Modeling Earth Systems and Environment

https://doi.org/10.1007/s40808-024-01977-6

ORIGINAL ARTICLE

Evaluation of regionalization parameters for stream flow prediction

in ungauged catchments of Rift Valley Lakes Basin, Ethiopia
Zerihun Makayno Mada1 · Abebe Temesgen Ayalew1 · Ayalkie Belete Amsie1 · Assefa Gedle Azeref1

Received: 31 October 2023 / Accepted: 12 February 2024

© The Author(s), under exclusive licence to Springer Nature Switzerland AG 2024

Abstract
Estimation of streamflow in ungauged catchments is receiving broad research attention for water resources planning and
management. Therefore, regionalization of model parameters and spot streamflow measurements have their importance in
filling such data gaps. In this study, a regionalization model was established for the estimation of streamflow in ungauged
catchments of the Rift Valley Lakes basin in Ethiopia. The Hydrologiska Byråns Vattenbalansavdelning-Integrated Hydro-
logical Modeling System (HBV-IHMS) was calibrated for 14 gauged catchments using bias-corrected satellite rainfall data
as model input. Out of which, 8 catchments were selected to develop a regional model due to a similar cluster with spot flow
measured catchments and was found that the model performance was acceptable. In these catchments, the calibrated model
reproduced the overall pattern and base flow of the observed hydrograph. To establish the regional model, 9 Model Param-
eters and 32 Physical Catchment Characteristics were used in multiple regression. Daily streamflow data were recorded for
three spot measurement rivers sites (Kulfo, Hare, and Hamessa) to serve as validation data for the regionalization model. The
validation result showed that the regional model can be applied for ungauged flow estimation in the basin since the Nash–
Sutcliffe Coefficient (NSE) is greater than 0.50 and the Relative Volume Error (RVE) is within − 10 to 10%. The combined
use of regionalization and spot streamflow measurements is a feasible approach for overcoming data gaps. This study is vital
to the Rift Valley Lakes basin authority, scientific community, and decision-makers to undertake spot streamflow measure-
ment in ungauged catchments for water resources management.

Keywords Regionalization · Spot stream flow measurements · HBV · Ungauged catchments

Introduction spot streamflow measurements can help to fill this growing

hydrological data gap (Du et al. 2020; Georgievová et al.
Water resources planning and management requires ade- 2023; Pool and Seibert 2021; Wu et al. 2022).
quate hydrological flow data for sustainable plane and devel- Any aspect of water resources studies requires reliable,
opment of water resources projects (Tsegaw et al. 2020). accurate, and consistent streamflow data (Viglione et al.
However, in developing countries, streamflow data are not 2013; Yang et al. 2019). However, the stream flow data are
sufficient in several basins, incomplete and too short of time largely inadequate in many African basins due to the impact
series data (Hrachowitz et al. 2013; Wagener and Montanari of human and natural activities (Abimbola et al. 2020). This
2011). Due to lack of understanding of the benefit of long- data gap affects the estimation of flood peaks, low flow esti-
time series data and not sufficient funds to install hydro- mation and causes a great challenge in the planning and
logic equipment (Mishra and Coulibaly 2009). Most rivers management of water resources. Therefore, estimation of
in Ethiopia are ungauged with insufficient data for hydro- streamflow in ungauged catchments is the initial task before
logic research (Gebregiorgis et al. 2013). For this reason, undertaking any hydrological studies and has potential appli-
generating hydrological data through regionalization and cations for estimation of hydrological extremes (Zhang and
Chiew 2009).
* Abebe Temesgen Ayalew During (2003–2012) the International Association of
[email protected] Hydrological Science (IAHS) was given great attention
for stream flow Prediction in Ungauged Basins (PUB)
1
Faculty of Hydraulic and Water Resources Engineering, (Hrachowitz et al. 2013). And also a vital research area for
Arba Minch University, P.O. Box 21, Arba Minch, Ethiopia

Vol.:(0123456789)
 Modeling Earth Systems and Environment

catchments with not enough hydrometric instruments and different hydrological seasons (Seibert and Beven 2009;
good ongoing progress has been achieved in the field of Viglione et al. 2013) and also important to understand catch-
regionalization for hydrological models (Taylor et al. 2013). ments behavior and develop new rating curve (Davids et al.
However, there is still a great challenge and ongoing pro- 2019; Haile et al. 2023). However, some factors that affect
gress in the field of hydrological regionalization in ungauged the installation of staff gauges are finance, technical, and
catchments (Swain and Patra 2017). To address data avail- awareness. These factors are minimized by engaging the
ability issues alternative hydrologic data (i.e. spot stream- local community using simple measurement techniques like
flow measurements) are suggested to supplement traditional tape meter and stick (Davids 2017; Lowry and Fienen 2012).
data sources (Pool et al. 2019; Viviroli and Seibert 2015). Most of the previous studies in Ethiopia conducted
In recent decades, many gauging stations have declined in regionalization with a small number of gauged catchments
developing countries due to this, the data collected from and small physical catchment characteristics. However, the
existing stations has also a problem associated with staff applicability of spot streamflow measurements for model
gauge, rating curve, and observation errors (Abimbola et al. parameterizations, validation with model estimated flow,
2020; Mishra and Coulibaly 2009). So, recently the advance- hydrological homogeneity, and practical datasets to moni-
ment of different regionalization science has overcome the tor gauging stations is not yet evaluated in Ethiopian basins.
challenges of the traditional regionalization approaches. As In addition to this, limitations related to the sample size of
a result, the science announces an extraordinary opportu- gauged catchments and physical catchment characteris-
nity for monitoring gauging stations by developing a new tics (PCCs) were too small and most studies used a simple
rating curve extensively applied and evaluated regionaliza- regression method to establish a regional model.
tion with spot streamflow measured data (Rojas-Serna et al. Looking at the recent streamflow gauging station's prob-
2016; Viviroli and Seibert 2015). lems and streamflow estimations is a challenge. The find-
Hydrological data collection is the basis for water ing of this study is to predicate streamflow and validate a
resources management. Due to this, recently installing staff regional model for ungauged catchment flow estimation
gauges and short runoff measurements have been recognized through observed data by monitoring spot measurement
as valuable options for the calibration of hydrologic models sites in the Rift Valley Lakes basin. The spot stream flow
(Viviroli and Seibert 2015). Participation of the local com- measurements were taken at three rivers Kulfo, Hare, and
munity (i.e., data collectors) makes it feasible to collect spot Hamessa gauging sites to provide data for validating the
flow measurements in a spatially distributed manner over accuracy of established regional. The purpose of this study
large areas (Meerveld et al. 2017). Because stream-level is to contribute to prediction in ungauged catchments by
data recording is easier to involve the community whereas addressing the hydrological data gap.
discharge measurement requires expert knowledge. Hence,
experts can develop the rating curve and convert the water
level data to streamflow data (Seibert and Vis 2016). Materials and methods
Stream flow prediction in ungauged catchments is chal-
lenging in the field of regionalization due to different meth- Description of study area
ods of regionalization generate different results in a single
basin and limited sample size to establish a regional model This study was conducted in catchments of the Rift Val-
are inherent limitations that cannot be ignored (Viglione ley Lakes Basin (RVLB). This basin is a part of Great East
et al. 2013). So, transferring hydrological information from African Rift Valley basin which extends towards south from
gauged to ungauged catchments of the country’s basins has the upper catchments of the Awash Basin to the Kenyan bor-
suffered from data shortage (Rientjes et al. 2011; Wale et al. der to the extreme south of Chew Bahir (Chorowicz 2005).
2009). For these reasons, a practical streamflow estimation The basin lies geographically between 222,000–570,000 N
method is advocated with additional flow measurements in and 468,000–916,000 E and covers an area of 52,985 ­km2,
target sites (Hughes et al. 2014; Nigussie et al. 2020; Tada stretching 489.1 km along the northeast to southwest with
and Beven 2012; Viviroli and Seibert 2015) to validate simu- an average width of 136.6 km. Its elevation ranges between
lated flows with observed discharge data and to trust trans- 453.62 and 4166.74 m above mean sea level. The basin con-
ferred model parameters. tains about 54 (10 lake level and 44 stream flow) gauging
In target catchments, spot streamflow measurements stations and 179 meteorological stations. For this study, 17
are extremely good because of hydrological information stream flow gauging stations and 37 meteorological stations
needed for practical purposes and can be most informative (Fig. 1) were used. This basin is hydrologically bounded by
for parameters estimation when the measurements cover eight terminal lakes.
Modeling Earth Systems and Environment

222000 280000 338000 396000 454000 512000 570000

916000 ±

916000
^_
^_ $
1 ^
_$
1
^_
$
1$
^_
^
_ _^_
1^
^_
860000

860000
^_ $
1
^$
_1 ^_
1$
^$
_ 1 ^_
^_
804000

804000
1^
$_ $
1

$
1
^_ ^_
748000

748000
^_
$
1
$
1^
_
2
%
$
1
1^
^$
_ _
692000

692000
^_
^_ 1^
$_
^_%
%
2
$
1 2
$
1 2
% Spot measurement sites
^_ ^
_
636000

636000
Meteorological stations
^_ $
1 Stream flow gauging stations
River network
580000

580000
Spot measurement catchments
Lakes
RVLB elevation range (m)
High : 4213
524000

524000
Low : 441

0 62.5 125 250 Km

468000

468000
222000 280000 338000 396000 454000 512000 570000

Fig. 1  Study area map showing meteorological and stream flow gauging stations and the three spot flow measurement catchments

Field data collections of corrosion in staff gauge, wear and tear, vandalism, and
lack of awareness in the local community.
Intensive fieldwork was conducted in Kulfo, Hare, and In the field survey, detailed information on channel
Hamessa rivers gauging stations, after visiting several bank and bed stability, presence of erosion and sedimen-
gauging stations in the Rift Valley lakes basin. The pur- tation, changes in zero-gauge effect, transport accessibil-
pose of visiting is to obtain data and information of the ity, site safety to take the record, presence of overflow,
watershed through direct observation and interviewing and effect of staff gauge corrosion were obtained. After
the people living near the sites. In the Kulfo River, three- inspecting a set of river gauging stations, three stations
stream flow gauging sites were available and the two gaug- were selected for spot flow measurements, and installed
ing stations near new and old bridges were well accessible 2 m staff gauge on each river gauging site (Fig. 2). The
to record head and flow velocity. But, data flow for several stations were selected due to the convenience of taking
years is not available. The third station is located near records, free from aquatic habitat and stable channel bank
Chamo Lake is not accessible and has not been function- and bed. In these sites, the daily water level measurements
ing for several years. Those three gauging sites were not and monitoring protocol were conducted from 2019 to
functional from 2014 to 2018. Also, the Hare and Hamessa 2022 re-trained on data collection, and were provided data
rivers have streamflow gauging sites. However, for sev- recording sheets. The observers regularly recorded stream
eral years those stational have become defunct, because flow level data for about 4 years.
 Modeling Earth Systems and Environment

