Sub-watershed prioritization using morphometric analysis, principal component analysis, hypsometric analysis, land use/land cover analysis, and machine learning approaches in the Peddavagu River Basin, India

Natural resources, such as water and land, are typically depleting because of population growth, urbanization, and so on; hence, desirable resource utilization is a precondition for sustainable growth. Natural resource managing has grown in importance for the planned expansion of water and land resources, as well as the prevention of land degradation in order to maintain environmental and ecological balance (Chakraborti 2003; Nookaratnam et al. 2005; Kudnar & Rajasekhar 2020). Remote sensing (RS) and geographic information system (GIS) are effective techniques for addressing specific natural resource management issues at the grassroots level of integrated land and water resource development (Diwakar & Mayya 2010; Kaya et al. 2019; Redvan & Mustafa 2021). To create proper water and land resource action plans, it is essential to first identify the area's geology and river system. Drainage basins are the fundamental water and land resource units managed, providing a managerial procedure for resource efficiency (Moore et al. 1994). Watersheds, for example, are regarded as a fundamental erosional environment aspect where water and land bodies interconnect. Watershed management is critical for protecting water and soil resources (Poongodi & Venkateswaran 2018). Geomorphology, land use/land cover (LULC), and hydrology information are extremely helpful in understanding the watershed's drainage system (Bhattacharya et al. 2020). It summarizes the river watershed, which is exceptionally useful in investigating watershed prioritization, environmental management, restoration, hydrological modelling, and conservation (Waiyasusri & Chotpantarat 2020).

The parameters of the watershed system are critical in the examination of various hydrological processes. Whereas a watershed is a crucial aspect of hydrology, watershed-scale morphometry parameters are much more preferable and beneficial than segment-scale morphometric evaluation (Poongodi & Venkateswaran 2018). Even without including taking into consideration soil sketch, morphometric analysis is an important approach for prioritizing sub-watersheds. Morphometry is a mathematical representation of the earth's surface topographic relief (Clarke 1966; Obi Reddy et al. 2004). A river basin's morphometric parameters are critical factors in watershed hydrology. They reflect the hydrological behaviour of the river basin and are useful in assessing the basin's hydrologic response. Watershed morphometric parameters and flood potential have relationships. For example, higher drainage density indicates faster runoff and a greater degree of channel abrasion for a particular amount of precipitation (Withanage et al. 2014; Samal et al. 2015). Watershed morphometric analysis results may be used as a strategic instrument in water managing, soil erosion sustainability, and watershed prioritization (Sreedevi et al. 2009; Sujatha et al. 2013; Iqbal & Sajjad 2014; Khanday & Javed 2016; Malik et al. 2019). Morphometric studies began in the mid-twentieth century, using a traditional method developed on the basis of conventional topographic maps (Horton 1945; Schumm 1956; Strahler 1964). The traditional method of analysing river morphology is a time-consuming and labour-intensive task. However, as geospatial technologies advance, more precise assessments can be performed with greater ease. Most researchers have studied morphometrical parameters of a watershed using GIS tools and RS (Aher et al. 2014; Mishra et al. 2018; Kadam et al. 2019).

Furthermore, several approaches to watershed prioritization have been developed, varying from simple analytical models to process-oriented physical-based models. For watershed prioritization, analytical as well as process-oriented models are challenging, information-driven, and nuanced. Nonetheless, they can be restored using watershed morphometric variables with less information requirements and more effective approaches. Several researchers have prioritized river basins using conventional methods such as the principal component analysis (PCA), used to minimize the dimension of morphometric parameters and discover variables that explain the majority of the variance seen in a wide range of parameters (Meshram & Sharma 2017; Arefn et al. 2020; Shekar & Mathew 2022c), weighted sum approach, and the fuzzy analytical hierarchy process (Rahaman et al. 2015; Farhan et al. 2018).

