Abstract
Land use and land cover (LULC) changes are important for gaining a perspective on environmental dynamics and the impact on climate, urbanization, and resources. To ensure that it is safe to monitor the changes over time and to adopt the right forceful changes in our area, remote sensing is one of the ways to monitor the local and regional level land use, land cover patterns, and landscape changes. This study investigates the temporal LULC changes in the Nowshera region of Pakistan for the years 2016–2023 using pixel and region-oriented classification methods. As a first step, freely available high-resolution multispectral data of Sentinel-2 satellite are acquired, which serves as input dataset for both pixel and region-oriented classifiers. The accuracy assessment scores confirm that for the classified data of the year 2016, the region-oriented technique demonstrated higher overall classification accuracy (89.6%) over pixel-based classification (80.77%). Moreover, for the dataset of the year, the region-oriented method achieved a higher overall Kappa hat score (0.88) as compared to the pixel-based method (0.71). Similarly, for the classified data of the year 2023, the region-oriented method achieved higher scores for both the overall accuracy and Kappa hat (93.6 and 0.92%) over the pixel-based method (77.18 and 0.66%). The study states that for the assessment of LULC changes in Nowshera, the region-oriented image analysis provides a higher level of classification accuracy than the pixel-based approach. These results illustrate that this tool is particularly effective in monitoring detailed land cover transformations, thereby enhancing the quality of environmental management. Furthermore, the regression analysis reveals a substantial correlation between LULC changes and alterations in temperature and precipitation, and this result suggests the necessity of the development of specific climate adaptation programs.
1 Introduction
Assessment and monitoring of land use and land cover (LULC) are crucial for evaluating and projecting the trends of LULC changes and their implications for the environment, ecological balance, resource management, and urban planning [1]. Human activities have played a pivotal role in changing landscapes and causing substantial and long-lasting environmental impacts [2]. The transformation of the earth’s environment has escalated significantly with the advent of agriculture and the growth of urbanized societies [3]. The analysis of LULC dynamics provides an understanding of the historical evolution of landscapes and the projection of the future land cover and its associated impacts on critical natural resources [4].
Climate change has a significant influence on the water, agriculture, ecosystems, infrastructure, and frequency of natural hazards exacerbated by evolving land use and cover. Over 35% of climate change impacts are attributable to land use activities based on the meta-analysis [5]. Tracking the LULC changes using geospatial techniques is vital for modeling climate sensitivities, projecting vulnerabilities, and enabling adaptation planning [6]. Without rigorous LULC monitoring through a climate risk lens, understanding and preparedness against the increasing climate threats is underestimated.
Assessment and monitoring of the LULC utilize both traditional and remote sensing methods to understand, characterize, and project the spatial and temporal distribution of LULC classes. Traditional methods consist of on-site field surveys, aerial photography, land use maps compiled from existing records, and surveys/with local communities [7]. These methods offer a detailed, realistic understanding of the landscape and its dynamics, often providing valuable socio-economic context; however, for a limited area, such methods require extensive cost and time.
On the contrary, space-borne remote sensing with the given long-temporal archive, global coverage, and often free accessibility of satellite imageries are frequently, efficiently, and effectively applied for assessing and monitoring the LULC at local, regional, and global scales.
A range of data and techniques, including manual, supervised, unsupervised, object-based, and machine-learning methods, are developed for assessing the temporal patterns of LULC in time and space domains [8,9,10,11].
Unsupervised classification algorithms include K-means cluster and hierarchical clustering, classifiers that identify and explain intrinsic structures or patterns in unlabeled datasets. These approaches help in deriving useful information from data with no previous labels; however, this is not the case wherever they occupy spectrally fewer elite regions [12].
Supervised classification relies on labeled data for training, and some of the most widely utilized methods include minimum distance, parallelepiped, maximum likelihood, support vector machine (SVM), random forest algorithm, and deep learning techniques such as convolutional neural networks (CNNs).
The advancement of RS classification, especially with deep learning methods such as CNNs and data cubes, provides increased power to extract complex patterns for LULC change detection [13]. However, CNNs are limited because their very nature is a black box that requires large amounts of data to train and compute intensively. In contrast, algorithms such as SVM and Random forest [14], despite their lower complexity, higher interpretability level, and generalization ability in small-data set classification limits, have proved competitive alternatives for learning accurate LULC classes [15].
SVM is a powerful and versatile machine-learning algorithm used for both classifications [16]. It contains finding hyperplanes that best separate classes [17], and it is very effective for high-dimensional data also [18]. It has great significance in classification to be able to learn nonlinear relationships by using kernel functions, which makes it one of the classical algorithms having vast applications on complex datasets [19]. The supervised classification methods give the best results for LULC classification, particularly for complex and higher LULC classes. Alshari and Gawali [20] also underscore the effectiveness of supervised classification methods in handling LULC classification tasks.
Pixel-based classification methods encounter several challenges; however, combining pixel-based methods with ancillary data such as topography or vegetation indices can enhance the accuracy of this type of classification method. Nevertheless, region-oriented image classification overcomes these challenges by grouping pixels into meaningful objects, considering spatial relationships and contextual information. Region orientation considers clusters of pixels and relies on the scale, shape, color, and contextual information of the objects and discretizes the image in multiple segments, which are later systematically and gradually grouped into LULC classes using ruler-based classification [14,21]. Region-oriented classification techniques utilize the appropriate satellite data considering their spatial and spectral resolution for monitoring urbanization patterns [22], forest monitoring, management, and modeling [23], crop monitoring, and agriculture patterns [24]. Space borne-sensors such as the Landsat, Sentinel-2, ASTER, and MODIS provide free, temporal, and global coverage; however, high-resolution commercial sensors including the IKONOS, World View, and Quick Bird [25] are effective for local-scale monitoring.
Nevertheless, the high cost of these mentioned commercial satellites poses challenges for conducting regional-scale monitoring [26]. Free accessibility, global coverage, and long temporal archive of the Landsat data series, with its increasing spatial and spectral capabilities, make it one of the most frequently and effectively utilized satellite data for monitoring the LULC changes from local and regional to global scales [27,28,29,30,31]. The onset of the Sentinel-2 multispectral data in 2015 provided emerging opportunities for monitoring the LULC dynamics and its impact on the environment, urbanization, food security, and climate change [32]. Equipped with a multi-spectral sensor capturing data across visible, near-infrared (NIR), and shortwave infrared bands, Sentinel-2 enhances discrimination among diverse land cover types based on their distinct spectral signatures.
Satellite technologies have improved our ability to monitor LULC, such as the case of the Pleiades Neo constellation. Detailed analysis is thus now possible over a larger swath of time using high-resolution imagery and six spectral bands from the new system. In addition, the system allows for less-than-1-day revisit, which is important to track fast LULC changes and, in turn, environmental impacts [33]. However, problems with data integration and processing may still exist, requiring research to alleviate land management.
There exists a vast body of literature assimilating and examining the influence of land cover change on climatic parameters, particularly temperature, precipitation, humidity, and radiation at local, regional, and global levels [34,35,36]. Urbanization at the expense of displacing agricultural and forest lands in view of increasing population is responsible for the rise in localized temperatures both spatially and temporally, conceptualized urban heat islands [37,38].
Pakistan is among the most prone countries to climate change with increasing temperature, changing patterns of precipitations, accelerated melting of glaciers, and increased occurrences of extreme weather events and disasters, posing a serious challenge to the sustainable socio-economic development of the country. The temporal changes in the intensity and distribution of climatic parameters are also influenced by the changing LULC patterns induced by anthropogenic activities, including deforestation, urbanization, and infrastructure development. Monitoring the LULC trends and evaluating its impacts on the climate indicators are critical to projecting the future climate and accordingly developing strategies to mitigate and adapt to minimize the negative impacts on society and the economy. However, such ideas are rarely exploited in developing countries; therefore, this study aims to evaluate the spatiotemporal trends of LULC in the climate change-prone district of Nowshera in Pakistan. The district Nowshera was selected for the study due to its diverse climate, land use practices, and notable land cover changes in recent decades that showcase broader regional trends. In this research work, Sentinel 2 satellite data have been utilized for pixel and region-oriented image classification to evaluate the impact of LULC on the local climate of the study site. The main objectives of this research work are highlighted as follows.
Objectives:
Quantification of LULC changes using object- and pixel-based supervised classification methods.
Regression analysis to estimate the Pearson coefficient matrix that shall quantify the temperature and precipitation data in our study area for the years 2016–2023
Explore relationships between observed LULC and climate variations.
Implement change detection between the multi-date LULC maps.
The rest of the article is organized as follows. The next section discusses the study area. Material and methods are discussed in Section 3. Section 4 discusses the results, and finally the discussions and conclusions are drawn based on the presented results.
