopes

The 60 accepted simulations were plotted against the observed values for the logarithmic
discharge and deuterium signatures in Figure 6. Generally, both the discharge and isotope
simulations reflect the seasonality of the system, peak events and baseflow conditions. The
first, and most depleted, streamflow isotope samples were not captured by the model
mainly due to a mismatch in the generated input isotope data (Figure 6B). Although
streamflow dynamics were captured well (Figure 6A; NSE of 0.67, KGE of 0.74 and lnNSE
equal to 0.39), the isotope simulations exhibited a larger range and a larger uncertainty.

A histogram showed that KGE values followed a Normal distribution with a maximum at
around 0.85 and a mean at 0.55 (Figure 7C). The NSE reached relatively high efficiencies
(maximum 0.75 and median 0.62) (Figure 7A). The lnNSE (Figure 7B) was the most stringent
criterion, producing maxima at only 0.53 indicating lower model efficiency in simulating
baseflows. Simulations that did not comply with the qualitative restrictions for isotopes
(described in 4.4) were rejected (Figure 7D) and retaining the 10 th percentile best remaining

This article is protected by copyright. All rights reserved.


simulations resulted in 60 accepted best parameter sets. The maximum overall efficiency
was close to a value of E = 1.8, with the NSE, KGE and lnNSE for the best simulation
indicated as black dots in Figure 7. Due to computational constraints, the MC analysis
produced 1319 simulations, which was not sufficient to fully sample the parameter space as
specified in Table 3, and precluded any formal uncertainty and parameter sensitivity
analysis.

5.4 Inferring high-resolution flow pathways, storage and runoff generation

In Figure 8A and 8B the best-fit simulation was used to visualize the internal model water
and isotope fluxes, respectively. The fluxes shown are throughfall (D), near-surface runoff
(QS), runoff generated from the soil store (QSTO) and groundwater store (QGW), respectively,
together with the total discharge (Qtot) at the catchment outlet. The filling and depletion of
all stores through incoming rain was influenced by the wet and dry climatic cycling with
most recharge (~400 mm seepage model parameter, Table 4B) generated during the wet
season (Figure S2). This seasonality is directly reflected in the discharge volumes of Figure 8.
Tables 4A, B and C summarise the annual water partitioning of water fluxes and storage.
Overall, most incoming rain directly converts into throughfall (model parameter D) and
subsequently into runoff as saturation overland flow (QS) without entering the soil store
(around 90% of the discharge). The estimated total AET was in the range of 400 to 500 mm
close to potential ET (Table 4B). Relatively small contributions to discharge were generated
from the deeper soil storage (between 6 and 9%). The average annual amounts of modelled
groundwater routed to the stream were around 12%. Storage volumes were highly variable
with a transient soil store resembling streamflow and a groundwater store in the range of
10 to 500 mm (Table 4C).

Simulated isotope dynamics are shown in Figure 8B as precipitation (concP), the


interception store (conci), the soil and groundwater stores (concs and concg) and at the
catchment outlet (Qconc). The catchment exhibited a clear isotope damping effect from
precipitation to groundwater. Generally, water from the interception store was most
enriched followed by the water from the soil store, groundwater store and incoming rain.

This article is protected by copyright. All rights reserved.


Isotope signatures in the soil store were limited to between -10‰ to -40‰. The isotopic
range was further damped in the groundwater store, where a large passive storage (~ 400
mm for the best-fit simulation) filters most of the incoming isotopic variability and produced
close to constant deuterium signatures of between only -16‰ to -22‰. The baseflow
stream isotope signatures were consistent with baseflow tributary signatures originating
from groundwater.

The spatial dynamics of soil moisture were largely conditioned by the vegetation
distribution and topographic characteristics of the catchment (Figure 2). The interception
storage, throughfall dynamics and evapotranspiration depended on the LAI and vegetation
height maps. The topography pre-conditioned runoff generation according to higher
infiltration and less near-surface discharge on the hillslopes and low infiltration and higher
discharge closer to the stream network. The latter is clearly visible in the spatial distribution
of the soil water storage for dry and wet conditions of contrasting dry and wet years (Figure
9).

6. Discussion
6.1 Adapting a spatially distributed tracer-aided rainfall-runoff model to the humid
tropics
An objective of this project was to explore the implementation of the STARR model as
developed by van Huijgevoort et al. (2016a and b) for application to a humid tropical
catchment. To comply with the general conditions of such a catchment (e.g. densely
vegetated, steep slopes) and the proposed model resolution (10 x 10 m grid cells and hourly
time steps), the model required several modifications. The focus of the new model structure
lays in the spatially-distributed evapotranspiration and interception modules needed to
conceptualise the heterogeneous, pristine rainforest of the study site.

Using Rutter’s approach for interception estimation allows the canopy to remain wet
between events, which according to Hoelscher et al. (2004) is a reasonable assumption for
the San Lorencito study catchment (located in a transitional zone above pre-montane
forests and slightly below cloud forest above 1800 masl). Due to the high epiphyte

This article is protected by copyright. All rights reserved.


vegetation, the area is comparable to tropical cloud forests (Salazar-Rodríguez, 2003).
Furthermore, the Rutter interception model is mass-balance based and could be directly
coupled to the isotope-enabled high resolution (hourly) evapotranspiration calculations,
while the original Gash approach only takes average evapotranspiration and rainfall
intensity into account and assumes the canopy dries. Muzylo et al. (2009) showed that both
approaches have been previously applied in tropical environments, but only the Rutter
model could be efficiently used with the fully distributed mass balance approach at high
temporal resolution. The interception is consequently (and in contrast to the original STARR
interception module) modelled as a storage with a certain capacity S and coupled to a tracer
mass balance. We used the empirical relationship to estimate evaporative enrichment in
throughfall (Figure 5B), which could be substituted by a physical representation of
interception fractionation in the future similar to Sprenger et al. (2017). A physical
fractionation mechanism would facilitate using measured throughfall for model evaluation.
The importance of interception as a physical store of water that causes fractionation was
also shown by Hervé-Fernández et al. (2016) in Chile and a temperate forest in Germany by
Stockinger et al. (2015). These authors, among others (see Allen et al., 2017 for a review on
the topic with examples of isotopic depletion in throughfall), also demonstrated the
importance of fractionated throughfall on streamflow isotope signatures and the error
introduced in model simulations that do not take throughfall into account for streamflow
isotope simulations and transit time estimates of forested catchments.

Given the steep, dissected topography, adequate spatial distribution of catchment structure
and organisation seemed an essential prerequisite for modelling the study site. For example,
the TWI was used to distinguish runoff generation close to the channel network from the
surrounding hillslopes similar to the original TOPMODEL formulation through calibration of
two different water holding capacity model parameters. Such a distribution was supported
by spatially distributed soil physical properties (30 samples), but in absence of groundwater
monitoring, our modified STARR model does not take variable water table depth
distribution and other possible deeper stratigraphy into account (Rinderer et al., 2014). A
relatively coarse 10 m x 10 m grid for a 3.2km2 catchment is still likely to introduce
uncertainties, but represented an improvement compared to previous applications using a
100m grid by van Huijgevoort et al. (2016a). Nevertheless, interception, soil and

This article is protected by copyright. All rights reserved.


groundwater stores exhibit heterogeneity within one grid cell which will not be captured.
This could be conceptualised using a sub-grid parameterization similar to Samaniego et al.
(2010). However, an increase in grid resolution would also lead to increased computation
time, which was already high (several days for 1000 runs on a cluster with 10 nodes each
deploying 32 CPU units). However, the hourly model time step was necessary to capture the
catchment response time, which is slightly below one hour.

