Highlights

  • Comparison of infrared remote sensing with microwave remote sensing for cloud cover.

  • Flood mapping for affected districts utilizing the Sentinel-1A satellite.

  • Assessment of the impact of floods on land cover, with a LULC prediction for 2030 and its interrelation with future floods.

Introduction

The most frequent natural catastrophe events are floods, which are increasing in frequency, severity, and financial costs worldwide. Flooding is a broad transitory phenomenon that causes partially or completely dry areas to be submerged due to inland or tidal water overflow, unusually quick accumulation, or runoff (Jeb and Aggarwal 2008). According to the European Commission, a flood is described as “a natural event that results in the temporary inundation with water at a place that does not occur under normal conditions.” Though flood occurs naturally, severe flood events may have negative effects like displacing people and harming the environment (Adeoye et al. 2009; Nmeribeh 2011). Floods can also be brought on by anthropogenic activities and human interference with natural processes, such as expanding human settlements, population growth, and alterations to flood-prone low-lying plains, which change the natural drainage and river basin patterns and contribute to deforestation and climate change (EC 2007; Kwak and Kondoh 2008; Balabanova and Vassilev 2010). Flood hazard mapping is a crucial element for proper land use planning in flood-prone locations. It generates quickly accessible, easily readable charts and maps that help administrators and planners identify risk regions and prioritize their response and mitigation activities. Geographic information systems (GIS) and remote sensing are two methods that are highly useful and effective in the evaluation and management of flood risk. Real-time flood disaster monitoring, early warning systems, and rapid damage assessments have all seen considerable improvements as a direct result of technological breakthroughs in remote sensing and geographic information systems (GIS). With the help of remote sensing, information may be gathered rapidly over large areas. There are a variety of potential applications for remotely sensed images, including delineating flood plains, mapping flood-prone areas, mapping land use, detecting and forecasting floods, mapping rainfall, organizing evacuation plans, and evaluating the damage. One of the many applications of remote sensing data is the mapping of a range of terrain features that are essential for flood analysis. GIS can be used to input, organize, analyze, and manipulate the vast amounts of data collected for controlling and assessing flood risk. The study by Demir and Ulke (2020) utilized remote sensing and GIS techniques to obtain the manning roughness coefficient for flood risk analysis. GIS has emerged as a crucial tool for flood monitoring as it provides a variety of methods for identifying or forecasting flood-prone areas as well as modeling the affected areas. Demir and Kisi (2016) utilized GIS and hydraulic modeling to map floods in the Mert River, Turkey. The results showed a significant impact, with certain areas highly affected by floods, impacting 650 houses. In the past, a variety of datasets, including the optical dataset and Landsat, have been used for flood mapping. However, due to cloud cover, poor temporal resolution, and low spatial resolution, none of these data sets can be used in all-weather scenarios (Anusha and Bharathi 2020). Technology advancements enable the use of the synthetic aperture radar (SAR) dataset to address these drawbacks. Synthetic aperture radar is better suited for flood detection due to the increased capacity of SAR images and high precision of cloud cover to distinguish between water and non-water areas (Rahman and Thakur 2018). The European Space Agency’s Sentinel-1A satellite is a polar-orbiting satellite that generates data with great temporal and spatial resolution. It is a radar-based imaging satellite that captures images and provides data in all-weather circumstances, including morning and night, at a C-band intensity ranging from 4 to 8 GHz in the microwave region (Andy 2016). A study by Arora et al. (2023) utilized Sentinel-1 SAR data to investigate flood inundation within Punjab’s Ghaggar River basin. Their findings highlighted significant land damage in Patiala and Sangrur districts, affecting 9600 and 2700 ha, respectively, resulting in substantial losses in paddy fields. Sugianto et al. (2022) examined the impact of land use and land cover (LULC) changes on floods in the Teunom watershed. Their research indicates a notable increase in residential, rice fields, and wetland areas, with varying infiltration rates contributing to variations in flood zone hazards. Anusha and Bharathi (2020) employed Sentinel-1A SAR data for flood mapping in the Rapti and Ghaghara rivers, identifying Gorakhpur as the most affected district, effectively captured by SAR imagery. Studies show that flood hazards are particularly affected by changes in land use and land management. The seasonal and annual distribution of stream flow can be considerably changed by deforestation, urbanization, and land use changes (Dunne and Leopold 1978). It is well recognized that converting land from dense forest cover to intensive pasture increases the rate and overall volume of runoff during storm occurrences (Selby 1972; Jackson 1972; Rowe 2003). A significant rise in impervious surface frequently accompanies the conversion of agricultural, forested, grassland, and wetland areas to urban areas, which might change the watershed’s natural hydrologic