Multi-scale fluvial remote sensing - a study on spatial scaling discrepancies between Sentinel-2 and UAV multispectral data on riparian zones in Northwest Portugal

Universidade do Minho Escola de Ciências Pedro Branco Saldarriaga Multi-scale fluvial remote sensing – A study on spatial scaling discrepancies between Sentinel-2 and UAV multispectral data on riparian zones in Northwest Portugal março de 2022 Multi-scale fluvial remote sensing – A study on spatial scaling discrepancies between Sentinel-2 and UAV multispectral data on riparian zones in Northwest Portugal Pedro Branco Saldarriaga UMinho | 2022 Pedro Branco Saldarriaga Multi-scale fluvial remote sensing – A study on spatial scaling discrepancies between Sentinel-2 and UAV multispectral data on riparian zones in Northwest Portugal Dissertação de Mestrado Mestrado em Biodiversidade, Ecologia e Alterações Globais Trabalho efetuado sob a orientação do Professor Doutor Renato Filipe Faria Henriques e do Doutor Giorgio Pace ~ março de 2022 ii DIREITOS DE AUTOR E CONDIÇÕES DE UTILIZAÇÃO DO TRABALHO POR TERCEIROS Este é um trabalho académico que pode ser utilizado por terceiros desde que respeitadas as regras e boas práticas internacionalmente aceites, no que concerne aos direitos de autor e direitos conexos. Assim, o presente trabalho pode ser utilizado nos termos previstos na licença abaixo indicada. Caso o utilizador necessite de permissão para poder fazer um uso do trabalho em condições não previstas no licenciamento indicado, deverá contactar o autor, através do RepositóriUM da Universidade do Minho. Licença concedida aos utilizadores deste trabalho Atribuição CC BY https://creativecommons.org/licenses/by/4.0/ iii AGRADECIMENTOS Dado como terminado o processo de investigação e escrita da presente tese de mestrado, resta-me apenas escrever breves palavras de agradecimento a todos os que de uma maneira direta ou indireta me ajudaram a concluir este processo. Primeiramente quero agradecer a ambos os professores orientadores por acreditarem nas minhas capacidades, pela ajuda fornecida ao longo destes 2 anos e pelo seu valioso contributo no desenvolvimento desta tese. Um especial agradecimento ao Professor Giorgio Pace pela constante disponibilidade durante todo este percurso e por todas horas despendidas a responder às questões que foram surgindo. Peço desculpa por qualquer cabelo branco ganho à minha custa! Por fim, mas não menos importante, à minha família e amigos que nunca me deixaram de surpreender com o apoio incondicional que me ofereceram durante todo o este processo. Esse mesmo apoio foi vezes sem conta o “empurrão” que não me deixou desistir dos objetivos inicialmente traçados. This work was supported by the “Contrato-Programa” UIDB/04050/2020 funded by national funds through the FCT I.P., the Centre of Molecular and Environmental Biology (CBMA) and the STREAMECO project (Biodiversity and ecosystem functioning under climate change: from the gene to the stream, PTDC/CTA-AMB/31245/2017). The work was also supported by CCDR-N (Norte Portugal Regional Coordination and Development Commission) and European Funds (FEDER/POCI/COMPETE2020) through the project AgriFoodXXI (NORTE-01-0145-FEDER000041) iv STATEMENT OF INTEGRITY I hereby declare having conducted this academic work with integrity. I confirm that I have not used plagiarism or any form of undue use of information or falsification of results along the process leading to its elaboration. I further declare that I have fully acknowledged the Code of Ethical Conduct of the University of Minho. v Remote sensing fluvial de escala múltipla – Um estudo das discrepâncias de escala espacial entre dados multiespectrais de Sentinel-2 e UAV em zonas ripárias no nordeste de Portugal RESUMO Dados provenientes de deteção remota e observação da Terra são cada vez mais utilizados para a monitorização e avaliação do estado da saúde de ecossistemas bem como das suas funções. Com o desenvolvimento de novas tecnologias, sensores montados em plataformas UAV fornecem dados de deteção remota a resoluções com maior precisão que aquela encontrada em satélites, apesar de não conseguirem cobrir tanta área como estes últimos. A ponderação destes prós e contras dá origem ao problema de correlacionar os dados provenientes de ambas as fontes. Nesta tese são apresentadas uma análise detalhada e uma comparação entre dados multiespectrais de habitats ripários captados em quatro afluentes (CAB1, RAB2, VEZ2 and VEZ3). Com base em dados multiespectrais capturados por um sensor Micasense Rededge™ montado num DJI Phantom 4 RTK e pelos satélites Sentinel-2, foi feita a caracterização dos afluentes utilizando o índice NDVI ( normalized difference vegetation index ). O NDVI foi considerado para este estudo uma vez que a sua relação com o estado de saúde das comunidades de plantas é bem conhecida. As imagens captadas por UAV foram redimensionadas em três processos diferentes para igualar a resolução espacial do satélite (10x10m): média, mediana e terceiro quartil. Para testar qual das imagens redimensionadas se aproximava mais à de satélite, foram usadas medidas de goodness of fit (RMSE e R2). Os resultados demonstram que nas resoluções nativas, os valores de NDVI apresetam o máximo de dispersão, o que é esperado dada a maior divergência na escala das resoluções. O método de upscale por