2 Study area
District Nowshera is in Khyber Pakhtunkhwa province of Pakistan, with a total area of 1,748 km2. The district is drained by the Kabul River (Figure 1). The area features a sub-tropical climate [39] with hot summers from June to August (average maximum temperature 35–40°C) and cold winters from December to February (average maximum temperature 15–20°C). Precipitation in the area varies seasonally, with monsoon rains in the summer season [40]. This climatic variability allows for a range of land use and agricultural practices in the study region [41]. Nowshera represents broader agricultural areas facing environmental change pressures amidst economic development and population growth [42]. As an important food-producing district, land use changes in Nowshera can impact regional and national food security and rural livelihoods. The area is covered by agricultural croplands, forests, grasslands/shrublands, wetlands, urban built-up areas, and barren lands [43]. Rural communities are relying extensively on groundwater for domestic water needs despite concerns of excessive salinity, hardness, or contamination. However, increasing population [39,44] and urbanization [45] have driven the expansion of settlements, roads, and commercial infrastructure at the cost of losing the forest and agricultural land [46]. From 1951 to 2017, the population of the Nowshera district grew from 0.2 million to over 1.5 million at a 2.96% annual growth rate [47]. Spatial analysis indicates only 21% of Nowshera land area overlies high-quality groundwater zones, suggesting land use changes may further degrade scarce water resources for residents and crops [48]. Such land changes also relate to climate risks from extreme heat, floods, and droughts [41,49,50]. Moreover, the recent floods in the district have contributed to the land cover changes along the river’s areas.
Geographical map of Nowshera district and its location in Pakistan, with (a) Pakistan, (b) Khyber Pakhtunkhwa, and (c) Nowshera.
3 Material and methods
This section presents the material and methods utilized in this research work. For more details, refer to the subsections discussed below.
3.1 Data acquisition
3.1.1 Sentinal-2 data
Satellite images from space and geographic information system datasets are useful for spatial event analysis and temporal variation estimation in various areas globally [51]. The 10 m resolution Sentinel-2 images in the study area are downloaded for the years 2016, 2019, 2021, and 2023, respectively, for use in these studies.
Sentinel-2 has characteristics that render the high-resolution imagery selected in 2016, 2019, 2021, and 2023 ideal for detecting changes in LULCs of the Nowshera district, Pakistan. Specific dates were carefully chosen based on considerations including timing, image quality, geographical focus, consistency with meteorological records, and new satellite observations. By comparing seasonal and annual variation over short- and long-term scales to local climate records from 2016 to 2021, greater insight could be gained regarding the interplay of climate drivers with landscape changes across space and time. In fact, continuous improvement in observational capabilities meant that the latest images from 2023 provided sharper details. The satellite images selected based on these criteria provide an adequate representation of the environmental changes and their effect on climate, enabling a comprehensive understanding of the dynamic environment in this region.
Table 1 shows the details of the acquired dataset. The dataset was used as input to the classifiers for the LULC classification. The Semi-Automatic Classification Plugin (SCP) (https://plugins.qgis.org/plugins/SemiAutomaticClassificationPlugin/) in QGIS 3.28 enabled access to atmospherically corrected Sentinel-2 images for the selected years (https://scihub.copernicus.eu) [52].
Satellite data specification
| S. no. | Acq. date | Zone/path | Sensor | Spectral bands | Spatial res. | Source |
|---|---|---|---|---|---|---|
| 1 | 21/10/2016 | 42SYC | Sentinel-2 | B1, B2, B3, B4, B5, B6, B7, B8, B8A, B9, B10, B11, B12 | 10 m | SNAP/scihub |
| 2 | 25/12/2019 | |||||
| 3 | 7/2/2021 | |||||
| 4 | 1/8/2023 |
3.1.2 Climate data
For the regression analysis, climate parameter data were sourced from the Pakistan Meteorological Department [53] (https://www.pmd.gov.pk/en/), encompassing temperature and precipitation metrics across the designated years. Population statistics were acquired from the Pakistan Bureau of Statistics [54] (https://www.pbs.gov.pk/census-2017-district-wise/results/017), providing demographic insights essential for evaluating the relationship between land use changes and population dynamics with the growth rate. The growth rate of the population in this census was used for extrapolation for population estimation in the study area for other selected years. This technical data acquisition process underpins the study’s analytical framework, ensuring a comprehensive understanding of the interplay between climate variations, population growth, land use, and land cover changes within the study area. WorldClim (https://www.worldclim.org/data/worldclim21.html) dataset was downloaded to obtain temperature and precipitation data for the Nowshera district for the years 2016, 2019, and 2021. Mean monthly temperature and precipitation were calculated using the zonal statistics plugin and then aggregated for annual averages. Time series analysis was used to evaluate temperature extremes and seasonal precipitation variability.
3.2 Image pre-processing and classification (pixel-based and region-oriented based)
For both pixel-based and region-oriented image analysis methodologies, Sentinel-2 multispectral bands underwent atmospheric correction. In the pre-processing phase within QGIS, we ensured accurate alignment of the Sentinel-2 Level-1C images with the WGS84 UTM Zone 42N coordinate system through several steps. First, we imported the images and visually inspected them against known geographic features to identify any misalignments. We then utilized the SCP for geometric verification, comparing the imagery with reference layers such as OpenStreetMap to confirm accurate spatial correspondence. Since the selected images all show cloud cover of about 0.0004–0.01 and no coverage of the site is compromised, we also checked for any residual cloud effects to maintain consistency across the dataset. The radiometric corrections include geometric corrections and cloud removal to ensure all images are well-matched in the same coordinate systems and with no pixels containing clouds. Also, special band combinations (RGB = 4-3-2, RGB = 3-2-1, and RGB = 7-3-2) were used for best land use pattern discrimination. Such a foundational step was of prime importance in preparing the data for later analysis and classification. The data spectral reflectance is a key principal parameter exploited in separating objects by their characteristic spectral reflectance curves [55,56].
For this analysis, atmospheric correction of the Sentinel-2 imagery was applied using SCP in QGIS. This plugin applies to the Sen2Cor algorithm internally in order to convert top-of-atmosphere (Level-1C) reflectance values into bottom-of-atmosphere (Level-2A) reflectance values.
3.3 Supervised image classification (pixel-based classification)
With the help of labeled data called regions of interest (ROIs), statistical models can be trained to classify pixels in the remote-sensing images into different categories via supervised classification [57,58]. ROIs are homogeneous areas delineated from the input dataset to sample and provide representative spectral signatures for the desired land cover classes [57]. We included additional sources (e.g., Google Earth) for cross-verification of the ROIs to further improve its accuracy, which is done by visual interpretation of the images.
ROIs were defined for five land cover classes (Table 2). Pixel-based classification of the pre-processed dataset was done using the SVM.
Description of land use land cover types
| Type | Description |
|---|---|
| Water bodies | Rivers, lakes, reservoirs, water in mine pits, ponds, wetlands |
| Urban land | Constructed areas including residential, commercial, industrial roads, etc. |
| Barren land | Land characterized as non-built-up land with no cover |
| Forest | Natural and manmade forests and grasslands vegetation of different types |
| Agriculture | Land covered by farmlands, croplands, and range lands |
Table 2 summarizes the defined land cover classes based on distinct characteristics observed within the region studied. The land cover classes were defined based on spectral characteristics from Sentinel-2 imagery and visualized using specific colors in QGIS with the SCP. The classification process grouped pixels with similar spectral reflectance, assigning colors for clarity. The criteria for each land cover class are as follows:
Water (dark blue): Identified by low reflectance in the NIR band and high absorption in the visible spectrum, water bodies like lakes and rivers were represented by a dark blue color, showing smooth, uniform patterns.
Urban/built-up land (light blue or cyan): Characterized by high reflectance in the visible and NIR bands, indicating built structures, these areas displayed block-like patterns and dense infrastructure, visualized in light blue or cyan.
Barren land (white, light blue, and light cream): Identified by high reflectance across the visible spectrum and lower NIR values, barren areas such as bare soil, sand, and rocky terrain were visualized using white, light blue, and light cream tones to indicate minimal vegetation.
Forest (maroon and blackish red): Vegetated areas with high NIR reflectance, indicative of healthy vegetation, were represented by maroon and blackish-red colors, denoting dense forests or natural vegetation cover.
Agriculture (red): Agricultural lands exhibited seasonal variability and moderate reflectance in the NIR and red bands, classified in red to distinguish cultivated fields from natural vegetation [59].
Although this classification protocol works well at the regional scale, subtle differences among land cover types are not distinguishable, and misclassification may result in more complex landscapes [60].
3.4 Region-oriented image analysis
Region-oriented involves segmenting images into objects, extracting features (spectral, shape, and texture), and classifying objects using algorithms like SVM. This approach was applied to Sentinel-2 data for LULC classification in 2016, 2019, 2021, and 2023. SVM’s capacity to effectively separate classes in high-dimensional feature spaces and incorporate spatial, textural, and contextual information made it appropriate for region-oriented image analysis [61].
3.4.1 Preprocessed image for segmentation
In region orientation, segmentation is a critical step where the image is partitioned into meaningful objects based on attributes like color, texture, and spatial proximity, facilitating a detailed analysis from a pixel-based to an object-oriented approach [62,63,64]. The Mean Shift algorithm was used for segmentation. The important parameters for this algorithm were a spatial radius of 5 to obtain the regions and then a range radius of 15.00. Further, the mode convergence was defined as 0.1 and maximum iterations at 100 with minimum region size being fixed to be 100. We used the processing mode vector and had a minimum object size of 1. In this study, we set these parameters mainly to make the objects generated by image segmentation more accurately correspond with land cover elements in our work region and will optimize segmentation results.
