Prediction of Land Use Land Cover Change Using a Coupled CA-ANN modeling in Dhanusha district of Nepal

doi:10.21203/rs.3.rs-5301403/v1

Land cover refers to the physical cover that is visible on the surface of the earth, whereas land use refers to how individuals use the land. Remote sensing (RS) and Geographical Information Systems (GIS) are proven tools for assessing the LULC change. We used Landsat 5, 8, and 9 satellite images and employed the Maximum Likelihood Supervised Classification algorithm to identify the LULC types and detect changes in the Dhanusha district of Nepal. The change prediction was done using the QGIS 2.18 version MOLUSCE plugin. The four criteria, namely elevation, slope, distance from the road, and built-up were used as spatial variable maps in the learning processes in CA-ANN to predict the LULC of 2033. We assessed five major LULC classes viz. Forest, Water, Cultivated land, Settlement, and barren land. We discovered that the Dhanusha district lost 173.4ha and 300.9ha of its forests, 1381.55ha and 1864.84ha of its cultivated land, and 303.21ha and 452.04ha of its water bodies over 20 years, from 2003 to 2013 and 2013 to 2023 respectively. Significant losses were absorbed by growing urbanized areas and barren land, which expanded by 1157.19ha and 700.95ha in 2003–2013 and by 1674.03ha and 943.8ha in 2013–2023 respectively. By 2033, forest cover is predicted to drop to 22.67%, water bodies to 0.6%, and barren land to 3.16%, with urbanized areas rising to 6.08%. Urban planners are recommended to incorporate nature nature-based solutions for adaptation and mitigation plans for cities that are supported by reliable funding and policy.

Landsat

Supervised classification

QGIS

MOLUSCE

CA-ANN

The land is a natural resource that humans have utilized for life and various activities (Nedd et al. 2021). Land cover refers to the current biophysical state of the Earth’s surface and immediate subsurface (Srivastava et al. 2020; Wang et al. 2021). On the other hand, land use refers to the amount of human activity that is directly tied to the land and the utilization of its resources (Boakye et al. 2018). Land use and land cover are considered central variables to understanding the physical environment and its interactions with anthropogenic activity (Fonte et al. 2024). Land use and land cover changes are ever-evolving, requiring consistent observation to fully understand the rapid developments in changing areas (Mohabey et al. 2015). The changes in LULC are derived from the interaction among socioeconomic, environmental, and institutional phenomena occurring on the land (Guevara Ochoa et al. 2020). Rapid population growth, migration, and the transformation of rural areas into urban centers are key drivers of land use and land cover change (LULCC). This alteration significantly affects interactions between the land and atmosphere, including climate dynamics (Nedd et al. 2021). In Nepal, Ecological and environmental impacts are the main causes of land-use change (Bist et al. 2021).

The dynamic nature of land use/land cover (LULC) changes requires continuous assessment, analysis, and monitoring employing socioeconomic and geospatial data for integrated and more accurate results. LULC changes have multidimensional effects on the present and future ecosystem balance (Yifru et al. 2021). The dynamic nature of LULC changes has a significant effect on spatiotemporal environmental stability due to its connection to local, regional, and international climate conditions, clean water supply, agricultural activities, stability of biodiversity, and food security (Abebe et al. 2022). Land Use Land Cover (LULC) maps are used as input for climate change monitoring and forecast (Moiceanu et al. 2021), population mapping, urban planning, and institutional guideline documents, among others (Naeem et al. 2018; Kidane et al. 2019).

Recent advancements in remote sensing (RS) and Geographical Information Systems (GIS) have proven to be effective tools for studying LULC dynamics by acquiring, analyzing, and quantifying data rapidly and regularly at lower cost and time than traditional ground survey methods (Mishra and Rai 2016). The suitability of RS and GIS in LULC dynamics has led to the evolution of several geospatial models to simulate and predict future spatial-temporal patterns and act as decision-support tools (Wang & Maduako 2018). The combined remote sensing (RS) and geographical information system (GIS) has the best tool for managing land-use change research and natural resources. Analyzing and tracking regional and temporal LULC shifts benefits scientists, environmentalists, agriculturalists, legislators, and urban planners (Guidingan et al. 2019). Remote sensing techniques assist in efficiently preparing natural resources, tracking land management, and long-term change dynamics (Bhattacharya et al. 2020).

In this study, the MOLUSCE (Modules for Land Use Change Simulation and Evaluation) plugin for QGIS was used to assess future land use and land cover (LULC) conditions, as well as spatial and temporal transitions (Kamaraj and Rangarajan 2022). MOLUSCE is a user-friendly plugin designed for QGIS versions 2.00 to 2.99, developed by Asia Air Survey (AAS), the plugin enables efficient analysis and simulation of land cover changes (Kulithalai Shiyam Sundar and Deka 2022). It incorporates several algorithms that support various techniques, tailored to meet the specific objectives of the study (Leta et al. 2021). A CA-ANN model in MOLUSCE is a reliable tool for predicting future LULC that may be utilized in land use planning and management. MOLUSCE is also used to investigate temporal LULC shifts and predict future land use, by anticipating prospective land cover and forest cover shifts, and detecting deforestation in sensitive locations (Aneesha Satya et al. 2020). This model predicts the spatial LULC shift by estimating the pixel’s current condition based on its initial situation, adjacent neighbourhood eventuality, and changeover laws. Moreover, this accurately depicts nonlinear spatial stochastic LULC change processes and produce complex patterns (Saputra and Lee 2019).

This study was carried out to detect and quantify the regional and temporal change in LULC changes from the years 2003, 2013, and 2023 using Landsat data series and to predict for the year 2033 using the MOLUSCE plugin of QGIS which is based on the CA-ANN model.

