Resolution effects on ox-bow lake mapping and inundation consistency analysis in moribund deltaic flood plain of India

doi:10.21203/rs.3.rs-2619802/v1

Research on investigating spatial resolution effect on image-based wetland mapping was done and found finer resolution image has more efficiency but how is it useful for spectral water indices based delineation of partially vegetation shaded ox-bow lake, a cut-off channels of a river, and mapping inundation frequency? In order to obtain these answers, the present study used Landsat and Sentinel images and computed spectral indices like Normalized differences water index (NDWI), Modified NDWI (MNDWI), Re-modified NDWI (RmNDWI) and ensemble Vegetation inclusive aggregated water index (ViAWI). For obtaining inundation consistency character, water presence frequency (WPF) approach was adopted. A set of accuracy matrices was applied for validating the resolution effect.

The study revealed that finer resolution sentinel images are more effective in ox-bow lake mapping and characterising inundation frequency but it was not significantly better. Accuracy difference between them was very minimum. Therefore, the study recommended that in a Sentinel image sparse condition, Landsat images could alternatively be used. Among the used indices MNDWI was found suitable for ox-bow lake mapping but vegetation-shaded ox-bow lake was mapped using ensemble ViAWI. Inundation frequency analysis exhibited that about 70% area is consistent in water presence and therefore be hydro-ecologically viable and economically remunerative.

Ox-bow Lake

Spatial resolution

Landsat and Sentinel images

Spectral indices

Accuracy matrices and Inundation frequency

Ox-bow lakes are the meander cuts of a river behave like a wetland of immense ecological significance (Mukherjee et al., 2018). It is usually found in the floodplain river reaches at different spatial extent but often the small-scale ox-bow lakes, a useful nutrient source of the local people is often neglected from both inventorying and consequent any planning. Due to lake of any such planned initiatives, these resources have been destroying over time since long back (Talukdar et al., 2020). Pal and Talukdar (2018) reported that many of the ox-bow lakes in Indian deltaic floodplain were completely reclaimed majorly due to agricultural and built-up encroachment. Increasing population density, and intensified anthropogenic perturbations are the dominant causes of diminishing ox-bow lake area (Fei et al., 2016) and its eco-hydrological functions (Talukdar et al., 2019). The conflict between man and ox-bow lakes has been aggravating due to lack of inventorying of those resources (Debanshi et al., 2022). So, it is highly required to properly delimit the ox-bow lakes. For doing that with precision, coarser resolution image data often creates an obstruction.

In the floodplain region, fertile soil, high moisture content attracts people to agriculture. This practice has been aggravating over time. Both horizontal expansion and intensification are frequently caused by transmitting of fertilizer and other agricultural wastes in the wetland which induce the growth of floating vegetation over the water surface. This floating vegetation creates an obstruction for oxbow lake mapping. How these vegetation-shaded water body could be delineated is a challenging task. For water body mapping usually supervised image classification using different decision rules was very common practice. For example, Wetland mapping is done with the help of context-based decision-making rules (Du et al., 2016). Supervised classification using a set of decision rules helps determine land use classification of wetlands (Rozenstein et al., 2011). Use of various machine learning and afterwards deep learning algorithms for the same was done to improve the classification. Land use and land cover change are monitored with the help of fuzzy decision-making rules (Feizizadeh et al., 2021). Wetlands area is delineated with the help of machine learning methods using a decision tree and rule-based process (Günen et al., 2022). From their works, machine learning processes often provided better scope for deriving good results. A comparison of the literature dealing with different source images with varying spatial resolution also revealed that finer resolution-based classification is more authentic (Drǎguţ et al., 2010). Another group of scholars tried to map water body using spectral indices from remote sensing products. For instance, multi-spectral satellite images are best for high spatial and temporal resolution water body mapping (Lu et al., 2011). Different water body mapping methods like NDWI and MNDWI have been developed to delineate water bodies from multispectral images (Du et al., 2016). These indices are not well equipped to delimit wetland shaded with floating vegetation. Even they did not judge the spatial resolution effect on the spectral indices. Expectedly, there is a high likelihood of having better results in finer resolution image-driven indices. However, there is a scope for proving the fact with ground evidence. The present work, therefore, tried to devise a spectral index for vegetation-shaded water body delineation and measure the resolution effect in mapping accuracy.

Moreover, in floodplain wetlands in subtropical climate, the inundation regime of water appearance is highly asymmetric. Since annual and seasonal rainfall pattern is highly skewed, areal extent of water body also varies accordingly. Mondal and Pal (2018) documented that areal extent of water body area in the seasonally inundated floodplain wetland dynamically changes by the amount of rainfall and its periodic concentration. Due to such a highly erratic nature, these wetlands are often neglected but these are ecologically valuable (Ennabili et al., 2021). For scientifically mapping the consistency of water present in a period. Borro et al., (2014) applied water presence frequency (WPF) approach. This approach counts how many years water has appeared in a pixel of the water body to the total length of a year (Borro et al., 2014). Highly consistent inundation means out of total year count, in maximum years, water was marked in a pixel and vice-versa in the case of inconsistent inundation frequency. A good many researchers adopted this approach in their field. But no one tried to investigate the resolution effect on inundation behaviour analysis which the present work attempted to do.

