Abstract
The precise identification of basin characteristics and climate factors that plays a significant role in determining water and sediment yield is of paramount importance. However, due to the enormous complexity associated with the hydrologic cycle, it is usually challenging to extricate the influence of individual parameters on the combined water and sediment yield responses. To accomplish this, a combined hydrological modelling and statistical approach was adopted in this study. The Soil and Water Assessment Tool (SWAT) model was adopted to simulate different components of the watershed and the results were utilized in Boosted Regression Trees (BRTs) to analyze the contribution of different parameters to water and sediment yield at spatio-temporal and seasonal scales in the upstream Teesta River basin. The outcomes of the analysis showed that precipitation and baseflow play a crucial role in regulating the water yield at all spatio-temporal scales. On the other hand, precipitation alone has a key role in determining the sediment yield, especially at the daily (49.30%) and monthly (21.14%) temporal scales. The relative contribution of the remaining parameters at a yearly temporal scale, and small and intermediate spatial scales showed relatively close results with an exception at the large basin scale (precipitation alone by 35.88%). The average contribution of actual evapotranspiration was found to be less on both water and sediment yield prediction in all spatio-temporal scales considered. The analysis also revealed that the precipitation, baseflow, and minimum temperature play a key role in regulating the water and sediment yield in both monsoon and non-monsoon seasons, whereas the contribution of maximum temperature and snowmelt was found less during monsoon and non-monsoon seasons. The outcomes of this study may assist policymakers and water managers in rational water management goals as well as in coping with soil degradation issues.
HIGHLIGHTS
A hydro-statistical approach to access the significance of basin and hydro-climatic factors on the water and sediment yield.
Precipitation and baseflow are the key contributors to the water yield.
Precipitation regulates sediment yield mechanisms at all scales.
Negligible role of evapotranspiration in determining water and sediment yield at all scales.
Weather and vegetation indices vary inversely to water and sediment yield at all scales.
Graphical Abstract
INTRODUCTION
The availability of freshwater is essential for the survival of humans and for maintenance of ecosystems. The mountainous regions which cover about 24% of landmass (Kapos et al. 2000) have enormous amounts of freshwater resources as they receive more precipitation than low lying areas, suffer less evapotranspiration, and hold huge supplies of water in the form of snow and ice (Somers & McKenzie 2020). Because of this, they are also known as ‘water tower’. The different components of surface and sub-surface runoff resulting from precipitation, snowmelt, and groundwater provide significant water resources to neighbouring areas which habitually includes arid and semi-arid areas. The mountainous areas play a significant role in providing water reserves in low lying areas (Viviroli et al. 2007) by regulating the snowmelt process for seasonal flow by storing and releasing water during dry and wet seasons (Painter et al. 2009; Armstrong et al. 2019). A unit millimetre of snowpack may produce 1–1.5 mm of water. It is reported that the dependency on mountainous water resources is increasing (Viviroli et al. 2020), the population of which is projected to increase to 1.4 billion in the 2050s. The distribution of population density is closely associated to the existence of a large water source which habitually originates from mountains (Meybeck et al. 2001) and the management of runoff is one way to lessen stress on existing freshwater resources (Vashisht & Ranjan 2020). The total flow contribution from mountains at global scale is estimated to be about 32% (Meybeck et al. 2001) and 95% at a regional scale (Liniger et al. 1998).
Globally, the estimated 1.386 billion cubic kilometres of total water on Earth covers about 71% of Earth's surface, and consists of 97.5% of ocean salt water, and only 2.5% available as freshwater (Gleick & Palaniappan 2010). Putting an emphasis on freshwater, about 69% is in the form of solid such as ice, snow cover, and glaciers. The remaining 31% of water comprises the freshwater lakes and groundwater components (Cavazza & Paglaria 2009). The surface freshwater is easily accessible; however, the sub-surface water, namely groundwater, requires external energy to extract especially in plain areas. In hilly areas, the components of sub-surface water can flow out as in the form of baseflow, orifice springs, and seepage springs. It has been well recognized that the surface and sub-surface water components of the watersheds are driven by climate conditions and various aspects of the basins (Sun et al. 2019). The water yield (WY), which implies the total volume of water that comes out of the hydrological response unit (HRU) and joining at some points from networks of streams in the stipulated time period (Arnold et al. 2011), is of great significance because it provides water reserve support to the ecological unit and for individual life. The water shortages and their increasing demand scenarios associated with increasing populations in hilly regions have brought negative aspects to long-term sustainable water resources management (Ranjan & Kumar Pandey 2020).
An attempt at WY services as well as impacts of multi-dimensional aspects over spatio-temporal scales employing different methodologies have been explored in the past. Sun et al. (2019) analyzed the basin and climate factor contributions to the WY using the coupled hydrological model SWAT and statistical tool Boosted Regression Trees (BRTs) at spatio-temporal scales; the WY and its dominant factors including land cover, precipitation, the Normalized Difference Vegetation Index (NDVI) by the Seasonal Water Yield Model (Lu et al. 2020) at spatio-temporal scales; long-term groundwater recharge studies using the partially-distributed water balance model WetSpass and investigated the influencing factors by the correlation technique (Zomlot et al. 2015). Other relationships between the WY and parameters like potential evapotranspiration (PET) (Wang et al. 2011), glacier, and snow melt (Anand et al. 2018; Li et al. 2021) have also been reported.
The sediment yield (SY), on the other hand, is the total quantity of sediment from a unit area detached from the basin by the action of flowing water all through a definite time period, signifying anthropogenic activities happening within the basin. The estimation of the SY is of utmost importance as it provides substrates for aquatic plants and animals, assessment of the reservoir sedimentation process that adversely affects the operational life of the hydro-electric dam, and recreational purposes. Soil erosion poses a serious threat to land degradation and agricultural land productivity (De Luis et al. 2010). The Himalayan range suffers erosion due to its undulating topographical features, slope, and improper management of watersheds (Chinnasamy & Sood 2020). Therefore, the prioritization of the basin as per erosion and SY is essential for executing soil and water conservation goals of the watersheds (Singh et al. 2019). Many studies have reported that precipitation (De Luis et al. 2010), snowmelt (Lana-Renault et al. 2011), precipitation and temperature (Hirschberg et al. 2021), land use land cover (LULC) change, and slope (Sok et al. 2020) play a vital role in determining the soil erosion and SY process.
