J. Geogr. Sci.

2023, 33(6): 1287-1312

DOI: https://doi.org/10.1007/s11442-023-2130-x

© 2023 Science Press Springer-Verlag

Dynamics of soil loss and sediment export as

affected by land use/cover change in Koshi River
Basin, Nepal
YIGEZ Belayneh1,2, *XIONG Donghong1,3, ZHANG Baojun1, BELETE Marye4,
CHALISE Devraj5, CHIDI Chhabi Lal6, GUADIE Awoke7, WU Yanhong1,3,
RAI Dil Kumar1,2
1. Institute of Mountain Hazards and Environment, CAS, Chengdu 610041, China;
2. University of Chinese Academy of Sciences, Beijing 100049, China;
3. Branch of Sustainable Mountain Development, Kathmandu Center for Research and Education, CAS-TU,
Kathmandu 44613, Nepal;
4. Department of Natural Resource Management, Debre Tabor University, Debre Tabor 272, Ethiopia;
5. Mackay Area Productivity Services, 26135 Peak Downs Highway, Te Kowai QLD 4740, Australia;
6. Central Department of Geography, Tribhuvan University, Kathmandu, Nepal;
7. Department of Biology, College of Natural Sciences, Arba Minch University, Arba Minch 21, Ethiopia

Abstract: How the dynamics in soil loss (SL) and sedimentation are affected by land
use/cover change (LULCC) has long been one of the most important issues in watershed
management worldwide, especially in fragile mountainous river basins. This study aimed to
investigate the impact of LULCC on SL and sediment export (SE) in eastern regions of the
Koshi River basin (KRB), Nepal, from 1990 to 2021. The Random Forest classifier in the
Google Earth Engine platform was employed for land use/land cover (LULC) classification,
and the Integrated Valuation Ecosystem Services and Trade-offs (InVEST) Sediment Delivery
Ratio model was used for SL and SE modeling. The results showed that there was a pro-
nounced increase in forest land (4.12%), grassland (2.35%), and shrubland (3.68%) at the
expense of agricultural land (10.32%) in KRB over the last three decades. Thus, the mean SL
and SE rates decreased by 48% and 60%, respectively, from 1990 to 2021. The conversion of
farmland to vegetated lands has greatly contributed to the decrease in SL and SE rates.
Furthermore, the rates of SL and SE showed considerable spatiotemporal variations under
different LULC types, topographic factors (slope aspect and gradient), and sub-watersheds.
The higher rates of SL and SE in the study area were observed mostly in slope gradient
classes between 8° and 35° (accounting for 83%–91%) and sunny and semi-sunny slope
aspects (SE, S, E, and SW) (accounting for 57%–65%). Although the general mean rate of SL
presented a decreasing trend in the study area, the current mean SL rate (23.33 t ha−1 yr−1) in
2021 is still far beyond the tolerable SL rate of both the global (10 Mg ha−1 yr−1) and the
Himalayan region (15 t ha−1 yr−1). Therefore, landscape restoration measures should be

Received: 2022-07-19 Accepted: 2023-03-06

Foundation: Chinese Academy of Sciences (CAS) Overseas Institution Platform Project, No.131C11KYSB20200033
Author: YIGEZ Belayneh, PhD Candidate, E-mail:[email protected]
*
Corresponding author: Xiong Donghong, PhD and Professor, E-mail: [email protected]

www.geogsci.com www.springer.com/journal/11442
1288 Journal of Geographical Sciences

integrated with other watershed management strategies and upscaled to hotspot areas to
regulate basin sediment flux and secure ecosystem service sustainability.
Keywords: soil loss; sediment export; land use/cover; InVEST model; Koshi River basin; Google Earth Engine

1 Introduction
Soil loss (SL) and sediment export (SE) are among the core global environmental problems
that severely affect food security, water quality, and climate change mitigation (Wang et al.,
2018; Zhao et al., 2018; Borrelli et al., 2020). Anthropogenic activities through land use and
land cover (LULC) conversion, among other factors, play the main role in controlling the
spatiotemporal variability of soil SL and SE (Latocha et al., 2016). Mountainous areas under
intense anthropogenic influences are particularly vulnerable, and the rates of SL and SE in
these regions are extremely variable and strongly linked to cover factors (Sun et al., 2013;
Teng et al., 2019; Zhou et al., 2019). Understanding the responses of SL and SE to LULCC,
especially in fragile mountainous river basins, is thus a major research concern for effective
river basin management.
Due to its rugged topography, young and fragile geology, active tectonics, and torrential
rains, the Koshi River basin (KRB) in Nepal is known to be one of the major SL and SE
hotspots both in Nepal and worldwide (Jain et al., 2001; Golosov and Walling, 2019). Its
annual sediment load to the Ganges River system is estimated to be 100–135 Mt yr−1 (nearly
25% of its total sediment budget) (ICIMOD, 2018). Anthropogenic activities through
LULCC have greatly increased their magnitude in this basin (ICIMOD, 2018). Apart from
acute land degradation, excessive SL and SE in this basin often result in severe riverbed ag-
gradation and cause severe flood hazards downstream (Sinha et al., 2019; Yigez et al., 2021).
To prevent agricultural intensification and the associated high rates of SL and SE, the Gov-
ernment of Nepal initiated various forest conservation plans and programs in the late 1980s,
such as the Master Plan for the Forestry Sector (1988) and the implementation of the com-
munity forestry program in the early 1990s (Bhawana et al., 2017). In addition, population
migration from the hill and mountainous areas to lowland and urban areas is the main phe-
nomenon that has triggered LULCC in the steep slope areas of this basin in recent decades
(Paudel et al., 2020). The abovementioned driving factors could have a significant impact on
LULCC as well as on the potential for SL and SE in the region. Therefore, it is necessary to
understand the spatiotemporal dynamics of LULCC along with the associated impacts on SL
and SE to provide reliable information for well-informed river basin management inter-
ventions in the future.
Several studies have examined the responses of SL and SE to LULCC (Bakker et al.,
2008; Xu et al., 2011; Borrelli et al., 2017; Zhou et al., 2019; Aneseyee et al., 2020), stating
that soil erosion is closely linked with LULC patterns and that its impact on the rate of SL
and SE is not homogeneous in space and among LULC types. In KRB, a handful of research
attempts have been made; however, most of them have only focused on the driving forces of
farmland abandonment (Paudel et al., 2020), LULCC, and their socioeconomic and ecosys-
tem service implications such as food production, carbon storage, and habitat quality (Wu et
al., 2017; Rimal et al., 2019; Xie et al., 2021). There have also been some isolated studies
that only focus on soil erosion, sediment dynamics, and the geo-environmental implications
YIGEZ Belayneh et al.: Dynamics of soil loss and sediment export in Koshi River Basin, Nepal 1289

of sediment dynamics (Chapagain and Banjade, 2009; Uddin et al., 2016; Rimal et al., 2019;
Yigez et al., 2021; Yuan et al., 2021). However, in the mountainous regions of KRB, SL and
SE rate evaluation with respect to LULCC are particularly sporadic. Due to the limited
availability of cloud-free satellite imagery for LULCC analysis, related studies have been
more spatially focused on the Hill and Terai regions than on their mountainous counterparts
(Virgo and Subba, 2016; Gilani et al., 2017; Yigez et al., 2021). As a result, studies inte-
grating these three aspects at different temporal scales in the mountainous regions of KRB
remain quite scarce.
Therefore, the objectives of this study are (1) to examine the spatiotemporal variability of
LULC dynamics; (2) to evaluate the spatial pattern and temporal dynamics of SL and SE;
and (3) to examine the relationship between SL, SE, and LULC dynamics as well as to
discriminate priority areas for conservation interventions using the random forest (RF)
classifier algorithm in the Google Earth Engine (GEE) cloud platform for LULCC analysis,
and the Integrated Valuation Ecosystem Services and Trade-offs (InVEST) and Sediment
Delivery Ratio (SDR) model (hereafter InVEST model) for SL and SE modeling. Studies
integrating the rates of SL and SE with respect to LULCC will have considerable signifi-
cance for policymakers and environmental practitioners to devise sustainable and targeted
watershed management strategies.

2 Materials and methods

2.1 Study area

This study was conducted in the mid- and high-mountain regions of the Tamor River basin
(TRB), one of the major sub-basins of the Koshi River (Figure 1). TRB originates from the
eastern Himalayan regions of the Kanchanjunga range and joins the main Koshi River at
Tribeni, Nepal. The study area is situated between 87.08°–88.06°E and 26.51°–27.42°N,
which drains an area of nearly 4189 km2. Its elevation ranges from 93 to 4647 m with a mean
elevation of 1814 m. The mean annual rainfall in the study area is approximately 1587 mm
yr−1. Northeasterly and southwesterly wind systems that occur in the winter and summer,
respectively, are the major sources of rainfall (Dahal et al., 2020). Summer (June–September)
is the major rainy season, providing nearly 80% of the annual rainfall (Dahal et al., 2020).
According to the FAO/UNESCO Digital Soil Map of the World, the dominant soil types in
the study area are dystric cambisols (66.8%), lithosols (23.6%), and humic acrisols (9.7%)
(FAO/UNESCO, 2007). Forest land, agricultural land, grassland, shrubland, barren land,
water, and built-up areas are the major LULC in the study area (Uddin et al., 2016). Forest
land covers the largest area, followed by agricultural land, while built-up areas constitute the
smallest share of the basin (Sinha et al., 2019). Forest, shrub, and grasslands predominantly
dominate the steep and high elevation areas of the basin, while agricultural lands and
built-up areas are concentrated in relatively less steep slopes and moderate elevation areas
along the side/banks of rivers. As part of the Himalayan region, the TRB is one of the key
SL and SE hotspot sites in the world because of its inherent unconsolidated and fragile ge-
ology, steep slopes, torrential rainfall, and intensive anthropogenic activities (Jain et al.,
2001; Golosov and Walling, 2019). This basin covers approximately 11% of the upper Koshi
1290 Journal of Geographical Sciences

River basin and contributes around 16% (16 Mt yr−1) of the total sediment load to the basin
(Sinha et al., 2019). The basin is generally the most sensitive and fragile environment. A
small disturbance to its natural components significantly affect its overall sediment dynam-
ics, making this basin the preferred choice for the current study.