Many studies have been conducted on sub-watershed prioritization and management in order to assess erosion-prone regions based on morphometry, LULC, soil type, and so on (Sreedhar et al. 2016; Said et al. 2018; Adhami et al. 2019; Sharma & Mahajan 2020). Change in LULC is a major issue in global environmental change. Land use shift, land cover shift, and land use managing interventions have significantly transformed a huge portion of the earth's surface in order to meet rising natural resource demands (Biswas & Chakraborty 2016; Song et al. 2018; Gupta et al. 2020; Mathew et al. 2022). The watershed's LULC is another essential aspect to consider when prioritizing sub-watersheds (Javed et al. 2009; Puno & Puno 2019; Shekar & Mathew 2022a).

The hypsometric curve of the watershed, which is a non-dimensional evaluation of the proportions of the surface region of a watershed over a given height, is used to analyse the geomorphic processes that occur inside watersheds as well as landforms (Willgoose & Hancock 1998). Langbein (1947) was the earliest to utilize hypsometric analysis (HA) to convey the overall gradient and drainage watershed forms. Hysometry is the measurement and analysis of relationships between the allocation of heights across a land surface and watershed area. HA was utilized to distinguish erosional landforms at various phases of growth in a watershed. The hypsometric integral (HI), hypsometric curve, and hypsometric skewness are all numerical parameters in HA (Pike & Wilson 1971). Strahler (1952) discussed various watersheds and classified various shapes of hypsometric curves as a youth when the watershed experiences erosion and landslide, equilibrium or mature, when the watershed experiences less erosion and landslide than the youth stage, and finally old when the watershed experiences very little erosion and landslide. The region under the hypsometric curve shows the percentage of landmass eroded from the watershed, and based on this, the watershed is classified into three stages: old (HI is less than 0.3), equilibrium or mature (HI is between 0.3 and 0.6), and in equilibrium or youth (HI greater than 0.6). As a result, the HI helps to clarify the erosion that takes place in the watershed over geological time as a result of hydrologic procedures and soil degradation aspects (Strahler 1964).

A well-designed watershed management strategy can be used to control natural disasters like floods and drought as well as watershed issues like increased soil erosion, low yield, excessive overflow, and inadequate infiltration (Gajbhiye & Sharma 2017). Therefore, it is vital for decision-makers to be knowledgeable about the parameters of the watershed and the hydrological processes that take place within it. The watershed's parameters, which could be properly evaluated through geomorphometric analysis, have an impact on how the hydrological processes within it work. The information obtained from geomorphometric analysis of watersheds could be an important tool for prioritizing watersheds (Ahmed et al. 2017). Therefore, in addition to conventional geomorphometric analysis, it is necessary to implement innovative aspects like machine learning techniques to prioritize sub-watersheds with the most accuracy in order to manage and protect natural resources (Javed et al. 2011).

The significance of parameters can vary across different sub-watersheds due to their specific characteristics. Recognizing this variability, the study aimed to effectively prioritize sub-watersheds based on their susceptibility to soil erosion and subsequent flooding using five distinct methods: morphometric analysis, PCA, linear discriminant analysis (LDA) as a novel approach, LULC analysis, and HA. Through the comprehensive utilization of these methods, the study sought to gain a holistic understanding of the sub-watersheds and their individual characteristics. Morphometric analysis involved assessing quantitative measurements like linear, relief, and areal parameters, while PCA identified highly correlated morphometric parameters for ranking sub-watersheds. LULC analysis focused on land use and land cover patterns to prioritize the sub-watersheds, and the novel approach of LDA introduced morphometric parameters to enhance the prioritization process. Furthermore, HA incorporated HI values for prioritization. Notably, the study made a unique contribution by applying machine learning techniques, specifically LDA, to sub-watershed prioritization in the Peddavagu watershed. The objective of this research was to prioritize sub-watersheds by conducting morphometric analysis, PCA, LULC, HA, and LDA. Moreover, the study utilized these five methods to identify sub-watersheds that shared common priorities, adding a valuable dimension to the analysis.

