ArticlePDF Available

Impact of Rain Gauges Distribution on the Runoff Simulation of a Small Mountain Catchment in Southern Ecuador

Authors:

Abstract and Figures

In places with high spatiotemporal rainfall variability, such as mountain regions, input data could be a large source of uncertainty in hydrological modeling. Here we evaluate the impact of rainfall estimation on runoff modeling in a small páramo catchment located in the Zhurucay Ecohydrological Observatory (7.53 km2) in the Ecuadorian Andes, using a network of 12 rain gauges. First, the HBV-light semidistributed model was analyzed in order to select the best model structure to represent the observed runoff and its subflow components. Then, we developed six rainfall monitoring scenarios to evaluate the impact of spatial rainfall estimation in model performance and parameters. Finally, we explored how a model calibrated with far-from-perfect rainfall estimation would perform using new improved rainfall data. Results show that while all model structures were able to represent the overall runoff, the standard model structure outperformed the others for simulating subflow components. Model performance (NSeff) was improved by increasing the quality of spatial rainfall estimation from 0.31 to 0.80 and from 0.14 to 0.73 for calibration and validation period, respectively. Finally, improved rainfall data enhanced the runoff simulation from a model calibrated with scarce rainfall data (NSeff 0.14) from 0.49 to 0.60. These results confirm that in mountain regions model uncertainty is highly related to spatial rainfall and, therefore, to the number and location of rain gauges.
Content may be subject to copyright.
water
Article
Impact of Rain Gauges Distribution on the
Runoff Simulation of a Small Mountain
Catchment in Southern Ecuador
Adrián Sucozhañay 1, 2, * and Rolando Célleri 1,2
1Departamento de Recursos Hídricos y Ciencias Ambientales, Universidad de Cuenca, Av. 12 de Abril,
Cuenca 010207, Ecuador; rcelleri@gmail.com
2Facultad de Ingeniería, Universidad de Cuenca, Av. 12 de Abril, Cuenca 010203, Ecuador
*Correspondence: addysc1@gmail.com
Received: 10 July 2018; Accepted: 19 August 2018; Published: 31 August 2018


Abstract:
In places with high spatiotemporal rainfall variability, such as mountain regions, input data
could be a large source of uncertainty in hydrological modeling. Here we evaluate the impact
of rainfall estimation on runoff modeling in a small páramo catchment located in the Zhurucay
Ecohydrological Observatory (7.53 km
2
) in the Ecuadorian Andes, using a network of 12 rain gauges.
First, the HBV-light semidistributed model was analyzed in order to select the best model structure
to represent the observed runoff and its subflow components. Then, we developed six rainfall
monitoring scenarios to evaluate the impact of spatial rainfall estimation in model performance and
parameters. Finally, we explored how a model calibrated with far-from-perfect rainfall estimation
would perform using new improved rainfall data. Results show that while all model structures
were able to represent the overall runoff, the standard model structure outperformed the others for
simulating subflow components. Model performance (NSeff) was improved by increasing the quality
of spatial rainfall estimation from 0.31 to 0.80 and from 0.14 to 0.73 for calibration and validation
period, respectively. Finally, improved rainfall data enhanced the runoff simulation from a model
calibrated with scarce rainfall data (NSeff 0.14) from 0.49 to 0.60. These results confirm that in
mountain regions model uncertainty is highly related to spatial rainfall and, therefore, to the number
and location of rain gauges.
Keywords:
rainfall-runoff modeling; rainfall monitoring; precipitation estimation; modeling
uncertainty; páramo ecosystem
1. Introduction
Páramo is a high-elevation tropical Andean ecosystem located in the upper belt of the northern
Andes approximately from 3500 to 5000 m a.s.l. [
1
,
2
]. In southern Ecuador, it is characterized by
the presence of tussock grasses, wetlands, and scarce patches of Polylepis sp. [
3
5
]. Like other
mountain ecosystems worldwide recognized as water suppliers for downstream populations [
6
],
the Andean páramo is the most important water source for Andean cities such as Quito, Bogota,
Mérida, and Cuenca, mainly due to the high water retention capacity of its soils and the constant
precipitation it receives throughout the year [
7
,
8
]. In Ecuador, it also provides water for some of
the most important hydropower projects such as Paute Integral (2353 MW) and Coca-Codo Sinclair
(1500 MW) and irrigation projects in the inter-Andean valley [
9
]. However, human activities such
as grazing, cultivation, and pine plantations can alter their normal hydrological regulation [
10
,
11
].
Therefore, a good understanding of páramo hydrology is critical for present and future water resources
development [12].
Water 2018,10, 1169; doi:10.3390/w10091169 www.mdpi.com/journal/water
Water 2018,10, 1169 2 of 19
Research on páramo hydrology is relatively new and has focused mainly on understanding
hydrological processes such as evapotranspiration [
13
,
14
], interception [
15
], and temporal and
spatial variability of precipitation [
16
,
17
]; hydrological functioning such as hydrological landscape
controls
[18,19]
and water provenance and transit times [
20
22
]; impact of land use change [
10
,
11
,
23
];
and weather and climate [
24
26
]. Nevertheless, rainfall-runoff modeling (and the uncertainties related
to model structure, input data and model parameters) has still been little studied, even though it is key
for hydrological applications and water resources management.
Modeling of páramo catchments has concentrated on studying the impact of land use [
27
29
] and
hypothesis testing [
30
,
31
]. Nevertheless, given the high rainfall variability found in this region [
16
,
32
],
that can reach up to 25% volume differences between rain gauges within small catchments [
33
] and
the scarcity of spatiotemporal hydrometeorological data, it is clear that the lack of rainfall monitoring
can have a large impact on modeling [12].
Several studies have analyzed the impact of rainfall observations in model calibration and results
(e.g., time to peak, peak flows, volume, performance, and parameters) [
34
39
]. Younger et al. [
40
] posits
that synthetic rainfall perturbations, mostly located in the upper and lower catchment, induce changes
in peaks and model parameters at the outlet, highlighting the importance of an adequate estimation
of spatial precipitation. Faurés et al. [
41
] found differences of 2% to 65% in runoff volume if just one
of five rain gauges was used in a small catchment. Similar conclusions were provided by Bárdossy
and Das [
42
] and Xu et al. [
43
], showing that the performance of runoff simulation improves and the
uncertainty is reduced by increasing the number of rain gauges until a given threshold. In order to
overcome this measuring problem, Berne et al. [
44
] recommended spatial and temporal sampling
for modeling purposes, making use of high quality and quantity of rainfall data, which is in most
cases difficult to obtain. Nevertheless, it is not possible to generalize these results to other ecosystems,
mainly due to the fact that the impact of precipitation on runoff modeling will be influenced by the
characteristics of the catchments and storms [45,46].
Singh [
47
] concluded that the hydrological response of a catchment is related to the spatial and
temporal rainfall variability, suggesting that the quality of hydrological modeling can be related to the
capacity to measure this variability. Lobligeois et al. [
48
] tested this hypothesis using 181 catchments.
It was found that those with higher rainfall variability needed a denser rain gauge network for
improving runoff simulations. Finally, other studies have found that the quality of the model
performance is indeed related to the quality of precipitation information [49,50].
These studies have clearly established the importance of analyzing the effect of rain gauge density
for hydrological modeling [
51
]. However, none of the studies cited previously has been carried out
in high mountainous regions. Indeed, only a few studies such as Arnaud et al. [
52
] and Girons,
Lopez, and Seibert [
53
] have been carried out in high altitudes (with a maximum altitude of 2503
and above 5000 m a.s.l., respectively), leaving a gap in knowledge about this topic in mountain areas.
Closing this gap will also have a positive impact on (i) operational hydrology, by providing solid
recommendations to water managers about the design of rainfall observation networks in the region,
a problem that has been previously identified by stakeholders [
54
], and (ii) the implementation and
evaluation of watershed management programs as well as to confront the ecohydrological challenges
of the region [55,56].
In this context, the objective of this study is to evaluate the impact of precipitation estimation
in the hydrological simulation of a mountain catchment. For this purpose we implemented a
dense rainfall network (12 rain gauges) in the Zhurucay Ecohydrological Observatory (7.53 km
2
),
in southern Ecuador. The semidistributed conceptual rainfall-runoff HBV-light (Hydrologiska Byråns
Vattenbalansavdelning) model was chosen because has been widely used with good results in several
ecosystems including mountain ecosystems [
53
,
57
59
], and offers a good tradeoff between the amount
of information required and the spatial representation. First, we identified the best model structure to
simulate the total discharge and its subflow components (fast flow, interflow, and slow flow). Secondly,
we evaluated model performance and parameter sensibility using six rainfall-monitoring scenarios
Water 2018,10, 1169 3 of 19
using from 1 to 11 rain gauges. Finally, we explored one open question of Andean water managers.
A common monitoring setting in mountain areas consists of having a single rain gauge station
outside or in the outlet of the catchment area. Models are then calibrated using this configuration,
evidently leading to low simulation efficiencies. However, as the catchment becomes more important
(e.g., for water resources development), additional rain gauges are installed within the catchment.
But, modelers will have to wait several years under this new monitoring system to recalibrate the
model. This raises the question of how a model calibrated with a far-from-perfect rainfall estimation
will perform using new improved rainfall data, i.e., without undergoing a recalibration process.
Thus, we sought to answer this question.
2. Study Area, Monitoring and Data Availability
2.1. Study Area
This study was carried out in the Zhurucay Ecohydrological Observatory located in the southern
Ecuadorian Andes, near the continental divide, draining into the Pacific Ocean. The observatory
consists of a nested catchment of 7.53 km
2
with elevation ranging from 3400 to 3900 m a.s.l. (Figure 1).
Mean annual precipitation is 1345 mm at 3780 m a.s.l. with weak seasonality. Rain mainly falls as
drizzle and occurs almost daily [
17
]. Mean annual temperature is 6.0
C [
14
]. The geomorphology
of the catchment is glacial U-shaped. The average slope is 10
, although slopes up to 73
are found.
The geology is compacted volcanic rock deposits formed during the last ice age [
60
]. In the northeastern
of the catchment there is a ponded wetland at a flat hilltop. This structure drains towards the outlet of
subcatchment S7 and occupies almost the entire subcatchment [21].
Water 2018, 10, x FOR PEER REVIEW 3 of 19
rain gauge station outside or in the outlet of the catchment area. Models are then calibrated using this
configuration, evidently leading to low simulation efficiencies. However, as the catchment becomes
more important (e.g., for water resources development), additional rain gauges are installed within
the catchment. But, modelers will have to wait several years under this new monitoring system to
recalibrate the model. This raises the question of how a model calibrated with a far-from-perfect
rainfall estimation will perform using new improved rainfall data, i.e., without undergoing a
recalibration process. Thus, we sought to answer this question.
2. Study Area, Monitoring and Data Availability
2.1. Study Area
This study was carried out in the Zhurucay Ecohydrological Observatory located in the southern
Ecuadorian Andes, near the continental divide, draining into the Pacific Ocean. The observatory
consists of a nested catchment of 7.53 km2 with elevation ranging from 3400 to 3900 m a.s.l. (Figure
1). Mean annual precipitation is 1345 mm at 3780 m a.s.l. with weak seasonality. Rain mainly falls as
drizzle and occurs almost daily [17]. Mean annual temperature is 6.0 °C [14]. The geomorphology of
the catchment is glacial U-shaped. The average slope is 10°, although slopes up to 73° are found. The
geology is compacted volcanic rock deposits formed during the last ice age [60]. In the northeastern
of the catchment there is a ponded wetland at a flat hilltop. This structure drains towards the outlet
of subcatchment S7 and occupies almost the entire subcatchment [21].
Figure 1. Study area and monitoring network. Coordinate system: UTM WGS84 17S.
The main soil type in Zhurucay are Andosols (Ah horizon), covering 80% of the area and
commonly located on the hillslopes. Histosols (H horizon) cover the 20% remaining area and are
commonly located in flat areas where geomorphology allows water accumulation. These soils,
formed in wetlands, are highly organic and are saturated most of the year [19]. Both soil types were
Figure 1. Study area and monitoring network. Coordinate system: UTM WGS84 17S.
Water 2018,10, 1169 4 of 19
The main soil type in Zhurucay are Andosols (Ah horizon), covering 80% of the area and
commonly located on the hillslopes. Histosols (H horizon) cover the 20% remaining area and are
commonly located in flat areas where geomorphology allows water accumulation. These soils, formed
in wetlands, are highly organic and are saturated most of the year [
19
]. Both soil types were formed by
volcanic ash accumulation; they are black, humic, acid, and rich in organic carbon and are located over
a mineral horizon (C horizon) commonly rich in clay [21,61].
The vegetation within the catchment is highly correlated with soil type [
19
] and is typical of
páramo grasslands. Tussock grasses grow in Andosols while cushion plants grow in Histosols
wetlands [
4
,
62
], covering 70% and 25% of the area, respectively. The remaining 5% of the area is
covered by polylepis trees and pine forest. Since the percentage of polylepis and pine forest is too low
within each subcatchment, these vegetation covers were not included in the study. Further description
of landscape characteristics are provided in Mosquera et al. [19].