6.2 Simulating the hydrological and isotopic response in a steep, pristine and volcanic
tropical catchment
A total of twelve model parameters were determined by calibration using uniform Monte
Carlo sampling. This relatively low number of iterations used in the calibration was likely not
enough to adequately sample the parameter space to assess sensitivity in a quantitative
way. This limited sampling probably results in the relatively low number of behavioural
simulations. Despite simulating the maximum peak flows reasonably well, the model could
not capture the extreme event in August 2015, which produced water levels up to 2.8 m,
mainly due to under catch of rain inputs induced by wind drifts (Frumau et al., 2011).

In addition to model structural and parameter uncertainty (e.g. Beven, 2012), the extreme
flood event raises potential uncertainties in the data and data quality (Beven and
Westerberg, 2011). We, therefore, excluded the re-constructed peak discharge and also did
not use the rating curve for evaluation avoiding bias towards that single extreme event.
Seibert and Beven (2009) found that few discharge points covering a certain range of the
hydrograph were sufficient to successfully calibrate a rainfall-runoff model. Therefore,
around 40 event-based streamflow isotope samples and 65 manual streamflow gaugings
were directly used for model evaluation. Although the data limitations and regression model
error propagation resulted in uncertainty, the regression relationship that generated hourly
model rainfall isotope input identified relative humidity together with wind speed as the
most important variables similar to Sanchez-Murillo et al. (2015) for three different sites in
Costa Rica.

This article is protected by copyright. All rights reserved.


The overall model performance can be compared (despite different time steps) to the best-
fit results obtained for a Scottish test catchment (van Huijgevoort et al., 2016a) with NSE
efficiencies of 0.52 and 0.56 for streamflow and isotopes in stream, respectively. The
adapted STARR tropics model resulted in a best-fit NSE and KGE values of 0.8 and almost 0.9
(Figure 6). A study by Ala-aho et al. (2017a), who used an adaptation of the original STARR
model to simulate isotopes in snow packs, reached values between 0.6 and 0.8,
respectively. Generally, the streamflow simulations were satisfactory, but stream isotopes
were less well simulated compared to previous studies using STARR (van Huijgevoort et al.,
2016a; Ala-aho et al., 2017a) and lumped TAMs in the humid tropics by Birkel and Soulsby
(2016). The latter is likely explained by data constraints at the time of the study, and this will
be addressed in future work as new isotope sampling is underway. Nevertheless, adding the
isotopes to the modelling framework supports the hypothesis that the San Lorencito
catchment can be characterized as a surface water-dominated system with most of the
water contributing to streamflow being generated as saturation overland flow (~90 %). The
latter runoff generation mechanism is in line with exceptionally high infiltration capacities
(measured at > 1000 mm/h in places) of the porous, undisturbed volcanic soils that exclude
any Hortonian overland flow mechanisms. Dominant saturation overland flow runoff
generation contrasts with the sub-surface dominated volcanic system studied by Muñoz-
Villers and McDonnell (2012) in the seasonal tropics in Mexico. However, runoff generation
studies from Panama (Zimmerman et al., 2012) and Brazil (Chaves et al., 2008) identified
overland flow as the main runoff generation mechanism due to the less conductive soils.
The quick near-surface runoff generation is also supported by the flashy isotope response in
the stream. A larger groundwater body is available for mixing (> 400 mm) and dampens out
the incoming precipitation signal. Moreover, it sustains baseflow when near-surface runoff
is switched off. Nonetheless, soil characteristics and moisture state largely control runoff
generation in the humid tropics as pointed out by Elsenbeer (2001) and Bruijnzeel (2004)
and tracer-aided rainfall-runoff models provide an integrated tool to assess future changes
due to environmental change.

7. Conclusions

This article is protected by copyright. All rights reserved.


This study aimed at providing a first insight into the hydroclimate, runoff generation and
isotope dynamics of a small humid tropical rainforest catchment in central Northern Costa
Rica by adapting and applying a spatially distributed Tracer-Aided rainfall runoff (STARR)
model (van Huijgevoort et al., 2016a):

- The original STARR model was adapted to the prevailing hydroclimatic and
topographic conditions using an hourly time step on a 10x10m spatial grid based on
the perceptual model of catchment functioning as an initial model hypothesis.
- A spatially distributed canopy storage capacity was introduced based on the Leaf
Area Index using a Rutter interception model.
- An empirical relationship estimated the enriched isotope signatures in throughfall
compared to gross precipitation to account for fractionation in the interception
storage.
- The model used a MLR model to reconstruct consistent hourly deuterium input time
series based on humidity and wind speed as the most important predictors.
- The model was calibrated using event-based, high-resolution isotope and stream
flow measurements.
- The San Lorencito streamflow was found to be largely controlled by quick near-
surface saturation overland flow (around 90%) with contributions of the
groundwater store (around 12%) and soil store (6%).
- The catchment storage is generally variable (10 to 500mm), but groundwater
contributions sustain baseflows during dry periods.
- The system’s isotope response is also quick and suggests limited mixing and near-
surface dominating flow pathways with a likely high event or “new” water
contribution to peak flows.
- The model inferred isotopic attenuation with depth increasing from the interception,
soil to groundwater storages (Figure 8), but due to quick near-surface runoff
dominating the streamflow isotope dynamics, there is a direct isotopic response to
rainfall events as young water dominates the hydrograph.
Although offering a first insight into hydrological functioning in terms of runoff generation,
flow paths and storage dynamics of a humid tropical, volcanic catchment, the study is just a
first step in exploiting the potential of STARR due to data and computational limitations. To

This article is protected by copyright. All rights reserved.


further improve the model, more computing power is needed to enable a more exhaustive
search of the parameter space. This would allow for a formal uncertainty and a parameter
sensitivity analysis. In addition, longer time series of event based precipitation and stream
isotope data are needed to improve the evidence base for the model by reducing the
uncertainty of interpolated input data and more available data for model evaluation. More
isotope data in form of soil, vegetation and groundwater samples would then also allow to
further evaluate eco-hydrologic water partitioning (e.g. Kuppel et al., 2018). Nevertheless,
coupling the event-based and high-resolution sampling with a spatially-distributed model
proved useful for a preliminary assessment of a challenging tropical catchment with difficult
access and limited economics to generate high-quality and resolution data sets.

Acknowledgements:
This project was initiated with a BGS fellowship to Josie Geris (University of Aberdeen) and
CB and an Ideawild grant to CB. Support by Tito Maldonado with the cluster computing
facility at CIGEFI, UCR is greatly acknowledged. Furthermore, we thank Vanessa Solano and
Sebastian Granados for processing the multispectral imagery. Many helping hands in the
field are acknowledged. The UCR-funded “Isotopes in tropical ecosystems network IsoNET”
and the project 217-B4-39 is greatly acknowledged. We also thank the VEWA project that
enabled early work on the STARR model. We thank one anonymous Reviewer and Matthias
Beyer for their constructive comments that helped improve the paper.