3.4.2 Feature extraction and classification
After segmentation, features of interest are extracted from each object, including spectral information, shape attributes, and texture properties. The integrity of this information is very important for the classification of objects correctly into a thematic category that improves semantic analysis and enhances classification accuracy [65]. The SVM is used to classify pixels in the classification stage by assigning the pixels of each object to a predefined class according to its features.
3.4.3 Post-processing and validation
The methodology concludes with post-processing and validation for both region-oriented and pixel-based classification, ensuring the accuracy of our LULC classifications. Region-oriented and Pixel-based classification utilized SVM, with the method showing enhanced segmentation and accuracy, proving essential for environmental and urban analysis.
The classified images were post-processed in QGIS using the SCP. For the reference raster, a satellite image from Google Earth was downloaded and georeferenced to match the spatial size of the study area. Validation samples across representative areas of each class of entered ground-truth data were developed in the Training Input Tool. The selections in the areas that were improperly classified, as determined by visual inspection, were further refined using multiple ROI tools. The classified raster and the reference raster produced were compared by means of the same plugin, and then an accuracy assessment was conducted by computing the confusion matrix to determine producer and user accuracy (UA) and the Kappa coefficient.
The metrics connected with the validation included producer accuracy (PA), UA, and Kappa coefficient. From a confusion matrix, they were calculated to explore the level of the classification’s precision. In this way, it was possible to ensure that the proposed results were as close to the real land cover in the study and that all the classifications were valid.
Figure 2 outlines our dual approach’s key steps, effectively capturing and analyzing Nowshera’s land cover changes.
Schematic representation of LULC.
3.5 Accuracy assessment using random points
Accuracy assessment is a postclassification step utilized for estimation of the performance indicators such as estimated areas proportion for each class, PA, UA, and Kappa hat [66]. The computation of PA and UA involves the utilization of a confusion matrix derived from the accuracy assessment process. In this study, an accuracy assessment was performed using the SCP plugin of the QGIS software. The accuracy assessment process involves the design of totally stratified samples and individual samples for each classified class.
The total sample count is determined by the following specified formula [66,67]:
In equation (1), A i shows the area proportion of each classified class, δ i represents the standard deviation of each class, while δ 0 shows the target standard deviation. In the second step, the samples for each classified class are derived using the following expression:
In equation (2), S i represents the sample for each classified class, i shows the class index, and n represents the total number of classes. Based on the above concepts, the designed samples are tabulated in Table 3.
Land cover classification statistics for different years
| Class | Sδ i | S/n | Total | Sδ i | S/n | Total | Sδ i | S/n | Total | Sδ i | S/n | Total |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2016 | 2016 | 2016 | 2019 | 2019 | 2019 | 2021 | 2021 | 2021 | 2023 | 2023 | 2023 | |
| Water | 29.13 | 53 | 41.065 | 37 | 69 | 53 | 40.76 | 73.5 | 57.13 | 87.19 | 143 | 115 |
| Urban land | 43.43 | 53 | 48.21 | 55 | 69 | 62 | 68.48 | 73.5 | 70.99 | 148.08 | 143 | 145 |
| Barren land | 99.82 | 53 | 76.41 | 127 | 69 | 98 | 120.79 | 73.5 | 97.145 | 300 | 143 | 221 |
| Forest | 31.45 | 53 | 42.22 | 54 | 69 | 61.5 | 49.29 | 73.5 | 61.4 | 106 | 143 | 124 |
| Agriculture | 57.43 | 53 | 55.21 | 72 | 69 | 70.5 | 88.22 | 73.5 | 80.86 | 215 | 143 | 179 |
| Total | 266.3 | 266.3 | 266.3 | 345 | 345 | 345 | 367.54 | 367.54 | 367.54 | 717 | 717 | 717 |
Table 4 specifies the percentage adjustment and validation samples that are used to conduct a full validation of the classification results. For validation, the percentage of samples is determined by the following formula (3):
Percentage of samples used for validation by class and year
| Class | 2016 (%) Pixel-based | 2016 (%) Region-oriented | 2019 (%) Pixel-based | 2019 (%) Region-oriented | 2021 (%) Pixel-based | 2021 (%) Region-oriented | 2023 (%) Pixel-based | 2023 (%) Region-oriented |
|---|---|---|---|---|---|---|---|---|
| Water | 4.81 | 18.31 | 4.85 | 19.05 | 4.85 | 18.46 | 4.85 | 20.80 |
| Urban land | 12.5 | 16.20 | 16.5 | 15.87 | 16.5 | 23.08 | 12.62 | 19.20 |
| Barren land | 54.81 | 26.41 | 48.54 | 21.83 | 44.17 | 15.77 | 54.37 | 18.40 |
| Forest | 4.81 | 18.31 | 5.34 | 21.83 | 6.8 | 16.15 | 4.85 | 23.60 |
| Agriculture | 23.08 | 20.75 | 24.76 | 21.43 | 27.67 | 26.54 | 23.3 | 18.00 |
Thus, the percentages provided reflect the distribution of samples for validation and adjustment. The results are more reliable and credible ways of the classification accuracy assessment.
3.6 Change detection
The postclassification comparison technique was used for the change analysis. This was adopted for this study due to its unique advantages of extracting quantitatively the conversions between the various LULC classes and deducing the LULC change rates [68]. The land use conversion graphs for the four periods between 2016 and 2023 were generated and pivoted in graphs Figure 5 (left and right) with area in square kilometers. The annual rate of change of the various classes was also calculated using equation (4) [69]:
where r is the rate of land cover changes, and A 1 and A 2 are the areas of LULC at time t 1 and t 2, respectively.
4 Results
4.1 LULC classification
The classified maps derived from the region-oriented and pixel-based workflows across the four periods exhibit visually distinct patterns. The qualitative assessment shows salt-and-pepper effects (Figure 3). Table 5 summarizes the area distribution of land cover classes in our study area for the years 2016, 2019, 2021, and 2023 based on classified data covering a total area of 1852.49 km². A total of five LULC classes, namely water bodies, barren land, urban land, agriculture, and forest, were identified using supervised classification. The LULC classification results revealed notable changes across the study area from 2016 to 2023. Water bodies showed a notable increase, expanding from 26.41 km² (region-oriented) and 36.27 km² (pixel-based) in 2016 to 38.88 km² and 49.00 km² in 2023, respectively. Urban land also exhibited substantial growth, increasing from 190.41 km² (region-oriented) and 242.49 km² (pixel-based) in 2016 to 244.42 km² and 283.03 km² in 2023. Similarly, forest cover experienced an increase, with region-oriented classification rising from 61.46 km² in 2016 to 94.96 km² in 2023, while pixel-based classification increased from 69.50 to 90.00 km² over the same period.
Qualitative assessment of classification outputs.
Area statistics of the land use/cover units in 2016, 2019, 2021, and 2023 (in km²)
| Classes | 2016 Region-oriented | 2016 Pixel-based | 2019 Region-oriented | 2019 Pixel-based | 2021 Region-oriented | 2021 Pixel-based | 2023 Region-oriented | 2023 Pixel-based |
|---|---|---|---|---|---|---|---|---|
| Water | 26.41 | 36.268 | 18.59 | 29.356 | 21.517 | 40.36 | 38.88 | 49.00 |
| Urban land | 190.41 | 242.489 | 258.34 | 285.00 | 283.57 | 318.76 | 244.42 | 283.03 |
| Barren land | 1227.80 | 1,055.997 | 1037.04 | 971.221 | 953.23 | 844.54 | 1123.87 | 980.65 |
| Forest | 61.46 | 69.496 | 132.93 | 90.00 | 160.61 | 126.06 | 94.96 | 90.00 |
| Agriculture | 346.41 | 448.039 | 405.59 | 476.712 | 433.58 | 523.14 | 350.36 | 450.00 |
| Total area (km²) | 1852.49 | — | — |
Bold value represents the total area.
Agricultural land expanded from 346.41 km² (region-oriented) and 448.04 km² (pixel-based) in 2016 to 350.36 and 450.00 km² in 2023. Conversely, barren land initially decreased from 1227.8 km² (region-oriented) and 1056.00 km² (pixel-based) in 2016 to 953.23 and 844.54 km² in 2021 but then increased again to 1123.87 km² (region-oriented) and 980.65 km² (pixel-based) in 2023, possibly due to the impact of the devastating floods of 2022.
Table 6 presents the annual changes in land cover areas from 2016 to 2023, highlighting the variations and transitions observed over this period. Each value in the table represents the difference in area for a specific land cover category between 2016 and 2023, providing insights into the dynamics and trends in land cover changes over the 7-year time limit.
Annual changes (2023–2016 change area)
| Classes | 2023–2016 Change area km² region-oriented | 2023–2016 Change area km² pixel-based | Annual change (%) region-oriented | Annual change (%) pixel-based |
|---|---|---|---|---|
| Water | 12.47 | 12.732 | 1.78 | 4.97 |
| Urban land | 54.01 | 40.541 | 7.72 | 2.37 |
| Barren land | −103.93 | −75.347 | −14.85 | −1.02 |
| Forest | 33.5 | 20.504 | 4.79 | 4.22 |
| Agriculture | 3.95 | 1.961 | 0.56 | 0.06 |
Note: The negative values (−) in the “Change Area” for Barren land represent the area declined from 2016 to 2023, demonstrating the conversion of barren land to other land use.
