The land is a complex and dynamic mixture of features that includes geology, topography, hydrology, soils, microclimates, and communities of plants and animals that are always interacting, because of weather change and human activity [1]. There is pressure on LULC due to the growing population and increasing socio-economic necessities, unplanned and uncontrollable alterations in LULC are the outcome of this pressure [15]. Information on LULC, as well as alternatives for optimal utilization, is crucial for selecting, planning, and implementing land use strategies to meet basic human needs and welfare. Additionally, this data is valuable for monitoring how evolving population requirements are impacting land utilization [5]. The LULC classification of an area is an output of natural and environmental factors and their utilization by human throughout time. Deforestation, biodiversity loss, global warming, and a rise in natural disastrous floods are the results of changes in LULC and human/natural processes.

Vegetation mapping aims to identify, describe, and evaluate the relative abundance of the many forms of vegetation that cover the surface of the earth. With the advancement in remote sensing technology, it is now possible to identify and define different phenomena and objects on the Earth's surface without physically visiting the location [4]. 

The main contribution of this paper is to quantitative mapping and change analysis of the LULC and vegetation cover in the Kasaragod district from 2000 to 2023. Kasaragod has undergone significant LULC transformations driven by socio-economic pressures and developmental activities. These changes contribute to deforestation, biodiversity loss, and escalating Human-Wildlife Conflicts (HWC), highlighting the necessity for spatially informed land-use policies. Remote Sensing (RS) and Geographic Information Systems (GIS) offer robust methodologies for monitoring and analyzing LULC changes. This study fills this gap by employing supervised classification methods and NDVI analysis to quantify LULC changes and assess their environmental impacts. 

Kasaragod, the northernmost district of Kerala, spanning an area of 1992 square kilometers. Kasaragod is geographically framed between North Latitude 12° 02' 25” and 12° 47’ 35”, and East Longitude 74° 25' 54” and 75° 25’ 25”. This area falls within the Survey of India toposheet numbers 48L and 48P. The district is enclosed by the Western Ghats to the east, and the Arabian Sea to the west, the northern boundary is shared with the Canara district of Karnataka, and to the south lies the Kannur District [Fig. 1]. 

Based on geological properties, the district is naturally separated into the eastern highlands, midlands, and coastal lowlands. The coastal region of the district is covered by coconut and arecanut gardens, while the midland has extensive paddy fields and long stretches of arecanut gardens.

Indian Remote Sensing (IRS) images acquired between 2000 and 2023 were utilized for the study. Multispectral satellite imageries of LISS III sensor from the Indian Remote Sensing Satellites IRS-1D, RESOURCESAT-1 (IRS-P6), RESOURCESAT-2 (IRS-R2) for years 2000, 2005, 2013 and 2023 were used. IRS data were obtained from the National Remote Sensing Center (NRSC). The 2000–2023 dataset includes four spectral bands and two scenes (Path-Row: 097-064 and 098-065) covering a 141km wide swath with a spatial resolution of 23m.

The study used supervised classification with maximum likelihood and parallelepiped techniques for each dataset. Satellite images were classified, LULC maps generated, and results compared using a pixel-by-pixel change detection matrix. The methodology involved image enhancement, interpretation, classification, NDVI analysis, and LULC and vegetation change detection [Fig. 2]. ArcMap 10 and ERDAS IMAGINE 9.1 software’s were used. A detailed LULC map was created from high-resolution satellite images [Table. 1]. The satellite images, each with different spectral bands in .tif format were stacked using the Layer Stack function in ERDAS Imagine, followed by image mosaicking. All topographic sheets were then mosaicked and used as a reference map for georeferencing satellite images [Fig. 3]. Geometric distortions were corrected with the Image Geometric Correction tool, using a polynomial geometric model and UTM WGS 84 north zone 43 projection. 