This article is protected by copyright. All rights reserved.


References

Aguilar E., Peterson T.C., Obando P.R., Frutos R., Retana J.A., Solera M., Soley J., García I.G.,
Araujo R.M., Santos A.R. and Valle V.E. (2005). Changes in precipitation and temperature
extremes in Central America and northern South America, 1961–2003. Journal of
Geophysical Research: Atmospheres. 110(D23).

Ala-aho P., Tetzlaff D., McNamara J., Laudon H., Kormos P. and Soulsby C. (2017a). Modeling
the isotopic evolution of snowpack and snowmelt: testing a spatially distributed
parsimonious approach. Water Resources Research, 53(7): 5813–5830.
doi: 10.1002/2017WR020650.

Ala-aho P., Tetzlaff D., Laudon H., McNamara J. and Soulsby C. (2017b). Using isotopes to
constrain water flux and age estimates in snow-influenced catchments using the STARR
(Spatially distributed Tracer-Aided Rainfall-Runoff) model. Hydrology and Earth System
Science, 21(10): 5089. doi: 10.5194/hess-2017-106.

Allen R.G., Pereira L.S., Raes D. and Smith M. (1998). Crop evapotranspiration -guidelines for
computing crop water requirements. FAO - Food and Agriculture Organization of the United
Nations, (56). FAO Irrigation and drainage paper 56. Food and Agriculture Organization,
Rome.

Allen S.T., Keim R.F., Barnard H.R., McDonnell J.J. and Brooks J.R. (2017). The role of stable
isotopes in understanding rainfall interception processes: a review. WIREs Water, 4(1). doi:
10.1002/wat2.1187.

Beck H.E., Bruijnzeel L.A., van Dijk A.I.J.M., McVicar T. R., Scatena F. N. and Schellekens J.
(2013). The impact of forest regeneration on streamflow in12 mesoscale humid tropical
catchments. Hydrological Earth System Sciences, 17(7):2613–2635.doi: 10.5194/hess-17-
2613-2013.

Bergoeing J. (2007). Geomorfologia de Costa Rica. Libreria Francesa, San Jose, Costa Rica.
ISBN: 9789968712057.

Bergström S. and Forsman A. (1973). Development of a conceptual deterministic rainfall-


runoff model. Nordic Hydrology, 4(3):147–170.

Beven K. (2012). Rainfall Runoff modelling: The primer, 2nd Edition. John Wiley and Sons.
ISBN: 978-0-470-71459-1.

Beven K. and Kirkby M. (1979). A physically based, variable contributing area model of basin
hydrology/ un modèle à base physique de zone d’appel variable de l’hydrologie du bassin
versant. Hydrological Sciences Bulletin, 24(1):43-69. doi:
dx.doi.org/10.1080/02626667909491834.

This article is protected by copyright. All rights reserved.


Beven K. and Westerberg I. (2011). On red herrings and real herrings: disinformation and
information in hydrological inference. Hydrological Processes, 25(10):1676-1680. doi:
10.1002/hyp.7963.

Birkel C. and Soulsby C. (2015). Advancing tracer-aided rainfall-runoff modelling: A review of


progress, problems and unrealised potential. Hydrological Processes, 29(25):5227–5240.
doi: 10.1002/hyp.10594.

Birkel C. and Soulsby C. (2016). Linking tracers, water age and conceptual models to identify
dominant runoff processes in a sparsely monitored humid tropical catchment. Hydrological
Processes, 30(24):4477–4493. doi: 10.1002/HYP.10941.

Birkel C., Soulsby C. and Tetzlaff D. (2011). Modelling catchment-scale water storage
dynamics: reconciling dynamic storage with tracer-inferred passive storages. Hydrological
Processes, 25(25):3924 - 3936. doi: 10.1002/hyp.8201.

Birkel C., Soulsby C. and Tetzlaff D. (2012). Modelling the impacts of land-cover change on
streamflow dynamics of a tropical rainforest headwater catchment. Hydrological Sciences
Journal, 57(8):1543-1561. doi: dx.doi.org/10.1080/02626667.2012.728707.

Birkel C., Soulsby C., and Tetzlaff D. (2014). Developing a consistent process‐based
conceptualization of catchment functioning using measurements of internal state variables.
Water Resources Research, 50(4):3481-3501. doi: 10.1002/2013WR014925.

Bruijnzeel L.A. (2004). Hydrological functions of tropical forests: not seeing the soil for the
trees?, Agriculture, Ecosystems & Environment, 104 (1), 185-228,
https://doi.org/10.1016/j.agee.2004.01.015.

Campling P., Gobin A., Beven K. and Reyen J. (2002). Rainfall-runoff modelling of a humid
tropical catchment: the TOPMODEL approach. Hydrological Processes, 16(2):231-253. doi:
10.1002/hyp.341.

Capra L., Lugo-Hubp J. and Borselli L. (2003). Mass movements in tropical volcanic terrains:
The case of Teziutlán (México). Engineering Geology, 69(3):359-379. doi: 10.1016/S0013-
7952(03)00071-1.

Chaves J., Neill C., Germer S., Neto S. G., Krusche A. and Elsenbeer H. (2008). Land
management impacts on runoff sources in small Amazon watersheds. Hydrol. Process., 22:
1766-1775. doi:10.1002/hyp.6803.

Craig H. (1961). Isotopic variations in meteoric waters. Science, 133(3465):1702-1703. doi:


10.1126/science.133.3465.1702.

Darcy H. (1856). Détermination des lois d'écoulement de l'eau á travers le sable. Les
fontaines publiques de la ville de Dijon. Victor Dalmont, Paris.

This article is protected by copyright. All rights reserved.


Denyer P. and Kussmaul S. (2000). Geologia de Costa Rica. Editorial Tecnológica de Costa
Rica, Cartago, Costa Rica. ISBN: 9789977662718.
Dolman A., Gash J., Roberts J. and Shuttleworth W. (1991). Stomatal and surface
conductance of tropical rainforest. Agricultural and forest Meteorology, 54(2-4):303-318.
doi: doi.org/10.1016/0168-1923(91)90011-E.

Dray S., Legendre P. and Blanchet G. (2009). packfor: Forward selection with permutation
(canoco p. 46) (r package version: 0.0-7/r58).

Dunn S. and Mackay R. (1995). Spatial variation in evapotranspiration and the influence of
land use on catchment hydrology. Journal of Hydrology, 171(1-2):49–73. doi:
doi.org/10.1016/0022-1694(95)02733-6.

Duran-Quesada A., Gimeno L., Amador J. and Nieto R. (2010). Moisture sources for central
america: Identification of moisture sources using a lagrangian analysis technique. Journal of
Geophysical Research, 115(D5). doi: 10.1029/2009JD012455.

Elsenbeer H. (2001). Hydrologic flowpaths in tropical rainforest soilscapes-a review.


Hydrological Processes, 15(10), 1751-1759.