2.2. Monitoring and Data Availability
The Zhurucay Ecohydrological Observatory was established in 2009 with the purpose of studying
the hydrological functioning of Andean mountain catchments. Over the years, it has been equipped
with two automatic meteorological stations, a laser disdrometer, five permanent rain gauges, a nested
hydrological network consisting of 10 weirs, a hillslope to study subsurface processes, and an
environmental water quality monitoring system to study runoff flowpaths and runoff formation
using isotopes and metals (in soils, streams, rainfall, and springs). Recent additions include an energy
balance and eddy covariance system, and small-scale lysimeters.
At the beginning of 2013, the hydrological network was fully operational. For this experiment,
the rain gauge network (five permanent rain gauges) was complemented with a total of 12 rain gauges:
11 were located inside the catchment and one was located 2 km downstream of the outlet. During the
experiment, this rain gauge network was arguably the densest network in the Andes mountains at this
spatial scale (1.46 rain gauges per km
2
). The extended network operated from 2014 to 2016. Hence,
daily data of precipitation, temperature, potential evapotranspiration, and runoff corresponding to
the period October 2013 to October 2016 were selected for this study. These variables were used
according to the requirements of the HBV-light model. Precipitation data for the first study year was
obtained from the five permanent rain gauges, and for the remaining years from the 12 rain gauges.
Temperature and meteorological variables to estimate potential evapotranspiration were acquired
from a weather station at 3780 m a.s.l. Runoff data were obtained from V-notch weirs at the outlets of
seven subcatchments (S1 to S7) and from one rectangular weir at the outlet of Zhurucay catchment
(S8) which is the focus of this study. Figure 1shows the distribution of sensors and subcatchments.
Subcatchments aggregated area and vegetation cover are presented in Table 1which was provided by
Mosquera et al. [19].
Table 1. Subcatchment aggregated area and vegetation cover.
Subcatchment Aggregated Area (km2)Vegetation Cover (%)
Wetland Tussock Grass
S1 0.20 15 85
S2 0.38 13 87
S3 0.38 18 82
S4 0.65 18 82
S5 1.40 17 83
S6 3.28 24 76
S7 1.22 65 35
S8 7.53 25 75
Water 2018,10, 1169 5 of 19
3. Methodology
First, we calibrated and validated each model structure using all rain gauges located inside the
catchment. Each model structure was evaluated and compared according to the ability to simulate
total runoff and its subflow components in order to select the model structure with the highest
performance. Then, the model structure selected previously was calibrated and validated using six
rainfall monitoring scenarios to evaluate the impact of rainfall estimation on the simulated runoff and
parameters. Finally, the calibrated parameters with the worst rainfall scenario were employed to run
the model with the remaining scenarios to analyze the possibility to enhance runoff simulation by
improving rainfall information. The methodological details are explained in the following subsections.
3.1. HBV-Light Model
The HBV-light model [
63
] is a semidistributed conceptual rainfall-runoff model based on the
original HBV model developed by Bergström [
64
] and Bergström [
57
] at the Swedish Meteorological
and Hydrological Institute (SMHI). This model can be distributed into different elevation and
vegetation zones as well as subcatchments. HBV-light uses four routines to simulate runoff: (i) the snow
routine represents snow accumulation and snow melt; (ii) the soil routine describes ground water
recharge and actual evaporation as function of water storage for each elevation or vegetation zone;
(iii) the response routine computes the runoff as function of water stored in reservoirs; and (iv) the
routing routine simulates the routing of the runoff to the outlet of the catchment by a triangular
weighting function. Before the routing routine, runoff from previous subcatchments are added to the
generated runoff of the current subcatchment. A detailed description of the model can be found in
Seibert and Vis [65].
The standard model consists of two serial reservoirs that receive water from the semidistributed
soil routine. Storage in the upper soil reservoir (SUZ) simulates fast flow and interflow, representing
the near surface and subsurface flow, respectively; while storage in the lower soil reservoir (SLZ)
simulates slow flow, representing baseflow. Both reservoirs are connected by a percolation rate. In total,
HBV-light offers 11 different structures of reservoirs varying from one to three reservoirs with varying
degrees of spatial distribution according to elevation and vegetation zones. Further details of model
structures are found in Uhlenbrook et al. [66].
For this study, the Zhurucay catchment was divided into eight subcatchments and two vegetation
zones representing tussock grasses (grassland) and cushion plants (wetlands). Runoff for each
subcatchment were simulated considering the contribution of previous subcatchments, and therefore,
the runoff simulation of subcatchment S8 is the simulation of the whole Zhurucay catchment. Structures
related to snow and very slow flow were not used since Zhurucay does not receive snow. Eight model
structures were used in this study, as presented in Table 2.
Table 2. HBV-light model structures.
ID Structure a
M1 Two boxes (SUZ and SLZ, standard).
M2 Two boxes (SUZ and SLZ). UZL threshold.
M3 Two boxes (SUZ and SLZ). SUZ distributed.
M4 Two boxes (SUZ and SLZ). SUZ distributed and UZL threshold.
M5 Three boxes (STZ, SUZ and SLZ).
M6 Three boxes (STZ, SUZ and SLZ). STZ distributed.
M7 Three boxes (STZ, SUZ and SLZ). STZ and SUZ distributed.
M8 One box. UZL and PERC thresholds.
a
STZ = Storage in top zone; SUZ = Storage in upper zone; SLZ = Storage in lower zone; UZL = Threshold parameter
above which overland flow is produced; PERC = Threshold parameter above which interflow is produced.
Water 2018,10, 1169 6 of 19
3.2. Model Calibration and Validation
The HBV-light model was calibrated using the standard split sample model calibration
procedure [
67
,
68
]. The first, second, and third year of data (October 2013–October 2016) were used for
model warming up, calibration, and validation, respectively. A year of data for each calibration and
validation period were considered sufficient because of they cover the hydrological cycle of the study
catchment. For the warming up period, spatial rainfall was estimated from the five permanent
rain gauges (R03, R06, R07, R08, and R12), by inverse distance weighting (IDW) interpolation.
This interpolation method was used due to the reduced number of rain gauges available. For the
calibration and validation periods (years two and three) rainfall was estimated from the 11 rain gauges
located inside the catchment, interpolated by ordinary Kriging. Potential evapotranspiration was
estimated by the Penman–Monteith method [69] which has been used previously in Zhurucay [14].
The Monte Carlo procedure established by the HBV-light model was used to select an optimal
parameter set after performing 10,000 simulations. Table 3lists all parameters used and their
calibration range. Parameter sets were generated using random numbers from a uniform distribution.
The recession coefficient for fast flow (K0) was not calibrated and was set to 0.99 day
1
(close to
one day). Interflow (K1) and slow flow (K2) coefficients were calibrated between relative fast
ranges. These considerations were taken due to the fast recession observed in these subcatchments.
The simulation results for each set of parameters in the Monte Carlo procedure were optimized by
selecting the Nash-Sutcliffe efficiency (NSeff) [70] as the objective function.
Table 3. HBV-light model parameters and their range used in the Monte Carlo procedure.
Parameter Description Unit Minimum Maximum
Soil routine a
FC Maximum soil moisture mm 250 400
LP
Soil moisture (MS) above which actual
evapotranspiration reaches potential
evapotranspiration (MS/FC)
- 0.5 1
BETA Relative contribution to runoff from
rain or snowmelt - 1 3
Response routine
K0 bRecession coefficient for quick flow day10.999 0.999
K1 Recession coefficient for interflow day10.33 0.999
K2 Recession coefficient for baseflow day10.066 0.2
ALPHA cNonlinearity coefficient - 0 1
UZL dThreshold parameter for K0 outflow mm 0 30
PERC ePercolation from SUZ to SLZ mm day10 2
UZL fPercolation from STZ to SUZ mm day10 30
PERC gThreshold parameter for K1 outflow mm 0 2
Routing routine
MAXBAS
Length of triangular weighting function
d 1 1.5
a
Parameters for each vegetation cover;
b
Parameter used only for structures M2, M4, M5, M6, M7, and M8;
c
Parameter used only for structures M1 and M3;
d
Parameter used only for structures M2, M4 and M8;
e
Parameter used only for structures M1, M2, M3, M4, M5, M6, and M7;
f
Parameter used only for structures
M5, M6, and M7;
g
Parameter used only for structure M8; FC is the previous parameter and the expression defines
the relationship to obtain LP.
Water 2018,10, 1169 7 of 19
3.3. Evaluation of Model Structures
Model structures were evaluated in two steps. First, for each model structure, HBV-light
was calibrated and validated based on NSeff. Nevertheless, other performance indexes such as
Nash-Sutcliffe efficiency with logarithms (Log NSeff), coefficient of determination (R
2
) and percentage
annual difference (%ADiff) were additionally calculated to provide more information about the quality
of the simulation [
71
,
72
]. NSeff has been widely used to evaluate the performance of hydrological
models and it is sensitive to differences in the observed and simulated means and variances. Log NSeff
is the NSeff using the logarithm of the runoff values reducing its sensitivity of extreme values.
R
2
describes how much of the observed dispersion is explained by the prediction. %ADiff is the
percentage difference of the total observed runoff against the total simulated runoff used to observe
the under or overestimation in the water balance.
Second, to identify the most behavioral models, the simulated runoff generated from the different
storages of the model structures (e.g., STZ, SUZ, and SLZ) representing fast flow (FF), interflow (IF),
and slow flow (SF) were compared to the flow components derived from measured runoff using the
Water Engineering Time Series PROcessing tool (WETSPRO) [
73
]. WETSPRO is a tool that extracts the
subflow components of a hydrograph based on Chapman filter [74] and using its recession constants
(K) and the average fraction of each subflow component over the total flow (w). This tool has been
used successfully to separate flow components and to test models by multicriteria approach [
75
77
].
Since model structures M1 and M3 simulate only two flow components, e.g., slow flow and the
combination of interflow and fast flow, WETSPRO-derived interflow and fast flow were added for an
adequate comparison.
3.4. Rainfall Monitoring Scenarios
We selected six rainfall-monitoring scenarios. These scenarios ranged from using all the rainfall
network to using a single rain gauge located outside the catchment; they are described in Table 4.
The first scenario (referred to as E0) uses 11 rain gauges inside the catchment and therefore provides
the best rainfall estimation. The remaining five monitoring scenarios (E1–E5) were selected based
on common practices used by water managers in the region. They simulate an scarce network by
reducing the number of gauges as follows: one configuration using three rain gauges located in
the upper, middle, and lower catchment (R01, R06, and R11) and four configurations using only
one rain gauge in each one, located in the upper (R01), middle (R06), lower (R11), and outside the
catchment (R12). For scenarios based on one rain gauge, spatial rainfall was considered uniform
throughout the catchment. For the three-rain-gauge scenario, spatial rainfall was interpolated by
inverse distance weighting (IDW). This method was selected rather than ordinary kriging due to the
low rain gauge density.
Table 4. Rainfall monitoring scenarios.
ID Rain Gauge(s) Spatial Rainfall Estimation Method
E0 R01 to R11 Ordinary kriging
E1 R01, R06, R11 IDW
E2 R01 Punctual
E3 R06 Punctual
E4 R11 Punctual
E5 R12 Punctual
3.5. Evaluation of the Quality of Spatial Rainfall Estimation
The best rainfall estimation (E0) was compared to the estimations obtained from the different
monitoring scenarios (E1 to E5) as a way to understand the impact of the degraded rain gauge network.
For this purpose, we used an analysis of scatter plot, R
2
and the Goodness of Rainfall Estimation
Water 2018,10, 1169 8 of 19
index (GORE) [
50
]. GORE was used to quantify the quality of the estimated rainfall time distribution.
GORE is the transposition of NSeff in the precipitation domain, using the square root of the rainfall
data to reduce the weight of extreme events and is expressed as:
GORE =1
n
i=1qPE
iPi2
n
i=1PiP2(1)
where, nis the number of time steps of the period,
PE
i
is the estimated rainfall from rainfall monitoring
scenario at time
i
,
Pi
is the best estimated rainfall available from rainfall monitoring scenario at time
i
,
and
P
is the mean of the best estimated rainfall available (considered as reference) over the study
period. Like NSeff, the GORE index varies between
and 1, where 1 represents that the estimated
rainfall is equal to the reference. GORE will be smaller as the rainfall estimates become poorer.
3.6. Evaluation of the Impact of Spatial Rainfall Estimation on Model Calibration
To evaluate the effect of the reduction of precipitation information by operating a sparse
precipitation network on hydrological simulation, the HBV-light structure selected in Section 3.3
was calibrated and validated using each of the rainfall scenarios in Section 3.4. The performance of the
simulated runoff (using NSeff) and calibrated parameters were related to the quality of the estimated
rainfall (GORE) used to run the model.
Besides GORE index, BALANCE index was used to quantify the overestimation or
underestimation of the total estimated rainfall and its relation to the model performance (NSeff) [
50
].
BALANCE index is greater than 1 in the case of rainfall overestimation and smaller than 1 in the case
of underestimation, and is expressed as:
BALANCE index =
n
i=1PE
i
n
i=1Pi
(2)
3.7. Effect of Improved Rainfall Estimation on A Model Calibrated with Scarce Data
For this objective, the model calibrated with the worst rainfall scenario was run using the input
rainfall of the estimates from the remaining rainfall scenarios in a step-wise fashion, i.e., by improving
the spatial rainfall estimation step-by-step up to the reference scenario (E0). Then, the performances
of runoff simulations (NSeff index) were analyzed regarding the quality of rainfall information
(GORE index) used to run the model.