4.2 Accuracy assessment of the classification results
Accuracy assessment in remote sensing evaluates the classification’s precision against reference data. Pixel-based classification matches pixel values to image classes, while region-oriented analysis groups pixels into meaningful segments. Derived confusion matrices for both methods (Tables 7 and 8) show pixel-based accuracy in barren land and agriculture and region-oriented in urban and barren land detection. Kappa statistics validate both methods’ consistency. Accuracy results, essential for method evaluation, are detailed in Tables 9 and 10 for Sentinel-2 classifications.
Consolidated pixel-based land classification confusion matrices (2016–2023)
| >Confusion matrix (pixel count) pixel-based | |||||||
|---|---|---|---|---|---|---|---|
| Year | Classification | Water | Urban land | Barren land | Forest | Agriculture | Total |
| 2016 | Water | 8 | 1 | 1 | 0 | 0 | 10 |
| Urban land | 0 | 26 | 0 | 0 | 0 | 26 | |
| Barren land | 2 | 10 | 93 | 1 | 8 | 114 | |
| Forest | 0 | 2 | 0 | 8 | 0 | 10 | |
| Agriculture | 0 | 4 | 9 | 2 | 33 | 48 | |
| 2019 | Water | 9 | 1 | 0 | 0 | 0 | 10 |
| Urban land | 0 | 26 | 7 | 0 | 1 | 34 | |
| Barren land | 3 | 6 | 81 | 0 | 10 | 100 | |
| Forest | 0 | 0 | 0 | 11 | 0 | 11 | |
| Agriculture | 0 | 5 | 7 | 3 | 36 | 51 | |
| 2021 | Water | 8 | 0 | 1 | 1 | 0 | 10 |
| Urban land | 0 | 29 | 5 | 0 | 0 | 34 | |
| Barren land | 1 | 4 | 81 | 1 | 4 | 91 | |
| Forest | 0 | 0 | 3 | 7 | 4 | 14 | |
| Agriculture | 2 | 4 | 6 | 7 | 38 | 57 | |
| 2023 | Water | 7 | 3 | 0 | 0 | 0 | 10 |
| Urban land | 0 | 23 | 1 | 1 | 1 | 26 | |
| Barren land | 2 | 3 | 82 | 5 | 20 | 112 | |
| Forest | 0 | 0 | 0 | 9 | 1 | 10 | |
| Agriculture | 0 | 3 | 5 | 2 | 38 | 48 | |
Consolidated region-oriented land classification confusion matrix (2016–2023)
| >Confusion matrix (pixel count) region-oriented | |||||||
|---|---|---|---|---|---|---|---|
| Year | Classification | Water | Urban land | Barren land | Forest | Agriculture | Total |
| 2016 | Water bodies | 45 | 2 | 5 | 0 | 0 | 52 |
| Urban land | 1 | 40 | 1 | 4 | 0 | 46 | |
| Barren land | 16 | 14 | 45 | 0 | 0 | 75 | |
| Forest | 0 | 6 | 0 | 46 | 0 | 52 | |
| Agriculture land | 10 | 0 | 0 | 0 | 49 | 59 | |
| 2019 | Water bodies | 47 | 0 | 1 | 0 | 0 | 48 |
| Urban land | 0 | 37 | 3 | 0 | 0 | 40 | |
| Barren land | 0 | 6 | 49 | 0 | 0 | 55 | |
| Forest | 4 | 0 | 0 | 51 | 0 | 55 | |
| Agriculture land | 0 | 7 | 47 | 0 | 0 | 54 | |
| 2021 | Water bodies | 0 | 0 | 0 | 48 | 0 | 48 |
| Urban land | 3 | 46 | 1 | 2 | 8 | 60 | |
| Barren land | 0 | 0 | 41 | 0 | 0 | 41 | |
| Forest | 0 | 0 | 0 | 42 | 0 | 42 | |
| Agriculture land | 47 | 4 | 18 | 0 | 0 | 69 | |
| 2023 | Water bodies | 0 | 0 | 51 | 1 | 0 | 52 |
| Urban land | 0 | 48 | 0 | 0 | 0 | 48 | |
| Barren land | 1 | 4 | 0 | 0 | 41 | 46 | |
| Forest | 50 | 2 | 0 | 0 | 7 | 59 | |
| Agriculture land | 0 | 44 | 1 | 0 | 0 | 45 | |
Accuracy assessment for land cover classification over different years
| Classes | Method | PA (%) 2016 | UA (%) 2016 | PA (%) 2019 | UA (%) 2019 | PA (%) 2021 | UA (%) 2021 | PA (%) 2023 | UA (%) 2023 |
|---|---|---|---|---|---|---|---|---|---|
| Water bodies | Region-oriented | 90 | 100 | 92.15 | 97.91 | 96 | 99 | 98.07 | 100 |
| Pixel | 80 | 80 | 75 | 90 | 72.73 | 80 | 77.78 | 70 | |
| Urban land | Region-oriented | 78.43 | 99 | 74 | 95 | 92 | 99 | 96 | 100 |
| Pixel | 60.47 | 78 | 68.42 | 76.47 | 78.38 | 85 | 71.88 | 88.46 | |
| Barren land | Region-oriented | 88.23 | 60 | 100 | 89.09 | 82 | 100 | 80.39 | 89.13 |
| Pixel | 80.21 | 70.5 | 85.26 | 81 | 81 | 89 | 93.18 | 73.21 | |
| Forest | Region-oriented | 92 | 88.46 | 100 | 92.72 | 84 | 98 | 100 | 84.74 |
| Pixel | 72.73 | 80 | 78.57 | 90 | 56.30 | 57 | 52.94 | 90 | |
| Agriculture land | Region-oriented | 100 | 83.05 | 92.15 | 87.03 | 94 | 70.14 | 86.27 | 97.77 |
| Pixel | 80.49 | 68.75 | 76.60 | 75 | 82.61 | 66 | 63.33 | 79 |
Bold values indicate the higher accuracy percentages achieved by the two selected methods, as this research involves comparison between two classification techniques.
Overall accuracy and Kappa hat scores over different years
| Statistical parameters | Method | 2016 | 2019 | 2021 | 2023 |
|---|---|---|---|---|---|
| Overall accuracy | Region-oriented | 89.6 | 91.7 | 85.6 | 93.6 |
| Pixel | 80.77 | 81.1 | 79.03 | 77.18 | |
| Kappa coefficient | Region-oriented | 0.88 | 0.9 | 0.94 | 0.92 |
| Pixel | 0.7053 | 0.6944 | 0.697 | 0.6616 |
Bold values indicate the higher accuracy percentages achieved by the two selected methods, as this research involves comparison between two classification techniques.
4.3 Regression analysis
LULC of Nowshera is controlled by various environmental and geophysical factors. For instance, rapid urbanization is primarily attributed to population explosion; however, the role of other drivers–environmental too needs evaluation. To assess correlation and regression analyses among the LULC changes in terms of underlying processes and between different chosen variables, including meteorological (population, temperature, and precipitation), categories were generated by IBM SPSS Statistics V22, and their outcomes are shown in Tables 11 and 12.
Correlation analysis
| Water | Agriculture | Forest | Urban land | Barren land | Precipitation | Temperature | Population | ||
|---|---|---|---|---|---|---|---|---|---|
| Water | Pearson correlation | 1 | 0.676 | −0.516 | 0.575 | −0.877 | −0.995* | 0.473 | 0.703 |
| Sig. (2-tailed) | 0.527 | 0.655 | 0.610 | 0.319 | 0.047 | 0.686 | 0.304 | ||
| Agriculture | Pearson correlation | 1 | −0.980 | 0.991 | −0.947 | −0.834 | 0.969 | 0.999* | |
| Sig. (2-tailed) | 0.128 | 0.083 | 0.208 | 0.372 | 0.159 | 0.023 | |||
| Forest | Pearson correlation | 1 | −0.998* | 0.864 | 0.707 | −0.999* | −0.972 | ||
| Sig. (2-tailed) | 0.037 | 0.336 | 0.500 | 0.031 | 0.151 | ||||
| Urban land | Pearson correlation | 1 | −0.897 | −0.755 | 0.993* | 0.996* | |||
| Sig. (2-tailed) | 0.291 | 0.455 | 0.050 | 0.042 | |||||
| Barren land | Pearson correlation | 1 | 0.967 | −0.838 | −0.958 | ||||
| Sig. (2-tailed) | 0.164 | 0.367 | 0.185 | ||||||
| Precipitation | Pearson correlation | 1 | −0.671 | −0.853 | |||||
| Sig. (2-tailed) | 0.431 | 0.349 | |||||||
| Temperature | Pearson correlation | 1 | 0.981 | ||||||
| Sig. (2-tailed) | 0.122 | ||||||||
| Population | Pearson correlation | 1 | |||||||
| Sig. (2-tailed) | |||||||||
*Correlation is significant at the 0.05 level (two-tailed).
