Monitoring Forest Cover Trends in Nepal: Insights from 2000–2020

Abstract

This study investigates the spatial relationship between population distribution and tree cover loss in Nepal from 2000 to 2020, using satellite-based forest cover and population data along with statistical and geospatial analysis. Two statistical methods—linear regression (LR) and Geographically Weighted Regression (GWR)—were used to assess the influence of population on forest cover change. The correlation between total population and forest loss at the national level suggested little to no direct impact of population growth on forest loss. However, sub-national analysis revealed localized forest degradation, highlighting the importance of spatial and regional assessments to uncover land cover changes masked by national trends. While LR showed a weak national-level correlation, GWR revealed substantial spatial variation, with the coefficient of determination values increasing from 0.21 in 2000 to 0.59 in 2020. In some regions, local R² exceeded 0.75 during 2015 and 2020, highlighting emerging hotspot clusters where population pressure is strongly linked to deforestation, especially along major infrastructure corridors. Using very high-resolution spatial data enabled pixel-level analysis, capturing fine-scale deforestation patterns, and confirming hotspot accuracy. Overall, the findings emphasize the value of spatially explicit models like GWR for understanding human–environment interactions guiding targeted land use planning to balance development with environmental sustainability in Nepal.

1. Introduction

Vegetation cover plays an important role in our environment. Several studies documented the role of vegetation in the removal of air pollution, urban cooling, climate change mitigation, stormwater management, biodiversity support, etc. [1,2,3,4]. Additionally, vegetation cover has been associated with economic benefits such as increased property values, economic resources, and a high return on investment in urban forestry [5]. However, by clearing vegetation, especially forests for agriculture or urbanization, humans are changing the Earth’s land cover in pervasive ways [6]. In particular, the Asian continent has seen explosive growth in population in recent decades. In the last 50 years, Asia’s population has grown by more than 2.5 billion people, from around 2.2 billion in 1970 [7]. As of 2023, Asia’s population was approximately 4.7 billion, accounting for nearly 60% of the world’s total population [8]. This dramatic population increase, combined with urban expansion, has significantly influenced the demand for resources and, consequently, environmental changes, including the loss of natural habitats in Asia. In particular, the expansion of infrastructure, agriculture, and industry has led to large-scale deforestation in Southeast Asia [9], exacerbating environmental challenges such as biodiversity loss, soil erosion, and climate change [10].

Specific to South Asian countries, the forest cover has experienced significant fluctuations over the past century. Historically, the region was covered by extensive forests, ranging from the dense tropical rainforests of the Western Ghats and Sri Lanka to the temperate forests of the Himalayas. However, deforestation has accelerated due to rapid agricultural expansion, logging, infrastructure development, and population growth. For example, in India, large-scale deforestation occurred during colonial rule, primarily for timber extraction, agriculture, and railway expansion [11]. Post-independence, agricultural expansion and urbanization further contributed to forest loss. However, in recent decades, afforestation initiatives such as the Joint Forest Management (JFM) program and the Green India Mission have attempted to reverse deforestation trends, leading to an increase in tree cover in some regions [12,13]. In the case of Afghanistan, the forest cover is relatively sparse, accounting for a small percentage of the country’s total land area. Most of its forests are concentrated in the eastern provinces, such as Nuristan, Kunar, and Paktia, where coniferous forests dominate the mountainous landscapes [14]. Over the past few decades, Afghanistan has experienced significant deforestation due to illegal logging, overgrazing, fuelwood collection, and land conversion for agriculture [15]. Ongoing conflict and political instability have further exacerbated forest degradation, limiting conservation efforts. In Bangladesh, the Sundarbans, the world’s largest mangrove forest, has faced deforestation due to shrimp farming, illegal logging, and rising sea levels caused by climate change [16]. The Chittagong Hill Tracts have also seen forest degradation due to shifting cultivation and encroachment [17]. In Pakistan, forest cover is among the lowest in the region, with deforestation driven by overgrazing, illegal logging, and infrastructure projects [18]. The country has launched afforestation initiatives such as the Billion Tree Tsunami project to restore degraded landscapes [19]. Sri Lanka, despite its rich biodiversity, has witnessed deforestation due to plantation agriculture (tea, rubber, and coconut), leading to habitat loss. Government-led conservation programs aim to protect remaining forests, especially in the central highlands [20].