Study Area

The study area was conducted in the Dhanusha district of Madhesh Pradesh. It lies in between latitude 26°50′31.56″ North, longitude 86°02′09.60″ East. Its elevation ranges from lower tropics, below 300 masl (occupies 92.5% of area) to upper tropical 1000 masl (occupies 7.5% of area) (Barnekow Lillesø et al. 2005). Dhanusha district covers an area of 1,180 km2 and has a population of 867,747 (Central Bureau of Statistics 2021). It is situated in the Outer Terai and its district headquarters is Janakpurdham.

The annual average temperature in the Dhanusha is around 29.68◦C and it is 7.68% higher than Nepal's averages. It typically receives about 142.16 millimeters (5.6 inches) of precipitation and has 136.86 rainy days (37.5% of the time) annually. The fertile plains support the cultivation of crops such as rice, wheat, maize, sugarcane, and various fruits and vegetables. The main vegetation types of the district are sal forest, riverine vegetation, bamboo groves, deciduous trees, orchards, and grassland. Dhanusha, like many other Terai districts, faces challenges related to agriculture, including issues with irrigation, land fragmentation, and vulnerability to floods. Socioeconomic development and infrastructure improvement are ongoing concerns in the district.

Data Acquisition

The Landsat data was downloaded from the site(https://earthexplorer.usgs.gov/). We collected three Landsat images of the years 2003, 2013, and 2023 in this study. Landsat Thematic Mapper was used for the year 2003 and Landsat Operational Land Imager/Thermal Infrared Sensor was used for the year 2013 and 2023. All of the downloaded Landsat images had less than 10% cloud cover. The Landsat images were available at 30m resolution in USGS Earth Explorer.

Table 1

Characteristics of acquired Landsat satellite images
S.N.	Satellite	Sensor Identifier	Image Date	Path/Row	Date of Data Acquisition
1.	Landsat 5	TM	2003/12/18	140/041, 140/042	2023/10/22
2.	Landsat 8	OLI and TIRS	2013/12/13	140/041, 140/042	2023/10/22
2.	Landsat 9	OLI and TIRS	2023/05/14	141/042, 140/042, 140/041	2023/10/22

Training Samples

Data for training and validation of land cover classification was collected from field data using GPS and from Google Earth Pro. About 30–50 training samples per land cover class were collected.

Tabel 2 Bands combination for different LULC Class Analysis

S.N.	CLASSES	BANDS			No. of training sample
S.N.	CLASSES	Landsat 5	Landsat 8 & 9	2003	2013	2023
1.	Forest	5,4,3	6,5,4	100	100	130
2.	Settlement	7,5,3	7,6,4	100	100	130
3.	Cultivated land	7,5,3	6,5,2	100	100	130
4.	Barren land	3,2,1	4,3,2	100	100	130
5.	Water	3,2,1	4,3,1	100	100	130

Classification of LULC

Maximum Likelihood Supervised Classification was used to identify the land use types in the study area. The maximum likelihood algorithm is one of the most popular methods of classification in remote sensing, in which a pixel with the maximum likelihood is classified into the corresponding class (Japan Association of Remote Sensing 1996). LULC classes were classified based on Food and Agriculture Organization guidelines, 2010(Table 3). Land Cover and Land Use classes Forest land, cultivated land, Water, Settlement and Barren land were considered 5 classes in our study. They are described in Table 3:

Table 3

Definitions of classified LULC classes
Forest	All land with woody vegetation that satisfies the criteria for defining forest land falls under this category.
Cultivated Land	Lands covered with temporary broadleaf or grass-type crops that are harvested at the completion of the growing season then remain idle until replanted.
Water	All areas within the landmass periodically water covered falls under this category.
Settlement	Land covered by buildings and other man-made structures and activities.
Barren-land	Lands exposed soil, sand, or rocks and has less than 10% vegetated cover during any time of the year.

Platform for Data Analysis

The processing and analysis of Landsat images and all the other geo-spatial analyses were done using ArcGIS and QGIS, also all the numerical data were analyzed using MS Excel. MS Excel was used to interpret the numerical results into bar graphs, charts, and tables.

Change detection and analysis

The classified images were prepared in ArcGIS 10.8. of three different dates were processed through a raster calculator to produce LULC change maps in land use land cover between 2003. 2013, and 2023. The numerical analysis of land use gains, losses, and net changes was then done through Microsoft Excel.

Accuracy Assessment

Accuracy assessment is an essential and crucial part of studying image classification and thus LULCC detection in order to understand and estimate the changes accurately (Owojori and Xie 2005). The classification accuracy was expressed as the overall accuracy, kappa coefficient, the user’s and producer’s accuracy. All these were calculated and analyzed for each land cover type. In this study, accuracy assessment was done for the LULC 2003, LULC 2013, and LULC 2023 year.

Post-processing

Accuracy assessment came out to be good enough for land use maps so it was exported to the GIS platform. After it was exported, it was subjected to more manipulation to obtain smooth images. The transition map was prepared over a time period of 2003, 2013, and 2023 with the help of Arc GIS 10.8. Finally, a layout map was made and extracted to obtain the Land Use map of Dhanusha District.

Prediction Analysis

This paper uses the MOLUSCE (Modules of Land Use Change Evaluation), a QGIS plugin that can estimate potential LULC changes. Various algorithms are embedded in the MOLUSCE plugin, which is essential for a variety of techniques as necessitated in the respective objective of the study (Leta et al. 2021). The MOLUSCE plugin in QGIS offers several well-known techniques, including Pearson’s correlation and Cramer’s coefficient, for assessing the connection between LULC data and geographic factors (Hakim et al. 2021). This study includes five basic steps for Modeling using MOLUSCE. Five primary steps make up the graphical user interface (GUI) explain the MOLUSCE plugin: Inputs, Evaluation Correlation, Area Changes, Transition Potential Modeling, Cellular Automata Simulation, and Simulation. For each simulation analysis LULC layers two different years were used as independent parameters and slope gradient, DEM (Digital Elevation Model), distance from urban, and distance from the road were used as dependent parameters.