Regression analysis for land use/cover change and underlying factors
| R | R 2 | Adjusted R 2 | SE of estimate | |
|---|---|---|---|---|
| Water | 0.971a | 0.94 | 0.884 | 1.95 |
| Agriculture | 0.969a | 0.94 | 0.878 | 1.91 |
| Forest | 0.999a | 1 | 0.995 | 3.51 |
| Urban land | 0.993a | 0.99 | 0.972 | 4.1 |
| Barren land | 0.838a | 0.7 | 0.405 | 2.68 |
a Predictors: Population, temperature, and rainfall.
The correlation analysis in Table 11 and Figure 4 shows a significant negative relationship between water and precipitation. In Figure 4, the visualization clearly depicts the relationship between various environmental and geographical categories, with color intensities indicating the strength and direction of correlations. Positive correlations are shown in warmer colors (toward red), indicating a direct relationship, while negative correlations are shown in cooler colors (toward blue), indicating an inverse relationship. Agriculture has a positive relationship with urban land and population, aligning with literature findings that urban expansion leads to increased agricultural practices nearby. The literature [70,71] also proved the fact that with the increase in the land clearing for human settlements, the agriculture practices near the Urban area have also increased. Forest shares a negative relationship with urban land. Urban land and population have a positive relationship with temperature, implying their contribution to temperature variations. Regression analysis examined the impact of climatic variables (temperature, precipitation) and population on land use changes. The selected variables could explain 99% of the changes in water, 98% in agriculture, 86% in forest, 89% in urban land, and 96% in barren land, indicating their substantial influence. Table 12 shows that 94.2% of water body dynamics, 93.9% of agricultural land changes, and over 98% of forest and urban land transitions are explained by the regression model incorporating population, temperature, and precipitation. However, the model could only explain 70.3% of barren land changes, suggesting the need to include additional variables like topography and soil properties for better robustness.
Heat map of Pearson correlation coefficients among land cover categories.
To study the impact of urban land and population on the selected climatic variables (temperature and precipitation), regression analysis was also performed (Table 13). Urban land was selected among all land use classes along with population as independent variables. The results revealed a stronger impact on both dependent variables, i.e., 84% on temperature and 45% on precipitation.
Regression analysis for the impact of land use (Urban Land) on local climate
| R | R 2 | Adjusted R 2 | SE of estimate | |
|---|---|---|---|---|
| Temperature | 0.959a | 0.92 | 0.841 | 0.61 |
| Precipitation | 0.853a | 0.73 | 0.456 | 9.12 |
a Predictor: Urban land and population.
4.4 Change detection
Land cover change analysis provides insights into landscape dynamics [72,73,74]. The post-classification change detection technique, utilizing cross-tabulation matrices, is efficient for identifying the nature and rate of land cover changes in urban areas [75,76,77,78]. Figure 5 illustrates the significant land use conversions in Nowshera from 2016 to 2023, providing insights into the dynamic transformations within the region during this time.
Major land use conversion in Nowshera from 2016 to 2023: (Left) region-oriented and (Right) pixel-based.
4.5 Land cover category changes
In the period spanning from 2016 to 2023, this study identified significant land cover transitions using two distinct methodological approaches: per-pixel and region-oriented analysis. Figure 6(a) delineates the transformations determined through the per-pixel method, highlighting three predominant transitions. From 2016 to 2023, significant land cover shifts were mapped using per-pixel and region-oriented methods. Per-pixel analysis showed major transitions: agriculture to barren land (11.22%), barren land to forest, and agriculture, urban land (4.29, 22.66, and 7.31%), highlighting urban expansion and greening efforts.
Quantitative assessment of land cover transitions: (a) pixel-based and (b) region-oriented based.
Conversely, Figure 6(b) encapsulates the major land cover changes ascertained through region-oriented, and it revealed different key changes: agriculture to barren land (10.27), an increase in agricultural land (14.61%), and barren to forests (3,32%), indicating shifts toward irrigation, cultivation, and reforestation.
4.6 Spatiotemporal dynamics of climate
Visualization of the spatiotemporal temperature and rainfall patterns supplemented by quantitative analysis. Figures 7 and 8, respectively, illustrate annual and monthly variations across Nowshera through area-wide heat maps and time-series plots for the years 2016, 2019, and 2021.
Spatial distribution of minimum and maximum temperature across Nowshera District in 2016, 2019, and 2021.
Time-series analysis of monthly minimum and maximum temperature in Nowshera District for 2016, 2019, and 2021.
In summary, a multi-faceted analysis of high-resolution climate grids elucidated local temperature and precipitation dynamics and relationships to observed land cover transitions from 2016 to 2021. The integration of earth observation and climate data provided a robust framework for investigating coupled human-environment processes shaping Nowshera’s landscape. The mean annual maximum temperature is recorded with a value of 23.45, 22.55, and 23.50 (in °C) in the years 2016, 2019, and 2021, respectively (Figures 7 and 8). There is an obvious effect of LULC change in the temperature of an area. The decrease in temperature from 23.45°C in 2016 to 22.55°C in 2019 might be due to the increase in forest and agriculture cover, which may cool down the temperature significantly (Figure 3 and Table 5).
High-resolution annual precipitation grids from the WorldClim dataset were analyzed to elucidate rainfall variability in the Nowshera District from 2016 to 2021. Interpolated raster surfaces visualized the spatial distribution of cumulative annual precipitation for each study year (Figure 9). Supplementing this, line charts characterized the range and distribution of precipitation across the district over time (Figure 10). Grid summary statistics revealed considerable inter-annual fluctuations, with total annual rainfall ranging from 554.5 to 706.2 mm.
Spatial distribution of annual precipitation in Nowshera District for 2016, 2019, and 2021.
Time-series analysis of monthly minimum and maximum precipitation in Nowshera District for 2016, 2019, and 2021.
5 Discussion
This study analyzed LULC changes in Nowshera, Pakistan, from 2016 to 2023, using Sentinel-2 satellite imagery with object-based image analysis region-oriented and pixel-based classification methods. The classified data derived from both region-oriented and pixel-based methods, substantiating the observed expansion in urban areas, the dynamic shifts in forest coverage, the fluctuation in agricultural land, and the variations in water bodies over the study period, are thoroughly documented.
The classification results are shown in Figure 3, Tables 5 and 6. Table 5 reveals significant land use changes in Nowshera between 2016 and 2023, with urban land expanding initially before stabilizing and notable increases in water bodies and forest areas, highlighting shifts toward urbanization and environmental conservation efforts. These changes reflect the critical need for sustainable land management practices in Nowshera to navigate the challenges of development and conservation.
This study found a pronounced increase in urban areas, from 54.01 to 40.541 km2, as documented in Table 6, correlating with Nowshera’s rapid population growth and urbanization pressures [44]. Agricultural land is subject to erratic patterns as it first grows and then declines, probably because of urban sprawl and the shocks of 2022 floods [79], highlighting food security and rural livelihood woes. With forest area gains of 33.50–20.504 km2, the achieved results are well inside of the Tsunami Tree Project goal in context to afforestation interventions improving land cover dynamics [80]. Additionally, a significant reduction in barren land indicates substantial greening efforts, further supported by forest cover increases. This analysis reveals that urban expansion and agricultural land reductions are primary concerns driven by demographic and development pressures, corroborating findings from previous Nowshera research [46,47].
As presented in Section 4, Figures 5 and 6 illuminate Nowshera’s land use evolution from 2016 to 2023, capturing barren into forest (3.32% and 4.29%) increases via afforestation, while barren to urban land is 8.38%, 7.31% in pixel-based and region-oriented, respectively, via urbanization.
The results of land classification from 2016 to 2023 are presented in Tables 7–10. Region-oriented methods show better performance than pixel-based ones in terms of water (PA: 90–98.07%) and urban land (PA: 74–96%), agriculture (70–100%) detection, as well as overall classification (93.6% accuracy in 2023) [63].
Accuracy discrepancies across classes are attributed to spectral confusion [81] and Sentinel-2’s resolution limitations [82]. Table 10 introduces overall accuracy and Kappa coefficients; the average yearly kappa coefficients on region resolved (0.88–0.94) results are always greater than the pixel-based (pixel level, 2023-year value = 0.6616), showing a better performance after region analysis and during all years, reflecting the precision of methodological selection in land cover classification.
Table 11 reveals significant correlations, with urban expansion and population growth closely linked (R = 0.996, p < 0.05), highlighting demand-driven development. Urban areas also significantly affect temperature increases (R 2 = 0.981), indicative of urban heat island effects, which influence urban dynamics and safety [83]. Agricultural intensification is strongly correlated with population (R = 0.999, p < 0.05) and contributes to regional warming (R = 0.969). Water bodies and vegetation decline as precipitation decreases (R = −0.995, p < 0.05), with deforestation linked to agricultural and urban development (R = −0.980 and R = −0.998, respectively).
The regression analysis in Tables 12 and 13 shows urban land use’s impact on climate, with significant correlations to temperature (R = 0.959 and R 2 = 0.920) and precipitation (R = 0.853 and R 2 = 0.728), underscoring urbanization’s climatic effects.
A heatmap of Pearson correlation coefficients (Figure 4) visually summarizes these dynamics, emphasizing the need for integrated management strategies to address climate change implications of land use alterations.
