Watershed prioritization of Kailali district through morphometric parameters and landuse/landcover datasets using GIS

Watershed prioritization is considered an important tool for soil and watershed management. This study focuses on the watershed prioritization of the Kailali district in terms of soil erosion, considering morphometric parameters and land use/landcover (LULC) datasets using GIS. ALOS DEM of 30 m resolution was used to delineate sub-watersheds and calculate linear, areal, and relief morphometric parameters. Similarly, Esri LULC 2021 (Sentinel-2 imagery at 10 m resolution) was used to calculate LULC parameters. An integrated approach of Principal Component Analysis (PCA) and Weighted Sum Analysis (WSA) was used for prioritization. PCA was used to reduce selected parameters, calculate the correlation matrix, and define the significant parameters. WSA was used to define weightage value, and Compound Value (CV) was calculated for the ranking of sub-watersheds. 22 sub-watersheds with at least 3rd order stream and 15 parameters were selected for prioritization. PCA integrated with WSA was found to be effective for prioritization. The findings showed that about 61.58% of the watershed area is in the high-priority category, suggesting those areas are at a higher risk of erosion. Therefore, different land rehabilitation programs and bioengineering techniques should be focused on the sub-watershed of high-priority categories followed by medium and low-priority categories to control further soil erosion. The adopted methodology of prioritization can also be performed for multi-hazard mapping.


Introduction
Nepal has fragile geography and diverse topography and climatic conditions. It is prone to natural disasters such as floods and landslides making it the 20th top most disaster-prone country in the world [1]. Chure-originated rivers like Mohana and its tributaries flow across the Kailali district. The flow of these rivers is dependent on monsoon precipitation and their flow level depletes significantly during the non-monsoon period [2]. The short duration of intense rainfall in these areas can cause sudden flash floods in the monsoon season which carry a huge volume of rocks, and debris [3] which deteriorate the soil and watershed health [4].
It is not feasible for the management of a watershed as a whole. So, watershed prioritization is necessary for the effective management of watersheds. Watershed prioritization is the process of identification and ranking of environmentally degraded Sub- watershed for the prevention of soil erosion, management of floods, drought, and application of varying levels of conservation treatment [5,6]. The prioritization uses the key issue of the watershed problem, including erosion, as the principal consideration [7].
Various studies [6][7][8][9][10][11] have been conducted using morphometric parameters and land use and landcover datasets to define priority level of a watershed using GIS technology and tools. Morphometric parameters provide an accurate quantitative description of a basin geometry [12]. It is the mathematical measurement of the configuration of the surface, shape, and dimension of the earth's landforms and its analysis [13] that provide knowledge about the characteristics of the watershed and the hydrological process within it. The morphometric parameters can be classified into linear, areal, and relief aspects. The linear aspect consists of bifurcation ratio, stream length, mean stream length, and stream length ratio; the areal aspect consists of stream frequency, drainage density, drainage texture, constant of channel maintenance, infiltration number, length of overland flow, and shape parameters such as form factor, elongation ratio, circularity ratio, and gravelius compactness coefficient. Similarly, the relief aspect consists of relief ratio, ruggedness number, and dissection index [7,9,10,14]. These parameters can be obtained and computed using DEM data with GIS software and different mathematical formula suggested by Refs. [12,[15][16][17][18][19][20]. Similarly, land use and land cover have also been considered important components for prioritization as they influence the hydrological process by influencing soil erosion and other factors [8,21,22]. Different analysis techniques like Principal Component Analysis [9], Weighted Sum Analysis [10], Fuzzy Analytic Hierarchy Process [23], and Multi-Criteria Ranking Method [5,24] have been applied for analysis in the prioritization. Integration of Principal Component Analysis (PCA) and Weighted Sum Analysis (WSA) is robust enough to define significant and more effective parameters for the watershed prioritization [7]. PCA was used to define the significant parameters whereas WSA was to determine the weights for significant parameters and the compound values for priority ranking.
So far, no research study has been conducted in the study area to analyze and prioritize the watershed. As prioritization of watersheds for their relative stability in terms of land erosion is an effective tool for soil and watershed management, a study for the prioritization of watersheds in the study area is necessary. This research is an attempt to prioritize the watersheds of the Kailali district using morphometric parameters and land use/landcover in terms of soil erosion through the integration of PCA and WSA approach for soil and water conservation practices. The study provides baseline information for making informed decisions toward effective soil and watershed management.

