Proposal for a new Green Red Water Index for geo-environmental surface water monitoring

One of challenges of today’s world is the long-term geo-monitoring of phenomena and processes that affect our environment after the closure of mining activities. Water resources are one of the components affected by post-mining processes. Moreover, land subsidence can be observed both during and after the cessation of mining activities. This phenomenon has an impact on the entire water management of a region. So far, radar or multispectral images have been used to identify water surfaces. This paper will present a methodology for using drones to detect water surfaces using vegetation indices such as NDVI, GRNDVI and NDWI. During their research, the authors modified the GRNDVI indicator by including the Red Edge band in the calculations. The newly developed Green Red Water Index – GRWI, makes it possible to identify water surfaces. This is important, because the change of water conditions makes it important focus more on the water supply and availability. However, analysis of the spectral bands of the different land-use classes in the Ruhr study area shown that the spectral profiles for water, soils, road and street surfaces have similar spectral characteristics and therefore difficulties may arise in distinguishing between the land-use classes shown. In this article a comparison of the indicators will be presented: NDVI, GRNDVI, NDWI and GRWI together with their statistical interpretation.


Introduction
One area of interest that utilizes of geo-environmental monitoring is the research of post-mining processes occurring on the ground surface after mining operations have ceased.Directive 2011/92/EU of European Parliament and of the Council of 13 December 2011 on the assessment of the effects of certain public and private projects on the environmental (codification) [1] indicates that extensive industry activities, including mining, should be subject to an assessment of their effects on the surrounding environment.
The assessment should be undertaken for the direct and indirect effects of the research on the following components [Ibidem]: " a) human beings, fauna and flora; b) soil, water, air, climate and the landscape; c) material assets and the culture heritage; d) the interaction between the factors to in points (a), (b) and (c)." An important component of the environment is water, both surface and groundwater.Mining activities also have a substantial impact on the water economy in a given area.During the exploitation of mineral deposits located below the water table, it is necessary to use ground drainage to dewater the mining excavations.However, this has consequences such as: 1295 (2024) 012013 IOP Publishing doi:10.1088/1755-1315/1295/1/012013 2  Lowering the groundwater level [2,3];  Dewatering of surface streams or a decrease in their flow rates [3,4];  Drying of wetlands [3];  Changes in the quality of surface and groundwater [5][6][7][8][9][10][11][12][13];  Formation of reservoirs (polders) on settled land [14];  The possibility of flooding adjacent mines [10,11,15].
Remote sensing is a geo-environmental monitoring method that can provide reliable information on environmental change at a regional and global scale [16,17].The use of drone flights enables data to be collected at a local scale, allowing a better understanding of phenomena occurring in a small area.Remote sensing methods use the basic reflectance properties of objects to observe the state of the environment, which depend on material and surface properties.Kuechly et al. [18] point out that each object (e.g.water, vegetation, soil) has a characteristic reflection and absorption pattern, which called a spectral signature.In the case of plants, this is characterized by the process of photosynthesis, in which plants absorb electromagnetic radiation to generate energy.In this way, they protect themselves from overheating [Ibidem].In order to enable life processes (photosynthesis, respiration and nutrient uptake) plants need water.This is taken up by the roots and transported to all parts of the plant [19].When there is insufficient water in the soil, "plants may suffer water stress" [Ibidem, p.741].Water stress causes stomatal closure, which reduces transpiration, resulting in reduced evaporative cooling and increased leaf temperature [20].A comparable problem can occur if there is too much water in the root zone.Such problems can occur during mining subsidence [15,[21][22][23][24][25][26][27][28].The observation of the environment and in particular of the locations of water surfaces, together with the integration of data from other sources, allows an understanding of the processes taking place after the closing of mining activities and the possibility of preventing the effects of mining activities in a given area.The occurrence of water surface can be related to various factors: the lowering of the land surface or the outflow of groundwater to the land surface."Monitoring the extent of surface water enables resource managers to detect perturbations and long-term trends in water availability and set consumption guidelines accordingly" [29, p.336].