Frumau K. F., Bruijnzeel L. A. and Tobón C. (2011). Precipitation measurement and


derivation of precipitation inclination in a windy mountainous area in northern Costa Rica.
Hydrol. Process., 25: 499-509. doi: 10.1002/hyp.7860.

Gash J., Lloyd C. and Lachaud G. (1995). Estimating sparse forest rainfall interception with an
analytical model. Journal of Hydrology, 170(1-4):79-86. doi: doi.org/10.1016/0022-
1694(95)02697-N.

Gibbs H. K., Ruesch A. S., Achard F., Clayton M. K., Holmgrem P., Ramankutty N. and Foley J.
A. (2010). Tropical forests were the primary sources of new agricultural land in the 1980s
and 1990s. Proceedings of the National Academy of Science, 107(38): 16732–16737.

Giorgi F. (2006), Climate change hot-spots. Geophysical Research Letters, 33(6). doi:
doi.org/10.1029/2006GL025734.

González J. E., Georgescu M., Lemos M. C., Hosannah N. and Niyogi D. (2017), Climate
change’s pulse is in Central America and the Caribbean, Eos, 98, doi:
doi.org/10.1029/2017EO071975.

Gupta H., Kling H., Yilmaz K. and Martinez G. (2009). Decomposition of the mean squared
error and nse performance criteria: Implications for improving hydrological modelling.
Journal of Hydrology, 377(1):80-91. doi: 10.1016/j.jhydrol.2009.08.003.

Hervé-Fernández P., Oyarzún C., Brumbt C., Huygens D., Bodé S., Verhoest N.E.C. and
Boeckx P. (2016). Assessing the ‘two water worlds’ hypothesis and water sources for native

This article is protected by copyright. All rights reserved.


and exotic evergreen species in south-central Chile. Hydrological Processes, 30(23):4227–
4241. doi: 10.1002/hyp.10984.

Hoelscher D., Koehler L., van Dijk A. and Bruijnzeel L. (2004). The importance of epiphytes to
total rainfall interception by a tropical montane rain forest in Costa Rica. Journal of
Hydrology, 292(1-4):308–322. doi: doi.org/10.1016/j.jhydrol.2004.01.015.

Hrachowitz M., Savenije H., Bogaard T.A., Tetzlaff D. and Soulsby C. (2013). What can flux
tracking teach us about water age distribution patterns and their temporal dynamics?
Hydrology and Earth System Sciences, 17:533–564. doi: 10.5194/hess-17-533-2013.

Hrachowitz M., Benettin P., van Breukelen B.M, Fovet O., Howden N.J.K., Ruiz L., van der
Velde Y. and Wade A.J. (2016). Transit times—the link between hydrology and water quality
at the catchment scale. WIREs Water, 3:629–657. doi: 10.1002/wat2.1155.

Karssenberg D., Schmitz O., Salamon P., de Jong K. and Bierkens M.F.P (2010). A software
framework for construction of process-based stochastic spatio-temporal models and data
assimilation. Environmental Modelling & Software, 25(4):489-502.
doi.org/10.1016/j.envsoft.2009.10.004.

Kirchner J.W. (2006). Getting the right answers for the right reasons: linking measurements,
analysis and models to advance the science of hydrology. Water Resources Research, 42(3).
doi: 10.1029/2005WR004362.

Kirchner J.W. (2016). Aggregation in environmental systems – Part 1: Seasonal tracer cycles
quantify young water fractions, but not mean transit times, in spatially heterogeneous
catchments. Hydrology and Earth System Sciences, 20(1):279-297. doi:
https://doi.org/10.5194/hess-20-279.

Kling H., Fuchs M. and Paulin M. (2012). Runoff conditions in the upper Danube basin under
an ensemble of climate change scenarios. Journal of Hydrology, 424-425:264-277. doi:
doi.org/10.1016/j.jhydrol.2012.01.011.

Knighton J., Saia S.M., Morris C.K., Archibald J.A. and Walter M.T. (2017). Ecohydrologic
considerations for modeling of stable water isotopes in a small intermittent watershed.
Hydrological Processes, 31(13): 2438-2452 doi: doi.org/10.1002/hyp.11194.

Kumagai T., Kanamori H. and Chappell, N. A. (2016). Tropical forest hydrology. In: Forest
Hydrology: Processes, Management and Assessment, 88-102, Editor(s): Amatya D.M.,
Williams T.M., Bren L. and Jong C. de, CAB International.

Kuppel S., Tetzlaff D., Maneta M. P. and Soulsby, C. (2018) EcH2O-iso 1.0: Water isotopes
and age tracking in a process-based, distributed ecohydrological model, Geosci. Model Dev.
Discuss. doi: https://doi.org/10.5194/gmd-2018-25, 2018.

This article is protected by copyright. All rights reserved.


Loescher H., Powers J. and Oberbauer S. (2002). Spatial variation of throughfall volume in an
old-growth tropical wet forest, Costa Rica. Journal of Tropical Ecology, 18(3):397–407. doi:
doi.org/10.1017/S0266467402002274.

McDonnell J. and Beven K. (2014). Debates—the future of hydrological sciences: A


(common) path forward? A call to action aimed at understanding velocities, celerities and
residence time distributions of the headwater hydrograph. Water Resources Research,
50(6):5342–5350. doi: 10.1002/2013WR015141.

McDonnell J., Sivapalan M., Vache K., Dunn S., Grant G., Haggerty R., Hinz C., Hooper R.,
Kirchner J., Roderick M., Selker J. and Weiler M. (2007). Moving beyond heterogeneity and
process complexity: A new vision for watershed hydrology. Water Resources Research,
43(7). doi: 10.1029/2006WR005467.

McGuire K. and McDonnell J. (2006). A review and evaluation of catchment transit time
modeling. Journal of Hydrology, 330(3-4):543-563. doi:
doi.org/10.1016/j.jhydrol.2006.04.020.

McMillan H., Tetzlaff D., Clark M. and Soulsby C. (2012) Do time variant tracers provide tools
for evaluating hydrological model structure? A multi-modelling approach. Water Resources
Research, 48(5). doi: 10.1029/2011WR011688.

Moličová H., Grimaldi M., Bonell M., and Hubert P. (1997). Using TOPMODEL towards
identifying and modelling the hydrological patterns within a headwater, humid, tropical
catchment. Hydrological Processes, 11(9), 1169-1196. doi:
https://doi.org/10.1002/(SICI)1099-1085(199707)11:9<1169::AID-HYP551>3.0.CO;2-W.

Muñoz-Villers L.E., McDonnell J.J. (2012). Runoff generation in a steep, tropical montane
cloud forest catchment on permeable volcanic substrate. Water Resources Research, 48(9).
doi: 10.1029/2011WR011316.

Muzylo A., Llorens P., Valente F., Keizer J., Domingo F. and Gash J. (2009). A review of
rainfall interception modelling. Journal of Hydrology, 370(1-4):191–206. doi:
doi.org/10.1016/j.jhydrol.2009.02.058.

Nash J. and Sutcliffe J. (1970). River flow forecasting through conceptual models part 1: A
discussion of principles. Journal of Hydrology, 10(3):282-290. doi: doi.org/10.1016/0022-
1694(70)90255-6.