Watershed delineations
First of all, DEM preparation was done for sub-watershed delineation using O'Callaghan and Mark model i.e., fill the sink and determine the flow direction raster using the D8 flow direction algorithm. Sub-watersheds were created using basin functions under hydrology tools in spatial analyst tools and vectorized [25]. For the extraction of the drainage network; a flow accumulation raster was created using a flow direction raster as an input. After a process of iterative and careful visual comparison between drainage networks from google earth online and streams generated with various cell thresholds using a raster calculator, a value of 1000 was found appropriate and selected for the study. The extracted drainage network was ordered based on Strahler's classification through the stream order function. The Sub-watersheds with at least 3rd order streams were selected for further analysis in the study. 22 Sub-watersheds were selected as it meets the required criteria as shown in Fig. 2. Morphometric analysis of the selected 22 Sub-watersheds was done using ArcGIS.

Landuse and landcover parameter
Esri 2021 Lulc map was clipped and projected to WGS 1984 UTM zone 44 N. The Lulc of the study area is shown in Fig. 3. Lulc raster map was converted to vector polygon and Lulc for each selected Sub-watershed was extracted. The Lulc classes with a similar relationship with erodibility were merged and two Lulc classes were defined to have a strong influence on the hydrological process in the watershed: trees (trees + rangelands) and crops (crops + barren lands + built-up). Flooded vegetation classes were excluded as they may not affect the prioritizations significantly due to their small areas. Fig. 4 Summarizes the entire methodology.

.Morphometric parameters calculations
Different basic, linear, areal, and relief morphometric parameters were calculated using the relation presented in Table 1.

Principal Component Analysis (PCA) and Weighted Sum Analysis (WSA)
PCA was used for reduction in the selected parameters to find the correlation matrix and to obtain Principal components (PCs) as well to get the most significant parameters. Dimension reduction tool in SPSS software was used for PCA calculations. Initially, KMO and Bartlett's Test was performed to measure sampling adequacy for overall datasets and check if data is suitable for data reduction for the applicability of PCA. Then correlation matrix, first-factor loading matrix, and rotated factor loading matrix were used for further analysis. Eigenvalue>1 was used to select PCs based on the Kaiser criterion and Varimax rotation was used as a rotated factor loading [7]. The significant parameters were calculated based on rotated factor loading. The morphometric parameters and LULC parameters of each sub-watershed were selected for PCA. Preliminary ranking of significant parameters (PRsp)was done on the relationship to the erodibility such that Fs, Dd, Dt, If, and Lo have a direct relationship with erodibility i.e., a higher value indicates more erodibility and higher ranking and vice versa whereas shape parameters and Ccm have indirect relationship i.e., a low value indicates high erosion and higher ranking and vice versa [7,9,10]. Similarly, crops have a direct relationship and trees have an indirect relationship with erodibility [7].
A weighted sum approach was applied to those significant parameters. The weighted value of significant parameters (W sp ) was calculated using the correlation matrix of significant parameters as explained in equation (1): The final priority ranking was done based on compound value (CV) such that the lowest value CV was given 1 priority ranking, the second-lowest value was given 2 priority ranking, and so on for all Sub-watershed. CV was calculated using equation (2): These Sub-watersheds were then classified into high, medium, and low categories of risk of erosion for priority mapping.

Morphometric and landuse landcover parameters
The morphometric parameter values of selected sub-watersheds are shown in Tables 2-4.