4. Results and Discussion
4.1. Evaluation of the Model Structures
The performances of the eight model structures, for each subcatchment and for calibration and
validation periods are shown in Figure 2. Each line represents a model structure and the overlapping
of these lines indicates that all model structures have very similar performances. Indexes values
can be found in supplementary Tables S1 and S2. From the magnitudes of NSeff, LogNSeff, R
2
,
and %ADiff, it is hard to identify a model structure that outperforms the others, as there is little
difference among them. Similar results were found by Uhlenbrook et al. [
66
] using six structures of
HBV model in a mountainous catchment in Germany where NSeff values varied between 0.825 and
0.876 in the calibration period. This suggests that model parameters compensate for the differences
among structures. Additionally, similar results are observed for all subcatchments, which can be
explained by two reasons. First, the reason explained previously and second, that physical differences
between subcatchments are not large enough to produce an impact in specific model structures [78].
Water 2018,10, 1169 9 of 19
Water 2018, 10, x FOR PEER REVIEW 9 of 19
The average of NSeff for all subcatchments shows low values for slow flow (SF) and fast flow
(FF), below 0.37 and 0.20, respectively. For most of the structures, NSeff values showed high
variability between subcatchments; for example, for the structure M2, slow flow NSeff ranges from
0.52 to 0.01. This shows that while model parameters can compensate the overall performance of a
given model structure, the simulated subflow components do not represent the physical functioning
of the catchments, and therefore these models are not behavioral [79].
Figure 2.
Performance indexes for the simulated runoff. Left column for calibration period and right
column for validation period. Each axis represents a subcatchment (S1–S8) and each colored line
represents a model structure (M1–M8).
For the catchment S8, which is the outlet of Zhurucay and the focus of this study, the NSeff values
from eight structures vary between 0.80 and 0.83 and 0.68 and 0.73 for calibration and validation,
Water 2018,10, 1169 10 of 19
respectively. These results were similar than those obtained by Buytaert and Beven [
30
] in a páramo
catchment of 2.53 km
2
with similar vegetation cover and altitude. The study evaluated nine model
structures with NSeff values between 0.72 and 0.87 and 0.45 and 0.77 for calibration and validation,
respectively; showing that HBV-light structures reached expected efficiencies for this ecosystem.
Considering these high NSeff values and the high R
2
values (over 0.82), one might consider that
the eight structures can represent the overall runoff dynamics. However, lower values of LogNSeff
compared to NSeff suggest that structures may have problems to simulate low flows [
71
]. Therefore,
we compared the observed and simulated subflow components for each model structure. This was
done for the S8 catchment and the average of S1–S8 (Table 5).
Table 5.
Nash–Sutcliffe efficiency for simulated flow components from each model structure. For outlet
subcatchment S8 and the average of S1–S8.
Model Structure Flow Components S8 Average Flow Component (S1–S8)
SF IF FF SF IF FF
M1 0.42 0.70 0.32 0.72
M2 0.17 0.68 0.67 0.07 0.61 0.08
M3 0.33 0.67 0.37 0.71
M4 0.28 0.64 0.72 0.13 0.60 0.27
M5 0.39 0.48 0.66 0.27 0.52 0.16
M6 0.35 0.53 0.65 0.05 0.55 0.20
M7 0.36 0.62 0.66 0.19 0.55 0.09
M8 0.31 0.63 0.70 0.05 0.60 0.30
The average of NSeff for all subcatchments shows low values for slow flow (SF) and fast flow
(FF), below 0.37 and 0.20, respectively. For most of the structures, NSeff values showed high variability
between subcatchments; for example, for the structure M2, slow flow NSeff ranges from 0.52 to
0.01. This shows that while model parameters can compensate the overall performance of a given
model structure, the simulated subflow components do not represent the physical functioning of the
catchments, and therefore these models are not behavioral [79].
For catchment S8, NSeff values for the eight model structures present similar results between
subflow components, ranging from 0.17 to 0.42, 0.48 to 0.70, and 0.65 to 0.72 for SF, IF, and FF,
respectively. These results indicate that SF was difficult to simulate; indeed, all model structures
had NSeff values below 0.42 for this subflow. M1 was the model structure that obtained the highest
performances (0.42 SF and 0.70 IF). Therefore, the M1 structure with two reservoirs that simulate
slow flow and the combination of interflow and fast flow was chosen for the remaining of the study.
The fact that the simplest structure showed the best performance in simulating all subflows suggests
that the páramo ecosystem is hydrologically relatively simple [
19
], and that its water storage–release
processes are mainly controlled by water moving laterally through the organic soils to the streams [
22
].
Thus, a two-reservoir model structure provided an accurate simulation of the observed runoff of
the catchment.
The best simulation according to NSeff (0.8 and 0.72 for calibration and validation, respectively)
using the M1 structure for the subcatchment S8 (Zhurucay catchment) is shown in Figure 3. In addition,
the optimal parameters found for this simulation are listed in Table 6. Temporarily we can observe
that in general the simulation fits well to the dry and humid periods. However, there is a slight
subestimation of the simulation in the peaks.
Water 2018,10, 1169 11 of 19
Water 2018, 10, x FOR PEER REVIEW 10 of 19
Figure 2. Performance indexes for the simulated runoff. Left column for calibration period and right
column for validation period. Each axis represents a subcatchment (S1–S8) and each colored line
represents a model structure (M1–M8).
For catchment S8, NSeff values for the eight model structures present similar results between
subflow components, ranging from 0.17 to 0.42, 0.48 to 0.70, and 0.65 to 0.72 for SF, IF, and FF,
respectively. These results indicate that SF was difficult to simulate; indeed, all model structures had
NSeff values below 0.42 for this subflow. M1 was the model structure that obtained the highest
performances (0.42 SF and 0.70 IF). Therefore, the M1 structure with two reservoirs that simulate slow
flow and the combination of interflow and fast flow was chosen for the remaining of the study. The
fact that the simplest structure showed the best performance in simulating all subflows suggests that
the páramo ecosystem is hydrologically relatively simple [19], and that its water storage–release
processes are mainly controlled by water moving laterally through the organic soils to the streams
[22]. Thus, a two-reservoir model structure provided an accurate simulation of the observed runoff
of the catchment.
Table 5. Nash–Sutcliffe efficiency for simulated flow components from each model structure. For
outlet subcatchment S8 and the average of S1–S8.
Model Structure Flow Components S8 Average Flow Component (S1–S8)
SF IF FF SF IF FF
M1 0.42 0.70
0.32 0.72
M2 0.17 0.68 0.67 0.07 0.61 0.08
M3 0.33 0.67
0.37 0.71
M4 0.28 0.64 0.72 0.13 0.60 0.27
M5 0.39 0.48 0.66 0.27 0.52 0.16
M6 0.35 0.53 0.65 0.05 0.55 0.20
M7 0.36 0.62 0.66 0.19 0.55 0.09
M8 0.31 0.63 0.70 0.05 0.60 0.30
The best simulation according to NSeff (0.8 and 0.72 for calibration and validation, respectively)
using the M1 structure for the subcatchment S8 (Zhurucay catchment) is shown in Figure 3. In
addition, the optimal parameters found for this simulation are listed in Table 6. Temporarily we can
observe that in general the simulation fits well to the dry and humid periods. However, there is a
slight subestimation of the simulation in the peaks.
Figure 3.
(
a
) Best simulated (according to NSeff) runoff for the subcatchment S8 (Zhurucay catchment)
using the M1 structure (with the optimal parameters) and the observed runoff. (
b
) Best simulated
(according to NSeff) runoff for the subcatchment S8 (Zhurucay catchment) using the M1 structure
(with the optimal parameters) and its uncertainty band corresponding to the 5–95% confidence limits
of the possible solutions from the 10,000 Monte Carlo simulations. The dashed line indicates the end of
the calibration period and the start of the validation period.
Table 6. Optimal parameters from the best runoff simulation according to NSeff.
Routine Parameter Value
Soil routine 1 (Wetland)
FC_1 294.44
LP_1 0.97
BETA_1 2.79
Soil routine 2 (Tussock grass)
FC_2 254.11
LP_2 1.00
BETA_2 2.46
Response routine
PERC 1.90
ALPHA 0.43
K1 0.42
K2 0.16
Routing routine MAXBAS 1.00
4.2. Evaluation of the Estimated Rainfall from Rainfall Monitoring Scenarios
The comparison of the daily rainfall estimated from the five monitoring scenarios (E1–E5) against
the best rainfall estimation for the catchment (E0) is shown in Figure 4. It is observed that the scenario
E1 using three rain gauges distributed in the upper, middle, and lower catchment represents well the
spatial rainfall. For scenarios that use only one rain gauge to represent the spatial rainfall, scenario E3
(placing the rain gauge in the middle of the catchment) gives the best estimation. This result is similar
to results observed by Hrachowitz and Weiler [
80
], and suggests that the rainfall observation in the
middle of the catchment is very similar to the average of the rainfall of the entire catchments. On the
other hand, it is observed that a rain gauge located in the upper catchment (E3) has better results than
a rain gauge located in the lower catchment (E4). The quality of the estimation is reduced significantly
when using rainfall observations from the rain gauge at 2 km downstream of the outlet (E5), producing
Water 2018,10, 1169 12 of 19
a large scatter and overestimation. These results indicate that in this mountain setting there is a large
rainfall variability at short distances, in line with results found by Buytaert et al. [
33
] who indicated that
páramo rainfall can be highly variable, even at short distances of 4 km, suggesting a strong orographic
influence in rainfall.
Water 2018, 10, x FOR PEER REVIEW 12 of 19
Figure 4. Scatter plot, determination coefficient (R2), and GORE of the estimated spatial rainfall of
catchment S8 from monitoring scenarios E1–E5 (axis X) against the reference estimated rainfall from
monitoring scenario E0 (axis Y).
4.3. Evaluation of the Impact of Rainfall Estimation on Hydrological Simulation
The impact of the rainfall estimation from the six monitoring scenarios on the performance of
the simulated runoff is presented in Figure 5. This figure shows the model performance (NSeff)
against the rainfall scenario (Figure 5a), the quality of the estimated rainfall (GORE, Figure 5b), and
the over- or underestimation of rainfall (Balance index, Figure 5c).
As can be seen in Figure 5a, rainfall scenarios E0 and E1 produce very similar, good results, for
both calibration and validation periods. This corroborates that the good rainfall quality of scenario
E1 identified in Section 4.2, is also translated into a good runoff simulation. On the other hand, and
as expected, the rainfall scenario with the worst rainfall quality (E5) produces the poorest runoff
simulation (below 0.31 NSeff). Similar to results of Zhang and Han [81], we can observe in the rainfall
scenario E5 that the model performance decreased drastically when the spatial rainfall variability is
not considered. Model performance using rainfall scenarios with one rain gauge deteriorates when
compared to the best spatial rainfall. And, although the validation period seems satisfactory for
scenarios E2 and E3, the efficiency in the calibration period is significantly reduced. These results
show how inadequate rainfall monitoring can produce high modeling uncertainty [36].
Additionally, we can observe in Figure 4 that E3 scenario shows better correlation than E4 with
E0, however, in Figure 5, the E4 scenario gives better model performance for the calibration period
compared with E3. The reason of the variation of the performance between calibration and validation
periods can be due to annual variations that can be strongly affected by specific orographic factors.
Results also suggest that there is a direct relation between the rainfall quality (GORE) and the
simulated runoff performance (NSeff), as can be seen in Figure 5b. Additionally, the slope of this
relation is 0.88 indicating that a slight reduction in the quality of the estimated rainfall can produce a
big reduction in the performance of the runoff simulation. This relation was also found in some
Mediterranean catchments in France [49,50]. Nevertheless, the slope found in this study is more
pronounced compared to the results of Andréassian et al. [50], which suggests a stronger influence
of the quality of rainfall input on the runoff simulation in this mountain catchment.
Finally, the highest performance of the simulated runoff was obtained when the over- or
underestimation of rainfall was minimal. According to the results of Andréassian et al. [50], it was
expected that the performance of runoff simulation would increase when the Balance Index is closer
to 1. This was found true when the Balance Index ranged from 1.07 to 0.97 (Figure 5c). However,
Figure 4.
Scatter plot, determination coefficient (R
2
), and GORE of the estimated spatial rainfall of
catchment S8 from monitoring scenarios E1–E5 (axis X) against the reference estimated rainfall from
monitoring scenario E0 (axis Y).
4.3. Evaluation of the Impact of Rainfall Estimation on Hydrological Simulation
The impact of the rainfall estimation from the six monitoring scenarios on the performance of the
simulated runoff is presented in Figure 5. This figure shows the model performance (NSeff) against the
rainfall scenario (Figure 5a), the quality of the estimated rainfall (GORE, Figure 5b), and the over- or
underestimation of rainfall (Balance index, Figure 5c).
Water 2018, 10, x FOR PEER REVIEW 13 of 19
above this range we found uncorrelated values of NSeff and BI (i.e., while for a BI of 1.11 we found a
very low NSeff of 0.31, for a higher BI of 1.19 we found a NSeff of 0.63).
Figure 5. Impact of rainfall quality on model performance (NSeff) against: (a) rainfall monitoring
scenario, (b) GORE index, and (c) Balance Index. For calibration and validation period.
The sensibility of model parameters to rainfall quality is shown in Figure 6. ALPHA and BETA
parameters are plotted against the GORE index. These two parameters were selected to illustrate their
sensibility to rainfall quality.