Fig. 2  Prepared staff gauge and

installation time

Stream flow velocity measurements The mean section method was used to estimate discharge
due to its reliability and relatively better estimation because of
The streamflow velocity measurement was taken three times the average velocity of each section multiplied by the average
in each section (­ V0.2d, ­V0.6d, and V
­ 0.8d) from the water sur- depth of the section with section width (Eq. 2) (Genç et al.
face. Because the accuracy of discharge increases with the 2015) and sample one day Kulfo River cross section found
number of points and vertical increases. The mean velocity on (Fig. 3).
of each section was determined by (Eq. 1). ( ) ( )
Yi + Yi+1 Vi + V i+1
V = 0.25 × V0.2d + 0.5 × V0.6d + 0.25 × V0.8d (1) ΔQi = Wi × × (2)
2 2
where, V = mean velocity in the sections and V0.2d, = velocity
where, ΔQi = is segment discharge, Wi = is segment width,
at 20%, 60%, and 80% from the water surface.
Yi + Yi+1 = depth measured in first and consecutive segments,

Fig. 3  Kulfo river cross-section Widith (m) River cross section

on 01/08/2022 used in the mean
method 0.0 2.0 4.0 6.0 8.0 10.0 12.0 14.0
0.0
0.2
Depth (m)

0.4
0.6
0.8
1.0
Modeling Earth Systems and Environment

and Vi + V i+1 = mean velocity of first and consecutive the value of r between 0.6 and 1 it is taken as good relation-
segments. ship between collected data sets.

Stage discharge relationship Collected times series data

Once the stage discharge relationship is established, the The meteorological data recorded at 37 stations of Rift
subsequent procedure consists of measuring the stage and Valley lakes basin catchments were collected from the
reading discharge from the rating curve. In each spot site, National Meteorological Agency (NMA) of Ethiopia.
discharge and head measurements are properly taken in However, the data contains a high amount of missing data
high, medium, and low flow conditions. To develop a rating and has uneven areal coverage. It was hence decided to
curve, it is convenient to derive a mathematical equation use gridded satellite rainfall data after bias correction. The
Q = f (Hw). Because of varying natural conditions during a Climate Hazards Group Infra-Red Precipitation (CHIRP)
series of measurements, the plotted points were not located was chosen for its highest spatial resolution of 0.05° (about
on one smooth curve. Therefore, a regression curve has to 5.3 km). This data was downloaded from http://​chg.​geog.​
be drawn that fits the points as well as possible. ucsb.​edu/​data/​chirps/ in netcdf format. Grad software was
In this study, the computation of the stage-discharge used to extract the data from the global data for a par-
relation is made by the power equation due to its widely ticular grid in the study area. Each catchment in the Rift
used and analytical determination of the rating curve. This Valley lake basin and its grid coverage was identified and
method consists determination of constant parameter ( Ho ) its satellite rainfall data was extracted.
and log-transformed parameters (α and b). Ho It represent- The satellite rainfall estimates are subjected to signif-
ing a gauge height for zero discharge in the stream and it is icant systematic error and this may lead to uncertainty
a hypothetical parameter that cannot be measured in field. It in the hydrologic model, which could result in under or
is affected by sediment deposition on the river bed ( Hs ) and overestimation of model parameters (Smitha et al. 2018;
the scouring from river bed ( Hc ). To determine Ho , three Teshome et al. 2022). Due to which, the power trans-
values of discharge Q1, Q2 and are selected from the curve formation method was selected for drizzle precipitation
which Q22 = Q1 × Q3. Also, the corresponding values of (Goshime et al. 2019). The technique matches the mean
the stages were H1 from Q1, H2 from Q2 and H3 from Q3 and coefficient variation of the CHIRP with observed rain
determined from rating curve though by manual trial and gauge data (Eq. 3).
error. The log- transformed parameters (α and b) determined
Pc = a × Pbo (3)
by least square method and all streamflow rating curve equa-
tion formulas found in Table 1. Here r indicates the extent of where: Pc = bias corrected CHIRP rainfall; Po = uncor-
linear relationship between the head and discharge data sets. rected satellite rainfall; and a and b bias factor. The value of
For perfect relationship, the correlation coefficient r = 1. If bias factor a was determined by such that the mean of the

Table 1  Principal compo- Components Equations

nents of rating curve equations
Rating curve ( )b
Q = a × Hw − H o
( )
Logarithmic transformation log Q = log (a) + b × log Hw − Ho

Y = a + bx
a = log (a)
( )
x = log Hw − Ho
∑ ∑ ∑
N ( XY ) − ( X ) ( Y )
b = ∑ 2 ∑ 2
N ( X ) − ( X)

Gauge height for zero discharge H0 =

H1 × H3 − H22
(H1 + H3 ) − 2 × H2
∑ ∑
Log-transformed parameters a =
Y − b( X )
N
a = antilog (a)
∑ ∑ ∑
Correlation coefficients r = ��
N ( XY ) − ( X )( Y )
∑ ∑ � � ∑ ∑ 2�
2
N ( X2 ) − ( X) × N ( Y 2 ) − ( Y )

Q discharge, Hw water level, Ho gauge height for zero discharge, (α and b) log-transformed parameters, Hs
sediment deposition on river bed, Hc scouring from the river bed, r correlation coefficients
 Modeling Earth Systems and Environment

transformed CHIP = mean of gauge observations, also bias data effect. These physical catchment characteristics influ-
factor b was determined iteratively such that coefficient of ences the runoff generation.
variance of the satellite that matched with that of the rain Catchment area: the entire watershed area water reach-
gauge daily data. ing the river from its contributing area.
Historically observed hydrological data was collected Perimeter: refers to the entire length of the catchment
from Ministry of Water Irrigation and Electricity (MoWIE) boundary. It affects the shape of catchments and large perim-
from Hydrology department. For this study, 18 hydrologi- eter generate large runoff.
cal gauging stations were selected on the Rift Valley Lakes Gravelius’s index (KG): affects the runoff contribution
basin due to the continuity of data and concurrent period of from narrow and wide catchments and related to perimeter
observations. Those selected stations have relatively good and area of the catchments (Eq. 4). It is always greater than
time series data from 1991 to 2006. However, recent stream- one and approaches unity when the basin approaches a cir-
flow data were not provided by MoWIE. For those three cular shape.
spot measurement sites recent stream flow data was gener-
P
ated from the developed rating curve and regional model KG = √ (4)
(Table 2). 2 (𝜋 × A)

where, KG = Gravelius shape index (km), A = watershed area

Selection of homogenous regions ­(km2),P = watershed perimeter (km).
Circularity index: the perimeter square divided into the
Hydrological regionalization is grouping of the catchments catchment area. If the catchment is flat shape, its contribu-
with required uniformity in a sense of flood generating tion is large runoff (Eq. 5).
mechanism, hydro-climatologically similar zone and deline-
ation of homogeneous region is the most important step. CI =
P2
(5)
A
Selection of physical catchment characteristics attributes where, all terms are as defined above.
Hmean: It affects runoff contribution by directly changing
For this study, 32 physical catchment characteristics (PCCs) rainfall to runoff and infiltration.
were identified for gauged and ungauged catchments and Hypsometric integral (HI ): it describes how the eleva-
these were categorized into four major group; geography and tion is distributed across the catchment area. It affects runoff
physiography, geology and soil, LULC and meteorological contribution by time of concentration (Eq. 6).

Table 2  Description of data and tools/models used in this study

Category Items Sources Methods of collection Applications

Secondary
Meteorology Precipitation, temperatures, sunshine, NMA Physical contact Input in to HBV-IHMS model
solar radiations and wind speed
Hydrology Stream flow MoWIE Physical contact
Satellite Precipitation Website http://​chg.​geog.​ucsb.​edu/​data/​chirps/ Input data to HBV-IHMS model
Geospatial DEM Website https://​gdem.​cr.​usgs.​gov/ To determine terrain characteristics of
the catchment
LULC and soil map EMA Physical contact To get different soil type and LULC
classification for regionalization
Primary
Site selection Staff gauge installation Rivers Field observation Selection of site for spot measure-
ments
Stream flow Water level, stream flow velocity Rivers Field observation and measurements To validation of regional model
measurements
Tools and models
Tools GrADs Website http://​sourc​eforge.​net/​proje​cts/​openg​ To process netcdf files
rads
Models HBV-IHMS SMHI Calibrations and validation To simulate stream flow
Modeling Earth Systems and Environment

( )
Hmean − Hmin group has been collected from EMWE GIS department. The
HI = ( ) (6) percent coverage of each soil type in the catchments were
Hmax − Hmin
used as PCCs for the regionalization.
where, Hmax = maximum elevation (m); Hmin = minimum LULC affects runoff generation by increasing and
elevation (m); and Hmean = mean elevation (m). decreasing infiltration. Rift valley lakes basin catchments
Catchment basin shape (Bshape): it describes the shape have different land cover types (shrub lands, intensively cul-
influence on the runoff generation at catchment outlet tivated land, moderately cultivated land, bar land, grass land,
(Eq. 7). urban area, marsh land and forest). The percent coverage
of each LULC in the catchment was used as PCCs for the
Hmax − Hmin regionalization.
Bshape = √ (7)
A Annual average rainfall (AAR): is the most widely
used meteorological catchments characteristics for gauged
where, all terms are as defined above. and ungauged catchments. Rainfall data is frequently avail-
Average slope of the watershed (AVg slope): slope con- able and calculated for 16 years (1991–2006) average of the
trol the water flow velocity and it’s determined from DEM annual rainfall of each catchment was used in this study.
of watershed. Average annual evapotranspiration (APET): has sig-
Longest flow path (LFP): This indirectly an indication of nificant impact on runoff generation and varied from catch-
time for water to reach the outlet or gauging stations. ments to catchments. Also, 16 years (1991–2006) average of
Elongation ratio (ER): It indicates how the shape of the the annual evapotranspiration of each catchment was used.
basin deviates from a circle (Eq. 8). Aridity index (AI): is directly related to mean annual
precipitation and mean annual potential evapotranspiration.
LL
ER = √ It indicates how much the atmospheric moisture demand is
(8)
4A
𝜋
satisfied by precipitation.