Figure 1 Location of the study area (Koshi River Basin, Nepal): (a) Location and elevation of the study area,
and (b) sub-watersheds (SW) of the study. Note: Sub-watersheds in the current study were generated from a 30 m
spatial resolution DEM by considering the minimum threshold area of 50 km2 using Arc Hydro tools.

2.2 Datasets

For this study, rainfall data for a period of 30 years (1987–2017) at 16 gauge stations were
obtained from the Department of Hydrology and Meteorology (DHM), Nepal. Similarly,
monthly scale discharge and sediment concentration data from 2006 to 2013 for the valida-
tion of our SE simulation results were obtained from DHM, Nepal. In addition, a 12.5 m
spatial resolution digital elevation model (DEM) and a digital soil map (scale 1:5,000,000)
were used to generate the topographic and soil erodibility (K) factor maps of the study area,
respectively. The LULC maps of the study area for the years 1990, 2002, 2010, and 2021
were generated using Landsat 5/7/8 images from the United States Geological Survey
(USGS) archive available in the GEE cloud platform. Detailed data types and the sources
used in this study are listed in Table S1. The overall study process from data acquisition to
the final SL and SE analysis is shown in Figure 2.

2.3 Land use/cover classification and accuracy assessment

2.3.1 Land use/cover classification
The LULC maps of the study area were derived from multi-temporal satellite imageries
YIGEZ Belayneh et al.: Dynamics of soil loss and sediment export in Koshi River Basin, Nepal 1291

Figure 2 Flow chart of the study methodology

using an RF classifier in the GEE data processing platform. For this study, stacked images
(from January to December) of Landsat 5 for the years 1990 and 2010, and Landsat 7 and
Landsat 8 images for the years 2002 and 2021, respectively, were used. All of the image
processing and classification tasks of this study were carried out using the GEE cloud
1292 Journal of Geographical Sciences

computing platform (https://earthengine.google.org/). The Fmask (function of mask) algo-

rithm available on the GEE platform was applied to manage clouds and cloud shadows (Zhu
et al., 2015). Previous studies suggested that the integration of spectral indices with spectral
bands can improve the classification accuracy (Teffera et al., 2018; Magpantay et al., 2019;
Kavhu et al., 2021). Moreover, stacking topographic features with spectral bands also en-
hances the level of classification accuracy, especially in topographically complex regions
(Fahsi et al., 2000). Therefore, we used a stacked image of five spectral indices, three
topographic variables, and six spectral bands to generate the LULC maps in this study. The
enhanced vegetation index (EVI), normalized difference vegetation index (NDVI), nor-
malized difference built-up index (NDBI), modified normalized difference water index
(MNDWI), soil-adjusted vegetation index (SAVI), elevation, slope, aspect, and all of the
visible, near-infrared (NI), short-wave infrared 1(SWI1), and short-wave infrared 2 (SWI2)
bands are all included in this image. The RF classifier algorithm was used for LULC clas-
sification because of its lower sensitivity to noise, good performance with high dimensional
and multi-source datasets, good ability in handling outliers, and higher classification accu-
racy than other popular classifiers (Rodriguez-Galiano et al., 2012; Zurqani et al., 2018;
Talukdar et al., 2020; Naboureh et al., 2021).
2.3.2 Accuracy assessment
In this study, using the Collect Earth software, a total of 2,332 reference points for each of
the study years were collected from very-high-resolution images embedded in Google Earth.
During the sample collection, a minimum threshold of 50 samples for minority LULC
classes was considered. To avoid the border effect (spectral mixing) within pixels of dif-
ferent classes, a 10 m radius buffer zone from neighboring classes was designed to lay out
the sampling plot. The collected reference datasets were then partitioned into 70%/30%
proportions. Of those, 70% were used for training the classifier and the remaining 30% for
validating the classified maps. The four classified maps were compared against the reference
data and the measure of accuracies was computed in a confusion matrix (Congalton, 1991).
Since there is no single optimal measure, the user’s accuracy (UA), producer’s accuracy (PA),
overall accuracy (OA), F1 score, and Kappa statistics were used to evaluate the classification
result (Table 4). Finally, the spatiotemporal dynamics of each of the LULC classes were
analyzed.

2.4 Soil loss and sediment export modeling

In this study, the InVEST model was chosen for SL and SE estimation because it is more
user-friendly and viable for data-scarce basins than other models. It also enables the evalu-
ation of sediment export, which is often considered one of the major limitations of the Re-
vised Universal Soil Loss Equation (RUSLE) model. The InVEST model calculates the SL
and SE for a specific pixel based on the concepts of the RUSLE (Renard et al., 1997) and
hydrological connectivity index (Borselli et al., 2008) models. The model requires input
datasets of geospatial layers consisting of the DEM, LULC maps, K factor, and rainfall
erosivity (R) factor maps in a raster file format, a study area polygon, and a look-up table
containing cover management (C) and conservation practice (P) factors in a CSV file format
(Figure S1). The parameters and equations used to generate input datasets for this model are
YIGEZ Belayneh et al.: Dynamics of soil loss and sediment export in Koshi River Basin, Nepal 1293

discussed below.
2.4.1 Soil loss estimation
The mean annual SL rate in each pixel in the study area was estimated as described in
Equation (1) (Renard et al., 1997).
SL = R × K × LS × C × P (1)
−1 −1
where SL is the mean annual soil loss (t ha yr ); R is the rainfall erosivity factor (MJ mm
ha−1 h−1 yr−1); K is the soil erodibility factor (t ha h MJ−1 ha−1 mm−1); and LS, C, and P are the
dimensionless topographic, vegetation cover, and conservation practice factors, respectively.
The computation for each factor is described in the remainder of this section.
Rainfall erosivity (R) factor: The R factor describes the influence of rainfall and runoff on
the erosion phenomenon. For this study, 30-year (1987–2017) gauge rainfall data from 16
stations were used to derive the R factor map of the study area. We adopted Equation (2),
developed by Harper (1987), to estimate the average annual R factor of the study area. Then,
we applied the co-kriging interpolation techniques with 30 m resolution DEM data by taking
into account the topographic complexity of the study area and the wider applicability of the
method in complex terrains (Yan et al., 2005; Teng et al., 2019).
R 38.5 + 0.35 P
= (2)
−1 −1 −1
where R is the rainfall erosivity factor (MJ mm ha h yr ) and P is the annual rainfall
(mm).
Soil erodibility (K) factor: The K factor indicates the vulnerability of the soil to raindrop
detachment and runoff wash losses. The K factor map of the study area was calculated using
the Sharpley and Williams (1990) equations (Equations S1–S5). Detailed equations are
provided in Supplementary Text 1.
Topographic (LS) factor: The LS factor mainly refers to the influence of slope length (L)
and gradient (S) on SL and SE (Wischmeier and Smith, 1978). The length and gradient of the
slope determine the speed and volume of the surface runoff and thereby affect the rate of SL
and SE. The LS factor of the study area was derived from DEM at a spatial resolution of 12.5
m, downloaded from the https://search.asf.alaska.edu/#/1website. The detailed equation can
be found in the InVEST manual (Sharp et al., 2018).
Cover management (C) and conservation practice (P) factors: The C factor represents the
effect of vegetation cover on controlling SL, whereas the P factor reflects the ratio of SL
under a specific support practice relative to the corresponding loss without soil erosion
control practices (Teng et al., 2019). For this study, we adopted these two factors’ values
from previous literature (Wischmeier and Smith, 1978; Kim et al., 2005; Ganasri and
Ramesh, 2016; Chalise and Kumar, 2020) and we assigned values to each LULC type based
on the LULC map generated in 1990, 2002, 2010, and 2021 (Table S2).
2.4.2 Sediment export estimation
Sediment export refers to the proportion of eroded sediment reaching the nearby outlet and it
is mainly a function of SL and SDR (Hamel et al., 2015; Sharp et al., 2018). The InVEST
model calculates the SL using the RUSLE (Equation 1). Then, the model applies the sig-
moidal function developed by Vigiak et al. (2012) (Equation 3) to estimate the SDR for a
given pixel as a function of the sediment connectivity index (IC) defined by Borselli et al.
1294 Journal of Geographical Sciences

(2008).
SDRmax
SDRi = (3)
 IC − ICi 
1 + exp  o 
 Kb 
where SDR represents the sediment delivery ratio; SDRmax represents the maximum theoret-
ical SDR; and IC0 and Kb are the calibration parameters that determine the shape of the sig-
modal function.
The IC is determined by the upslope contributing area and downslope flow path to the
nearby stream based on the input DEM (https://earthexplorer.usgs.gov/). The IC in the In-
VEST model is calculated by Equations 4–6.
 Dup 
IC = log10   (4)
 Ddn 
Dup = WS Ac (5)
di
Ddn = ∑ WSi
i i
(6)

where Dup and Ddn represent the upslope and downslope components, respectively; W and
S denotes the average weighting and slope steepness factor of the upslope contributing area
(m m–1), respectively; Ac denotes the upslope contributing area (m2); di represents the length
of the flow path along the ith cell according to the steepest downslope direction (m); and Wi
and Si are the weighting factor (C factor) and the slope steepness of the ith cell, respectively.
Finally, the model generated the rate of SEi from each pixel by multiplying the results of
SLi and SDRi in the study area (Hamel et al., 2015) (Equation 7).
SE
=i SLi × SDRi (7)
where SEi and SLi represent the mean rate of sediment export and soil loss, respectively, for
a given pixel.
In this study, the key parameters required were calibrated as follows. We set the threshold
flow accumulation to 500. The second parameters that needed to be determined in this model
were the two calibration factors (IC0 and Kb). Of the two, the model is more sensitive to Kb,
which is highly dependent on landscape characteristics (Vigiak et al., 2012). The reverse is
true for IC0. Thus, a default value of 0.5 for IC0 and then the value of Kb was adjusted back
and forth from the default value of 2 until we obtained a value (Kb = 1.9) closer to the ob-
served rate of SE in the basin. As suggested by Vigiak et al. (2012), the SDRmax was set to
the default value of 0.8. Detailed explanations of the InVEST model can be found in the user
manual (Sharp et al., 2018).