The only parameter sensible to rainfall quality was ALPHA (Figure 6a). In the study of
Andréassian et al. [50], the sensibility is expressed as the reduction of the variability of the parameter
values as the rainfall quality increases until reaching an optimal parameter value. Our study
indicated that the higher the GORE, the higher the ALPHA. On the other hand, Figure 6b shows
BETA as example of a parameter showing no sensibility to rainfall quality. In contrast to the findings
of Andréassian et al. [50], we found that the majority of parameters were not sensible to rainfall
quality. This may be caused by the relative high number of parameters which allows an
overadjustment to the input data compared to the three and six parameters used in Andréassian et
al. [50]. Although, it is still an open question if the sensibility of parameters to the rainfall quality is a
function of the number of parameters used by the model. Nevertheless, at least one parameter was
found sensible, showing that despite this situation, rainfall quality may still have impact on the
parameterization. This can be due to a correct estimation of precipitation which allows a better
description of ALPHA parameter (which controls the amount of water that recharges the streams) to
be adapted to the high water recharge of these soils.
Figure 6. Sensibility of model parameters to rainfall quality (GORE). (a) For a parameter sensible,
ALPHA and (b) for a parameter no sensible, BETA.
4.4. Effect of Improved Rainfall Estimation in A Calibrated Model
In this section, we analyzed the performance of the runoff simulation obtained from a calibrated
model with far-from-perfect rainfall using new improved rainfall estimations. In this way, the model
Figure 5.
Impact of rainfall quality on model performance (NSeff) against: (
a
) rainfall monitoring
scenario, (b) GORE index, and (c) Balance Index. For calibration and validation period.
As can be seen in Figure 5a, rainfall scenarios E0 and E1 produce very similar, good results,
for both calibration and validation periods. This corroborates that the good rainfall quality of scenario
E1 identified in Section 4.2, is also translated into a good runoff simulation. On the other hand,
and as expected, the rainfall scenario with the worst rainfall quality (E5) produces the poorest runoff
Water 2018,10, 1169 13 of 19
simulation (below 0.31 NSeff). Similar to results of Zhang and Han [
81
], we can observe in the rainfall
scenario E5 that the model performance decreased drastically when the spatial rainfall variability is
not considered. Model performance using rainfall scenarios with one rain gauge deteriorates when
compared to the best spatial rainfall. And, although the validation period seems satisfactory for
scenarios E2 and E3, the efficiency in the calibration period is significantly reduced. These results show
how inadequate rainfall monitoring can produce high modeling uncertainty [36].
Additionally, we can observe in Figure 4that E3 scenario shows better correlation than E4 with
E0, however, in Figure 5, the E4 scenario gives better model performance for the calibration period
compared with E3. The reason of the variation of the performance between calibration and validation
periods can be due to annual variations that can be strongly affected by specific orographic factors.
Results also suggest that there is a direct relation between the rainfall quality (GORE) and the
simulated runoff performance (NSeff), as can be seen in Figure 5b. Additionally, the slope of this
relation is 0.88 indicating that a slight reduction in the quality of the estimated rainfall can produce
a big reduction in the performance of the runoff simulation. This relation was also found in some
Mediterranean catchments in France [
49
,
50
]. Nevertheless, the slope found in this study is more
pronounced compared to the results of Andréassian et al. [
50
], which suggests a stronger influence of
the quality of rainfall input on the runoff simulation in this mountain catchment.
Finally, the highest performance of the simulated runoff was obtained when the over- or
underestimation of rainfall was minimal. According to the results of Andréassian et al. [
50
], it was
expected that the performance of runoff simulation would increase when the Balance Index is closer
to 1. This was found true when the Balance Index ranged from 1.07 to 0.97 (Figure 5c). However,
above this range we found uncorrelated values of NSeff and BI (i.e., while for a BI of 1.11 we found a
very low NSeff of 0.31, for a higher BI of 1.19 we found a NSeff of 0.63).
The sensibility of model parameters to rainfall quality is shown in Figure 6. ALPHA and BETA
parameters are plotted against the GORE index. These two parameters were selected to illustrate their
sensibility to rainfall quality.
Water 2018, 10, x FOR PEER REVIEW 13 of 19
above this range we found uncorrelated values of NSeff and BI (i.e., while for a BI of 1.11 we found a
very low NSeff of 0.31, for a higher BI of 1.19 we found a NSeff of 0.63).
Figure 5. Impact of rainfall quality on model performance (NSeff) against: (a) rainfall monitoring
scenario, (b) GORE index, and (c) Balance Index. For calibration and validation period.
The sensibility of model parameters to rainfall quality is shown in Figure 6. ALPHA and BETA
parameters are plotted against the GORE index. These two parameters were selected to illustrate their
sensibility to rainfall quality.
The only parameter sensible to rainfall quality was ALPHA (Figure 6a). In the study of
Andréassian et al. [50], the sensibility is expressed as the reduction of the variability of the parameter
values as the rainfall quality increases until reaching an optimal parameter value. Our study
indicated that the higher the GORE, the higher the ALPHA. On the other hand, Figure 6b shows
BETA as example of a parameter showing no sensibility to rainfall quality. In contrast to the findings
of Andréassian et al. [50], we found that the majority of parameters were not sensible to rainfall
quality. This may be caused by the relative high number of parameters which allows an
overadjustment to the input data compared to the three and six parameters used in Andréassian et
al. [50]. Although, it is still an open question if the sensibility of parameters to the rainfall quality is a
function of the number of parameters used by the model. Nevertheless, at least one parameter was
found sensible, showing that despite this situation, rainfall quality may still have impact on the
parameterization. This can be due to a correct estimation of precipitation which allows a better
description of ALPHA parameter (which controls the amount of water that recharges the streams) to
be adapted to the high water recharge of these soils.
Figure 6. Sensibility of model parameters to rainfall quality (GORE). (a) For a parameter sensible,
ALPHA and (b) for a parameter no sensible, BETA.
4.4. Effect of Improved Rainfall Estimation in A Calibrated Model
In this section, we analyzed the performance of the runoff simulation obtained from a calibrated
model with far-from-perfect rainfall using new improved rainfall estimations. In this way, the model
Figure 6.
Sensibility of model parameters to rainfall quality (GORE). (
a
) For a parameter sensible,
ALPHA and (b) for a parameter no sensible, BETA.
The only parameter sensible to rainfall quality was ALPHA (Figure 6a). In the study of
Andréassian et al. [
50
], the sensibility is expressed as the reduction of the variability of the parameter
values as the rainfall quality increases until reaching an optimal parameter value. Our study indicated
that the higher the GORE, the higher the ALPHA. On the other hand, Figure 6b shows BETA as
example of a parameter showing no sensibility to rainfall quality. In contrast to the findings of
Andréassian et al. [50]
, we found that the majority of parameters were not sensible to rainfall quality.
This may be caused by the relative high number of parameters which allows an overadjustment to
the input data compared to the three and six parameters used in Andréassian et al. [
50
]. Although,
it is still an open question if the sensibility of parameters to the rainfall quality is a function of the
Water 2018,10, 1169 14 of 19
number of parameters used by the model. Nevertheless, at least one parameter was found sensible,
showing that despite this situation, rainfall quality may still have impact on the parameterization.
This can be due to a correct estimation of precipitation which allows a better description of ALPHA
parameter (which controls the amount of water that recharges the streams) to be adapted to the high
water recharge of these soils.
4.4. Effect of Improved Rainfall Estimation in A Calibrated Model
In this section, we analyzed the performance of the runoff simulation obtained from a calibrated
model with far-from-perfect rainfall using new improved rainfall estimations. In this way, the model
calibrated with input rainfall from scenario E5 was run for the validation period with the remaining
scenarios. Figure 7shows the NSeff index against the rainfall monitoring scenario and the GORE index.
Water 2018, 10, x FOR PEER REVIEW 14 of 19
calibrated with input rainfall from scenario E5 was run for the validation period with the remaining
scenarios. Figure 7 shows the NSeff index against the rainfall monitoring scenario and the GORE
index.
Figure 7. Impact of rainfall quality on model performance (NSeff) of a calibrated model with low
rainfall quality against (a) rainfall monitoring scenario and (b) GORE index.
Rainfall scenarios E4 to E0 produce a considerably better runoff simulation compared to scenario
E5 (Figure 7a). Furthermore, NSeff values obtained from these scenarios were very similar and
increased to 0.49–0.60 from an original value of 0.14. Additionally, in Figure 7b it is observed that
high values of GORE are related to high values of NSeff, although, it was not found a direct relation
between these two indexes.
In this way, better rainfall estimations produced an improvement in model performance. Similar
results were found by Bárdossy and Das [42], who used 10 and 20 rain gauges for calibration and
validation, respectively. Nevertheless, the improvement in the model efficiency was higher in our
case. In this way, any improvement in rainfall monitoring will produce a big gain in modeling
efficiency, even when using a model calibrated with far-from-perfect rainfall. Additionally, the fact
that bad and good efficiencies can be obtained for the same parameters suggests that for this
mountain ecosystem the rainfall input could be the biggest source of model uncertainty.
5. Conclusions
The present study was designed to assess the impact of rainfall estimation on hydrological
modeling using six rainfall monitoring scenarios in a small headwater páramo catchment. This was
achieved by installing a dense network of 12 rain gauges in the Zhurucay Ecohydrological
Observatory in southern Ecuador, and creating rainfall scenarios by withdrawing a given number of
rain gauges.
This study showed that all model structures of HBV-light model can represent total runoff.
However, the capacity to simulate subflow components strongly varied between structures. The
simplest model structure M1 (standard model) had the highest performance to represent subflow
components, although all model structures have problems properly representing slow flow.
The research has also shown that having good spatial rainfall measurements is essential to
achieve good modeling results in mountainous areas. We can conclude that a limited number of rain
gauges can produce acceptable modeling performance. However, it strongly depends on the location
of the rain gauges. In this case, three rain gauges in the upper, middle, and lower catchment worked
well, but this has to be confirmed in other páramo catchments in order to generalize this knowledge.
Furthermore, a better description of the spatial rainfall of the catchment not only enhances the runoff
simulation but also the possibility to select an optimal parameter value.
Another significant finding that emerges from this study is that a calibrated model based on far-
from-perfect rainfall estimates can produce acceptable runoff simulations when was run with new
improved rainfall data, something that can be highly valuable by water managers.
Figure 7.
Impact of rainfall quality on model performance (NSeff) of a calibrated model with low
rainfall quality against (a) rainfall monitoring scenario and (b) GORE index.
Rainfall scenarios E4 to E0 produce a considerably better runoff simulation compared to scenario
E5 (Figure 7a). Furthermore, NSeff values obtained from these scenarios were very similar and
increased to 0.49–0.60 from an original value of 0.14. Additionally, in Figure 7b it is observed that
high values of GORE are related to high values of NSeff, although, it was not found a direct relation
between these two indexes.
In this way, better rainfall estimations produced an improvement in model performance. Similar
results were found by Bárdossy and Das [
42
], who used 10 and 20 rain gauges for calibration and
validation, respectively. Nevertheless, the improvement in the model efficiency was higher in our case.
In this way, any improvement in rainfall monitoring will produce a big gain in modeling efficiency,
even when using a model calibrated with far-from-perfect rainfall. Additionally, the fact that bad and
good efficiencies can be obtained for the same parameters suggests that for this mountain ecosystem
the rainfall input could be the biggest source of model uncertainty.
5. Conclusions
The present study was designed to assess the impact of rainfall estimation on hydrological
modeling using six rainfall monitoring scenarios in a small headwater páramo catchment. This was
achieved by installing a dense network of 12 rain gauges in the Zhurucay Ecohydrological Observatory
in southern Ecuador, and creating rainfall scenarios by withdrawing a given number of rain gauges.
This study showed that all model structures of HBV-light model can represent total runoff.
However, the capacity to simulate subflow components strongly varied between structures.
The simplest model structure M1 (standard model) had the highest performance to represent subflow
components, although all model structures have problems properly representing slow flow.
Water 2018,10, 1169 15 of 19
The research has also shown that having good spatial rainfall measurements is essential to achieve
good modeling results in mountainous areas. We can conclude that a limited number of rain gauges
can produce acceptable modeling performance. However, it strongly depends on the location of
the rain gauges. In this case, three rain gauges in the upper, middle, and lower catchment worked
well, but this has to be confirmed in other páramo catchments in order to generalize this knowledge.
Furthermore, a better description of the spatial rainfall of the catchment not only enhances the runoff
simulation but also the possibility to select an optimal parameter value.
Another significant finding that emerges from this study is that a calibrated model based on
far-from-perfect rainfall estimates can produce acceptable runoff simulations when was run with new
improved rainfall data, something that can be highly valuable by water managers.
Despite that rain gauges are considered the most trusted source of rainfall information, they are
subject to errors. However, differences between rain gauges found here were larger (GORE from
0.41 to 0.98) than those caused by measurement errors. The effect of factors such as wind, elevation,
topographic location, and mechanical errors on rainfall measuring were out of the scope of this study.
Nevertheless, authors encourage the study of such factors mainly due to the lack of information about
rainfall measuring on páramo ecosystem.
Overall, this study has showed that rainfall input could be the largest source of model uncertainty
for this mountain ecosystem. Therefore, our findings have provided a first insight of the importance of
rainfall monitoring for hydrological modeling in páramo catchments, which are the main water supply
for millions of people.