where, LL = Longest basin length (km), and A = area in ­km2. Catchments clustering
Drainage density (DD): high amount drainage density
represents fast amount of hydrologic response to rainfall, The political administrative boundaries of watershed is not
while, low drainage density shows slow hydrologic response guarantee about hydrological similarity. Due to this stream-
(Eq. 9). flow estimation and flood statistics have been used to region-
∑ alize watershed characteristics used to check hydrological
L
DD = (9) similarity (Isik and Singh 2008; Kebebew and Awass 2022).
A Catchments in one group cluster are similar behavior while
∑ the catchments in different clusters are different behavior.
where, L = summation of all stream length (km), and
A = area in ­km2. For different climatological and topographically Rift Valley
Bifurcation ratio (Rb): is the ratio of number of stream Lakes basin catchments, hybrid (hierarchical and k-mean)
of given order to the number of streams to the next higher cluster analysis is used to cluster dataset. In this study, catch-
order (Eq. 10). ments are preliminary grouped using hierarchical method
then using number of clusters from agglomeration schedule
Ni then apply to k-means finally using SPSS software.
Rb = (10)
Ni+1 Wards method of hierarchical clustering was in study to
get optimum number of cluster. The distinctions made at
where, Ni is the number channels of order i and Ni+1 is the each successive stage of study area represented by a two
number channels of order i + 1. dimensional diagram known as a dendrogram or tree dia-
Soil type directly influences runoff and infiltration rate gram. For this study, K-mean clustering was implemented to
through controlling the degree of permeability. In this study, classify catchments in homogeneous clusters after number
nine major soil group (Leptosols: LEPS, Nitisols: NITS, of cluster is fixed by hierarchical clustering. Because this
Vertisols: VERS, Xerosols: CLXRS, Acrisols: ACRS, Cam- method is most commonly used clustering algorithms and
bisols: CMBS, Fluvisols: FLVS, Solonchaks: SLNKS, and is widely used due to its simplicity, ease of execution, effec-
Regisols: RGS) of rift valley lake basin as per FAO soil tiveness and empirical realization.
 Modeling Earth Systems and Environment

HBV‑IHMS hydrological model selection and the remaining 1/3 is specified for validation. In this
and its application study, the stream flow data from 1991 to 2006 was divided
in to three periods. The first year (1991) was specified as
The conceptual models are good compromise in scarcely warm-up the model, 1992–2000 was specifies as calibration
gauged basin due to reducing model complexity (Beven period and 2001–2006 was used for model validation. The
2001). These models are usually able to capture the domi- model was calibrated for fourteen catchments. A manual
nating hydrological processes at the appropriate scale and model calibration was applied to obtain the optimum cali-
most time used in regionalization (Pool et al. 2017; Seib- brated parameter values by changing one parameter value at
ert 1999). Here the selected model for this study is HBV a time within the allowable ranges.
model, because it is continuous, and conceptual approach
in which the hydrological process is simplified to algebraic Sensitivity analysis
and mathematical function.
The main reason for selecting HBV model is limited The identification of sensitivity model parameters is usu-
data in Rift Valley Lake basin. Also, HBV model has been ally important to know the most influential parameters and
applied in Ethiopia with success in several studies. It is the extent to which the model realistically represents the
successfully applied in regionalization for hydrological real world (Tibangayuka et al. 2022). In this study, first
balance and streamflow prediction in ungauged catch- sensitive model parameters were identified based on the
ments in Upper Blue Nile sub basin (Rientjes et al. 2011; model objective function and sensitive model parameters
Wale et al. 2009). It also received applications for esti- were identified in calibration event by changing the value
mation of streamflow in ungauged catchments in Dedessa of model parameters plus or minus 20%, 40%, 60%, 80%
(Tamalew and Kemal 2016) in upper Tekeze sub basin. and 100% within the allowable range. The most sensitive
Mada and Nannawo (2023) also applied HBV-IHMS for model parameters are highly controlling the stream flow
regionalization of model parameters for hydrological pro- calibration and showed steep slope (i.e. great variation
cess ungauged Shafe catchment in Rift Valley Lakes basin. between intervals in RVE and NSE) while, low sensitive
Teshome et al. (2022) also used HBV-IHMS model for model parameters are less controlling stream flow simu-
prediction of stream flows in ungauged Bilate catchment lation and moderate to gentle slopes (i.e. low variation
for assess the potential of renewable energy in Rift Val- between intervals in RVE and NSE).
ley Lakes basin. Therefore, those studies showed that the
HBV-IHMS model is capable to simulate streamflow in
gauged and ungauged catchments. Model performance evaluations
HBV- IHMS model comprises four routines precipita-
tion subroutines, soil moisture accounting, runoff genera- In this study, HBV-IHMS hydrologic model performances
tion or response and routing routine or transform excess records Relative Volume Error (RVE) and Nash-Sutcliffe
water from the soil moisture zone to local runoff. In the Efficiency (NSE) in addition to visual examination of the
rift valley lakes basin catchments, precipitation was simu- observed and simulated hydrograph. Also, graphical and
lated only in the form of rainfall. In this study, the input statistical comparisons were conducted to evaluate the per-
of HBV- IHMS model used is daily bias corrected precipi- formance of regional model simulated flow with observed
tation, daily mean temperature, daily potential evapora- discharge data (spot flow measured and historical archived
tion, geographic zone (forest and field) and stream flow data).
for calibration. Nash–Sutcliffe efficiency (NSE); the simulated flows
were quantitatively evaluated against observed discharge.
It indicates how well the plot of observed versus simulated
HBV‑ IHMS model calibration and validation value. If the measured value is the same as all predictions,
NSE is 1. If the NSE between 0 and 1, it indicates deviations
Model calibration was conducted by selected model param- between measured and predicted values. If NSE is negative,
eters within their given respective ranges. The selected predictions are very poor, and the average value of output is
parameters values are accepted, if they lead to reasonable a better estimate than the model prediction. (Eq. 11).
much between various features of the simulated and observed ∑n � �2
hydrograph. Otherwise, the values are changed manually for i=1
Qobs(i) − Qsim(i)
∑n � �2
NSE = 1 − (11)
one parameter at a time until acceptable match is attained.
Q − Q
According to Arnold et al. (1998) from the total N years, i=1 obs(i) obs(i)

the calibration period can be conducted for 2/3 of the period

Modeling Earth Systems and Environment

where, NSE = Nash–Sutcliffe coefficient, Qobs(i) = observed Multiple linear regression method: is used in this study
discharge, Qsim(i) = simulated discharge, Qobs(i) = is the because runoff affected several catchment characteristics and
observed average discharge values i = the time step and the independent variables (PCCs) which are used to deter-
n = total number of time steps used during calibration. mine dependent variables (MPs). Independent variables are
Relative Volume Error (RVE): measures the average needs to be determined from different characteristics using
volume of difference between the simulated and the observed Arc GIS and dependent variables are those having fixed
streamflow. It is varied between − ∞ and + ∞ but performs value determined from MPs. In this study, both forward
is best when a value of 0 is generated since the accumulated selection and backward elimination were used in multiple
difference between simulated Q ­ sim(i) and observed Q
­ obs(i) dis- regression analysis to select the best independent variables
charges. Value between + 5 and − 5% indicates that a model using excel data analysis due to most widely used in sev-
performs very well while value between ± 10 and ± 10% eral hydrological studies. The general regression equations
indicates that a model has reasonably good performance described as follows (Eq. 13).
(Eq. 12). Y = 𝛽0 + 𝛽1 × X1 + 𝛽2 × X2 + ...... 𝛽n × Xn (13)
� ∑n ∑n �
i=1 Qsim(i) − i=1 Qobs(i) where,𝛽1, 𝛽2, 𝛽n are regression coefficient, X1, X2, Xn is inde-
RVE = ∑n × 100 (12)
i=1 Qobs(i)
pendent variable (PCCs), Y is dependent variable (MPs) and
𝛽0 is intercepting of regression line.
where, RVE = Relative volume error, Qobs(i) = observed dis-
charge, Qsim(i) = simulated discharge.
Results and discussion
Regionalization approaches In this section, hydrological similar regions were identified
using cluster analysis and results of model calibration and
The regional model method is extensively used and con-
validation presented for 12 gauged catchments. However,
siders catchments characteristics attributes of gauged
8 gauged catchments are used to develop regional model
and ungauged catchments (Merz 2004). The main task in
due to similar cluster with spot measurements conducted
regional model methods is the preparation of appropriate
catchments. Regionalization equations are also presented by
catchments characteristics for the regression technique and
relating the model parameters and PCCs. The equations are
possible interactions between hydrologic model parameters
validated by testing their applicability to three gauged catch-
(Oudin et al. 2008). This approach is transferring model
ments whose data were not used for developing the multi-
parameters from gauged to ungauged catchments and most
ple regression equations. Finally, the use of spot streamflow
widely used and give best results (Halefom et al. 2017;
measurements for validation of regional developed model
Oudin et al. 2008; Zhang and Chiew 2009). In contrary, it
for regionalization.
takes a lot of time and energy to achieve reasonable hydro-
logical simulation due to the effort involved in parameter
calibration. Rating curve developed using spot measured data
The main thing in regionalization is to transfer the
hydrological information from gauged to ungauged catch- The stage-discharge relationship or rating curves for three
ments (Bloschl and Sivapalan 1995; Farfán and Cea 2023; spot measurement stations (Kulfo, Hare, and Hamessa) were
Hrachowitz et al. 2013). In this study, regression- based developed. The Hare gauging station has a 2 m staff gauge
regionalization is selected due to widely used and give bet- installed and used to measure water level. The site is free
ter results (Rientjes et al. 2011; Swain and Patra 2017). In from sediment deposition and scouring of riverbeds. For this
regression based regionalization, the highest correlation reason, only the hypothetical parameter ­(Ho) is considered
between PCCs and MPs were entered first and its statistical as zero-gauge height.
significance was examined. The evaluation criteria for the In the Kulfo gauging site, there is extreme scouring and
PCCs collinearity was tested using statistical significance vandalism of the staff gauge. Thus, a new staff gauge was
P-values of (≤ 0.1 for 95% confidence interval) and coef- installed close to the previous site. At the Hamessa gauging
ficient of determination ­(R2 ≥ 0.80). Statistically significant station, two meters of staff gauges were used to measure the
relationship was established between PCCs and calibrated water level. These staff gauges measure river water levels
MPs by using excel in data analysis. For this study, multiple of up to 2 m height to provide data for low, medium, and
type of regression methods was used to develop regional high flood conditions. At this site, the riverbed is affected by
model. gradual sediment deposition of 0.10 m depth above the zero
 Modeling Earth Systems and Environment

Table 3  Rating curve equations for the spot measurement site

Catchments Log parameters Correlation (r) Zero-gauge (H0 + Hs ) a = anti log (a) Rating curve quations
b α