terceiro quartil foi o que mais se aproxima aos dados de satélite. Uma segunda análise foi feita para avaliar qual era a maior causa da dispersão de valores dentro do terceiro quartil. Foi encontrada uma maior influência do tipo de uso do solo que na localização dos rios, sendo os campos agrícolas os que apresentam maior discrepância, maioritariamente devido a diferenças no uso do solo (rotação de baldios) e a diferentes estádios de crescimento das colheitas. Este método comparativo devia ser utilizado em diferentes ecossistemas, índices e intervalos temporais para avaliar a sua fiabilidade Palavas-chave: deteção remota; UAV; Sentinel-2; vegetação ripária; escalamento vi Multi-scale fluvial remote sensing – A study on spatial scaling discrepancies between Sentinel-2 and UAV multispectral data on riparian zones in Northwest Portugal ABSTRACT Remote sensed data is increasingly being used to monitor and evaluate ecosystem health and functions. With the dawn of new technologies, UAV platform mounted sensors provide remote sensing data at spatial resolutions that are far more precise than satellite, however the spatial extent to which a UAV covers is diminutive when compared to that of satellite. This trade-off between pros and cons raises a problem in correlating image data from both sources. In this thesis, a detailed analysis and comparison of riparian habitat multispectral data between UAV and satellite at four different river reaches (CAB1, RAB2, VEZ2 and VEZ3) is presented. Based on multispectral data, captured from a Micasense Rededge™ sensor mounted on a DJI Phantom 4 RTK and Sentinel-2 satellites, the characterization of the stream reaches was possible using the normalized difference vegetation index (NDVI) maps. NDVI was considered due to its well-known relationship to plant community health. UAV images were rescaled to match satellite resolution (10x10m pixel) by three distinct methods: average, median and third quartile. To test which one was closer to satellite values, goodness of fit measures (RMSE and R2) were considered. Results show that at native resolutions, NDVI values differ the most, as is expected due to the higher divergence of spatial resolution. The method that best fitted the satellite values was upscaling by third quartile. A second analysis was made to evaluate what caused dispersion within the third quartile upscale. Significantly higher influences of land cover type were confirmed when compared to river location, with farmland showing the greatest discrepancy mainly because of differences in farm plot use (fallow rotation) and crop growth stage. This proposed comparative method should be extended to different ecosystems, indices and time frames in future studies to further evaluate his reliability. Keywords: remote sensing; UAV; Sentinel-2; riparian vegetation; rescaling TABLE OF CONTENTS vii DIREITOS DE AUTOR E CONDIÇÕES DE UTILIZAÇÃO DO TRABALHO POR TERCEIROS ...... ii AGRADECIMENTOS ........................................................................................................ iii STATEMENT OF INTEGRITY ............................................................................................ iv RESUMO .......................................................................................................................... v ABSTRACT ...................................................................................................................... vi LIST OF FIGURES ............................................................................................................ ix LIST OF TABLES .............................................................................................................. x LIST OF SUPPLEMENTARY MATERIAL FIGURES .............................................................. xi LIST OF SUPPLEMENTARY MATERIAL TABLES .............................................................. xii 1. INTRODUCTION ....................................................................................................... 1 1.1. Anthropogenic influence in riparian ecosystem functions and services ................................... 1 1.2. Remote sensing for monitoring ecosystems ........................................................................ 2 1.3. Vegetation spectral indices ................................................................................................ 4 1.4. Study Objectives and Hypothesis ........................................................................................ 5 2. MATERIALS AND METHODS .................................................................................... 6 2.1. Study sites ...................................................................................................................... 6 2.2.2. Satellite time series imagery .................................................................................... 10 2.3. Earth observation data ................................................................................................ 10 2.4. Normalized Difference Vegetation Index ........................................................................ 11 2.5. Data processing and analysis ........................................................................................... 11 2.5.1. Data processing ...................................................................................................... 