Consequently, it is obvious from the vast review of the literature that the WY and SY vary with different spatio-temporal scales. Thus, a reasonable assessment of the WY is necessary to grasp the complex inter-relationships between different basin and climate parameters and for developing better management plans. In this research, a coupled hydrological (SWAT model) and statistical approach (BRTs) is adopted to study the influence of different parameters considered in WY and SY prediction. The SWAT model has been adopted to simulate WY (Shawul et al. 2013; Adeogun et al. 2014; Abeysingha et al. 2015; Jain et al. 2017) and SY (Duan et al. 2009; Chandra et al. 2014; Liu & Jiang 2019; Kuti & Ewemoje 2021) in different mountainous regions over the world. The BRTs model is adopted in multi-disciplinary field of studies like ecological modelling (De'ath 2007; Elith et al. 2008; Franklin 2010), and in various hydrological applications like quantifying the influence of basin and climate characteristics on the WY (Sun et al. 2019); mapping of groundwater potential zones (Naghibi et al. 2016, 2018), anthropogenic change impacts on aquatic life (Hale et al. 2014); defining non-linear relationships between nutrients concentration and basin variables (Golden et al. 2016), because of its superiority in capturing dynamic correlation between complex dependent and independent variables. BRTs use the combined statistical regression trees and the machine learning techniques to rank the variables and thereby enhance the predictions (Elith et al. 2008). Despite the superiority of BRTs in hydrological applications, they are still not often used.
Although various models and techniques are available to explore the WY and SY studies, limited studies have been conducted upon the contributions of various factors at different spatial and temporal scales especially in the context of the Indian Himalayan regions as they are one of the largest suppliers of freshwater (Bandyopadhyay & Gyawali 1994). In a previous study by Sun et al. (2019), only basin characteristics and climate parameters were considered to explore the WY. In this study, a comprehensive investigation was considered to understand the influence of different aspects, inclusive of hydrological components including snowmelt, groundwater flow, actual evapotranspiration, and the vegetation indices influence on the WY and SY. It could serve to enhance the knowledge of underlying processes governing watersheds, especially soil and water managements goals, prioritizing the areas which need special soil and water management aspects.
MATERIALS AND METHODS
Study area
(a) Study area showing drainage networks, elevation range, and sub-basins. (b) Slope map of the study area.
(a) Study area showing drainage networks, elevation range, and sub-basins. (b) Slope map of the study area.
Monthly variations of precipitation, maximum, and minimum temperature in study area. Note that precipitation consists of nine stations and temperature with single station near the basin outlet.
Monthly variations of precipitation, maximum, and minimum temperature in study area. Note that precipitation consists of nine stations and temperature with single station near the basin outlet.
Hydro-meteorological data for the SWAT model
The SWAT model requires comprehensive data about weather, LULC, soil properties, topographical aspects, and various land management practices adhered to the basin. All these records were collected from different sources and are presented in Table 1 and Figure 2.
Descriptions of data used in SWAT modelling
Sl . | Date type . | Data period . | Description . | |
---|---|---|---|---|
1. | Meteorological data | IMD-gridded dataset | ||
(i) | Precipitation | 2000–2020 | ||
(ii) | Maximum and minimum temperature | 2000–2020 | ||
2. | Hydrological data | Central Water Commission, Kolkata | ||
(i) | Streamflow | 2009–2013 | ||
(ii) | Sediment yield | 2009–2013 | ||
3. | Digital elevation model | Shuttle Radar Topography Mission (1 arc-second) | ||
4. | LULC | 2010 | Landsat 5 | |
5. | Soil map | 2000 | National Bureau of Soil Survey and Land Use Planning, Kolkata (Scale 1:250,000) |
Sl . | Date type . | Data period . | Description . | |
---|---|---|---|---|
1. | Meteorological data | IMD-gridded dataset | ||
(i) | Precipitation | 2000–2020 | ||
(ii) | Maximum and minimum temperature | 2000–2020 | ||
2. | Hydrological data | Central Water Commission, Kolkata | ||
(i) | Streamflow | 2009–2013 | ||
(ii) | Sediment yield | 2009–2013 | ||
3. | Digital elevation model | Shuttle Radar Topography Mission (1 arc-second) | ||
4. | LULC | 2010 | Landsat 5 | |
5. | Soil map | 2000 | National Bureau of Soil Survey and Land Use Planning, Kolkata (Scale 1:250,000) |
The meteorological data required to run the SWAT model are precipitation, maximum–minimum temperature, relative humidity (RH), solar radiation, and wind speed. The precipitation (Pai et al. 2014) and maximum–minimum temperature (Srivastava et al. 2009) data used in this model were acquired from the gridded dataset of India Meteorological Department (IMD), Pune. The weather generator of IMD-gridded dataset processed into SWAT format (https://swat.tamu.edu/data/india-dataset/) has been used to generate the missing data for RH, solar radiation, and wind speed. The discharge (two stations) and sediment data (one station) for a common period starting from 2009 to 2013 have been acquired from the Central Water Commission, Kolkata for calibration and validation of the model.
Digital elevation model
The Shuttle Radar Topography Mission global 1 arc-second (∼30 m resolution) digital elevation model (DEM) was obtained from www.earthexplorer.usgs.gov to delineate the basin boundary, preparation of drainage network, stream length, channel widths, slope, and elevation aspects of the basin. The DEM was converted to Universal Transverse Mercator zone 45 °N. The topography of the study area is characterized by steep slopes and high relief. The slope was categorized into the following five classes, namely (0–20), (20–40), (40–60), (60–80), and >80 covering 16, 20.42, 21.97, 17.65, and 23.97%, respectively.