In this modeling section Cellular Automata – Artificial Neural Network (CA-ANN) model, was employed to predict future LULC scenarios of the study area. The CA-ANN is a hybrid modelling approach which accurately defines nonlinear spatial stochastic LULC change by estimating pixels present values based on the initial values and their neighborhood pixels (Abbas 2021; Saputra and Lee 2019).

Fig. 3, depicts several geographical variables included in the model, such as DEM, Slope, Euclidean distance to road, and Euclidean distance to existing settlements. This model utilized these data as inputs to produce a simulation/prediction map of land use and land cover in 2033, allowing the identification of the study area’s shifting pattern between 2003, 2013, and 2023.

Results from the classification of the satellite images in 2003, 2013, and 2023 are presented in the following figure. We classified 5 land use class types as Forest, Water, Cultivated land, Settlement, and Barren land. Areas of land under different land use categories as derived from the classification results are also given in Table.

Land Use and Land Cover Classes for 2003

The result of 2003 image showed that Crop land and Forest dominated the landscape of the Dhanusha District. The Cropland was found to be confined at the lower elevations with (82,318ha) 69.3% of total land area and a majority of Forest land was situated at a higher elevation around the Chure region with an area (27,799ha) 23.4%. The Barren land was present with (3,807ha) 3.2% of the total area. And water includes the area with (3,199 ha) 2.7%. The settlement areas were found to be comparatively less than that of forest and cropland represented by (1679ha) 1.4%. The distribution patterns of different LULC classes throughout the district are seen as in the (Fig. 4) below:

Land Use and Land Cover Classes for 2013

In 2013, the land cover distribution exhibited a significant predominance of cultivated land, which occupied 80,936.5 hectares, accounting for 68.13% of the total area. Forested areas were the second largest land cover type, encompassing 27,625.8 hectares, which equated to 23.25% of the total area. There was only a slight change in water bodies and barren land over the decade. Water bodies were spread across 2,896.2 hectares, making up 2.44% of the area which, while barren land covered 4,507.68 hectares, constituting 3.79%. Built-up areas expanded significantly over the decade. By 2013, this had increased to 2,836.5 hectares, representing 2.39% of the total area. This growth of 1,831.1 hectares (1.54%) reflects increased urbanization and infrastructure development.

Land Use and Land Cover Classes for 2023

In 2023, the land use land cover (LULC) analysis continued to show dynamic changes in the landscape. The area under cultivated land further decreased to 75,642.8 hectares, constituting 63.68% of the total area compared to 2003 and 2013 statistics. The Settlement area increased to (4,510.53ha) 3.80%. There is also a slight increase in Barren land with an area of (5451.48ha) 4.59%. The rest of the land use area (Forest, cultivated land, and Water) decreased. Although the area of forest and cultivated land decreased compared to the image of 2003 and 2013, forestland and cultivated land both were seen as dominating land use. The water decreased and had an area of (2,444.13ha) 2.06%. Cultivated land includes (79,071.33ha) 66.56% of the total area and forestland also decreased and had (27,324ha) 23% of the area.

Land Use and Land Cover Class Change between (2003–2023)

The result of the Land Use Land Cover Change Trend between 2003, 2013, and 2023 shows that all land use coverage decreased except for Settlement and Barren land. Cultivated land decreased from 82,318.05 ha in 2003 to 80,936.5 ha in 2013 and to 79,071.66 ha in 2023, indicating a reduction of 3246.39ha. Forest cover decreased from 27,799.2 ha in 2003 to 27,625.8 ha in 2013 and to 27,324.9 ha in 2023, reflecting a reduction of 474.3 ha. Built-up areas saw substantial growth from 1,679.31 ha in 2003 to 2,836.5 ha in 2013 and to 4,510.53 ha in 2023, reflecting an increase of 2831.22 ha. The water decreased by 0.64% i.e. 755.28ha within the given time frame whereas Barren land has seen to be increased from the year of 2003 and 2023 by 1.38% i.e. 1644.75ha.

Table 4

Composite table of area statistics (Hectare) of the Dhanusa district in 2003, 2013, and 2023 and their net change
LULC Classes	LULC 2003	LULC 2013	LULC 2023	Area Change (2003–2013)	Area Change (2013–2023)
LULC Classes	Area (Hectare)	Area (Hectare)	Area (Hectare)	Area (Hectare)	Area (Hectare)
Forest	27799.2	27625.8	27324.9	-173.4	-300.9
Water	3199.4	2896.2	2444.1	-303.21	-452.07
Barren land	3806.7	4507.7	5451.5	700.95	943.8
Settlement	1679.3	2836.5	4510.5	1157.19	1674.03
Cultivated land	82318.1	80936.5	79071.7	-1381.55	-1864.84

Land Use and Land Cover Change Transition Matrix between 2003–2013 and 2013–2023

The LULC transition matrix for 2003–2013 and 2013–2023 depicted the transition from one class to another. In 2003–2013, the highest transition from barren land to cultivated land was 1736.2ha, followed by the conversion of forest to cultivated land by 1726.5ha, while in 2013–2023, the highest transition from forest to cultivated land (4737.6 ha) and the second highest transition from cultivated land to barren land by 3023.8ha occurred. There was also a large change of cultivated area to settlement during both transition years, with 1132.4ha and 3005ha in 2003–2013 and 2013–2023, respectively. It was mostly due to urbanization. Water and barren land occupy a small area in comparison to other land use classes. So, no significant conversion was seen between them. The area statistics of the transition map below show the transition from one land use to another.