The possibility of detecting and monitoring water surfaces using radar data derived from satellite imagery is a field in development.Their undoubted advantage is the independence from sunlight and atmospheric condition (cloud cover), which allows timely data acquisition and, in the long-term, continuous monitoring.The dieelectric constant of free water is very high compared to vegetation and dry soil, so any change in water content has a strong effect on radar scattering characteristics, making it possible to "detect small changes in content for soil or vegetation targets" [46, p.131].Morley et al. [48] indicate that SAR interferometry at C-band interferometry can be used to delineate flooded and nonflooded areas.This phenomenon -for example -is used after heavy rain disasters to create water masks for different purposes [49].Moreover, to distinguish between vegetation types in wetlands, including flooded forests and aquatic macrophytes, combinations of L and C bands should be used [46,50].Brisco et al. [46] show that the use of the HH/VV band is preferred for mapping floodplains and wetlands.Li-Chee-Ming et al. [47] point out the possibility of validating radar data (RADARSAT-2) and data from small unmanned aerial systems (sUAS), which enables flood mapping and is also one method of environmental monitoring.
The second method of identifying water surfaces are vegetation indices (VI), which are calculated on the basis of two or more spectral bands.They are used to determine the state of vegetation.Healthly vegetation absorbs red and blue wavelengths and reflects green and near-infrared (NIR) wavelengths.Vegetation components (leaf structure, leaf pigments) and leaf water content can cause changes in the spectral signature, making vegetation indices sensitive to changes in a vegetation structure or leaf water content [51,52].The Normalized Difference Vegetation Index (NDVI) is one of the oldest vegetation indices based on measurements of red and near-infrared reflectance [33,[53][54][55][56].The NDVI is calculated using Formula 1 presented in Table 1, with values in the range of <-1.1>.The higher the value, the better the condition of the vegetation [33,53].One of the best known indices for detecting and mapping parts of surface water is the Normalized Difference Water Index -NDWI [18,57].Both Gao [34] and McFeeters [35] named the index -NDWI, but to distinguish between them, it is important to note the formula of their research basis [18].McFeeters [35,58] presents the NDWI in Formula 2 referred to in Table 1.The NDWI uses green and near-infrared reflectance to provide information on vegetation condition [18,35] and to identify surface water associated with wetlands [57].NDWI index values are in the range of <-1.1> [Ibidem].Xu [39] points out an important disadvantage of the method, namely that it is unable to separate built-up objects from water bodies.Gao [34] presents NDWI in Formula 3 presented in Table 1.In its calculations, this formula takes into account the bands in the range of 0.86 μm and 1.24 μm, which in the case of satellite imagery corresponds to the NIR and SWIR bands.However, Tucker [59] already pioneered the use of the SWIR band for water abundance calculations.The SWIR band (range 1300-2500 μm) shows strong absorption of light by liquid water and is therefore sensitive to the amount of liquid water in vegetation and its background in soil [36,39,[60][61][62][63][64][65].Based on the Formula 3 proposed by Gao [34], Chen et al. [36] presented Formulas 4 and 5 (Table 1).However, the authors identify the SWIR bands of the ranges: 1640 and 2130 nm as bands that are sensitive to changes in Vegetation Water Content (VWC).Jürgens [37] calls the combination of NIR and MIR bands: mNDWI -modified Normalized Difference Water Index (Formula 6 presented in Table 1) [Ibidem].Examples of the use of the mNDWI proposed by Jürgens [Ibidem] include the work of Xiao et al. [63,66] on a study of the Northeast China area.Xu [39], based on McFeeters [35], replaced the near-infrared (NIR) band with the shortwave infrared (SWIR) band and named these Formula 8 (Table 1), as modified NDWI (MNDWI).This made it possible to distinguish between built-up areas and water bodies, but the occurrence of shadows in mountainous terrain caused problems in interpretation [67].Yang and Xu [38] proposed another indicator, the General Water Index-GWI.This index is based on green, red, near-infrared and MIR channels.The Formula 7 of the GWI is presented in Table 1.In order to make full use of the information in the visible range, Wang et al. [40] proposed replacing the red band in Formula 1 by combinations of each sum of the green bands, and the results are presented as Formula 9 in Table 1.Shen et al. [41] proposed an indicator -the Water Ratio Index, based on the GWI indicator.An important difference is the use of the SWIR1 band instead of the MIR band.The Formula 10 of the WRI is shown in Table 1.To develop the new Automated Water Extraction Index-AWEI, Feyisa et al. [42] used five spectral bands from the Landsat 5 TM satellite.They proposed two separate Formulas: 11 and 12 in Table 1, to extract surface water.The main objective was to increase the separability of water and dark surfaces (shadows or buildings).Also, the choice of coefficients was intended to stabilise the threshold needed to distinguish between water and non-water pixels.In this case, the threshold was 0, meaning that values below 0 were non-aquatic pixels and positive values were water pixels.