LULC map
Soil map
The soil map acquired from the National Bureau of Soil Survey and Land Use Planning, Kolkata, for the year 2000 with a scale of 1:250,000 was digitized using the ArcMap 10.2.2 software. Six different types of soil based on the Food and Agriculture Organization classification have been identified (see Table 2). Additionally, two complementary categories comprising glaciers with rocky mountains covering snow and glaciers throughout the year and water bodies with inclusion of lakes and rivers were also identified as shown in Figure 3(b).
Hydrological and physical properties of soil types in study area
Sl . | Soil type . | Area (km2) . | Basin area (%) . | Texture . | Saturated hydraulic conductivity (mm/h) . | Available water holding capacity (mm water/ mm soil) . | Hydrologic Soil Group . |
---|---|---|---|---|---|---|---|
1 | Bd32-2bc-3662 | 108.74 | 2.86 | Loam | 61.87 | 0.118 | C |
2 | Be84-2a-3685 | 53.97 | 1.42 | Loam | 8.31 | 0.175 | D |
3 | Bh25-2bc-3024 | 1,094.49 | 28.76 | Loam | 56.36 | 0.072 | C |
4 | GLACIER-6998 | 907.55 | 23.85 | Unweathered bedrock | – | – | D |
5 | Hh11-2bc-3711 | 71.62 | 1.88 | Clay loam | 2.88 | 0.16 | D |
6 | Hl38-2bc-5889 | 39.12 | 1.03 | Loam | 5.27 | 0.144 | D |
7 | I-Bh-U-c-3717 | 1,523.75 | 40.04 | Loam | 33.91 | 0.064 | C |
8 | WATER-6997 | 6.39 | 0.17 | Water | – | – | D |
Sl . | Soil type . | Area (km2) . | Basin area (%) . | Texture . | Saturated hydraulic conductivity (mm/h) . | Available water holding capacity (mm water/ mm soil) . | Hydrologic Soil Group . |
---|---|---|---|---|---|---|---|
1 | Bd32-2bc-3662 | 108.74 | 2.86 | Loam | 61.87 | 0.118 | C |
2 | Be84-2a-3685 | 53.97 | 1.42 | Loam | 8.31 | 0.175 | D |
3 | Bh25-2bc-3024 | 1,094.49 | 28.76 | Loam | 56.36 | 0.072 | C |
4 | GLACIER-6998 | 907.55 | 23.85 | Unweathered bedrock | – | – | D |
5 | Hh11-2bc-3711 | 71.62 | 1.88 | Clay loam | 2.88 | 0.16 | D |
6 | Hl38-2bc-5889 | 39.12 | 1.03 | Loam | 5.27 | 0.144 | D |
7 | I-Bh-U-c-3717 | 1,523.75 | 40.04 | Loam | 33.91 | 0.064 | C |
8 | WATER-6997 | 6.39 | 0.17 | Water | – | – | D |
Hydrological modelling of the upstream Teesta River basin using the SWAT model

















The hydrological simulations of the basin comprise two major components, namely land and water. The land component simulates water, nutrients, pesticides, and sediment resulting from the surface runoff towards the main channel. Similarly, the water components predict these through the movement of water from the networks of the channel in the watershed (Neitsch et al. 2011).










SWAT model setup
In this study, Arc SWAT 2012 extension was used in the ArcMap 10.2.2 software to set up the model. The watershed characteristics were created based on the 30 m spectral resolution of the DEM. All other input raster components were resampled to same spatial resolution using the resampling tool in ArcMap. A threshold area of 30 km2 was defined for the generation of stream networks. The topographical components, stream and channel networks were also derived from the DEM. A total of 21 monitoring points were defined manually based on the observed discharge gauging stations for ease of model calibration and validation. Furthermore, the sub-basins were divided into 299 HRUs with land-use, soils, and slope thresholds of 20, 10, and 15%, respectively. The water balance mechanisms processed at the HRU stage are collected at the corresponding sub-basin which is further directed to the basin outlet (Grusson et al. 2015). For computation of water balance components, runoff was estimated using the curve number method, PET by the Penman–Monteith method, and channel routing by the variable storage method. The daily surface runoff was estimated by the SCS-CN method based on daily observed precipitation data. The curve number value for the particular day was estimated as per the hydrologic group of soil and the mean antecedent soil moisture condition. For the estimation of PET, the input data used are observed precipitation, maximum–minimum temperature; and the generated RH, solar radiation, and wind speed data from IMD-gridded weather generators. The sub-surface flow components were estimated using the variable storage method.