Linear parameter:
All the Sub-watershed has a low value of mean bifurcation ratio ranging from 1.58 to 3.67. Only sw2 has the highest value i.e., 10.34 which signifies low permeability and high erosion of sw2 [10].
2. Areal parameters: Table 8 shows low stream frequency (Fs) ranging from 0.43 to 0.83. Low drainage density (Dd) suggests a very coarse texture of Sub-watersheds [27]. The constant of channel maintenance (Ccm) suggests the least erodible nature of sub-watersheds based on Schumm's classification [28]. The infiltration number (If) value ranges from 0.39 (sw22, least erosive) to 0.74 (sw8, most erosive) [14]. Length of overland flow (Lo) value ranges from 0.44 to 0.71; sw1 being the least erodible and sw12 and sw13 as the most erodible Sub-watersheds. The form factor (Rf) value ranges from 0.10 to 0.89 indicating an elongated shape except for sw18 which is circular [29]. The elongation ratio indicates sw14 and sw18 as circular shapes; sw16 and sw15 are oval; and other Sub-watersheds are elongated in shape [30]. The circularity ratio (Rc) value ranges from 0.15 to 0.62 indicating sw17 is the most circular whereas sw1 is the most elongated [10]. Similarly, all the Sub-watershed are in the youth stage except sw17 which is in the mature stage of topography [11]. The 1.27 to 2.61 value of gravelius compactness coefficient (Cc) suggests sw17 as the most elongated Sub-watershed and erosive [27]. 3. Relief parameters: Table 8 shows relief ratio (Rr) value ranges from 0.002 to 0.33; sw18 with high relief and steep slope, indicating the most erosive nature [30]. Similarly, the dissection index (DI) value ranges from 0.23 to 0.92 indicating sw8 as being the least erosive and sw7 and sw3 as being the most erosive nature [14]. The ruggedness number (Rn) ranges from 0.05 to 1.60 indicating sw8, sw4, sw9, sw13, sw17, sw14, sw18, and sw16 has low Rn value and hence are less vulnerable to soil erosion while other Sub-watersheds have higher  [20] Source: [7,10,14].   values with sw7 being more vulnerable to erosion [31].

Landuse landcover
The landuse landcover was represented as percentage coverage to each Sub-watershed. Table 8 shows the percentage area of merged Lulc classes for each Sub-watershed.
Intercorrelation among the geomorphic parameters: The correlation matrix of selected 15 geometric and landuse parameters was obtained using SPSS statistics 23 software. The correlation coefficient >0.9 suggests the parameter is strongly correlated, the correlation coefficient >0.75 suggest good and the correlation coefficient >0.6 suggest a moderate correlation.
Total variance or first-factor loading: PCA application on the selected parameters resulted in 4 principal components (PCs) Fig. 5. These components explained about 92.858% of total variance with Eigenvalues greater than 1 Table 6.
The correlation explained that some of the parameters have a strong correlation, some have a good and some have a moderate correlation, while some do not correlate at all with any of the components. So, the rotated matrix is necessary to identify the significant parameters.
Rotation of 1st factor loading matrix: The rotated factor loading ( Table 7) shows that the first component has a strong correlation with Rf, Rc, Re, and Cc; the second component has a strong correlation with Rbm; the third component has a strong correlation with Rn and DI; and fourth component has good correlation with Fs. Thus, Re, Rbm, Rn, and Fs were selected as the most important parameters for watershed prioritization.
Preliminary ranking of significant parameters: Preliminary ranking of significant parameters was done based on their relationship with erodibility. Rbm, Rn, and Fs have a direct relationship with erodibility and Re has an indirect relationship.

Weighted Sum Analysis (WSA)
Wsp was calculated using equation (1). Table 8 shows an intercorrelation matrix of significant parameters and weighted sum parameters of different parameters. Table 9 shows the preliminary ranking, CV values, and final priority ranking based on the Wsp and PRsp for each Sub-watershed. CV value was calculated using equation (2). The Sub-watershed having the lowest CV value i.e., sw7 was ranked 1 whereas the Subwatershed having the highest value i.e., sw17 was ranked 22 (Fig. 6).