2.5 Model calibration and comparison

A long-term average of the measured sediment data should be imperative to calibrate and
validate the InVEST model (Sharp et al., 2018). The model results of this study were cali-
brated using average observed sediment load data measured at the final outlet of the study
area (at Mulghat station) from 2006 to 2013. As there are no plot-based experimental studies
and long-term observed sediment data in the study area, the simulation results were validat-
YIGEZ Belayneh et al.: Dynamics of soil loss and sediment export in Koshi River Basin, Nepal 1295

ed by comparison to the results of other studies conducted in KRB using the radionuclide
techniques and RUSLE model. In this study, we adopted the approach proposed by Quilbe et
al. (2006) to manage the missing sediment load data. Based on the observed sediment load
data together with the results of previously published studies in KRB (Sinha et al., 2019;
Yigez et al., 2021; Yuan et al., 2021), it was possible to confirm that the model and parame-
ters are suitable for simulating the spatiotemporal variation of SL and SE in this case study
area. The deviation between the simulated data in this study and the reported SL and SE in
the previous studies varies from 1.67 to 46.84 t ha−1 yr−1 and 0.06 to 14.32 t ha−1 yr−1, re-
spectively (Table 1).

Table 1 Model performance compared with results of previous studies that have been conducted in the Koshi
Basin and Nepal
SL rate SE rate Difference
Study area Methods Author name
(t ha−1 yr−1) (t ha−1 yr−1) (t ha−1 yr−1)
Pakarbas catchment Radionuclide tracing
31.29 –13.92 to 7.96 Yuan et al. (2021)
(KRB) (137Cs and 210Pbex)
Khajuri stream Ghimire et al.
Plot based study 16 –29.21 to –7.33
catchment (KRB) (2013)
RUSLE and Wil- –14.21–7.67 and
Triyuga watershed
liams and Berndt 31 3.04 –4.45–0.06, re- Yigez et al. (2021)
(KRB)
(1972) SDR models spectively
The whole Nepal RUSLE model 25 –20.21–1.67 Koirala et al. (2019)
Kaligandaki River
Chinnasamy and
basin (Nepales SWAT model 17.3 9.81–14.32
Sood (2020)
Himalaya)
Jhikhu Khola water- Radionuclide tracing
70.17 24.96–46.84 Su et al. (2016)
shed (KRB) (137Cs)

Current study InVEST SDR model 23.33–45.21 7.49–2.98

2.6 Evaluating impacts of land use/cover change on soil loss and sediment export

To evaluate the spatiotemporal association between the change in SL and SE with LULCC,
we calculated the mean SL and SE change rates in each LULC type and compared them with
the corresponding LULC dynamics from 1990 to 2021 in KRB. An average of the annual
rainfall data from the year 1987 to 2017 was used as model input to isolate the impact of the
temporal variation in rainfall on SL and SE rates and independently examine the impacts of
LULCC on SL and SE dynamics (Gashaw et al., 2018). In this study, we used a correlation
analysis to present the association between simulated SL, SE, and the proportion of different
LULC classes in the study area. The zonal statistics and fishnet analysis tools in the ArcGIS
10.2 environment were used to generate the statistical samples. Finally, IBM SPSS Statistics
20 (IBM Corp., Armonk, NY, USA) was employed for the correlation analysis in this study.
In this study, the normality of the data was initially tested, and the variables that failed to
meet the normality criteria were logarithmically transformed.
2.7 Grading of soil erosion severity and slope gradients
To examine the severity of SL in the study area, the SL rate was further categorized into six
classes, as proposed by Singh et al. (1992) (Table 2). To evaluate the variability of SL along
1296 Journal of Geographical Sciences

the slope gradients in this study, we categorized the slope gradient into six steepness classes
(i.e., 0°–5°, 5°–8°, 8°–15°, 15°–25°, 25°–35°, >35°) (Wang et al., 2016; Teng et al., 2019).
The variation in SL rate analysis among these categories was computed using the zonal sta-
tistics tools in the ArcGIS environment.

Table 2 Soil loss rate, severity classes and priority classes as proposed by Singh et al. (1992)
Severity class Slight Moderate High Very high Severe Very severe
SL rate (t ha−1 yr−1) <5 5–10 10–20 20–40 40–80 >80
Priority class VI V IV III II I

3 Results
3.1 Land use/cover change detection

The LULC maps of the study area in the years 1990, 2002, 2010, and 2021 are depicted in
Figure 3. The overall accuracy and Kappa coefficient values of each of these maps range
from 90 to 93% and from 0.86 and 0.91, respectively (Tables 3 and S3). This indicated that
the classified maps had a very good agreement with the reference data (Monserud and
Leemans, 1992).

Figure 3 Land use/cover for 1990, 2002, 2010, and 2021 in Koshi River Basin, Nepal
YIGEZ Belayneh et al.: Dynamics of soil loss and sediment export in Koshi River Basin, Nepal 1297

Table 3 Confusion matrix for classified images

1990 2002 2010 2021
LULC
FS PA UA FS PA UA FS PA UA FS PA UA
FT 90 90 89 90 90 91 94 93 94 95 97 93
SH 87 79 96 88 85 91 87 83 90 81 75 88
GL 87 83 90 90 88 92 89 87 91 90 90 90
AG 91 95 88 91 94 88 92 94 90 95 95 94
BL 93 88 98 87 87 87 88 87 89 92 94 90
WT 97 97 98 93 95 90 94 97 92 96 96 96
BU 90 84 96 84 72 100 91 83 100 97 94 100
OA 90 90 91 93
KAP 0.86 0.87 0.88 0.91
Note: FT–forest land, SH–shrubland, GL–grassland, AG–agricultural land, BL–barren land, BU–built-up area,
KAP–Kappa coefficient, OA–overall accuracy, UA–user accuracy, PA–producer accuracy, and FS–F1 score

The proportion and spatiotemporal dynamics of each LULC type at the four time frames
are shown in Table 4. The change detection result depicted that the LULC in the study area
has undergone a considerable change during the last three decades (1990–2021), since nearly
21% of the study area has experienced the LULC transition (Figure 3; Table S4). The largest
area loss was detected for agricultural land. The area of forest land, shrubland, grassland, and
built-up areas showed a continuous increase during the study period at the expense of ag-
ricultural and barren lands. From 1990 to 2021, the amount of agricultural land decreased
from 28% to 18% (–432 km2), which is an average loss of approximately 14 km2 yr−1.
Conversely, the area of forest land, grassland, and shrubland increased by ~4% (173 km2),
2.4% (154 km2), and 3.7% (99 km2), respectively (Table 4). Out of the total agricultural land
loss, 69%, 19%, and 10% were converted to forest land, grassland, and shrubland, respec-
tively. The largest proportion of change was observed for built-up areas (> six-fold from the
initial state) and the smallest change was identified for forest lands (Table 4).

Table 4 Areal extent of land use/cover and proportion of land use/cover changes, average soil loss (SL),
and sediment export (SE) rate in Koshi River Basin, Nepal from 1990 to 2021
Area (km2) Area proportion ∆ (%)
LU class
1990 2002 2010 2021 1990–2002 2002–2010 2010–2021 1990–2021
AG 1176.99 973.28 892.38 744.51 –4.86 –1.93 –5.46 –10.32
BL 19.00 13.28 10.56 16.07 –0.14 –0.06 0.07 –0.07
BU 1.00 2.31 2.51 7.54 0.03 0 0.12 0.16
FT 2607.41 2702.7 2736.2 2779.96 2.28 0.8 1.84 4.12
GL 272.65 348.35 411.49 426.61 1.81 1.51 1.87 3.68
SH 100.71 136.27 124.67 199.26 0.85 –0.28 1.5 2.35
WT 11.49 13.04 10.13 14.70 0.04 –0.07 0.04 0.08
−1 −1
SL (t ha yr ) 45.21 32.18 30.31 23.33 –13.03 –1.87 –6.98 –21.88
SE (t ha−1 yr−1) 7.49 4.78 4.20 2.98 –2.71 –0.58 –1.22 –4.51
Note: FL–forest lands, SH–shrublands, GL–grasslands, AG–agricultural land, BL–barren lands, WT–water, BU–
built-up area
1298 Journal of Geographical Sciences

Compared to the three study periods (time intervals) (1990–2002, 2002–2010, and
2010–2021), the smallest and largest changes in area proportion were detected in the second
and third study periods, respectively (Table 4). During the third time interval, the extent of
grassland (+1.87%), shrubland (+1.5%), and built-up (+0.12%) areas substantially expanded
over the other two periods (Table 4). Meanwhile, agricultural land experienced a drastic loss
(–5.46%) during this time frame. The major hotspot areas of LULCC were mostly observed
in sloping areas and areas adjacent to natural forests and far from settlement centers and river
networks (Figures 3a–3d).