From the above literature review part, it is very evident fact that a good many researchers applied spectral water indices (Zhou et al., 2017; Pal et al., 2020; Paul et al., 2020), decision rules (Labib et al., 2018; Bijeesh et al., 2020; Wei et al., 2020), machine learning (Acharya et al., 2019; Mallick et al., 2021) and deep learning (Wang et al., 2020; Li et al., 2021) algorithm for defining wetland. In order to obtain better results in classification of images, Muster et al (2013) applied advanced techniques for water body mapping and they proved that in all environments, coarser resolution images are more potential for providing precise wetland maps. Although a comparison of wetland maps produced from the image indices with varying resolutions was not done so far, but a comparison of literatures on image-driven water indices done with images of diverse resolutions proved that a finer resolution image has better efficacy for precise wetland map (López-Tapia et al., 2021). Agriculture-dominated ox-bow lakes are often covered by both endemic and exotic floating vegetation (Shan et al., 2021) and it causes hindrance to water body mapping from image (Li et al., 2022). Researchers yet not investigated deeply how to demarcate vegetation shaded ox-bow lake and effect of resolution on it. Considering this research gap, the first objective of this present work is to assess spatial resolution effect on ox-bow lakes delineation shaded by floating vegetation cover using waterbody extracting indices from remote sensing images.

In a floodplain region, where inundation behaviour (frequency and magnitude) in a wetland is highly irregular based on rainfall variability, scientifically mapping wetland area is a difficult task (Onojeghuo et al., 2021). In a year due to having greater rainfall wider part of the wetland domain may be inundated but in the immediate next year its extent may be shrinked (Vera-Herrera et al., 2021). So, the proper distinction between consistently and inconsistently inundated wetland is highly necessary to comprehend the spatial extent and hydrological behaviour of a wetland. For this analysis, the role of the resolution effect was not assessed so far. Second objective of this work is to assess the resolution effect on inundation frequency analysis of the ox-bow lakes.

Indian Ganges moribund delta covers major parts of 24 Pargana, Nadia and Murshidabad covering about 6642.58 km² of geographical areas (Bagchi and Mukherjee., 1983) (Fig. 1). This part is characterized by numerous numbers of dead channels, ox-bow lakes, river scours and marshy lands. Frequent in filing and shifting of channels (Bhagirathi-Hooghly, Jalangi, Mathabhanga, Churni, Ichamati etc.) is the main cause behind origin of this deltaic land (Bagchi 1944). Hydrological modifications caused by natural and anthropogenic factors and growing human pressure are the serious causes behind identity crisis of the wetland and qualitative transformation of the wetland in general and ox-bow lakes in particular. So, inventorying such resources is very essential. Considering the decaying trend of such resources, moribund deltaic region was taken as a study unit. Existing ox-bow lakes are also highly dynamic in nature in relation to seasonal rainfall regime of this sub humid monsoon climate. Annual average precipitation of this study area ranges from 1160 mm to 1180 mm, out of which more than 75% happens during the monsoon season (June to September).

Landsat OLI and Sentinel 2A (2017 to 2021) were taken into account. Satellite images were collected for the post-monsoon (October to January) season from the US Geological Survey (USGS). Sentinel images of the same years were taken from 2017 to 2021. A detail list of image acquisition dates, resolutions was mentioned in Table 1. Arc Gis was deployed for mapping ox-bow lakes and inundation frequency analysis. To assess the accuracy of the ox-bow lake maps prepared using spectral indices we have used google earth images.

Table 1

Satellite images details
Year	Landsat 8				Sentinel
	Sensor	Resolution (m)	Path/Row	Date of Acquisition	Sensor	Bands	Resolution (m)	Tiles Name	Date of Acquisition
2017	OLI	30	138/43–44	08-11-2017	2A			QXF, QXG, RXH	01-12-2017
			139/43	01-12-2017
2018			138/43–44	11-11-2018				QXF, QXG, RXH	20-11-2018
			139/43	17-10-2018
2019			138/43–44	14-11-2019		1–12	10–60	QXF, QXG, RXH	21-11-2019
			139/43	05-11-2019
2020			138/43–44	16-11-2020				QXF, QXG, RXH	15-11-2020
			139/43	07-11-2020
2021			138/43–44	19-11-2021				QXF, QXG, RXH	31-10-2021
			139/43	10-11-2021

4.1. Ox-bow Lake mapping using conventional spectral indices

Five spectral water indices like Water Index (NDWI) of McFeeters (1996), Modified Normalized Difference Water Index (MNDWI) of Xu (2006), Re-modified Normalized Difference Water Index (RmNDWI) of Debanshi and Pal (2020), Normalized Difference Vegetation Index (NDVI) of Townshend and Justice (1986) and Vegetation inclusive Aggregated Water Index (ViAWI) of Ghosh and Pal (2023) were applied in order to map ox-bow lakes partly covered with floating vegetation. Each index has some merits and demerits. NDWI maps open water more clearly but often mixed with built area (Zhou et al., (2017). MNDWI sometimes overestimates water bodies (Li et al., 2022). RmNDWI in better way capable to segregate water body from built area (Debanshi and Pal 2020). ViAWI is capable to map water body even covered with floating vegetation (Ghosh and Pal, 2023). Equations of all the used indices are mentioned in Eq. 1 to 5. These indices were applied both in the case of Landsat and Sentinel image in order to justify the resolution effect on ox-bow lake mapping.

$$NDVI=\frac{{bNIR - bred}}{{bNIR+bred}}$$

1

$$NDWI=\frac{{{b_{green}} - {b_{NIR}}}}{{{b_{green}}+{b_{NIR}}}}$$

2

$$MNDWI=\frac{{{b_{green}} - {b_{MIR}}}}{{{b_{green}}+{b_{MIR}}}}$$

3

$$RmNDWI=\frac{{{b_{red}} - {b_{MIR}}}}{{{b_{red}}+{b_{MIR}}}}$$

4

$$ViAWI=\frac{{NDWI+MNDWI+RmNDWI+NDVI}}{{No{._{}}o{f_{}}Indices}}$$

5

where b_green denotes brightness value of Green band, b_red denotes brightness value of Red band, b_NIR denotes brightness value of Infra-red band, b_MIR denotes brightness value of middle Infra-red band.