Penman H. (1947). Natural evaporation from bare soil, open water and grass. Proceedings of
the Royal Society of London. Series A, Mathematical and Physical Sciences, 193(1032). doi:
10.1098/rspa.1948.0037.

Pereira D., Martinez M.A., Pruski F.F. and da Silva D.D. (2016). Hydrological simulation in a
basin of typical tropical climate and soil using the SWAT model part I: Calibration and

This article is protected by copyright. All rights reserved.


validation tests. Journal of Hydrology: Regional Studies, 7:14-37. doi:
doi.org/10.1016/j.ejrh.2016.05.002.

R Core Team (2013). R: A language and environment for statistical computing. R Foundation
for Statistical Computing, Vienna, Austria. URL http://www.R-project.org/.

Rhodes A.L., Guswa A.J. and Newell S.E. (2006). Seasonal variation in the stable isotopic
composition of precipitation in the tropical montane forests of Monteverde, Costa Rica.
Water Resources Research, 42(11). doi: 10.1029/2005WR004535.

Rinderer M., Van Meerveld H. J. and Seibert J. (2014). Topographic controls on shallow
groundwater levels in a steep, prealpine catchment: When are the TWI assumptions valid?
Water resources research, 50(7):6067-6080. doi: https://doi.org/10.1002/2013WR015009.

Rutter A., Kershaw K., Robins P. and Morton A. (1971). A predictive model of rainfall
interception in forest, 1. derivation of the model from observation in a plantation of
Corsican pine. Agricultural Meteorology, 9:367–384. doi: doi.org/10.1016/0002-
1571(71)90034-3.

Rutter A., Morton A., and Robins P. (1975). A predictive model of rainfall interception in
forests. ii. generalization of the model and comparison with observations in some
coniferous and hardwood stands. Journal of Applied Ecology, 12(1):367–380. doi:
10.2307/2401739.

Saito K., Ogawa S., Aihara M. and Otowa K. Estimates of LAI for forest management in
Okutama. In Proceedings of the 22nd Asian Conference on Remote Sensing, Singapore, 5–9
November 2001.

Salazar-Rodríguez A. (2003). Reserva biologica alberto ml. brenes: Una excepción en Costa
Rica. Revista InterSedes, 5(8). ISSN: 2215-2458.

Samaniego L., Kumar R. and Attinger S. (2010). Multiscale parameter regionalization of a


grid-based hydrologic model at the mesoscale. Water Resources Research, 46(5):W05523.
doi: 10.1029/2008WR007327.

Sanchez-Murillo R., Esquivel-Hernandez G., Welsch K., Brooks E., Boll J., Alfaro-Solis R. and
Valdes-Gonzalez J. (2013). Spatial and temporal variation of stable isotopes in precipitation
across Costa Rica: An analysis of historic GNIP records. Open Journal of Modern Hydrology,
3(4):226-240. doi: 10.4236/ojmh.2013.34027.

Sanchez-Murillo R., Birkel C., Esquivel-Hernandez G., Corrales-Salazer J., Boll J., Brooks E.,
Roupsard O., Saenz-Rosales O., Katchan I., Arce-Mesen R., Soulsby C, and Araguas-Araguas L.
(2015). Key drivers controlling stable isotope variations in daily precipitation of Costa Rica:
Caribbean sea versus eastern pacific ocean moisture sources. Quaternary Science Reviews,
131(Part B):250–261. doi: doi.org/10.1016/j.quascirev.2015.08.028.

This article is protected by copyright. All rights reserved.


Sanchez-Murillo R. and Birkel C. (2016). Groundwater recharge mechanisms inferred from
isoscapes in a complex tropical mountainous region: Tropical isoscapes. Geophysical
Research Letters, 43(10):5060–5069. doi: 10.1002/2016GL068888

Schellekens J., Scatena F., Bruijnzeel L., and Wickel A. (1999). Modelling rainfall interception
by a lowland tropical rain forest in northeastern Puerto Rico. Journal of Hydrology, 225(3-
4):168–184. doi: doi.org/10.1016/S0022-1694(99)00157-2.

Seibert J. and Beven K. (2009). Gauging the ungauged basin: how many discharge
measurements are needed? Hydrology and Earth System Sciences, 13(6):883-892. doi:
doi.org/10.5194/hess-13-883-2009.

Seibert J. and Vis M (2012). Teaching hydrological modeling with a user-friendly catchment-
runoff-model software package. Hydrology and Earth System Sciences, 16:3315-3325. doi:
10.5194/hess-16-3315-2012.

Shuttleworth W.J. (1978). A simplified one-dimensional theoretical description of the


vegetation-atmosphere interaction. Boundary-Layer Meteorology, 14(1): 3-27.

Solano V., Granados S., Geris J., Brenes L., Artavia G., Sánchez R. and Birkel C. (in Review).
Sediment transport dynamics in steep and pristine, tropical volcanic catchments. Submitted
to: Earth surface processes and landforms.

Sørensen R., Zinko U. and Seibert J. (2006). On the calculation of the topographic wetness
index: evaluation of different methods based on field observations. Hydrology and Earth
System Sciences, 10(1):101-112. doi: 1607-7938/hess/2006-10-101.

Soulsby C., Tetzlaff D., Rodgers P., Dunn S. and Waldron S. (2006). Runoff processes, stream
water residence times and controlling landscape characteristics in a mesoscale catchment:
an initial evaluation. Journal of Hydrology, 325(1-4):197-221. doi:
doi.org/10.1016/j.jhydrol.2005.10.024.

Sprenger M., Tetzlaff D., and Soulsby C. (2017). Soil water stable isotopes reveal evaporation
dynamics at the soil-plant-atmosphere interface of the critical zone. Hydrology and Earth
System Sciences, 21(7):3839-3858. doi: doi.org/10.5194/hess-21-3839-2017.

Stockinger P., Lücke A., McDonnell J.J., Diekkrüger B., Vereecken H. and Bogena H.R. (2015).
Interception effects on stable isotope driven streamwater transit time estimates.
Geophysical Research Letters, 42(13):5299–5308. doi: 10.1002/2015GL064622.

Tang H., Dubayah R., Swatantran A., Hofton M., Sheldon S., Clark D.B. and Blair B. (2012).
Retrieval of vertical LAI profiles over tropical rain forests using waveform lidar at La Selva,
Costa Rica. Remote Sensing of Environment, 124:242-250. doi:
doi.org/10.1016/j.rse.2012.05.005.

This article is protected by copyright. All rights reserved.


