Detection of Water Spread Area Changes in Eutrophic Lake Using Landsat Data

Adequate water resource management is essential for fulfilling ecosystem and human needs. Nainital Lake is a popular lake in Uttarakhand State in India, attracting lakhs of tourists annually. Locals also use the lake water for domestic purposes and irrigation. The increasing impact of climate change and over-exploration of water from lakes make their regular monitoring key to implementing effective conservation measures and preventing substantial degradation. In this study, dynamic change in the water spread area of Nainital Lake from 2001 to 2018 has been investigated using the multiband rationing indices, namely normalized difference water index (NDWI), modified normalized difference water index (MNDWI), and water ratio index (WRI). The model has been developed in QGIS 3.4 software. A physical GPS survey of the lake was conducted to check the accuracy of these indices. Furthermore, to determine the trend in water surface area for a studied period, a non-parametric Mann–Kendall test was used. San’s slope estimator test determined the magnitude of the trend and total percentage change. The result of the physical survey shows that NDWI was the best method, with an accuracy of 96.94%. Hence, the lake water spread area trend is determined based on calculated NDWI values. The lake water spread area significantly decreased from March to June and July to October at a 5% significance level. The maximum decrease in water spread area has been determined from March to June (7.7%), which was followed by the period July to October (4.67%) and then November to February (2.79%). The study results show that the lake’s water spread area decreased sharply for the analyzed period. The study might be helpful for the government, policymakers, and water experts to make plans for reclaiming and restoring Nainital Lake. This study is very helpful in states such as Uttarakhand, where physical mapping is not possible every time due to its tough topography and climate conditions.


Introduction
Lakes are important water bodies assets; therefore, their conservation is crucial in ensuring water security to maintain both ecological and economic benefits [1,2]. A substantial reduction or augmentation of lake areas and their status in water quality can reflect the global environment and climate change impacts [3]. Thus, shrinkage in the water surface areas of lakes is caused by natural phenomena or anthropogenic processes [4][5][6]. Hence, the mappings of surface water bodies are extremely important for policy-making and research purposes.
Remote sensing technology is broadly utilized for surface features mapping, extracting and monitoring the changes, and real-time information acquisition due to its high accuracy and cost-effectiveness [7][8][9][10]. Remote sensing applications in water resources include rainwater and natural base flow. It is fed by twenty water channels, from which only tw are perennial open drains [44]. The average annual rainfall of Nainital is 3500 mm, which 85% is received during the southwest monsoon season [45].

Dataset
Landsat-7 images from 2001 to 2012 and Landsat 8 images from 2013 to 2018 ha been used in this study. These images were level-1 from the United States Geological Su vey (https://earthexplorer.usgs.gov/) accessed between 2 March 2019 to 10 March 20 and 15 April 2019. Specifications of these imageries are shown in Table 1.   Table 1. To estimate the spatiotemporal changes in Nainital lakes, each year is divided into three periods: November to February, March to June, and July to October. The days at which cloud-free Landsat imagery was used in the study every year are in Table 2. Table 2. Cloud-free Landsat imageries were used in this study.

Satellite
Year Note: "-" indicates no cloud-free Landsat image available for that duration.
"QGIS 3.4 Madeira" software has created an open-source model for analyzing satellite imagery. The QGIS software can extract water bodies from non-water bodies to calculate land cover area, built-up area, and temperature variations.

Water Body Extraction
The multiband rationing technique has been used to extract the water area from Landsat imageries. Three water indices, namely NDWI, MNDWI, and WRI, have been used to extract the lake water surface area from the image. NDWI, MNDWI, and WRI are based on the spectral index method using QGIS software. NDWI, MNDWI, and WRI were developed primarily to delineate water bodies and enhance their presence in remotely sensed imagery while eliminating soil and vegetation features.
NDWI is primarily used to monitor and detect slight changes in the water content area of the water bodies. Using NIR (near-infrared) and green spectral bands, NDWI can enhance the water bodies in a satellite image. MNDWI uses green and SWIR bands to enhance open water feature extraction. MNDWI can enhance open water features while efficiently suppressing and removing built-up land, vegetation, and soil noise. In the WRI technique, four spectral reflectance bands were used. It is the ratio of the total spectral reflectance of two visible bands (i.e., green and red) to the spectral reflectance of the nearinfrared and middle infrared bands (MIR). This method only applies to satellite images with a middle infrared band.
The formulas for the calculation of NDWI, MNDWI, and WRI are described as follows: where Green, NIR, SWIR_1, and SWIR_2 represent the reflection of band 3, band 5, band six, and band 7, respectively, for Landsat 8 images and band 2, band 4, band 5, and band 7 for Landsat 7 images, respectively.