3.2 Variation of soil loss and sediment export from 1990 to 2021

The rate of SL and SE from 1990 to 2021 at a basin and sub-watershed (SW) scale are shown
in Figures 4 and 5 and Tables 3 and S5. Overall, the mean rates of SL and SE in the study
area steadily decreased over the study period. It can be seen that the mean SL and SE rates in
the study basin dropped by approximately 48% and 60%, respectively, between 1990 and
2021. The mean SL decreased from 45.21 t ha−1 yr−1 in 1990 to 32.18, 30.31, and 23.33 t
ha−1 yr−1 in 2002, 2010, and 2021, respectively. Similarly, the corresponding mean rate of SE
gradually decreased from 7.49 t ha−1 yr−1 in 1990 to 4.78, 4.20, and 2.98 t ha−1 yr−1 in 2002,

Figure 4 Spatial distribution of soil loss in 1990 (a), 2002 (b), 2010 (c), and 2021 (d) in Koshi River Basin,
Nepal
YIGEZ Belayneh et al.: Dynamics of soil loss and sediment export in Koshi River Basin, Nepal 1299

Figure 5 Spatial distribution of sediment export in 1990 (a), 2002 (b), 2010 (c), and 2021 (d) in Koshi River
Basin, Nepal

2010, and 2021, respectively. As a result, the total SL in the study area dropped from
18.94×106 t in the base year of 1990 to 13.48×106, 12.69×106, and 97.73×106 t in 2002, 2010,
and 2021, respectively, which is equivalent to an average decline of 2.9×105 t yr−1. Likewise,
the corresponding total SE decreased from 3.14× 106 in 1990 to 2.0× 106, 1.76 × 106, and
1.25 ×106 t yr−1 in 2002, 2010, and 2021, respectively.
At the SW scale, the severe SL and SE hotspots were concentrated in the middle and
mid-western parts of the river basin. The maximum mean SL rates were all observed in SW5
in the years 1990, 2002, 2010, and 2021 and their SL rates were 88.66, 62.13, 55.03, and
52.72 (t ha−1 yr−1), respectively. The corresponding maximum SE rates in 1990, 2002, 2010,
and 2021 were observed in SW5, SW5, SW14, and SW5, respectively, while the minimum
rate was observed in SW7, SW2, SW7, and SW18, respectively (Table S5). These maximum
SL and SE hotspot SWs were all located in the middle part of the study area, as they were
largely dominated by sloping farmlands (~40%, 58%, 30%, and 28%, respectively) and less
vegetation coverage (Figure 2).
1300 Journal of Geographical Sciences

Table 5 Mean (t ha−1 yr−1) and total (*104 t ha−1 yr−1) rate of soil loss and sediment export for different land
use/cover from 1990 to 2021 in Koshi River Basin, Nepal
Outputs Year Forestland Shrubland Grassland Agricultural land Barren land Built-up land
1990 10.24 16.43 18.99 130.77 75.45 76.35
2002 7.81 11.09 18.31 107.41 78.28 10.67
Mean SL
2010 7.44 11.35 18.53 107.14 69.42 13.60
2021 6.36 9.74 12.92 95.20 68.81 12.36
1990 1.43 2.79 2.83 22.25 16.16 17.18
2002 0.98 1.66 2.36 16.52 14.91 1.36
Mean SE
2010 0.90 1.59 2.30 15.47 13.35 1.57
2021 0.71 1.20 1.39 12.65 10.68 1.30
1990 267.04 16.54 51.76 1539.10 14.33 0.76
2002 211.16 15.11 63.79 1045.45 10.40 0.25
Total SL
2010 203.68 14.15 76.24 956.11 7.33 0.34
2021 176.80 19.41 55.10 708.79 11.06 0.93
1990 37.23 2.81 7.72 261.86 3.07 0.17
2002 26.61 2.26 8.21 160.80 1.98 0.03
Total SE
2010 24.56 1.98 9.48 138.03 1.41 0.04
2021 19.81 2.39 5.94 94.17 1.72 0.10

The study also presented a temporal variation in SL and SE rates among the severity
classes (Table 1 and Figure 6). During the study period, the proportion of slight, moderate,
and high SL severity classes steadily increased (by 8.22%, 1.28%, and 0.56%, respectively)
(Figures 6a and 6c, Table 6) The remaining three classes (very high, severe, and very severe)
gradually decreased by 0.43%, 1.96%, and 7.67% for SL and 2.98%, 1.98%, and 1.09% for
SE from 1990 to 2021 (Figure 6, Tables 4 and S6). More specifically, the two dominant
classes, slight and very severe, experienced the largest areal variation (an increase of 8.22%
and a decrease of 7.67%, respectively) (Figure 6a and Table 6). It can be seen from Figure 6
that approximately 31% of the study area in the base year of 1990 was found under the cat-
egory of high to very severe SL rate. However, the areal proportion under this category
dropped to 21% in 2021. These intense SL severity classes were predominantly concentrated
in the middle and mid-western parts of the study basin comprising SW5, SW8, SW12,
SW13, SW14, SW16, and SW17 (Table S5 and Figures 4a–4d). Despite these classes cov-
ering a small percentage of the basin area, their contribution to the total basin SL was very
large (> 88%). Conversely, the north and northeastern parts of the study basin (SW–13,
SW6–7, and SW9–10) had relatively low rates of severity (Table S5). Since SE is strongly
determined by the SL rate, the spatial dynamics of SE in this basin generally showed a sim-
ilar spatial distribution pattern to the SL rate (Figures 4 and 5a–5d).

3.3 Land use/cove change and its relationship with the change in soil loss and
sediment export

During the last three decades (1990–2021), remarkable changes in LULC were experienced
in the study area (Figures 3a–3d). Along with these changes, considerable dynamics in the
YIGEZ Belayneh et al.: Dynamics of soil loss and sediment export in Koshi River Basin, Nepal 1301

Figure 6 Area proportion of soil loss (a), sediment export (b), and the percentage contribution of each severity
class to the basin soil loss (c) and sediment export (d) in 1990, 2002, 2010, and 2021

Table 6 Soil loss severity class transition matrix from 1990 to 2021
2021
Year
Severity class Slight Moderate High Very high Severe Very severe Total (km2) (%)
Slight 2378.45 48.15 30.01 21.13 36.15 74.69 2588.57 62.00
Moderate 31.41 266.60 6.70 7.96 0.07 5.79 318.53 7.63
High 23.63 3.64 140.11 2.72 2.16 1.45 173.71 4.16
1990 Very high 48.54 3.10 1.89 111.92 2.60 0.84 168.89 4.05
Severe 120.48 0.05 0.87 2.81 133.70 4.17 262.07 6.28
Very severe 329.29 50.37 17.43 4.42 5.85 255.81 663.17 15.88
Total (km2) 2931.79 371.92 197.02 150.95 180.53 342.74 4174.94 100.00
Total (%) 70.22 8.91 4.72 3.62 4.32 8.21 100.00

rates of SL and SE were also exhibited in this basin (Figures 4 and 5a–5d). Table 5 summa-
rizes the rate of SL and SE from each LULC type as a function of time. We found that the
mean SL and SE rates varied with LULC types in decreasing order from agriculture > barren
land > built-up > grassland > shrubland > forest land over the study periods. The change in
SL and SE rates in the study area was principally linked with the dynamics in the proportion
of the two dominant LULC (agricultural and forest lands) (Figure 3 and Table 5). The SL and
1302 Journal of Geographical Sciences

SE rates detected in agricultural land were much higher than that of all other LULC, ranging
from 95 to 130.77 and 12.65 to 22.25 (t ha−1 yr−1), respectively (Table 5). This rate is more
than six times larger than the maximum SL tolerance rate of the Himalayan region (15 t ha−1
yr−1) and four times larger than that of the mean SL rate estimated in the current study (22.22
t ha−1 yr−1) for the study area. On the contrary, the minimum SL and SE rates were observed
in forest land, varying from 6.36 to 10.24 t ha−1 yr−1 and 0.57 to 0.75 t ha−1 yr−1, respectively.