Model calibration, validation, and sensitivity analysis
Performance of SWAT model in calibration and validation of streamflow and sediment yield
Daily time step . | ||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
. | Calibration . | Validation . | ||||||||||
. | NSE . | R2 . | PBIAS . | NSE . | R2 . | PBIAS . | ||||||
Station . | Qsr . | Qsy . | Qsr . | Qsy . | Qsr . | Qsy . | Qsr . | Qsy . | Qsr . | Qsy . | Qsr . | Qsy . |
Chungthang | 0.67 | – | 0.71 | – | −4.7 | – | 0.68 | – | 0.73 | – | −6.6 | – |
Sangkalang | 0.70 | 0.55 | 0.76 | 0.56 | 6.1 | −8.6 | 0.66 | 0.60 | 0.68 | 0.62 | 13.0 | −12.7 |
Monthly time step | ||||||||||||
. | Qsr . | Qsy . | Qsr . | Qsy . | Qsr . | Qsy . | Qsr . | Qsy . | Qsr . | Qsy . | Qsr . | Qsy . |
Chungthang | 0.81 | – | 0.82 | – | 7.9 | – | 0.83 | – | 0.88 | – | −5.3 | – |
Sangkalang | 0.79 | 0.70 | 0.84 | 0.73 | 14.1 | −8.2 | 0.72 | 0.71 | 0.74 | 0.76 | 13.5 | −12.6 |
Daily time step . | ||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
. | Calibration . | Validation . | ||||||||||
. | NSE . | R2 . | PBIAS . | NSE . | R2 . | PBIAS . | ||||||
Station . | Qsr . | Qsy . | Qsr . | Qsy . | Qsr . | Qsy . | Qsr . | Qsy . | Qsr . | Qsy . | Qsr . | Qsy . |
Chungthang | 0.67 | – | 0.71 | – | −4.7 | – | 0.68 | – | 0.73 | – | −6.6 | – |
Sangkalang | 0.70 | 0.55 | 0.76 | 0.56 | 6.1 | −8.6 | 0.66 | 0.60 | 0.68 | 0.62 | 13.0 | −12.7 |
Monthly time step | ||||||||||||
. | Qsr . | Qsy . | Qsr . | Qsy . | Qsr . | Qsy . | Qsr . | Qsy . | Qsr . | Qsy . | Qsr . | Qsy . |
Chungthang | 0.81 | – | 0.82 | – | 7.9 | – | 0.83 | – | 0.88 | – | −5.3 | – |
Sangkalang | 0.79 | 0.70 | 0.84 | 0.73 | 14.1 | −8.2 | 0.72 | 0.71 | 0.74 | 0.76 | 13.5 | −12.6 |
Qsr, runoff; Qsy, sediment yield.
Calibration and validation of streamflow for (a) Chungthang and (b) Sangkalang gauging stations. (c) Calibration and validation of sediment yield at the Sangkalang gauging station.
Calibration and validation of streamflow for (a) Chungthang and (b) Sangkalang gauging stations. (c) Calibration and validation of sediment yield at the Sangkalang gauging station.
Sensitivity analysis was carried out to govern the effect of chosen parameters on calculating the WY and SY. SWAT Calibration Uncertainty Programmes embedded with SUFI-2 were employed to carry out calibration and sensitivity analysis in this study.
The SUFI-2 optimization algorithm is widely used to calibrate the SWAT model using a Bayesian system as it required a lower number of model runs with good prediction capability (Yang et al. 2008). Generally, the sensitivity analysis is carried out in two ways: local sensitivity or one-at-a-time (OAT) analysis (Abbaspour et al. 2017) or global sensitivity or all-at-a-time (AAT) analysis. In OAT analysis, while keeping all other parameters constant and changing one parameter at a time, the effect on output is accessed simultaneously. While in AAT analysis, all parameters are changed simultaneously to access the effect of individual parameters on the model output. Based on the analysis, the sensitive parameters were identified, and reduction of model parameters was achieved with further improving the objective function. The sensitive parameters were further used to calibrate the SWAT model for prediction of the WY and SY (see Tables 4 and 5).
SWAT parameters, range, and calibrated values for streamflow
Parameters . | Min. . | Max. . | Calibrated value . | Sensitivity rank . | t-stat . | p-value . | Description . |
---|---|---|---|---|---|---|---|
Management parameters | |||||||
r__CN2 | −0.4 | 0.4 | −0.01300 | 4 | −3.495 | 0.001 | Antecedent moisture condition-II SCS runoff curve number |
Soil parameters | |||||||
r__SOL_AWC | −0.20 | 0.20 | −0.04160 | 13 | 0.361 | 0.719 | Soil available water capacity |
r__SOL_K | −0.40 | 0.40 | 0.30057 | 17 | 0.092 | 0.927 | Saturated hydraulic conductivity |
HRUs parameters | |||||||
r__HRU_SLP | 0 | 0.6 | 0.40825 | 6 | 2.198 | 0.031 | Average slope steepness |
v__CANMX | 0 | 20 | 8.86925 | 7 | 1.957 | 0.054 | Maximum canopy storage |
v__EPCO | 0 | 1 | 0.56157 | 14 | −0.252 | 0.802 | Plant uptake compensation factor |
Groundwater parameters | |||||||
v__GW_SPYLD | 0 | 0.4 | 0.38634 | 11 | −0.760 | 0.449 | Shallow aquifer specific yield |
v__GWQMN | 0 | 400 | 1.88539 | 18 | −0.072 | 0.943 | Shallow aquifer's threshold depth for return flow to occur |
v__GW_DELAY | 0 | 50 | 29.03028 | 10 | 1.189 | 0.238 | Groundwater delay time |
v__ALPHA_BF | 0 | 1 | 0.43385 | 2 | −5.209 | 0.000 | Baseflow alpha factor |
Main channel parameters | |||||||
v__CH_N2 | 0.01 | 0.3 | 0.28865 | 5 | 2.439 | 0.017 | Mannings's ‘n’ for main channel |
v__CH_K2 | 50 | 250 | 239.47588 | 8 | 1.607 | 0.112 | Effective hydraulic conductivity of main channel |
Sub-basin parameters | |||||||
v__TLAPS | −8 | 8 | −5.21234 | 1 | −8.010 | 0.000 | Temperature lapse rate |
v__PLAPS | −40 | 200 | 185.97263 | 12 | −0.611 | 0.543 | Precipitation lapse rate |
Basin parameters | |||||||
v__SFTMP | −2.5 | 2.5 | −0.20370 | 3 | 3.644 | 0.000 | Snowfall temperature |
v__SMFMN | 0 | 4 | 2.21313 | 15 | 0.154 | 0.878 | Minimum snowmelt occurring during winter season |
v__SMFMX | 2 | 8 | 4.06792 | 9 | −1.528 | 0.130 | Maximum snowmelt occurring during summer season |