Accuracy Analysis
In remote sensing techniques, physical mapping is important to verify and accurately check satellite data. For this purpose, a survey of the Nainital Lake has been completed using GPS. The GPS survey of Nainital Lake was completed on 20 March 2019, and the area obtained by this survey was compared with the area obtained by the nearest Landsat-8 image on 20 March 2019. The comparison was made to find the error. The best method was determined among NDWI, MNDWI, and WRI by comparing GPS results.

Trend Analysis
To detect the non-parametric trend in hydro-climate data, the Mann-Kendall test [46][47][48][49] has been widely used by researchers and recommended by the World Meteorological Organization [50][51][52][53]. The Sen slope method is used to obtain the magnitude of the trend and percentage change of the lake. Sen's slope method is a non-parametric method in which slope calculation carries trend prediction [54]. Detailed mathematical expressions have been given in [55,56].

Comparison of Different Band Ratios for Accurate Water Spread Mapping
The water spread area from 2001 to 2018, from November to February, March to June, and July to October, has been shown in Tables 3-5, respectively. A threshold value between −0.05 and 0.9 has been used to estimate the water spread mapping. From Table 3, it could be suggested that the maximum and minimum area obtained by NDWI was 0.   Similarly, Table 4 depicted that the average lake water spread area estimated from March to June by NDWI was 0.409 km 2 with a standard deviation (SD) of 0.01833 km 2 . The maximum and minimum water spread area obtained by MNDWI was 0.423 km 2 in 2014-2015 and 0.383 km 2 in 2011-2012, respectively.
The average area for the studied period by the MNDWI method is 0.3998 km 2 with an SD of 0.0100 km 2 . By the WRI method, the maximum and minimum water spread area is 0.431 km 2 in the study year 2002-2003 and 0.388 km 2 in 2017-2018, respectively. This method's average water spread area was 0.4119 km 2 with an SD of 0.0124 km 2 .
Further, Table 5 shows Nainital Lake's calculated water spread area for July to October. The maximum and minimum area obtained by NDWI was 0.467 km 2 in 2007-2008 and 0.432 km 2 in 2015-2016, respectively. This method's average lake water spread area for the The water spread area of Nainital Lake was conducted using a GPS survey on 16-17 March 2019, and its results are presented in Table 6. Landsat-8 imagery of 20 March 2019, nearest to the day of the GPS survey, was obtained from earth explorer and then extracted to identify the lake's area using different multiband water indices. A comparison between surveyed and estimated water areas by Landsat imagery is presented in Table 6. Table 6 also shows the overall errors in estimating water spread mapping using different water indices. It was also found that the accuracy of all methods used was more than 80%. However, the overall accuracy of NDWI was higher, followed by the WRI method. A 3.06% error or deviation from the original surface water of the lake was noted by NDWI.
In contrast, a 3.16% deviation was noted by using WRI. Due to the higher accuracy, the NDWI method's areas were considered accurate areas of Nainital Lake in different seasons from 2001 to 2018. The water spread mapping and spatial changes in water spread from 2001-2002 to 2017-2018 of Nainital Lake using NDWI are shown in Figures 2-4 for the periods November to February, March to June, and July to October, respectively.

Seasonal Fluctuation in Water Spread Area Using NDWI
Since NDWI estimated the water spread area more accurately in this study than other indices, this rationing technique has been used to analyze water spread fluctuations further. Figure 5 depicts the fluctuations in water spread area from November to February, March to June, and July to October estimated using NDWI.
Since July to October is a monsoon month in India, the water spread area is expected to be maximum for these months. A remarkable change in water spread could be noticed from March to June. From 2001 to 2018, from July to October, there was a reduction in the water spread area of 0.028154 km 2 from March to June. Similarly, from November to February, there was a decrease of 0.016846 km 2 .

Seasonal Fluctuation in Water Spread Area Using NDWI
Since NDWI estimated the water spread area more accurately in this study than other indices, this rationing technique has been used to analyze water spread fluctuations further. Figure 5 depicts the fluctuations in water spread area from November to February, March to June, and July to October estimated using NDWI. Since July to October is a monsoon month in India, the water spread area is expected to be maximum for these months. A remarkable change in water spread could be noticed from March to June. From 2001 to 2018, from July to October, there was a reduction in the water spread area of 0.028154 km 2 from March to June. Similarly, from November to February, there was a decrease of 0.016846 km 2 .

Trend and Spatio-Temporal Variability in Water Spread Area
Trend analysis of lake spread area has been conducted using the non-parametric test The main results of the trend analysis based on the Mann-Kendall test are presented in Table 7. The Z-statistics value for November to February was −1.090, March to June was −2.818, and July to October was −2.961. The Mann-Kendall test results indicated a significant negative trend in the water surface area of Nainital Lake for the period March to June and for the period July to October, whereas for November to February, the trend was not significant at a 5% significance level. March-June July-0ct.