Kulfo 2.08 1.35 0.99 0.11 22.28 ( )2.08

Q = 22.28 × Hw − 0.11
Hare 1.21 0.92 0.99 0.19 8.32 ( )1.21
Q = 8.32 × Hw − 0.19
Hamessa 0.87 1.11 0.99 0.10 + 0.12 = 0.22 12.96 ( )0.87
Q = 12.96 × Hw − 0.22

reading of the gauge. Thus, the zero-gauge height is larger velocity measurements were taken side bridge method for
than 0.10 m for this station. The total zero-gauge height at water depths > 1.2 m. High ­R2 values are achieved for the
this site equals to the summation of hypothetical parameters developed rating curves, with better fit for low flows than
­(Ho) and silt deposition (­ Hs) detail shown below in Table 3. high flows (Fig. 4). The discharge measurements in which
During the observation period, the maximum water lev- water level greater than 1.2 m was far apart as compared
els were 1.70 m, 1.72 m and 1.68 m for Kulfo, Hare and to water level less than 1.2 m because the measurements
Hamessa stations respectively. Figure 4 showed the devel- require high skilled persons, equipment, cost, time and
oped rating curve using different (low, medium and high) related to risk. This study developed new rating curve for
flow seasons. In Kulfo river 4-year spot water level measure- three spot measurement sites found in Fig. 4.
ments, 28-time simultaneous stage and discharge measure- In point flow measurement from 2019 to 2022, for Hare
ments were taken for all flow ranges. Current meter was used river 26 times simultaneously water level and discharge
for measuring velocity for water depths < 1.2 m. However, measurements were conducted and for Hamessa river 24
times simultaneously head and discharge measurements
were conducted to develop rating curve.
2.5
(a)
2.0
Clustering analysis
Head (m)

1.5

1.0 This study used hierarchical clustering analysis used to

0.5 get optimum number of cluster. After obtaining optimum
0.0 number of cluster, the k-means clustering have been used
0 10 20 30 40 50 60 70
to identify homogenous regions after several iterations.
Discharge (m3/s)
Using the proximity matrix squared Euclidean distance
2.0
(b)
was calculated between all pairs of attributes. The screen
1.6
plot of distances (coefficients column) against stage using
Head (m)

1.2
agglomeration schedule is used to decide number of clus-
0.8
ters as shown in(Fig. 5.
0.4 According to Fig. 5 elbows or changes on screen plot
0.0 occur in number of clusters becomes 2, which is indirectly
0 2 4 6 8 10 12 14 16
Discharge (m3/s) showed hydrological homogenous region. From the 18
2.1 catchments in the study area, 5 catchments are one clus-
1.8 (c) ter and the rest 13 catchments are anther cluster which
1.5
including spot measurements catchments (Kulfo, Hare and
1.2
Hamessa) (Table 4 and Fig. 7).
Figure 6 revealed the dendogram produced by SPSS
Head (m)

0.9
with an added crosses horizontal lines specifying number
0.6
of clusters and the numbers on the y-axis indicating the
0.3
lists of catchment numbers prior to initial cluster member-
ship. And also presents the dendrogram of Ward’s method,
0.0
0 3 6 9 12 15 18 21 24 27
Discharge (m3/s showing the cluster distances for each number of clusters
identified by the branch intersection. A k-means cluster
Fig. 4  Rating curve developed by the current study for Kulfo river analysis was conducted classify regions. This cluster-
(a), Hare river (b) and Hamessa river (c) ing was performed using a 2-cluster solution, which was
Modeling Earth Systems and Environment

Fig. 5  Scree plot of coefficients 75.0

(distance) by stage (number of

Behavioral distances or
cluster)
60.0

Coefficients
45.0

30.0

15.0

0.0
0 2 4 6 8 10 12 14 16 18
Number of Cluster or Stage

Table 4  Cluster membership and Khq) were moderate sensitive and base flow parameters
Case Catch- Cluster Case Catch- Cluster
(Alfa, PERC and CFLUX) were least sensitive model param-
number ments number ments eters in rift valley lakes basin catchments.
The result indicated that the parameters controlling the
1 Bedessa 1 10 Bilate Al 2 water balance (BETA, FC, and LP) were found to be the
2 Dedeba 1 11 U Gelana 1 most sensitive parameters, while routing parameters (K4 and
3 Kolla 1 12 Katar Ab 2
Khq) were relatively less sensitive. Parameters Alfa, PERC,
4 Batena 1 13 Weira 1
and CFLUX were the least sensitive model parameters.
5 Hare 1 14 Hamessa 1
Goshime et al. (2019) reported similar results at Ziway and
6 Gidabo A 1 15 T wuha 1
Meki catchments upper rift valley lakes basin in Ethiopia
7 Chifu Ar 1 16 Kulfo 1
and Worqlul et al. (2018) stated the same results in Gilgel
8 Ferfuro 1 17 Meki 2
Abbay and Gumara catchments in Ethiopia. Also, in this
9 Katar Fete 2 18 Duguda 2
study sensitivity of the model parameters were tested by
visual inspection of the plot of NSE and RVE against values
of model parameters.
obtained from hierarchical clustering. ANOVA, which is Steep and bolded line plot in Fig. 8 indicates that the
used as a vital variable to view clusters, provides k-means model simulations are sensitive to changes in the respective
clustering using SPSS. parameter value at least in terms of NSE and RVE. Figure 8
The study area is classified into two homogeneous groups showed that, the NSE and RVE values for rift valley lakes
of region. Region one includes spot measurements catch- basin catchments Bilate at Alaba and Dedeba nearly con-
ments (Kulfo, Hare, Hamessa, Gelan, Gidabo, many Abay stant for broad range of Perc, Alfa, K4 and Cflux values.
Chamo lakes sub basin catchments and upper Rift Valley This inducts that, these four parameters do not controlled
Lakes sub basin catchments. Region two includes (Bilate the simulated hydrograph pattern study areas. However,
Meki, katar, Duguda and Katar Fete) catchments. The rest the changes in the value of Beta, FC, LP and Khq showed
catchments are having most of ungauged areas in down- abrupt changes in NSE indicating that these model param-
stream parts near to Keny border (Fig. 7). eters should be used to match the pattern of observed and
simulated hydrograph. Variation in the values of FC, Beta
HBV model sensitive analysis and evaluations and LP also resulted in substantial changes in RVE indicat-
ing that, the simulated streamflow volume is controlled by
In each catchment, the HBV-IHMS model parameters were these three parameters. Due to these, those parameters (FC,
independently calibrated for bias corrected rainfall inputs Beta and LP) are the most sensitive parameters in rift valley
by comparing each simulated stream flow time series with lakes basin catchments.
observed runoff data. Initially, the sensitive model param-
eters were identified through changing the value of each Calibration and validation results
model parameter was in allowable range while, the value
of another model parameters are fixed. The study addressed This study used manual model calibration through chang-
that, the water balance/volume parameters (Beta, FC and ing the value of each parameter within allowable range to
LP) were the most sensitive parale routing parameters (K4 get optimized model parameters. Figure 9 revealed the daily
 Modeling Earth Systems and Environment

Fig. 6  Dendrogram using Ward

method

observed and simulated hydrograph of Bedessa and Upper

Gelana catchments in rift valley lakes basin. For the calibra-
tion period, the recession limb and base flow of the hydro-
graph are well captured by the model.
Table 5 presents the calibrated model parameters and the
values of the objective functions for all catchments used
in this study. In the calibration period, the NSE exceeded
0.6 for all catchments indicating very good performance of
the model in capturing the observed hydrograph pattern.
For most catchments, the values of RVE are within − 5 and
5% indicating the model well reproduced observed stream-
flow volumes. For six catchments, RVE is within − 10 and
− 5% and 5 and 10% indicating good performance. How-
ever, the volume error is found excessive high for for Weira
whereas, the NSE is extremely low for Tikur Wuha catch-
ment (Table 5). The total volume of simulated discharge is
underestimated in many catchments with RVE is negative
%. It is mainly due to the deficiency optimum parameters set
tend to underestimates both peak and base flows.
In terms of objective function (NSE), the model perfor-
mance improved in the validation period for Bilate and Katar
Fete catchments and deteriorated for Kolla, Gidabo and
Bedessa catchments as compared to the calibration period.
However, the model performance is still acceptable even
outside the calibration period except for Weira catchment.
Fig. 7  Delineated homogenous regions
Modeling Earth Systems and Environment

(a) 1.0 (a) 200

0.8 150

RVE (%)
100
NSE (-)
0.6
50
0.4
0
0.2 -100% -50% -50 0% 50% 100%
0.0
-100
-100% -50% 0% 50% 100%
Alfa Beta Cflux FC Alfa Beta Cflux FC
K4 Khq LP Perc K4 Khq LP Perc

(b) 1.0 (b) 200

0.8 150
NSE (-)

RVE (%)
0.6 100
0.4 50
0.2 0
0.0 -100% -50% -50 0% 50% 100%
-100% -50% 0% 50% 100% -100
Alfa Beta Cflux FC Alfa Beta Cflux FC
K4 Khq LP Perc

Fig. 8  Sensitivity of the HBV response to its parameters in terms of the NSE and RVE objective functions for Bilate Alaba (a) and Dedeba (b)
catchments

Initialization Validation
Calibration Q obs Q sim RF
75 0
Streamflow (m3/s)

60 50

Rainfall (mm)
45 100
Bedessa
30 150

15 200

Initialization Calibration Q obs Q sim RF Validation

120 0

100
Rainfall (mm)
Streamflow (m3/s)

60
80

60 U Gelana 120

40
180
20

0 240

Fig. 9  Observed and simulated hydrograph for some catchments

 Modeling Earth Systems and Environment

Table 5  Model performance and calibrated values of model parameters

No. Catchments Model performances Remarks
Calibration Validation
NSE RVE NSE RVE

1 Bedessa 0.66 − 8.61 0.57 − 17.36

2 Dedeba 0.75 0.38 0.72 3.16
3 Kolla 0.66 3.76 0.62 − 28.25
4 Batena 0.72 1.52 0.68 − 10.47
5 Gidabo A 0.65 9.63 0.57 5.58
6 Chifu Ar 0.68 7.23 0.67 − 5.60
7 Ferfuro 0.62 − 5.42 0.57 − 5.57
8 U Gelana 0.66 1.20 0.56 7.96
9 Bilate at Alaba 0.79 5.98 0.82 − 0.25
10 Katar Fete 0.72 − 9.34 0.72 7.65
11 Katar Abura 0.80 − 1.28 0.74 3.04 Goshime et al. (2019
12 Meki 0.71 − 1.47 0.64 3.84 Goshime et al. (2019
13 Weira 0.64 − 26.41 0.56 − 26.27
14 T Wuha − 0.14 0.24 − −
No. Catchments HBV-IHMS model parameters
Alfa Beta Cflux FC K4 Khq LP Perc Hq