conditions. The runoff generation process is altered as a result of increased human intervention and catchment area expansion, particularly due to a reduction in the soil cover’s ability to infiltrate. Global land use pressure is still rising, and as ecosystems and human livelihoods are intertwined, additional unfavorable LULC changes are anticipated to occur soon. As a result of this, having an understanding of the patterns and trends of LULC is necessary to steer responsible resource use trends both in the present and in the future. Finding a technique to measure the key factors that influence local land use is a necessary step before building models. In modeling, a variety of factors are taken into account that may serve as land use change catalysts. These factors include topography, elevation, slope, distance from a riverbed, proximity to metropolitan areas, and distance from water bodies, among others. It evaluates how these proxy factors collectively affect predictions of the change’s type and location. Therefore, they are important and helpful for resource managers and land use planners in developing proper policies and decisions for managing land use (Guan et al. 2011; Mohamed and Worku 2019). They can be used to prioritize conservation areas, predict urban and rural developments, and examine evolving agriculture dynamics (Oyana et al. 2014). Model building based on knowledgeable predictions of the future behavior of the driving elements is necessary for forecasting future LULC trends (Theobald and Hobbs 2002; Verburg et al. 2004; Munthali et al. 2020). It is significant to note that LULC modeling aids natural resource planners and managers in mitigating or preventing the harmful effects of unfavorable future, in addition to providing checks on current land use plans (Oyana et al. 2014). Hence, the main objective of LULC modeling is to guarantee an adequate provision of natural resources for both present and future generations (Verburg et al. 2004). Moreover, regulating changes in land use and land cover is also crucial for minimizing flood occurrences (Adnan et al. 2020). The global focus on flood management strategies has intensified in response to catastrophic events in multiple regions worldwide. Hence, it is imperative to comprehend the relationship between LULC and floods. The study by Hasan et al. (2023) investigated the connection between alterations in LULC and flooding in the Jamuna River for a 30-year duration. Findings indicate that changes in LULC in areas distant from the river substantially influenced alterations in riverine flooding, while LULC changes near the river exhibited a weaker correlation with changes in riverine flood potential. To make accurate future forecasts of LULC on a global scale, researchers have turned to a wide range of models, including agent-based, Markov models, and cellular automata models (Sohl and Claggett 2013; Li et al. 2016; Mwanjalolo et al. 2018). Integrated models, such as the cellular automata CA-Markov model, have been shown to perform much better in forecasting long-term temporal and spatial land use and land cover fluctuations (Yang et al. 2012; Liping et al. 2018; Munthali et al. 2020). The CA-Markov model is reliable and applicable in a diverse array of land use and land cover planning situations (Kamusoko et al. 2009; Fu et al. 2018; Fitawok et al. 2020). Adnan et al. (2020) performed the flood risk assessment for present and future LULC scenarios, specifically for 2030, using the CA-Markov model. The results showed a projected increase in expected annual damage, rising from $903 million in 2005 to $2096 million in 2030. Another study by Dalanhol et al. (2020) projected future land use and land cover to map flood-prone areas in Pato Branco City, Brazil. The results suggest a future expansion of urban areas, resulting in soil impermeabilization and flooding. In the existing literature, very limited studies have been conducted on future land use and land cover (LULC) changes in conjunction with flood events, particularly within the context of the Indian subcontinent. Therefore, this study seeks to bridge this gap by examining the association between projected LULC changes for 2030 and future flood occurrences. The change detection and thresholding (CDAT) method has been adopted to detect floods utilizing SAR datasets (Clement et al. 2018; Singha et al. 2020). This technique aims to identify a specific backscatter intensity value (in dB) that serves as a threshold, allowing the image to be categorized into flooded and non-flooded areas (Islam and Meng 2022). A key advantage of employing this technique is the reduced computational time, making it applicable to satellite images of varying complexity (Haile et al. 2023). Unlike other processes that use a single SAR image, CDAT highlights temporal land cover changes by comparing flood images to preceding dry images (Schlaffer et al. 2015; Wu et al. 2019). This study contributes significantly as it employs a GIS-based technique to detect floods in the affected districts of Saharsa, Supaul, and Madhepura within the Kosi River Basin. It aims to identify all land classes prone to flooding and offers future projections using the Land Change Modeler of Terrset version 18.31. Therefore, this study aims to: (a) examine the extent of the flooding using Sentinel-1A SAR data and its repercussions on different land classes; (b) forecast LULC patterns for the year 2030 and its potential impact on future floods. The findings will assist both the government and the local communities in formulating effective long-term strategies to mitigate the damage caused by flooding in these districts.