Trend and Spatio-Temporal Variability in Water Spread Area
Trend analysis of lake spread area has been conducted using the non-parametric test The main results of the trend analysis based on the Mann-Kendall test are presented in Table 7. The Z-statistics value for November to February was −1.090, March to June was −2.818, and July to October was −2.961. The Mann-Kendall test results indicated a significant negative trend in the water surface area of Nainital Lake for the period March to June and for the period July to October, whereas for November to February, the trend was not significant at a 5% significance level.
The magnitude of the trend and percentage changes of the lake water spread area over the studied period was found using Sen's slope estimator test. The results have been tabulated in Table 8. The magnitude of the trend for the period November to February, March to June, and July to October was −0.00070 km 2 /year, −0.00187 km 2 /year, and −0.00123 km 2 /year, respectively. The maximum decrease in water spread area has been obtained for the period March to June (−7.70%), which was followed by the period July to October (−4.67%) and the period November to February (−2.79%).

Discussion
Accurately estimating surface water bodies becomes essential as water problems increase globally. The important objective of this study is to estimate changes in the surface area of water in Nainital Lake using remote sensing and satellite data. Lake data have been extracted from Landsat images using NDWI, MNDWI, and WRI. Furthermore, a GPS survey of the lake has been completed to estimate the most accurate index for Nainital Lake. Compared to the result of the GPS survey with the area obtained by water indices, the overall accuracy of all used indices is more than 95%. The accuracy of 96.94% when using the NDWI approach compared to 94.31% when using MNDWI and 96.29% using WRI indicates that NDWI is better at extracting small water bodies in the Himalayas region's water bodies and at separating water bodies from water-like features such as forest shadows. This study's findings also agree with Rokani et al. [21]. They observed that for modeling temporal lake area change, the NDWI performed better than other indices. Acharya et al. [57] employed the same indices as NDVI, NDWI, MNDWI, and WRI. NDWI can be easily applied to satellite imagery to detect water features from the land surface.
Nevertheless, it should be emphasized that extracting water bodies using a multiband pixel-based method has inherent challenges because they require expertise from analysts to select the accurate training data set [58]. Using a water index such as NDWI is crucial for extracting water bodies to provide information for stockholders, policymakers, and decision-makers. Namely, the developed methodology in this study might be very useful for decision-makers and water managers to take effective measures such as reducing pollution and overwater exploitation from such important and fragile water bodies.
On the other hand, Deoli et al. [59] observed that WRI is the best method to calculate the water extraction area of some Himalayan lakes of India among the studied NDWI, MNDW, WRI, and NDVI indices based on a physical survey of lakes. Furthermore, they calculated the trend in all studied lakes based on Mann-Kendall and Sen's slope test. The data show a strong downward tendency, and their results align with this study. In addition, Elshabi et al. [60] extract the water surface area using the same water indices. Their overall study accuracy varied from 97.90 to 99.40%, and they recommended these techniques in the areas with the same conditions to extract water bodies. Bijeesh and Narasimhamurthy [61] performed a comparative study to identify water suitability for NDWI, MNDWI, WRI, normalized difference forest index (NDFI), automated water extraction index (AWEI), tasseled cap water index, EWI, and weighted normalized difference water index (WNDWI). All these indices are important for water extraction. All these indices are quick and beneficial in extracting the water bodies. The Mann-Kendall and Sen's slope estimator shows their efficiency in calculating the trend and magnitude of the trend in hydrological data.

Conclusions
In this study, the dynamic change in the water spread area of Nainital Lake of India has been completed using the band rationing (i.e., multiband water indices) technique in QGIS software by Landsat images. The results evaluate that the area found by NDWI is very close to the actual area of the lake determined by the GPS survey. The Mann-Kendall test and Sen's slope were employed to detect a trend in Nainital Lake water spread obtained from NDWI. It has been observed that the water spread area of Nainital Lake shrunk for all studied periods by 0.00070 km 2 /year for the period November to February; −0.00187 km 2 /year for the period March to June, and −0.00123 km 2 /year for the period July to October. The overall decrease in the water spread area of Nainital Lake in the studied years (from 2001 to 2019) was 7.70% for the period March to June, followed by 4.67% for the period July to October and 2.79% for the period November to February. The decreasing lake water area is not a good sign for Nainital Lake. It attracts lakhs of tourists annually and is one of the famous hill stations in India.
Furthermore, the study can also assist policymakers and water information specialists in managing surface water resources. This work will focus on the fusion of the data of studied indices using machine learning techniques to increase efficiency and accuracy. In further work, we will try new indices with machine learning for water surface mapping and monitoring.