In the current study, 18 morphometric parameters were reduced to 4 important components using the statistical programme for the social sciences (SPSS) software version 22. The rotated component matrix shows that each component considers one highly correlated parameter. Following the collection of four parameters, the sub-watershed with the highest result in the relief parameter as well as the linear parameter was ranked 1st, and so on. The sub-watershed with the lowest shape parameter result was ranked 1st, and so on. The result of the C_p is then computed. The C_p results of the sub-watersheds were used to categorize them into three groups: high, medium, and low.

The current research introduces a new technique to predict the priority of sub-watersheds utilizing an efficient method by the name LDA (Subasi 2020). It is a supervised machine learning technique used for classification problems. LDA accomplishes classification by determining the directions or linear discriminants that represent axes, hence maximizing class separation (Fu et al. 2020; Wang et al. 2022). LDA is a dimensionality reduction approach that keeps as much information as feasible while reducing the number of dimensions or parameters in a dataset. Since the volume of data to be trained grows as the dimensionality of the data grows, an exemplary parameter extractor is needed to adequately capture the complete data properties in a smaller number of variables.

LDA assumes that each parameter in the dataset is Gaussian distributed and that the value of each parameter varies around the mean, with the same variance. Therefore, prior to applying LDA to a dataset, it is necessary to standardize it. Standardizing the dataset is done by subtracting every data value by its mean and dividing the difference by its standard deviation. After standardization of data, the following are the steps to perform LDA:

Calculate the mean vectors of each dependent variable class.

• 1.

  Determine within-class and between-class scatter matrices.

• 2.

  Calculate eigen values and eigen vector for the matrix, where S_W is the scatter matrix within-class and S_B is the scatter matrix between-class.

• 3.

  The top k eigen values are chosen after sorting the eigen values in descending order.

• 4.

  Construct a new matrix with eigen vectors that map to the k eigen values.

• 5.

  Obtain the linear discriminants by taking the dot product of the data and the matrix.

The LDA method produces two highly correlated parameters. The sub-watershed with the highest result in relief as well as linear parameters ranked 1st, and so on. The sub-watershed with the lowest shape parameter result was ranked first, and so on. The value of the C_p is then computed. C_p of the sub-watersheds were used to categorize them into three groups: high, medium, and also low.

The prioritization of sub-watersheds depends on information from Sentinel-2 imagery. The resolution of LULC is 10 metres. The maximum result for the built-up area parameter was allocated a rank of 1, and so on. The minimum result was assigned a rank of 1 in the case of trees, crops, rangeland parameters, and so on. C_p results of the sub-watersheds were used to categorize them into three groups: high, medium, and low.

The elevation-relief ratio approach was utilized to calculate the HI values. The hypsometric curve and HI were calculated using the SRTM-DEM. The maximum HI value is ranked as 1, and so on. C_p results of the sub-watersheds were used to categorize them into three groups: high, medium, and also low.

Morphometric analysis of shape, linear, as well as relief parameters, was performed in this study using mathematical formulae. The specifics of the various parameters are discussed further below.

It represents the total number of streams in each ordered stream. According to Horton's (1945) laws of stream numbers, the number of stream segments of individual order forms an inverse geometric sequence that can be plotted against the order. In this study, SW10 and SW3 have the maximum and minimum stream numbers, respectively.

It is defined as the overall length of streams in a specific order. It is an important hydrological parameter of the area because it provides information about surface overflow parameters. Horton's (1945) law was used to calculate the length of the stream. SW6 and SW4 are the lengths of the largest and smallest streams in this study, respectively.

It is the proportion of stream channel of the given order (N_u) to streams of the next higher-order (N_u + 1). Horton (1945) utilized it as a relief as well as a dissertation index. Strahler (1957) illustrated that it tends to vary only slightly across regions, except for where geological regulation is intense. The R_b varies from order to order; these improprieties are influenced by the geomorphological growth of drainage watersheds (Strahler 1964). SW6 and SW3 have the highest bifurcation ratios, indicating a higher proportion of stream channels of a given order compared to streams of the next higher order in those watersheds. This suggests a more complex and branched stream network in SW6 and SW3. On the other hand, SW4 has the lowest bifurcation ratio, indicating a lower proportion of stream channels of a given order compared to streams of the next higher order in that watershed. This implies a less branched and hierarchical stream network in SW4 compared to the other locations studied in your research.