Tetzlaff D., Soulsby C., Waldron S., Malcolm I.A., Bacon P.J., Dunn, S.M. and Lilly A. (2007).
Conceptualisation of runoff processes using GIS and tracers in a nested mesoscale
catchment. Hydrological Processes, 21(10):1289-1307. doi: doi:10.1002/hyp.6309.

van Huijgevoort M., Tetzlaff D., Sutanudjaja E. and Soulsby C. (2016a). Using high resolution
tracer data to constrain water storage, flux and age estimates in a spatially distributed
rainfall-runoff model. Hydrological Processes, 30(25):4761–4778. doi: 10.1002/hyp.10902.
van Huijgevoort M., Tetzlaff D., Sutanudjaja E. and Soulsby C. (2016b). Visualization of
spatial patterns of connectivity and runoff ages derived from a tracer-aided model.
Hydrological Processes, 30(25):4893–4895. doi: 10.1002/hyp.10961.
Von Hoyningen-Huene J. (1981). Die Interzeption des Niederschlages in landwirtschaftlichen
Pflanzenbestanden. Arbeitsbericht Deutscher Verband fuer Wasserwirtschaft und
Kulturbau, DVWK, Braunschweig, Germany.
Waylen P., Caviedes C. and Quesada M. (1996). Interannual variability of monthly
precipitation in Costa Rica. Journal of Climate, 9:2606-2613. doi:
https://doi.org/10.1175/1520-0442(1996)009<2606:IVOMPI>2.0.CO;2.
Wohl E., Barros A., Brunsell N., Chappell N., Coe M., Giambelluca T., Goldsmith S., Harmon
R., Hendrickx J., Juvik J., McDonnell J. and Ogden F. (2012). The hydrology of the humid
tropics. Nature Climate Change, 2:655–662. doi: 10.1038/nclimate1556.
Zheng G. and Moskal L. (2009). Retrieving leaf area index (LAI) using remote sensing:
Theories, methods and sensors. Sensors, 9(4):2719-2745. doi: 10.3390/s90402719.
Zimmermann A., Francke T. and Elsenbeer H. (2012). Forests and erosion: Insights from a
study of suspended-sediment dynamics in an overland flow-prone rainforest catchment.
Journal of Hydrology, 27:170-181. doi: 10.1016/j.jhydrol.2012.01.039

This article is protected by copyright. All rights reserved.


Table 1: Overview of the topographical, observed hydrometeorological and isotope
characteristics of the San Lorencito catchment.

Descriptor Unit Mean [range]


Topography
Elevation masl. 1133.5 [873.7-1472.4]
Slope ◦ 22.3 [0.15-52]
Drainage Density (1:200000) km/km2 0.016
Stream slope ◦ 20.5
Sediment transport Index - 17.04 [0-390]
Terrain Ruggedness Index - 1.63 [0.01-5.03]
Hydroclimatic Data (2013-2016)
Annual gross precipitation mm 2589 [1935-3787]
Annual PET mm 490 [332-579]
Annual modelled streamflow mm 2099 [1343-2446]
Isotope signatures (deuterium)
Precipitation ‰ -26.65 [-74.73 - +8.33]
Streamflow ‰ -26.70 [-46.87 - -20.77]

This article is protected by copyright. All rights reserved.


Table 2: The modified STARR tropics model components and equations.

Process Equation Variables


PET (Allen et al., 1998)
FAO Penman-Monteith (𝑒𝑠 − 𝑒𝑎 ) 𝑅𝑛 : net incoming radiation
𝛥(𝑅𝑛 − 𝐺) + 𝜌𝑎 𝑐𝑝 𝑟𝑎
(Penman, 1947) 𝜆𝐸𝑇 = 𝜌𝑎 : air density
𝑟𝑠 𝑐𝑝 : specific heat capacity
𝛥 + 𝛾 (1 + 𝑟 )
𝑎 𝛾: psychrometric constant
𝑒𝑠 : saturated vapour pressure
𝑒𝑎 : actual vapour pressure
𝛥: slope saturation pressure curve
Ground Heat Flux 𝐺 = 0.1𝑅𝑛 (day)
𝐺 = 0.5𝑅𝑛 (night)
Aerodynamic 𝑑 𝑑 𝑧𝑚 : height above ground of
ln (𝑧𝑚 − 𝑧 ) ln (𝑧ℎ − )
resistance 𝑜𝑚 𝑧𝑜ℎ windspeed measurement
𝑟𝑎 =
𝑘 2 𝑢10 𝑧ℎ : height of humidity measurement
𝑑: displacement height
𝑘: von Karman constant
𝑢10: windspeed
Momentum roughness 𝑧𝑜𝑚 = 0.123 ∗ ℎ ℎ: vegetation height
Roughness heat and 𝑧𝑜ℎ = 0.1 ∗ 𝑧𝑜𝑚
water vapor
Bulk surface resistance 𝑟𝐼 𝑟𝑙 : stomatal resistance
𝑟𝑠 =
𝐿𝐴𝐼𝑎𝑐𝑡𝑖𝑣𝑒 𝐿𝐴𝐼: leaf area index
Interception Storage (Rutter, 1971 and 1975)
Evaporation from if 𝐼𝑁𝑇 < 𝑆: 𝐸𝑇 = 𝐸𝑇𝑎 𝐼𝑁𝑇: Interception storage
canopy else: 𝐸𝑇 = 𝑃𝐸𝑇 𝐸𝑇𝑎 : evaporation including
canopy roughness
𝑃𝐸𝑇: potential evaporation, excludes
canopy roughness
Maximum Canopy 𝑆 = 0.935 + 0.498 ∗ 𝐿𝐴𝐼 − 0.00575 ∗
Storage Capacity (Von 𝐿𝐴𝐼 2
Hoyningen-Huene,
1981)
Drainage if 𝐼𝑁𝑇 > 𝑆: 𝐷 = 𝐷𝑆 ∗ 𝑒 𝑏(𝐼𝑁𝑇−𝑆) 𝐷𝑠 : drainage when INT=S
else: 𝐷 = 0 𝑏: empirical Rutter parameter
Interception Mass 𝐼𝑁𝑇(𝑡) = 𝐼𝑁𝑇(𝑡 − 1) + 𝑐 ∗ 𝑃 − 𝐸𝑇 − 𝐷 𝑐: 1-Canopy Gap Fraction = 0.95
Balance 𝑃: gross precipitation
Soil Storage
Direct Surface Runoff 𝑄𝑆 = max(𝑆𝑇𝑂 − 𝐹𝐶, 0) 𝑆𝑇𝑂: soil water store
𝐹𝐶: field capacity
Evaporation from soil 𝑃𝐸𝑇 𝑃𝐸𝑇: potential evapotranspiration
if 𝑆𝑇𝑂 < 𝐿𝑃 ∗ 𝐹𝐶: 𝑆𝑒𝑣𝑎𝑝 = 𝐿𝑃∗𝐹𝐶
else: 𝑆𝑒𝑣𝑎𝑝 = 𝑃𝐸𝑇 𝐴𝐸𝑇: actual evapotranspiration
𝐿𝑃: fraction of field capacity above
which 𝐴𝐸𝑇 = 𝑃𝐸𝑇
Seepage 𝑆𝑇𝑂 𝛽: exponent describing soil
𝑆𝑒𝑒𝑝𝑎𝑔𝑒 =
𝐹𝐶𝛽 groundwater recharge
Soil discharge 𝑄𝑆𝑇𝑂 = 𝑆𝑇𝑂 ∗ 𝑘𝑠 𝑘𝑠 : recession coefficient

This article is protected by copyright. All rights reserved.