4.2 Comparing the accuracy of ox-bow lake maps based on validation approach

In case of validation of the existing spectral indices, different methods like Kappa co-efficient (Eq. 6), Area under curve in Receiver Operating Characteristics (AUC-ROC) Curve, and confusion matrices (True Positive Rate (TPR) (Eq. 7), True Negative Rate (TNR) (Eq. 8), Positive Predictive Value (PPV) (Eq. 9), F1 Score (Eq. 10), Matthews Correlation Coefficient (MCC) (Eq. 11)) were applied. Total 250 sites from higher resolution Google earth image and 70 field sites were chosen for validating the ox-bow lake maps prepared based on spectral indices.

TPR is plotted against FPR in case of ROC and the value ranges from 0–1. This can be classified into some subclasses like fail, which ranges from 0.50–0.60; poor, which ranges from 0.60–0.70; fair, which ranges from 0.70–0.80; good, which ranges from 0.80–0.90; and excellent, which ranges from 0.90-1.00 following Rasyid et al. (2016).

The Kappa statistics (K) was also used for the same purpose for assessing the classification performance. K value ranges from 0–1. It can be classified into some sub classes according to the degree of accuracy like very poor (< 0.40), poor (0.4–0.55), good (0.55–0.70), very good (0.70–0.85) and excellent (> 0.85) Monserud and Leemans (1992). F1 score and MCC method is applied to correctly measure the accuracy of map (Chicco et al., 2021).

$$Kapp{a_{}}Statistics(K)=\frac{{(TS \times TSC) - \sum {CT \times RT)} }}{{T{S^2} - \sum {(CT \times RT)} }} \times 100$$

6

where TS is the Total Reference pixel, TSC is the Total Corrected Classified pixel, CT is the Colum total and RT is the Row total

$$TPR=\frac{{TP}}{{TP+FN}}$$

7

$$TNR=\frac{{TN}}{{TN+FP}}$$

8

$$PPV=\frac{{TP}}{{TP+FP}}$$

9

$$F1=\frac{{2TP}}{{2TP+FP+FN}}$$

10

$$MCC=\frac{{TP*TN - FP*FN}}{{\sqrt {(TP+FP)(TP+FN)(TN+FP)(TN+FN)} }}$$

11

where TP means True positive, TN means True negative, FP means False positive, FN means False negative

4.3 Histogram comparison approach (HCA) and scattered plot approach (SPA) for comparing spectral indices of two image sets

For this histogram of different spectral and ensemble, indices were developed and compared to the result yielded from Landsat and Sentinel images. Distribution patterns will help to understand the difference or similarities between two image products. Similarity can encourage to adopt any one of the products for ox-bow lake mapping. A scattered plot was made between two image products in reference to each spectral index. Nature of scattered plots will help to understand how they are associated. A high association can help to decide whether any one of the products could be taken into account.

4.4 Water presence frequency framework for consistency analysis

Inundation frequency analysis of the ox-bow lake map of each year (2017 to 2021) driven from both Landsat and Sentinel images was done using water presence frequency (WPF) (Eq. 12) framework of Borro et al (2014) and subsequently applied by Pal et al., (2022), Saha et al (2022). WPF refers to the water presence frequency of each wetland pixel in a specific period. It is computed by rationing the frequency of water presence to the total number of years considered in each pixel. WPF value ranges from 0–1, where 1 signifies water appeared in each pixel in all the years considered (consistent inundation) and a value near to 0 signifies inconsistent water presence. Consistency of inundation indicates regular water support from the source and thereby, hydrologically and ecologically more viable (Pal et al., 2021). In floodplain regions, this WPF framework is very useful.

$WP{F_j}=\frac{{\sum\limits_{{i=1}}^{n} {{I_j}} }}{n} \times 100$ $WP{F_j}=\frac{{\sum\limits_{{i=1}}^{n} {{I_j}} }}{n} \times 100$

(12)

5.1 Ox-bow Lake maps

Ox-bow Lake maps derived through the NDWI, MNDWI, RmNDWI and ViAWI from both Landsat and Sentinel images are not uniform in their areal extents (Figs. 2,3,4 and 5). Similarly, geolocation of the derived ox-bow lakes is not absolutely uniform in all the indices. For instance, the maximum area of each ox-bow lake was found in the case of ViAWI followed by MNDWI. Not only in terms of their geolocation and areal extent, different spectral indices revealed diverse results but also the area of the same index is varied in the case of Landsat and Sentinel images due to their varied resolution. Most of the cases, the areal extent of the ox-bow lakes was found in the case of finer resolution Sentinel image-driven spectral indices (Table 2). However, the areal extent doesn’t confirm the accuracy of the indices and hence demands an accuracy test. It may help to be confirmed which index is the best representative and is resolution matters while mapping ox-bow lake.

Table 2

Seasonal areal dynamics of the ox-bow lake in different resolution images
Year	Satellite Series	Used Indices Area (km²)
Year	Satellite Series	NDWI	MNDWI	RmNDWI	ViAWI
2017	Landsat	14.10	22.18	18.98	28.62
2017	Sentinel	15.15	23.01	19.31	28.68
2019	Landsat	12.46	20.06	16.98	28.61
2019	Sentinel	13.44	22.15	17.72	28.70
2021	Landsat	16.02	23.40	21.08	28.61
2021	Sentinel	16.14	24.03	20.84	28.69