Table 5

Statistical table of transition map from 2003–2013
LULC Classes		LULC 2013
LULC Classes		Barren land	Settlement	Cultivated land	Forest	Water
LULC 2003	Barren land	1761.83	48.25	1736.22	79.79	149.32
	Settlement	93.59	235.18	984.07	50.86	62.66
	Cultivated land	491.53	1132.44	77050.80	4072.85	228.53
	Forest	5.56	9.43	1726.45	26021.05	11.96
	Water	152.59	38.32	1934.75	224.91	450.66

Table 6

Statistical table of transition map from 2013–2023
LULC Classes		LULC 2023
LULC Classes		Barren land	Settlement	Cultivated land	Forest	Water
LULC 2013	Barren land	1499.89	11.03	883.88	22.45	87.68
	Settlement	65.16	1130.85	224.80	18.14	24.74
	Cultivated Land	3023.81	3005.02	73703.41	2020.46	1679.86
	Forest	313.91	46.19	4737.62	25248.74	102.22
	Water	169.21	26.61	278.97	26.35	402.60

Accuracy assessment

For the accuracy assessment, 100 of the sample points were taken for the validation. The overall accuracy obtained for the year 2003 is 82%, for the year 2013 is 87%, and for the year 2023 is 87%. The kappa value of the classified image of year 2003 obtained is 0.78, kappa value for 2013 is 0.81, and kappa value for 2023 is 0.84. Kappa values between 0.6 to 0.80 is substantial and 0.81 to 1 is almost perfect, our obtained kappa values are substantial for year 2000 and almost perfect for 2010 and 2020. Kappa values of < 0 reflect no agreement, 0–0.2 as slight, 0.2–0.41 as fair, 0.41–0.60 as moderate, 0.60–0.80 as substantial, and 0.81–1.0 as almost perfect agreement (Maingi et al. 2002; Manonmain and Suganya 2010). These estimates indicate that classification accuracies were of substantial and perfect agreement.

Table 7

Accuracy assessment of LULC map 2003
Classes	Forest	Water	Barren land	Settlement	Cultivated land	Total	User Accuracy (%)
Forest	20	0	0	0	0	20	100
Water	1	15	0	0	4	20	75
Barren land	1	1	15	0	3	20	75
Settlement	0	0	2	13	5	20	65
Cultivated land	1	0	0	0	19	20	95
Total	23	16	17	13	31	100
Producer Accuracy (%)	86.96	93.75	88.24	100.00	61.29	Overall Accuracy 82%
Kappa Coefficient: 0.775

Table 8

Accuracy assessment of LULC map 2013
Classes	Water	Barren land	Settlement	Forest	Cultivated land	Total	User Accuracy (%)
Water	17	3	0	0	0	20	85
Barrenland	0	18	0	0	2	20	90
Settlement	0	2	17	1	0	20	85
Forest	0	0	1	19	0	20	95
Cropland	0	2	1	1	16	20	80
Total	17	25	19	21	18	100
Producer Accuracy (%)	100	72	89.5	90.5	88.9	Overall Accuracy: 87%
Kappa Coefficient: 0.813

Table 9

Accuracy assessment of LULC map 2023
Classes	Forest	Settlement	Cultivated land	Barren land	Water	Total	User Accuracy (%)
Forest	20	0	0	0	0	20	100
Settlement	0	18	2	0	0	20	90
Cultivated land	0	1	18	1	0	20	90
Barren land	1	0	7	12	0	20	60
Water	0	0	1	0	19	20	95
Total	21	19	28	13	19	100
Producer Accuracy (%)	95.2	94.7	64.3	92.3	100.0	Overall Accuracy 87%
Kappa Coefficient: 0.838

Spatial Variables

The spatial variables selected for the current study are DEM, slope, distance to roads, and the distance to existing settlements. After relevant spatial variables (Fig. 3) were selected, geometry matching was conducted, which included the cell size of the raster data, NoData value, and coordinate reference system. The cell size, No Data value, and coordinate reference system are 30 m, 0, and WGS_1984_UTM_Zone_45 N, respectively, for all spatial variables.

In this study, Pearson’s correlation was used to measure the correlation between the variables which is one of the three correlation measurements that is available in the MOLUSCE plugin of QGIS.

Table 10

Evaluation of the correlation between variables using Pearson’s correlation method.
Spatial Variables	DEM	Slope	Distance to Road	Distance to settlements
DEM	- -	0.5138	0.7453	0.5466
Slope		- -	0.3431	0.305
Distance to Road			- -	0.4239
Distance to settlements				- -

LULC Change Transition Potential Modeling based on CA-ANN

The MOLUSCE approach was used to model transition potentials using an artificial neural network (ANN). The ANN employs a multilayer perception method with neighborhood (1 px), learning rate (0.1), maximum iteration (1000), hidden layer (10), momentum (0.05), fixed overall accuracy (-0.00341), minimum error for validation (0.03498), and validation kappa (0.965), as shown in Figure 10. The model employed 5000 randomly distributed samples to generate a spatial representation for the ANN.

The 2023 LULC was modeled using the initial and final rasters from the 2003 and 2013 LULC maps, respectively. The simulation model was constructed and evaluated against the real 2023 LULC. That yields a validation kappa of 0.965, an overall accuracy of 96.43%, and an overall kappa value of 0.927. (Fig. 11) depicts the real and simulated LULC maps for 2023, with their respective area coverages reported in (Table 11) and (Fig. 12) depicts the validation graph between actual and simulated LULC 2023.

Table 11

Actual and Simulated LULC of 2023
LULC Classes	Actual LULC		Simulated LULC		Area change (Ha)
LULC Classes	Area (Ha)	Area (%)	Area (Ha)	Area (%)	Area change (Ha)
Forest	27324.90 ha	23.00	26144.46 ha	22.01	-1180.44 ha
Water	2444.13 ha	2.06	2465.55 ha	2.08	21.42 ha
Barren land	5451.48 ha	4.59	4898.79 ha	4.12	-552.69 ha
Settlement	4510.53 ha	3.80	4423.86 ha	3.72	-86.67 ha
Cultivated land	79071.66 ha	66.56	80870.04 ha	68.07	1798.38 ha

LULC Prediction

The LULC prediction for 2033 was carried out after receiving acceptable model validation results, with an overall kappa value of 0.927 being regarded as good accuracy by many researchers (Alam et. al., 2021; Aneesha Satya et. al., 2020; Perović et. al., 2018; Rahman et. al., 2017). Model validation combines the contents of two data sources while taking into consideration the features of actual and simulated LULC data (Lukas et. al., 2023). The LULC for 2033 was predicted using the LULC of 2013 and 2023 and the spatial variables were chosen for simulation. A kappa value of 0.75 was obtained. The predicted LULC map of 2033 along with its area can be seen in (Fig. 13) and (Table 12) respectively.