Problem statement
The overview of vegetation indicators presented above makes it possible to carry out studies on the identification of water surfaces.Most of the vegetation indicators use the SWIR and MIR spectral channel [34, 36-39, 41, 42], which so far has not been used as a sensor in drone flights.Due to this fact, the authors in this publication decided to review vegetation indicators for the identification of water surfaces, based on the spectral bands that are used in the most popular multispectral cameras used in drone flights, namely: Blue, Green, Red, Red Edge and NIR.Of the vegetation indices listed in Table 1, the following meet this criterion: NDVI, NDWI and GRNDVI.Vegetation indices have certain limitations:  NDVI according to Kuechly et al. [18] does not distinguish between soils or streets and water. The NDWI indicates that water surfaces range from 0 to 1 [35]. The GRNDVI indicator has no classification.

Contribution
The aim of this paper is to propose a Green Red Water Index (GRWI), based on the GRNDVI formula, the model of which is shown in Table 1 under Formula 13.The new indicators allows observation and identification of the water surface in a selected area of the disused Prosper-Haniel mine, based on multispectral camera drone without need to use SWIR and MIR bands.Monitoring the water surface with a drone equipped with a multispectral camera will enable data to be acquired in less time and with better spatial resolution than with optical or radar satellite imagery.Vegetation indices calculated from satellite imagery have a specifv spatial resolution that depends on the characteristics of the space mission.In the case of drone flights equipped with multispectral camera, centimeter-level accuracy can be achieved by using RTK (Real Time Kinematic) module.A selection of vegetation indices will also be calculated in this article: NDVI, NDWI and GRNDVI for comparison with the newly created index -GRWI.The results of the calculated indices will be statistically analysed.

Materials and Methods
The following steps were carried out in this study, as shown in Figure 1, which included: site selection, collection and pre-processing of drone data, calculation of water indices and their subsequent validation.The processes described above were carried out in the following environments:

Test area
The study area was selected as the area between 51 • 35'04" N to 51 • 35'12" N latitude and 6 • 54'40" E to 6 • 54'50" E longitude.Figure 2 shows the location of the study area.The research area is a section of the Boye River, located southwest of Kirchhellen, Germany.This site belongs to the Prosper-Haniel coal mine, which was in operation until 2018 [68].Due to mining activities in this area, subsidence can be observed, which has a significant impact on the water management in this region (pumping stations have been established, which enable the supply of water to the downstream stages of the river).The area in question contains vegetation such as plants and mixed forest on the periphery.It belongs to the cathment area of the River Boye, which flows through the area.It is part of the wider context of the near-natural restoration of the Emscher and its tributaries, such as the Boye brook, which recently developed itself close to nature.

Data
Drone flights are a modern method of data acquisition, enabling data to be acquired in a short time and showing centimetre accuracy, depending on the flight settings selected (altitude) and the use of the RTK module (Real Time Kinematic), which increases the accuracy of the results obtained.The research presented in this paper is based on two drone flights carried out on 5 May 2022 and 6 October 2022 (Figure 5).The drone that has been used for this research is the DJI Phantom 4 with RTK module and multispectral camera.The spectral characteristics of the multispectral camera are shown in Table 2.Both drone flights were carried out at a height of 100 m and the spatial resolution achieved is 5 cm.