v__ESCO | 0 | 1 | 0.83547 | 16 | −0.135 | 0.893 | Soil evaporation compensation factor |
Parameters . | Min. . | Max. . | Calibrated value . | Sensitivity rank . | t-stat . | p-value . | Description . |
---|---|---|---|---|---|---|---|
Management parameters | |||||||
r__CN2 | −0.4 | 0.4 | −0.01300 | 4 | −3.495 | 0.001 | Antecedent moisture condition-II SCS runoff curve number |
Soil parameters | |||||||
r__SOL_AWC | −0.20 | 0.20 | −0.04160 | 13 | 0.361 | 0.719 | Soil available water capacity |
r__SOL_K | −0.40 | 0.40 | 0.30057 | 17 | 0.092 | 0.927 | Saturated hydraulic conductivity |
HRUs parameters | |||||||
r__HRU_SLP | 0 | 0.6 | 0.40825 | 6 | 2.198 | 0.031 | Average slope steepness |
v__CANMX | 0 | 20 | 8.86925 | 7 | 1.957 | 0.054 | Maximum canopy storage |
v__EPCO | 0 | 1 | 0.56157 | 14 | −0.252 | 0.802 | Plant uptake compensation factor |
Groundwater parameters | |||||||
v__GW_SPYLD | 0 | 0.4 | 0.38634 | 11 | −0.760 | 0.449 | Shallow aquifer specific yield |
v__GWQMN | 0 | 400 | 1.88539 | 18 | −0.072 | 0.943 | Shallow aquifer's threshold depth for return flow to occur |
v__GW_DELAY | 0 | 50 | 29.03028 | 10 | 1.189 | 0.238 | Groundwater delay time |
v__ALPHA_BF | 0 | 1 | 0.43385 | 2 | −5.209 | 0.000 | Baseflow alpha factor |
Main channel parameters | |||||||
v__CH_N2 | 0.01 | 0.3 | 0.28865 | 5 | 2.439 | 0.017 | Mannings's ‘n’ for main channel |
v__CH_K2 | 50 | 250 | 239.47588 | 8 | 1.607 | 0.112 | Effective hydraulic conductivity of main channel |
Sub-basin parameters | |||||||
v__TLAPS | −8 | 8 | −5.21234 | 1 | −8.010 | 0.000 | Temperature lapse rate |
v__PLAPS | −40 | 200 | 185.97263 | 12 | −0.611 | 0.543 | Precipitation lapse rate |
Basin parameters | |||||||
v__SFTMP | −2.5 | 2.5 | −0.20370 | 3 | 3.644 | 0.000 | Snowfall temperature |
v__SMFMN | 0 | 4 | 2.21313 | 15 | 0.154 | 0.878 | Minimum snowmelt occurring during winter season |
v__SMFMX | 2 | 8 | 4.06792 | 9 | −1.528 | 0.130 | Maximum snowmelt occurring during summer season |
v__ESCO | 0 | 1 | 0.83547 | 16 | −0.135 | 0.893 | Soil evaporation compensation factor |
Min., minimum value; Max., maximum value; r, multiplicative operator; v, replace operator.
SWAT parameters, range, and calibrated values for sediment yield
Parameters . | Min. . | Max. . | Calibrated value . | Sensitivity rank . | t-stat . | p-value . | Description . |
---|---|---|---|---|---|---|---|
Management parameters | |||||||
r__CN2 | −0.4 | 0.4 | −0.0919 | 2 | 9.324 | 0.000 | Antecedent moisture condition-II SCS runoff curve number |
v__USLE_P | 0.3 | 0.8 | 0.5984 | 4 | 5.201 | 0.000 | USLE support practice factor |
v__USLE_C | 0 | 0.5 | 0.2778 | 8 | −0.005 | 0.227 | USLE cover management factor |
Soil parameters | |||||||
r__SOL_AWC | −0.20 | 0.20 | 0.0829 | 9 | 1.356 | 0.881 | Soil available water capacity |
r__SOL_K | −0.40 | 0.40 | −0.0959 | 5 | −2.978 | 0.003 | Saturated hydraulic conductivity |
r__SOL_Z | −0.4 | 0.4 | −0.3286 | 11 | −0.105 | 0.946 | Soil surface to bottom layer depth |
r__SOL_ZMX | −0.4 | 0.4 | 0.0109 | 10 | 1.008 | 0.920 | Soil profile maximum rooting depth |
r__USLE_K | −0.4 | 0.4 | 0.1539 | 6 | 1.895 | 0.061 | USLE soil erodibility factor |
HRUs parameters | |||||||
r__OV_N | −0.2 | 0.2 | −0.0087 | 12 | −0.062 | 0.996 | Average slope steepness |
Main channel parameters | |||||||
v__CH_S2 | −0.4 | 0.4 | −0.1363 | 7 | −1.355 | 0.178 | Mannings's ‘n’ for main channel |
Sub-basin parameters | |||||||
v__TLAPS | −8 | 8 | −7.5006 | 3 | −6.617 | 0.000 | Temperature lapse rate |
Basin parameters | |||||||
v__SPEXP | 1 | 1.5 | 1.0600 | 1 | −13.720 | 0.000 | Exponential factor for calculating sediment re-entrained in channel |
Parameters . | Min. . | Max. . | Calibrated value . | Sensitivity rank . | t-stat . | p-value . | Description . |
---|---|---|---|---|---|---|---|
Management parameters | |||||||
r__CN2 | −0.4 | 0.4 | −0.0919 | 2 | 9.324 | 0.000 | Antecedent moisture condition-II SCS runoff curve number |
v__USLE_P | 0.3 | 0.8 | 0.5984 | 4 | 5.201 | 0.000 | USLE support practice factor |
v__USLE_C | 0 | 0.5 | 0.2778 | 8 | −0.005 | 0.227 | USLE cover management factor |
Soil parameters | |||||||
r__SOL_AWC | −0.20 | 0.20 | 0.0829 | 9 | 1.356 | 0.881 | Soil available water capacity |
r__SOL_K | −0.40 | 0.40 | −0.0959 | 5 | −2.978 | 0.003 | Saturated hydraulic conductivity |
r__SOL_Z | −0.4 | 0.4 | −0.3286 | 11 | −0.105 | 0.946 | Soil surface to bottom layer depth |
r__SOL_ZMX | −0.4 | 0.4 | 0.0109 | 10 | 1.008 | 0.920 | Soil profile maximum rooting depth |
r__USLE_K | −0.4 | 0.4 | 0.1539 | 6 | 1.895 | 0.061 | USLE soil erodibility factor |
HRUs parameters | |||||||
r__OV_N | −0.2 | 0.2 | −0.0087 | 12 | −0.062 | 0.996 | Average slope steepness |
Main channel parameters | |||||||
v__CH_S2 | −0.4 | 0.4 | −0.1363 | 7 | −1.355 | 0.178 | Mannings's ‘n’ for main channel |
Sub-basin parameters | |||||||
v__TLAPS | −8 | 8 | −7.5006 | 3 | −6.617 | 0.000 | Temperature lapse rate |
Basin parameters | |||||||
v__SPEXP | 1 | 1.5 | 1.0600 | 1 | −13.720 | 0.000 | Exponential factor for calculating sediment re-entrained in channel |
Min., minimum value; Max., maximum value; r, multiplicative operator; v, replace operator.