1 Bedessa 1.5 2.15 0.5 1132 0.001 0.036 0.72 0.32 5.38
2 Dedeba 0.29 2.63 0.14 1285 0.001 0.052 0.68 0.11 1.56
3 Kolla 0.01 1.35 0.001 875 0.043 0.11 0.44 1.51 3.15
4 Batena 0.47 1.13 0.34 468 0.002 0.056 0.67 0.18 3.31
5 Gidabo A 0.01 1.13 0.001 1285 0.079 0.066 0.39 2.82 1.45
6 Chifu Ar 0.11 1.79 0.01 1500 0.001 0.051 0.31 0.92 1.86
7 Ferfuro 0.16 1.97 0.09 876 0.064 0.100 0.63 1.43 4.44
8 U Gelana 0.22 1.00 0.11 1500 0.001 0.066 0.58 0.85 3.87
9 Bilate at Alaba 0.19 1.76 0.031 1295 0.001 0.043 0.61 0.09 0.87
10 Katar Fete 0.37 1.37 0.3 485 0.001 0.063 0.44 0.78 1.11
11 Katar Abura 1.10 3.05 0.005 820 0.100 0.120 0.70 2.75 0.59
12 Meki Meki 0.80 1.96 0.01 860 0.100 0.10 0.50 1.15 0.87
13 Weira 0.65 1.00 0.001 100 0.05 0.130 0.99 0.001 2.43
14 T Wuha 0.01 3.71 0.001 1500 0.002 0.002 0.56 2.350 1.12

The model performance did not show any dependence on model parameters are further entered in to regional model
catchments size. and validated with observed stream flow data from three spot
The optimized values of the routing parameters (K4 and measurements sites (Kulfo, Hare, Hamessa).
Khq and Cflux) varied in a narrow range of values for all
catchments (Table 5). However, the values of parameters that Regionalization model results
control base flow (Perc), water balance (FC, Beta, LP) and
peak flow parameter (Alfa) showed wide range of variation Based on the results of calibration (Table 5), eight catch-
across the catchments. For instance, FC varied from 468 mm ments were selected for regionalization. This selection was
for Batena and 1500 mm for Chiufa and Upper Gelana catch- based on the clustering and model performance as evaluated
ments. This suggests that catchment characteristics can have using the objective function of RVE between + 10 up to
significant effect of the calibrated value of these parameters. − 10% and the NSE value greater ≥ 0.60. The catchments
Those eight catchments (Bedessa, Dedeba, Kolla, Batena, which are eliminated from further analysis are Tikur Wuha
Gidabo, Chifu, Ferfuro and Upper Gelana) which are found and Weira based on objective functions and the catchments
in similar cluster are well performed in NSE and RVE in which are dis similar group from spot measurements catch-
calibration and validation periods. These eight catchments ments are Bilate Alaba, Katar Abura, Dugda, Meki, and
Modeling Earth Systems and Environment

Table 6  Values of Pearson MPs

correlation coefficient between PCCs Alfa Beta Cflux Fc K4 Khq Lp Perc Hq
MPs and PCCs for gauged Area -0.45 -0.38 -0.53 0.36 0.68 0.08 -0.56 0.86 -0.55
catchments
Per -0.63 0.06 -0.66 0.36 0.48 0.14 -0.38 0.54 -0.82
KG -0.30 0.53 -0.19 -0.07 -0.16 0.03 0.22 -0.33 -0.54
CI -0.22 0.58 -0.13 -0.01 -0.20 -0.04 0.26 -0.37 -0.51
H mean -0.47 -0.07 -0.40 0.15 -0.35 -0.16 -0.57 -0.15 -0.66
HI -0.14 0.33 -0.07 -0.61 0.13 0.38 -0.06 -0.12 -0.10
B shape 0.60 0.30 0.41 0.25 -0.60 -0.50 -0.08 -0.44 0.24
Avg slop 0.57 -0.25 0.44 0.03 -0.10 -0.05 0.23 0.00 0.61
LFP -0.51 0.35 -0.55 0.39 0.13 -0.05 -0.34 0.18 -0.89
ER -0.14 0.63 -0.08 0.02 -0.38 -0.19 0.13 -0.49 -0.54
DD 0.98 0.45 0.86 -0.01 -0.40 -0.53 0.63 -0.49 0.68
Rb 0.56 0.62 0.36 0.23 0.12 -0.26 0.31 -0.02 0.44
LEP 0.08 -0.33 0.41 -0.78 -0.20 -0.10 0.34 -0.35 0.11
NIT -0.06 0.02 -0.04 0.49 -0.39 -0.14 0.32 -0.31 0.06
VER -0.32 0.24 -0.47 0.19 0.34 0.24 -0.56 0.33 -0.17
FLV -0.36 0.04 -0.46 0.37 0.46 -0.11 -0.47 0.59 -0.79
ACR -0.10 -0.45 -0.09 0.44 -0.28 -0.02 0.07 -0.07 0.21
RG 0.10 -0.36 0.43 -0.73 -0.27 -0.18 0.31 -0.37 0.05
CMB 0.58 0.15 0.40 -0.16 -0.07 0.06 0.13 -0.09 0.51
SLNK 0.00 -0.25 0.31 -0.76 -0.10 0.02 0.30 -0.30 0.17
CLXR 0.08 -0.33 0.41 -0.78 -0.20 -0.10 0.35 -0.35 0.12
SHRL -0.19 0.10 -0.31 0.44 -0.28 -0.26 -0.64 -0.04 -0.36
INTC -0.22 -0.83 -0.14 -0.38 0.47 0.52 -0.24 0.51 0.36
MDRC 0.39 0.85 0.30 0.31 -0.46 -0.56 0.38 -0.53 -0.20
BRL -0.17 -0.53 -0.17 0.48 -0.12 -0.02 -0.03 0.12 0.10
GRL -0.05 0.68 -0.02 0.19 -0.28 -0.24 0.34 -0.41 -0.44
URBA -0.19 0.10 -0.31 0.44 -0.28 -0.26 -0.64 -0.04 -0.36
MRL -0.20 0.10 -0.31 0.43 -0.28 -0.25 -0.64 -0.04 -0.36
FRT -0.05 0.68 -0.03 0.20 -0.27 -0.24 0.33 -0.39 -0.45
AAR 0.14 -0.24 -0.02 0.68 -0.41 -0.20 -0.12 -0.10 0.23
PET -0.52 0.20 -0.41 -0.31 0.28 0.22 -0.27 0.15 -0.73
AI 0.26 -0.25 0.12 0.59 -0.40 -0.22 0.03 -0.14 0.39

Strong relations are shaded with gray color

Katar Fete and these catchments are not used to develop with drainage density. Since Alfa is related to peak flow
regional model. The remaining eight catchments were used parameter, DD is expected to control peak flow, this is also
for regionalization. physically explainable relationship. Hq has strong negative
The correlation between model parameters (MPs) and correlation with longest flow path and perimeter, which can
Physical Catchment Characteristics (PCCs) is found strong be physically explained relationship.
only for 13 out of 288 relations which means the r values Many scholars suggested that values of model parameters
between (0.7 up to 1) and (− 0.7 up to − 1) and moderate for (MPs) can be better determined by multiple PCCs than only
123 out of 288 relationships which means r values between one PCC (Patel et al. 2016; Rientjes et al. 2011; Tamalew
(0.3 up to 0.7) and (− 0.3 up to − 0.7) (Table 6). The cor- and Kemal 2016) since runoff generation is affected by
relation is weak for most pairs of MPs and PCCs (152 out several catchment characteristics. For this reason, relations
of 288) which means r values between (0 up to 0.3) and (0 between PCCs and MPs were assessed through multiple lin-
up to − 0.3), whereas, no relationship was found for 0 out ear regression analysis. To establish regional model using
of 288 relationship. This suggests the need to implement calibrated MPs and PCCs of gauged catchments, stepwise
multiple regression instead of simple regression to establish regression method is applied, which involves progressively
relationship between MPs and PCCs. adding or removing one PCC at a time based pre- speci-
Table 6 shows a strong correlation that can be physically ­ 2 greater than
fied criteria (i.e. P-value of less than 0.1and R
explained existing between several MPs and PCCs. For the 0.80).
sake of brevity some of these strong and expected relation- This study, evaluated developed regional model in gauged
ships are discussed here. Alfa has strong positive correlation catchments before transferring the model parameters to
 Modeling Earth Systems and Environment

Table 7  Validating the No Gauged Model parameters (MPs)

regionalization model in gauged catchments
catchments Alfa Beta Cflux Fc K4 Khq Lp Perc Hq

1 Bedessa 1.48 2.16 0.505 1012.8 0.008 0.042 0.72 0.64 5.33
2 Dedeba 0.32 2.61 0.115 1289.3 0.004 0.051 0.62 0.11 1.56
3 Kolla 0.04 1.36 0.019 1017.5 0.020 0.102 0.56 1.21 2.99
4 Batena 0.46 1.25 0.336 484.7 0.023 0.054 0.60 0.22 3.37
5 Gidabo A 0.07 1.11 0.005 1171.2 0.084 0.088 0.38 2.85 1.30
6 Chifu Ar 0.04 1.79 0.020 1483.7 0.001 0.051 0.31 1.21 2.10
7 Ferfuro 0.24 1.90 0.115 894.2 0.053 0.093 0.67 1.22 4.01
8 U Gelana 0.22 1.00 0.087 1463.0 0.004 0.055 0.56 0.65 4.44

