Geospatial Assessment of Spatiotemporal LULC Dynamics in the Mula and Pravara River Basins - Gatha Cognition

3.1 Data acquisition

The methodology involves data acquisition, preprocessing, validation through ground surveys, and generating thematic maps for the present study area analysis (Samarinas et al., 2023). The toposheet data acquired from the Survey of India (SOI) (https://surveyofindia.gov.in) toposheets (1:50,000) (47 E/10, 47 E/11, 47 E/14, 47 E/15, 47 I/1, 47 I/2, 47 I/3, 47 I/4, 47 I/6, 47 I/7, 47 I/8, 47 I/10, 47 I/11, 47 I/12, 47 I/14, 47 I/15, and 47 I/16) were georeferenced via ArcMap 10.8.1. The spatial datasets were projected to the WGS 1984 UTM Zone 43 N coordinate system. Georeferencing accuracy was assessed with a minimum of 8 GCPs per toposheet, with an RMSE of no more than 0.5 pixels. These toposheets were then mosaicked to generate a continuous spatial dataset to extract watershed boundaries (Legese et al., 2026). For the multi-temporal LULC analysis, the necessary satellite images were selected from the USGS Earth Explorer online geospatial portal (https://earthexplorer.usgs.gov). Landsat satellite images from 1994 and 2025 were utilized to assess temporal changes in LULC within the study area.

The images from the Landsat 4-5 Thematic Mapper (TM) satellite were selected for the year 1994, whereas images acquired by the Landsat-9 Operational Land Imager-2 (OLI-2) and Thermal Infrared Sensor-2 (TIRS-2) sensors were used for 2025. The image selected for 1994 dates back to February (2 and 9 February 1994) and had a cloud cover of less than 10% (Murad et al., 2024). For 2025, Landsat-9 OLI-2/TIRS-2 imagery acquired on January 29 and 30, 2025 comprising a total of 11 spectral bands. Out of these, the initial seven bands essential for LULC classification were utilized (Figure 2 The study area was covered by Landsat scenes corresponding to WRS-2 Paths 146–147 and Rows 46–47. The satellite imagery for the month of July exhibits the maximum waterlogged area, while the image data for March reveals the maximum vacant land surface, attributed to the harvesting of all crops during the summer season (Ferrant et al., 2017).

Figure 2. Methodology

Therefore, the periods of March and July are deemed unsuitable for clear-cut delineation of LULC changes (Sohl et al., 2025). Both the TM and MSI sensors find their primary applications in forest management, agricultural monitoring, coastal studies, inland water resource assessment, and the mapping and monitoring of LULC. The pixel-based radiometric measurements are provided in Top of Atmosphere (TOA) reflectance along with the parameters to transform them into radiance (Gupta and Gupta, 2026).

3.2 Data analysis

3.2.1 Image classification

The satellite images were downloaded from the USGS and georeferenced using the SOI toposheets and corresponding ground control points (Manjunatha, 2021). The projection of both images were UTM 43 N with WGS 84 datum. The Earth Resource Data Analysis System (ERDAS 2014) Imagine was utilized for image pre-processing, band combination, and LULC analysis. False Color Composites (FCC) were generated using band combinations of 3-2-1 for the 1994 Landsat 4-5 TM image and 5-4-3 for the 2025 Landsat-9 OLI-2 image. Based on these images, various LULC classes were identified and classified. (USGS, 2020; Singh and Vyas, 2022). Multispectral satellite imagery was effectively utilized to classify LULC. A ‘supervised classification’ method was adopted, based on the spectral reflectance characteristics of various land cover elements within the imagery (Reddy, 2018). To facilitate the classification, “training sites” representing different land cover classes were identified (Maury et al., 2026). Among the available classification algorithms, the Maximum Likelihood Classification (MLC) method was employed to classify the satellite imagery. This method operates on the assumption that training samples for each land cover class follow a normal distribution, classifying pixels into their respective categories based on their spectral characteristics. The classification includes six major classes, i.e., water body, roads and settlement, natural vegetation, barren land, agricultural land, and scrub land (Waza et al., 2026). For each class, more than fifty training sites were used. Accuracy represents how close the results are to the true values and determines the quality of information obtained from remotely sensed data (Wang et al., 2024). The researcher assessed accuracy by visually interpreting images with support from ground truth data collected during fieldwork. The assessment of image classification involved using an error matrix, which compares referenced and classified data (Congalton and Green 2019). It helped determine user accuracy, producer accuracy, overall accuracy, and the Kappa coefficient. Overall accuracy is calculated by dividing the total correctly classified pixels on the diagonal by the total samples (Barman et al., 2026). The Kappa coefficient offers a more comprehensive assessment by considering off-diagonal observations in the error matrix, providing a robust evaluation compared to overall accuracy measures (Tokluoglu et al., 2026).