5.2 Analyzing resolution effect on mapping using validation approaches

Kappa coefficient, AUC-RoC, TPR, FNR, PPV, F1, and MCC were computed for all the spectral indices derived from Landsat and Sentinel images and tabulated in Table 3. It was computed for the years 2017, 2019 and 2021 in order to obtain more confirmation about the representativeness of the indices. All the accuracy matrices always not figure out a single spectral index as the best suited. However, in most of the cases, amongst the individual spectral indices, MNDWI was found as the best representative both in Landsat and Sentinel image-driven indices (Table 3). Ensemble spectral index (ViAWI) was recognized as a more fit for floating vegetation cover ox-bow lakes. In the case of MNDWI, AUC value was around 0.7, while it was 0.85 (Table 3). Since a larger part of the ox-bow lake was shaded with floating vegetation, the ensemble index was found more prominent than the individual spectral water index. Comparison of accuracy values between Landsat and Sentinel image-driven spectral indices and ensemble index showed a slightly enhanced accuracy in the later case signifying the fact that finer resolution image produces better output. For instance, MNDWI values were 0.66, 0.64, and 0.65, respectively in 2017, 2019 and 2021 in the case of Landsat image-driven products, and it was 0.69, 0.73 and 0.66 in same time orders. AUC values ranged from 0.83–0.84 in the case of MNDWI and 0.93 in the case of ViAWI in reference to Landsat output, and it was 0.83–0.87 in the former case and 0.93 in the later case in reference to the Sentinel image-based output. The comparative view clarifies the fact that although finer resolution Sentinel image produced slightly better result while delineating ox-bow lake, Landsat-based product in reference to the Sentinel is not obsolete.

Table 3

Accuracy matrices of different spectral indices used in order to highlight the difference between Landsat and Sentinel products
Validation Methods	Satellite Series	2017					2019					2021
Validation Methods	Satellite Series	NDWI	MNDWI	RmNDWI	ViAWI	Floating Vegetation	NDWI	MNDWI	RmNDWI	ViAWI	Floating Vegetation	NDWI	MNDWI	RmNDWI	ViAWI	Floating Vegetation
Kappa Co-efficient	Landsat	0.48	0.66	0.59	0.87	0.70	0.40	0.64	0.59	0.87	0.64	0.29	0.65	0.65	0.81	0.76
Kappa Co-efficient	Sentinel	0.55	0.69	0.61	0.88	0.74	0.40	0.73	0.61	0.88	0.71	0.52	0.66	0.65	0.88	0.80
AUC- ROC	Landsat	0.75	0.84	0.79	0.93	0.85	0.71	0.83	0.80	0.93	0.82	0.66	0.84	0.84	0.93	0.88
AUC- ROC	Sentinel	0.79	0.85	0.82	0.93	0.86	0.76	0.87	0.82	0.93	0.86	0.77	0.83	0.83	0.93	0.90
TPR	Landsat	0.52	0.75	0.67	1.00	0.80	0.45	0.72	0.66	1.00	0.79	0.36	0.78	0.71	1.00	0.84
TPR	Sentinel	0.60	0.78	0.67	1.00	0.81	0.46	0.81	0.67	1.00	0.82	0.57	0.80	0.78	1.00	0.89
TNR	Landsat	0.99	0.94	0.96	0.86	0.92	0.98	0.95	0.95	0.86	0.87	0.97	0.94	0.97	0.86	0.92
TNR	Sentinel	0.99	0.94	0.97	0.87	0.95	0.98	0.94	0.98	0.87	0.90	0.98	0.87	0.89	0.87	0.92
PPV	Landsat	0.98	0.94	0.95	0.90	0.93	0.93	0.89	0.95	0.90	0.89	0.93	0.94	0.96	0.90	0.88
PPV	Sentinel	0.99	0.94	0.97	0.91	0.96	0.96	0.95	0.97	0.91	0.92	0.98	0.89	0.90	0.91	0.94
F1	Landsat	0.68	0.83	0.79	0.95	0.86	0.61	0.82	0.78	0.95	0.84	0.51	0.83	0.81	0.95	0.86
F1	Sentinel	0.74	0.85	0.79	0.95	0.88	0.62	0.87	0.79	0.95	0.87	0.72	0.84	0.84	0.95	0.91
MCC	Landsat	0.36	0.50	0.45	0.72	0.54	0.30	0.48	0.44	0.72	0.51	0.23	0.50	0.49	0.72	0.52
MCC	Sentinel	0.41	0.53	0.46	0.72	0.56	0.30	0.56	0.45	0.72	0.55	0.39	0.52	0.51	0.72	0.63

5.3 Comparison of two products using HCA and SPA

Figure 6 shows the pixel frequency under different ViAWI classes derived from Landsat and Sentinel images. Figure prominently shows that in all the years, the ViAWI value is significantly higher in the case of Sentinel-driven ViAWI images than Landsat-driven images. The distribution pattern shows an almost normal distribution of pixel value in sentinel-driven ViAWI, and it is negatively skewed in Landsat driven image. Negatively skewed means, frequency of low ViAWI value is much greater. Research also revealed that intensity of the spectral value often is associated with water depth (Talukdar and Pal., 2019). So, pixels with shallower depth are much greater than mean depth. But the situation is quite different in sentinel image. Here majority of pixels are concentrated around mean value. Since the pictures are different in two ViAWIs, it needs further investigation for establish the truth. Ground truth information based accuracy assessment is the main way. Departure of ViAWI value of each pixel from the mean also reveals the uniform distribution pattern as ViAWI. More departure is found in the majority of the pixels in Landsat-driven ViAWI. So it may likely indicate a high variation of water depth. This variation is not so high in the case of sentinel-driven ViAWI. Although depth measurement and comparison from both indices was not the objective of this current work, but field investigation revealed that the depth data is more closely associated with sentinel data.

A scattered plot between ViAWI driven from Landsat and Sentinel images in the years 2017, 2019 and 2021 revealed that there is a strong association between ViAWI driven from both the images. Since the indices value is more in the case of the sentinel image, distribution of the scattered points are more upright toward the Y axis which represents the ViAWI of sentinel image (Fig. 7). Almost similar trend is also noticed in co-variance plot of ViAWIs derived from both the images. This pattern signifies that the degree of variance is lesser in the case of Landsat images.