target catchments and found that the regionalization model added to the regression equation, which caused R ­ 2 increased
output in permissible range that means not underestimate up to 0.99 with P-value of less than 0.1. Addition of another
and overestimate HBV-IHMS model parameters in gauged PCCs did not improve the relationship. Hence, the collin-
catchments (Table 7). The evaluation of regional model earity found here are feasible and the regression equation
performance using pseudo-ungauged catchments is done by is developed (Table 8). The resulting statistical relationship
many researchers (Rientjes et al. 2011; Samuel et al. 2011). between Cflux with CLXRS soil is much more physically
Alfa: In this study, Alfa has the strongest relationship explainable than the relation with DD because of capillary
with drainage density (Table 6). Thus, drainage density rise is affected by soil properties.
is taken as the initial variable for executing forward entry Field capacity (FC): describes the maximum water
method because of simple linear regression coefficient R ­ 2 is storage in the catchment. The value of FC can be estimated
0.95 between Alfa and DD. Next, LEPS was added to the based on soil type and the rooting depth of the predominant
regression equation, which caused ­R2 increase up to 0.99 vegetation. For FC, the backward elimination method is exe-
with P-value of less than 0.1. cuted with adding variables like AI, VERS and LFP with ­R2
Addition of other variables did not improve the ­R2. The increased up to 0.94 and P-value ≤ 0.1. Addition of another
regression equation is accepted, and the statistical charac- PCCs did not improve the relationship. For this reason, the
teristics are shown in Table 7. Since Alfa is quick runoff collinearity found here is feasible and the regression equa-
parameter this equation appears to be plausible. Catchments tion is developed (Table 8). The relation between FC with
with high DD can generate quick runoff (i.e. small time of soil and meteorological characteristics is plausible than the
concentrations) while high Leptos soil (%) indicates high relation between FC with topography.
runoff contributions due to its low infiltration characteristics. K4: is the recession coefficient in the lower zone. It has
Beta: Has significant relationship with moderately cul- no significant relationship with any of the single catchment
tivated land but addition of another catchments character- characteristics. Therefore, the backward elimination method
istics is not increased. Therefore, the backward elimination is executed. After adding Area of the catchment, Fluvious
method is executed. After adding CI, VERS and DD catch- soil, AI and ­Hmean catchment characteristics ­R2 increased
ment characteristics ­R2 increased up to 0.98 with P-value to 0.84 with P- value of less than 0.1. Addition of other
of less than 0.1. Addition of another PCCs did not improve variables did not improve the ­R2. The regression equation
the relationship. Hence, the collinearity found here are fea- is accepted, and the statistical characteristics are shown
sible and the regression equation is developed Table 7. The in Table 8. Since K4 is base flow parameter this equation
resulting statistical relationship between Beta and CI, VERS appears to be plausible.
and DD can be physically explained since Beta affects the Khq: has no significant relation with any of the single
runoff generation which is determined by topography and catchment characteristics. For this reason, the backward
soil characteristics. elimination method is executed. After adding Bshape, cam-
Capillary rise coefficient (Cflux): is a model parameter bisols and marsh land catchment characteristics R ­ 2 increased
which is used to correct water rise by capillary through soil to 0.83 with P-value of less than 0.1. Addition of other vari-
matrix. In this study, Cflux has the strongest relationship ables did not improve the ­R2. Hence, the collinearity found
with drainage density with R ­ 2 is 0.74 between Cflux and DD here are feasible and the regression equation is developed
when evaluating simple linear regression correlation coef- (Table 8). The resulting statistical relationship between Khq
ficient. Thus, drainage density is taken as the initial vari- and Bshape, CMBS and MRL can be physically explained.
able for executing forward entry method. Next, CLXRS was Since Khq is peak flow parameter, catchments with high
Modeling Earth Systems and Environment

Table 8  The regional model and Coefficients 95% confidence Co-linearity

its statistical characteristics R2
β Std error t stat P-value L bound U bound

Alfa = 𝛽0 + 𝛽1 × DD + 𝛽2 × LEPS
𝛽0 − 0.32 0.04 − 9.27 0.00 − 0.41 − 0.23 0.99
𝛽1 1.59 0.07 23.47 0.00 1.41 1.76
𝛽2 0.22 0.05 4.28 0.01 0.09 0.35
Beta = 𝛽0 + 𝛽1 × CI + 𝛽2 × VERS + 𝛽3 × DD
𝛽0 − 0.10 0.10 − 0.97 0.04 − 0.39 0.19 0.98
𝛽1 0.01 0.00 14.30 0.00 0.01 0.01
𝛽2 0.02 0.00 10.21 0.00 0.01 0.02
𝛽3 1.48 0.11 13.00 0.00 1.17 1.80
Cflux = 𝛽0 + 𝛽1 × DD + 𝛽2 × CLXRS
𝛽0 − 0.10 0.01 − 7.52 0.00 − 0.14 − 0.07 0.99
𝛽1 0.53 0.03 20.29 0.00 0.47 0.60
𝛽2 0.02 0.00 11.26 0.00 0.01 0.02
Fc = 𝛽0 + 𝛽1 × AI + 𝛽2 × VERS + 𝛽3 × LFPS
𝛽0 − 616.19 216.30 − 2.85 0.05 − 1216.75 − 15.64 0.94
𝛽1 772.41 104.58 7.39 0.00 482.05 1062.78
𝛽2 6.29 2.00 3.14 0.03 0.72 11.85
𝛽3 17.22 3.04 5.67 0.00 8.79 25.66
K4 = 𝛽0 + 𝛽1 × Area + 𝛽2 × FLVS + 𝛽3 × AI + 𝛽4 × Hmean
𝛽0 0.19 0.12 1.62 0.02 − 0.19 0.57 0.84
𝛽1 0.00 0.00 2.49 0.09 0.00 0.00
𝛽2 0.00 0.00 − 1.35 0.03 0.00 0.00
𝛽3 − 0.04 0.02 − 2.24 0.01 − 0.10 0.02
𝛽4 0.00 0.00 − 1.27 0.03 0.00 0.00
Khq = 𝛽0 + 𝛽1 × Bshape + 𝛽2 × CMBS + 𝛽3 × MRL
𝛽0 0.16 0.02 7.03 0.00 0.10 0.23 0.83
𝛽1 0.00 0.00 − 4.19 0.01 0.00 0.00
𝛽2 0.00 0.00 3.63 0.02 0.00 0.00
𝛽3 0.15 0.05 2.85 0.05 0.00 0.30
LP = 𝛽0 + 𝛽1 × Rb + 𝛽2 × URBA + 𝛽3 × Area
𝛽0 0.58 0.08 6.92 0.00 0.35 0.82 0.85
𝛽1 0.05 0.04 1.52 0.02 − 0.04 0.15
𝛽2 − 0.48 0.14 − 3.44 0.03 − 0.87 − 0.09
𝛽3 0.00 0.00 − 2.95 0.04 0.00 0.00
Perc = 𝛽0 + 𝛽1 × Area + 𝛽2 × FRST + 𝛽3 × CLXRS + 𝛽4 × AI
𝛽0 1.45 0.62 2.36 0.01 − 0.51 3.41 0.94
𝛽1 0.00 0.00 4.07 0.03 0.00 0.01
𝛽2 − 0.10 0.04 − 2.86 0.06 − 0.22 0.01
𝛽3 − 0.06 0.03 − 1.89 0.01 − 0.16 0.04
𝛽4 − 0.67 0.36 − 1.89 0.02 − 1.81 0.46
Hq = 𝛽0 + 𝛽1 × LFP + 𝛽2 × CLXRS + 𝛽3 × FRST
𝛽0 7.75 0.48 15.98 0.00 6.40 9.09 0.97
𝛽1 − 0.13 0.01 − 9.54 0.00 − 0.17 − 0.09
𝛽2 − 0.07 0.03 − 2.73 0.05 − 0.15 0.00
𝛽3 0.18 0.05 3.87 0.02 0.05 0.32
 Modeling Earth Systems and Environment

marshland can generate high runoff (i.e. covers large areas land percent covers have increases the Hq due to high mois-
and small-time of concentration). While catchments with ture content.
low percent covers of CMBS can generate quick runoff due
to high water holding capacity and high infiltration rates Validation of regional model with spot measured
respectively. and historically archived data
Limit for potential evaporation (LP): It has no signifi-
cant relation with any of the single catchment characteristics. The developed regional model also evaluted in addition to
Therefore, the backward elimination method is executed. pseudo-ungauged catchments. Regional model simulated
After adding bifurcation ratio, urban area (%) and area of flow was validated with observed discharege data (i.e. spot
catchment characteristics ­R2 increased to 0.85 with P-value measurements and historical data) in three target catchments,
of less than 0.1. Addition of other variables did not improve which were not used in the establishemente of the regional
the ­R2. Hence, the collinearity found here are feasible and model. Each model parameters were estimated by regional
the regression equation is acceptable (Table 8). The resulting model method derived equation (Table 9). The estimated
statistical relationship between LP and Rb, URBA and Area model parameters values were in allowable ranges.
can be physically explained. Since LP is affects potential The established regional model (Table )8 was used to
evaporation, the relation between LP with soil (FLVS) and estimate the model parameters (Table 9) of target catchments
URBA is much more plausible than the relation between LP using their PCCs. Next discharge was simulated HBV-IHMS
with topography and geology (Rb). based on the estimated model parameters. The simulated
Percolation capacity (Perc): is water movement in soil flow was compared with spot flow measured data for the
matrix from upper to lower response zone. It has signifi- period from 01 February 2019 to 31 December 2022 in each
cance relation with Area of catchment. Therefore, the for- spot flow measurements sites.
ward elimination method is executed. After adding Forest, In Kulfo catchment, regional model simulated stream-
CLXRS and AI catchment characteristics R2 increased up flow well captured the measured streamflow especially in
to 0.94 with P-value of less than 0.1. Addition of other vari- base flows and recession limb of the hydrograph (Fig. 10).
ables did not improve the ­R2. Hence, the collinearity found However, peak flows are not reasonably captured. In Hare
here are feasible and the regression equation is acceptable and Hamessa catchments, the simulated flows reasonably
(Table 8). The resulting statistical relationship between captured the overall pattern of the observed hydrograph in
Perc and Area, FRST, CLXRS and AI can be physically the wet season (Fig. 10). However, the base flow is relatively
explained. Since Perc affects water movement in soil, the overestimated. Note that there is significant water abstraction
relation is more plausible with AI, CLXRS and FRST than for irrigation in the Hamessa catchment. This abstraction is
AREA. Where higher AI values may result in higher perco- not modelled in HBV. For Kulfo and Hare, the model cap-
lation. Catchments with high FRST % covers can increase tured the overall pattern but overestimated the stream flow
the Perc values due to high amount of ground water recharge. prior to the rainy season. In generally, the overall pattern
Hq: has the strongest relationship with longest flow path of simulated and observed hydrographs shows acceptable
with ­R2 is 0.80 between Hq and Lonest flow path (LFP) agreement for three catchments.
when evaluating simple linear regression correlation coeffi- The performance of the regional model was also evalu-
cient. Thus, LFP is taken as the initial variable for executing ated for the three catchments using objective functions
forward entry method. Next, CLXRS and FRST was added (Table 10). In terms of NSE, the model performance is
to the regression equation, which caused R ­ 2 increased to acceptable for all three catchments with good performance.
0.97 with P-value of less than 0.1. In terms of RVE, the model performance was acceptable for
Addition of another PCCs did not improve the relation- NSE greater than 0.5 and RVE with between − 10 and + 10.
ship. Hence, the collinearity found here are feasible and the The regional model performance is evaluated using
regression equation is developed (Table 8). The resulting historically archived observed discharge data for 15 year
statistical relationship between Hq with LFP, CLXRS and (1991–2005) of spot measured catchments. Figure 11 shows
FRST is physically explained. Catchments with high forest well agreement between the pattern of the simulated and

Table 9  MPs estimated in target No. Catchments Estimated model parameters

catchments using the regional
model Alfa Beta Cflux FC K4 Khq LP Perc Hq

1 Kulfo 0.61 1.20 0.14 760.4 0.09 0.06 0.52 2.17 2.32
2 Hare 0.41 1.28 0.15 724.1 0.04 0.03 0.64 1.36 3.10
3 Hamessa 0.98 3.51 0.27 1495.7 0.14 0.07 0.49 2.48 0.11
Modeling Earth Systems and Environment