3.2.2 LULC Change Detection

In this study, temporal changes in LULC were analyzed using the Post-Classification Change Detection method, based on supervised classified satellite imagery from 1994 and 2025. This method enables the determination of how an area belonging to a specific LULC class in 1994 transformed into a different class by 2025. Consequently, it enables the precise identification of changes in the total area of individual LULC classes as well as specific class-to-class conversions and their respective areas.

The analysis utilized the ‘Combine’ tool (located under Spatial Analyst Tools → Local) in ArcMap 10.8.1. By using the ‘Combine’ tool to merge the classified raster images from 1994 and 2025, a unique combination of LULC classes for both time periods was obtained for each pixel. This process made it possible to determine whether a class from 1994 remained unchanged by 2025 or transformed into a different class. For instance, if an area classified as agricultural land in 1994 was converted into a settlement, barren land, or any other category by 2025, that specific conversion was identified, and its area was quantified. Thus, a quantitative analysis was conducted regarding the direction of change, the nature of class conversion, and the associated area for each land use and land cover class during the 1994–2025 period. Several scholars have used the Equation (1) to perform the change detection (Hassen and Assen, 2018; Bagwan and Gavali 2023).

C = [(A_t2 - A_t1) / A_{t1 ×} 100]                           (1)

A_t1= Area of one type of land cover at t¹ time

A_t2= Area of one type of land cover at t² time

C = Rate of change.

3.2.3Transition matrix for LULC conversion monitoring

‘Transition Matrix for Landscape Conversion Monitoring’ is an effective method for the quantitative assessment of land-use changes within a specific geographic area over a defined period, utilizing multi-temporal LULC imagery (Bektas and Goksel 2005; Kalpana, 2018). This method enables the identification of “from-to” transitions, indicating how the LULC class changes from an initial time period to a subsequent period. Based on classified data derived from multi-temporal satellite imagery, the nature, direction, and magnitude of land-use changes can be clearly established.

In the analysis of LULC changes, the transition matrix serves as a systematic table where rows represent LULC classes from the initial stage, and columns represent the corresponding classes from the final stage. Diagonal values within the matrix indicate areas where no change occurred in the respective LULC class during the period, whereas off-diagonal values represent the transition or conversion from one class to another (Suleiman et al., 2017; Bagwan and Gavali 2023). Consequently, it is possible to clearly identify which land-use class transformed into another and the extent of the area involved in such a conversion. Mishra et al., (2020) and Naikoo et al., (2020) have adopted similar methods. Through this process, the area of transition for each LULC class was determined, and a 'LULC Conversion Transition Matrix' was generated. The resulting transition matrix was visualized as a heat map using the Python programming language. Recently, similar studies were also carried out to analyze LULC using the ‘Conversion Transition Matrix’ heat map (Arsiso, 2025; Hussain et al., 2025; de Bodas et al., 2026). This heat map visually illustrates the magnitude and nature of transitions across various LULC classes.

3.3 Land use land cover analysis

3.3.1 Landsat 4-5 (TM) Image (February 1994):

The Landsat 4-5 Thematic Mapper (TM) image from 1994 was utilized to generate the LULC map for the study area (Rigala et al., 2026). Figure 3 illustrates the spatial distribution of LULC classes for 1994. The northern and southern regions are predominantly characterized by barren land, while cropland is primarily concentrated in the eastern and central zones. Scrub land exhibits a scattered distribution (Zhai et al., 2026), while cropland and scrub land predominantly occupy areas adjacent to water bodies. The presence of built-up land is minimal across the study area (Arif and Toersilowati, 2024).

