Keywords

3.1 Introduction

In India, the continuous demand for economic growth coupled with the population explosion has resulted in substantial land use/land cover (LULC) changes in the past century (Tian et al. 2014). The ramifications of LULC changes include influencing ecosystem services, altering hydrological components, triggering and intensifying the natural hazards, complicating the hydroclimatic predictions, affecting quantity and quality of available water resources, etc. (Aadhar et al. 2019; Astuti et al. 2019; Bahita et al. 2021; Chen et al. 2020; Hengade and Eldho 2016; Saputra and Lee 2019; Sharma et al. 2020; Singh et al. 2020; Swain et al. 2018, 2019a, b; Talukdar et al. 2020; Tripathi et al. 2020). The impacts of droughts are more severe, where the LULC is mostly dedicated to agriculture (Swain et al. 2017, 2020a, 2021a, b, b). An improved understanding of the LULC and climatic changes of a particular area can be pivotal for effective policy framing, specifically in water resources planning and management (Anand et al. 2018; Himanshu et al. 2018, 2019; Dayal et al. 2019, 2021; Guptha et al. 2021; Kalura et al. 2021; Sahoo et al. 2021). Due to all these reasons, a detailed assessment of LULC change patterns has become necessary, which is typically carried out by analyzing historical LULC changes through multi-temporal remotely sensed images. Several research works have been carried out in the last few years to investigate the LULC changes, their future predictions, and consequential effects (Dutta et al. 2019; Liping et al. 2018; Palmate et al. 2017a, b; Pandey and Khare 2017; Pandey and Palmate 2018; Rimal et al. 2017; Rwanga and Ndambuki 2017; Singh et al. 2018; Tran et al. 2017).

The Himalayan catchments have immense significance from the point of view of water resources planning and management (Singh and Pandey 2021; Swain et al. 2021c). However, the natural and anthropogenic changes and their effects upon these reservoirs are often not explored, mainly due to limitations of data availability. With the advancement of remote sensing and geospatial technologies, the LULC information of these catchments have become easily accessible. Recently, Mishra et al. (2020) used Landsat 5 and Sentinel 2A for supervised LULC classification over the Rani Khola watershed located in the Sikkim Himalaya, India. They reported a series of complicated changes in LULC over the watershed during 1988–2017. Therefore, this study aims to carry out a detailed assessment of LULC changes over a Himalayan catchment considering multi-temporal satellite images and supervised classification. Further, it is wise to cross-check the LULC classification of the recent period by ground truth verification. In this regard, the details of the study area, methodology, results and discussion, ground-truthing information, and the conclusions derived from this study are presented in the subsequent sections.

3.2 Study Area and Data

The Tehri catchment located in the state of Uttarakhand, India, is considered as the study area. The catchment covers an area of 7295 km2. The catchment lies in the lower Himalayan region and thus is associated with very steep slopes. This is the main reason for the very high velocity of flow, which consequently leads to mass erosion. The location of the study area is shown in Fig. 3.1. The maximum and minimum temperatures over the catchment are 36 °C in summer and 0 °C in winter, respectively. A good amount of rainfall is received all over the catchment, though there are remarkable spatial variations (Kumar and Anbalagan 2015; Rautela et al. 2002).

Fig. 3.1
figure 1

Location map of the study area

Full size image

The satellite-based imageries were collected from the website of United States Geological Survey (USGS) EarthExplorer. While the image for 2008 was taken from Landsat 5, the images for 2014 and 2020 were taken from Landsat 8 and Sentinel 2 satellites, respectively.

3.3 Methodology

The extraction of LULC information from the imageries is carried out by remote sensing and GIS techniques. The two softwares, viz. ERDAS IMAGINE and ArcGIS are widely used to carry out different image processing and geospatial operations, which were also used in this study. The overall methodology adopted for LULC classification and change detection is presented in Fig. 3.2. First of all, the satellite images for different years are collected and their preprocessing is carried out. The area of interest may not fall under a single satellite image, and the collected data may be available in different file formats or projection systems. Therefore, stacking, mosaicing, and adjusting the coordinate systems, etc., were performed using ERDAS IMAGINE and ArcGIS 10.2.4 softwares. Moreover, for better interpretation of the imageries, false-color composites, contrast stretching, and image enhancement operations were also carried out.

Fig. 3.2
figure 2

Overall methodology for analyzing the multi-temporal LULC changes

Full size image

The next step is the supervised classification, where the sample pixels in an image representing particular classes are selected by the user based on his/her knowledge. These are called the input classes or the training sites. The classification of all the remaining pixels can be carried out using these training sites through an image processing software. The reflectance of each pixel is the core of the image classification. For a particular class, the higher the number of training sites, the better is the precision of the classification. Therefore, LULC classification is based on the concept of segmenting the spectral domain into distinct ground cover classes. In this study, the study area is divided into eight different LULC classes, viz. open forest, dense forest, water bodies, shrubland, agricultural land, settlements, barren land, and snow covers.

The next step is the accuracy assessment, whose purpose is to validate the classification results. This justifies the utility of the classified maps for further applications. This can be achieved by ground truth verification in terms of field visits. However, it is practically infeasible to collect information on the entire study area through field visits. Moreover, ground-truthing of the LULC for past years is almost impossible. Therefore, Google Earth images for a particular period can be used as a reference for validating the classifications. Using Google Earth as the reference is convenient and requires minimal cost. Nevertheless, for assessing the classification accuracy relevant to the recent period, it is always wise to conduct a field visit to some portions of the study area.