Supplementary Materials:
Supplementary materials can be found at http://www.mdpi.com/2073-4441/10/
9/1169/s1. Table S1: Performance indexes (NSeff, Log NSeff, R
2
and %ADiff) for calibration period. For eight
sub-catchments (S1–S8) and eight model structures (M1–M8), Table S2: Performance indexes (NSeff, Log NSeff,
R2and %ADiff) for validation period. For eight sub-catchments (S1–S8) and eight model structures (M1–M8).
Author Contributions:
A.S. analyzed the data and wrote the manuscript. R.C. designed and supervised the study
and provided a critical revision of the manuscript.
Funding:
This research was part of the project “Identificación de los procesos hidrometereológicos que
desencadenan crecidas extremas en la ciudad de Cuenca” funded by the Dirección de Investigación of Universidad
de Cuenca (DIUC) and the Empresa Pública Municipal de Telecomunicaciones, Agua Potable, Alcantarillado y
Saneamiento de Cuenca (ETAPA-EP).
Acknowledgments:
This manuscript is an outcome of UCuenca’s Master in Ecohydrology. The authors thank
Johanna Orellana and Daniela Ballari for their collaboration in the development of the project. We are grateful
to the staff and students that contributed to the hydrometeorological monitoring. We also thank Zhurucay’s
Chumblin community that allowed the installation of equipment on its lands.
Conflicts of Interest:
The authors declare no conflicts of interest. The founding sponsors had no role in the design
of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, and in the
decision to publish the results.
References
1.
Hofstede, R.; Mena, V.; Segarra, P. Los Paramos del Mundo; Global Peatland Initiative/NC-IUCN/EcoCiencia:
Quito, Ecuador, 2003.
2.
Josse, C.; Cuesta, F.; Navarro, G.; Barrena, V.; Cabrera, E.; Chacón-Moreno, E.; Ferreira, W.; Peralvo, M.;
Saito, J.; Tovar, A. Atlas de los Andes del Norte y Centro. Bolivia, Colombia, Ecuador, Perúy Venezuela;
Secretaría General de la Comunidad Andina, Programa Regional ECOBONA, CONDESAN-Proyecto Páramo
Andino, Programa BioAndes, EcoCiencia, NatureServe, LTA-UNALM, IAvH, ICAE-ULA, CDC-UNALM,
RUMBOL SRL: Lima, Perú, 2009.
3. Pinos, J.; Studholme, A.; Carabajo, A.; Gracia, C. Leaf litterfall and decomposition of polylepis reticulata in
the treeline of the Ecuadorian Andes. Mt. Res. Dev. 2016,37, 87–96. [CrossRef]
4.
Sklenáˇr, P.; Jørgensen, P.M. Distribution patterns of paramo plants in Ecuador. J. Biogeogr.
1999
,26, 681–691.
[CrossRef]
5.
Sarmiento, L.; Llambí, L.D.; Escalona, A.; Marquez, N. Vegetation patterns, regeneration rates and divergence
in an old-field succession of the high tropical Andes. Plant Ecol. 2003,166, 145–156. [CrossRef]
Water 2018,10, 1169 16 of 19
6.
Viviroli, D.; Du, H.H.; Messerli, B.; Meybeck, M.; Weingartner, R. Mountains of the world, water towers for
humanity: Typology, mapping, and global significance. Water Resour. Res. 2007,43, W07447. [CrossRef]
7.
Buytaert, W.; De Bièvre, B. Water for cities: The impact of climate change and demographic growth in the
tropical Andes. Water Resour. Res. 2012,48, W08503. [CrossRef]
8.
Sarmiento, L. Water balance and soil loss under long fallow agriculture in the Venezuelan Andes. Mt. Res. Dev.
2000,20, 246–253. [CrossRef]
9.
De Bièvre, B.; Alvarado, A.; Timbe, L.; Célleri, R.; Feyen, J. Night irrigation reduction for water saving in
medium-sized systems. J. Irrig. Drain. Eng. 2003,129, 108–116. [CrossRef]
10.
Buytaert, W.; Wyseure, G.; De Bièvre, B.; Deckers, J. The effect of land-use changes on the hydrological
behaviour of Histic Andosols in south Ecuador. Hydrol. Process. 2005,19, 3985–3997. [CrossRef]
11.
Crespo, P.; Célleri, R.; Buytaert, W.; Feyen, J.; Iñiguez, V.; Borja, P.; De Bièvre, B. Land use change impacts
on the hydrology of wet Andean páramo ecocystems. In Status and Perspectives of Hydrology in Small Basins;
IAHS Publ. 336; International Association for Hydrological Sciences: London, UK, 2009; pp. 71–76.
12.
Célleri, R.; Feyen, J. The hydrology of tropical Andean ecosystems: Importance, knowledge status,
and perspectives. Mt. Res. Dev. 2009,29, 350–355. [CrossRef]
13.
Carrillo-Rojas, G.; Silva, B.; Córdova, M.; Célleri, R.; Bendix, J. Dynamic mapping of evapotranspiration
using an energy balance-based model over an Andean páramo catchment of southern Ecuador. Remote Sens.
2016,8, 160. [CrossRef]
14.
Córdova, M.; Carrillo-Rojas, G.; Crespo, P.; Wilcox, B.; Célleri, R. Evaluation of the Penman-Monteith
(FAO 56 PM) method for calculating reference evapotranspiration using limited data. Application to the wet
páramo of southern Ecuador. Mt. Res. Dev. 2015,35, 230–239. [CrossRef]
15.
Ochoa-Sánchez, A.; Crespo, P.; Célleri, R. Quantification of rainfall interception in the high Andean tussock
grasslands. Ecohydrology 2018,11. [CrossRef]
16.
Campozano, L.; Célleri, R.; Trachte, K.; Bendix, J.; Samaniego, E. Rainfall and cloud dynamics in the Andes:
A southern Ecuador case study. Adv. Meteorol. 2016. [CrossRef]
17.
Padrón, R.S.; Wilcox, B.P.; Crespo, P.; Célleri, R. Rainfall in the Andean páramo: New insights from
high-resolution monitoring in southern Ecuador. J. Hydrometeorol. 2015,16, 985–996. [CrossRef]
18.
Crespo, P.; Feyen, J.; Buytaert, W.; Bücker, A.; Breuer, L.; Frede, H.G.; Ramírez, M. Identifying controls of the
rainfall-runoff response of small catchments in the tropical Andes (Ecuador). J. Hydrol.
2011
,407, 164–174.
[CrossRef]
19.
Mosquera, G.M.; Lazo, P.X.; Célleri, R.; Wilcox, B.P.; Crespo, P. Runoff from tropical alpine grasslands
increases with areal extent of wetlands. Catena 2015,125, 120–128. [CrossRef]
20.
Correa, A.; Windhorst, D.; Crespo, P.; Célleri, R.; Feyen, J.; Breuer, L. Continuous versus event based
sampling: How many samples are required for deriving general hydrological understanding on Ecuador’s
páramo region? Hydrol. Process. 2016,30, 4059–4073. [CrossRef]
21.
Mosquera, G.M.; Segura, C.; Vaché, K.B.; Windhorst, D.; Breuer, L.; Crespo, P. Insights on the water mean
transit time in a high-elevation tropical ecosystem. Hydrol. Earth Syst. Sci. 2016,20, 2987–3004. [CrossRef]
22.
Mosquera, G.M.; Célleri, R.; Lazo, P.X.; Vaché, K.B.; Perakis, S.S.; Crespo, P. Combined use of isotopic
and hydrometric data to conceptualize ecohydrological processes in a high-elevation tropical ecosystem.
Hydrol. Process. 2016,30, 2930–2947. [CrossRef]
23.
Ochoa-Tocachi, B.F.; Buytaert, W.; De Bièvre, B.; Célleri, R.; Crespo, P.; Villacís, M.; Llerena, C.A.; Acosta, L.;
Villazón, M.; Guallpa, M.; et al. Impacts of land use on the hydrological response of tropical Andean
catchments. Hydrol. Process. 2016,30, 4074–4089. [CrossRef]
24.
Avilés, A.; Célleri, R.; Paredes, J.; Solera, A. Evaluation of Markov chain based drought forecasts in an
Andean regulated river basin using the skill scores RPS and GMSS. Water Resour. Manag.
2015
,29, 1949–1963.
[CrossRef]
25.
Avilés, A.; Célleri, R.; Solera, A.; Paredes, J. Probabilistic forecasting of drought events using Markov chain-
and Bayesian network-based models: A case study of an Andean regulated river basin. Water
2016
,8, 37.
[CrossRef]
26.
Flores-López, F.; Galaitsi, S.E.; Escobar, M.; Purkey, D. Modeling of Andean páramo ecosystems’ hydrological
response to environmental change. Water 2016,8, 94. [CrossRef]
Water 2018,10, 1169 17 of 19
27.
Buytaert, W.; De Bièvre, B.; Wyseure, G.; Deckers, J. The use of the linear reservoir concept to quantify the
impact of changes in land use on the hydrology of catchments in the Andes. Hydrol. Earth Syst. Sci.
2004
,8,
108–114. [CrossRef]
28.
Buytaert, W.; Iñiguez, V.; Célleri, R.; Wyseure, G.; Deckers, J. Analysis of the water balance of small páramo
catchments in south Ecuador. In Environmental Role of Wetlands in Headwaters; Krecek, J., Haigh, M., Eds.;
Springer: Dordrecht, The Netherlands, 2006; pp. 271–281.
29.
Espinosa, J.; Rivera, D. Variations in water resources availability at the Ecuadorian páramo due to land-use
changes. Environ. Earth Sci. 2016,75, 1173. [CrossRef]
30.
Buytaert, W.; Beven, K. Models as multiple working hypotheses: hydrological simulation of tropical alpine
wetlands. Hydrol. Process. 2011,25, 1784–1799. [CrossRef]
31.
Crespo, P.; Feyen, J.; Buytaert, W.; Célleri, R.; Frede, H.-G.; Ramírez, M.; Breuer, L. Development of a
conceptual model of the hydrologic response of tropical Andean micro-catchments in southern Ecuador.
Hydrol. Earth Syst. Sci. Discuss. 2012,9, 2475–2510. [CrossRef]
32.
Célleri, R.; Willems, P.; Buytaert, W.; Feyen, J. Space–time rainfall variability in the Paute basin, Ecuadorian
Andes. Hydrol. Process. 2007,21, 3316–3327. [CrossRef]
33.
Buytaert, W.; Célleri, R.; Willems, P.; De Bièvre, B.; Wyseure, G. Spatial and temporal rainfall variability in
mountainous areas: A case study from the south Ecuadorian Andes. J. Hydrol.
2006
,329, 413–421. [CrossRef]
34.
Arnaud, P.; Bouvier, C.; Cisneros, L.; Dominguez, R. Influence of rainfall spatial variability on flood prediction.
J. Hydrol. 2002,260, 216–230. [CrossRef]
35.
Beven, K.; Homberger, G.M. Assessing the effect of spatial pattern of precipitation in modeling stream flow
hydrographs. Water Resour. Bull. 1982,18, 823–829. [CrossRef]
36.
Chang, C.-L.; Lo, S.-L.; Chen, M.-Y. Uncertainty in watershed response predictions induced by spatial
variability of precipitation. Environ. Monit. Assess. 2007,127, 147–153. [CrossRef] [PubMed]
37.
Chaubey, I.; Haan, C.T.; Salisbury, J.M.; Grunwald, S. Quantifying model output uncertainty due to spatial
variability of rainfall. J. Am. Water Resour. Assoc. 1999,35, 1113–1123. [CrossRef]
38.
Obled, C.; Wendling, J.; Beven, K. The sensitivity of hydrological models to spatial rainfall patterns:
An evaluation using observed data. J. Hydrol. 1994,159, 305–333. [CrossRef]
39.
Rouhier, L.; Le Lay, M.; Garavaglia, F.; Le Moine, N.; Hendrickx, F.; Monteil, C.; Ribstein, P. Impact of
mesoscale spatial variability of climatic inputs and parameters on the hydrological response. J. Hydrol.
2017
,
553, 13–25. [CrossRef]
40.
Younger, P.M.; Freer, J.E.; Beven, K. Detecting the effects of spatial variability of rainfall on hydrological
modelling within an uncertainty analysis framework. Hydrol. Process. 2009,23, 1988–2003. [CrossRef]
41.
Faurés, J.-M.; Goodrich, D.C.; Woolhiser, D.A.; Sorooshian, S. Impact of small-scale spatial rainfall variability
on runoff modeling. J. Hydrol. 1995,173, 309–326. [CrossRef]
42.
Bárdossy, A.; Das, T. Influence of rainfall observation network on model calibration and application.
Hydrol. Earth Syst. Sci. 2008,12, 77–89. [CrossRef]
43.
Xu, H.; Xu, C.-Y.; Chen, H.; Zhang, Z.; Li, L. Assessing the influence of rain gauge density and distribution
on hydrological model performance in a humid region of China. J. Hydrol. 2013,505, 1–12. [CrossRef]
44.
Berne, A.; Delrieu, G.; Creutin, J.-D.; Obled, C. Temporal and spatial resolution of rainfall measurements
required for urban hydrology. J. Hydrol. 2004,299, 166–179. [CrossRef]
45.