Study area description

The Kosi is one of the largest rivers that flow into the Ganges, and after draining a significant portion of land in Nepal, the Kosi eventually makes its way into Bihar, close to the city of Bhimnagar (Towheed and Roshni 2021a, b). The drainage area encompasses portions of Tibet, Nepal, and India and is approximately 52,731 km2 in size. The region of north Bihar is home to one of the most vulnerable areas to flooding in India, which is caused by the Kosi River. The Kamlabalan, Bagmati, and Bhutibalan rivers are three of the Kosi River’s most important tributaries. With an annual rainfall of 1205 mm on average and 53 rainy days, the state of Bihar, which is located in eastern India, is considered to be one of the most flood-prone states in the country (Ghosh and Mukhopadhyay 2014). The topographical gradient in the Bihar plains is quite low (0.01°–0.05°), which contributes to the widespread flooding as the water volume in the rivers increases. The rivers carry significantly higher amounts of water, ranging from 10 to 20 times more, during the monsoon compared to the winter season. However, in cases of intense rainfall in the river’s catchment regions, the river flow can surge up to a hundred times its normal volume. A predominately westerly movement across around 150 km of alluvial plains has been documented in approximately 200 years due to the fact that the Kosi River, in particular, is renowned for shifting the course that it travels (Wells and Dorr 1987; Chakraborty et al. 2010). Due to the regular movement of the channel and the very high sediment flux of 102 Mt/yr (Sinha 2009), a megafan has formed in the alluvial plain of Kosi (Gohain and Parkash 1990; Mishra and Sinha 2020). This megafan is primarily located in the plains of north Bihar. Some of the worst flood-affected districts include Supaul, Saharsa, Madhepura, Purnia, Katihar, and some parts of Nepal that are nearby. The flood-prone area is densely populated, and casualties are common. Saharsa, Supaul, and Madhepura districts of the Kosi Basin have been selected as an area of study as these areas get inundated each year due to flooding. Geographically, the Kosi Basin extends from 86° 20ʹ to 87° 41ʹ east longitudes and 25° 20ʹ to 26° 48ʹ north latitudes, as shown in Fig. 1, which depicts the Digital Elevation Model (DEM), Kosi River, and drainage network of the study area. Saharsa is located at 25.87° N, 86.5° E, and is typically 41 m above sea level. Saharsa and the area surrounding it are located on a level alluvial plain that is a part of the Kosi River Basin. Even though the soil is very fertile, changes in the course of the Kosi, one of the Ganges’ main tributaries, have led to problems with soil erosion. Flooding is also a big reason why the area isn’t well connected. The Supaul District, which is located in the Kosi division in northern Bihar and spans the coordinates 25° 37ʹ N to 26° 25ʹ N and 86° 22ʹ E to 87° 10ʹ E, has an area of 2,103 square kilometers and is bordered to the north by Nepal, the east by Araria district, the south by Madhepura and Saharsa districts, and the west by Madhubani district. Madhepura district, on the other hand, covers an area of 1,930 square kilometers and is located in the plains of the Kosi River in north-eastern Bihar at 26.07° N 87.07° E. Within the Kosi River Basin, it is crucial to acknowledge that reducing the risk and mitigating damage caused by floods also relies on altering land use rather than only implementing structural measures (Sudhakar Reddy et al. 2008; Mishra and Sinha 2020). Agriculture constitutes the primary economic activity in these districts, serving as the primary source of livelihood for the residents. However, encroachments on the natural resources of the region have escalated. The unregulated pace of LULC alterations within rural and urban areas, particularly the conversion from natural to artificial land cover, has emerged as a significant global issue (Gharbia et al. 2016). Therefore, it is crucial to conduct regional mapping of LULC to assess both favorable and adverse changes, analyze the effects of transitions from one land cover type to another, identify the factors driving these changes, and comprehend the relationship between human activities and the environment (Alipbeki et al. 2020). Moreover, these encroachments promote flooding, and addressing them in vulnerable regions helps build resilience to extreme weather. Hence, it is essential to implement regulatory frameworks to uphold land use regulations, building codes, and floodplain management ordinances. The spatial analysis regarding flood occurrences and alterations in land use presented in this study can assist government officials and local communities in identifying areas necessitating intervention through land use policies. Furthermore, the suggested approach for evaluating the consequences of evolving land use and flood risks in the future should be useful for land use planners.

Fig. 1
figure 1

Location map of the study area

Full size image

Data used and methodology

Flood assessment

Data

This study conducted a comparative analysis of three distinct satellite datasets, namely, Sentinel-1A, Sentinel-2B, and Landsat-8 to assess cloud cover during the 2020 flood. Landsat-8 imagery was also employed to generate land cover maps for the year 2020, facilitating the estimation of inundated land cover during flooding. The list of datasets used in this study is shown in Table 1.

Table 1 Types of data utilized in the study
Full size table

Flood mapping and assessing land cover inundation

During flooding events, extracting information from satellite images becomes challenging due to dense cloud cover. This study compared images captured by Sentinel-1A, Landsat-8, and Sentinel-2B on July 22, 2020, amidst the flood event, to assess the extent of cloud cover within the images. All three satellite images were projected to WGS 84/UTM zone 45N, mosaiced, and then clipped to the study area. The cloud free satellite data can be utilized to identify flood-affected areas and develop further preparedness strategies by providing crucial information about the floods (Klemas 2014). Due to Indian monsoon abnormalities, 3–4 months of the monsoon season in India account for 5% of the country’s annual rainfall. Therefore, during this time, rivers release heavy discharges, which leads to extensive flooding. Heavy rains caused the Kosi River to surge above the red mark on July 18, inundating the main regions of the districts of Saharsa, Supaul, and Madhepura. Flood mapping require two types of remotely sensed data. One set comprises images captured prior to the event, whereas the other set comprises images taken during the flood event (Klemas 2014). The Sentinel-1A SAR image from March 1, 2020, was chosen for pre-flood analysis, and the SAR image from July 22, 2020, was chosen for flood analysis. The Landsat-8 image of March 2020, was chosen for LCLU classification to estimate the different land classes affected by flooding. Erdas Imagine version 15 is used for pre-processing and image processing, while ArcGIS version 10.3 is used for post-processing. The sequential approach outlining the methodology is shown in Fig. 2.

Fig. 2
figure 2

Flow chart for processing LandSat-8 and Sentinel-1A image

Full size image
Pre-processing

The Sentinel-1A image for Saharsa, Supaul, and Madhepura districts is subsetted to choose the specific area of interest (AOI), and speckle correction is applied. Speckle noise, a type of granular noise that is inherent to SAR images, reduces the image quality and poses a significant challenge due to the coherent nature of scattering phenomena (Ahmed et al. 2011). A median filter with a 3 × 3 kernel size is used to remove the speckle noise from the SAR image. Then, image-to-image rectification is used to automatically correct the image distortions. To achieve positional precision, the despeckled image is geometrically corrected to a predetermined projection system using a master reference image with a root mean square (RMS) error of less than 0.5 pixels.

Image processing

Utilizing the digitization technique, a water mask is generated from the pre-flood image and change detection and thresholding (CDAT) approach has been adopted. The CDAT method needs two satellite images to identify alterations in flood inundation. One image, denoted as the reference image (R1), represents land cover without flooding, while the other, referred to as the target image (R2), is acquired on the flood date. In this investigation, a pre-flood image from March 1, 2020, served as the reference image due to its minimal variability in signal backscatter intensity during dry conditions. Bekele et al. (2022) provide comprehensive information on the selection of reference images. Using the conditional function shown below, thresholding is carried out in the model maker in Erdas Imagine version 15:


EITHER < argument1 > IF (<test>) OR < argument2 > OTHERWISE where,


argument1-> set water pixels a value of 1,


argument2-> set non-water pixels a value of 0.