soil discharge
Soil Storage Mass 𝑆𝑇𝑂(𝑡) = 𝑆𝑇𝑂(𝑡 − 1) + 𝑃𝑒𝑓𝑓 − 𝑄𝑆 − 𝑃𝑒𝑓𝑓 : sum of gross rainfall and
Balance 𝑆𝑒𝑣𝑎𝑝 − 𝑄𝑆𝑇𝑂 + 𝐶𝑎𝑝𝐹𝑙𝑢𝑥 throughfall
Groundwater Storage
Capillary Flux 𝐹𝐶 − 𝑆𝑇𝑂 𝐶𝑓𝑙𝑢𝑥 : maximum capillary rise
𝐶𝑎𝑝𝐹𝑙𝑢𝑥 = 𝐶𝑓𝑙𝑢𝑥 ∗
𝐹𝐶
Groundwater 𝑄𝐺𝑊 = 𝑘𝑔 ∗ 𝐺𝑊 𝐺𝑊: groundwater storage
Discharge 𝑘𝑔 : recession constant baseflow
Lateral Flow 𝐷𝐸𝑀 𝐾𝑠𝑎𝑡 : horizontal saturated
𝑄𝑙𝑓 = 𝐾𝑠𝑎𝑡 ∗ 𝑠𝑙𝑜𝑝𝑒 ( + 𝐺𝑊)
1000 conductivity
𝐷𝐸𝑀: Digital Elevation Model
Groundwater Mass 𝐺𝑊(𝑡) = 𝐺𝑊(𝑡 − 1) − 𝐶𝑎𝑝𝐹𝑙𝑢𝑥 − 𝑄𝐺𝑊 + 𝛥𝑄𝑙𝑓 : net lateral flow
Balance 𝛥𝑄𝑙𝑓 + 𝑆𝑒𝑒𝑝𝑎𝑔𝑒
Routing
Discharge at Outlet 𝑄𝑡𝑜𝑡𝑎𝑙 = 𝑄𝑆 + 𝑄𝑆𝑇𝑂 + 𝑄𝐺𝑊
Isotope Mass Balances
𝑖 𝑇𝐹 𝑖𝑃 empirical 𝑖 𝑇𝐹 = −8.8 + 0.76 ∗ 𝑖𝑃 𝑖 𝑇𝐹 : predicted throughfall
relation isotope signature (Fig. 5B)
𝑖𝑃 : isotope signature gross
precipitation
Interception Store 𝑖𝐼𝑁𝑇 (𝑡) ∗ 𝐼𝑁𝑇(𝑡) = 𝑖 𝑇𝐹 ∗ 𝑃 ∗ 𝑐 𝑖𝐼𝑁𝑇 : isotope signature in
interception store
Soil Store 𝑖𝑆𝑇𝑂 (𝑡) ∗ 𝑆𝑇𝑂(𝑡) 𝑖𝑆𝑇𝑂 : isotope signature in soil store
= 𝑖𝑆𝑇𝑂 (𝑡 − 1) 𝑆𝑇𝑂𝑝𝑎𝑠 : passive soil water storage
∗ (𝑆𝑇𝑂(𝑡 − 1) + 𝑆𝑇𝑂𝑝𝑎𝑠 ) 𝑖𝐺𝑊 : isotope signature in ground
+ 𝑖𝐼𝑁𝑇 (𝑡) ∗ 𝐷(𝑡) + 𝑖𝑃 (𝑡) water store
∗ 𝑃(𝑡) ∗ (1 − 𝑐)
− 𝑖𝑆𝑇𝑂 (𝑡 − 1) ∗ 𝑆𝑒𝑣𝑎𝑝 (𝑡)
− 𝑖𝑆𝑇𝑂 (𝑡 − 1) ∗ 𝑆𝑒𝑒𝑝𝑎𝑔𝑒(𝑡)
+ 𝑖𝐺𝑊 (𝑡 − 1) ∗ 𝐶𝑎𝑝𝐹𝑙𝑢𝑥(𝑡)
− 𝑖𝑆𝑇𝑂 (𝑡 − 1) ∗ 𝑄𝑆𝑇𝑂 (𝑡)
Groundwater Store 𝑖𝐺𝑊 (𝑡) ∗ 𝐺𝑊(𝑡) = 𝑖𝐺𝑊 (𝑡 − 1) ∗ (𝐺𝑊(𝑡 − 𝐺𝑊𝑝𝑎𝑠 : passive groundwater storage
1) + 𝐺𝑊𝑝𝑎𝑠 ) + 𝑖𝑆𝑇𝑂 (𝑡 − 1) ∗ 𝑆𝑒𝑒𝑝𝑎𝑔𝑒(𝑡) −
𝑖𝐺𝑊 (𝑡 − 1) ∗ 𝐶𝑎𝑝𝐹𝑙𝑢𝑥(𝑡) − 𝑖𝐺𝑊 (𝑡 − 1) ∗
𝑄𝐺𝑊 (𝑡) + 𝑖𝐺𝑊 (𝑡 − 1) ∗ 𝛥𝑄𝑙𝑓

This article is protected by copyright. All rights reserved.


Table 3: Overview of the STARR model parameters, units and initial ranges per module (11
calibrated and 4 fixed parameters). Parameters are briefly described and referenced in case
of literature values used.

Parameter Unit Description Initial Range


Evapotranspiration
𝑧ℎ [m] Height above ground of humidity measurement 2
𝑧𝑚 [m] Height above ground of windspeed 10
measurement
𝑟𝐼 [mm/s] Stomatal resistance 140 (Dolman et al.,
1991)
Interception Storage
𝐷𝑆 [mm/hr] Drainage from canopy when the storage is full 0.08-0.12
(Schellekens et al.,
1999)
𝑏 [-] Exponent in Rutter interception module 3.7-5.1 (Dunn and
Mackay, 1995)
Soil Storage
𝐹𝐶 [mm] Water holding capacity of the soil, distinguishes if 𝑇𝑊𝐼>6.5: 0-100
between valley and hillslope soils else: 100-500
𝐿𝑃 [-] Fraction of soil saturation above which 0.3-1
evaporation happens unlimited
𝛽 [-] Betaseepage, non-linear exponent for soil store 0.01-4
runoff generation
𝑘𝑠 [1/hr] Recession coefficient discharge from soil store 0.000042-0.021
𝑆𝑇𝑂𝑝𝑎𝑠 [mm] Passive soil storage 0-300
𝑓𝑟𝑎𝑐𝑆𝑇𝑂 [-] Differentiates passive storage in valley and if 𝑇𝑊𝐼>6.5: 1
hillslope else: 0-1
Groundwater Storage
𝑘𝑔 [1/hr] Recession coefficient for discharge from 0.000042-0.0042
groundwater store
𝐾𝑠𝑎𝑡 [mm/hr] Horizontal saturated hydraulic conductivity 0.0000042-8.33
𝐺𝑊𝑝𝑎𝑠 [mm] Passive groundwater storage water 0-1000
𝐶𝑎𝑝𝐹𝑙𝑢𝑥 [mm/hr] Capillary rise from groundwater module to soil if 𝑇𝑊𝐼>6.5: 1
store, differentiated for hillslopes and valley else: 0-1
bottom

This article is protected by copyright. All rights reserved.