In contrast to the above countries in South Asia, Nepal presents a unique case where community forestry programs have successfully contributed to forest regeneration [21]. Since the 1990s, local communities have been empowered to manage forests, leading to an increase in forest cover, particularly in the mid-hills [22]. Despite these successes, forest loss continues in specific pockets [23]. As of mid-2024, Nepal’s population was estimated at approximately 30 million, growing at about 1.0% per year, with a median age of 25 years, highlighting a youthful and expanding labor force [24]. Nepal’s economy reached USD 40.9 billion in Gross Domestic Product (current prices) with per capita income at USD 1378 (nominal) [25]. After a slowdown in 2023 to 2.0% real GDP growth, projections for FY2024/25 show a rebound to 5.1%, supported by rising remittances, over USD 11 billion, or 25–26% of GDP [26], and tourism recovery. Structurally, the economy remains dominated by the services sector (62%), followed by agriculture (25–30%) and industry. Nepal’s forest resources are among its most valuable natural assets, covering approximately 5.96 million hectares, or 40.36% of the total land area, according to the National Forest Inventory 2021 [27]. Forests are concentrated in the mid-hills (37%), High Mountains (32%), and Terai (31%), and support over 118 ecosystems and 7000+ plant species, including 180 species of mammals [28]. Nepal’s community forestry program, initiated in the late 1970s, is globally recognized; over 22,000 Community Forest User Groups (CFUGs) now manage nearly 2.9 million hectares, benefiting more than 3 million households [29]. Studies show forest cover increased by over 5% between the 1990s and 2020s, largely due to CFUG-led regeneration [30,31]. However, recent literature suggests that population pressures, especially in the Terai and mid-hills, continue to strain forest ecosystems. Over 60% of rural households still depend on fuelwood as their primary energy source [32], contributing to forest degradation. The deforestation is driven by agricultural expansion, infrastructure development, and illegal logging, exacerbated by population pressures and land encroachment [33]. Additionally, in some highland regions, overgrazing, fuelwood collection, and the expansion of road networks have contributed to localized forest degradation. Climate change-induced factors, such as forest fires and shifting monsoon patterns, further threaten Nepal’s forests [34,35]. Nepal has responded with policy reforms including the Forest Act [36], Forest Policy [37], and participation in REDD+ initiatives. The country received over USD 45 million from the World Bank under REDD+ for verified emission reductions between 2015 and 2020 [38]. These advances show promise, yet balancing conservation and development remains a challenge. Without integrating population dynamics into national land use and forest management plans, Nepal risks undermining decades of progress in community-led sustainable forestry.

This study focuses on Nepal with the main goal to quantify the relationship between population dynamics and forest loss over time (2000–2020). It specifically aims to determine whether the increasing population is one of the primary drivers of tree cover loss. It also seeks to understand how this relationship has changed over time. While population growth is often linked to deforestation, it is unclear how this relationship has evolved in Nepal over different periods (2000–2020). A literature review suggests that some regions, particularly the Terai plains, have experienced rapid population expansion [39]; however, it is not clear if such an increase also led to significant forest loss in neighboring districts due to population demand. This study seeks to analyze such a complex relationship by integrating remote sensing data with population data to identify spatial and temporal trends in forest cover change. The quantitative analysis focused on whether increased population in specific regions directly contributed to deforestation during different years. The study identifies hotspots where population–deforestation relationships are strong. Understanding these dynamics is crucial for developing sustainable forest management strategies that balance population growth with conservation efforts. The findings provide valuable insights for policymakers, conservationists, and land use planners aiming to ensure the long-term sustainability of Nepal’s forests.