Table 12

Statistical table of Predicted LULC map of 2033
LULC Classes	Area (Ha)	Area (%)
Forest	26941.54	22.67
Water	719.24	0.60
Barren land	3763.73	3.16
Settlement	7227.63	6.08
Cultivated land	80150.56	67.46

The results of the Land Use Land Cover Change (LULCC) trend analysis between 2003, 2013, and 2023 indicate a decrease in all land cover types coverages except for Settlement and Barren land. Notably, there was a substantial increase in settlement area (Table 4). This rapid expansion of built-up areas aligns with previous reports by the United Nations Department of Economic and Social Affairs (UNDESA 2014) and Paudel et al. (2016), which highlighted the unchecked and unplanned urbanization occurring in Nepal.

In Dhanusa, land use patterns have undergone significant shifts, with forests being cleared for agriculture, often leading to subsequent abandonment and resulting in barren land (Table 6). Rawat et al. (2011) reported a significant increase in barren land, primarily due to human activities and climate change, resulting in a 56% increase in non-forest areas. The rural communities are increasingly turning to non-agricultural activities, leading to neglect of forest management (Bardsley et al. 2014). These findings collectively emphasize the urgent need for sustainable land management practices to mitigate the negative effects of such transformations on water resources.

Our study shows a continuous decrease in water sources (Table 4). This trend is consistent with research findings by Tanksali and Soraganvi (2021), who observed a reduction in water yield areas due to land use changes, leading to a noticeable decrease in surface runoff and water availability in semi-arid regions. Additionally, Sumitra et al. (2018) highlighted that land use changes, particularly the conversion of forests to barren land, have a substantial impact on water discharge patterns. This emphasizes the influence of land use and climate change on hydrological processes.

This study also concluded the shrinking of agricultural and forest lands (Fig. 6), which is primarily caused by rapid urbanization and a lack of robust land use regulations (Paudel et al. 2016). Unstable agricultural production due to natural disasters and limited resources further contributes to the conversion of farmland to barren land (Bhusal et al. 2022). Kayet and Patak (2015) analyzed land use changes in India from 1992 to 2014, finding that the expansion of built-up areas and wasteland, along with the decline of agricultural land, is reducing the dense forest cover of the region. A study conducted in Linyi, China, using the QGIS MOLUSCE plugin and remote sensing big data, revealed an increase in the population number and density, leading to the expansion of urban areas and a significant decrease in forest resources and grasslands (Muhammad et al. 2022). Pandey et al. (2016) identified encroachment, illegal harvesting, infrastructural development, population growth, forest grazing, and forest fires as the main drivers of LULCC change in the Tehrathum district, while six drivers were identified for the Dhanusha district.

Our results indicate that areas with lower elevations are associated with rapid LULCC changes due to their suitability for anthropogenic activities. The highest changes occurred in the plain areas of the Dhanusha district, such as Jankapur, Dhanushadham, and Dhalkebar (Fig. 8, 9).

This research predicts that both forest and water bodies will decrease in Dhanusha by 2033, while cultivated land and settlements are expected to increase (Table 12). A study in Birendranagar Municipality reported a loss of 1146 ha of forest and 280.53 ha of water bodies from 2000 to 2020, alongside gains in settlement and agricultural land (Sapkota and Dahal, 2024). Similarly, in the Chure region, significant decreases in barren land and water areas were noted, while agricultural and built-up areas increased (Kafle et al. 2023). In the Ritung Khola sub-watershed, agricultural land and human settlements surged by 313.54% and 367.14%, respectively, with corresponding reductions in forest cover and water bodies (Timilsina et al. 2023). A study conducted in the Garhwal Himalayan region of India found that agriculture and built-up areas increased at rates of 1.23% and 3.75%, respectively, while water and forest decreased at a rate of -0.98% and − 0.64%, respectively (Batar et al. 2017). These findings collectively suggest a trend of land use transformation driven by urbanization and agricultural expansion, highlighting the urgent need for sustainable land management practices to mitigate these changes (Kafle et al. 2023; Timilsina et al. 2023).

Our study used the CA-ANN model for the prediction of future LULCC changes and found it to be effective and reliable. However, it is suggested that future studies incorporate more spatial and socioeconomic variables such as climate variability, economic development, technological advancement, political economy, and GDP (gross domestic product) to provide more comprehensive and valuable information regarding the positive and negative impacts of LULCC changes in the future.

The research conducted on detecting and predicting LULC change in the respective study area has revealed significant transformations over the past two decades. The study utilized RS and GIS to analyze LULC changes from 2003 to 2023, highlighting the dynamic nature of land use in the region. The CA ANN model has demonstrated significant effectiveness in simulating LULC changes, particularly in the context of the Dhanusha district, Nepal. The CA-ANN model, integrated with the MOLUSCE plugin, has provided reliable predictions of future LULC scenarios, specifically for the year 2033. The model's validation showed a high kappa value of 0.927, indicating good accuracy in simulating LULC transitions based on historical data from 2003, 2013, and 2023. We discovered that the Dhanusha district lost 173.4ha and 300.9ha of its forests, 1381.55ha and 1864.84ha of its cultivated land, and 303.21ha and 452.04ha of its water bodies over the course of 20 years, from 2003 to 2013 and 2013 to 2023. This decline in natural resources is alarming and indicates a pressing need for sustainable management practices. According to the research, urbanized areas increased by 1157.19ha between 2003 and 2013 and by 1674.06ha between 2013 and 2023, while bare land increased by 700.95ha between 2003 and 2013 and by 943.8ha between 2013 and 2023.