The relationship between LULC changes and climate variations in Nowshera, Pakistan, referencing climate data from 2016 to 2021 and LULC data up to 2023, are shown in Figures 7–10 and Table 5. The study identifies a notable temperature increase (max rise: 0.92°C from 23.5 to 24.42°C) alongside urban land growth, suggesting urban heat island effects. Urban expansion, particularly between 2016 and 2019 and 2021 and 2023, parallels rising temperatures, affirming a correlation with warming (R = 0.993 and p < 0.05).
The climate data for the study cover the period from 2016 to 2021. It is acknowledged that such a time frame is relatively short to perform a comprehensive climate trend analysis. However, the period was selected for the study due to the availability of reliable data, which is representative of understanding the past and present climate states in the region. The time frame helps to capture significant recent land use changes in Nowshera and reflect upon rapid urbanization and conservation. Thus, addressing the changes in this specific period helps determine the short-term effects to be compared to long-term outcomes that may be analyzed in extended study periods in future research.
Contrary to initial observations, water bodies and forest areas expanded by 2023, indicating enhanced environmental conservation, which could influence local climate, especially moisture dynamics. Despite a perceived decrease in precipitation (75 mm drop from 480 to 406 mm), these LULC shifts (notably, the increase in forests from 61.46 to 94.96 region-oriented and in water bodies from 26.41 to 38.88 region-oriented) suggest complex interactions affecting precipitation and possibly mitigating aridity through enhanced moisture recycling capacities.
The data reveal discrepancies between observed temperature trends and LULC changes, with urban areas showing a clear correlation with temperature increases. This complexity highlights the need for nuanced understanding and further research into how LULC changes contribute to local climate patterns, acknowledging potential delays in climate responses to land modifications and the role of external climatic drivers. In addition, the accurate results of the classification process could be affected by the sample size being very small.
In summary, this analysis underscores the dynamic interplay of human activities, LULC changes, and climatic factors in shaping Nowshera’s environment, emphasizing the importance of informed land use planning and climate resilience strategies amidst data limitations and the need for higher-resolution datasets for accurate analysis.
6 Conclusion
In this study, LULC changes were quantified using proper supervised classification methods based on regions and pixels and yielded convincing results. The classification results revealed that region-oriented showed superiority to pixel-based methods with better accuracies and Kappa coefficients for all years, which confirms the applicability of this approach method in LULC mapping in Nowshera.
The key findings comprise the urban land expansion, change in the water body area, and barren land decrease, representing LULC changes between multi-date maps from 2016 to 2023. Moderate correlations between LULC changes and climate variables were observed using the Pearson coefficient matrix, especially in regions experiencing land-use conversions to urbanized lands and increased cultivation of the area.
The research demonstrated that Sentinel-2 data and QGIS have a high potential for mapping LULC, with classification accuracies ranging from 77.18 to 93.60%. The results are thus able to provide an understanding of the relationships between LULC changes and climate variations observed, which was one of the main aims of this study.
This signifies that the region-oriented image analysis technique has clearly outperformed pixel-based classification and appeared to be highly accurate in analyzing LULC changes regarding Nowshera. This result highlights that more of the advanced classifiers should be used for improving LULC information.
The study highlights the critical role of high-resolution imagery and spatial analysis in shaping Nowshera’s environmental management and policy, influenced by climate and human activity. Limitations noted include the potential benefits of higher-resolution imagery, additional field surveys for validation, and the exploration of more diverse data sources for enhanced accuracy. Future research should focus on expanding the scope of change detection and regression analysis to incorporate additional climate factors and socioeconomic data, providing a more comprehensive understanding of the land-climate nexus in Nowshera.
Acknowledgements
The authors are grateful for the access to Sentinel-2 satellite imagery provided by the European Space Agency (ESA), which served as a foundational element for the spatial analysis conducted in this study. We also acknowledge the use of QGIS software, an open-source geographical information system, which enabled the comprehensive processing and analysis of the imagery. Special thanks to the Pakistan Meteorological Department and the Pakistan Bureau of Statistics for supplying the climate and population data, respectively, crucial for our regression analysis. The efforts of Rabia Shabbir in facilitating the acquisition of these datasets were invaluable to the success of this research.
-
Funding information: The Prince of Songkla University, Hat Yai, Songkhla, Thailand, provided Ms. Farnaz with a scholarship to conduct this research as part of her Ph.D degree program, along with partial funding for publication.
-
Author contributions: Farnaz: conceptualization, writing original draft, data collection, and classification of the data and analysis. Benazeer Iqbal: analysis and writing. Rabia Shabbir and Narissara Nuthammachot: proofreading and project supervision.
-
Conflict of interest: The authors declare that there is no conflict of interest regarding materials presented in the manuscript. All authors make a significant contribution to the manuscript.
-
Data availability statement: The authors confirm that the datasets that support the findings of this research, are publicly available and all sources are cited in the text and generated data, which is further analysed is available within the article.
References
[1] Schirpke U, Tscholl S, Tasser E. Spatio-temporal changes in ecosystem service values: Effects of land-use changes from past to future (1860–2100). J Env Manag. 2020;272:111068.10.1016/j.jenvman.2020.111068Search in Google Scholar PubMed
[2] Stephens L, Fuller D, Boivin N, Rick T, Gauthier N, Kay A, et al. Archaeological assessment reveals Earth’s early transformation through land use. Science. 2019;365(6456):897–902.10.1126/science.aax1192Search in Google Scholar PubMed
[3] Syed A, Raza T, Bhatti TT, Eash NS. Climate Impacts on the agricultural sector of Pakistan: Risks and solutions. Env Chall. 2022;6:100433.10.1016/j.envc.2021.100433Search in Google Scholar
[4] Rufin P, Müller H, Pflugmacher D, Hostert P. Land use intensity trajectories on Amazonian pastures derived from Landsat time series. Int J Appl Earth Obs Geoinf. 2015;41:1–10.10.1016/j.jag.2015.04.010Search in Google Scholar
[5] Devaraju N, Bala G, Nemani R. Modelling the influence of land‐use changes on biophysical and biochemical interactions at regional and global scales. Plant, Cell Env. 2015;38(9):1931–46.10.1111/pce.12488Search in Google Scholar PubMed
[6] Verburg PH, Alexander P, Evans T, Magliocca NR, Malek Z, Rounsevell MD, et al. Beyond land cover change: towards a new generation of land use models. Curr Opin Environ Sustainability. 2019;38:77–85.10.1016/j.cosust.2019.05.002Search in Google Scholar
[7] Tobias TN. Chief Kerry’s moose: a guidebook to land use and occupancy mapping, research design, and data collection. Vancouver, B.C., Canada: Union of BC Indian Chiefs and Ecotrust Canada; 2000.Search in Google Scholar
[8] Fahad S, Li W, Lashari AH, Islam A, Khattak LH, Rasool U. Evaluation of land use and land cover Spatio-temporal change during rapid Urban sprawl from Lahore, Pakistan. Urban Clim. 2021;39:100931.10.1016/j.uclim.2021.100931Search in Google Scholar
[9] Halefom A, Teshome A, Sisay E, Ahmad I. Dynamics of land use and land cover change using remote sensing and GIS: a case study of Debre Tabor Town, South Gondar, Ethiopia. J Geogr Inf Syst. 2018;10(2):165.10.4236/jgis.2018.102008Search in Google Scholar
[10] Liu D, Cai S. A spatial-temporal modeling approach to reconstructing land-cover change trajectories from multi-temporal satellite imagery. Ann Assoc Am Geogr. 2012;102(6):1329–47.10.1080/00045608.2011.596357Search in Google Scholar