5.4 Inundation consistency characters

Figure 8 displays the inundation frequency maps of the ox-bow lakes executed from both Landsat and Sentinel images using WPF approach. WPF map was classified into five subclasses which are very low (< 20%), low (20–40%), moderate (40–60%), high (60–80%) and very high (> 80%). Higher frequency indicates the consistency of water appearance in the considered years in the post-monsoon season as considered here. Table 4 shows the area under different WPF zones in order to understand the proportion of area under different WPF or inundation consistence classes. About 70 to 72% area was identified under high to very high consistent WPF zones indicating greater consistency over wider area of the ox-bow lakes and hence, hydro-ecologically more viable. Very low and low WPF zones appeared in the ox-bow lakes’ fringe area water depth is lower and human intervention is quite greater. High and very high WPF zones were identified majorly in the middle part of the ox-bow lakes and these were sometimes recognized even at the fringe area. Comparison between inundation frequency maps of both Landsat and Sentinel products did not provide any significant variation. About 2% more area was recognized under higher WPF classes in the case of Sentinel image over Landsat. No separate validation approach was used for this purpose since the individual yearly ox-bow lake maps were validated. Since the Sentinel image-based spectral indices showed slightly better accuracy, therefore, the WPF map produced from the same could be treated as the representative.

Table 4

Area under different WPF zones computed based on Landsat and Sentinel images
WPF classes (%)	Landsat		Sentinel
WPF classes (%)	Area (km²)	Area (%)	Area (km²)	Area (%)
< 20	1.63	6.21	1.63	6.06
20–40	2.07	7.87	2.08	7.73
40–60	4.24	16.10	3.93	14.61
60–80	5.63	21.42	5.88	21.86
> 80	12.73	48.39	13.38	49.74
Total	26.30	100	26.90	100

Precise inventorying ox-bow lakes is very necessary since it is a major ecosystem service provider in a rural ecological and economic system. In this perspective, scientific delineation of ox-bow lakes is very important. Many of the previous literatures established that spatial resolution is a major concern in inventorying wetlands (Banks et al., 2019; Johnson et al., 2022). Cavallo et al (2021) and Long et al (2021) while working on wetland delineation in Flood plain regions, reported that finer resolution Sentinel (2) image was more effective. Not only in case of wetland inventorying, the resolution effect was also established in the case of image-driven multi-parametric flood modelling (Pal et al., 2021), flood simulation (Wang et al., 2018) etc. Since finer resolution revealed the object more prominently, such results were recorded. In the present finding, the same trend was noticed but it was so prominent. Relatively stepper bank slope and deeper water body may be the cause behind such less resolution effect in ox-bow lake mapping. In this study only two satellite products were taken, but the incorporation of other images with more resolution differences may produce the resolution effect more prominently.

Amongst the used spectral indices, MNDWI was found as the best representative but the presence of aquatic vegetation over the ox-bow lakes often obstructs mapping with existing spectral indices used in this study. Ensemble ViAWI represented the ox-bow lakes in a precise manner, including the floating vegetation. It is a long-standing problem in the agriculturally dominated floodplain wetlands in general and ox-bow lakes in particular. Deposition of agricultural residues encourages the growth of floating vegetation particularly water hyacinth (Ajithram et al., 2021). It creates a barrier to precise inventorying ox-bow lakes (Yamazaki et al., 2015). Ensemble ViAWI did not show any resolution effect. Accuracy level was found uniform both in Landsat and Sentinel based products. It does signify that the floating vegetation mapping using NDVI was uniformly and efficiently done from both images.

Inundation frequency analysis is a reliable approach for water presence consistency analysis in seasonally inundated floodplain wetlands (Das et al., 2017), mainly controlled by irregular rainfall behaviour (Talukdar et al., 2020). Many a scholar like Du et al. (2018) applied this technique in order to focus on the consistency character of the water presence of a wetland. Sometimes, this technique is also used for delineating wetlands. In this case, highly infrequent water pixels are used to exclude from wetland map. In the present case, since the study aimed to delineate ox-bow lake with typical morphological characters, this technique was not used to assess the delineation of ox-bow lake rather used it for assessing the resolution effect on inundation consistency characters. As like ox-bow lake mapping, a finer resolution Sentinel image was found more appropriate for this analysis than a coarser resolution Landsat image but the effect is not too prominent. WPF analysis revealed the fact that a wider part (70% or more) of the ox-bow lakes was characterised by high to very high-water presence frequency and therefore, more reliable for fishing services from this natural capital. Rest about 30% with infrequent water presence could be taken into consideration for immediate restoration to enhance the benefits.

Spectral indices value is often used for calibrating water depth (Khatun et al., 2021) based on field data. Field data and image-driven spectral values are used for computing co-efficient of water depth and the computed coefficient is used for calibrating water depth (Kundu et al., 2021). So, spectral indices value derived from Landsat and Sentinel images is a matter of concern. HCA and SPA analysis exhibited that there is a similar trend of indices value derived from both the images, but the projected trend line did not show 1:1 ratio and hence a question comes from which image products would be used for calibrating water depth. Reflectance character in signal-receiving bands used for Landsat and Sentinel images is the major cause behind such differences. The present study raised the question but did not answer it. Field depth data-based validation could solve this. Hence further research is needed in this regard. Similarity in spectral indices value-based histogram of two products showed that both the images have almost similar capability of delineating ox-bow lakes as well as inundation frequency. HCA was used by Pal and Sarda (2021) for comparing the image character in order to obtain the similarity between two bands of an image of the same bands of two images.

Since the present study explored the spatial resolution effect on spectral indices for determining the ox-bow lakes, in particular, it could be treated as a new knowledge addition in this state-of-arts. Moreover, no one so far has explored the effect of spatial resolution for analysing inundation frequency as the present study was done. How the resolution effect can create impacts on spectral indices-based calibration is unanswered that needs to be answered with further research. Besides this, for inundation analysis, total analysis period was very narrow, which could be enhanced where data is available.