In this study, 400 random points from various classes were taken across the LULC maps. Considering their corresponding points from Google Earth image or ground-truthing information, a confusion matric is prepared. The producer accuracy (AP) and the user accuracy (AU) are calculated for each LULC class, whereas the overall classification accuracy (AO) and the Kappa coefficient (KC) are calculated to assess the LULC classification of the entire area. AP, AU, and AO are expressed in percentage with a range from 0 to 100. On the other hand, KC ranges from 0 to 1. The formula for these accuracy measures is provided in Table 3.1. The detailed procedure of accuracy assessment may be referred from literature (Manandhar et al. 2009; Rwanga and Ndambuki 2017; Sarkar 2018). Following the accuracy assessment, the spatiotemporal LULC changes are detected and analyzed.

Table 3.1 Measures for accuracy assessment of classification
Full size table

3.4 Results and Discussion

The multi-temporal remotely sensed imageries were used for the detailed LULC classification. The classified maps of the years for 2008, 2014, and 2020 are presented in Fig. 3.3. The spatial variation of LULC classes (open forest, dense forest, water bodies, shrubland, agricultural land, settlements, barren land, and snow covers) in different years can be visualized clearly.

Fig. 3.3
figure 3figure 3

Multi-temporal supervised classification of LULC over Tehri catchment

Full size image

The areal coverage details of the individual classes in 2008, 2014, and 2020 are presented in Table 3.2. For all three years, it can be observed that the dense forest is the most dominant LULC class over the catchment, followed by shrubland and barren land. A significant portion of the catchment is covered by snow, which is inherent in the Himalayan conditions. The settlement constitutes the least portions of the catchment among all the classes.

Table 3.2 Areal coverage details of LULC classes over the Tehri catchment
Full size table

From Table 3.2, the temporal changes in individual LULC classes over the Tehri catchment can be noticed. The percentage of catchment area under each of these classes during 2008, 2014, and 2020 is also presented. The dense forest has decreased by nearly 60 km2 from 2008 to 2014, whereas there is no change between 2014 and 2020. On the other hand, there is a clear increase in open forests from 2008 to 2014. Hence, it can be fairly inferred that the canopy density has reduced over the years. As a result, the dense forests have been converted to open forests. An increase in the settlement is also observed. These may be attributed to anthropogenic activities, resulting in aggravated soil erosion. No appreciable change in agricultural land is observed between 2008, 2014, and 2020. There is a decrease in barren lands over the years. The shrubland has witnessed a remarkable increase from 2008 to 2014. Similarly, there is a clear increase in the snow covers from 2014 to 2020. The LULC changes were drastic from 2008 to 2014 over most of the classes, whereas there is hardly any change in LULC classes from 2014 to 2020, excluding barren land and snow (Table 3.2).

The results of the accuracy assessment are presented in Table 3.3. Considering the Google Earth images and the information collected during the field visits, the accuracy measures were estimated. The AP and AU values for individual classes are quite encouraging (Table 3.3). The results of the accuracy assessment reflect a precise identification of LULC classes over the catchment for all three periods. The overall classification accuracy (Kappa coefficient) for 2008, 2014, and 2020 are found to be 83.6 (0.821), 82.5 (0.803), and 88.9 (0.873), respectively. These high values of AO and KC confirm that the LULC changes reported in this study can be justifiably regarded as accurate and, hence, are utilizable for further applications.

Table 3.3 Accuracy assessment results of LULC classification
Full size table

3.5 Ground Truth Verification

It is always wise to cross-check the supervised detailed LULC classification by ground truth verification. Therefore, a field visit was made to some portions of the catchment area to collect the land use/land cover observations along with their appropriate location details so that it would be helpful for proper validation and accuracy assessment. Moreover, it aimed to obtain relevant information from the local people regarding the causes of LULC changes, other hydrological problems over the study area (particularly soil erosion), and steps taken to combat those issues. This can be very helpful in preparing an effective catchment area treatment plan. Thus, a field visit was made to accomplish the aforementioned objectives. The details of the locations are provided in Fig. 3.4 and Table 3.4.

Fig. 3.4
figure 4

Locations covered during the field visit

Full size image
Table 3.4 Location details of the points covered during the field visit
Full size table

Due to the constraints of cost and time, only a portion of the catchment was covered during the field visit. The photographs collected during the field visits along with their location details were useful for the accuracy assessment of the LULC classification pertaining to 2020. A few photographs are presented in Fig. 3.5. As the catchment is prone to soil erosion, several protection measures were adopted, which is evident from some of the photographs (Fig. 3.5).

Fig. 3.5
figure 5

Photographs of some locations in the Tehri catchment collected during the field visit

Full size image

3.6 Conclusion

The detailed supervised LULC classification for the Tehri catchment is carried out using the Landsat 5, Landsat 8, and Sentinel 2 data pertaining to 2008, 2014, and 2020, respectively. The various land covers that the catchment area is classified into are water bodies, agricultural land, dense forest, open forest, shrubland, settlement, barren land, and snow. The LULC changes are found to be drastic from 2008 to 2014 over most of the classes, whereas no appreciable changes in classes are found from 2014 to 2020 except for snow and barren lands. Comparing 2008 and 2020, an increase in the open forests, water bodies, shrubland, snow, and settlement is observed, whereas a decrease in dense forests and barren land is noticed. The accuracy assessment results confirm that the LULC changes reported in this study are justifiably accurate and utilizable for further applications.