12 2.5.2. Data analysis .......................................................................................................... 13 3. RESULTS ............................................................................................................... 14 3.1. Evaluating discrepancies between satellite and UAV images. .................................... 14 3.2. Influence of land cover .................................................................................................... 16 4. DISCUSSION ......................................................................................................... 21 4.1. Comparing NDVI values from UAV and Satellite ......................................................... 21 4.2. Finding the best rescaling method and influence factors. ................................................... 23 4.3. Land cover and Heterogeneity ......................................................................................... 24 5. CONCLUSIONS AND FUTURE PERSPECTIVES ........................................................ 25 SUPPLEMENTARY MATERIAL ....................................................................................... 27 ANNEX I – Orthophotos of study sites ................................................................................... 27 ANNEX II - NDVI maps of study sites based on UAV images ................................................ 31 ANNEX III – NDVI maps of study sites based on Sentinel-2 images .................................... 35 ANNEX IV – Land cover maps of study sites ......................................................................... 39 2 plant communities that inhabit riparian zones. Climatic change also leads to significant alterations at the physiological level of plants, namely leaf unfolding and flowering of plants in spring or colour changing and leaf fall in autumn (Gordo & Sanz, 2010), as well as an increase in the susceptibility of plant species to pathogens and pests, causing tree die-offs and changes in the distribution of vegetation at a regional level (Bodner & Robles, 2017; Breshears et al. , 2005). It is therefore of major importance to monitor these river corridors to understand their processes, characterize evolutionary trajectories, maintain their ecological sustainability and preserve them as a resource for future generations (Piégay et al. , 2020), (Tomsett & Leyland, 2019). 1.2. Remote sensing for monitoring ecosystems Earth observation (EO) can be defined as the gathering of information about the physical, chemical, and biological systems of planet Earth. It can be performed via remote-sensing technologies and by ground-based techniques ( International Journal of Applied Earth Observation and Geoinformation , 2012). As such, EO is instrumental in for monitoring ecosystems, for it provides information about changes in ecosystems at local, regional, and global scales, being a powerful tool for conservation planning (Vihervaara et al. , 2017). Due to the rapid changes occuring throughout Earth’s biosphere, quick spatio-temporal assessment is difficult using conventional methods, however, thanks to technological advances, remote sensing platforms now come with higher spatial resolutions (which in turn translates to a decrease in pixel area, and an increase in homogeneity of soil/vegetation cover characteristics inside the pixel), broad coverages and high revisit frequency, which facilitates in the acquisition of data (Bollas et al. , 2021; Torresani et al. , 2019; Westoby et al. , 2012).However, this also raises new challenges in terms of processing and software needs for conservation and biodiversity activities (Corbane et al. , 2015; He et al. , 2015; Lang et al. , 2015; Rocchini et al. , 2015). One of the growing fields in EO techniques in the past decades is remote sensing (from now on referred to as RS). RS can be defined as a range of techniques and methods used to monitor the earth’s resources and to acquire information about spatial objects and phenomena without physical contact (usually through platform mounted sensors) (Fig.1) (Bollas et al. , 2021; Piégay et al. , 2020; Pinter et al. , 2003). RS uses the electromagnetic spectrum (visible, infrared and microwaves) to extract data from the spectral reflectance characteristics of targets at a distance (Bollas et al. , 2021; Shanmugapriya et al. , 2019). 3 Figure 1. Elements and processes of a remote sensing system (modified from Walton, 1989). With a growth in multispectral and hyperspectral sensors, RS applications have been employed in different fields, such as crop growth monitoring, land use pattern and land cover changes, mapping of water resources and water status under field condition, monitoring of diseases and pest infestation, forecasting of harvest date and yield estimation, precision farming and weather forecasting purposes along with field observations (Atzberger, 2013; Di Gennaro et al. , 2019; Kingra et al. , 2016; Messina et al. , 2020; Shanmugapriya et al. , 2019). RS techniques (along with GIS) have recently been applicable in riparian zones with good results, for they allow the creation of spatio-temporal basic informative layers which can be successfully applied to diverse fields including flood plain mapping, hydrological