Figure 3. LULC (1994)

The LULC classification statistics, as presented in Table 1, indicate that the total study area PRB encompasses 2,650.7 km². The dominant land cover categories include cropland (36%) and barren land (25%), followed by scrub land (18%), natural vegetation (15.5%), and built-up land (3.6%) (Hussien and Woyessa, 2026). Water bodies constitute only 1.2% of the total area, specifically along the Pravara River (Yamazaki et al., 2015). In the case of the Mula River basin, the study area extends over 2,900.6 km², where scrub land (43%) and barren land (25%) are the prevailing land cover types. The remaining area comprises cropland (15.5%), natural vegetation (12.8%), and built-up land (0.9%). The absence of remarkable urban settlements results in a negligible proportion of built-up areas within both watersheds (Table 1) (Waqas et al., 2026).

3.3.2 Landsat 8-9 (OLI / TIRS) Image (January 2025):

The Landsat 9 Operational Land Imager (OLI-2) and Thermal Infrared Sensor 2 (TIRS-2) imagery was utilized to generate the LULC map for the year 2025. Figure 4 is highlighting considerable land cover transformations over time (Głowienka and Maj, 2026; Zahrani et al., 2026). A noticeable increase in settlements is observed, particularly in the central and eastern regions, which may be attributed to urban expansion and favorable topographic conditions (Kong et al., 2026; Tripta et al., 2026). Cropland, depicted in yellow, remains the dominant land cover, indicating the region's continued reliance on agriculture (Cui et al., 2026). Additionally, water resources are distinctly visible, contributing to agricultural sustainability. Vegetation is predominantly concentrated along water bodies, forming a natural buffer, and is also dispersed across other parts of the study area (Maceseanu et al., 2026; AbdelRahman, 2026; Das, 2026).

Figure 4. LULC (2025)

Fieldwork plays a crucial role in providing ground truth data, enabling a deeper understanding of land use and land cover changes in the study area (Ekka et al., 2026). During this stage, the study area is visited at various times or seasons using Global Positioning System (GPS) surveys, and digital photographs are taken (Fretwell et al., 2017). This helps in understanding the current land use and land cover, as well as the actual changes that have occurred (Tahir et al., 2025) in the study area from 1994 to 2025.

4.1 Field Verification

4.1.1 Built-up Land

During the field survey, built-up areas were identified and validated using fifty GPS coordinates and photographs, based on the presence of houses, roads, and related infrastructure that indicate significant anthropogenic expansion (NRSC, 2006; Utomo et al., 2025).

4.1.2 Cropland

Cropland covers areas dedicated to the cultivation of crops such as grains, vegetables, fruits, and nuts, as well as land used for livestock grazing (Bringezu et al., 2021). It consists of varied agricultural practices, including irrigated and unirrigated agriculture, fallow fields, and plantations, which jointly support farming, manufacturing, and rural living (Sivaraman et al., 2023). Field verification was conducted using GPS coordinates and photographic documentation, supported by observations of crop characteristics, cultivation patterns, and land-management practices (Photo 1) (Anderson, 1976).

Photo 1. Agricultural practices

4.1.3 Scrub Land

Scrub land is characterized by shallow and skeletal soils, rare chemical degradation (Wielgos and Kulik, 2026), steep slopes, severe erosion, and extreme aridity, which favor the expansion of sparse scrub vegetation (NRSC 2006; Zubair et al., 2026). These areas are generally characterized by elevated, relatively level terrain and are distinct from hilly or mountainous terrain (Gujar et al., 2025). Field verification was conducted using GPS coordinates and geotagged photographs to validate these features and support the accuracy assessment of the LULC classification (Photo 2) (Ganga et al., 2026).

Photo 2. Scrub Land Near (Sakur Mandve)

4.1.4 Natural Vegetation

Vegetation was defined as land with >10% tree canopy cover, an area >0.5 ha, and trees ≥5 m in height (NRSC, 2006; Goparaju and Ahmad, 2020). It included evergreen/semi-evergreen, deciduous, plantation, and scrub forests (Jara et al., 2017). Field observations and GPS-referenced photographs were used to verify vegetation type, canopy cover, and spatial characteristics at selected locations, supporting the satellite-based LULC classification. (Photo 3) (Sarun and Nat, 2025).