Table 7 Pearson correlation coefficients for the relationship between land use proportion and the average
values of soil loss and sediment export in all sub-watersheds from 1990 to 2021
1990 2002 2010 2021
LULC type
SL SE SL SE SL SE SL SE
FL –0.45* –0.51* –0.39 –0.47* –0.55** –0.57** –0.21 –0.32
SH –0.18 –0.24 –0.28 –0.36 –0.33 –0.33 –0.24 –0.31
GL –0.06 –0.09 –0.23 –0.33 –0.46* –0.47* –0.20 –0.30
AG 0.64** 0.63** 0.69** 0.77** 0.68** 0.70** 0.71** 0.75**
BL 0.15 0.15 0.12 0.17 0.04 0.08 0.02 0.01
WT 0.18 0.16 0.21 0.23 0.07 0.07 0.31 0.25
BU –0.08 –0.05 0.29 0.18 –0.08 –0.07 0.36 0.36
Note: *p<0.05, ** p<0.01; FL–forest lands, SH–shrublands, GL–grasslands, AG–agricultural land, BL–barren lands,
WT–water, BU–built-up area

The correlation analysis also showed a clear linkage between LULCC with SL and SE
dynamics in the study area (Table 7). It can be seen that SL and SE showed a significant
(p<0.05) negative correlation with the proportion of forest land with R2 values ranging from
–0.45 to –0.55, and –0.47 to –0.57, respectively, from 1990 to 2010. On the other hand, the
SL and SE rates had a significant (p<0.01) positive correlation with the proportion of agri-
cultural land change with R2 values ranging from 0.64 to 0.71 and 0.63 to 0.77, respectively
(Table 7). Moreover, the relationship between grassland with SL and SE was significant
(p<0.05) and moderately correlated in the year 2010. Generally, this analysis suggested that
SWs dominated with a high proportion of vegetated LU types were less subjected to SL and
SE, and vice versa.

3.4 Variation of soil loss and sediment export across different slope aspects and
gradients

The spatial distribution of SL and SE for the years 1990, 2002, 2010, and 2021 across topo-
graphic factors was evaluated by overlaying the spatial distribution map of SL and SE with
the slope aspect and slope gradient maps of the study area (Table 8; Figure7a and 7b). The
mean SL and SE in different slope gradients generally showed an increasing trend with in-
creasing slope gradients up to 25°, and then gradually decreased with increasing slope gra-
dients beyond this gradient in the study area (Table 8). The mean SE rate showed a decreas-
ing trend from 1990 to 2021 across all slope classes. Concurrently, the mean SL rate also
showed a decreasing trend from 1990 to 2021 under slope gradient classes beyond 25°. In
the current study, SL and SE also varied with slope aspects (Figures 7a and b). Sun-facing
aspects, which receive better solar radiation (SE, S, E, and SW), showed higher SL and SE
YIGEZ Belayneh et al.: Dynamics of soil loss and sediment export in Koshi River Basin, Nepal 1303

rates than shady slopes in the study basin. On the contrary, shady slopes comprising flat, NW,
and N aspect slopes exhibited the lowest rates of SL and SE in the study area. We also found
that the SL and SE rates under sun-facing and semi-sunny slopes experienced a decreasing
trend from the base year of 1990 to the final year of 2021. A slope gradient from 8°–35° and
sunny and semi-sunny slope aspects including S, SE, E, and SW accounted for nearly 91%
and 57%, respectively, of the total SL and SE in the final year of this study.

4 Discussion
4.1 Impact of land use/cover change on soil loss and sediment export

In environmentally fragile regions, LULCC has a significant impact on sediment flux (Bak-
ker et al., 2008; Xu et al., 2011). This study investigated the responses of SL and SE to
LULCC in the study area of Nepal for a time window of 32 years (1990–2021). The results
of the current study showed a substantial decrease in the average rates of SL (48%) and SE
(60%) from 1990 to 2021. Concomitantly, the corresponding total SL and SE decreased by
91.66×105 and 18.89×105 t, respectively. The study suggested that there has been a consid-
erable shift in SL severity classes (about 10% of the high to very severe classes transforming
to moderate to slight classes) over the study period in the study area. The significant drop in
SL and SE rates was attributed to the continuous conversion of intensified (agricultural land)
to de-intensified (forest, grass, and shrublands) LULC types associated with the implement-
ed government and community-based forest development, protection, and other soil and
water conservation interventions in the late 1980s and early 1990s (such as the Master Plan

Table 8 Soil loss and sediment export on different slope gradients in 1990, 2002, 2010, and 2021 in Koshi
River Basin, Nepal

Slope Area MSL TSL

(°) (km2) 1990 2002 2010 2021 1990 2002 2010 2021
0–5 5.94 5.78 4.72 4.82 5.10 0.34 0.27 0.23 0.25
5–8 9.35 16.49 13.38 13.44 13.72 1.54 2.21 1.80 1.84
8–15 58.62 43.83 34.23 34.44 33.21 25.70 15.00 11.79 11.44
15–25 224.57 61.13 46.39 46.30 38.37 137.29 28.36 21.48 17.77
25–35 249.02 43.75 30.43 26.13 18.57 108.94 13.31 7.95 4.85
>35 177.34 30.40 17.89 16.32 8.81 53.91 5.44 2.92 1.44

Slope Area MSE TSE

(°) (km2) 1990 2002 2010 2021 1990 2002 2010 2021
0–5 5.94 0.99 0.76 0.73 0.72 0.06 0.01 0.01 0.01
5–8 9.35 2.79 2.08 1.99 1.89 0.26 0.06 0.04 0.04
8–15 58.62 7.02 4.96 4.83 4.34 4.11 0.35 0.24 0.21
15-25 224.57 10.04 6.84 6.51 5.00 22.53 0.69 0.44 0.33
25–35 249.02 7.40 4.63 3.64 2.36 18.41 0.34 0.17 0.09
>35 177.34 5.01 2.61 2.10 0.97 8.88 0.13 0.05 0.02
−1 −1 4 −1
Note: MSL–Mean soil loss (t ha yr ), TSL–Total soil loss (10 t), MSE–Mean sediment export (t ha yr−1),
TSE–Total sediment export (104 t)
1304 Journal of Geographical Sciences

Figure 7 Soil loss in different slope aspects (a) and sediment export in different slope aspects (b) in 1990,
2002, 2010, and 2021 in Koshi River Basin, Nepal
(N–north, NE–northeast, E–east, SE–southeast, S–south, SW–southwest, W–west, NW–northwest)

for the Forestry Sector in 1988 and the Forest Protection Law in 1993) in the study area
(Gautam et al., 2004; Li et al., 2017; Chalise et al., 2019). According to Gautam et al.
(2004), the formalization of local forestry institutions by forest user groups after the imple-
mentation of the community forestry program also contributed positively to the increase in
vegetation cover. In addition to the above factors, migration-triggered farmland abandon-
ment and associated stimulation in LULCC also had positive implications on the decline in
SL and SE in the study area. During the study period, other LULC classes also showed both
gradual declining (in barren land) and increasing (in built-up and water area) trends, but they
had no significant effect on the SL and SE rates because of their small area.
Although the mean rate of SL in this study showed a continuous decreasing trend from
1990 to 2021, the mean SL rates of all four study years (about 21% of the study area in 2021)
YIGEZ Belayneh et al.: Dynamics of soil loss and sediment export in Koshi River Basin, Nepal 1305

are already beyond the maximum SL tolerance of 15 t ha−1 yr−1 in the Himalayan region
(Jasrotia and Singh, 2006) and almost double the global tolerance rate of 10 Mg ha−1 yr−1
(Borrelli et al., 2017). The estimated rate of SL in the study area is in line with other
sub-basin and watershed-scale study results; however, it is higher (nearly 89%) than previ-
ous results reported by Uddin et al. (2016) in the entire KRB, as this study utilized coarse
resolution DEM (90 m) and incorporated the Terai region of KRB (an area of large flat to-
pography), which led to the low rates of SL and SE. Moreover, the different models used for
estimation might be the other possible reason for the variation.
The type of land use and management practice applied to a specific watershed strongly
determines the variability in vegetation cover and, thereby, the process of SL and SE (Zhou
et al., 2019). LULC with higher vegetation cover has higher soil conservation services
(Bakker et al., 2008). For example, a study by Zuo et al. (2016) found that reforestation de-
creased the rate of SE by more than 40% in the Huangfuchuan watershed, China. Similarly,
vegetation coverage greater than 60% has the potential to reduce annual SE by up to 60%
(Chen et al., 2021). In the current study, the strongest soil conservation function was ob-
served in forest land use followed by shrubland and grassland, since these LULC categories
can provide better ground cover against erosive agents than agricultural and barren lands.
The intensity of SL and SE was considerably higher in the middle and mid-western parts
of the study area in all four study years, since these areas are dominated by a high proportion
of sloping farmlands and other LULC with low vegetation coverage. Conversely, a lower
intensity of SL and SE was observed in the north and northeastern parts of the study area. In
addition to agricultural land proportion, the inherent lower inter-annual ground/vegetation
cover variability of forest, grass, and shrubland might also be another possible reason for the
relatively lower rates of SL and SE in the north and northeastern part of the study area, de-
spite the topographic and K factor potential (steep slope and high elevation) of these regions
being very prone to soil erosion.
In this study, the variability of SL, SE, and LULCC among the three study periods was
also evaluated. Accordingly, the period from 1990 to 2002 was highly characterized by the
largest drop in SL and SE over the other two periods. During this period, the extent of agri-
cultural land drastically decreased (a drop of nearly 33.32%) at the expense of forest, grass-
land, and shrubland expansion. Concurrently, the total amount of SL and SE substantially
decreased by approximately 32% and 39%, respectively (Table 5). Comparable results have
also been reported in other studies in different regions (Bakker et al., 2008; Sun et al., 2013;
Yan et al., 2018; Teng et al., 2019; Zhou et al., 2019). The LULCC and corresponding de-
crease in SL and SE during this period were mostly attributed to the implementation of dif-
ferent forest, soil, and water conservation interventions such as the community, leasehold,
and private forestry programs introduced in the late 1980s and early 1990s. Reduced pres-
sure on the natural landscape encourages natural regeneration, which, in turn, enhances the
conversion of some shrublands into mature forest trees (Bakker et al., 2008; Sun et al., 2013;
Li et al., 2017). In this regard, the application of improved agricultural technologies and
practices that enable farmers to produce high yields (such as agroforestry practices, the
promotion of improved seed/vegetative varieties, the expansion of irrigation systems, and
pesticide and fertilizer application) in recent decades could also be another reason for the
increase in vegetation cover, resulting in a decrease in SL and SE in the study basin (Gilani
1306 Journal of Geographical Sciences