2. Study Area

Nepal is situated between the longitudes 80.08 E and 88.17 E and latitudes 26.39 N and 30.42 N (Figure 1). The country is mountainous (3265 m above sea level) and bordered by Tibet and India. It has an area of 147,180 sq. km [40], of which nearly 41.6% of the land area is estimated to be forest area. The total population of Nepal is estimated to be 29,694,614 [40]. Figure 1 displays the land use land cover map of Nepal obtained from the Sentinel-2 2020 ESA World Cover Map, along with provincial boundaries. There are a total of 7 provinces and 77 districts in Nepal [41].

3. Data and Sources

The datasets that were used for conducting this study are the LandScan Population Data Global 1 km [42] and the Global Forest Change (GFC) [43] dataset, the former used to extract population data and the latter used for extracting the tree cover information. The LandScan dataset is provided by the Oak Ridge National Laboratory (ORNL), has a spatial resolution of 30 arc seconds (approximately 1000 m), and is available yearly from 2000 to 2023. The LandScan data has one band which is named ‘b1’ and is described as the estimation population count per pixel. This dataset supports scientific remote sensing applications in urban planning, disaster response, epidemiology, and environmental research [44].

The GFC dataset is a product that is derived from the collaboration of Global Land Analysis & Discover (GLAD) at the University of Maryland (UMD), Google, USGS, and NASA. It measures tree cover loss per pixel and has global coverage except for Antarctica and Arctic areas. The spatial resolution of this dataset is approximately 30 m. The data is obtained from analyzing Landsat’s image timeseries. The dataset is available at a yearly level starting from the base year of 2000 to 2023. The two important bands in the GFC dataset are ‘treecover2000’ and ‘lossyear’. The ‘treecover2000’ band is a percentage. It is described as the tree canopy for the year 2000, defined as the canopy closure for all vegetation that is taller than 5 m. The ‘lossyear’ band, which is a number between 0 and 23, is described as the year of gross forest cover loss. The two datasets are in a raster format.

The two datasets are available in the Google Earth Engine catalog, which makes it convenient to procure and preprocess the data. This data is predominantly used to monitor and analyze the change in forest cover across the globe. Figure 2 and Figure 3 show the population map and tree cover percentage map for the year 2000 obtained from the LandScan and GFC datasets, respectively.

4. Materials and Methods

The time slices selected for this study were 2000, 2005, 2010, 2015, and 2020. These years were selected to map and monitor the tree cover and population dataset changes periodically, considering the dataset availability. First, the two datasets were imported into Google Earth Engine and clipped to the boundary of Nepal for the five selected years. Both the tree cover and population datasets are annual datasets. For consistency, the population dataset for the year 2000 was used as the reference or master grid. Tree cover rasters for all years were resampled and aligned to match the spatial resolution of the year 2000 population grid, which served as the master reference. Similarly, the population datasets for the other years were reprojected to the 2000 population grid. This ensured that all ten rasters (five for tree cover and five for population) were perfectly aligned pixel-by-pixel.

To extract the vegetation or tree cover percentage for a given year, the base year 2000 dataset is masked to exclude pixels where forest loss occurred prior to or during the target year, retaining only those where no loss occurred or where loss happened afterward. The following expression was used to compute vegetation tree cover for a target year:

$$TreeCove{r}_t=TreeCove{r}_{2000}\times \left(LossYear>t OR LossYear=0\right)$$

(1)

where TreeCover_t represents the tree cover raster for the target year ‘t’, TreeCover₂₀₀₀ is the baseline raster for tree cover in the year 2000, and LossYear is the loss year raster, with a value of 0 indicating no loss and values from 1 to 20 representing forest loss in the corresponding year from 2001 to 2020. Equation (1) is written as pseudocode to represent the raster calculation performed to derive tree cover for a target year. It expresses the logic used in Google Earth Engine, where the 2000 tree cover raster (TreeCover₂₀₀₀) is masked using the forest loss year raster (LossYear). This raster expression allowed the generation of annual tree cover layers based on cumulative forest loss.

For each corresponding year, only the intersecting pixels between the tree cover and population rasters were selected, ensuring the analysis focused on areas where data from both datasets were available.

Table 1. Number of valid pixels for tree cover, population, and their intersection for each selected year. The intersecting pixels represent areas where both datasets have overlapping, aligned values, used for further analysis.