This urban expansion is primarily driven by rapid urbanization and rural-urban migration, which poses challenges for land management and environmental sustainability. Predictions for 2033 suggest further declines in forest, barren land, and water bodies, with urbanized areas expected to increase by 6.08%. These projections underscore the urgency for effective urban planning and resource management strategies to mitigate adverse impacts on the environment and local communities.

In conclusion, the study emphasizes the critical need for continuous monitoring of LULC changes to inform sustainable land management practices. The integration of RS, GIS & Machine learning techniques has proven effective in this analysis, providing valuable insights for policymakers and urban planners in addressing the challenges posed by rapid land use changes in the Dhanusha district. Prediction of LULC is critically required for the creation of plans capable of striking a balance between attempts to preserve resources, conflicting user requirements, and development pressures. Future studies should consider integrating more spatial and socioeconomic variables, such as climate change impacts, economic development trends, and technological advancements. This would provide a more comprehensive understanding of the factors driving LULC changes and improve prediction accuracy. Implementing comparative studies across different regions with varying socio-economic contexts could yield insights into the adaptability and limitations of the CA ANN model in diverse settings. This would enhance the model's generalizability and applicability in various geographical contexts.

The results obtained from current and future land use and land cover changes in the Dhanusha could be taken as inputs for policymakers to revise land use policies. We urge that the Dhanusha district employ an ecosystem-based approach to addressing socioeconomic and ecological challenges in the city and that such techniques be incorporated into both short and long-term development plans.

Funding Statement: This research did not receive a specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Ethical Statement: This research was conducted ethically, adhering to standards of honesty, integrity, and respect for all participants, animals, and the environment. All sources are accurately cited, and any potential conflicts of interest are disclosed.

Consent Statement: The publishing of this research article has the complete approval of all its authors. We all concur that the material, conclusions, and submission for publication are sound.

Author Contribution

O.M. and B.K. performed data collection. O.M., N.K., and D.J. conceptualized the article and critically revised the work. O.M., S.G., performed the data curation. O.M., N.K., and S.G. carried out data analysis & software operation and prepared the original manuscript. O.M., D.J., N.K., and B.K. did the review and editing.

Acknowledgement

We would like to sincerely thank everyone who has generously shared their knowledge and provided guidance, as they have all been helpful in shaping this research. We also like to thank the Faculty of Forestry, Agriculture, and Forestry University for providing an educational setting that has helped this research come to fulfillment. We would especially like to express our gratitude to Rajeev Kumar Jha, our field advisor, and the Ministry of Forest and Environment, Madhesh Province, for their unwavering support and direction during the fieldwork phase.