Test-> the threshold limit value and the input image.

The conditional function in the model maker is used to classify the pixels as water pixels if they fall below or equal to the specified threshold value T, which in this instance is set at 112. Since CDAT technique has been used which uses one threshold value for entire image in which value less than 112 gives weaker backscatter signal from water and pixels are labeled as inundated. The remaining pixels are given a value of 0 using conditional function in ERDAS Imagine version 15. If digital number is less than or equal to threshold then it can be considered as water pixels else non-water pixels.

Post-processing

A land use and land cover (LULC) map of March 1, 2020 is generated through supervised classification, categorizing the area into four classes: water bodies, built-up areas, barren land, agricultural and vegetation areas. Subsequently, this LULC map is superimposed on the inundated water layer and flood assessment is performed by calculating the inundated area of different land classes in each district.

Land use projections

Anicillary data

This study examined the socioeconomic factors that directly impact the pattern of future growth. The model utilized the following five factors for 2030 prediction:

  1. (a)

    Road Proximity: The presence of a road network plays a vital role in facilitating and connecting other land classes. Data on road networks were obtained from OpenStreetMap (https://www.openstreetmap.org/) as of March 1, 2020. The Euclidean distance tool in QGIS version 3.22.4 was used to compute the proximity to roads, considering the nearest cell (Singh et al. 2022).

  2. (b)

    Elevation: The elevation map was prepared utilizing ArcGIS version 10.3 from the Digital Elevation Model (DEM) to determine the terrain topography. DEM, of 30 m resolution is derived from the Shuttle Radar Topographic Mission (SRTM) can be downloaded from https://earthexplorer.usgs.gov.

  3. (c)

    Slope: The slope map was generated from the DEM using a slope tool in ArcGIS version 10.3.

  4. (d)

    Land use land cover: LULC maps for March 2013, 2016 and 2020 are prepared using the supervised classification tool in ArcGIS version 10.3. These maps can be utilized to identify and analyze the changes in the landscape.

  5. (e)

    Urban Distance Map: An urban distance map is prepared using the Euclidean tool in ArcGIS version 10.3. The map shows how an urban area would expand in any direction, i.e., there is a strong likelihood of urban expansion in nearby areas due to the influence of the neighborhood effect (Singh et al. 2022).

Image processing and classification

All Landsat-8 images utilized in this study were Level 1 Terrain-corrected data, signifying that they had undergone geometric correction (Zhu 2017). These images were projected to WGS 84/UTM zone 45N, processed through RGB color composition, mosaicked, and then clipped to the study area. In ArcGIS version 10.3, the images underwent classification using the maximum likelihood algorithm for supervised classification. Training samples were chosen by creating polygons around representative class pixels, with classes including water bodies, built-up areas, barren land, agriculture and vegetation areas.

Future LULC Projections

Using the Land Change Modeler in TerrSet version 18.31, future LULC scenarios for Saharsa, Supaul, and Madhepura for the year 2030 were predicted using the hybrid cellular automata–Markov model (CA-Markov) (Guan et al. 2011). Markov chain analysis (MCA) and multi-layer perceptron (MLP) are used to create a soft prediction of the future (2030) LULC distribution (Eastman 2016). The modeler utilizes Markov chain analysis so as to assess the probability of a transition from one category to another based on the classified land cover maps from the past and present (Beeram et al. 2023). Whereas the MLP algorithm builds a neuron network using weights, then employs those weights to calculate the training error and adjust the weight accordingly. A change potential map was generated for each transition between different LULC types, creating individual transition submodels. Various transitions can occur based on the same underlying driver variables and are influenced by the susceptibility of LULC to transition into other land uses. In such cases, the submodels are combined into a single composite map, representing the change potential (or transition suitability) for that land use (Eastman 2016). The multi-objective land allocation (MOLA) and cellular automata included in the land change modeler (LCM) were employed to spatially distribute the Markov transitions. These transitions consist of a probability matrix, which calculates the likelihood that a class of land cover will change into other classes, and an area matrix, which accounts for the number of pixels that are anticipated to switch from one land cover class to all other classes. The model offers the following four forecasts for LULC change in the future: (i) change evaluation; (ii) development of potential transition matrices; (iii) validation of data; (iv) future prediction. To account for the changes that occurred during the period, the LULC maps of the time periods 2013, 2016, and 2020 were prepared. Urban distance map, road distance map, elevation, slope map, and LULC maps for 2013, 2016, and 2020 were exported to TerrSet version 18.31. To further identify the potential factors contributing to changes in land use, the Cramer’s V test is employed. This non-parametric statistical tool is utilized in Terrset version 18.31 to assess the effectiveness of drivers. The variables tested include distance to roads, distance to urban centers, elevation, and slope. Using the land cover images from 2013 and 2016 as a reference and a transition probabilities matrix, a Markov module was utilized to model the LULC of 2020. By comparing the generated LULC map of 2020 (Terrset) with the actual LULC map of 2020 (Arc GIS) based on the Kappa coefficient, the Markov model was validated. The kappa coefficient provides a quantitative assessment of the accuracy of pixel classification into various LULC (Land Use Land Cover) classes, thereby assessing the reliability of the classified image (Singh et al. 2022). The kappa coefficient’s value ranges from −1 to +1, with +1 indicating perfect agreement (Lillesand et al. 2015). The CA-Markov prediction module in the Land Change Modeler of TerrSet was then used to forecast LULC changes for 2030 using the validated LULC map of 2020 (Terrset) as a reference. A step-by-step methodology for the same is shown in Fig. 3.