Pechlivanidis, I.G.; McIntyre, N.; Wheater, H.S. The significance of spatial variability of rainfall on runoff:
An evaluation based on the Upper Lee catchment, UK. Hydrol. Res. 2016,48, 478–485. [CrossRef]
46.
Syed, K.H.; Goodrich, D.C.; Myers, D.E.; Sorooshian, S. Spatial characteristics of thunderstorm rainfall fields
and their relation to runoff. J. Hydrol. 2003,271, 1–21. [CrossRef]
47.
Singh, V.P. Effect of spatial and temporal variability in rainfall and watershed characteristics on stream flow
hydrograph. Hydrol. Process. 1997,11, 1649–1669. [CrossRef]
48.
Lobligeois, F.; Andréassian, V.; Perrin, C.; Tabary, P.; Loumagne, C. When does higher spatial resolution
rainfall information improve streamflow simulation? An evaluation using 3620 flood events. Hydrol. Earth
Syst. Sci. 2014,18, 575–594. [CrossRef]
49.
Anctil, F.; Lauzon, N.; Andréassian, V.; Oudin, L.; Perrin, C. Improvement of rainfall-runoff forecasts through
mean areal rainfall optimization. J. Hydrol. 2006,328, 717–725. [CrossRef]
50.
Andréassian, V.; Perrin, C.; Michel, C.; Usart-Sanchez, I.; Lavabre, J. Impact of imperfect rainfall knowledge
on the efficiency and parameters of watershed models. J. Hydrol. 2004,250, 206–235. [CrossRef]
Water 2018,10, 1169 18 of 19
51.
Segond, M.-L.; Wheater, H.S.; Onof, C. The significance of spatial rainfall representation for flood runoff
estimation: A numerical evaluation based on the Lee catchment, UK. J. Hydrol.
2007
,347, 116–131. [CrossRef]
52. Arnaud, P.; Lavabre, J.; Fouchier, C.; Diss, S.; Javelle, P. Sensitivity of hydrological models to uncertainty in
rainfall input. Hydrol. Sci. J. 2011,56, 397–410. [CrossRef]
53.
Girons Lopez, M.; Seibert, J. Influence of hydro-meteorological data spatial aggregation on streamflow
modelling. J. Hydrol. 2016,541, 1212–1220. [CrossRef]
54.
Célleri, R.; Buytaert, W.; De Bièvre, B.; Tobón, C.; Crespo, P.; Molina, J.; Feyen, J. Understanding the
hydrology of tropical Andean ecosystems through an Andean Network of Basins. IAHS-AISH
2010
,336,
209–212. [CrossRef]
55.
Aparecido, L.M.T.; Teodoro, G.S.; Mosquera, G.; Brum, M.; Barros, F.d.V.; Pompeu, P.V.; Rodas, M.; Lazo, P.;
Müller, C.S.; Mulligan, M.; et al. Ecohydrological drivers of Neotropical vegetation in montane ecosystems.
Ecohydrology 2018,11. [CrossRef]
56.
Wright, C.; Kagawa-Viviani, A.; Gerlein-Safdi, C.; Mosquera, G.M.; Poca, M.; Tseng, H.; Chun, K.P. Advancing
ecohydrology in the changing tropics: Perspectives from early career scientists. Ecohydrology
2018
,11.
[CrossRef]
57.
Bergström, S. The HBV Model: Its Structure and Applications; SMH1 Reports Hydrology No. 4; Swedish
Meteorological and Bydrological Institute: Norrkoping, Sweden, 1992.
58.
Plesca, I.; Timbe, E.; Exbrayat, J.-F.; Windhorst, D.; Kraft, P.; Crespo, P.; Vaché, K.B.; Frede, H.-G.; Breuer, L.
Model intercomparison to explore catchment functioning: Results from a remote montane tropical rainforest.
Ecol. Model. 2012,239, 3–13. [CrossRef]
59.
Steele-Dunne, S.; Lynch, P.; Mcgrath, R.; Semmler, T.; Wang, S.; Hanafin, J.; Nolan, P. The impacts of climate
change on hydrology in Ireland. J. Hydrol. 2008,356, 28–45. [CrossRef]
60.
Coltorti, M.; Ollier, C.D. Geomorphic and tectonic evolution of the Ecuadorian Andes. Geomorphology
2000
,
32, 1–19. [CrossRef]
61.
Quichimbo, P.; Tenorio, G.; Borja, P.; Cárdenas, I.; Crespo, P.; Célleri, R. Efectos sobre las propiedades físicas
y químicas de los suelos por el cambio de la cobertura vegetal y uso del suelo: Páramo de Quimsacocha al
sur del Ecuador. Suelos Ecuatoriales 2012,42, 138–153.
62.
Ramsay, P.M.; Oxley, E.R.B. The growth form composition of plant communities in the Ecuadorian paramos.
Plant Ecol. 1997,131, 173–192. [CrossRef]
63.
Seibert, J. Estimation of parameter uncertainty in the HBV model. Nord. Hydrol.
1997
,28, 247–262. [CrossRef]
64.
Bergström, S. Development and Application of a Conceptual Runoff Model for Scandinavian Catchments;
SMHI Report RHO 7; Swedish Meteorological and Hydrological Institute: Norrköping, Sweden, 1976.
65.
Seibert, J.; Vis, M.J.P. Teaching hydrological modeling with a user-friendly catchment-runoff-model software
package. Hydrol. Earth Syst. Sci. 2012,16, 3315–3325. [CrossRef]
66.
Uhlenbrook, S.; Seibert, J.; Rodhe, A. Prediction uncertainty of conceptual rainfall-runoff models caused by
problems in identifying model parameters and structure. Hydrol. Sci. J. 1999,44, 779–797. [CrossRef]
67.
Ewen, J.; Parkin, G. Validation of catchment models for predicting land-use and climate change impacts.
1. Method. J. Hydrol. 1996,175, 583–594. [CrossRef]
68.
Klemeš, V. Operational testing of hydrological simulation models. Hydrol. Sci. J.
1986
,31, 13–24. [CrossRef]
69.
Allen, R.G.; Pereira, L.; Raes, D.; Smith, M. Crop Evapotranspiration: Guidelines for Computing Crop Requirements;
FAO Irrigation and Drainage Paper No. 56; Food and Agriculture Organization of the United Nations: Rome,
Italy, 1998.
70.
Nash, J.E.; Sutcliffe, J.V. River flow forecasting through conceptual models part I—A discussion of principles.
J. Hydrol. 1970,10, 282–290. [CrossRef]
71.
Krause, P.; Boyle, D.P.; Bäse, F. Comparison of different efficiency criteria for hydrological model assessment.
Adv. Geosci. 2005,5, 89–97. [CrossRef]
72.
Legates, D.R.; McCabe, G.J., Jr. Evaluating the use of “goodness of fit” measures in hydrologic and
hydroclimatic model validation. Water Resour. Res. 1999,35, 233–241. [CrossRef]
73.
Willems, P. A time series tool to support the multi-criteria performance evaluation of rainfall-runoff models.
Environ. Model. Softw. 2009,24, 311–321. [CrossRef]
74.
Chapman, T.G. Comment on “Evaluation of automated techniques for base flow and recession analyses” by
R. J. Nathan and T. A. McMahon. Water Resour. Res. 1991,27, 1783–1784. [CrossRef]
Water 2018,10, 1169 19 of 19
75.
Shakti, P.C.; Shrestha, N.K.; Gurung, P. Step wise multi-criteria performance evaluation of rainfall-runoff
models using WETSPRO. J. Hydrol. Meteorol. 2010,7, 18–29. [CrossRef]
76.
Rubarenzya, M.H.; Willems, P.; Berlamont, J.; Feyen, J. Application of multi-criteria tool in MIKE SHE model
development and testing. In Proceedings of the 2006 World Environmental and Water Resources Congress,
Omaha, Nebraska, 21–25 May 2006. [CrossRef]
77.
Chawanda, C.J.; Nossent, J.; Bauwens, W. Baseflow Separation Tools; What do they really do? In Proceedings
of the 19th EGU General Assembly, EGU2017, Vienna, Austria, 23–28 April 2017; Volume 19, p. 18768.
78.
Seibert, J. Regionalisation of parameters for a conceptual rainfall-runoff model. Agric. For. Meteorol.
1999
,
98–99, 279–293. [CrossRef]
79.
Gupta, H.V.; Beven, K.; Wagener, T. Model calibration and uncertainty estimation. In Encyclopedia of
Hydrological Sciences; John Wiley & Sons, Ltd.: Chichester, UK, 2006.
80.
Hrachowitz, M.; Weiler, M. Uncertainty of precipitation estimates caused by sparse gauging networks in a
small, mountainous watershed. J. Hydrol. Eng. 2011,16, 460–471. [CrossRef]
81.
Zhang, J.; Han, D. Assessment of rainfall spatial variability and its influence on runoff modelling: A case
study in the Brue catchment, UK. Hydrol. Process. 2017,31, 2972–2981. [CrossRef]
©
2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
... This characterization is more important where atmospheric conditions are complex, as in the Tropical Andes climate Matovelle et al, 2021) where there is still a lack of hydrometeorological monitoring and insufficient knowledge of hydrological processes (Celleri et al, 2007;Padrón et al, 2015;Matovelle et al, 2021). Meteorological forecast, especially rainfall, is essential for agricultural planning, early warning systems, hydroelectric, drinking water plant management, and general monitoring of water resources (Moya-Álvarez et al, 2018;Sucozhañay and Célleri, 2018;Yáñez-Morroni et al, 2018). Due to several factors, extreme weather conditions currently affect industrial and agricultural activities, leading to economic losses as a result of investments to counteract their effects. ...
... The city covers 3665.33 km 2 , with 73.01 km 2 (2%) of the urban zone. Rainfall has a bimodal structure with a dry season between July and August and then a long humid season (Sucozhañay and Célleri, 2018). ...
Preprint
Full-text available
Rainfall nowcasting is a challenging task due to the time-dependencies of the variables and the stochastic behavior of the process. The difficulty increases when the geographic area of interest is characterized by a large spatio-temporal variability of its meteorological variables, causing large variations of rainfall even within a small area in places such as the Tropical Andes. To address this problem, we propose a methodology for building a group of models based on Long Short-Term Memory (LSTM) neural networks using Bayesian optimization. We optimize the model hyperparameters using accumulated experience to reduce the hyperparameter search space over successive iterations. The result is a large reduction in modeling time that allows the building of specialized LSTM models for each zone and nowcasting time. We evaluated the method by nowcasting rain events in the urban area of Cuenca City in Ecuador, a city with large spatio-temporal variability. The results show that our proposed model offers better performance over the trivial forecaster for up to 9 hours of future forecasts with an accuracy of up to 84.4%
... Precipitation is bimodal with unmarked dry seasons-similar to the pattern seen in other páramo ecosystems of the region (Celleri et al., 2007)-and fog and drizzle are present almost daily (Padrón et al., 2015). Average annual precipitation is 1345 mm (Sucozhañay and Célleri, 2018) and average annual drizzle amounts to 340.1 mm (Berrones et al., 2021). Annual actual evapotranspiration averages 634.7 ± 9.0 mm . ...
... The Zhurucay encompasses one of the most complete super-sites for ecohydrological and meteorological monitoring in the Andean region, detailed in Carrillo-Rojas et al. (2019) and Sucozhañay and Célleri (2018). This super-site enables not only conventional measurement of weather, topography, soil taxonomy, and vegetation species within the observatory, but also monitoring of the energy and water fluxes of the land surface and atmosphere . ...
Article
In high-altitude Andean grasslands (páramo), overgrazing leads to alterations in both vegetation and microclimate. These alterations need to be identified to devise land management strategies that will preserve and enhance ecosystem processes. To elucidate this issue, we designed an overgrazing experiment: we selected two plots covered with native grass (pajonal), in one of which we mowed to the ground surface. We left the second plot undisturbed to serve as a control. For both plots, we continuously monitored albedo and ancillary energetic components to generate quarterly and yearly comparisons for the following parameters: (a) impacts on albedo and resilience of grass; (b) radiative forcing of albedo; and (c) land surface temperature feedback during the recovery period. In the first quarter following removal, when the soil was covered with light litter, median albedo increased 38.81% (0.16 ± 0.02), then began a gradual decrease, which continued until its full recovery 1.75 years later (0.10 ± 0.01). During the first year of the experiment, a strong mean negative instantaneous radiative forcing was observed (−7.08 ± 6.03 Wm⁻²), signifying a reduction in net shortwave energy. This forcing returned to normal, pre-intervention conditions (−0.55 ± 0.97 Wm⁻²) after 1.75 years, equal to the energetic recovery period of the grass. Both the amount (from 133.0 ± 44.72 to 119.67 ± 39.30 Wm⁻²) and the partitioning (net shortwave decreased 5%; net longwave increased 9.7%) of net energy were altered after removal, evidence of cooling feedback during the recovery period. This feedback indicated that the decrease in albedo (1.25%) or instantaneous radiative forcing (4.67 Wm⁻²) resulted in a decrease in land surface temperature of 1 °C. Thus, our overgrazing experiment without soil destruction followed by a natural recovery time has identified the energetic recovery period for grass in the páramos; suggesting the albedo as a good indicator of grass resilience.