et al., 2017; Li et al., 2017). Comparatively, the magnitude of all three variables (SL, SE,
and LULCC) in the period from 2002 to 2010 was lower than both the first and third study
periods. The decrease in mean SL and SE was only 1.87% and 0.58%, respectively (Table 4).
Differing from the other periods, the extent of shrubland in this period decreased by 0.28%
compared with that of 2002. This could be due to the shift in human pressure from forest
land to shrubland. The major sources of construction and fuelwood for rural communities in
developing countries including Nepal are mostly forests. However, forest resources are now
under better management conditions since the launch of the community forestry program (in
the early 1990s) in Nepal, and thus, the pressure on forests for the abovementioned con-
sumptions have drastically decreased. This is likely to have resulted in a shift in pressure
from forest land to shrubland to compensate for the demand for wood obtained from com-
munity forest resources. This might lead to the selective clearing of shrubs and scant tree
cover in this land use type, leading to areas with no shrubs that are dominated by grass cover.
This could be one reason for the change in shrubland to grassland and the decrease in its
coverage in the second period of this study. However, the intensity of LULCC from 2010 to
2021 was more substantial than both of the previous two periods. This might be due to not
only the implementation of the national forest conservation measures (community forestry,
afforestation, and reforestation), but also the other socioeconomic drivers such as the in-
creasing rate of internal migration between 2010 and 2021 (Bhawana et al., 2017; Chidi,
2017; Paudel et al., 2020). In recent decades, there has been a high outflow of the active
labor force from the fragile hill and mountain regions to plains and urban areas of the coun-
try in search of better infrastructure, fertile land, and access to off-farm income activities, etc.
(Maharjan et al., 2020; Paudel et al., 2020). This high rate of outmigration creates a consid-
erable shortage in terms of an active labor force for on-farm activities and, in turn, leads to
an increased de-intensification and abandonment of agricultural land. The subsequent aban-
donment of farmland in marginal areas of the study basin has enhanced vegetation succes-
sion, replacing historically highly erosive cereal fields with grasses, dense shrubs, and forest
communities, thereby reducing SL and SE. In addition to natural vegetation colonization,
farmland abandonment in the study area also encourages farmers to grow trees such as uttis
(Alnus nepalensis), chiraito (Swertia chirayita), loth salla (Taxus wallichiana), and carda-
mom (Elettaria cardamomumon) on these abandoned lands, thereby increasing vegetation
cover and decreasing the rate of SL and SE in recent years (Pandey et al., 2016). Similar
experiences of landscape transitions following depopulation/migration and a reduction in the
rate of SL and SE have been reported in other parts of the world (Li et al., 2016; Bhawana et
al., 2017; Kolecka et al., 2017; Xu et al., 2019).
Although the LULCC from 2010 to 2021 exhibited a relatively higher magnitude than the
previous two periods, the changes in SL and SE rates from 2010 to 2021 were not as strong
as in the first interval. This could be due to the concentration of the changing area in less
SL- and SE-sensitive areas (steep slope, high rainfall, or areas with high erodible soils)
compared to the previous two periods, since the severely eroded areas have already been
changed to less intensified land use types (better vegetation coverage) during the first and
second study periods. Moreover, the lower SL and SE rates exhibited during the third inter-
val were attributed to the high change area proportion of non-vegetated LULC types (agri-
cultural land) to less erosion-sensitive vegetated classes compared to the other two periods.
YIGEZ Belayneh et al.: Dynamics of soil loss and sediment export in Koshi River Basin, Nepal 1307

Comparable results have also been reported in previous studies (Bakker et al., 2008; Zhu et
al., 2018; Chen et al., 2021) stating that de-intensification (such as afforestation and aban-
donment) leads to a more favorable landscape pattern with respect to the reduction of SL and
SE. On top of the above rationale, the plot size of the changed area also has a significant
impact on the rate of SL and SE by influencing runoff and sediment connectivity (Wang et
al., 2018; Fang, 2020). Due to the rugged nature of the study area, most of the agricultural
plots that were abandoned during the third interval were mostly concentrated in high eleva-
tion and steep areas and they were small in size and fragmented. Thus, the response in SL
and SE might not be as significant as the change in large-sized agricultural land plots.

4.2 Variation of soil loss and sediment export across different slope aspects and slope
gradients

In the study area, SL and SE varied with different slope aspects and gradients. Generally, the
mean rates of SL and SE gradually increased with increasing slope gradients up to 25° and
then decreased beyond this gradient. As can be seen from the spatial distribution LULC
maps (Figure 3) in this study, the decreasing trend in the rate of SL and SE under slope gra-
dients >25° might be due to the presence of high vegetation coverage in steep slope areas,
since steep slope areas are mostly recommended for the use of forest, shrubland, or grass-
land in the study area. This result suggested that the increase in vegetation recov-
ery/coverage can effectively weaken the impact of slope gradient on SL and SE rates. In ad-
dition to the observed variation in SL and SE across slope gradients, they also varied within
the same slope gradient classes during the study periods, with a general decreasing trend
from 1990 to 2021. This might be attributed to the improvement and expansion of vegetation
coverage in steep and high-elevation areas associated with the recently implemented con-
servation measures. The slope aspect also showed a clear variation in mean SL and SE rates
in this basin. High SL and SE were observed in S-, SE-, E-, and SW-facing slopes than in the
north-facing counterparts. The slope aspect controls the solar angle and determines the rate
of SL and SE through changing the local climate, and consequently impacts vegetation cov-
erage and human activities (Chidi, 2017). Most crops need sufficient sunlight for proper
growth, flowering, and fruiting, especially in areas where there is high rainfall; thus,
sun-facing slopes, mostly south-facing in Nepal, are more suitable for most crop types than
north-facing slopes (Chidi, 2017). This indicates that south-facing slopes are highly sub-
jected to intensive human activities such as agricultural practices. This could be the reason
for the high rate of SL and SE in sunny slopes in the study area. On the contrary, shady
slopes are more suitable for non-farming practices. Chidi (2017) and Yan et al. (2018) found
that the rate of farmland abandonment and vegetation recovery is higher on shady slopes
than on sunny slopes. The result thus suggests that the ecological restoration and conserva-
tion measures should be continued and should focus more on sunny slopes, since this slope
aspect is one of the hotspot areas of SL and SE in this basin.

4.3 Limitations and future research directions

Sediment sources in most mountainous basins such as the study area are very complex. Un-
managed road construction, gullies, landslides, and debris flow in mountainous basins con-
1308 Journal of Geographical Sciences

tribute a significant amount of sediment load to the river channel. Thus, the sediment con-
tribution from these sources should be accurately quantified using other modeling ap-
proaches in the future for better planning and management interventions. The temporal dy-
namics of structural soil and water conservation measures (P factor) play a major role in re-
ducing SL and SE (Chen et al., 2021); however, they were not taken to account in this study
due to reliable data limitations. Remote sensing data may not effectively detect existing tree
crop areas, which have widely expanded in recent decades in the study site, such as tea
plantations and extensive cardamom (Elettaria cardamomumon) and chiraito (Swertia chi-
rayita) farming. Therefore, rigorous field verification and secondary data at the local level
will be very helpful in addressing this problem. Although this study has the abovementioned
limitations, they do not influence the general trend of the results and the conclusions drawn.

5 Conclusions
The contributions of LULCC to the changes in SL and SE in the far eastern mountainous
regions of KRB, Nepal, from 1990 to 2021 were evaluated using a combination of the In-
VEST SDR model and the RF classifier algorithm in the GEE cloud platform. The study
highlighted that the increased vegetation cover associated with the rapid expansion of forest,
shrubs, and grasslands at the expense of agricultural land reduced the rates of SL and SE
throughout the study period. Concurrently, the mean rates of SL and SE in the study area fell
drastically by about 48% and 60%, respectively, over the last three decades. The rates of SL
and SE in the study area showed a noticeable spatiotemporal variation, with higher rates in
1990 and lower rates in 2021. Spatially, higher SL and SE rates were observed in the middle
and mid-western parts of the study area, while lower rates were observed in the north and
northeastern parts. The variations were principally linked with the dynamics in the propor-
tion of the two dominant LULC, i.e., agricultural and forest lands. Furthermore, the rates of
SL and SE varied significantly under different slope gradients, slope aspects, and
sub-watersheds. Generally, the slopes with gradients >8°, sunny and semi-sunny aspects, and
sub-watersheds dominated by a large proportion of agricultural land (low vegetation cover)
exhibited much higher SL and SE rates. However, with the vegetation restoration, the im-
pacts of topographic factors on SL and SE were effectively weakened.
Although the overall rates of SL and SE decreased dramatically during the study period,
the mean SL rate in the final year (2021) of the study period was still far beyond both the
global and Himalayan region SL tolerance rates. Therefore, we also suggest that more veg-
etative conservation measures should be reinforced in the severe SL and SE hotspot areas,
such as the middle and mid-western parts of the basin, areas dominated by a large proportion
of agricultural lands, south-facing slope aspects, and slope gradients between 8 and 35⁰. This
research provides useful information in supporting land-use planners, restoration managers,
and policymakers to set conservation priorities and make informed decisions for sustainable
basin management interventions.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal rela-
tionships that could have appeared to influence the work reported in this paper.
YIGEZ Belayneh et al.: Dynamics of soil loss and sediment export in Koshi River Basin, Nepal 1309

Acknowledgments

We thank QIN Xiaomin (a PhD student from the Institute of Mountain Hazards and Envi-
ronment, Chinese Academy of Sciences) for her kind help during the revision work.