Fig. 3
figure 3

Flowchart illustrating the processing steps for the 2030 LULC projection

Full size image

Results

Flood assessment

The satellite-based imagery is being utilized to analyze the massive flood crisis that happened in Bihar, India, in July 2020. The comparison of Sentinel-1A, Landsat-8, and Sentinel-2B satellite datasets for the cloud cover of all three selected districts was chosen as the AOI, which is shown in Fig. 4. Landsat-8 and Sentinel-2B data consist of 80–90% cloud cover, as shown in Fig. 4b, c, which makes it difficult to extract information, whereas the Sentinel-1A SAR image, as shown in Fig. 4a, consists of <1% cloud cover, making it ideal for flood monitoring. Effective duration and flood dynamics assessment requires a number of images captured by the satellite at various points throughout the floods (Voigt et al. 2008). Sentinel-1A satellite images covering the regions of Saharsa, Supaul, and Madhepura were impacted by floods in July 2020. The pre-flood image of 1 March 2020 and the flood image of 22 July 2020 are shown in Fig. 5. Utilizing a Landsat-8 image from March 1, 2020, the LULC (ArcGIS) map of Saharsa, Supaul, and Madhepura for the year 2020 is shown in Figs. 6, 7 and 8. The results show that 34.9, 34, and 46.2% of the areas in Saharsa, Supaul, and Madhepura were flooded, respectively. Barren land was predominantly affected by flooding in all three districts. Following this, agriculture and vegetation areas in Madhepura (49.8%) and Saharsa (39.7%) were submerged, with Supaul (29.5%) being affected subsequently. In the case of the built-up area, Madhepura (39.9%) experienced higher inundation than Supaul (36.2%) and Saharsa (28.6%). Overall, Madhepura emerged as the most affected district, followed by Saharsa and Supaul. Area-wise inundation is shown in Table 2.

Fig. 4
figure 4

a–c Cloud cover of Sentinel 1A (a), Landsat 8 (b) and Sentinel 2B (c) flood image of Saharsa, Supaul, and Madhepura district

Full size image
Fig. 5
figure 5

Pre-flood image and during flood image of Saharsa, Supaul, and Madhepura of 1-03-2020 and 22-07-2020, respectively

Full size image
Fig. 6
figure 6

Land use land cover of Saharsa District for the year 2013–2020 and Projected land use for the year 2030

Full size image
Fig. 7
figure 7

Land use land cover of Supaul District for the year 2013–2020 and Projected land cover for the year 2030

Full size image
Fig. 8
figure 8

Land use land cover of Madhepura District for the year 2013–2020 and Projected land use for the year 2030

Full size image
Table 2 Inundated area-wise land use land classification
Full size table

LULC projections

The Land Change Modeler, utilizing Terrset version 18.31, is used to forecast future land use and estimate the likelihood of geographical trend changes. It offers Cramer’s V test as a non-parametric statistical method to assess the effectiveness of the drivers. The variables with Cramer’s V values of 0.15 and higher are satisfactory, while those with values of 0.4 are good (Eastman 2016). The values for the employed drivers in all three districts are listed in Table 3. The values of Cramer’s V variables are very close, and “distance to urban” was the most prominent factor. The kappa coefficients between LULC 2020 (ArcGIS) and LULC 2020 (Terrset) for Saharsa, Supaul, and Madhepura are 0.91, 0.94, and 0.95, respectively. Kappa coefficients exceeding 0.5 can be deemed satisfactory, with values greater than 0.8 considered excellent (Abdelkareem et al. 2018; Bunyangha et al. 2021). Initially, these districts were predominantly comprised of water bodies, built-up areas, barren land, agriculture, and vegetation areas. The LULC maps of 2013, 2016, 2020, and 2030 for Saharsa, Supaul, and Madhepura are shown in Figs. 6, 7 and 8. However, several changes occurred in the LULC composition. From 2013 to 2020, there was a slight decrease in water bodies observed in Saharsa and Madhepura, whereas there was a slight increase in Supaul. Concurrently, built-up areas experienced growth across all three districts, with the most significant increase recorded in Madhepura, followed by Supaul and Saharsa. In contrast, there was a notable decrease in barren areas observed in Saharsa and Madhepura, with Supaul experiencing a subsequent decline from 2016 to 2020. Additionally, agricultural and vegetative areas expanded in Saharsa but declined in Supaul and Madhepura during the same period. Area-wise land cover composition of these districts is shown in Table 4. The actual land cover for the year 2020 is compared with the forecasted LULC utilizing Terrset version 18.31 (Table 5). A positive change (%) signifies an overestimation in land cover area, whereas a negative change (%) indicates an underestimation. The analysis shows that water bodies have been slightly overestimated in Saharsa and underestimated in Supaul and Madhepura. Built-up areas have been consistently underestimated across all three districts, whereas barren land has been overestimated in Saharsa and underestimated in Supaul and Madhepura. As for agriculture and vegetation, there has been a slight overestimation in all three districts. The transition probability matrix was used to analyze the probable percentage changes in LULC for the years 2020–2030. The findings indicate that by 2030, water bodies are projected to undergo a marginal increase in Saharsa, while decreasing in Supaul and Madhepura. Built-up areas are expected to experience a substantial increase across all three districts. Barren land is anticipated to undergo a significant decrease in all districts. Additionally, agriculture and vegetation are predicted to marginally increase in Saharsa and Supaul, with a slight decrease expected in Madhepura. Projected area coverage for 2030, and change analysis are shown in Table 6.