.Prioritization of sub-watershed using PCA-WSA
The final priority ranking suggests that sw7 is at the highest risk of erosion whereas sw17 is at the lowest risk of erosion. Those with the highest ranking require the highest priority for soil conservation measures than a medium and low priority. Similarly, about 163511.22 ha (i.e., 61.58%) of Sub-watersheds fall under high priority categories, whereas about 77390.13 ha (i.e., 29.15%) fall under medium categories and about 24603.85 ha (i.e., 9.27%) falls under low priority categories (Table 10; Fig. 7).

Data quality issue
This study uses O'Callaghan and Mark models with ALOS DEM of 30 m resolution for basin and drainage network extractions. The resulting drainage network doesn't lie along the natural river network in some parts of southern flat areas. Similar issues of undefined flow during automated extraction of drainage networks in flat areas using DEM are also discussed by Ref. [33]. These issues are unavoidable because flat areas are connected regions with no internal gradients and the depression filling process in DEM actually increases the size and number of flat surfaces in DEM [34]. Therefore, the basins and streams network derived from DEM may not agree with ground features. Different algorithms have been generated by authors [35,36] to reduce this issue to some extent.

Comparison of approaches of prioritization
Three approaches of prioritization i.e., without PCA or WSA (general method), with PCA only (PCA method), and using PCA-WSA approaches were compared. CV values for the general method were calculated by taking the average of the selected parameter's rankings and CV values for the PCA method were calculated by taking an average of the significant parameter's ranking. Similarly, the priority ranking for each method was calculated based on CV values. When sub-watersheds were ranked using the general method, different sub-watersheds were found to have a common ranking. When ranked using PCA only, sw1 and sw11 as well as sw20 and sw21 were found to have a common ranking and when ranked using PCA-WSA approaches unique ranks were observed as shown in Table 11.

Conclusion
Watershed prioritization plays a key role in soil and watershed conservation. It helps in the identification and ranking of different degraded watersheds or sub-watersheds into different risk categories which can be used to prioritize the conservation treatments and budgets effectively. This study was conducted for the prioritization of watersheds of the Kailali district in terms of erosion risk. ALOS DEM and LULC maps were used for the calculation of morphometric and Lulc parameters. An integrated PCA-WSA approach was used for prioritization, where PCA was used to calculate the significant parameters and WSA was used in ranking these parameters based on CV values. Elongation ratio (Re), Mean bifurcation ratio (Rbm), and Stream frequency (Fs) were found to be significant parameters. Sub-watershed 7 was found to be at the highest risk of erosion whereas Sub-watershed 17 was at the lowest risk of erosion. About 61.58% of the area of Sub-watersheds falls under high-priority categories, 29.15% of the area falls under medium categories and 9.27% of the area falls under low-priority categories. Different land rehabilitation programs and bioengineering techniques should be focused on the Sub-watershed of high-priority categories followed by medium and low-priority categories to control further soil erosion.
A comparison of different approaches of prioritization was also performed to identify their effectiveness in prioritization. The integrated approaches of PCA-WSA were found to be more effective than the general methods, and the method used PCA only. Future   *Strong correlation (r > 0.9), ** Good correlation (0.90≥ r > 0.75), *** Moderate correlation (0.75 ≥ r > 0.60).     studies can be performed for other issues such as flash floods, groundwater potentiality, landslide susceptibility, and drought using the adopted methodology.

Author contribution statement
Susil Ojha: conceived and designed the experiments; performed the experiments; analyzed and interpreted the data; contributed reagents, materials, analysis tools or data; wrote the paper.
Lila Puri: conceived and designed the experiments; analyzed and interpreted the data. Suraj Prasad Bist, Arjun Prasad Bastola, Bishwabandhu Acharya: analyzed and interpreted the data; wrote the paper.

Funding statement
This work was supported by Tribhuvan University.

Declaration of competing interest
The authors declare no conflict of interest.

Table 11
Comparative approaches of priority ranking.