... In addition, researchers started to evaluate data uncertainties related to the location and technical properties of equipment in the increasingly denser monitoring networks (Guallpa, 2013;Muñoz, Célleri, & Feyen, 2016;Sucozhañay & Célleri, 2018). As a result, the evaluation of land-use and climate change scenarios and their impact on the hydrological system became more feasible and helped decision-makers to predict potential economic benefits for several service providers (Bremer et al., 2019;Flores-López, Galaitsi, Escobar, & Purkey, 2016;Kroeger et al., 2019). ...
... Improvements in the spatial-temporal estimation of precipitation were possible thanks to the identification of the best satellite products, model images (Ballari, Giraldo, Campozano, & Samaniego, 2018;Manz et al., 2017;Nerini et al., 2015;Ulloa, Ballari, Campozano, & Samaniego, 2017;Ulloa, Samaniego, Campozano, & Ballari, 2018), the use of dense and/or extensive rain gauges networks (Manz et al., 2016;Sucozhañay & Célleri, 2018), and more sophisticated equipment such as radars and disdrometers (Orellana-Alvear et al., 2017;Padrón et al., 2015). Precipitation forecasting and projections use statistical and dynamical downscaling applications (Campozano, Tenelanda, Sánchez, Samaniego, & Feyen, 2016;Ochoa, Campozano, Sánchez, Gualán, & Samaniego, 2016) and forecasting of daily precipitation occurrence (Urdiales & Célleri, 2018). ...
Article
Full-text available
Páramos, a neotropical alpine grassland‐peatland biome of the northern Andes and Central America, play an essential role in regional and global cycles of water, carbon, and nutrients. They act as water towers, delivering water and ecosystem services from the high mountains down to the Pacific, Caribbean, and Amazon regions. Páramos are also widely recognized as a biodiversity and climate change hot spots, yet they are threatened by anthropogenic activities and environmental changes. Despite their importance for water security and carbon storage, and their vulnerability to human activities, only three decades ago, páramos were severely understudied. Increasing awareness of the need for hydrological evidence to guide sustainable management of páramos prompted action for generating data and for filling long‐standing knowledge gaps. This has led to a remarkably successful increase in scientific knowledge, induced by a strong interaction between the scientific, policy, and (local) management communities. A combination of well‐established and innovative approaches has been applied to data collection, processing, and analysis. In this review, we provide a short overview of the historical development of research and state of knowledge of the hydrometeorology, flux dynamics, anthropogenic impacts, and the influence of extreme events in páramos. We then present emerging technologies for hydrology and water resources research and management applied to páramos. We discuss how converging science and policy efforts have leveraged traditional and new observational techniques to generate an evidence base that can support the sustainable management of páramos. We conclude that this co‐evolution of science and policy was able to successfully cover different spatial and temporal scales. Lastly, we outline future research directions to showcase how sustainable long‐term data collection can foster the responsible conservation of páramos water towers.
... We selected the simplest structure to have few parameters that could be related to the hydrological processes mimicking the effect of land use change. This structure was also selected because it has been proven to work well for the Zhurucay catchment [22]. ...
Article
Full-text available
To gain insights into the impact of afforestation on the hydrology of a mountain river basin, the Mpinos catchment was modeled to obtain parameters that represent pine land use. Thereafter, the parameters were applied in the Zhurucay catchment to answers the following questions: 1) How can the parameters of hydrological models subject to land use change be calibrated? 2) What is the impact on peaks, total flow, and baseflow when land use gradually changes from tussock grass to pine plantations? 3) Is the impact different when land use changes gradually from upstream to downstream (U-D) or downstream to upstream (D-U)? Based on our results, the total flow, baseflow, and peaks were reduced by 21%, 66%, and 21%, respectively. Overall, this study presents a calibration approach to predict the effects of land-use change prior to its occurrence.
... Spatial rainfall in páramo ecosystems is well represented with one rainfall station within distances of less than 4 km (Buytaert et al., 2006a(Buytaert et al., , 2006b(Buytaert et al., , 2006c. Additionally, the rain gauge, located in the mid-upper part of the catchment, was consistent with the streamflow response (Sucozhañay and Célleri, 2018). ...
Article
Monitoring solute fluxes in water quality studies is essential to reveal potential ecosystem disturbances, and is particularly important in Andean headwater catchments as they are the main sources of water for downstream populations. However, such studies have mainly focused on organic matter and nutrients, disregarding other solutes that can threaten water quality (e.g. arsenic, lead, calcium or magnesium). Additionally, routine low-resolution (weekly or monthly) sampling schemes may overlook important solute dynamics. Therefore, we collected water samples every four hours for the analysis of twenty-four solutes in a pristine tropical Andean páramo catchment. Solute fluxes were calculated using five different methods. The 4-hourly data set was filtered to test for an optimum sampling frequency without compromising export rates. Based on the available 4-hourly data, the results showed that the interpolation export method was best suited, due to a weak correlation with discharges. Of the twenty-four solutes analyzed, Dissolved Organic Carbon (DOC), Total Nitrogen bound (TNb), Si, Ca, Mg, K, and Na presented the highest input rates (with DOC = 4.167E+08 mEq km⁻² yr⁻¹ and Si = 1.729E+07 mEq km⁻² yr⁻¹) and export rates (with DOC = 2.686E+08 mEq km⁻² yr⁻¹ and Si = 2.953E+08 mEq km⁻² yr⁻¹). Moreover, DOC, TNb, NH4-N, NO2-N, NO3-N, PO4, Al, B, Cu, Fe, Zn, As, Cd, Cr, Pb, and V presented more input than export, while Ca, K, Mg, Na, Rb, Si, Sr, and Ba presented more export than input (geogenic sources). Filtered sampling frequencies demonstrated that a minimum of daily grab samples would be required to obtain reliable export rates with differences consistently below 10%, when compared to the 4-hourly solute export. These findings can be particularly useful for the implementation of long-term monitoring programs at low cost, and they provide high-quality information, for the first time, on biogeochemical budgets in a pristine páramo catchment.
... Both models allow for including glacier and non-glacier routines, require few input data, and have been implemented to study a large variety of topics including climate change impacts and water management (e.g. Astorayme et al., 2015;López López et al., 2018;Lujano Laura et al., 2016;Nauditt et al., 2017;Ochoa-Sánchez et al., 2019;Sucozhañay and Célleri, 2018). In addition, the Shaman model has been developed to model a full water balance accounting for, among others, a parsimonious glacier routine with focus on the modeling of tropical glacier melt, a subsurface reservoir, and sectorial water demand including back flows. ...
Article
Full-text available
Study region: Glaciated headwaters of the Vilcanota-Urubamba river basin, Southern Peru Study focus: A pivotal question is if robust hydrological simulation of streamflow in data-scarce and glaciated catchments can be achieved using parsimonious or more complex models. Therefore , a multi-model assessment of three glacio-hydrological models of different complexity was conducted thoroughly analyzing model performance, flow signatures and runoff components. New hydrological insights for the region: In data-scarce catchments, such as in the tropical Andes, parsimonious glacio-hydrological models can provide more robust results than complex models. While the overall performance of all models was reasonably good (R²: 0.65-0.70, Nash-Sutcliffe: 0.65-0.73, Nash-Sutcliffe-ln: 0.73-0.78), with increasing data scarcity more complex models involve higher uncertainties. Furthermore, complex models require substantial understanding of the underpinning hydrological processes and a comprehensive calibration strategy to avoid apparently high model performance driven by inadequate assumptions. Based on these insights we present a framework for robust glacio-hydrological simulation under data scarcity. This stepwise approach includes, among others, a multi-model focus with a comprehensive assessment of flow signatures and runoff components. Future modeling needs to be further supported by alternative data collection strategies to substantially improve knowledge and process understanding. Therefore, the extension of sensor and station networks combined with the integration of co-produced knowledge represents a meaningful measure to robust decision-making for climate change adaptation and water management under high uncertainty.
... The initial model, Hydrologiska Byråns Vattenbalansavdelning (HBV) was primarily developed by the Swedish Meteorological and Hydrological Institute (SMHI) [55,56]. Although the HBV-light model is a relatively recent version, the research community already uses it widely for flood forecasting, and producing valid output results [57][58][59][60][61][62]. The HBV-light version uses a warming-up period so that the initial state values progress to their proper values based on the meteorological data and parameter calibration [54,63]. ...
Article
Full-text available
The hydrological response of a medium-sized watershed with both rural and urban characteristics was investigated through event-based modeling. Different meteorological event conditions were examined, such as events of high precipitation intensity, double hydrological peak, and mainly normal to wet antecedent moisture conditions. Analysis of the hydrometric features of the precipitation events was conducted by comparing the different rainfall time intervals, the total volume of water, and the precedent soil moisture. Parameter model calibration and validation were performed for rainfall events under similar conditions, examined in pairs, in order to verify two hydrological models, the lumped HEC-HMS (Hydrologic Engineering Center’s Hydrologic Modeling System model) and the semi-distributed HBV-light (a recent version of Hydrologiska Byråns Vattenbalansavdelning model), at the exit of six individual gauged sub-basins. Model verification was achieved by using the Nash–Sutcliffe efficiency and volume error index. Different time of concentration (Tc) formulas are better applied to the sub-watersheds with respect to the dominant land uses, classifying the Tc among the most sensitive parameters that influence the time of appearance and the magnitude of the peak modeled flow through the HEC-HMS model. The maximum water content of the soil box (FC) affects most the peak flow via the HBV-light model, whereas the MAXBAS parameter has the greatest effect on the displayed time of peak discharge. The modeling results show that the HBV-light performed better in the events that had less precipitation volume compared to their pairs. The event with the higher total precipitated water produced better results with the HEC-HMS model, whereas the rest of the two high precipitation events performed satisfactorily with both models. April to July is a flood hazard period that will be worsened with the effect of climate change. The suggested calibrated parameters for severe precipitation events can be used for the prediction of future events with similar features. The above results can be used in the water resources management of the basin.
... Local networks have produced specific hydro-climatic knowledge on small catchments in the Ecuadorian Andes (see e.g. Correa et al., 2017;Sucozhañay and Célleri, 2018;González-Zeas et al., 2019). ...
Thesis
This thesis investigates methods to represent the past and future hydro-climatic variability in space and over time in poorly-gauged regions. It proposes a complete and reproducible procedure applied to the continental Ecuador to deal with observed and simulated hydro-climatic data in order to represent past variability and project the potential impact of climate change on water resources by the end of the 21st century. Up-to-date techniques were identified in a literature review and were integrated in a chain protocol to obtain continuous space-time series of air temperature, precipitation and streamflow over past and future multi-decadal periods. Three central chapters are dedicated to this objective according to the following topics: (1) regionalization of air temperature and precipitation from in situ measurements by comparing deterministic and geostatistical techniques including orographic corrections; (2) streamflow reconstruction in various catchments using conceptual hydrological models in a multi-model, multi-parameter approach; and (3) hydro-climate projections using climate model simulations under a high range emission scenario. Climate regionalization revealed the importance of calibrating parameters and of assessing interpolated fields against independent gauges and via hydrological sensitivity analyses. Streamflow reconstruction was possible with the regionalized climate inputs and the combined simulations of three hydrological models evaluated in contrasting climate conditions. Future medium term (2040-2070) and long term (2070-2100) hydro-climatic changes were analysed with confidence intervals of 95% using scenarios from nine climate models and transferring the model parameters calibrated for streamflow reconstruction. Analysis of hydro-climatic variability over the period 1985-2015 showed a slight increase in temperature, while precipitation variability was linked to the main modes of El Niño and La Niña phases at inter-annual scale and to the displacement of the inter-tropical convergence zone (ITCZ) at seasonal scale. Under climate change, a general increase in temperature (+4.4 °C) and precipitation (+17%) is expected by the end of the 21st century, which could lead to between +5% and 71% increase in mean annual streamflow depending on the catchments. These results are discussed in terms of significance for water management before suggesting future hydrological research such as regionalizing streamflow, better quantifying uncertainties and assessing the capacity to meet future water requirements.
... Systematic monitoring of flood events is necessary for the validation of flood risk and hazard models that will help in the decision-making process. Since a high spatial and temporal variability of rainfall is expected within the Andean mountain range, efficient distribution of new types of hydrometeorological networks is required [e.g., Contreras et al. (2019)] to reduce investment and maintenance costs, but without compromising modeling accuracy (Sucozhañay and Célleri, 2018). In recent years, citizen science has shown an increasing potential for hydrological data collection in remotes areas such as mountains by applying simple downloading procedures from ...
Article
Full-text available
Increasing urbanization and development along rivers, together with climate change, exacerbate future flood risk in Ecuador. Current policy strategies in the highlands greatly need improvement. Politicians must rethink if the governmental environmental institutions and inhabitants of the highlands are prepared to respond to future small-and large-scale flood episodes. The purpose of this paper is to identify the issues facing flood risk management (FRM) in the Ecuadorian highlands with a view to finding approaches for overcoming them. We focus on three specific concerns: an assessment of the deficiencies of current flood risk management, the development of diverse strategies to combat flooding, and the need for an overarching vision for future actions and research. These are presented within a theoretical framework together with the authors' recommendations of adaptation options for Ecuador's newly emerging flood challenges. Traditional and novel flood risk management approaches are required, including smart land-use planning, implementation of structural and non-structural measures, coordinated water governance systems with public participation, and the development and improvement of resilience to flooding. Moreover, academic input is a fundamental component of FRM to fill current knowledge gaps regarding mountain floods. This article addresses further research developments that are required to articulate and enable targeted FRM. To our knowledge, no previous publication has specifically dealt with FRM issues in the Ecuadorian highlands, thus, an overview of the current status of FRM is necessary. It is hoped that the challenges associated with flood management discussed here can be addressed over time.