Horton (1945) identifies the length ratio as the ratio of the mean of order channels to the mean length of the next lower order channels that also appear to be consistent all through the watershed's consecutive orders. In this study, the maximum and minimum Rl are SW2 and SW8, respectively.

Stream frequency (Fs) refers to the number of stream segments per unit area (Horton 1945). It is a metric commonly used in hydrology and geomorphology to describe the density of streams within a given region. The value of stream frequency can provide insights into the drainage pattern and network complexity of an area. SW10 has the highest stream frequency value, indicating that it has the highest density of stream segments per unit area. On the other hand, SW3 has the lowest stream frequency value, suggesting that it has the lowest density of stream segments per unit area among the studied locations.

According to Horton (1945), it is calculated as the ratio of the overall length of streams of all orders per area, and it represents an indicator of the closeness of channel spacing. It is calculated as D_d = L_u/A, where L_u is the overall length of all streams and A is the watershed area. Based on the information you provided, SW10 has the highest drainage density value, indicating that it has a higher overall length of streams relative to its watershed area. This suggests that SW10 has a higher concentration of streams and a closer spacing between channels. On the other hand, SW2 has the lowest drainage density value, indicating that it has a lower overall length of streams relative to its watershed area and a more dispersed channel network compared to the other locations studied.

It is the overall number of stream channels of all orders per area perimeter (Horton 1945). Smith (1950) categorizes D_d into five textures: very coarse (less than 2), coarse (between 2 and 4), moderate (between 6 and 8), and fine (greater than 8). SW10 has the highest value in this study, while SW3 has the lowest.

It is approximately half of the reciprocal of D_d (Horton 1945). It refers to the length of water flow over land before it becomes concentrated in definite stream channels. SW2 has the highest value of the length of the overland flow. This indicates that water in SW2 tends to travel a greater distance over the land surface before it becomes concentrated in stream channels compared to the other locations studied in this research. On the other hand, SW10 has the lowest value of the length of the overland flow. This suggests that water in SW10 has a shorter distance to travel over the land surface before it becomes concentrated in stream channels compared to the other locations.

According to Faniran (1968), it is described as the ratio of F_s to D_d. A watershed with low D_d, texture, and intensity is vulnerable to flooding and erosion. SW10 has the highest drainage intensity value, indicating a relatively higher stream frequency compared to the drainage density. This suggests that SW10 has a higher concentration of streams in relation to its channel spacing, which can indicate a higher potential for flooding and erosion in the watershed. On the other hand, SW7 has the lowest drainage intensity value, indicating a lower stream frequency relative to the drainage density. This suggests that SW7 has a lower concentration of streams compared to its channel spacing, which may imply a lower vulnerability to flooding and erosion compared to the other locations studied in this research.

It is an important parameter that relates D_d to the physiographic growth of a watershed, allowing for the evaluation of drainage system storage capacity and, as a result, determining the ultimate degree of drainage growth in a given watershed (Horton 1945). It is greater and lesser at SW2 and SW8, respectively, in this study.

It is described as the inverse of the D_d with the dimension of length (Schumm 1956). In this study, the C_cm is larger and lower at SW2 and SW10, respectively.

It is calculated as the sum of F_s and D_d. Higher I_f results in lower infiltration rates and higher surface overflow (Faniran 1968). SW10 has the highest infiltration number, indicating a higher sum of stream frequency and drainage density. This suggests that SW10 may have a relatively lower infiltration rate and a higher potential for surface overflow compared to the other locations studied in this research. On the other hand, SW3 has the lowest infiltration number, indicating a lower sum of stream frequency and drainage density. This implies that SW3 may have a relatively higher infiltration rate and a lower potential for surface overflow compared to the other locations.

It is measured as the variation among the maximum and minimum elevations of the watershed. SW7 provides the most relief in this study, while SW2 provides the least.