The result clearly revealed that Sentinel image (finer resolution) based product is slightly better performed while delineating ox-bow lakes and making inundation frequency maps but the no significant difference in accuracy was found between Landsat and Sentinel image based spectral indices. So, for ox-bow lake mapping, Landsat images could be alternatively used in a Sentinel images sparse situation. Previous literatures mostly recommended resolution is a major issue for wetland mapping. The present finding is not any exception, but while comparing Landsat and Sentinel products, a considerably significant difference was noticed, particularly while mapping ox-bow lakes and inundation frequency analysis. Among the spectral indices used for ox-bow lake mapping, MNDWI was found as the best suited, and ViAWI was found as the best representative method for delineating the vegetation shaded ox-bow lakes. It was found true for both the image products. Moreover, HCA and SPA also exhibited that there is a symmetric pattern in spectral indices values, and therefore, these could be treated as compatible. So, spectral indices-based calibration, whenever needed, any one of the images could be consideration for mapping the ox-bow lakes. Here, it could be mentioned that this may not be true for all kinds of wetland mapping, inundation frequency and calibration works.

The spatial resolution effect found in ox-bow lake mapping and inundation frequency analysis is a new dimension. A good many works focused on resolution effect on wetland mapping as a whole but finding in reference to ox-bow lakes in particular and inundation characters were presented newly. Although the study carried out HCA and SPA for association analysis between Landsat and Sentinel products, but did not conduct any spectral indices-based calibration in order to present the real ground situation. Moreover, while the inundation frequency analysis, the considered study length was narrower. But a long period of time could explore the inundation character more clearly.

Both the images clearly delineated the ox-bow lakes and their consistency of inundation; this could be useful for ecological planning since ox-bow lakes are a reliable biodiversity hotspot and major nutrient source of rural livelihood in India.

Ethical approval and consent to participate: Not applicable.

Consent to publish: All the co-authors agreed to publish the manuscript.

Author’s Contribution: Conceptualization, Swades Pal and Ripan Ghosh; Formal analysis, Swades Pal and Ripan Ghosh; Methodology, Swades Pal, and Ripan Ghosh; Software, Ripan Ghosh; Supervision, Swades Pal; Validation: Ripan Ghosh; Writing – original draft, Swades Pal and Ripan Ghosh; Writing – review & editing, Swades Pal, and Ripan Ghosh.

Competing interests: The authors declare that they have no conflict of interests.

Availability of data and materials All the data and materials related to the manuscript

Are published with the paper, and available from the corresponding author upon request ([email protected]).

Funding: There is no funding source to declare. Not applicable.

Acknowledgement:

For this study the corresponding author thankful to the USGS for providing satellite imagery. The corresponding author of the article would like to thank the UGC, New Delhi, India for giving economical help as a junior research fellowship to carry on the research work given in this paper. Special thanks are extended to Manabendra Let and Susmita Ghosh for cooperating authors in different purposes.