BRTs for statistical analysis
BRTs are capable of scrutinizing the high non-linearity between dependent and independent variables. Nowadays, due to its superiority in separating complex and highly interdependent variables, the BRTs model has been employed in various hydrological studies (Sun et al. 2019) to detect the importance of each environmental factor in predicting the WY.






The model calculates the significance of each independent variable over recursive binary splitting techniques which improved the discrepancies among variables. A forward stage-by-stage technique reserves already-built trees at each stage and adds new trees by reweighing the residuals from earlier trees. As a result, the BRT models generate thousands of trees, from which a mean variable importance estimate is extracted across all trees (De'ath 2007; Elith et al. 2008). In order to run BRTs, five key parameters needed to be defined: (1) probability distribution function (PDF) of response variable, (2) learning rate, (3) tree complexity, (4) bag fraction, and (5) cross-validation folds (Leathwick et al. 2006; Elith et al. 2008). The estimated parameters of the selected PDF as the response variables were assumed to be Gaussian distributed. As suggested by Elith et al. (2008), BRTs with the learning rate 0.005, bag fraction 0.5, tree complexity 1, and the cross-validation fold, which is the number of times the cross-validation was carried out to get the optimum number of trees to be assembled for the final model was set to 10 folds, and is considered in this study. The analysis of BRTs was carried out using dismo package (Hijmans et al. 2020) in R software (R Core Team 2021).
Spatial map showing small, intermediate, and large basins considered in this study.
Spatial map showing small, intermediate, and large basins considered in this study.
In this study, to analyze the significance of different variables on predicting WY, few hydro-climatic and basin factors including precipitation, maximum and minimum temperature, snowmelt, actual evapotranspiration (ETa), baseflow, NDVI, Enhanced Vegetation Index (EVI) have been considered. These variables have been revealed to have an influence on WY and SY predictions. The daily NDVI and EVI dataset of the Moderate Resolution Imaging Spectroradiometer (MODIS) Terra platform with an average resolution of about 463.31 m were obtained from the Google Earth Engine platform. The NDVI value is generated from surface reflectance composites in Near-Infrared and red bands as (NIR–red band)/(NIR+red band), which ranges from −1.0 to 1.0. EVI, on the other hand, is superior to the NDVI as it also incorporates visible blue band in depicting the vegetation index, which allows the aerosol-scattering effects to be compensated.
Comparative contribution of variables to the WY at spatio-temporal scales
(a)–(h) Percent contributions of parameters to the WY at 45 small sub-basins. Note: Tmin: minimum temperature (°C), Tmax: maximum temperature (°C), ETa: actual evapotranspiration.
(a)–(h) Percent contributions of parameters to the WY at 45 small sub-basins. Note: Tmin: minimum temperature (°C), Tmax: maximum temperature (°C), ETa: actual evapotranspiration.
Relative contributions of parameters to the WY at intermediate basins during monsoon and non-monsoon periods.
Relative contributions of parameters to the WY at intermediate basins during monsoon and non-monsoon periods.
Relative contributions of parameters to the WY at small basins during monsoon and non-monsoon periods. Note: M indicates monsoon; NM indicates non-monsoon season; BF indicates baseflow; Prc indicates precipitation; and Snow indicates snowmelt.
Relative contributions of parameters to the WY at small basins during monsoon and non-monsoon periods. Note: M indicates monsoon; NM indicates non-monsoon season; BF indicates baseflow; Prc indicates precipitation; and Snow indicates snowmelt.
(a–h) Relative contributions of parameters to the WY at intermediate basins.
The BRTs analysis of the monsoon season at a large basin scale revealed that precipitation, baseflow, and minimum temperature contribute about half to total WY with precipitation being the highest contributor (16.50%). On the other hand, the contribution of ETa was the least among all parameters with 7.90%. During the non-monsoon season, baseflow (15.19%) and precipitation (14.68%) contribute more to WY. The contribution of snowmelt and maximum temperature were equally observed least among all the parameters by about 10.23%. The climatic factors contribute about 41.38% during monsoon and 35.31% during the non-monsoon season to WY. On the other hand, about 20.98% of total WY was contributed by vegetation index during monsoon and 27.31% during non-monsoon season.
Contribution of variables to SY at spatio-temporal scales
(a–h) Relative contributions of parameters to the SY at small basins.
Relative contributions of parameters to the SY at small basins during monsoon and non-monsoon periods.
Relative contributions of parameters to the SY at small basins during monsoon and non-monsoon periods.
(a–h) Relative contributions of parameters to the SY at intermediate basins.
Relative contributions of parameters to the SY at intermediate basins during monsoon and non-monsoon periods.
Relative contributions of parameters to the SY at intermediate basins during monsoon and non-monsoon periods.