It has been determined using the perimeter and watershed relief (Melton 1957). SW8 has a greater value in this study, while SW6 has a lesser value.

It is described as the ratio of a watershed's overall relief to its longest dimension parallel to the main drainage line (Schumm 1956). SW8 has a higher value in this study, while SW1 has a lower value.

It is the result of the greatest possible watershed relief and D_d (Strahler 1964). A basin with a low R_n is less prone to soil erosion, whereas a basin with a high R_n is highly susceptible to erosion due to the structural complexity of the terrain. SW10 has the greater value of ruggedness ratio, indicating a higher ratio of watershed relief to drainage density. This suggests that SW10 has a more structurally complex terrain, which can make it more susceptible to soil erosion compared to the other locations studied in this investigation. On the other hand, SW2 has the lower value of ruggedness ratio, indicating a lower ratio of watershed relief to drainage density. This implies that SW2 has a less structurally complex terrain, which may make it less prone to soil erosion compared to the other locations.

The watershed area is the watershed parameter for hydrologic layout because it represents the volume of water that can be created from precipitation. It is denoted by the letter ‘A’. The watershed's total area is 3,150 km². The maximum and minimum sub-watershed areas in the current study are SW5 and SW4, as shown in Supplementary Figure S1.

The watershed perimeter is the watershed's outer boundary that encloses its region. It is calculated across watershed categorized and can be used to determine watershed size and shape. As shown in Supplementary Figure S2, the maximum and minimum sub-watershed perimeters of the 13 Peddavagu basins are SW6 and SW4, respectively.

The distance determined along the main channel from the watershed outlet to the watershed divide is commonly used to define basin length. It is represented by the letter L_b. According to the current research, SW5 has the longest sub-watershed length, while SW4 has the shortest, as shown in Supplementary Figure S3.

It is the dimensionless ratio of the watershed area to the region of a circle with an equal circumference perimeter as the watershed. It expresses the degree of circularity of the basin (Miller 1953). The greater the circularity value, the greater the flood hazard at the outlet point during the peak time. In the present investigation, SW13 has a maximum R_c than SW6, which has a minimum R_c.

It is described as the ratio of the diameter of a circle of the equal region as the watershed to the maximum watershed length. Watershed slopes can be categorized utilizing the index of R_e, which includes circular (between 0.9 and 0.10), oval (between 0.8 and 0.9), less elongated (between 0.7 and 0.8), elongated (between 0.5 and 0.7), and more elongated (less than 0.5). SW4 has a higher R_e in this study, while SW5 has a lower R_e.

It is the ratio of watershed area (A) to watershed length squared (L_b). The longer the watershed, the lower the F_f value (Strahler 1964). A higher F_f watershed has a maximum peak stream flow in a short time (Horton 1945). SW4 has a higher form factor in this study, while SW5 has a lower form factor, as shown in Supplementary Table S2.

There are three types of morphometric parameters (shape, relief, and linear parameters). The erosion of soil has a direct effect on the linear as well as relief parameters (Nookaratnam et al. 2005; Javed et al. 2009). The watershed with the high result of linear as well as relief parameters has more erosion of soil. As a result, the sub-watershed with the highest result in relief as well as linear parameters ranks 1st, and so on. The shape parameters are related to soil erosion in an indirect way. The soil with the lowest shape parameter results in a watershed, the more erosion of soil. Therefore, the sub-watershed with the lowest shape parameter result is ranked 1st, and so on. C_p value is calculated after assigning a rank to each parameter. When all of the ranks of SW1 are combined and divided by 18, C_p is 7.28. The process was repeated for the other sub-watersheds.

The information available is for 13 sub-watersheds, with 18 parameters representing each sub-watershed. The priorities of sub-watersheds have been enlisted as low, medium, and high within the dataset. The objective of this work is to prioritize the sub-watersheds based on their parameters, which is achieved by LDA-based parameter extraction and support vector machine (SVM) classification. The method has been implemented in the Jupyter integrated development environment (IDE) on a desktop personal computer running at 3.4 GHz with 8 GB of random-access memory.