Table 4: A) Overview of the annual water fluxes (𝑃, gross precipitation; 𝑄𝑆 , direct surface
runoff; 𝑄𝑆𝑇𝑂 , soil storage runoff and 𝑄𝐺𝑊 , groundwater storage runoff) in mm and as a % of
total discharge 𝑄 using the best model runs. B) Mean and range of additional water fluxes
simulated by the model in mm. 𝐷, drainage from canopy storage, 𝐴𝐸𝑇, the actual
evapotranspiration, 𝑆𝑒𝑒𝑝𝑎𝑔𝑒 and 𝐶𝑎𝑝𝐹𝑙𝑢𝑥, the Capillary Flux are given. C) Mean and range
of interception (𝐼𝑁𝑇), soil (𝑆𝑇𝑂) and groundwater (𝐺𝑊) storage.

𝑷 𝑸𝑺 𝑸𝑺𝑻𝑶 𝑸𝑮𝑾 𝑸
A) Mean [Range] Mean [Range] Mean [Range] Mean [Range]
2013 [mm] 2845 2044 [1424-2295] 122 [24-288] 307 [50-840] 2254 [2183-2290]
[%] 91 [65-103] 5 [1-13] 14 [2-38]
2014 [mm] 1916 1248 [885-2295] 44 [9-136] 183 [58-750] 1343 [1303-1534]
[%] 93 [58-103] 3 [1-10] 13 [4-49]
2015 [mm] 3088 2248 [1539-2530] 142 [35-330] 284 [65-778] 2446 [2270-2499]
[%] 92 [68-105] 6 [1-14] 12 [3-33]
2016 [mm] 3105 2162 [1529-2428] 106 [10-272] 313 [75-908] 2352 [2243-2401]
[%] 92 [66-104] 4 [0.4-11] 13 [3-40]

B) 𝑷 𝑫 𝑨𝑬𝑻 𝑺𝒆𝒆𝒑𝒂𝒈𝒆 𝑪𝒂𝒑𝑭𝒍𝒖𝒙


Mean [Range] Mean [Range] Mean [Range] Mean [Range]
2013 2845 2483 [2477-2487] 367 [332-431] 431 [107-1127] 118 [31-218]
2014 1916 1445 [1438-1499] 435 [428-499] 255 [75-717] 102 [28-218]
2015 3088 2711 [2706-2715] 370 [332-433] 464 [119-1212] 127 [32-215]
2016 3105 2618 [2612-2622] 489 [442-578] 464 [126-1214] 121 [32-214]

C) 𝑷 𝑰𝑵𝑻 𝑺𝑻𝑶 𝑮𝑾 𝑸
Mean [Range] Mean [Range] Mean [Range] Mean
2013 2845 0.90 [0.85-0.96] 19 [11-30] 65 [15-557] 2254
2014 1916 0.86 [0.83-0.90] 11 [7-18] 46 [8-506] 1343
2015 3088 1.14 [1.09-1.96] 21 [12-33] 61 [16-510] 2446
2016 3105 0.84 [0.80-0.90] 19 [12-32] 68 [16-613] 2352

This article is protected by copyright. All rights reserved.


Figure 1: A) The continental divide over the Central Costa Rican mountain range (Cordillera)
is visible in the regional overview map with the location of the San Lorencito study catchment.
B) The topography, stream network and monitoring location at the outlet is also shown. C)
The monthly precipitation regime at San Lorencito indicates a moderate dry season from
January to April followed by a wet season from May to December with virtually no variation
in temperature. D) An air photo shows the dense primary rainforest of both hillslopes.

This article is protected by copyright. All rights reserved.


Figure 2: The spatial maps at 10m grid size developed and used for modelling with identical
scales and the location of measurements at the outlet for reference: A) The Topographical
Wetness Index (TWI), B) the slope map in degrees, C) the Leaf Area Index (LAI) and D) the
vegetation height in meters.

This article is protected by copyright. All rights reserved.


This article is protected by copyright. All rights reserved.
Figure 3: Flow chart of the modified STARR tropics model. After interception INT of incoming
rainfall by the vegetation, water reaches the soil store STO via throughfall. Together with the
gross rainfall, this water enters the soil store and/or is discharged at a rate Qs. The water
entering the soil generates discharge QSTO, reaches the groundwater GW via seepage or is
evaporated back to the atmosphere at the soil surface. The groundwater storage contributes
QGW to total discharge QTOT. Furthermore, lateral flow from QLF,in and to QLF,out neighbouring
cells is calculated using slope and elevation. The groundwater store also returns some water
to the soil store as capillary flux Capflux. Isotope signatures i are calculated as mass balances
within each storage volume upon instantaneous and complete mixing.

This article is protected by copyright. All rights reserved.


Figure 4: The corrected climatic variables A) precipitation, B) relative humidity, C) solar
radiation and D) air temperature are shown as monthly boxplots summarizing the variability
observed over the complete period of meteorological measurements from 2008 to 2016. Each
box is restricted by the first and third quartile of data with the median indicated as a line. The
boxplot whiskers stretch to 1.5 times the height of the box. Extreme values outside the range
of the whiskers are defined as outliers and displayed as points.

This article is protected by copyright. All rights reserved.


Figure 5: A) The local MWL in red (P_gross) is compared to the GMWL (thin black line) with
throughfall samples (P_through) in blue, stream water samples (Q) in green and some
tributary streams (Q_trib) at low flow (7) in light purple. B) Regression relationship of
observed gross rainfall and throughfall (sampled at the same time) together with the 1:1 line
is directly used to parameterize the isotopic enrichment effect of the interception storage in
the STARR model.

This article is protected by copyright. All rights reserved.


Figure 6: The retained (after evaluation) 60 hourly A) streamflow (log-scale) and B) stream
deuterium simulations from 2015 (day 500) to January 2017 are plotted on a daily scale
against observations (black dots). The black line represents the best-fit simulation according
to the sum of all discharge and isotope performance criteria. The grey area presents the reach
of the other 59 simulations.

This article is protected by copyright. All rights reserved.


Figure 7: The MC simulation results expressed as histograms of performance indicators (A-
Nash Sutcliffe efficiency NSE, B-logarithmic NSE, C- the Kling Gupta efficiency KGE and D- the
isotope criterion) used to evaluate the model with the best-fit model used for further analysis
displayed as a single black dot.

This article is protected by copyright. All rights reserved.


Figure 8: The best-fit simulated model A) water fluxes and B) isotope signatures are plotted
for the different model components. Throughfall, direct runoff, soil store runoff and
groundwater runoff are given as daily averages in mm of all grid cells in the catchment with
discharge plotted in m3/s at the catchment outlet. Similarly, the isotope signatures
(deuterium in per mil) of gross precipitation (concP), the interception store (conci), the soil
store (concs) and in the groundwater store (concg), respectively are catchment averages
except for Qconc, which is the simulated stream deuterium at the catchment outlet.

This article is protected by copyright. All rights reserved.


Figure 9: The spatial variability of the simulated soil storage over time is shown for two
contrasting years (dry year 2014 and wet year 2016) and dates (end of the dry season 15 th
march and peak rainy season 15th October, both at noon). All maps are given at a 10 m pixel
size and with individual, colour-coded scales in mm.

This article is protected by copyright. All rights reserved.

You might also like