Photo 3. Natural Vegetaion in Kalsubai Harishchandragad wildlife sanctuary

4.1.5 Water body

Water bodies include natural water features such as rivers and streams, which carry water through defined channels either continuously or seasonally (Singh et al., 2026). They also include ponds, lakes, tanks, and reservoirs (Ninawe et al., 2026). During the field survey, these water features were directly observed and verified, including the Bhojapur Dam and other water bodies, based on their presence, flow conditions, and spatial distribution (Photos 4 and 5) (Manandhar et al., 2024).

Photo 4. Water body (Bhojapur Dam)

Photo 5. Water body (Randha Falls)

4.1.6 Barren Land

Barren land is characterized by little or no vegetation and generally poor soil fertility, limiting its suitability for agricultural activities (Naik et al., 2025). Barren land may include rocky terrain, sand dunes, exposed surfaces, and other sparsely vegetated areas (Singh and Sharma, 2025). Field verification was conducted using GPS-referenced observations and geotagged photographs at selected locations, based on vegetation cover, surface characteristics, and agricultural use (Photo 6).

Photo 6. Barren Land (Near Sakur)

4.2 Accuracy assessment

Accuracy is defined as the degree of proximity between the obtained results and the true values (Zhang et al., 2026). It serves as a measure of the quality of the information extracted from remotely sensed data. Accuracy assessment relied on visual interpretation of images supplemented by ground truth data gathered from fieldwork. The accuracy assessment of the classified images was conducted using the ERDAS Imagine application (Othow et al., 2017). For each of the identified images in 2025, the accuracy evaluation tool of the supervised classifier randomly generated 100 reference points using stratified random sampling. The error matrix and kappa statistics for all classified images were also generated using the ERDAS Imagine software (Foody, 2002). Due to the unavailability and unclear nature of Google Earth images, the accuracy assessment of the images from 1994 was not feasible. The classified image attained an overall classification accuracy of 89% for the year 2025 (Table 3). Subsequent sections present the individual producer’s accuracy and user’s accuracy values for all classes.

4.2.1 User’s and Producer’s Accuracy

Another procedure derived from the error matrix includes the “error of omission” and the “error of commission” (Bhat et al., 2026). Omission error refers to the proportion of reference pixels belonging to a specific class that are incorrectly classified into other classes. Commission error refers to the proportion of pixels classified as a specific class that actually belong to other classes. Omission and commission errors are complementary to Producer’s Accuracy and User’s Accuracy, respectively, and can be calculated as 100 minus the corresponding accuracy value (Rwanga and Ndambuki, 2017).

4.2.2 Kappa statistics

Kappa statistics were produced to assess class accuracy within the broader context of classifier accuracy. According to the criteria outlined by Landis and Koch (1977), agreement is considered poor if Kappa is below 0.4, satisfactory if it falls between 0.4 and 0.7, and excellent if it exceeds 0.75. Table 4 shows the kappa statistics for the 2025 LULC classification.

For the year 2025, the overall Kappa statistics value is 0.8959. Remarkably, the water body category exhibits a kappa value of 1 for 2025, signifying an outstanding agreement. All other classes have kappa values exceeding 0.5, indicating good to excellent agreement.

Monitoring and managing natural resources and urban development rely on change detection, which provides a quantitative analysis of how the population and other environmental activities are distributed spatially (Huang et al., 2008).

4.3 LULC Change Detection and Dynamics

By comparing the classified LULC images from 1994 and 2025, the area of each category was determined in terms of both km² and percentage. Subsequently, to quantify the temporal changes in the area of ​​each LULC category, the percentage increase or decrease for the respective category was calculated (Desalegn et al., 2014; Hailu et al., 2020). For this purpose, the percentage change in land use and land cover was determined using the following Equation (2), based on a comparison of the category-wise areas from 1994 and 2025.