The contribution of precipitation to SY at the small basin scale was highest amongst all the parameters by 17.24%, followed by baseflow, minimum temperature, and EVI by 15.33, 14.13, and 12.23%, respectively. The least contributor to SY was snowmelt with 7.19% and the remaining three parameters contributes from 11.09 to 11.59%, respectively. In the non-monsoon season, ETa contributes more to SY by 13.67%, though the relative contribution of minimum temperature, snowmelt, and baseflow is very close to ETa. The precipitation contribution to SY was relatively low at 12.94% and the NDVI with the least by 10.53%, respectively. The contribution of both climate factor and vegetative index decreases from monsoon to non-monsoon season by 42.96–38.53% and 23.32–21.36%, respectively (see figure 13).
In the intermediate basin, the climate factors were dominant with an overall contribution of about 67.62%. The snowmelt and baseflow together contribute about 22.17%, followed by a vegetative factor (7.17%) and ET with the least significance of 3.04% only. The sequence of significant parameters was same at both monthly and annual scales. However, precipitation contributes more at monthly scale (21.48%) as compared to the annual scale by 16.69%. Baseflow and snowmelt contribute slightly more at annual scale (26.76%) when compared to the monthly scale (23.37). However, the snowmelt contribution approximately remains the same in both time scales. Taking the climatic and vegetative factors into account, the former accounts for 48.69% (monthly) and 39.25% (yearly) only, and the latter accounts for 18.03% (monthly); and 24.02% (annual) which follows the same order of increasing (vegetative factors) and decreasing (climate) over the small basin scale. The ET was found to be less significant at about 10% approximately in both temporal scales. The graphical representation of the relative contribution of each parameter to SY is shown in Figure 14.
In the monsoon season, at the intermediate basin scale, precipitation contributes more to SY by 16.39% and snowmelt contributes least by 9.86%, respectively. On the other hand, during the non-monsoon season, the contribution of baseflow (13.77%) and precipitation (13.53) were highest to SY. The role of snowmelt and minimum temperature were felt to be low with a contribution of about 11.13% and 11.48%, respectively. In the intermediate basin scale, the contribution of climatic factors decreases from monsoon to non-monsoon season by 42.10–37.10%, and the vegetation index increases from 21.80 to 24.90% to SY (see Figure 15).
Relative contributions of parameters to the SY at the large basin during monsoon and non-monsoon periods.
Relative contributions of parameters to the SY at the large basin during monsoon and non-monsoon periods.
The contribution of precipitation to SY at large basin scale during monsoon season was highest (15.75%) with the lowest contributors being maximum temperature (10.44%) and NDVI (10.49%). On the other hand, during the non-monsoon season, precipitation was the highest contributor to SY with a slight increase in contrast to monsoon season by about 1.16%. The least contributor to SY during non-monsoon was found to be same parameters with the same value as in the case of monsoon season. The overall contribution of climate and vegetation index decreases from monsoon to non-monsoon season from 40.07 to 38.33% and 22.88 to 21.34%, respectively (see Figure 17).
Discussion
The statistical analysis report of BRTs demonstrates that the precipitation and baseflow were two of the principal factors that influenced WY on all spatio-temporal scales. Precipitation is one of the important factors that governs WY as previously reported by Jiang et al. (2016); Sun et al. (2019) and Lu et al. (2020) is used by policymakers, especially in water resources management and goals in conducting apportioning of water resources. The baseflow is one of the key components of the groundwater system, which provides sub-surface flow and other delayed sources like snowmelt into the stream, which is the second most important factor contributing to WY. Its importance is attributed to the improvement of water management plans, particularly for drought situations, assessment of small to medium water supplies, water quality (Santhi et al. 2008), water supply, generation of hydro-electric power, and recreational purposes (McMahon & Mein 1986). The significance of baseflow was felt next to precipitation in all spatial and temporal scales. The third key variable to WY, temperature, plays a crucial role in monitoring the water cycle of the watershed by altering the different components. The importance of minimum temperature was perceived as slightly higher than that of snowmelt and maximum temperature to WY. This may be attributed to the fact that minimum temperature helps to generate more snow and glaciers (in cold regions), and maximum temperature drives the snowmelt process. Thereby, storing water in the form of snow and glaciers when the temperature is minimum and melts in the course of a rise in temperature. The contribution of vegetation indices was comparatively less at daily timescale (5.12–8.89%) and gradually increases at monthly (18.29–18.98%) and yearly (22.27–27.22%) timescales. The contribution of ET was less than WY in all spatio-temporal scales. The average ETa values at daily, monthly, and annual scales are much less than precipitation values. A small change in precipitation and ETa will cause disproportionate changes in runoff and water storage within the basin (Rungee et al. 2021). For example, the average annual precipitation is about 215 mm with an SD of ∼593, and ETa value was 21 mm with an SD of ∼17. The large deviation of precipitation obviously depicts the significant role played by precipitation in overall runoff variations in the basin as compared to ETa. Moreover, as per our analysis, almost 60% of the total basin area is predominantly covered by bare land and snow/glaciers throughout the year and the temperature remained well below 0 °C especially in the mid- to northern part of the basin. In the southern part of the basin, the average temperature within the basin seldom exceeds 21 °C, which could limit the process of ETa. Therefore, the low degree of variations in ETa data series results in a low coefficient of variation, which resulted in low dependency by runoff, and then the SY. Further, the regression analysis between ETa with WY and SY revealed the coefficient of variation value ranges from 0.04 to 0.06 for both WY and SY. This may be one of the important reasons for an insignificant contribution of ETa to both WY and SY.
In contrast to monsoon and non-monsoon seasons, on average, baseflow, precipitation, and minimum temperature were the key parameters governing half of the total WY in monsoon season. On average, the overall contribution of precipitation, baseflow, and vegetation index was about 55.78% of the total WY during the non-monsoon season. However, the average contribution of maximum temperature, ETa, and NDVI were the lowest (<9%) during monsoon season and in the case of non-monsoon, snowmelt contributes less to WY. Putting an emphasis on climate and vegetation indices, the contribution of climate factors decreases from monsoon to non-monsoon seasons from 41.10 to 36.89%. On the other hand, the contribution of the vegetation index increases by 4.82% from 21.60% during the monsoon season, respectively.