LDA has determined the eigen value/vector of the product of scatter matrices and it has been found that the first two eigen values corresponding to mean bifurcation ratio, and mean stream length ratio have significant values. Since LDA extracts only N − 1 parameters for N-class data, we select the first two most significant eigen values and their eigen vectors for further calculation. Thus, the first eigenpair is the most informative one and is selected to form a one-dimensional parameter space based on this eigenpair. A percentage of variance is also calculated to ascertain the significance of the obtained eigen values. Supplementary Table S5 shows the eigen value and explains the variance in percentage of each corresponding eigen vector. The linear discriminants are then obtained by multiplying the standardized data with the eigen vector matrix. Table 4 shows the two linear discriminants representing the entire 18 parameter data to prioritize the sub-watersheds.

To determine the accuracy of classification, we have used a confusion matrix. A confusion matrix is a table (refer to Table 5) that shows how well a classification model (or ‘classifier’) performs on a set of test data for which the real values are known. The accuracy obtained for prediction based on our test data is 75%. The obtained confusion matrix is plotted below.

Table 5

Confusion matrix

View Large

The important morphometric parameters derived from LDA are the mean bifurcation ratio and mean stream length ratio. As a result, these factors are used to prioritize the 13 sub-watersheds of the Peddavagu watershed.

For the parameters of trees, crops, and rangelands, the lowest result received a rank of 1, and so on. In the case of the built-up area parameter, the highest result received a rank of 1, and so on. As shown in Table 7, the C_p result was calculated for sub-watershed prioritization.

In reality, the importance of parameters may vary across different sub-watersheds depending on their specific characteristics. Therefore, the study used five methods – morphometric analysis, PCA, LDA, LULC, and HA – to prioritize sub-watersheds. Using five methods helps to identify sub-watersheds that have different characteristics and prioritize them based on their specific susceptible to soil erosion and subsequent flooding. In the sub-watershed priority assessment in Peddavagu watershed, different methods may yield varying priority levels. However, when three out of the five methods consistently indicate a similar priority for a particular sub-watershed, it is reasonable to consider that priority. For instance, if SW2 is consistently assigned a high priority by three out of the five methods, it can be considered as having a high priority based on the convergence of those methods. By identifying the overlapping results and considering the consistency among multiple methods, a more robust and reliable assessment of sub-watershed priority can be achieved. The most similar sub-watersheds were identified using morphometric analysis, PCA, LDA, LULC, and HA. Based on the results presented in Table 9, SW2, SW3, SW5, SW6, and SW10 have been determined to have a high priority. SW13 has been assigned a medium priority, while SW11 has been categorized as having a low priority. Ultimately, SW10 should be allocated a high priority in the list of high priority sub-watersheds for conservation measures and land management practices. This decision is based on the fact that SW10 consistently exhibits a similar priority across all five distinct methods. This alignment across multiple assessment approaches highlights the importance of prioritizing SW10 for focused conservation efforts and effective land management practices.

The traditional method of watershed prioritization includes activities like digitizing contours in the toposheet or thorough field survey, digitizing stream networks, segmenting streams to separate stream orders, and manually measuring stream length. It also involves counting streams, calculating watershed area and perimeter, as well as other related variables to calculate morphometric parameters. As a result, this traditional method was time-consuming, expensive, and labour-intensive. GIS can be used to establish watershed priorities in a less expensive, less time-consuming, and less labour-intensive manner. For instance, watersheds are used in forestry, agriculture, recreation, infrastructure development, and other activities. Therefore, it is important to know the purpose of the watershed in its prioritization. In this paper, five techniques have been used for the Peddavagu River Basin prioritizing, wherein a novel strategy is introduced to rank the priority of sub-watersheds using a machine learning algorithm based on LDA written in Python script. The results indicate that SW2, SW3, SW5, SW6, and SW10 have been identified as high priority sub-watersheds. SW13 has been assigned a medium priority, while SW11 has been categorized as low priority. Considering the consistent results obtained from all five distinct methods, it is recommended to assign a high priority to SW10 in the list of high priority sub-watersheds for conservation measures and land management practices. The convergence of assessments across multiple approaches underscores the significance of prioritizing SW10. By doing so, focused conservation efforts and effective land management practices can be directed towards SW10, ensuring the optimal allocation of resources and maximizing the potential for positive environmental outcomes. Among the sub-watersheds identified as high priority, immediate priority is assigned to SW10.