Year	Tree Cover Pixels	Population Pixels	Tree Cover ∩ Population Pixels
2000	122,588	163,960	121,990
2005	112,277	129,268	94,121
2010	94,560	131,352	81,097
2015	74,074	126,889	61,132
2020	66,455	119,109	51,324

presents the intersecting pixels for each of the five years. The selected data were then processed into 1 km × 1 km grid files using QGIS (v3.34), with each file containing latitude, longitude, tree cover percentage (hereafter treecover%), and population as attributes. These yearly grid files served as inputs for the regression analysis, with tree cover percentage as the dependent variable and population as the independent variable.

4.1. Linear Regression

A linear regression (LR) analysis was conducted using Python’s (3.12.3) scikit module with ‘population’ as an independent variable and ‘treecover%’ as a dependent variable for each of the five years.

4.2. Geographically Weighted Regression

To capture spatial heterogeneity in the relationship between the tree cover and population across the entirety of Nepal, Geographically Weighted Regression (GWR) has been applied. It is an extension of linear regression (LR) that allows model coefficients to change across geographic space [45]. Unlike ordinary least squares (OLS), which estimates a single, global model, GWR generates localized models tailored to each spatial coordinate. The GWR model is formulated as

$$y_i={\beta }_0\left(u_i,v_i\right)+\sum _{k=1}^p{\beta }_k\left(u_i,v_i\right)x_{ik}+{\epsilon }_i$$

(2)

Here, (u_i, v_i) represents the spatial coordinates of each observation, and i and β_k(u_i, v_i) denote the spatially varying coefficients estimated for each location. These coefficients are derived through a weighted least-squares method:

$$\widehat{\beta }\left(u_i,v_i\right)={\left(X^T{W}_{i}X\right)}^{-1}{X}^T{W}_{i}Y$$

(3)

In this formulation, X is the design matrix of the predictor variable (population), Y is the response vector (tree cover percent), and W_i is a spatial weighting matrix that defines the influence of surrounding data points. The weights are computed using a kernel function, most commonly the Gaussian kernel:

$$w_{ij}=\exp (-{\displaystyle \frac{d_{ij}^2}{2b^2}})$$

(4)

where d_ij is the Euclidean distance between observation points i and j, and b is the bandwidth parameter that dictates the spatial scale of influence. The optimal bandwidth is selected by minimizing the Akaike Information Criterion (AIC), balancing model fit and complexity.

GWR analyses were conducted for 2000, 2005, 2010, 2015, and 2020, using an adaptive kernel bandwidth optimized through the corrected AIC (AICc). By accounting for spatial variation in relationships, GWR enables more nuanced interpretations of the link between tree cover and population, uncovering localized patterns that global models such as LR may miss. Two different outputs were presented: a derived spatial map and a summary table. The spatial map explores the local relationship between the variables, whereas the summary table gives a summary of the statistics. To validate the findings from the GWR analysis, supplementary visual inspections were conducted using high-resolution imagery available through Google Earth. This additional step allowed for a more detailed spatial verification of the GWR results. Specifically, areas that exhibited high R² values in the GWR output, indicating strong model performance, were closely examined to assess actual changes in forest cover and land use/land cover. By visually comparing these regions with time-series imagery, the spatial patterns and accuracy of the GWR-predicted changes are assessed, thereby strengthening the overall validity of the analysis.

5. Results and Discussion

Forest loss by year, using a threshold of tree cover greater than 50%, derived from University of Maryland GLAD data (2001–2020), is shown in Figure 4. The plot indicates a clear decline in forest loss over time. Additionally, the relationship between Nepal’s total population and forest loss during the corresponding years showed a slight negative slope with a very weak correlation (R² = 0.000290), suggesting that population growth had little to no direct impact on forest loss at the national level during this period (Figure 5). However, detailed spatial analysis revealed distinct localized patterns that suggest more complex patterns. While national trends indicate overall stability or improvement, several regions, particularly in the mid-hills and along urban fringes, exhibited concentrated forest degradation. These losses were often associated with infrastructure expansion, agricultural encroachment, and settlement growth and, in some cases, showed relatively strong correlations with population increases. These localized dynamics highlight the importance of sub-national analysis, as aggregated national statistics can obscure critical land cover change processes occurring at finer spatial scales.