Abbas Z, Yan, G., Zhong Y, Zhao Y (2021) Spatiotemporal change analysis and future scenario of LULC using the CA-ANN approach: A case study of the greater bay area, china. Land, 10(6), 584. https://doi.org/10.3390/land10060584
Abebe G, Getachew D, Ewunetu A (2022) Analysing land use/land cover changes and its dynamics using remote sensing and GIS in Gubalafito district, Northeastern Ethiopia. SN Applied Sciences, 4(1), 30. https://doi.org/10.1007/s42452-021-04915-8
Aneesha Satya B, Shashi M, Deva P (2020) Future land use land cover scenario simulation using open source GIS for the city of Warangal, Telangana, India. Applied Geomatics, 12(3), 281-290. https://doi.org/10.1007/s12518-020-00298-4
Bardsley DK, Cedamon E, Paudel N, Nuberg I (2024) Forest livelihood values and management opportunities to support the agrarian transition in Nepal. Journal of Rural Studies, 107, 103253. https://doi.org/10.1016/j.jrurstud.2024.103253
Barnekow Lillesø J-P, Shrestha TB, Dhakal LP, Nayaju RP, Shrestha R (2005) The map of potential vegetation of Nepal: a forestry/agro ecological/biodiversity classification system. Center for Skov, Landskab og Planlægning/Københavns Universitet. Development and Environment No. 2/2005
Barros JX (2004) Urban Growth in Latin American Cities-Exploring urban dynamics through agent-based simulation. University of London, University College London (United Kingdom).
Batar AK, Watanabe T, Kumar A (2017) Assessment of land-use/land-cover change and forest fragmentation in the Garhwal Himalayan region of India. Environments - MDPI, 4(2), 1–16. https://doi.org/10.3390/environments4020034
Bhattacharya RK, Das Chatterjee N, Das K (2021) Land use and land cover change and its resultant erosion susceptible level: An appraisal using RUSLE and Logistic Regression in a tropical plateau basin of West Bengal, India. Environment, Development and Sustainability, 23(2), 1411-1446.https://doi.org/ 10.1007/s10668-020-00628-x
Bhusal S, Raju R, Nithyashree ML, Singh DR (2022) Agricultural scenario of Nepal: Performance and SWOT analysis. Agricultural Economics Research Review, 35(conf), 180-180.
Bist SP, Adhikari R, Regmi RR, Subedi R (2021) Land use/land cover change assessment of Mohana watershed (Far-Western Nepal) using GIS and remote sensing. Archives of Agriculture and Environmental Science, 6(3), 320-328. https://doi.org/10.26832/24566632.2021.060309
Boakye E, Anornu GK, Quaye-Ballard JA, Donkor EA (2018) Land use change and sediment yield studies in Ghana. Journal of Geography and Regional Planning, 11(9), 122-133. https://doi.org/10.5897/JGRP2018.0707
Central Bureau of Statistics (2018) Dhanuha District Profile. https://cbs.gov.np/district-profile/CoverChanges.
Dos Santos V, Laurent F, Abe C, Messner F (2018) Hydrologic response to land use change in a large basin in eastern Amazon. Water, 10(4), 429.
FAO (2010) Global forest resource assessment, country report Nepal. Food and Agriculture Organization of the United Nations, FRA2010/144 Rome, 2010.
Fonte CC, Duarte D, Jesus I, Costa H, Benevides P, Moreira F, Caetano M (2024) Accuracy Assessment and Comparison of National, European and Global Land Use Land Cover Maps at the National Scale—Case Study: Portugal. Remote Sensing, 16(9), 1504. https://doi.org/10.3390/rs16091504
Guevara Ochoa C, Medina Sierra A, Vives L, Zimmermann E, Bailey R (2020) Spatio‐temporal patterns of the interaction between groundwater and surface water in plains. Hydrological Processes, 34(6), 1371-1392. https://doi.org/10.1002/hyp.13615
Guidigan MLG, Sanou CL, Ragatoa DS, Fafa CO, Mishra VN (2019) Assessing land use/land cover dynamic and its impact in Benin Republic using land change model and CCI-LC products. Earth Systems and Environment, 3(1), 127-137. http://dx.doi.org/10.1007/s41748-018-0083-5
Hakim AMY, Baja S, Rampisela D, Arif S (2021) Modelling land use/land cover changes prediction using multi-layer perceptron neural network (MLPNN): a case study in Makassar City, Indonesia. International Journal of Environmental Studies, 78(2), 301-318. http://dx.doi.org/10.1080/00207233.2020.1804730
Japan Association on Remote Sensing (1996) Remote Sensing Note. Tokyo, Japan: JARS
Kafle S, Sandeep KC, Poudyal B, Devkota S (2023) Machine learning approach to detect land use land cover (LULC) change in Chure region of Sarlahi district, Nepal. Archives of Agriculture and Environmental Science, 8(2), 168-174. https://doi.org/10.26832/24566632.2023.0802012
Kamaraj M, Rangarajan S (2022) Predicting the future land use and land cover changes for Bhavani basin, Tamil Nadu, India, using QGIS MOLUSCE plugin. Environmental Science and Pollution Research, 29(57), 86337-86348. http://dx.doi.org/10.21203/rs.3.rs-616393/v1
Kayet N, Pathak K (2015) Remote sensing and GIS based land use/land cover change detection mapping in Saranda forest, Jharkhand, India. International research journal of earth sciences, 3(10),1-6.
Khawaldah HA, Farhan I, Alzboun NM (2020) Simulation and prediction of land use and land cover change using GIS, remote sensing and CA-Markov model. Global Journal of Environmental Science and Management, 6(2), 215-232. https://doi.org/10.22034/gjesm.2020.02.07
Kidane M, Bezie A, Kesete N, Tolessa T (2019) The impact of land use and land cover (LULC) dynamics on soil erosion and sediment yield in Ethiopia. Heliyon, 5(12). https://doi.org/10.1016/j.heliyon.2019.e02981
Kulithalai Shiyam Sundar P, Deka PC (2022) Spatio-temporal classification and prediction of land use and land cover change for the Vembanad Lake system, Kerala: a machine learning approach. Environmental Science and Pollution Research, 29(57), 86220-86236. http://dx.doi.org/10.21203/rs.3.rs-581788/v1
Kumar S (2018) Remote sensing and GIS-based groundwater prospects and quality assessment in Fatehabad district, Haryana”, i-Manager’s Journal on Future Engineering and Technology, Vol. 14(1), pp.48
Kumar S, Singh R (2021) Geospatial Applications in Land Use/Land Cover Change Detection for Sustainable Regional Development: The Case of Central Haryana, India.
Leta MK, Demissie TA, Tränckner J (2021) Modeling and prediction of land use land cover change dynamics based on land change modeler (Lcm) in nashe watershed, upper blue nile basin, Ethiopia. Sustainability, 13(7), 3740. https://doi.org/10.3390/su13073740
Lv ZQ, Zhou QG (2011) Utility of Landsat image in the study of land cover and land surface temperature change. Procedia Environmental Sciences, 10(PART B). 1287–1292. https://doi.org/10.1016/j.proenv.2011.09.206
Maingi JK, Kepner SE, Edmonds WG (2002) Accuracy Assessment of 1992 Landsat-MSS Derived Land Cover for the Upper San Pedro Watershed(US/Mexico). Sponsored by Environmental Protection Agency, Las Vegas, NV. National Exposure Research Lab, 2002.
Manonmani R, Suganya G (2010) Remote sensing and GIS application in change detection study in urban zone using multi temporal satellite. International journal of Geomatics and Geosciences, 1(1), 60-65.