Image enhancement techniques play a critical role in improving the quality of imagery and expanding the scope for accurate image interpretation. In this study, techniques such as brightness/contrast adjustment, edge enhancement, and histogram equalization were employed. The brightness and contrast adjustment functions were applied to enhance the interpretability and perception of the imagery, making the visual data more comprehensible [Fig. 4]. A histogram, which visually represents the brightness levels of an image, provides a deeper understanding of its tonal distribution [Fig. 5]. Through histogram equalization, the intensity values are spread across the full range (0 to 255), where the darkest pixel is assigned a value of 0 (black) and the brightest pixel is assigned 255 [14]. Edge enhancement, another crucial technique used in this study, amplifies intensity differences across features or expands the width of linear features [14]. This adjustment made critical features more distinct and easier to interpret in the imagery [Fig. 6].

Image interpretation is a vital tool for performing various LULC classifications, enabling the identification of features such as vegetation, water bodies, built-up areas, and more. In this study, the Indian Scheme of LULC Classification was applied [13] categorizing the study area into six distinct groups [Table. 2]. 

Image analysis involves identifying groups of pixels with specific spectral properties and determining the features or land cover classes these groups represent [12].  Image classification methods are broadly divided into two categories: supervised classification and unsupervised classification. This study adopted the supervised classification approach, utilizing the parallelepiped algorithm and the Maximum Likelihood Classifier (MLC) to ensure precise categorization of land cover features.

Maximum likelihood rule is widely used parametric method in RS and GIS [6, 10]. It based on the principle of determining the probability that a pixel belongs to a specific class. The basic formula assumes that these probabilities are equal in all classes and that the input bands have normal distributions. The equation for the ML/Bayesian classifier is as follows [8]:

D = ln (ac) − [0.5 ln(|Covc|)] − [0.5 (X − Mc) T (Covc-1) (X−Mc)]         (1)

Where:

D= Weighted distance 

c = A particular class

X = Measurement vector for the candidate pixel

Mc= Mean vector of the sample of class c

ac = Percent probability that any candidate pixel is a member of class c (defaults to 1.0, or is entered from a priori knowledge)

Covc = Covariance matrix of pixels in the sample of class c

|Covc| = Determinant of Covariance of class c (matrix algebra)

Covc-1 = Inverse of Covariance of class c (matrix algebra) 

T = Transposition function (matrix algebra)

The parallelepiped decision rule classifies pixels based on their values compared to upper and lower limits for each band. Each signature (class) has defined limits for all bands. If a pixel's values fall within these limits for all bands, it is assigned to that class [10]. Pixels outside these limits are marked as unclassified. Large rectangles, or parallelepipeds, represent the boundaries of these limits [Fig. 7]. Two image bands, A and B, define the pixel values for training areas based on their maximum and minimum pixel values.

In supervised classification, there are three main steps:

This process ensures accurate grouping of pixels into predefined land-use categories.

In this study, LULC change analysis was identified by employing post-classification change detection method. The Post classification method has been effectively applied by numerous studies in urban areas because of its efficiency in recognizing the location, nature, and rate of change [7]. The following equation was used to calculate the magnitude of change (C) for each class i:

                  (2)

The percentage change (Pi) in each land use class was determined using [18]:

          (3)

Where:

Li = Base image of 2000                                    

Bi = Current image of 2023 

NDVI is a widely used remote sensing method to evaluate the health and productivity of vegetation. It is calculated from the reflectance measurements in the red (RED) and the near-infrared (NIR). NDVI values range from -1 to +1, where higher values (between 0.1 and 1) indicate an abundance of healthy and dense vegetation due to high reflectance in the NIR portion of the electromagnetic spectrum. Bare soil areas correlate with NDVI values closer to zero because of their high reflectivity in both the visible and NIR areas of the electromagnetic spectrum (EMS) [12]. A negative NDVI value indicates non vegetation like water, snow, developments, and barren lands. NDVI method is based on the premise that healthy vegetation has low reflectance in the visible region of the EMS because of chlorophyll and other pigment absorption, but high reflectance in the NIR due to internal reflection by the mesophyll spongy tissue of green leaves [3]. In current study, the NDVI value is calculated according to the following formula.