Acharya, T. D., Subedi, A., & Lee, D. H. (2019). Evaluation of machine learning algorithms for surface water extraction in a Landsat 8 scene of Nepal. Sensors, 19(12), 2769.
Ajithram, A., Jappes, J. W., & Brintha, N. C. (2021). Water hyacinth (Eichhornia crassipes) natural composite extraction methods and properties–A review. Materials Today: Proceedings, 45, 1626-1632.
Bagchi, K. (1944). The ganges delta. University of Calcutta.
Bagchi, K., & Mukerjee, K. N. (1983). Diagnostic survey of West Bengal (s). Department of Geography, Calcutta University, Pantg Delta & Rarh Bengal, 42(58), 17-19.
Banks, S., White, L., Behnamian, A., Chen, Z., Montpetit, B., Brisco, B., ... & Duffe, J. (2019). Wetland classification with multi-angle/temporal SAR using random forests. Remote Sensing, 11(6), 670.
Bijeesh, T. V., & Narasimhamurthy, K. N. (2020). Surface water detection and delineation using remote sensing images: A review of methods and algorithms. Sustainable Water Resources Management, 6(4), 1-23.
Borro, M., Morandeira, N., Salvia, M., Minotti, P., Perna, P., & Kandus, P. (2014). Mapping shallow lakes in a large South American floodplain: a frequency approach on multitemporal Landsat TM/ETM data. Journal of Hydrology, 512, 39-52.
Cavallo, C., Papa, M. N., Gargiulo, M., Palau-Salvador, G., Vezza, P., & Ruello, G. (2021). Continuous monitoring of the flooding dynamics in the Albufera Wetland (Spain) by Landsat-8 and Sentinel-2 datasets. Remote Sensing, 13(17), 3525.
Chang, L., Cheng, L., Huang, C., Qin, S., Fu, C., & Li, S. (2022). Extracting urban water bodies from Landsat imagery based on mNDWI and HSV transformation. Remote Sensing, 14(22), 5785.
Chicco, D., Tötsch, N., & Jurman, G. (2021). The Matthews correlation coefficient (MCC) is more reliable than balanced accuracy, bookmaker informedness, and markedness in two-class confusion matrix evaluation. BioData mining, 14(1), 1-22.
Das, R. T., & Pal, S. (2017). Exploring geospatial changes of wetland in different hydrological paradigms using water presence frequency approach in Barind Tract of West Bengal. Spatial Information Research, 25(3), 467-479.
Debanshi, S., & Pal, S. (2020). Wetland delineation simulation and prediction in deltaic landscape. Ecological indicators, 108, 105757.
Debanshi, S., & Pal, S. (2022). Assessing the role of deltaic flood plain wetlands on regulating methane and carbon balance. Science of The Total Environment, 808, 152133.
Drǎguţ, L., Tiede, D., & Levick, S. R. (2010). ESP: a tool to estimate scale parameter for multiresolution image segmentation of remotely sensed data. International Journal of Geographical Information Science, 24(6), 859-871.
Du, J., Kimball, J. S., Galantowicz, J., Kim, S. B., Chan, S. K., Reichle, R., ... & Watts, J. D. (2018). Assessing global surface water inundation dynamics using combined satellite information from SMAP, AMSR2 and Landsat. Remote sensing of environment, 213, 1-17.
Du, Y., Zhang, Y., Ling, F., Wang, Q., Li, W., & Li, X. (2016). Water bodies’ mapping from Sentinel-2 imagery with modified normalized difference water index at 10-m spatial resolution produced by sharpening the SWIR band. Remote Sensing, 8(4), 354.
Ennabili, A., Libiad, M., & Khabbach, A. (2021). Importance of Wetlands in Maintaining the Richness of Morocco’s Vascular Flora. Wetlands, 41(8), 1-10.
Fei, L., Shuwen, Z., Jiuchun, Y., Kun, B., Qing, W., Junmei, T., & Liping, C. (2016). The effects of population density changes on ecosystem services value: A case study in Western Jilin, China. Ecological indicators, 61, 328-337.
Feizizadeh, B., Mohammadzade Alajujeh, K., Lakes, T., Blaschke, T., & Omarzadeh, D. (2021). A comparison of the integrated fuzzy object-based deep learning approach and three machine learning techniques for land use/cover change monitoring and environmental impacts assessment. GIScience & Remote Sensing, 58(8), 1543-1570.
Ghosh, R., & Pal, S. (2023). Delineation of vegetation shaded ox-bow lakes in Ganges flood plain, India. Ecological Informatics, 74, 101954.
Günen, M. A. (2022). Performance comparison of deep learning and machine learning methods in determining wetland water areas using EuroSAT dataset. Environmental Science and Pollution Research, 29(14), 21092-21106.
Johnson, L. K., Mahoney, M. J., Bevilacqua, E., Stehman, S. V., Domke, G. M., & Beier, C. M. (2022). Fine-resolution landscape-scale biomass mapping using a spatiotemporal patchwork of LiDAR coverages. International Journal of Applied Earth Observation and Geoinformation, 114, 103059.
Khatun, R., Talukdar, S., Pal, S., & Kundu, S. (2021). Measuring dam induced alteration in water richness and eco-hydrological deficit in flood plain wetland. Journal of Environmental Management, 285, 112157.
Kundu, S., Pal, S., Talukdar, S., & Mandal, I. (2021). Impact of wetland fragmentation due to damming on the linkages between water richness and ecosystem services. Environmental Science and Pollution Research, 28(36), 50266-50285.
Labib, S. M., & Harris, A. (2018). The potentials of Sentinel-2 and LandSat-8 data in green infrastructure extraction, using object based image analysis (OBIA) method. European Journal of Remote Sensing, 51(1), 231-240.
Le, C., Zha, Y., Li, Y., Sun, D., Lu, H., & Yin, B. (2010). Eutrophication of lake waters in China: cost, causes, and control. Environmental management, 45(4), 662-668.
Li, A., Fan, M., Qin, G., Xu, Y., & Wang, H. (2021). Comparative analysis of machine learning algorithms in automatic identification and extraction of water boundaries. Applied Sciences, 11(21), 10062.
Li, W., Zhang, W., Li, Z., Wang, Y., Chen, H., Gao, H., ... & Wu, X. (2022). A new method for surface water extraction using multi-temporal Landsat 8 images based on maximum entropy model. European Journal of Remote Sensing, 55(1), 303-312.
Li, Z., Zhang, X., & Xiao, P. (2022). Spectral index-driven FCN model training for water extraction from multispectral imagery. ISPRS Journal of Photogrammetry and Remote Sensing, 192, 344-360.
Long, X., Li, X., Lin, H., & Zhang, M. (2021). Mapping the vegetation distribution and dynamics of a wetland using adaptive-stacking and Google Earth Engine based on multi-source remote sensing data. International Journal of Applied Earth Observation and Geoinformation, 102, 102453.
López-Tapia, S., Ruiz, P., Smith, M., Matthews, J., Zercher, B., Sydorenko, L., ... & Katsaggelos, A. K. (2021). Machine learning with high-resolution aerial imagery and data fusion to improve and automate the detection of wetlands. International Journal of Applied Earth Observation and Geoinformation, 105, 102581.
Lu, S., Wu, B., Yan, N., & Wang, H. (2011). Water body mapping method with HJ-1A/B satellite imagery. International Journal of Applied Earth Observation and Geoinformation, 13(3), 428-434.