Table 3 Selection and evaluation of spatial drivers of flood-affected districts
Full size table
Table 4 Area-wise land use land classification for 2013, 2016, and 2020
Full size table
Table 5 Comparison of actual and projected LULC types for year 2020
Full size table
Table 6 Comparison of actual and projected Land Cover Change for year 2020 and 2030
Full size table

Discussion

Flood Assessment

In this study, a comparison of cloud cover during a flood event showed that infrared-based Sentinel-2B and Landsat-8 typically have 80–90% cloud cover, posing challenges in extracting information. In contrast, microwave-based Sentinel-1A has less than 1% cloud cover, making them well suited for flood monitoring. Due to infrared sensors’ inability to see through clouds, early infrared datasets could not accurately depict the extent of a flood. The majority of passive sensor satellites experience cloud cover, which contaminates the image and makes it challenging to extract the data. There are numerous methods to remove clouds, which include cloud masking, temporal aggregation, and other interpolation methods (Wan 2014). However, with technological improvement, microwave-based sensors are now regarded as the most effective tool for flood detection and mapping. The SAR satellite imagery is highly effective in evaluating flooded areas as it can capture images consistently at all times (day and night), irrespective of weather conditions and almost negligible cloud cover. In recent times, there have been substantial modifications in the Kosi Basin region. These alterations are predominantly influenced by natural factors like fluctuations in rainfall and human-made factors such as population growth, urbanization, and variable agricultural techniques. It has been acknowledged that regulating LULC shifts is a crucial spatial phenomenon for directing socioeconomic growth (Corner et al. 2014). The type and amount of land use and land cover (LULC) greatly impact the way in which water flows in rivers and surrounding areas (Yalcin and Akyurek 2004). This study has also examined the impact of floods on different land classes in Saharsa, Supaul, and Madhepura. The results indicate a significant impact on barren land in all three districts. The barren area is mainly affected by abundant rainfall, gentle terrain, and a well-connected network of active channels, which are identified as major factors amplifying flood risks, rendering about 67% of the Kosi Basin area vulnerable to flooding (Mishra and Sinha 2020). The Kosi River is characterized by the presence of paleochannels. During the monsoon season, many of these river channels experience a resurgence in flow and become active. However, their flow into the current Kosi River channel is impeded by the embankments constructed along the left bank. These embankments are positioned very close to the active stream channel, leaving the stream with limited space to accommodate its overbank flow. Furthermore, there are settlements located directly on the embankment, which not only compromises the integrity of the embankment but also poses a heightened risk to human life and property in the event of a breach (Sinha 2009). Hence, an optimal flood management approach would entail a thorough assessment of flood hazards, followed by well-planned floodplain zoning. The government of Bihar reports that 78% of the population engages in agriculture and related activities, either as cultivators or agricultural labor (Kansal et al. 2017). This situation is particularly critical in North Bihar, an area prone to significant flooding. The Kosi Basin is primarily focused on agriculture, and there has been a substantial increase in farming activities in the region over the past 40–50 years (Uddin et al. 2015). Compared to fallow or marshy land, urban and agricultural areas are more vulnerable to flooding risks (Ghosh and Kar 2018). Madhepura experiences the most significant flooding in the agriculture and vegetation regions compared to Saharsa and Supaul. Intensive agricultural activities, such as soil tilling, compact the soil and reduce permeability, resulting in increased surface runoff during rainfall events and contributing to flooding. Additionally, the loss of riparian vegetation along rivers can elevate erosion and sedimentation in water bodies, leading to localized flooding during periods of high flow. Agricultural practices often include drainage alterations intended to enhance productivity by expelling excess water from fields, but these modifications may also contribute to downstream flooding by hastening the discharge of water into rivers and streams. Improper irrigation practices, inadequate drainage systems, or poorly designed water management strategies in agriculture can induce waterlogging and escalate the risk of localized flooding (Ares et al. 2024). Therefore, sustainable land management practices, such as soil conservation, reforestation of riparian areas, and controlled drainage, are essential for mitigating the impact of agriculture on flooding. Floods have significantly impacted built-up areas in all three districts. The exposure to flood risk in urban areas will increase as spatially impermeable land increases and natural drainage pathways change (Ashaolu et al. 2019). This occurs because flooding is exacerbated when water infiltration is inhibited after a precipitation event. Due to rapid urbanization, vegetation is being converted into built-up areas. If vegetation is removed, there will be more runoff compared to areas with abundant vegetation, as it affects the ability of water to infiltrate the soil and the speed at which it can flow through roots and pores. Improper urban development can disrupt natural hydrological processes, such as surface flow patterns and groundwater recharge, resulting in alterations to runoff timing and volume (Mabrouk et al. 2024). Therefore, proper land use planning strategies should be employed to restrict development in flood-prone regions and safeguard natural floodplains. Changes in land cover characteristics are crucial for determining how much a region is at risk from flooding. To identify the affected areas, disaster management organizations need precise flood mapping. They rely on this vital information to take the appropriate action and assist the flood victims in the impacted areas (Halounova and Holubec 2014).