Article
Due to climate change and human disturbances, rainfall spatial and temporal variabilities at the watershed have become increasingly prominent, significantly impacting the patterns of runoff generation and pollutant transportation. Originally, rainfall variabilities in space and time were frequently neglected or idealized in the traditional applications of hydrology and water quality models. This research attempted to investigate the responses of runoff quality and quantity simulation to rainfall spatial and temporal variabilities. We used Rainfall Hazard Modeling System (RainyDay model) combined with TRMM (Tropical Rainfall Measuring Mission) rainfall products to generate rainfall events for 500 return periods. Such events were resampled into eight rainfall scenarios with multiple levels of spatial and temporal variabilities. They were used to drive the Soil and Water Assessment Tool (SWAT) to simulate runoff quality (TP as the index) and quantity under multiple watershed scales and return periods at the Dongjiang River watershed. Analysis of Variance was employed to analyze the relative contributions of rainfall spatial and temporal variabilities to the total variabilities of runoff quality and quantity. The results highlighted that rainfall spatial and temporal variabilities would have notable impacts on runoff quality and quantity under multiple watershed scales and event magnitudes. For headwater at small watershed scales, rainfall spatial variability would have a relatively significant impact on runoff quality and quantity. For relatively larger watershed scales, rainfall temporal variability showed an increasingly important impact than spatial variability for smaller return periods, whereas the opposite results can be obtained for relatively larger return periods. The results can be caused by the following reasons:1) rainfall was concentrated in the middle and lower reaches of the watershed; 2) it was based on the daily scale. Our research suggested that attention should be paid to spatial and temporal variabilities of rainfall inputs in hydrological and water quality models.
Article
Full-text available
Tropical ecosystems offer a unique setting for understanding ecohydrological processes, but to date such investigations have been limited. The purpose of this paper is to highlight the importance of studying these processes—specifically, how they are being affected by the transformative changes taking place in the tropics—and to offer an agenda for future research. At present, the ongoing loss of native ecosystems is largely due to agricultural expansion, but parallel processes of afforestation are also taking place, leading to shifts in ecohydrological fluxes. Similarly, shifts in water availability due to climate change will affect both water and carbon fluxes in tropical ecosystems. A number of methods exist that can help us better understand how changes in land use and climate affect ecohydrological processes; these include stable isotopes, remote sensing, and process-based models. Still, our knowledge of the underlying physical mechanisms, especially those that determine the effects of scale on ecosystem processes, remains incomplete. We assert that development of a knowledge base concerning the effects of transformative change on ecological, hydrological, and biogeochemical processes at different spatio-temporal scales is an urgent need for tropical regions, and should serve as a compass for emerging ecohydrologists. To reach this goal, we advocate a research agenda that expands the number and diversity of ecosystems targeted for ecohydrological investigations and connects researchers across the tropics. We believe that the use of big data and open source software—already an important integrative tool/skill for the young ecohydrologist—will be key in expanding research capabilities.
Article
Full-text available
This study explores rainfall spatial variability and its influence on runoff modelling. A novel assessment scheme integrated with coefficient of variance and Moran's I is introduced to describe effective rainfall spatial variability. Coefficient of variance is widely accepted to identify rainfall variability through rainfall intensity, whereas Moran's I reflects rainfall spatial autocorrelation. This new assessment framework combines these two indicators to assess the spatial variability derived from both rainfall intensity and distribution, which are crucial in determining the time and magnitude of runoff generation. Four model structures embedded in the Variable Infiltration Capacity model are adopted for hydrological modelling in the Brue catchment of England. The models are assigned with 1, 3, 8, and 27 hydrological response units, respectively, and diverse rainfall spatial information for 236 events are extracted from 1995. This study investigates the model performance of different partitioning based on rainfall spatial variability through peak volume (Qp) and time to peak (Tp), along with the rainfall event process. The results show that models associated with dense spatial partitioning are broadly capable of capturing more spatial information with better performance. It is unnecessary to utilize models with high spatial density for simple rainfall events, though they show distinct advantages on complex events. With additional spatial information, Qp experiences a notable improvement over Tp. Moreover, seasonal patterns signified by the assessment scheme imply the feasibility of seasonal models.
Conference Paper
Full-text available
Baseflow separation plays a major role in calculation of runoff coefficients and in the component-wise calibration of hydrologic models. The tools which are used in separating baseflow are based on various assumptions which may not be fully met across various catchments. When used in the cases where the hypotheses behind their operation are not fully met, what will the tools do? This study investigated the performance of selected baseflow separation tools if some of the assumptions on which they were based are not met. Different natures of catchments were represented by models created in a flexible modelling platform, SUPERFLEX. Flow time series for the base flow component, the quick flow component and the total discharge were generated from the models at daily and hourly time step. Three baseflow separation tools, namely; the BFI tool, the SWAT baseflow separation tool and WETSPRO, were used to filter baseflow from the total discharge time series and the filtered baseflow was compared to the original baseflow from the model. The results show that the assumption of exponential recession holds true across a wide range of catchment structures. Thus, WETSPRO provides better estimates of baseflow in a wider range of catchments. It was also shown that the assumptions behind the SWAT baseflow separation tool, and hence performance, heavily depend on the time step of the discharge time series. The BFI was found to mostly overestimate the baseflow during wet periods when used in most catchments although modifying the block size improved its performance.
Article
Full-text available
Leaf litterfall contributes significantly to carbon fluxes in forests. A crucial open question for the sustainability of mountain forests is how climate change will affect this and other carbon fluxes (eg photosynthesis and respiration). Leaf litterfall and decomposition of Polylepis reticulata, an endemic species of the Andes, were analyzed during a period of 1 year at 6 experimental plots located in the Andean páramo between 3700 and 3900 m above sea level in Cajas National Park, Ecuador. Litterfall was collected in each plot using 5 randomly distributed traps. Every trap had a 40-cm diameter (0.125 m2) and was suspended 0.8 to 1.0 m above the ground. The decomposition rate of the leaf litter was analyzed using litter bags. Eighteen bags with approximately 20 g of dry litter were placed in the litter layer in each experimental plot and collected 30, 60, 90, 150, 210, 300, and 365 days after they were installed. The mean annual litterfall recorded was 3.77 Mg ha−1, representing 51% of the leaf biomass present in the canopy, so the leaf life span of P. reticulata in Cajas National Park is 1.98 years. Litterfall occurred all year, with no significant seasonal pattern. The mean decomposition rate (k) obtained for this study period was 0.38 year−1. This study contributes to the information gap on litterfall and decomposition in natural forests located at the highest elevations in the world.
Article
Full-text available
Precipitation is the key factor controlling the high-frequency hydrological response in catchments, and streamflow simulation is thus dependent on the way rainfall is represented in a hydrological model. A characteristic that distinguishes distributed from lumped models is the ability to explicitly represent the spatial variability of precipitation. Although the literature on this topic is abundant, the results are contrasting and sometimes contradictory. This paper investigates the impact of spatial rainfall on runoff generation to better understand the conditions where higher-resolution rainfall information improves streamflow simulations. In this study, we used the rainfall reanalysis developed by Météo-France over the whole country of France at 1 km and 1 h resolution over a 10 yr period. A hydrological model was applied in the lumped mode (a single spatial unit) and in the semidistributed mode using three unit sizes of subcatchments. The model was evaluated against observed streamflow data using split-sample tests on a large set of French catchments (181) representing a variety of sizes and climate conditions. The results were analyzed by catchment classes and types of rainfall events based on the spatial variability of precipitation. The evaluation clearly showed different behaviors. The lumped model performed as well as the semidistributed model in western France, where catchments are under oceanic climate conditions with quite spatially uniform precipitation fields. By contrast, higher resolution in precipitation inputs significantly improved the simulated streamflow dynamics and accuracy in southern France (Cévennes and Mediterranean regions) for catchments in which precipitation fields were identified to be highly variable in space. In all regions, natural variability allows for contradictory examples to be found, showing that analyzing a large number of events over varied catchments is warranted.
Article
Full-text available
Changes in land use and land cover are major drivers of hydrological alteration in the tropical Andes. However, quantifying their impacts is fraught with difficulties because of the extreme diversity in meteorological boundary conditions, which contrasts strongly with the lack of knowledge about local hydrological processes. Although local studies have reduced data scarcity in certain regions, the complexity of the tropical Andes poses a big challenge to regional hydrological prediction.This study analyses data generated from a participatory monitoring network of 25 headwater catchments covering three of the major Andean biomes (páramo, jalca, and puna), and link their hydrological responses to main types of human interventions (cultivation, afforestation and grazing). A paired catchment setup was implemented to evaluate the impacts of change using a “trading space-for-time” approach. Catchments were selected based on regional representativeness and contrasting land use types. Precipitation and discharge have been monitored and analysed at high temporal resolution for a time period between 1 and 5 years.The observed catchment responses clearly reflect the extraordinarily wide spectrum of hydrological processes of the tropical Andes. They range from perennially humid páramos in Ecuador and northern Peru with extremely large specific discharge and baseflows, to highly seasonal, flashy catchments in the drier punas of southern Peru and Bolivia. The impacts of land use are similarly diverse and their magnitudes are a function of catchment properties, original and replacement vegetation, and management type. Cultivation and afforestation consistently affect the entire range of discharges, particularly low flows. The impacts of grazing are more variable, but have the largest effect on the catchment hydrological regulation. Overall, anthropogenic interventions result in increased streamflow variability and significant reductions in catchment regulation capacity and water yield, irrespective of the hydrological properties of the original biome. This article is protected by copyright. All rights reserved.
Article
The páramo ecosystem provides most of the water for the tropical Andean highlands in South America. While the comprehension of this environment has increased lately, there remains an urgent need to quantify the processes involved in the hydrological cycle. Interception loss (IL) is one of the least studied processes in the páramo, and more generally, in grasslands globally. The main objective of this study was to quantify IL at event scale by estimating it indirectly from precipitation (P) and effective rainfall (ER). Furthermore, the following questions were assessed: (1) how much of the P becomes ER, (2) what is the impact on IL calculations of using a raingauge instead of a disdrometer?, (3) which meteorological variables are related to the IL process?, and (4) is it possible to estimate IL from meteorological variables?. High percentages of IL in relation to P were found (10 – 100%). The canopy storage capacity of tussock grasses was approximately 2 mm. The disdrometer observations led to more accurate results than the raingauge observations since only the disdrometer registers light precipitation, horizontal precipitation, and drizzle which increases the amount of P, ER, and IL estimates. Also, we found that IL is more strongly correlated with P; and IL can be estimated with a multiple linear regression (R²=0.9) from P and relative humidity for events where 1.7 < P < 8.5 mm. These findings show the important role of IL in the páramo and provide a stepping stone to modelling of water resources.
Article
Montane ecosystems are known for their high numbers of endemic species, unique climate conditions, and wide variety of ecosystem services such as water supply and carbon storage. Although many ecohydrological and climatic studies of montane environments have been carried out in temperate and boreal regions, few have been done in Neotropical regions. Hence, the objective of this review is to synthesize the existing literature on the main factors (biotic and abiotic) that influence vegetation distribution, functional traits, and ecohydrological processes and feedbacks in tropical montane ecosystems (TME) and to identify key knowledge gaps. Most of the literature used includes work conducted in Neotropical montane rainforests, cloud forests, and grass/scrublands (e.g., páramos, punas, and campos de altitude/rupestres). Fog is a major climatic attribute in tropical montane habitats. We found that fog regimes (frequency and intensity of fog events) influence both water inputs (i.e., canopy interception, foliar water uptake) and outputs (evapotranspiration), and represent an important driver of local species composition, dominance of plant functional types, and ecological functioning. The stability and conservation of TME depends on such ecohydrological fluxes, which are sensitive to increases in air temperature and changing precipitation and fog regimes. Furthermore, to better inform effective conservation and restoration strategies, more work is needed to elucidate how key ecohydrological processes are affected by land use conversion to agriculture and pasture lands, as human activities influence the water budgets in Neotropical montane watersheds not only at regional-scales, but also globally.
Article
The spatial variability of the hydrological response is controlled by the interaction of two spatial variabilities: climatic forcing and physiographic characteristics. This paper investigates their impact on streamflow modelling throughout a large catchment. They are assessed by successively refining the spatial information of climatic inputs and model parameters, going from the lumped to the sub-catchment or mesh scale. The analysis is conducted according to a spatial split-sample test. For each resolution, the distributed model first calibrated on calibration station(s) is then evaluated in light of this(these) station(s) in addition to validation stations previously treated as ungauged. For that purpose, simulated and observed streamflows are compared focusing on four hydrological signatures: daily runoff, daily regime, flood and low flow.
Article
Data availability is important for virtually any purpose in hydrology. While some parts of the world continue to be under-monitored, other areas are experiencing an increased availability of high-resolution data. The use of the highest available resolution has always been preferred and many efforts have been made to maximize the information content of data and thus improve its predictive power and reduce the costs of maintenance of hydrometric sensor networks. In the light of ever-increasing data resolution, however, it is important to assess the added value of using the highest resolution available.