Ordinary least-squares linear regression explained almost none of the variance in tree cover loss, with R² climbing only marginally from 0.01 in 2000 to 0.025 in 2005, 0.027 in 2010, 0.028 in 2015, and 0.039 in 2020, demonstrating that LR fails to capture any substantive relationship between population counts and tree cover loss over time.

In contrast, the GWR results demonstrate a consistent increase in R² values over time, revealing a stronger spatial association between tree cover loss and population density. Specifically, the R² value for GWR starts at 0.21 in the base year 2000, rises to 0.30 in 2005, and then to 0.38 in 2010. A notable increase occurs in 2015, with an R² of 0.54, followed by a further rise to 0.59 in 2020. These findings underscore the spatial dependency inherent in the data and highlight the effectiveness of GWR in capturing localized relationships between population and deforestation.

Figure 6a–e illustrate the spatial results of the GWR model at the pixel level for different years. These maps reveal spatial variability in R² values across the country. In 2000 and 2005, the highest local R² values reached 0.64 and 0.77, respectively; in 2010, they peaked at 0.83; and in 2015 and 2020, they stayed at 0.82 and 0.81, respectively.

To better understand the regions where population is most strongly correlated with tree cover loss, a detailed analysis was conducted for 2020. To validate the GWR findings, supplementary visual analysis was performed using high-resolution imagery from Google Earth to identify high-R² areas, which were examined more closely through spatial zoom-ins (Figure 7a–d).

The local R² values for the identified hotspots range from 0.67 to 0.75. Specifically, Figure 7a–d illustrate four major hotspot locations characterized by notable land use/cover dynamics:

Hotspot (7a): Masuriya (80°49′5.72″ E, 28°44′58.32″ N): Located in the Gauriganga Municipality of Kailali District in Sudurpaschim Province, this area shows a local R² of 0.67. The region is primarily characterized by agricultural expansion and settlement growth.

Hotspot (7b): Imiliya (82°59′13.07″ E, 27°39′36.66″ N): As a town in Bhudabhumi Municipality, Kapilbastu District of Lumbini Province, showing a local R² of 0.69, this hotspot is experiencing transitions from agricultural lands to mixed-use development, with visible growth in built-up areas. The region’s proximity to transportation corridors has also contributed to land fragmentation and urbanization.

Hotspot (7c): Pipara Simara (84°59′28.64″ E, 27°9′3.50″ N): As a rapidly expanding urban area in Jitpur Simara Sub-Metropolitan City of Bara District, Madhesh Province, with a local R² of 0.72, this hotspot is undergoing significant urbanization, with a marked increase in impervious surfaces including a decline in open/agricultural lands.

Hotspot (7d): Urlabari (87°33′38.74″ E, 26°39′38.96″ N): As a city located along the Mahendra Highway in Patahrishanishchare Municipality, Morang District, Koshi Province, with a local R² of 0.72, this region has shown consistent growth in built-up areas. The land use transitions here are closely tied to urban migration and the expansion of transport and commercial infrastructure.

These hotspots represent regions where land use/cover dynamics are shifting rapidly. The relatively high local R² values confirm the spatial relevance of the GWR model in capturing these patterns.

In addition to the previously discussed hotspots, the table below presents the top 10 locations with the highest local R² values, sorted in descending order. Interestingly, all of these high-correlation points are located within Madhesh Province, as shown in

Table 2. Top 10 locations in Madhesh Province with the highest local R² values between tree loss and population in 2020.

Latitude	Longitude	District	Municipality	Local R2
27.1559	85.0794	Bara	Jitpur Simara	0.812
27.1738	85.0524	Bara	Jitpur Simara	0.812
27.1469	85.0794	Bara	Jitpur Simara	0.802
27.1738	85.0704	Bara	Jitpur Simara	0.797
27.1559	85.0883	Bara	Kolhabi	0.796
27.1648	85.2680	Rautahat	Gujara	0.796
27.1828	85.2680	Rautahat	Gujara	0.794
27.1738	85.2590	Rautahat	Gujara	0.791
27.1648	85.2770	Rautahat	Gujara	0.790
27.1828	85.2590	Rautahat	Gujara	0.787

.