Future Projections

The Land Change Modeler is used to project future land use as it predicts the probability of geographically oriented changes in land use and land cover. To confirm the accuracy of the model, both simulated and actual LULC maps from 2020 were employed. The Kappa (K) accuracy coefficients of Saharsa, Supaul, and Madhepura are 0.91, 0.94, and 0.95, respectively, which indicates that the model’s prediction accuracy is reasonable. The Kappa values, all above 0.8, indicated a high level of agreement between the projected and actual land use and land cover maps of 2020 (Araya and Cabral 2010; Bunyangha et al. 2021). Based on the simulated map, the modeled result has been projected for the year 2030. It is evident that the hydrological regimes of Supaul (−1.26%) and Madhepura (−5.31%) are anticipated to exhibit a decline, while an augmentation in water bodies is expected in Saharsa (2.4%). This depends on multiple factors, such as hydrological management practices, altered precipitation patterns, etc. The built-up area has undergone a marked escalation in all three districts, with Supaul (13.43%) displaying the most significant rise, followed by Madhepura (10.32%) and Saharsa (2.70%). This spike in the built-up area may be attributed to rapid urbanization and the conversion of other land classes into urban terrain. The barren area has witnessed a drastic reduction in Supaul (−63.94%), followed by Madhepura (−28.88%) and Saharsa (−14.53%). The conversion of barren area into built-up and agricultural/vegetation areas is evident from the illustrations Figs. 6, 7 and 8. The agriculture and vegetation areas are projected to experience a marginal increase in Supaul (4.57%) and a slight decline in Madhepura (−3.23%), but remain relatively stable in Saharsa (0.008%). Since these districts rely primarily on agriculture, the likelihood of change in these regions is low, as their principal occupation is agro-based. Moreover, it is equally essential to highlight that as future changes in land use and land cover (LULC) persistently reduce the extent of land cover (including barren areas, agriculture, and vegetation) and replace it with urban development, it becomes crucial for the communities residing in these regions to adopt appropriate strategies to mitigate adverse climate conditions. These unfavorable circumstances will continue to impact the well-being of the local population. Consequently, both local and central governments should establish and oversee measures aimed at mitigating environmental degradation, such as reforestation and responsible management of alterations in land cover. Furthermore, the development of infrastructure in these areas, particularly in the vicinity of urban centers, should be properly planned, with a special focus on preserving green spaces in and around urban areas and promoting environmentally friendly green cities. In order to establish a relationship between policy and LULC change, it is necessary to have a thorough comprehensive understanding and monitoring of all the relevant factors.

Linkage of projected LULC with future floods

An endeavor has been made to establish a relationship between land cover changes and prospective floods using the simulated 2030 map. The land use and land cover change scenario employed in this study was predicated on the assumption that the alterations in land use and land cover within the coming decade (2020–2030) would follow patterns similar to those that have driven changes in land use and land cover in previous years. It was estimated that the built-up area of Madhepura has increased, and given that it was the district most impacted by flooding in 2020, there is a likelihood that flooding in 2030 could be worse because the expansion of man-made structures will compact the area, thereby causing a higher concentration of discharge. The conversion of land into built-up areas often results in a rise in impermeable surfaces, which reduces the ability of water to seep into the ground, leading to a decline in the replenishment of groundwater (GWYB 2019). In urban areas, the land is predominantly covered with impermeable materials like pavement and concrete, hindering the infiltration of rainwater into the soil. The solidified soil surface takes a considerable amount of time to absorb water during rainy periods (Portelinha and Zornberg 2017). The majority of rainfall in residential zones transforms into surface runoff, gathers in lower areas, hampers water absorption, and leads to flooding (Dawson et al. 2017). According to Seidl et al. (2020), runoff velocity is at its peak in residential localities due to the challenges of water penetration into the ground. Consequently, a significant portion of the water is directed through canals to other low-lying areas, often at a high speed and in substantial quantities. This accumulation of water in lower regions results in flooding and erosion. Frequently, the drainage canals and culverts in residential zones are too narrow to effectively manage rainwater, causing overflow within these areas. This predicament is exacerbated by the improper disposal of solid waste by residents, leading to blockages in the drainage system. The appearance of land cover and the activities carried out on the land further heighten the risk of flooding due to reduced infiltration capacity coupled with the topographical features of the areas where these activities are conducted (Wu et al. 2022). The agricultural and vegetation areas in Supaul will grow for the year 2030, despite the vegetation class suffering less damage from the 2020 floods than built-up and arid areas. However, when agriculture and vegetation areas grow, land cover vegetation will reduce the intensity of flooding. In districts like Madhepura, the vegetation and agricultural coverage are diminishing. This decline in grassland and shrubs is linked to an increase in water discharge. Vegetation, including forests and shrubs, plays a vital role in mitigating peak water discharge levels (Zhang et al. 2018). Watersheds with adequate vegetation tend to slow down the water discharge, resulting in a gradual decrease in peak flow. Conversely, watersheds lacking sufficient vegetation experience a sharp and sudden surge in discharge rates (Yang and Nepf 2019). This underscores that areas with reduced vegetation, which impacts water infiltration, swiftly convert precipitation into runoff within the watershed, which then accumulates in basins and rivers. Consequently, water volume accumulates rapidly in areas with limited capacity, leading to runoff and subsequent flooding in the future. The barren area in all three districts has decreased drastically and has mostly been converted to built-up areas. To address the potential flood risk associated with the conversion of barren areas into built-up zones, several preventive measures should be adopted to ensure proper utilization of such land and reduce the likelihood of flooding in the future. Integrate green infrastructure strategies within urban areas, such as the use of permeable pavements and eco-friendly building designs. Promote sustainable development principles that prioritize the creation of green spaces and minimize impermeable surfaces. Enhance public awareness regarding flood vulnerabilities and the significance of responsible urban planning. Encourage property owners to embrace flood-resilient practices. Formulate adaptable land use and zoning blueprints that take into account shifting climate patterns and forthcoming flood hazards. This adaptability in planning can assist communities in adapting to evolving challenges. By implementing a blend of these preventive measures, communities can efficiently repurpose barren land for development while mitigating the potential for future flooding. Achieving this goal necessitates a collaborative effort involving government authorities, urban planners, and the local community to establish sustainable, resilient, and flood-resistant urban environments. However, further extensive study is required to interlink future floods with LULC. Overall, it may be argued that this study can act as a foundation for further research in the area.