[11] Zhou Y, Li Y, Li W, Li F, Xin Q. Ecological responses to climate change and human activities in the arid and semi-arid regions of Xinjiang in China. Remote Sens. 2022;14(16):3911.10.3390/rs14163911Search in Google Scholar
[12] Lang R, Shao G, Pijanowski BC, Farnsworth RL. Optimizing unsupervised classifications of remotely sensed imagery with a data-assisted labeling approach. Comput Geosci. 2008;34(12):1877–85.10.1016/j.cageo.2007.10.011Search in Google Scholar
[13] Acuña-Alonso C, García-Ontiyuelo M, Barba-Barragáns D, Álvarez X. Development of a convolutional neural network to accurately detect land use and land cover. MethodsX. 2024;12:102719.10.1016/j.mex.2024.102719Search in Google Scholar PubMed PubMed Central
[14] Bacha AS, Van Der Werff H, Shafique M, Khan H. Transferability of object-based image analysis approaches for landslide detection in the Himalaya Mountains of northern Pakistan. Int J Remote Sens. 2020;41(9):3390–410.10.1080/01431161.2019.1701725Search in Google Scholar
[15] Sheykhmousa M, Mahdianpari M, Ghanbari H, Mohammadimanesh F, Ghamisi P, Homayouni S. Support vector machine versus random forest for remote sensing image classification: A meta-analysis and systematic review. IEEE J Sel Top Appl Earth Obs Remote Sens. 2020;13:6308–25.10.1109/JSTARS.2020.3026724Search in Google Scholar
[16] Campbell C, Ying Y. Learning with support vector machines. Switzerland AG: Springer Nature; 2022.Search in Google Scholar
[17] Awad M, Khanna R, Awad M, Khanna R. Support vector machines for classification. Efficient learning machines: Theories, concepts, and applications for engineers and system designers. Berkeley, CA: Apress; 2015. p. 39–66.10.1007/978-1-4302-5990-9_3Search in Google Scholar
[18] Ghaddar B, Naoum-Sawaya J. High dimensional data classification and feature selection using support vector machines. Eur J Oper Res. 2018;265(3):993–1004.10.1016/j.ejor.2017.08.040Search in Google Scholar
[19] Bhavsar H, Panchal MH. A review on support vector machine for data classification. Int J Adv Res Comput Eng Technol (IJARCET). 2012;1(10):185–9.Search in Google Scholar
[20] Alshari EA, Gawali BW. Development of classification system for LULC using remote sensing and GIS. Glob Transit Proc. 2021;2(1):8–17.10.1016/j.gltp.2021.01.002Search in Google Scholar
[21] Hay GJ, Castilla G. Geographic Object-Based Image Analysis (GEOBIA): A new name for a new discipline. Object-based image analysis: Spatial concepts for knowledge-driven remote sensing applications. Berlin/Heidelberg, Germany: Springer; 2008. p. 75–89.10.1007/978-3-540-77058-9_4Search in Google Scholar
[22] Li D, Wu S, Liang Z, Li S. The impacts of urbanization and climate change on urban vegetation dynamics in China. Urban For Urban Green. 2020;54:126764.10.1016/j.ufug.2020.126764Search in Google Scholar
[23] Tariq A, Jiango Y, Li Q, Gao J, Lu L, Soufan W, et al. Modelling, mapping and monitoring of forest cover changes, using support vector machine, kernel logistic regression and naive bayes tree models with optical remote sensing data. Heliyon. 2023;9(2):e13212.10.1016/j.heliyon.2023.e13212Search in Google Scholar PubMed PubMed Central
[24] García-Berná JA, Ouhbi S, Benmouna B, Garcia-Mateos G, Fernández-Alemán JL, Molina-Martínez JM. Systematic mapping study on remote sensing in agriculture. Appl Sci. 2020;10(10):3456.10.3390/app10103456Search in Google Scholar
[25] Huggel C, Kääb A, Salzmann N. Evaluation of QuickBird and IKONOS imagery for assessment of high-mountain hazards. EARSeL eProceedings. 2006;5(1):51–62.Search in Google Scholar
[26] Sawaya KE, Olmanson LG, Heinert NJ, Brezonik PL, Bauer ME. Extending satellite remote sensing to local scales: land and water resource monitoring using high-resolution imagery. Remote Sens Env. 2003;88(1–2):144–56.10.1016/j.rse.2003.04.006Search in Google Scholar
[27] Amri K, Rabai G, Benbakhti IM, Khennouche MN. Mapping geology in Djelfa District (Saharan Atlas, Algeria), using Landsat 7 ETM + data: an alternative method to discern lithology and structural elements. Arab J Geosci. 2017;10(4):87.10.1007/s12517-017-2883-6Search in Google Scholar
[28] Bhardwaj A, Joshi PK, Sam L, Singh MK, Singh S, Kumar R. Applicability of Landsat 8 data for characterizing glacier facies and supraglacial debris. Int J Appl Earth Obs Geoinf. 2015;38:51–64.10.1016/j.jag.2014.12.011Search in Google Scholar
[29] Burns P, Nolin A. Using atmospherically-corrected Landsat imagery to measure glacier area change in the Cordillera Blanca, Peru from 1987 to 2010. Remote Sens Env. 2014;140:165–78.10.1016/j.rse.2013.08.026Search in Google Scholar
[30] Forkuor G, Dimobe K, Serme I, Tondoh JE. Landsat-8 vs. Sentinel-2: examining the added value of sentinel-2’s red-edge bands to land-use and land-cover mapping in Burkina Faso. GISci Remote Sens. 2018;55(3):331–54.10.1080/15481603.2017.1370169Search in Google Scholar
[31] Khan A, Hansen MC, Potapov P, Stehman SV, Chatta AA. Landsat-based wheat mapping in the heterogeneous cropping system of Punjab, Pakistan. Int J Remote Sens. 2016;37(6):1391–410.10.1080/01431161.2016.1151572Search in Google Scholar
[32] Nasiri V, Deljouei A, Moradi F, Sadeghi SMM, Borz SA. Land use and land cover mapping using Sentinel-2, Landsat-8 Satellite images, and google earth engine: A comparison of two composition methods. Remote Sens. 2022;14(9):1977.10.3390/rs14091977Search in Google Scholar
[33] Cantrell SJ, Sampath A, Vrabel JC, Bresnahan P, Anderson C, Kim M, Park S. System characterization report on the Pléiades Neo Imager. USA: US Geological Survey Open-File Report 2021–1030, 52 p.; 2023.10.3133/ofr20211030PSearch in Google Scholar
[34] Barati AA, Zhoolideh M, Azadi H, Lee J-H, Scheffran J. Interactions of land-use cover and climate change at global level: How to mitigate the environmental risks and warming effects. Ecol Indic. 2023;146:109829.10.1016/j.ecolind.2022.109829Search in Google Scholar
[35] Thapa P. The relationship between land use and climate change: A case study of Nepal. The Nature, causes, effects and mitigation of climate change on the environment. Norderstedt, Germany: BoD – Books on Demand; 2021.10.5772/intechopen.98282Search in Google Scholar
[36] Salazar A, Baldi G, Hirota M, Syktus J, McAlpine C. Land use and land cover change impacts on the regional climate of non-Amazonian South America: A review. Glob Planet Change. 2015;128:103–19.10.1016/j.gloplacha.2015.02.009Search in Google Scholar
[37] Asad A, Ullah K, Butt MJ, bin Hussin Labban A. Analysis of urban heat island effects in high altitude areas of Pakistan. Remote Sens Appl: Soc Env. 2023;32:101071.10.1016/j.rsase.2023.101071Search in Google Scholar
[38] Tariq A, Mumtaz F, Zeng X, Baloch MYJ, Moazzam MFU. Spatio-temporal variation of seasonal heat islands mapping of Pakistan during 2000–2019, using day-time and night-time land surface temperatures MODIS and meteorological stations data. Remote Sens Appl: Soc Env. 2022;27:100779.10.1016/j.rsase.2022.100779Search in Google Scholar
[39] Zulfiqar M, Mian IA, Khan N, Hanan NGDF, Qazi MA, Fahad S, et al. District Nowshera-Pakistan. USA: IISTE (International Institute for Science, Technology, and Education); 2019.Search in Google Scholar
[40] Basharat M, Siddique M, Ashraf J, Ismail M. Assessment of groundwater development potential in district Nowshera using groundwater modeling. Pak J Water Resour. 2009;13(2):17–27.Search in Google Scholar
[41] Ali Shah I, Muhammad Z, Khan H. Impact of climate change on spatiotemporal variations in the vegetation cover and hydrology of district Nowshera. J Water Clim Change. 2022;13(11):3867–82.10.2166/wcc.2022.229Search in Google Scholar
[42] Ahmad W, Iqbal J, Nasir MJ, Ahmad B, Khan MT, Khan SN, et al. Impact of land use/land cover changes on water quality and human health in district Peshawar Pakistan. Sci Rep. 2021;11(1):16526.10.1038/s41598-021-96075-3Search in Google Scholar PubMed PubMed Central
[43] Ali S, Shah SZ, Khan MS, Khan WM, Khan Z, Hassan N, et al. Floristic list, ecological features and biological spectrum of district Nowshera, Khyber Pakhtunkhwa, Pakistan. Acta Ecol Sin. 2019;39(2):133–41.10.1016/j.chnaes.2018.08.007Search in Google Scholar
[44] Khan MA, Khan MS, Khan K. Economic aspects of literacy on humans in District Nowshera. Econ Res. 2020;4(3):267–80.10.29226/TR1001.2020.205Search in Google Scholar
[45] Afridi AK, Khan R, Mussawar S. A comparative study of the disparities between urban-rural education at elementary level: A case study of District Nowshera, NWFP Pakistan. JL Soc’y. 2007;36:1.Search in Google Scholar
[46] Khattak RH, Teng L, Mehmood T, Ahmad S, Bari F, Rehman EU, et al. Understanding the dynamics of human–Wildlife conflicts in North-Western Pakistan: Implications for sustainable conservation. Sustainability. 2021;13(19):10793.10.3390/su131910793Search in Google Scholar