NDVI=(NIR-RED)/(NIR+RED)      (4)

Where NIR is near infrared reflectance and RED is visible red reflectance. RS satellite imagery utilized band 3 (red) and band 4 (near-infrared).  

Prominent changes in LULC classes investigated in this study for 2000, 2005, 2013 and 2023 based on the supervised classification of satellite images [Fig. 9]. 

The results indicate significant changes in LULC within study area during these periods. In 2000, majority of the study area was covered with plantations and sparse vegetation, representing 60.46% of the total area, followed by mixed crop with settlement with a proportion of 20.43% and barren land at 10.31%. Water body, dense vegetation and built-up land was limited to 2.53%, 3.48%, and 2.79% respectively, of the total area. By 2005, plantation and sparse vegetation covered 56.25% of the area, followed by mixed crop with settlement occupied 23.39%. The result demonstrates the presence of both increased agricultural activities and infrastructural developments. In 2013 plantation and sparse vegetation remained the dominant land cover comprising 51.83% of the total area, showing a decrease compared to the previous years of 2000 and 2005. Mixed crop with settlement expanded significantly, accounting for 26.69% of total land area. The area under water bodies and dense vegetation decreased to 2.23% and 2.63% respectively. The built-up area increased to 7.57% while barren land decreased to 9.05%, indicating that non-vegetated land converted to urban or built-up land. By 2023 plantation and sparse vegetation were still the major land cover, covering an area of 51.17%. Mixed crop with settlement indicates a slight decrease with an area of 24.23% and built-up area significantly increased to 15.55%, the result shows an increasing demand for land from the growing population. The area under water bodies and dense vegetation, covers an area of 1.79% and 2.13% respectively. Barren land covered 5.13%, represents that barren land was increasingly converted to agricultural or built-up land [Fig. 10]. 

The change detection analysis highlights significant transformations in LULC over 23 years. Water bodies declined by 14.65 km² (-41.09%) due to conversion into other land uses and changes in water management practice. Dense vegetation decreased by 26.92 km² (-63.48%) due to deforestation, wildfires, and natural disturbances like cyclones [Table. 3]. Mixed crops with settlements increased by 75.58 km² (15.66%), reflecting agricultural growth and urban sprawl, though their extent declined after 2013. Plantations and sparse vegetation showed a net loss of 185.19 km² (-18.17%) due to land conversion. Built-up areas expanded by 254.26 km² (82.06%), driven by residential, commercial, and industrial development particularly in the western plain, which benefits from better connectivity by major highways, railways, water ways, and infrastructure. This concentration has led to rapid urbanization and industrial activities in the area. Barren land shown the most significant loss, declining by 103.08 km² (-100.83%) as it was converted into settlements and agricultural areas. Between 2000 and 2023, barren land, dense vegetation, water bodies, and plantations saw major reductions, while built-up areas and mixed crops experienced notable gains [Fig. 11].

The vegetation index calculated using NDVI was applied to analyze vegetation changes from 2000 to 2023 [Fig. 12]. Based on NDVI values, images were classified into three classes: Dense vegetation (NDVI > 0.4), sparse vegetation (NDVI ranging from 0.1 to 0.4), and no vegetation (NDVI < 0.1). The area-wide details for each category are presented in [Table. 4]. In 2000, dense vegetation covered 804.32 km² (40.38%), sparse vegetation accounted for 952.26 km² (47.80%), and no vegetation occupied 235.42 km² (11.82%). By 2005, dense vegetation decreased to 757.42 km² (38.02%), while sparse vegetation and no vegetation increased to 957.38 km² (48.06%) and 277.20 km² (13.92%), respectively. In 2013, dense vegetation further decreased to 652.42 km² (32.75%), with sparse vegetation and no vegetation increasing to 984.63 km² (49.43%) and 354.95 km² (17.82%), respectively. By 2023, dense vegetation significantly reduced to 432.41 km² (21.72%), sparse vegetation declined slightly to 904.16 km² (45.38%), and no vegetation expanded to 655.43 km² (32.90%). These changes highlight significant variations in land use and vegetation cover over the past 23 years. The dense vegetation class showed a considerable decline from 804.32 km² in 2000 to 432.41 km² in 2023, reflecting a loss of -371.91 km² (-86.01%) over the 23-year period [Table. 4]. This decline indicates to factors such as deforestation and urbanization, with forest areas being converted into agricultural land, implying encroachment into forest reserves through cultivation. Sparse vegetation showed an initial increase from 952.26 km² in 2000 to 984.63 km² in 2013, followed by a decrease to 904.16 km² in 2023, resulting in a net loss of -48.10 km² (-5.32%). The no vegetation category expanded consistently, increasing by 420.01 km² (64.08%) during the study period [Fig. 13]. This upward trend due to the factors like urbanization, leading to the transformation of vegetated land into urban infrastructure, including roads, buildings, and residential areas. Additionally, deforestation and soil erosion have further contributed to vegetation loss. 