Mishra VN, Rai PK (2016) A remote sensing aided multi-layer perceptron-Markov chain analysis for land use and land cover change prediction in Patna district (Bihar), India. Arabian Journal of Geosciences, 9, 1-18. http://dx.doi.org/10.1007/s12517-015-2138-3
Mohabey DP, Sharma NK, Kumar A (2015) Change Detection in Lulc of Sahibganj District , Jharkhand , India Using Remote Sensing & Gis. International Journal of Advancement in Remote Sensing, GIS and Geography, 3(2), 59–70.
Moiceanu G, Dinca MN (2021) Climate change-greenhouse gas emissions analysis and forecast in Romania. Sustainability, 13(21), 12186. https://doi.org/10.3390/su132112186
Muhammad R, Zhang W, Abbas Z, Guo F, Gwiazdzinski L (2022) Spatiotemporal change analysis and prediction of future land use and land cover changes using QGIS MOLUSCE plugin and remote sensing big data: a case study of Linyi, China. Land, 11(3), 419. https://doi.org/10.3390/land11030419
Naeem S, Cao C, Fatima K, Najmuddin O, Acharya BK (2018) Landscape greening policies-based land use/land cover simulation for Beijing and Islamabad—An implication of sustainable urban ecosystems. Sustainability, 10(4), 1049. https://doi.org/10.3390/su10041049
NASA, USGS (2021) Landsat 9 | Landsat Science. https://landsat.gsfc.nasa.gov/satellites/landsat-9.
Nedd R, Light K, Owens M, James N, Johnson E, Anandhi A (2021) A synthesis of land use/land cover studies: Definitions, classification systems, meta-studies, challenges and knowledge gaps on a global landscape. Land, 10(9), 994. https://doi.org/10.3390/land10090994
Nguyen TTH, Ngo TTP (2018) Land use/land cover change prediction in Dak Nong Province based on remote sensing and Markov Chain Model and Cellular Automata. Journal of Vietnamese Environment, 9(3), 132-140. https://doi.org/10.13141/jve.vol9.no3.pp132-140
Owojori A, Xie H (2005) Landsat image-based LULC changes of San Antonio, Texas using advanced atmospheric correction and object-oriented image analysis approaches. In 5th international symposium on remote sensing of urban areas, Tempe, AZ.
Pandey D, Heyojoo BP, Shahi H (2016) Drivers and dynamics of land use land cover in Ambung VDC of Tehrathum district, Nepal. Banko Janakari, 26(1), 90–96. http://dx.doi.org/10.3126/banko.v26i1.15508
Paudel B, Zhang YL, Li SC, Liu LS, Wu X, Khanal NR (2016) Review of studies on land use and land cover change in Nepal. Journal of Mountain Science, 13, 643-660. https://doi.org/10.1007/s11629-015-3604-9
Rawat PK, Tiwari PC, Pant CC (2011) Climate change accelerating hydrological hazards and risks in Himalaya: A case study through remote sensing and GIS modeling. International Journal of Geomatics and Geosciences, 1(4), 678-699.
Sapkota L, Dahal RP (2024) Assessment of forest cover change, key drivers of change and perception of locals in Birendranagar Municipality, Surkhet District, Nepal. Archives of Agriculture and Environmental Science, 9(2), 308-316. http://dx.doi.org/10.26832/24566632.2024.0902015
Saputra MH, Lee HS (2019) Prediction of land use and land cover changes for north sumatra, indonesia, using an artificial-neural-network-based cellular automaton. Sustainability, 11(11), 3024. https://doi.org/10.3390/su11113024
Şatır O, Berberoğlu S (2012) Land use/cover classification techniques using optical remotely sensed data in landscape planning. Landscape Planning. Rijeka: InTech, 21-54.
Shafizadeh-Moghadam H, Khazaei M, Alavipanah SK, Weng Q (2021) Google Earth Engine for large-scale land use and land cover mapping: an object-based classification approach using spectral, textural and topographical factors. GIScience and Remote Sensing, 58(6), 914–928. https://doi.org/10.1080/15481603.2021.1947623
Shrestha DP, Alfred Zinck J (2001) Land use classification in mountainous areas: Integration of image processing, digital elevation data and field knowledge (application to Nepal). ITC Journal, 3(1), 78–85.
Shrestha R, Bhandari S, Twayana R (2021) Monitoring Land Cover / Land Use Change Using Remote Sensing and GIS Technique for Preserving Agriculture Land at Dhulikhel and Banepa. 05, 1–8.
Singh A (1989) Review article digital change detection techniques using remotely-sensed data. International journal of remote sensing, 10(6), 989-1003.
Singh K, Anji M, Naveen KB (2019) Remote sensing and GIS based land use and land cover information of Medchal Mandal of Medchaldistrict. International journal of engineering and advanced technology, 9 (2), 4682-4685.
Singh P, Thakur JK (2012) Assessment of land use/land cover using geospatial techniques in a semi-arid region of Madya Pradesh, India. Geospatial techniques for managing environmental resources,6(10) 152-162.
Srivastava R, Singh S, Oran A (2020) Changes in vegetation cover using GIS and remote sensing: a case study of South Campus BHU, Mirzapur, India. Journal of Scientific Research, 64(2), 135-141. http://dx.doi.org/10.37398/JSR.2020.640219
Sulistiyanto (2021) Sistem Informasi Geografis Teori dan Praktek Dengan Quantum GIS. Ahlimedia Press.
Sumitra KC, Shrestha RP, Shrestha S (2018) Stream Discharge Response to Climate Change and Land Use Change in Tamor Basin, Nepal. International Journal of Engineering Technology and Sciences, 5(2), 50-62. http://dx.doi.org/10.15282/ijets.5.2.2018.1006
Tanksali A, Soraganvi VS (2021) Assessment of impacts of land use/land cover changes upstream of a dam in a semi-arid watershed using QSWAT. Modeling Earth Systems and Environment, 7(4), 2391-2406. https://doi.org/10.1007/s40808-020-00978-5
Timilsina S, Shrestha S, TripathiS, Bhattarai R, Mishra SK, Regmi, RR, Miya MS (2023) Assessment of land use land cover dynamics and its impact on springs water in Ritung Khola Sub-Watershed, Myagdi district, Nepal. Eurasian Journal of Soil Science, 12(2), 190-204. https://doi.org/10.18393/ejss.1244373
UN Department of Economics and Social Affairs. World Urbanization Prospects: The 2014 Revision; UNDESA:
Wang J, Maduako IN (2018) Spatio-temporal urban growth dynamics of Lagos Metropolitan Region of Nigeria based on Hybrid methods for LULC modeling and prediction. European journal of remote sensing, 51(1), 251-265. https://doi.org/10.1080/22797254.2017.1419831
Wang SW, Munkhnasan L, Lee WK (2021) Land use and land cover change detection and prediction in Bhutan's high altitude city of Thimphu, using cellular automata and Markov chain. Environmental Challenges, 2, 100017. https://doi.org/10.1016/j.envc.2020.100017
Yifru BA, Chung IM, Kim MG, Chang, SW (2021) Assessing the effect of land/use land cover and climate change on water yield and groundwater recharge in East African Rift Valley using integrated model. Journal of Hydrology: Regional Studies, 37, 100926. https://doi.org/10.1016/j.ejrh.2021.100926

No competing interests reported.

Prediction of Land Use Land Cover Change Using a Coupled CA-ANN modeling in Dhanusha district of Nepal

Status:

Version 1

Abstract

Figures

Introduction