References
Aneseyee A B, Elias E, Soromessa T et al., 2020. Land use/land cover change effect on soil erosion and sediment
delivery in the Winike watershed, Omo Gibe Basin, Ethiopia. Science of the Total Environment, 728: 138776.
https://doi.org/10.1016/j.scitotenv.2020.138776.
Bakker M M, Govers G, van Doorn A et al., 2008. The response of soil erosion and sediment export to land–use
change in four areas of Europe: The importance of landscape pattern. Geomorphology, 98: 213–226.
https://doi.org/10.1016/j.geomorph.2006.12.027.
Bhawana K C, Wang T, Gentle P, 2017. Internal migration and land use and land cover changes in the Middle
Mountains of Nepal. Mountain Research and Development, 37(4): 446–455. https://doi.org/10.1659/MRD-
JOURNAL-D-17-00027.1.
Borrelli P, Robinson D A, Fleischer L R et al., 2017. An assessment of the global impact of 21st century land use
change on soil erosion. Nature Communications, 8(1). https://doi.org/10.1038/s41467-017-02142-7.
Borrelli P, Robinson D A, Panagos P et al., 2020. Land use and climate change impacts on global soil erosion by
water (2015–2070). PNAS, 117(36). https://doi.org/10.1073/pnas.2001403117.
Borselli L, Cassi P, Torri D, 2008. Prolegomena to sediment and flow connectivity in the landscape: A GIS and
field numerical assessment. Catena, 75(3): 268–277. https://doi.org/10.1016/j.catena.2008.07.006.
Chalise D, Kumar L, 2020. Land use change affects water erosion in the Nepal Himalayas. PLoS One, 15(4).
https://doi.org/10.1371/journal.pone.0231692.
Chalise D, Kumar L, Kristiansen P, 2019. Land degradation by soil erosion in Nepal: A review. Soil Systems, 3(1):
1–18. https://doi.org/10.3390/soilsystems3010012.
Chapagain N, Banjade M R, 2009. Community forestry and local development: Experiences from the Koshi Hills
of Nepal. Journal of Forest and Livelihood, 8(2): 78–92. https://doi.org/10.3126/jfl.v8i2.2310.
Chen J, Li Z, Xiao H et al., 2021. Effects of land use and land cover on soil erosion control in southern China:
Implications from a systematic quantitative review. Journal of Environmental Management, 282: 111924.
https://doi.org/10.1016/j.jenvman.2020.111924.
Chidi C L, 2017. Patch analysis of cultivated land abandonment in the Hills of Western Nepal. In: Li A, Deng W,
Zhao W eds. Land Cover Change and Its Eco-environmental Responses in Nepal. Singapore: Springer,
149–162.
Chinnasamy P, Sood A, 2020. Estimation of sediment load for Himalayan Rivers: Case study of Kaligandaki in
Nepal. Journal of Earth System Science, 129: 181.https://doi.org/10.1007/s12040-020-01437-6.
Congalton R G, 1991 A review of assessing the accuracy of classifications of remotely sensed data. Remote Sens-
ing of Environment, 37(1): 35–46. https://doi.org/10.1016/0034-4257(91)90048-B.
Dahal N M, Donghong X, Neupane N et al., 2020. Spatiotemporal analysis of drought variability based on the
standardized precipitation evapotranspiration index in the Koshi River basin , Nepal. Journal of Arid Land, 13:
433–454. https://doi.org/10.3390/agronomy11091691.
Fahsi A, Tsegaye T, Tadesse W et al., 2000. Incorporation of digital elevation models with Landsat-TM data to
improve land cover classification accuracy. Forest Ecology and Management, 128(1/2): 57–64. https://doi.org/
10.1016/S0378-1127(99)00272-8.
Fang H, 2020. Impact of land use changes on catchment soil erosion and sediment yield in the northeastern China:
A panel data model application. International Journal of Sediment Research, 35(5): 540–549. https://doi.org/
10.1016/j.ijsrc.2020.03.017.
FAO/UNESCO, 2007. Digital Soil Map of the World (DSMW), (version 3.6). Rome, Italy.
Ganasri B P, Ramesh H, 2016. Assessment of soil erosion by RUSLE model using remote sensing and GIS: A case
1310 Journal of Geographical Sciences

study of Nethravathi Basin. Geoscience Frontiers, 7(6): 953–961. https://doi.org/10.1016/j.gsf.2015.10.007.

Gashaw T, Tulu T, Argaw M et al., 2018. Modeling the hydrological impacts of land use/land cover changes in the
Andassa watershed, Blue Nile Basin, Ethiopia. Science of the Total Environment, 619/620: 1394–1408.
https://doi.org/10.1016/j.scitotenv.2017.11.191.
Gautam A P, Shivakoti G P, Webb E L, 2004. A review of forest policies, institutions, and changes in the resource
condition in Nepal. International Forestry Review, 6(2): 136–148. https://doi.org/10.1505/ifor.6.2.136.38397.
Gilani H, Qamer F M, Sohail M et al., 2017. Review of ecosystem monitoring in Nepal and evolving earth ob-
servation technologies. In: Li A, Deng W, Zhao W eds. Land Cover Change and Its Eco-environmental Re-
sponses in Nepal. Singapore: Springer, 165–183.
Golosov V N, Walling D E, 2019. Erosion and sediment problems: Global hotspots. Paris, France.
Hamel P, Chaplin-Kramer R, Sim S et al., 2015. A new approach to modeling the sediment retention service (In-
VEST 3.0): Case study of the Cape Fear catchment, North Carolina, USA. Science of the Total Environment,
524/525: 166–177. https://doi.org/10.1016/j.scitotenv.2015.04.027.
Harper D, 1987. Improving the accuracy of the universal soil loss equation in Thailand. In: Proceedings of the
Fifth International Conservation Conferences, Bangkok, Thailand.
ICIMOD, 2018. Understanding Sediment Management. Kathmandu, Nepal.
Jain S K, Kumar S, Varghese J, 2001. Estimation of soil erosion for a Himalayan Watershed using GIS technique.
Water Resources Management, 15: 41–54. https://doi.org/10.1023/A:1012246029263.
Jasrotia A S, Singh R, 2006. Modeling runoff and soil erosion in a catchment area, using the GIS, in the Himala-
yan region. Environmental Geology, 51(1): 29–37. https://doi.org/10.1007/s00254-006-0301-6.
Kavhu B, Mashimbye Z E, Luvuno L, 2021. Climate-based regionalization and inclusion of spectral indices for
enhancing transboundary land-use/cover classification using deep learning and machine learning. Remote
Sensing, 13(24). https://doi.org/10.3390/rs13245054.
Kim J B, Saunders P, Finn J T, 2005. Rapid assessment of soil erosion in the Rio Lempa Basin, Central America,
using the universal soil loss equation and geographic information systems. Environmental Management, 36(6):
872–885. https://doi.org/10.1007/s00267-002-0065-z.
Koirala P, Thakuri S, Joshi S et al., 2019. Estimation of soil erosion in Nepal using a RUSLE modeling and geo-
spatial tool. Geosciences, 9: 147. doi: 10.3390/geosciences9040147.
Kolecka N, Kozak J, Kaim D et al., 2017. Understanding farmland abandonment in the Polish Carpathians. Ap-
plied Geography, 88: 62–72. https://doi.org/10.1016/j.apgeog.2017.09.002.
Latocha A, Szymanowski M, Jeziorska J et al., 2016. Effects of land abandonment and climate change on soil
erosion-An example from depopulated agricultural lands in the Sudetes Mts., SW Poland. Catena, 145:
128–141. https://doi.org/10.1016/j.catena.2016.05.027.
Li A, Lei G, Cao X et al., 2017. Land cover change and its driving forces in Nepal since 1990. In: Li A, Deng W,
Zhao W eds. Land Cover Change and Its Eco-environmental Responses in Nepal. Singapore: Springer, 41–65.
Li S, Sun Z, Tan M et al., 2016. Effects of rural–urban migration on vegetation greenness in fragile areas: A case
study of Inner Mongolia in China. Journal of Geographical Sciences, 26(3): 313–324. https://doi.org/10.1007/
s11442-016-1270-7.
Magpantay A T, Adao R T, Bombasi J L et al., 2019. Analysis on the effect of spectral index images on improve-
ment of classification accuracy of Landsat-8 OLI image. Korean Journal of Remote Sensing, 35(4): 561–571.
https://doi.org/10.7780/kjrs.2019.35.4.6.
Maharjan A, Kochhar I, Chitale V S et al., 2020. Understanding rural outmigration and agricultural land use
change in the Gandaki Basin, Nepal. Appllied Geography 124:102278. https://doi.org/10.1016/j.apgeog.
2020.102278.
Monserud R A, Leemans R, 1992. Comparing global vegetation maps with the Kappa statistic. Ecological Model-
ling, 62(4): 275–293. https://doi.org/10.1016/0304-3800(92)90003-W.
Naboureh A, Li A, Ebrahimy H et al., 2021. Assessing the effects of irrigated agricultural expansions on Lake
Urmia using multi-decadal Landsat imagery and a sample migration technique within Google Earth Engine.
International Journal of Applied Earth Observation and Geoinformation, 105: 102607. https://doi.org/
YIGEZ Belayneh et al.: Dynamics of soil loss and sediment export in Koshi River Basin, Nepal 1311