[47] Pakistan Bureau of Statistics. Pakistan Census 2023. Islamabad, Pakistan, Census Report; 2023.Search in Google Scholar
[48] Nasir MJ, Tufail M, Ayaz T, Khan S, Khan AZ, Lei M. Groundwater quality assessment and its vulnerability to pollution: A study of district Nowshera, Khyber Pakhtunkhwa, Pakistan. Env Monit Assess. 2022;194(10):692.10.1007/s10661-022-10399-9Search in Google Scholar PubMed
[49] Qazi IA, Gul N, Zulfiqar NK, Hanan F, Jehan SFNAS, Naushad M. Impacts of climate change on social life of older people in District Nowshera-Pakistan. Nairobi, Kenya: IISTE; 2019.Search in Google Scholar
[50] Fida M, Hussain I, Rashid A, Shah SAA, Khan S. Change detection in land use and land cover of District Charsadda Pakistan along River Kabul (2010 flood): taking advantage of geographic information system and remote sensing. Geol Behav (GBR). 2021;5(2):35–41.10.26480/gbr.02.2021.40.46Search in Google Scholar
[51] Lai J-S, Tsai F. Improving GIS-based landslide susceptibility assessments with multi-temporal remote sensing and machine learning. Sensors. 2019;19(17):3717.10.3390/s19173717Search in Google Scholar PubMed PubMed Central
[52] Congedo L. Semi-Automatic Classification Plugin: A Python tool for the download and processing of remote sensing images in QGIS. J Open Source Softw. 2021;6(64):3172.10.21105/joss.03172Search in Google Scholar
[53] Department PM. Home. Islamabad: Pakistan Meteorological Department.Search in Google Scholar
[54] Statistics PBo. Census 2017 district wise results. Census Report; 2017.Search in Google Scholar
[55] Cardoso-Fernandes J, Lima A, Teodoro A, eds. Potential of Sentinel-2 data in the detection of lithium (Li)-bearing pegmatites: a study case. Earth resources and environmental remote sensing/GIS applications IX. Bellingham, WA: SPIE (Society of Photo-Optical Instrumentation Engineers); 2018.10.1117/12.2326285Search in Google Scholar
[56] Weinmann M, Maier P, Florath J, Weidner U. Investigations on the potential of hyperspectral and Sentinel-2 data for land-cover/land-use classification. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial. Inf Sci. 2018;4:155–62.10.5194/isprs-annals-IV-1-155-2018Search in Google Scholar
[57] Li J, Bioucas-Dias JM, Plaza A. Semisupervised hyperspectral image segmentation using multinomial logistic regression with active learning. IEEE Trans Geosci Remote Sens. 2010;48(11):4085–98.10.1109/TGRS.2010.2060550Search in Google Scholar
[58] Maxwell AE, Warner TA, Fang F. Implementation of machine-learning classification in remote sensing: An applied review. Int J Remote Sens. 2018;39(9):2784–817.10.1080/01431161.2018.1433343Search in Google Scholar
[59] Guizani D, Buday-Bódi E, Tamás J, Nagy A. Land cover modelling with Sentinel 2 in water balance calculations of Urban sites. J Cent Eur Green Innov. 2023;11(3):70–83.10.33038/jcegi.4520Search in Google Scholar
[60] Wieland M, Fichtner F, Martinis S, Groth S, Krullikowski C, Plank S, et al. S1S2-Water: A global dataset for semantic segmentation of water bodies from Sentinel-1 and Sentinel-2 satellite images. IEEE J Sel Top Appl Earth Obs Remote Sens. 2023;17:1084–99.10.1109/JSTARS.2023.3333969Search in Google Scholar
[61] Tzotsos A, Argialas D. Support vector machine classification for object-based image analysis. Object-based image analysis: Spatial concepts for knowledge-driven remote sensing applications; 2008. p. 663–77.10.1007/978-3-540-77058-9_36Search in Google Scholar
[62] Marangoz A, Alkış Z. Detection of Urban Details from Satellite Images using Objectbased Classification Methods and Integration to GIS. Zonguldak, Turkey: The 4th Remote Sensing and GIS Symposium; 2012.Search in Google Scholar
[63] Ez-zahouani B, Teodoro A, El Kharki O, Jianhua L, Kotaridis I, Yuan X, et al. Remote sensing imagery segmentation in object-based analysis: a review of methods, optimization, and quality evaluation over the past 20 years. Remote Sens Appl: Soc Environ. 2023;101031. 10.1016/j.rsase.2023.101031.Search in Google Scholar
[64] Mohd Rahim MA. Assessment of building detection accuracy based on different altitude of high-resolution images. Malaysia: Universiti Teknologi Mara Perlis; 2022.Search in Google Scholar
[65] Liu T, Yang L, Lunga D. Change detection using deep learning approach with object-based image analysis. Remote Sens Environ. 2021;256:112308.10.1016/j.rse.2021.112308Search in Google Scholar
[66] Olofsson P, Foody GM, Herold M, Stehman SV, Woodcock CE, Wulder MA. Good practices for estimating area and assessing accuracy of land change. Remote Sens Environ. 2014;148:42–57.10.1016/j.rse.2014.02.015Search in Google Scholar
[67] Mateen S, Nuthammachot N, Techato K, Ullah N. Billion Tree Tsunami Forests classification using image fusion technique and random forest classifier applied to Sentinel-2 and Landsat-8 images: A case study of Garhi Chandan Pakistan. ISPRS Int J Geo-Information. 2022;12(1):9.10.3390/ijgi12010009Search in Google Scholar
[68] Hassan MM, Nazem MNI. Examination of land use/land cover changes, urban growth dynamics, and environmental sustainability in Chittagong city, Bangladesh. Environ Dev Sustainability. 2016;18(3):697–716.10.1007/s10668-015-9672-8Search in Google Scholar
[69] Munsi M, Malaviya S, Oinam G, Joshi P. A landscape approach for quantifying land-use and land-cover change (1976–2006) in middle Himalaya. Reg Environ Change. 2010;10:145–55.10.1007/s10113-009-0101-0Search in Google Scholar
[70] Bren d’Amour C, Reitsma F, Baiocchi G, Barthel S, Güneralp B, Erb K-H, et al. Future urban land expansion and implications for global croplands. Proc Natl Acad Sci. 2017;114(34):8939–44.10.1073/pnas.1606036114Search in Google Scholar PubMed PubMed Central
[71] Patel SK, Verma P, Shankar Singh G. Agricultural growth and land use land cover change in peri-urban India. Environ Monit Assess. 2019;191:1–17.10.1007/s10661-019-7736-1Search in Google Scholar PubMed
[72] Milly P, Shmakin A. Global modeling of land water and energy balances. Part I: The land dynamics (LaD) model. J Hydrometeorol. 2002;3(3):283–99.10.1175/1525-7541(2002)003<0283:GMOLWA>2.0.CO;2Search in Google Scholar
[73] Ellis EC. Anthropogenic transformation of the terrestrial biosphere. Philos Trans R Soc A: Math Phys Eng Sci. 2011;369(1938):1010–35.10.1098/rsta.2010.0331Search in Google Scholar
[74] Sas-Bojarska A. Landscape as a potential key concept in urban environmental planning: The case of Poland. Urban Plan. 2021;6(3):295–305.10.17645/up.v6i3.4044Search in Google Scholar
[75] Panuju DR, Paull DJ, Griffin AL. Change detection techniques based on multispectral images for investigating land cover dynamics. Remote Sens. 2020;12(11):1781.10.3390/rs12111781Search in Google Scholar
[76] Taloor AK, Kumar V, Singh VK, Singh AK, Kale RV, Sharma R, et al. Land use land cover dynamics using remote sensing and GIS Techniques in Western Doon Valley, Uttarakhand, India. In: Advances in Geographical and Environmental Sciences (AGES), Geoecology of Landscape Dynamics. Springer; 2023. p. 37–51. 10.1007/978-981-15-2097-6_23.Search in Google Scholar
[77] Peiman R. Pre-classification and post-classification change-detection techniques to monitor land-cover and land-use change using multi-temporal Landsat imagery: a case study on Pisa Province in Italy. Int J Remote Sens. 2011;32(15):4365–81.10.1080/01431161.2010.486806Search in Google Scholar
[78] Serra P, Pons X, Sauri D. Post-classification change detection with data from different sensors: some accuracy considerations. Int J Remote Sens. 2003;24(16):3311–40.10.1080/0143116021000021189Search in Google Scholar
[79] Nation T. Villages flooded in Nowshera as embankment breaks. Pakistan: The Nation; 2022 [cited 2023 Nov 2]. https://www.nation.com.pk/27-Aug-2022/villages-flooded-in-nowshera-as-embankment-breaks.Search in Google Scholar
[80] Haris J. Socioeconomic impacts of the Ten Billion Tree Tsunami (TBTT) plantation project on the participating communities. Uppsala, Sweden: SLU (Swedish University of Agricultural Sciences); 2023.Search in Google Scholar
[81] Powell RL, Roberts DA, Dennison PE, Hess LL. Sub-pixel mapping of urban land cover using multiple endmember spectral mixture analysis: Manaus, Brazil. Remote Sens Environ. 2007;106(2):253–67.10.1016/j.rse.2006.09.005Search in Google Scholar
[82] Radoux J, Chomé G, Jacques DC, Waldner F, Bellemans N, Matton N, et al. Sentinel-2’s potential for sub-pixel landscape feature detection. Remote Sens. 2016;8(6):488.10.3390/rs8060488Search in Google Scholar
[83] Shiau Y-H, Yang S-F, Adha R, Peng G-S, Muzayyanah S. Dynamic and non-linear analysis of the impact of Diurnal temperature range on road traffic accidents. Climate. 2023;11(10):199.10.3390/cli11100199Search in Google Scholar
© 2025 the author(s), published by De Gruyter
This work is licensed under the Creative Commons Attribution 4.0 International License.