120 0

100 40
Stream flow (m3/s)

Rainfall (mm)
80 80
Q obs Q sim RF (a)
60 120

40 160

200
20
240
0
01-01-2019 29-08-2019 25-04-2020 21-12-2020 18-08-2021 15-04-2022 11-12-2022

30 0

25 40

Rainfall (mm)
20
Stream flow (m3/s)

Q obs Q sim RF (b) 80

15
120
10

5 160

0 200
01-01-2019 29-08-2019 25-04-2020 21-12-2020 18-08-2021 15-04-2022 11-12-2022

60 0

50 40

Rainfall (mm)
Stream flow (m3/s)

40 Q obs Q sim RF (c ) 80

30 120

20 160

10 200

0 240
01-01-2019 29-08-2019 25-04-2020 21-12-2020 18-08-2021 15-04-2022 11-12-2022

Fig. 10  Simulated and observed hydrograph from spot measurements for Kulfo (a), Hare (b) and Hamessa (c) catchments

Table 10  Performance of the No. Catchments Model per- Spot measurements Historical period
regional model in capturing formances
the pattern and volume of the 2019–2022 1991–1995 1996–2000 2001–2005
observed spot measurements
and historical archived data 1 Kulfo NSE 0.57 0.56 0.41 0.43
RVE 7.61 − 13.03 − 28.47 − 6.64
2 Hare NSE 0.51 0.45 0.39 0.40
RVE − 8.86 27.90 5.44 1.67
3 Hamessa NSE 0.52 0.42 0.45 0.43
RVE − 8.66 24.46 27.92 26.50
 Modeling Earth Systems and Environment

Fig. 11  Simulated and observed 180 0

hydrograph for historical period 40
150
for Kulfo (a), Hare (b) and

Stream flow (m3/s)

Hamessa (c) catchments 120 80

Rainfall (mm)
120
90 Q obs Q sim RF (a)
160
60
200
30
240
0
01-01-1991 11-11-1992 22-09-1994 02-08-1996 13-06-1998 23-04-2000 04-03-2002 13-01-2004 23-11-2005

48 0

40 40
Stream flow (m3/s)

Rainfall (mm)
32 80

24 120
Q obs Q sim RF (b)
16 160

8 200

0 240
01-01-1991 11-11-1992 22-09-1994 02-08-1996 13-06-1998 23-04-2000 04-03-2002 13-01-2004 23-11-2005

90 0

75 40

Rainfall (mm)
Streamflow (m3/s)

60 80
Q obs Q sim RF (c)
45 120

30 160

15 200

0 240
01-01-1991 11-11-1992 22-09-1994 02-08-1996 13-06-1998 23-04-2000 04-03-2002 13-01-2004 23-11-2005

observed hydrograph for those three target catchments. period archived discharge data. The reasons for this improve-
For Kulfo catchment, the model well captured rising limb, ment include appropriate supervision of data collectors, the
recession, medium peaks and base flows for the period from site safely to recorded water level, and new rating curve
1991 to 2002. However, there is under estimate the base and were developed. Many literatures suggested that (Melsen
medium flow and some mismatches in 2003–2004 which et al. 2016; Nigussie et al. 2020; Seibert and Vis 2016) flow
suggests the need for inspecting the reliability of the stream- measurements are better informative for catchments and most
flow data for these 2 years. The model also well captured informative model parameters were determined if it consid-
the overall pattern of the observed hydrograph for Hare and ers dominant hydrological processes and covers a range of
Hamessa catchments. However, it consistently overestimated hydrological conditions (Meerveld et al. 2017; Negatu et al.
the base flow especially in dry and prior to rainy seasons. 2022). Because river system is changing over time and space.
In historical period, the simulated flow was validated with
historical archived data with model performances objec-
tive functions in Table 10. The evaluation was conducted Conclusions
for 3 different year classes (1991–1995), (1996–2000) and
(2001–2005). In terms of NSE, the model performance is Regional model was developed for Rift Valley Lakes basin
relatively weak for Kulfo, Hare and Hamessa catchments catchments to predict stream flow in ungauged watershed.
as compared to spot measured discharge data. In terms of The performance of the developed regional model was eval-
RVE, largest volumetric error in Hamessa catchment differ- uated by comparing the simulated flow with observed dis-
ent year classes and large volume error in Kulfo and Hare charge data obtained from spot measurements and historical
catchments. archived data. In each spot site, the current study developed
Note that, spot flow measured data were better related new rating curve and the accuracy of the rating curves was
to regional model simulated flow as compared to historical found adequate to estimate daily flow data in each spot sites.
Modeling Earth Systems and Environment

The HBV-IHMS model well has performed in the 8 Declarations

gauged watersheds with reasonably reproduced the volume
and pattern of the observed hydrograph. 9 calibrated model Conflict of interest The authors declare there is no conflict.
parameters and 32 physical catchments characteristics were
correlated in simple regression but the relationships are
weak for most pairs. For this reasons, multiple regression References
method was selected to develop the regional model. The
Abimbola OP, Wenninger J, Venneker R, Mittelstet AR (2020) The
relationship between MPs and PCCs was not only statisti- assessment of water resources in ungauged catchments in Rwanda.
cally significant but also the relationship was justified with J Hydrol Region Stud 13(August 2017):274–289. https://​doi.​org/​
a strong hydrological explanation. 10.​1016/j.​ejrh.​2017.​09.​001
The study also evaluated the accuracy of regional model Arnold JG, Srinivasan R, Muttiah RS, Williams JR (1998) Large area
hydrologic modeling and assessment part i: model development.
simulated flows with historical period for different time peri- Wiley 34(1):73–89. https://​doi.​org/​10.​1111/j.​1752-​1688.​1998.​
ods for Kulfo, Hare and Hamessa catchments. However, poor tb059​61.x
relationship was found between regional model simulated Beven K (2001) Dalton lecture: how far can we go in distributed hydro-
flows and historical archived streamflow data as the model logical modelling? Keith Beven Lancaster University. Computer,
5(Figure 2), pp 1–12. http://​eprin​ts.​lancs.​ac.​uk/​4420/
performances were beyond the acceptable ranges (NSE ≤ 0.5 Bloschl G, Sivapalan M (1995) Scale issues in hydrological modelling:
and RVE outside of − 10 and 10%). This can be partly a review. Hydrol Process 9(February 1994):251–290
related with poor rating curve which was not updated to Chorowicz J (2005) The East African rift system. J Afr Earth Sci
43(1–3):379–410. https://d​ oi.o​ rg/1​ 0.1​ 016/j.j​ afrea​ rsci.2​ 005.0​ 7.0​ 19
accommodate changes in river cross-sections due to siltation
Davids JC (2017) Continuity vs. the crowd—tradeoffs between con-
and scouring. The field visit under this study also revealed tinuous and intermittent citizen hydrology stream flow obser-
that the monitoring stations were unused for several years. vations. Environ Manag 60:12–29. https://​d oi.​o rg/​1 0.​1 007/​
The regional model was found acceptable when the spot s00267-​017-​0872-x
Davids JC, Rutten MM, Pandey A, Devkota N, Van Oyen WD (2019)
stream flow measurements were used as reference for Kulfo,
Citizen science flow—an assessment of simple streamflow meas-
Hare and Hamessa stations. This was verified by visual urement methods. Hydrol Earth Syst Sci 23:1045–1065
inspection of joint plots, the simulated and observed stream- Du TLT, Lee H, Bui DD, Arheimer B, Li H, Stephen E (2020) Stream-
flow. The values of the objective functions were also within flow prediction in “ geopolitically ungauged ” basins using sat-
ellite observations and regionalization at subcontinental scale. J
the acceptable range of values (NSE ≥ 0.5 and RVE within
Hydrol 588:125016
− 10 and 10%) indicating that the regional model perfor- Farfán JF, Cea L (2023) Regional streamflow prediction in north-
mance is satisfactory to be used for estimating of ungauged west Spain: a comparative analysis of regionalisation schemes. J
flows in the basin. The main reasons for these improvements Hydrol Reg Stud 47(June):101427. https://​doi.​org/​10.​1016/j.​ejrh.​
2023.​101427
were the spot measurements involved updating the rating
Gebregiorgis AS, Moges SA, Awulachew SB (2013) Basin regionaliza-
curves and the good monitor the sites. tion for the purpose of water resource development in a limited
In general, the findings of this study showed that spot data situation: case of Blue Nile River Basin, Ethiopia. J Hydrol
flow measurements are one of viable option to fill data Eng 18(10):1349–1359. https://​doi.​org/​10.​1061/​(asce)​he.​1943-​
5584.​00007​30
gaps in ungauged flow estimation. The spot measurements
Genç O, Ard M, Necati A (2015) Calculation of mean velocity and dis-
can be used to increase the confidence on the applicability charge using water surface velocity in small streams. Flow Meas
regionalization and hence the measurements can serve as an Instrum 41:115–120
alternative data source for different water resources studies Georgievová I, Hanel M, Pavlík P, Vizina A (2023) Streamflow simula-
tion in poorly gauged basins with regionalised assimilation using
including water budget studies. Thus, the study addressed
Kalman filter. J Hydrol 620(PA):129373. https://d​ oi.o​ rg/1​ 0.1​ 016/j.​
the problem of classical regionalization methods by seeking jhydr​ol.​2023.​129373
practical solution for filling data gaps. Goshime DW, Absi R, Ledésert B (2019) Evaluation and bias correc-
tion of CHIRP rainfall estimate for rainfall-runoff simulation over
Acknowledgements We would like to thank Arab Minch University Lake Ziway Watershed, Ethiopia. Hydrology 6(3):68. https://​doi.​
Water Resources Research Center for providing support during field org/​10.​3390/​hydro​logy6​030068
work and data analysis. We also extend our thanks to Rift Valley Lakes Haile AT, Geremew Y, Wassie S, Fekadu AG, Taye MT (2023) Filling
Bain Authority and site data collectors streamflow data gaps through the construction of rating curves
in the Lake Tana sub-basin, Nile basin. J Water Clim Change
Funding This research was supported by Arba Minch University, 00:1–14. https://​doi.​org/​10.​2166/​wcc.​2023.​372
Water Resources Research Center. Halefom A, Sisay E, Khare D, Singh L, Worku T (2017) Hydrologi-
cal modeling of urban catchment using semi-distributed model.
Availability of data and materials The datasets used in the current study Model Earth Syst Environ 3(2):683–692. https://​doi.​org/​10.​1007/​
are available from the first and corresponding author on reasonable s40808-​017-​0327-7
request. Hrachowitz M, Savenije HHG, Blöschl G, McDonnell JJ, Sivapalan M,
Pomeroy JW, Arheimer B, Blume T, Clark MP, Ehret U, Fenicia
 Modeling Earth Systems and Environment