Limitations of the study

Although SAR has operational benefits over infrared systems, there are a few limitations in detecting flooding. The SAR systems operate in a side-looking manner, and terrain characteristics may lead to issues like radar shadowing, layover, and foreshortening, depending on the incidence angle (Giustarini et al. 2015; Haile et al. 2023). Specifically, radar shadows can pose challenges in hydrological analyses, causing dark areas within radar images that may be incorrectly classified as water. Detecting flooding can be challenging in regions where nearby structures above the water’s surface can influence the radar signal. This results in the double-bounce effect, where the signal reflects off the water surface before encountering the vertical structure, forming a corner reflector and resulting in a powerful return signal at the satellite (Jung et al. 2010; Pulvirenti et al. 2016). The dense concentration of buildings in urban regions can result in both radar shadowing and the double-bounce effect, restricting the accurate mapping of flood extents unless employing costly ultra-high-resolution SAR data, like TerraSAR-X, RADARSAT-2, or COSMO-SkyMed (Clement et al. 2018). Nevertheless, SAR sensor’s ability to penetrate through cloud cover, coupled with enhancements in the spatial and temporal resolution of freely accessible datasets, underscores the significant role SAR plays in flood monitoring and management. Consequently, for 2030 future projections, the model employed in this study proves to be an efficient method for conducting simulations, but there are uncertainties in predicting future land type classes. Firstly, the model cannot estimate future alterations resulting from governmental modifications in land development policies and master plans, along with natural processes impacting land use patterns. The model forecasts future changes based solely on the current identified LULC patterns (Singh et al. 2022). Secondly, the driving factors evolve over time. For instance, alterations in the transportation system directly impact the direction and shape of urban growth, leading to shifts in settlement locations and density. Therefore, selecting the most suitable set of drivers is crucial for enhancing the accuracy of model predictions. Failure to do so may result in inaccuracies in estimating accurate changes (Keshtkar and Voigt 2016; Wang et al. 2019).

Conclusions

This study has compared the cloud coverage of Sentinel-1A, Landsat-8, and Sentinel-2B datasets during the 2020 floods, as cloud cover often affects satellite observations. The Sentinel-1A dataset (<1% cloud cover) is preferred for detecting floods compared to the other two datasets, which have 80–90% cloud cover. Hence, irrespective of cloud cover, Sentinel-1A can operate effectively under all-weather conditions for detecting floods. Additionally, it has also assessed flooding in Saharsa, Supaul, and Madhepura by utilizing SAR datasets, examining its impact on different land classes to estimate the extent of the flooded area. From the findings, Madhepura was the most impacted district, with overall 46.2% of the area experiencing flooding, particularly affecting barren areas the most at 57.6%. Consequently, Madhepura should be designated for limited occupancy purposes due to its classification as a high-risk zone. Areas with a moderate risk zone, particularly Saharsa (34.9% inundation) and Supaul (34% inundation), can adopt a “living with floods” approach for effective management that includes improving the drainage network and decreasing impervious surfaces. The government should formulate a flood insurance policy targeting residents in flood-prone zones, providing coverage for losses and damages associated with flooding. Enhancing environmental management in the affected landscapes requires an understanding of historical LULC changes, patterns, and trends. A persistent global challenge is the constant alteration of the natural environment required to support urbanization, population growth, and the unrelenting demand for environmental resources. This work has demonstrated the numerous spatial–temporal LULC alterations that occurred in the flood-affected areas of Saharsa, Supaul, and Madhepura. The study also predicted future changes using the CA-Markov model of TerrSet’s Land Change Modeler and what would happen to land classes in 2030 if flooding occurred. The findings of the forecast reiterated that there would be remarkable variations in land use in these districts, with built-up areas projected to increase by 2030. The built-up area (10.32% increase) in Madhepura has already grown, which could lead to more severe flooding by 2030. The reason behind this is the fast-paced urbanization, mainly driven by the proliferation of artificial structures and the rise in population density. It is expected that the agricultural and vegetation regions (4.57% increase) in Supaul will expand by 2030. This expansion will have the added benefit of reducing the severity of floods in the area, as increased vegetation cover allows water to seep into the ground and makes the area less susceptible to flooding. By the year 2030, some portion of the desolate area of Saharsa will have been transformed into a built-up region (2.70% increase). This could potentially lead to a decrease in its natural drainage capacity, and if flooding were to occur in the future, it could result in an increase in water flow. Policymakers and planners need to evaluate the disparity between the current situation and anticipated changes, as ongoing land use conversions will continue to minimize the amount of land cover (barren land, agricultural land, and vegetative area) and replace it with built-up regions. This study utilized a GIS-based approach to generate maps that predict how land use and land cover (LULC) will change in the future. The maps can be utilized in future studies to investigate the repercussions of the expansion or reduction of different land classes on flooding. Additionally, the maps can aid in urban planning as a way to prevent or reduce the effects of floods and other types of disasters. Whereas the effectiveness, potential, and capacity of SAR imagery in tracking floods, precisely pinpointing the affected area and flood duration through multiple SAR data coverages, have also been confirmed by this study. This affirms the value of SAR data in water-related research, particularly in delineating and analyzing flood water. It is anticipated that the identification of flood-prone regions and their characteristics will serve as useful inputs for future flood modeling and analysis, making it beneficial for flood risk reduction planning and flood disaster management.