​\(Percentage LULC change= {Area final year- Area initial year \over Area initial year} \times 100\)​   (2)

4.3.1 Pravara Basin

The LULC change analysis of the PRB reveals remarkable land cover transitions over time. Barren land has decreased by 10.7%, indicating conversion to more productive land uses, likely cropland (+4.0%) and scrub land (+6.6%) (Table 5, Figure 5). The increase in cropland suggests agricultural expansion, possibly driven by improved irrigation and farming practices. A minor fall in natural vegetation (-0.9%) implies limited deforestation, while the growth in scrub land may indicate land degradation. Built-up areas (roads and settlements) have increased slightly by 0.4%, reflecting limited urban expansion. Additionally, water bodies have expanded by 0.7%, possibly due to improved watershed management or seasonal water retention. Overall, the observed changes suggest a shift toward agricultural intensification, natural regeneration, and infrastructure expansion, with a potential impact on land sustainability and hydrological balance in the region.

 

Figure 5. Overall LULC change in Pravara basin (1994–2025)

 

Table 5. LULC (Pravara basin)

Class Name	2025		1994		Change (%)
	Area (km2)	Area (%)	Area (km2)	Area (%)
Barren Land	387.0	14.6	670.6	25.3	-10.7
Cropland	1068.2	40.3	962.2	36.3	4.0
Natural Vegetation	387.0	14.6	410.9	15.5	-0.9
Roads and Settlement	103.4	3.9	95.4	3.6	0.4
Scrub Land	654.7	24.7	479.8	18.1	6.6
Water body	50.4	1.9	31.8	1.2	0.7

 

 

A transition matrix was prepared to quantify the changes in LULC classes between 1994 and 2025. The matrix records the area of each LULC class in 1994 that was subsequently classified into the same or a different LULC class in 2025. The rows represent the LULC classes of 1994, while the columns represent the corresponding LULC classes of 2025. The diagonal elements indicate areas that remained under the same LULC class, whereas the off-diagonal elements represent conversions from one LULC class to another. The transition matrix was generated using the classified LULC maps in GIS and was used to quantify the magnitude and direction of landscape transformation during the study period.

An analysis of LULC changes in the PRB between 1994 and 2025 reveals considerable transformations driven by human intervention and environmental factors. The barren land decreased by 41.9%, with its area dropping from 670.6 km² to 387.0 km² (Figure 6). This barren land was primarily converted into cropland (171.4 km²) and scrub land (180.4 km²). Key drivers behind this shift include the expansion of irrigation facilities, water conservation and watershed management measures, and agricultural expansion. Consequently, the cropland area saw a significant increase, rising from 962.2 km² in 1994 to 1,067.1 km² (40.3% increase) by 2025 (Figure 7). This highlights the expansion of agriculture and improved water resource availability within the river basin. Concurrently, changes have occurred in areas covered by natural vegetation and scrub land. Due to increasing urbanization, population growth, and infrastructure development, the area occupied by roads and settlements expanded by approximately 103.6 km² (11%). Specifically, a conversion of cropland and scrub land into built-up areas is evident. However, the water surface area shows relatively minimal change. Overall, it is evident that agricultural expansion, irrigation, water conservation, watershed management, and increasing urbanization are the primary drivers of LULC changes in the PRB.

 

Figure 6. LULC dynamics by confusion matrix in PRB (1994–2025)

 

Figure 7. LULC Transition Matrix (Pravara Basin)

 

4.3.2 Mula Basin

The LULC change analysis of the Mula River Basin (MRB) indicates significant land cover transitions, reflecting shifts in land utilization patterns. Table 6 and Figure 8 show a prominent decline in barren land (-9.4%) and scrub land (-7.5%), suggesting large-scale land conversion, primarily into cropland (+15.5%), indicating intensified agricultural expansion, likely driven by improved irrigation practices and increasing food demand. The increase in built-up areas (roads and settlements) by +3.7% signifies urban growth and infrastructure development, potentially influenced by population expansion and economic activities. A marginal decline in natural vegetation (-1.6%) suggests minimal deforestation, while the slight decrease in water bodies (-0.7%) may be attributed to seasonal hydrological variations. These LULC changes highlight a transition from fallow and degraded lands to more productive agricultural and urban landscapes, which may have implications for land productivity, ecological stability, and watershed hydrology in the MRB.