F, Freer JE, Gelfan A, Gupta HV, Hughes DA, Hut RW, Montan- Water Resour Res 52(6):4765–4784. https://​doi.​org/​10.​1002/​
ari A, Pande S, Tetzlaff D et al (2013) A decade of predictions 2015W​R0185​49
in ungauged basins (PUB)—a review. Hydrol Sci J 58(6):1198– Samuel J, Coulibaly P, Asce M, Metcalfe RA (2011) Estimation of
1255. https://​doi.​org/​10.​1080/​02626​667.​2013.​803183 continuous streamflow in Ontario ungauged basins: comparison
Hughes DA, Gush M, Tanner J, Dye P (2014) Using targeted short- of regionalization methods. J Hydrol Eng 16:447–459. https://d​ oi.​
term field investigations to calibrate and evaluate the structure of org/​10.​1061/​(ASCE)​HE.​1943-​5584.​00003​38
a hydrological model. Hydrol Process 28(May 2013):2794–2809. Seibert J (1999) Regionalisation of parameters for a conceptual rainfall-
https://​doi.​org/​10.​1002/​hyp.​9807 runoff model. Agric for Meteorol 98–99:279–293
Isik S, Singh VP (2008) Hydrologic Regionalization of Watersheds Seibert J, Beven KJ (2009) Gauging the ungauged basin: how many
in Turkey. J Hydrol Eng 13(9):824–834. https://​doi.​org/​10.​1061/​ discharge measurements are needed? Hydrol Earth Syst Sci
(asce)​1084-​0699(2008)​13:​9(824) 13:883–892
Kebebew AS, Awass AA (2022) Regionalization of catchments for Seibert J, Vis MJP (2016) How informative are stream level obser-
flood frequency analysis for data scarce Rift Valley Lakes Basin, vations in different geographic regions? Hydrol Process
Ethiopia. J Hydrol Region Stud 43(February):101187. https://​ 2508(June):2498–2508. https://​doi.​org/​10.​1002/​hyp.​10887
doi.​org/​10.​1016/j.​ejrh.​2022.​101187 Smitha PS, Narasimhan B, Sudheer KP, Annamalai H (2018) An
Lowry CS, Fienen MN (2012) CrowdHydrology : crowdsourcing improved bias correction method of daily rainfall data using a
hydrologic data and engaging citizen scientists. Groundwater sliding window technique for climate change impact assessment.
51:151–156. https://​doi.​org/​10.​1111/j.​1745-​6584.​2012.​00956.x J Hydrol 556:100–118. https://​doi.​org/​10.​1016/j.​jhydr​ol.​2017.​11.​
Mada ZM, Nannawo AS (2023) Enhancing understanding of hydro- 010
logic processes in the Shafe Watershed, Ethiopia. Adv Civ Eng Swain JB, Patra KC (2017) Streamflow estimation in ungauged catch-
2023:1–18. https://​doi.​org/​10.​1155/​2023/​55778​51 ments using regionalization techniques. J Hydrol. https://​doi.​org/​
Melsen LA, Teuling JA, Torfs JJFP, Uijlenhoet R, Mizukami N, Clark 10.​1016/j.​jhydr​ol.​2017.​08.​054
PM (2016) HESS opinions: the need for process-based evalua- Tada T, Beven KJ (2012) Hydrological model calibration using a short
tion of large-domain hyper-resolution models. Hydrol Earth Syst period of observations. Hydrological Process 892(November
Sci 20(3):1069–1079. https://​doi.​org/​10.​5194/​hess-​20-​1069-​2016 2011):883–892. https://​doi.​org/​10.​1002/​hyp.​8302
Merz R (2004) Regionalisation of catchment model parameters. J Tamalew C, Kemal A (2016) Estimation of discharge for Ungauged
Hydrol 287:95–123. https://d​ oi.o​ rg/1​ 0.1​ 016/j.j​ hydro​ l.2​ 003.0​ 9.0​ 28 catchments using rainfall-runoff model in Didessa Sub-basin: the
Mishra AK, Coulibaly P (2009) Developments in hydrometric net- case of Blue Nile River. Int J Innov Eng Res Technol [IJIERT]
work design: a review. Rev Geophys 2007:1–24. https://​doi.​org/​ 3(9):62–72
10.​1029/​2007R​G0002​43.1.​INTRO​DUCTI​ON Taylor P, Hrachowitz M, Savenije HHG (2013) A decade of predictions
Negatu TA, Zimale FA, Steenhuis TS (2022) Establishing stage-dis- in ungauged basins (PUB)— a review. Hydrol Sci J. https://​doi.​
charge rating curves in developing countries: Lake Tana Basin, org/​10.​1080/​02626​667.​2013.​803183
Ethiopia. Hydrology. https://​doi.​org/​10.​3390/​hydro​logy9​010013 Teshome M, Zerihun A, Mada M, Dagalo S, Abdella H, Mohammed K
Nigussie L, Haile AT, Gowing J, Walker D, Parkin G (2020) Citizen (2022) A compendious approach for renewable energy assessment
science in community-based watershed management: an institu- based on satellite and ground truth data : Bilate catchment, Rift
tional analysis in Ethiopia. IWMI Working Papers, 2020(191), Valley Basin, Ethiopia. Int J Energy Environ Eng 13(3):1081–
IWMI-1-IWMI-18. https://​doi.​org/​10.​5337/​2020.​207 1100. https://​doi.​org/​10.​1007/​s40095-​022-​00484-7
Oudin L, Andre V, Perrin C, Michel C (2008) Spatial proximity, physi- Tibangayuka N, Mulungu DMM, Izdori F (2022) Performance evalua-
cal similarity, regression and ungauged catchments: a comparison tion, sensitivity, and uncertainty analysis of HBV model in Wami
of regionalization approaches based on 913 French catchments. Ruvu basin, Tanzania. J Hydrol Region Stud 44(August):101266.
Water Resour Res 44:1–15. https://d​ oi.o​ rg/1​ 0.1​ 029/2​ 007WR ​ 0062​ https://​doi.​org/​10.​1016/j.​ejrh.​2022.​101266
40 Tsegaw AT, Pontoppidan M, Kristvik E, Alfredsen K, Muthanna TM
Patel S, Hardaha MK, Seetpal MK, Madankar KK (2016) Multiple (2020) Hydrological impacts of climate change on small ungauged
linear regression model for stream flow estimation of Wainganga. catchments—results from a global climate model-regional climate
Am J Water Sci Engi. https://​doi.​org/​10.​11648/j.​ajwse.​20160​201.​ model-hydrologic model chain. Nat Hazard 20(8):2133–2155.
11 https://​doi.​org/​10.​5194/​nhess-​20-​2133-​2020
Pool S, Viviroli D, Seibert J (2017) Prediction of hydrographs and Van Meerveld HJI, Vis MJP, Seibert J (2017) Information content of
flow-duration curves in almost ungauged catchments: which run- stream level class data for hydrological model calibration. Hydrol
off measurements are most informative for model calibration? J Earth Syst Sci 21:4895–4905
Hydrol 554:613–622. https://​doi.​org/​10.​1016/j.​jhydr​ol.​2017.​09.​ Viglione A, Parajka J, Rogger M, Salinas JL, Laaha G, Sivapalan
037 M, Blöschl G (2013) Comparative assessment of predictions in
Pool S, Seibert J (2021) Gauging ungauged catchments—active learn- ungauged basins—part 3: runoff signatures in Austria. Hydrol
ing for the timing of point discharge observations in combination Earth Syst Sci 17(6):2263–2279. https:// ​ d oi. ​ o rg/ ​ 1 0. ​ 5 194/​
with continuous water level measurements. J Hydrol 598(Janu- hess-​17-​2263-​2013
ary):126448. https://​doi.​org/​10.​1016/j.​jhydr​ol.​2021.​126448 Viviroli D, Seibert J (2015) Can a regionalized model parameterisation
Pool S, Viviroli D, Seibert J (2019) Value of a limited number of dis- be improved with a limited number of runoff measurements ? J
charge observations for improving regionalization: a large-sample Hydrol 529:49–61. https://​doi.​org/​10.​1016/j.​jhydr​ol.​2015.​07.​009
study across the United States. Water Resour Res 55(1):363–377. Wagener T, Montanari A (2011) Convergence of approaches toward
https://​doi.​org/​10.​1029/​2018W​R0238​55 reducing uncertainty in predictions in ungauged basins. Water
Rientjes THM, Perera BUJ, Haile AT, Reggiani P, Muthuwatta LP Resour Res 47:1–8. https://​doi.​org/​10.​1029/​2010W​R0094​69
(2011) Regionalisation for lake level simulation—the case of Wale A, Rientjes THM, Gieske ASM, Getachew HA (2009) Ungauged
Lake Tana in the Upper Blue Nile, Ethiopia. Hydrol Earth Syst catchment contributions to Lake Tana’s water balance. Hydrol
Sci 2009:1167–1183. https://​doi.​org/​10.​5194/​hess-​15-​1167-​2011 Process. https://​doi.​org/​10.​1002/​hyp.​7284
Rojas-Serna C, Lebecherel L, Perrin C, Andréassian V, Oudin L (2016) Worqlul AW, Bayabil HK, Yimam YT, Seifu A (2018) Advances in
How should a rainfall-runoff model be parameterized in an almost water resources research in the Upper Blue Nile basin and the
ungauged catchment? A methodology tested on 609 catchments.
Modeling Earth Systems and Environment

way forward : a review. J Hydrol. https://​doi.​org/​10.​1016/j.​jhydr​ Publisher's Note Springer Nature remains neutral with regard to
ol.​2018.​03.​042 jurisdictional claims in published maps and institutional affiliations.
Wu H, Zhang J, Bao Z, Wang G, Wang W, Yang Y (2022) Runoff mod-
eling in ungauged catchments using machine learning algorithm- Springer Nature or its licensor (e.g. a society or other partner) holds
based model parameters regionalization methodology. Engineer- exclusive rights to this article under a publishing agreement with the
ing. https://​doi.​org/​10.​1016/j.​eng.​2021.​12.​014 author(s) or other rightsholder(s); author self-archiving of the accepted
Yang X, Magnusson J, Xu CY (2019) Transferability of regionalization manuscript version of this article is solely governed by the terms of
methods under changing climate. J Hydrol 568(August 2018):67– such publishing agreement and applicable law.
81. https://​doi.​org/​10.​1016/j.​jhydr​ol.​2018.​10.​030
Zhang Y, Chiew FHS (2009) Relative merits of different methods for
runoff predictions in ungauged catchments. Water Resour Res.
https://​doi.​org/​10.​1029/​2008W​R0075​04