Methods
In this article, research will be carried out on the basis of calculated vegetation indices: NDVI, NDWI, GRNDVI and GRWI.The calculation of the vegetation indices is based on a combination of spectral bands, so at the beginning of the research it is necessary to familiarise oneself with the spectral profiles of the different land use classes, which are shown in Figure 6.It is worth noting that the spectral profiles corresponding to different types of vegetation show increased pixel values in the RedEdge and NIR bands (Figure 6).In contrast, the water classes and street show reduced values in the RedEdge and NIR bands (Figure 6) and increased values for the blue band (Figure 6).The highest pixel values in the red band were observed for the soil and street classes.The NDVI index proposed by Rouse et al. [33] is based on the NIR and RED bands.According to Kuechly et al. [18], values below 0.1 indicate that these can be land use classes: water, soil, rock, sand or snow.However, the water classes cannot be distinguished from the other classes mentioned earlier.The second indicator that has been analysed is the NDWI indicator proposed by McFeeters [35], which is based on green and near-infrared bands.The NDWI equation results in positive values for water features and negative (or zero) values for soil and terrestrial vegetation.The third indicator used in this study is the GRNDVI proposed by Wang et al. [40].GRNDVI values oscillate between -1 and 1. Taking into account the spectral profiles for the different land use classes presented in Figure 4, the authors modified the GRNDVI formula, focussing the RedEdge channel and the order of the signs, so that the range of values corresponds to -1 to 1.

Statisitcal interpretation
The calculated vegetation indices described in subsection 2.3 will be statistically interpreted.Namely, pairs of indicators will be compiled for two time points: 05.05.2022 and 06.10.2022.As a result of this study, a correlation analysis between the indicators will be carried out using the Pearson correlation coefficient (PCC).The Pearson coefficient is a measure of the linear correlation between two sets of data.It is expressed as the ratio of the covariance of two variables to the product of their standard deviations (Formula 14).Here,  ̅   ̅ are indicating average value, (, ) are indicating the covariance and     are indicating standard deviation.The Pearson coefficient score has a value between -1 and 1.On the basis of the coefficient, it is possible to assess the strength of the correlation (Table 3).

Strong correlation
The following descriptive statistics will also be presented for each land-use class: minimum value, maximum value, mean ( ̅ ) and standard deviation (σ).Therefore, the values of the individual indices in the land use classes are examined.So far, vegetation indices have not yet been classified.The exception is the NDVI, for which a classification was presented in Kuechly et al. [18].Therefore, the authors will attempt to classify the GRWI indicator.

Results
In this chapter, the results of the calculations of selected vegetation indices based on drone flights from two time points will be presented: 05.05.2022 and 06.10.2022.Figure 7 shows the indices A) NDVI, B) NDWI, C) GRNDVI and D) GRWI for the time: 05.05.2022.As a result of the analyses carried out, correlations between the various indicators were examined, as shown in Table 4.As a result of the analyses carried out, correlations between the various indicators were examined, as shown in Table 5.The relationship between the pairs of indicators is characterised by a very strong correlation.
In order to attempt classifying the new GRWI indicator, test objects were selected based on land use classes (Figure 3).The land use classes were selected based on the classification provided by the WMS Landbedeckung NW service published by Open NRW.The following classes were recognised in the study area: vegetation (grass, shrub, hardwood), street and water.
Indices were calculated for each of the test areas shown in Figure 3: • NDVI; • NDWI; • GRNDVI; • GRWI.The results were statistically interpreted as shown in Table 6 and in Figure 9.For this purpose, the values of each indicator were calculated: minimum, maximum; average and standard deviation.The Figure 9 present a different behaviour between May and October in the case: shrub, hardwood and inland water, as the sun was low in October, allowing shadows to form.