 

Figure 8. Overall LULC change in the MRB (1994–2025)

 

Table 6. LULC (Mula basin)

Class Name	2025		1994		Change (%)
	Area (Km2)	Area (%)	Area (Km2)	Area (%)
Barren Land	456.0	15.7	728.1	25.1	-9.4
Cropland	899.2	31.0	449.6	15.5	15.5
Natural Vegetation	324.6	11.2	371.3	12.8	-1.6
Roads and Settlement	134.4	4.6	26.1	0.9	3.7
Scrub Land	1033.4	35.6	1250.2	43.1	-7.5
Water body	55.7	1.9	75.4	2.6	-0.7

 

Anthropogenic activities and intensive agricultural practices have led to significant changes in LULC within the MRB between 1994 and 2025. A major shift observed is the expansion of agricultural land, with the area increasing from 449.6 km² to 899.2 km² (Figure 9). 

 

Figure 9. LULC dynamics by confusion matrix in MRB (1994–2025)

 

Against the backdrop of this agricultural expansion, barren land decreased by 37.4%, dropping from 728.1 km² to 456.0 km². Furthermore, rapid infrastructure development and urbanization have resulted in a substantial increase in the area covered by roads and settlements, rising from 26.1 km² to 134.4 km².

The impact of human and economic development is also evident in the natural ecosystem. The area under natural vegetation declined from 371.3 km² to approximately 324 km², while scrub land saw a reduction of 17.5%. Similarly, a decline of approximately 26% was recorded in the area covered by surface water bodies (Figure 10).

 

Figure 10. LULC Transition Matrix (Mula Basin)

 

Overall, the LULC changes in the MRB highlight the mounting pressure that agricultural expansion, urbanization, and infrastructure development exert on the natural ecosystem. This situation highlights the importance of sustainable land and water resource management alongside socio-economic progress.

 

The present study demonstrates substantial and spatially heterogeneous LULC changes in the MPRBs during 1994–2025. In the PRB, cropland increased by 4.0% and scrubland by 6.6%, whereas barren land decreased by 10.7%. In the MRB, cropland showed a more pronounced increase of 15.5%, while barren land and scrubland decreased by 9.4% and 7.5%, respectively. Roads and settlements increased by 3.7% in the MRB, indicating a notable expansion of developed land.

The transition-matrix analysis provided further insight into the direction and magnitude of these changes. Using the ArcMap 10.8.1 Combine tool, the 1994 and 2025 classified LULC maps were integrated on a pixel-by-pixel basis to identify class-to-class transitions. The analysis indicated substantial conversion of barren land toward cropland and scrubland in the PRB, while the MRB showed considerable conversion of barren and scrubland toward cropland and developed land. Thus, the transition analysis complemented the overall area-based comparison by identifying the specific pathways of landscape transformation. The 2025 LULC classification achieved an overall accuracy of 89%, with a Kappa coefficient of 0.8959, indicating good agreement between the classified map and reference information. The integration of multi-temporal Landsat imagery, supervised classification, GIS-based spatial analysis, the Combine tool, transition-matrix analysis, and field observations provided an effective framework for assessing long-term LULC dynamics.

Overall, the findings demonstrate contrasting LULC trajectories between the two basins, characterized mainly by agricultural expansion, reduction of barren and semi-natural surfaces, and increasing developed land, particularly in the MRB. The results provide a quantitative baseline for future land and water-resource planning. Continued monitoring using higher-resolution satellite imagery, field observations, and GIS-based techniques is recommended to identify emerging agricultural, settlement, vegetation, and hydrological change hotspots and to support sustainable watershed management.

The authors declare that they have no conflict of interest.

The authors sincerely acknowledge the support and cooperation extended by Sangamner College, Sangamner, and the Department of Geography and its staff throughout the completion of this research work. The author also gratefully acknowledges Ms. Sonali Parad, Department of Computer Applications, Sangamner College, Sangamner, for her valuable assistance in data analysis. The author is also thankful to the reviewers for their valuable comments, constructive suggestions, and insightful feedback, which helped improve the quality and clarity of the manuscript.

ERDAS 2014: The Earth Resource Data Analysis System 2014; GIS: Geographic Information Systems; GPS: Global Positioning System; LULC: Land Use and Land Cover; MPRBs: Mula and Pravara River Basins; MRB: Mula River Basin; OLI-2: Operational Land Imager-2; PRB: Pravara River Basin; RS: Remote Sensing; SOI: Survey of India; TIRS-2: Thermal Infrared Sensor-2; TM: Thematic Mapper.