The scarcity of natural prey and food sources often compels wildlife to seek alternative resources, while human-generated food and waste inadvertently attract wild animals, leading to conflict. In regions where human and 

wildlife populations overlap, such interactions frequently result in physical conflicts and negative consequences, including loss of life, agricultural damage, and habitat destruction. Key drivers of human-wildlife conflict (HWC) include agricultural expansion, human settlement, overgrazing, deforestation, illegal grass collection, and improper waste disposal. This study identified five HWC-affected locations in the Kasaragod district: Delampady, Muliyar, Kuttikol, Bedadka, and Karadka, based on recent reports from local newspapers. Four of these locations lie in the midland region, while Delampady is situated in the Western Ghats [Fig. 14]. 

Farmers in these HWC zones have reported frequent wildlife incursions, primarily from elephants and wild boars, causing extensive damage to crops and property. Crops such as arecanut, plantains, and coconuts were particularly affected, prompting some farmers to abandon their cultivation [17]. Over the past two decades, elephants from the Karnataka forests have increasingly crossed into the Kasaragod district. A group of 22 elephants has been responsible for recurrent damage, and previous efforts to redirect these herds have been largely unsuccessful. Consequently, many farmers have shifted to alternative income sources, while others have left their agricultural lands fallow [16]. The findings emphasize the need for further research in the HWC zones of Kasaragod to better understand the severity of the issue and develop effective, sustainable mitigation strategies to address the conflict.

This study utilized RS and GIS techniques to analyze LULC and vegetation changes in Kasaragod district, Kerala, from 2000 to 2023. The findings reveal significant declines in dense vegetation, water bodies, and barren land, coupled with a substantial increase in built-up areas and mixed crops with settlements. Built-up areas expanded by 82.06% over 23 years, reflecting growing population pressure and infrastructural development. NDVI analysis highlighted a drastic reduction in dense vegetation (-86.01%) and a notable increase in non-vegetation areas (64.08%), primarily driven by deforestation, urbanization, and other anthropogenic activities.

The study also identified HWC zones impacted by these LULC changes, emphasizing the need for localized land-use planning to address crop protection and reduce conflict. These results underscore the urgency for sustainable management practices to mitigate urbanization impacts, prevent further deforestation, and conserve agricultural lands and water resources. By providing critical baseline data, this study offers valuable insights for policymakers and planners. Continued monitoring of LULC changes is imperative to support effective policymaking and ensure balanced developmental planning for Kasaragod district, promoting both ecological sustainability and socio-economic growth.

We would like to express our sincere gratitude to the Kerala State Council for Science, Technology, and Environment (KSCSTE) - National Transportation Planning & Research Centre (NATPAC) for providing the technological materials and financial support for this study. We also extend our heartfelt thanks to the National Remote Sensing Center (NRSC) in Hyderabad for their invaluable assistance in acquiring the IRS data.

Funding: The purchase of IRS Data for this study was funded by the Kerala State Council for Science, Technology, and Environment (KSCSTE) - National Transportation Planning and Research Centre.

Competing interest: The author declares no competing interests.