When soil erosion occurs, it leads to the loss of topsoil and exposes the underlying layers, which are often less permeable. As a result, the ability of the soil to absorb and retain water decreases, leading to increased surface runoff during rainfall events. When heavy rainfall occurs on eroded or compacted soil, the water is unable to infiltrate the ground effectively. Instead, it runs off the surface, accumulating in streams, rivers, and other water bodies. This increased volume of water can overwhelm the natural drainage system, causing rivers to overflow their banks and resulting in flooding. Additionally, the sediments eroded from the soil can be carried by the runoff and deposited in water channels, reducing their capacity and further exacerbating the risk of flooding. There are several measures that can be implemented to mitigate soil erosion and subsequent flooding in high priority sub-watersheds (SW2, SW3, SW5, SW6, and SW10). These include: Promoting sustainable land management practices; Constructing soil conservation structures: Installing check dams, gabions, and percolation tanks can help intercept and slow down water flow, reducing erosion and sedimentation. These structures help to retain water within the landscape, promoting infiltration and minimizing runoff; Managing water flow: Implementing measures such as flood control reservoirs, channel maintenance, and riverbank stabilization can effectively manage water flow and reduce the risk of flooding. These actions help to regulate and control the movement of water, mitigating the impact of heavy rainfall events. By implementing these measures, it is possible to mitigate soil erosion and subsequent flooding, leading to improved land and water management, enhanced agricultural productivity, and reduced risks to human settlements and infrastructure.

Prioritization of sub-watersheds within a larger watershed is an essential step towards effective watershed planning, implementation, and allocation of resources. The current research aims to introduce a novel strategy to determine the priority of sub-watersheds by utilizing an efficient tool written in Python (the LDA method). The study employed RS and GIS techniques to improve the precision and cost-effectiveness of drainage basin delineation and extraction. For the prioritization of the Peddavagu river basin, five methods were utilized and prioritized. Utilizing morphometric analysis, higher priorities were designated for SW5, SW8, SW10, and SW13; using PCA, higher priorities were assigned to SW5, SW6, SW7, SW9, and SW10. HI has given high priority to SW1, SW2, SW3, SW4, SW10, and SW11. Utilizing LULC examination, higher priorities have been designated for SW4, SW6, SW7, SW10, SW1, SW2, SW3, and SW12. Utilizing LDA, higher priorities have been designated for SW2, SW3, SW5, SW6, and SW10. Based on the results, SW2, SW3, SW5, SW6, and SW10 have been determined to have a high priority. SW13 has been assigned a medium priority, while SW11 has been categorized as having a low priority. Among the sub-watersheds identified as high priority, immediate priority is assigned to SW10. As a result, the high priority sub-watershed may be used by decision-makers for conservation measurement and land management practices for planning and development as well as to protect the land from further erosion and flooding.

A limitation of the study is the limited availability of data, which restricts the exploration of deep learning models in the research. However, as a future scope, there is a plan to expand the data collection efforts to include more micro-level watersheds. This expansion will enable the continuation of the classification study using deep learning models, offering the potential for enhanced analysis and insights.

The authors would like to thank the anonymous reviewers and editor for their instructive comments, which helped to improve this paper. In addition, the authors wish to thank the USGS Earth Explorer for making available the satellite data. Finally, the authors also want to thank ESRI for providing land use land cover data.

P.R.S.: Conceptualization, Methodology, Software, Data curation, Writing – original draft. A.M.: Supervision, Visualization, Investigation, Writing – review & editing. A.P.S.: Data curation, Writing – review & editing. V.P.G.: review & editing.

There was no funding for this project.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.