Mallick, J., Talukdar, S., Pal, S., & Rahman, A. (2021). A novel classifier for improving wetland mapping by integrating image fusion techniques and ensemble machine learning classifiers. Ecological Informatics, 65, 101426.
McFeeters, S. K. (1996). The use of the Normalized Difference Water Index (NDWI) in the delineation of open water features. International journal of remote sensing, 17(7), 1425-1432.
Mondal, D., & Pal, S. (2018). Monitoring dual-season hydrological dynamics of seasonally flooded wetlands in the lower reach of Mayurakshi River, Eastern India. Geocarto International, 33(3), 225-239.
Monserud, R. A., & Leemans, R. (1992). Comparing global vegetation maps with the Kappa statistic. Ecological modelling, 62(4), 275-293.
Mukherjee, K., Pal, S., & Mukhopadhyay, M. (2018). Impact of flood and seasonality on wetland changing trends in the Diara region of West Bengal, India. Spatial Information Research, 26(4), 357-367.
Muster, S., Heim, B., Abnizova, A., & Boike, J. (2013). Water body distributions across scales: A remote sensing based comparison of three arctic tundra wetlands. Remote Sensing, 5(4), 1498-1523.
Onojeghuo, A. O., Onojeghuo, A. R., Cotton, M., Potter, J., & Jones, B. (2021). Wetland mapping with multi-temporal sentinel-1 &-2 imagery (2017–2020) and LiDAR data in the grassland natural region of alberta. GIScience & Remote Sensing, 58(7), 999-1021.
Pal, S., & Sarda, R. (2020). Damming effects on the degree of hydrological alteration and stability of wetland in lower Atreyee River basin. Ecological Indicators, 116, 106542.
Pal, S., & Sarda, R. (2021). Measuring the degree of hydrological variability of riparian wetland using hydrological attributes integration (HAI) histogram comparison approach (HCA) and range of variability approach (RVA). Ecological Indicators, 120, 106966.
Pal, S., & Sarda, R. (2021). Modelling water richness in riparian flood plain wetland using bivariate statistics and machine learning algorithms and figuring out the role of damming. Geocarto International, 1-24.
Pal, S., & Sarda, R. (2021). Modelling water richness in riparian flood plain wetland using bivariate statistics and machine learning algorithms and figuring out the role of damming. Geocarto International, 1-24.
Pal, S., & Singha, P. (2022). Image-driven hydrological parameter coupled identification of flood plain wetland conservation and restoration sites. Journal of Environmental Management, 318, 115602.
Pal, S., & Talukdar, S. (2018). Drivers of vulnerability to wetlands in Punarbhaba river basin of India-Bangladesh. Ecological Indicators, 93, 612-626.
Pal, S., Sarkar, R., & Saha, T. K. (2022). Exploring the forms of wetland modifications and investigating the causes in lower Atreyee river floodplain area. Ecological Informatics, 67, 101494.
Paul, S., & Pal, S. (2020). Exploring wetland transformations in moribund deltaic parts of India. Geocarto International, 35(16), 1873-1894.
Rasyid, A. R., Bhandary, N. P., & Yatabe, R. (2016). Performance of frequency ratio and logistic regression model in creating GIS based landslides susceptibility map at Lompobattang Mountain, Indonesia. Geoenvironmental Disasters, 3, 1-16.
Rozenstein, O., & Karnieli, A. (2011). Comparison of methods for land-use classification incorporating remote sensing and GIS inputs. Applied Geography, 31(2), 533-544.
Saha, T. K., Pal, S., & Sarda, R. (2022). Impact of river flow modification on wetland hydrological and morphological characters. Environmental Science and Pollution Research, 1-21.
Shan, V., Singh, S. K., & Haritash, A. K. (2021). Present Status, Conservation, and Management of Wetlands in India. In Advances in Energy and Environment (pp. 235-256). Springer, Singapore.
Talukdar, S., & Pal, S. (2019). Effects of damming on the hydrological regime of Punarbhaba river basin wetlands. Ecological Engineering, 135, 61-74.
Talukdar, S., & Pal, S. (2019). Effects of damming on the hydrological regime of Punarbhaba river basin wetlands. Ecological Engineering, 135, 61-74.
Talukdar, S., & Pal, S. (2020). Modeling flood plain wetland transformation in consequences of flow alteration in Punarbhaba river in India and Bangladesh. Journal of Cleaner Production, 261, 120767.
Townshend, J. R., & Justice, C. O. (1986). Analysis of the dynamics of African vegetation using the normalized difference vegetation index. International journal of remote sensing, 7(11), 1435-1445.
Vera-Herrera, L., Soria, J., Pérez, J., & Romo, S. (2021). Long-Term Hydrological Regime Monitoring of a Mediterranean Agro-Ecological Wetland Using Landsat Imagery: Correlation with the Water Renewal Rate of a Shallow Lake. Hydrology, 8(4), 172.
Wang, Y., Chen, A. S., Fu, G., Djordjević, S., Zhang, C., & Savić, D. A. (2018). An integrated framework for high-resolution urban flood modelling considering multiple information sources and urban features. Environmental modelling & software, 107, 85-95.
Wang, Z., Gao, X., Zhang, Y., & Zhao, G. (2020). MSLWENet: A novel deep learning network for lake water body extraction of Google remote sensing images. Remote Sensing, 12(24), 4140.
Wei, X., Xu, W., Bao, K., Hou, W., Su, J., Li, H., & Miao, Z. (2020). A water body extraction methods comparison based on FengYun Satellite data: a case study of Poyang Lake Region, China. Remote Sensing, 12(23), 3875.
Xu, H. (2006). Modification of normalised difference water index (NDWI) to enhance open water features in remotely sensed imagery. International journal of remote sensing, 27(14), 3025-3033.
Yamazaki, D., Trigg, M. A., & Ikeshima, D. (2015). Development of a global~ 90 m water body map using multi-temporal Landsat images. Remote Sensing of Environment, 171, 337-351.
Zhou, Y., Dong, J., Xiao, X., Xiao, T., Yang, Z., Zhao, G., ... & Qin, Y. (2017). Open surface water mapping algorithms: A comparison of water-related spectral indices and sensors. Water, 9(4), 256.

Resolution effects on ox-bow lake mapping and inundation consistency analysis in moribund deltaic flood plain of India

Status:

Journal Publication

Version 1

Abstract

Figures

1. Introduction

2. Study Area

3. Materials

4. Methods

4.1. Ox-bow Lake mapping using conventional spectral indices

4.2 Comparing the accuracy of ox-bow lake maps based on validation approach

4.3 Histogram comparison approach (HCA) and scattered plot approach (SPA) for comparing spectral indices of two image sets

4.4 Water presence frequency framework for consistency analysis

5. Results

5.1 Ox-bow Lake maps

5.2 Analyzing resolution effect on mapping using validation approaches

5.3 Comparison of two products using HCA and SPA

5.4 Inundation consistency characters

6. Discussion

Conclusion

Declarations

References

Status:

Journal Publication

Version 1