Further, a closer examination of spatial patterns reveals that the identified GWR hotspots are predominantly concentrated along Nepal’s major transportation corridor—the Mahendra Highway (East–West Highway). This highway stretches from Bhim Dutta in the far west to Mechinagar in the far east, connecting key urban centers and driving development across the Terai region. Many of the high-R² areas in the 2020 GWR analysis are situated near this corridor, suggesting that proximity to major roads is a critical factor in accelerating deforestation. These regions are likely experiencing rapid urbanization, agricultural expansion, and infrastructure development, all facilitated by increased accessibility. The spatial extent of these hotspots and their alignment with the East–West Highway are illustrated in Figure 8. A shapefile containing the coordinates and boundaries of these identified hotspots is available upon request for further research, planning, or policy development.

The observed relationship between population distribution and tree cover loss in Nepal highlights a pressing environmental concern: as population increases, particularly in rapidly developing or accessible areas, there is a corresponding rise in tree cover loss and vegetation degradation. This trend is especially evident in regions with expanding urban footprints, agricultural frontiers, and infrastructure projects. The results reveal that the tree cover loss is not uniform across the country, but is instead strongly influenced by localized human activities and demographic pressures. To effectively capture and analyze this spatially complex relationship, traditional global modeling approaches such as Linear Regression (LR) are insufficient, as they assume a uniform relationship across space and fail to account for local variations. In contrast, Geographically Weighted Regression (GWR) demonstrated that it provides a more nuanced, spatially adaptive framework by allowing regression parameters to vary over space, thereby identifying region-specific patterns and hotspots of tree cover loss associated with population pressure, as captured in this study. The increasing R² values from 2000 to 2020 observed in the GWR analysis also indicate that the influence of population on deforestation has become more spatially structured over time. Furthermore, integrating very high-resolution (VHR) spatial data significantly enhances the precision and reliability of such analyses. By enabling pixel-level assessments, VHR data allows for the detection of fine-scale patterns and localized environmental impacts typically obscured in coarser datasets. This combination of GWR modeling and VHR imagery offers a powerful toolset for identifying deforestation hotspots, validating spatial trends with visual evidence, and supporting targeted environmental management. Notably, many of the areas with high local R² values are situated along the Mahendra Highway corridor, a key axis of development in the Terai region. This spatial alignment suggests that accessibility and infrastructure development are major drivers of forest loss, and that spatially informed planning is essential to balance growth with ecological sustainability.

6. Conclusions

• This study demonstrates that the relationship between population growth and tree cover loss in Nepal is spatially variable and has become increasingly structured over time.
• The correlation between total population and forest loss (2000–2020) at the national level in Nepal suggests little to no direct impact of population growth on forest loss. However, sub-national analysis revealed localized forest degradation, especially in the mid-hills and urban fringes, highlighting the importance of thorough spatial and regional assessments to uncover forest and land cover changes masked by national trends.
• Traditional linear regression models fail to capture these complexities. In contrast, Geographically Weighted Regression (GWR), supported by very high-resolution (VHR) spatial data, provides a more accurate and insightful understanding of localized environmental dynamics.
• Increasing R² values between the tree cover loss and population over the past two decades in some regions of Nepal highlight a growing spatial association between population distribution and tree cover loss, particularly in areas experiencing rapid development and improved accessibility.
• High-R² hotspots identified using GWR along the Mahendra Highway corridor reveal the critical role of infrastructure in shaping land use change.
• These findings not only validate the effectiveness of GWR and VHR data in capturing fine-scale human–environment interactions but also highlight the urgent need for spatially informed decision-making.
• As Nepal continues to develop, integrating spatial analytics into land management and policy frameworks will be essential to mitigate environmental degradation and promote sustainable development across the country.