A change in SY at the outlet to a maximum extent indicates the anthropogenic activities linked to the watershed. An analysis of BRTs established that precipitation is the key factor that has a greater influence on SY in all spatio-temporal scales. As reported by Jiang et al. (2016) and Zhang et al. (2022), precipitation is one of the principal factors affecting the degree of soil erosion. The contribution of maximum and minimum temperatures tends to be more emphasized at daily and monthly scales than yearly scales. Subsequently, the importance of vegetation indices firmly remained low to SY in all the spatio-temporal scales. This may be attributed to the fact that where precipitation increases erosion, vegetation inhibits the erosion process. The contribution of ETa to SY constantly remained low to negligible, in contrast to other predictors at all spatio-temporal scales. The relation between ETa and SY is similar to the relation between ETa and WY. Based on our analysis, a very high correlation exists between SY and WY (correlation coefficient is equal to 0.93 at the monthly timescale). This is one of the important reasons for the lower contribution of ETa to SY as well. This may be attributed to the fact that the correlation coefficient between ETa and SY was found to be closer to zero (0.04–0.06) in all spatio-temporal scales in the study area.
As analogous to WY, the average contribution of precipitation, baseflow, and minimum temperature tends to be more during the monsoon season than SY at all spatial scales. However, during the non-monsoon season, only precipitation and baseflow were found to be dominant parameters contributing to SY. The maximum temperature during monsoon and snowmelt during non-monsoon were found to be the lowest contributors to SY. It is worth mentioning that the contribution of climate factors during monsoon to non-monsoon season was reduced by 3.37% and that of vegetation index increased by 2.19%, respectively. This clearly demonstrates the influence of climate and vegetation parameters on SY.
In this study, by combining the SWAT with BRTs, the relationships between dependent (WY and SY) and independent variables at multiple spatio-temporal scales are quantitated and evaluated. Our investigations undoubtedly discover that the response of WY and SY to hydro-climatic and basin parameters is physically flexible and essentially relies on the scale.
Evidently, the impact of hydro-climatic and basin parameters on WY and SY is basin-specific, making it impractical to characterize the quantitative relationships employing a standard acceptable rule. In this circumstance, we argue that it is inappropriate to discuss how different parameters affect the WY and SY without addressing the precise site and time scale. It is therefore suggested that any extrapolation of WY and SY relationships from one basin to another must be done with caution (Berg et al. 2016). Compared to coarser scales, simulations at daily scale are more detailed in their representation of hydrological processes. The opportunity to analyze and observe the relative contributions of parameters that might be misinterpreted at coarser resolutions is provided by the daily scale. It makes sense to advise researchers looking at changes in WY and SY to give priority to performing their research at smaller temporal scales.
CONCLUSION
The analysis was undertaken at different spatial (small, intermediate, and large basins), temporal (daily, monthly, and yearly), and seasonal (monsoon and non-monsoon) scales. Based on the results obtained from the BRTs model, the following major outcomes have been identified:
Out of the eight parameters considered, precipitation and baseflow to a larger extent contributes more to WY at all spatio-temporal scales except for (monthly and yearly scale) at the intermediate basin, baseflow, and minimum temperature at monthly timescale; precipitation and actual evapotranspiration at yearly temporal scale. On the other hand, the contribution of ETa was consistently low to WY at all spatio-temporal scales considered.
Similarly, the contribution of precipitation was much higher than any other parameters to SY at all spatio-temporal scales. Although the ranking of precipitation is higher in all spatio-temporal scales, the relative contributions of all parameters at yearly timescale are close to each other. The importance of ETa was similar to that of WY.
The seasonal analysis of eight parameters to WY and SY revealed that precipitation, baseflow, and minimum temperature contribute about 40–49% of the total WY, and 42–47% of the total SY at all spatial scales. The average contribution of maximum temperature was relatively low during monsoon and snowmelt during the non-monsoon season. The relative significance of the remaining parameters to WY and SY differs among monsoon and non-monsoon seasons.
At the small basin scale, weather parameters including precipitation, maximum and minimum temperature account for 55.53, 43.65, and 38.86% to WY at daily, monthly, and yearly timescales. Similarly, the same parameters account for 69.58, 48.46, and 38.73% to SY at the same order of temporal scales. At the intermediate basin, they contribute about 55.19, 49.79, and 35.08% to WY, and 67.62, 48.69, and 45.76% to SY at daily, monthly, and yearly temporal scales. In view of the large basin scale, their contributions were about 53.10, 42.35, and 38% to WY; and 85.96, 44.53, and 39.83% to SY at daily, monthly, and yearly temporal scales.
The overall influence of weather parameters gradually decreases from daily, monthly, and yearly at all spatial scales for both WY and SY. On the other hand, the contributions of vegetation indices were more pronounced to yearly scale than monthly and daily scale at all spatial scales for both WY and SY. In fact, gradually increasing importance was observed from daily, monthly to yearly timescales in all spatial scales. The significance of baseflow to WY was felt to be higher (next to precipitation) to WY at all spatial-temporal scales, but in the case of SY, was suppressed by other predictors except at yearly timescales. The relative significance of snowmelt to WY and SY differs among spatial and temporal scales. In contrast, the significance was more pronounced to SY than WY.
AUTHOR CONTRIBUTIONS
P.T.L. acquired and interpreted the data, and analyzed and drafted the manuscript; P.K.P. helped in study conception and design; V.P. edited and critically revised the manuscript. All authors of this paper have directly participated in the writing, editing, planning, execution, and analysis of this study.
DATA AVAILABILITY STATEMENT
Data cannot be made publicly available; readers should contact the corresponding author for details.
CONFLICT OF INTEREST
The authors declare there is no conflict.