Discussion
The index calculations are based on data from two time points.It is worth noting that the drone raids were carried out in two time points.For the drone data from 05.05.2022 and 06.10.2022,calculations were made with Agisoft Metashape software.For the raid carried out on 06.10.2022, a low Sun position (<30 degrees) was observed, therefore shadows appear in the southern part of the image, making it difficult to identify locations.Therefore, during data pre-processing in Agisoft Metashape, sun correction was also applied.
In the summaries of all the indicators for the various time periods shown in Figures 7 and 8, it can be seen that there are differences between them.
The NDVI, GRNDVI and GRWI indicators show positive values and NDWI show negative values for the area shown in Figure 10.The area in question is classified as having grass and a gravel road.Can be seen that the GRNDVI values are negative (Figure 11), which is also observed in other indicators, but not on such a large scale.This is related to the occurrence of soils in the area.There are wetlands in the central part of the test area (Figure 12), where there is vegetation that is surrounded by water.In all the indicators shown in Figures 7 and 8, the street, which is located in the eastern part of the study area, is coloured red, indicating negative values for the NDVI, GRNDVI and GRWI indicators and positive values for the NDWI indicator.A study of the relationships for each pair of indicators was carried out using Pearson's correlation coefficient.Table 4 and 5 show the relationship between each indicator, which showed a strong correlation.We can observe occurrences where there are no values.This is the case for the relationship between NDVI and GRNDVI (Figure 12).This is due to the strong correlation between the pair of indicators, as the values of the GRNDVI indicator are not greater than those of the NDVI.A similar situation exists between the GRNDVI and GRWI indicators (Table 4 and 5).The NDWI values for water areas are positive, i.e. greater than 0. The opposite is true for the GRWI, which is negative for water classes.This paper also attempts to present a classification of the new indicator.An analysis of Figure 6, which shows the spectral profiles of the different land-use classes, reveals that the water layer as well as the streets have similar spectral profiles.Therefore, in a given study it is expected that the values of the calculated vegetation indices will have similar values in the land use classes water and street.This is observed and shown in Table 6 and in Figure 9.The average values for the water class are: NDVI -0.06 and 0.33, NDWI 0.12 and -0.22,GRNDVI -0.41 and -0.06 and GRWI 0.24 and 0.42.

Conlusions
The aim of this study was to present a new vegetation indicator, the GRWI, for special questions in mining and post-mining, especially as a base for data fusion in Geo-monitoring such areas using satellite and drone imageries.The application was observing and identifying water surfaces for a selected area of the Prosper-Haniel mine.On the basis of raids with a DJI Phantom 4 drone equipped with a multispectral camera, carried out in two periods: May and October 2022, multispectral analyses were performed using the calculated vegetation indices: GRWI, NDVI, NDWI and GRNDVI.As shown in Figure 6, the Red Edge band also influences the presentation of a given pixel in the image.It can be seen that the spectral profiles of the water, soil and street classes have similar spectral profiles.Consequently, there are difficulties in distinguishing between the different classes using the selected vegetation indices.The newly developed index, which has been given the name Green Red Water Index -GRWI, has been developed on the basis of the GRNDVI index formula, also taking into account the Red Edge band.The GRWI has values ranging from -1 to 1.As shown in Figures 7 and 8, it enables the detection of water surfaces.The authors suggest that the water values for GRWI should be determined between 0.2 to 0.6 (Figure 13).As shown on Figure 14, GRWI also recognized streets.The above conclusions are presented based on the results shown in Table 6 and Figure 9.The authors note that analyses should be carried out for other areas to verify the following classification.The vegetation indices presented in this study, calculated on the basis of drones flights, can be used to monitor water surfaces, allowing services to quickly determine the extent of the action and to predict the effects of flooding.It is worth noting that this method is dependent on the prevailing weather conditions, such as wind speed and sunshine in a given region and season (spring = more vegetation, autumn = less vegetation).

Figure 2 .
Figure 2. Location of the research area.

Figure 3 .
Figure 3. Classification of the land use in the research area and test class in the research area.Source: Open NRW (WMS Survey: Landbedeckung NW).

Figure 6 .
Figure 6.Spectral profiles of various types of land use (a) location; (b) spectral profiles.

Table 4 .
Correlation between vegetation indices from 05.05.2022.the pairs of indicators is characterised by a very strong correlation.Figure 8 shows the indices A) NDVI, B) NDWI, C) GRNDVI and D) GRWI for the time: 06.10.2022.

Figure 9 .
Figure 9.The figure presents the statistical interpretation of the test class in the research area.

Figure 13 . 20 Figure 14 .
Figure 13.Comparison of the GRWI indicator for water classes from 05.05.2022(b) with land use (a).