10.1016/j.jag.2021.102607.
Pandey D, Heyojoo B P, Shahi H, 2016. Drivers and dynamics of land use land cover in Ambung VDC of Teh-
rathum district, Nepal. Banko Janakari, 26(1): 90–96. https://doi.org/10.3126/banko.v26i1.15508.
Paudel B, Wu X, Zhang Y et al., 2020. Farmland abandonment and its determinants in the different ecological
villages of the Koshi river basin, central Himalayas: Synergy of high-resolution remote sensing and social
surveys. Environmental Research, 188: 109711. https://doi.org/10.1016/j.envres.2020.109711.
Quilbe R, Rousseau A N, Duchemin M et al., 2006. Selecting a calculation method to estimate sediment and nu-
trient loads in streams : Application to the Beaurivage River. Journal of Hydrology, 326: 295–310.
https://doi.org/10.1016/j.jhydrol.2005.11.008.
Renard K G, Foster G A, Weesies D K et al., 1997. Predicting Soil Erosion by Water: A Guide to Conservation
Planning with the Revised Universal Soil Loss Equation (RUSLE). U.S. Departement of Agriculture, Agricul-
ture Handbook No.703, pp. 404. United States Department of Agriculture Research Service, Washington D.C.
Rimal B, Sharma R, Kunwar R et al., 2019. Effects of land use and land cover change on ecosystem services in
the Koshi River basin, eastern Nepal. Ecosystem Services, 38: 100963. https://doi.org/10.1016/j.ecoser.
2019.100963.
Rodriguez-Galiano V F, Ghimire B, Rogan J et al., 2012. An assessment of the effectiveness of a random forest
classifier for land-cover classification. ISPRS Journal of Photogrammetry and Remote Sensing, 67(1): 93–104.
https://doi.org/10.1016/j.isprsjprs.2011.11.002.
Sharp R, Tallis H T, Ricketts T et al., 2018. InVEST 3.5.0 User’s Guide. The Natural Capital Project, Stanford
University, University of Minnesota, The Nature Conservancy, and World Wildlife Fund.
Sharpley A N, Williams J R, 1990. EPIC: The erosion-productivity impact calculator: I. Model documentation.
Beltsville, MD: U.S. Department of Agriculture Technical Bulletin (No.1768). Washington DC.
Singh G, Babu R, Narain P et al., 1992. Soil erosion rates in India. Journal of Soil and Water Conservation, 47:
97–99.
Sinha R, Gupta A, Mishra K et al., 2019. Basin-scale hydrology and sediment dynamics of the Kosi river in the
Himalayan foreland. Journal of Hydrology, 570: 156–166. https://doi.org/10.1016/j.jhydrol.2018.12.051.
Su Z, Xiong D H, Deng W et al., 2016. 137Cs tracing dynamics of soil erosion, organic carbon, and total nitrogen
in terraced fields and forestland in the Middle Mountains of Nepal. Journal of Mountain Sciences,
13:1829–1839. https://doi.org/10.1007/s11629-015-3581-z.
Sun W, Shao Q, Liu J, 2013. Soil erosion and its response to the changes of precipitation and vegetation cover on
the Loess Plateau. Journal of Geographical Sciences, 23(6): 1091–1106. https://doi.org/10.1007/s11442-013-1065-z.
Talukdar S, Singha P, Mahato S et al., 2020. Land-use land-cover classification by machine learning classifiers for
satellite observations: A review. Remote Sensing, 12(7). https://doi.org/10.3390/rs12071135.
Teffera Z L, Li J, Debsu T M et al., 2018. Assessing land use and land cover dynamics using composites of spec-
tral indices and principal component analysis: A case study in middle Awash sub-basin, Ethiopia. Applied Ge-
ography, 96: 109–129. https://doi.org/10.1016/j.apgeog.2018.05.015.
Teng M, Huang C, Wang P et al., 2019. Impacts of forest restoration on soil erosion in the Three Gorges Reservoir
area, China. Science of the Total Environment, 697(1): 134164. https://doi.org/10.1016/j.scitotenv.2019.134164.
Uddin K, Murthy M S R, Wahid S M et al., 2016. Estimation of soil erosion dynamics in the Koshi Basin using
GIS and remote sensing to assess priority areas for conservation. PLoS One, 11(3). https://doi.org/10.1371/
journal.pone.0150494.
Vigiak O, Borselli L, Newham L T H et al., 2012. Comparison of conceptual landscape metrics to define
hillslope-scale sediment delivery ratio. Geomorphology, 138(1): 74–88. https://doi.org/10.1016/j.geomorph.
2011.08.026.
Virgo K, Subba K J, 2016. Land-use 1978 and 1990 change between in Dhankuta Koshi Eastern Nepal District.
International Mountain Society, 14(2): 159–170.
Wang J, Zhong L, Zhao W et al., 2018. The influence of rainfall and land use patterns on soil erosion in mul-
ti-scale watersheds: A case study in the hilly and gully area on the Loess Plateau, China. Journal of Geo-
graphical Sciences, 28(10): 1415–1426. https://doi.org/10.1007/s11442-018-1553-2.
1312 Journal of Geographical Sciences

Wang X, Zhao X, Zhang Z et al., 2016. Assessment of soil erosion change and its relationships with land
use/cover change in China from the end of the 1980s to 2010. Catena, 137: 256–268. https://doi.org/10.1016/
j.catena.2015.10.004.
Wischmeier W H, Smith D D, 1978. Predicting rainfall erosion losses. Agriculture Handbook No.537: 285–291.
Wu X, Gao J G, Zhang Y L et al., 2017. Land cover status in the Koshi River basin, central Himalayas land cover
status in the Koshi River basin. Journal of Resources and Ecology, 8(1): 10–19. https://doi.org/10.5814/j.issn.
1674-764x.2017.01.003.
Xie F D, Wu X, Liu L S et al., 2021. Land use and land cover change within the Koshi River basin of the central
Himalayas since 1990. Journal of Mountain Science, 18(1): 159–177. https://doi.org/10.1007/s11629-
019-5944-3.
Xu D, Deng X, Guo S et al., 2019. Labor migration and farmland abandonment in rural China: Empirical results
and policy implications. Journal of Environmental Management, 232: 738–750. https://doi.org/10.1016/j.
jenvman.2018.11.136.
Xu Y, Luo D, Peng J, 2011. Land use change and soil erosion in the Maotiao River watershed of Guizhou prov-
ince. Journal of Geographical Sciences, 21(6): 1138–1152. https://doi.org/10.1007/s11442-011-0906-x.
Yan H, Nix H A, Hutchinson M F et al., 2005. Spatial interpolation of monthly mean climate data for China. In-
ternational Journal of Climatology, 25(10): 1369–1379. https://doi.org/10.1002/joc.1187.
Yan R, Zhang X, Yan S et al., 2018. Estimating soil erosion response to land use/cover change in a catchment of
the Loess Plateau, China. International Soil and Water Conservation Research, 6(1): 13–22.
https://doi.org/10.1016/j.iswcr.2017.12.002.
Yigez B, Xiong D, Zhang B et al., 2021 Spatial distribution of soil erosion and sediment yield in the Koshi River
basin, Nepal: A case study of Triyuga watershed. Journal of Soils and Sediments, 21(12): 3888–3905.
https://doi.org/10.1007/s11368-021-03023-9.
Yuan Y, Xiong D, Wu H et al., 2021. Using 137Cs and 210Pbex to trace soil erosion rates for a small catchment in
the mid-hills of Nepal. Journal of Soils and Sediments, 21(1): 403–418. https://doi.org/10.1007/
s11368-020-02760-7.
Zhao H, Tang Y, Yang S, 2018. Dynamic identification of soil erosion risk in the middle reaches of the Yellow
River Basin in China from 1978 to 2010. Journal of Geographical Sciences, 28(2): 175–192. https://doi.org/10.
1007/s11442-018-1466-0.
Zhou M, Deng J, Lin Y et al., 2019. Identifying the effects of land use change on sediment export: Integrating
sediment source and sediment delivery in the Qiantang River Basin, China. Science of the Total Environment,
686: 38–49. https://doi.org/10.1016/j.scitotenv.2019.05.336.
Zhu X, Liu W, Jiang X J et al., 2018. Effects of land-use changes on runoff and sediment yield: Implications for
soil conservation and forest management in Xishuangbanna, Southwest China. Land Degradation and Devel-
opment, 29(9): 2962–2974. https://doi.org/10.1002/ldr.3068.
Zhu Z, Wang S, Woodcock C E, 2015. Improvement and expansion of the Fmask algorithm: Cloud, cloud shadow,
and snow detection for Landsats 4–7, 8, and Sentinel 2 images. Remote Sensing of Environment, 159: 269–277.
https://doi.org/10.1016/j.rse.2014.12.014.
Zuo D, Xu Z, Yao W et al., 2016. Assessing the effects of changes in land use and climate on runoff and sediment
yields from a watershed in the Loess Plateau of China. Science of the Total Environment, 544: 238–250.
https://doi.org/10.1016/j.scitotenv.2015.11.060.
Zurqani H A, Post C J, Mikhailova E A et al., 2018. Geospatial analysis of land use change in the Savannah River
basin using Google Earth Engine. International Journal of Applied Earth Observation and Geoinformation, 69:
175–185. https://doi.org/10.1016/j.jag.2017.12.006.