modelling, surface energy flux, urban development and stress detection (Alleaume et al. , 2018; Barton, 2012; He et al. , 2015; Kingra et al. , 2016; Nezhad et al. , 2018; Pinter et al. , 2003; Rocchini et al. , 2015; Tomsett & Leyland, 2019). The usage of satellites as platforms for remote sensing is not a novelty for the scientific world (Brekke & Solberg, 2005; Holmgren & Thuresson, 1998; Martin, 2008; Tucker & Sellers, 1986; Verbyla, 1995), with studies dating back more than 40 years. Since the launch of the first civilian earth-observing satellite in 1972, satellite remote sensing has provided an ever increasing sophisticated information on the structures and functions of the earth’s surface (Iverson et al. , 1989), with modern satellites systems like Pléiades 1, KOMPSAT-3 and SuperView-1 offering an impressive resolution of just 0,5m/pxl. Sentinel-2 satellite has been extensively used (Bollas et al. , 2021; Cavur et al. , 2019; Di Gennaro et al. , 2019; Ghoussein et al. , 2019; Khaliq et al. , 2019; Messina et al. , 2020, 2020; 4 Nezhad et al. , 2018, 2019; Pace et al. , 2021; Revill et al. , 2020; Xu et al. , 2021) in different fields of knowledge (conservation, engineering, urban planning, etc) and it is a well-established and powerful remote sensing tool, most of the times chosen based on its decametric resolution in terms of space and time, with a ground sample distance of up to 10m, revisit time of six days, field of view of 290km and a free access dataset that is easily available. Similar to satellite platforms, UAV mounted sensors are being used more and more in scientific studies (Abdullah et al. , 2021; Berni et al. , 2009; Casado et al. , 2015; De Luca et al. , 2019; Dubbini et al. , 2015; Kislik et al. , 2018, 2020; Pontoglio et al. , 2021; Themistocleous, 2014) in recent years, mainly because these platforms are becoming increasingly more available and reliable, offering unrivalled spatial resolution over small and medium sized areas and a revisit time that’s basically defined by the user (Berni et al. , 2009; Klemas, 2015; Piégay et al. , 2020; Shanmugapriya et al. , 2019). However, both technologies have a series of pros and cons that involve technological, economic and operational factors. UAV platforms come with limitations that hinder wide scale implementation, such as a limited payload and short flight endurance (Matese et al. , 2015), while satellite surveys still present coarse resolutions for finer scale classifications, are subject to cloud cover and the fixed-timing acquisitions can, for instance, miss out on specific growth stages of vegetation (Matese et al. , 2015). Although there is familiarity with both platforms for ecological purposes, the conjoined use and, most importantly, the comparison of both Sentinel-2 and UAV images is still a very recent endeavour that scientists are trying to understand (Alvarez-Vanhard et al. , 2020; Bansod et al. , 2017; Bollas et al. , 2021; Di Gennaro et al. , 2019; Khaliq et al. , 2019; Messina et al. , 2020; Revill et al. , 2020). This comparison of data with different native resolution involves the application of spatial statistics and requires tackling the problem of spatial autocorrelation and although methods are becoming available to compare maps accounting for the spatial structures present in the data, the most practiced procedures still rely on cell-by-cell evaluations (Matese et al. , 2015). It’s also important to note that, even though UAV and satellite studies are being more commonplace (as shown above), studies using comparisons of both platforms in a riparian setting are still scarce (Gómez-Sapiens et al. , 2021; Huylenbroeck et al. , 2020). 1.3. Vegetation spectral indices Multispectral reflectance of the canopies is related to two important plant physiological processes (photosynthesis and evapotranspiration) (Kingra et al. , 2016). Several studies (Asner, 1998; Ceccato et al. , 2001; Datt, 1998; Gupta et al. , 2003; Pu et al. , 2003; Stimson et al. , 2005) 5 focused on the spectral reflectance properties of the plants, identifying key spectral wavebands related to plant physiological and structural properties and from there, derived vegetation spectral indices for their non-destructive estimation. The potential to spectrally estimate plant physiological properties over relatively large areas, and to predict plant water status and plant water stress has already been demonstrated in forestry species (Stimson et al. , 2005). RS data has been used to estimate canopy characteristics by using spectral indices based approach (D’Urso et al. , 2004). Chlorophyll pigments absorb radiation in the blue and red part of the electromagnetic spectrum and reflects in the green; nevertheless, the percentage of radiation reflected from the leaf is higher in the NIR than in the green (Chappelle et al. , 1992; Gausman et al. , 1971). The spectral reflectance of the leaf in healthy plants is characterized by high values of reflectance in the NIR region and low values in red portion (absorption) (Pinter et al. , 2003), while the opposite behaviour (more red light reflectance and more absorption in NIR) can be expected in plants subjected to stress. Numerous spectral vegetation indices (VIs) have been developed to characterize vegetation (Kingra et al. , 2016), but for the sake of this investigation, we shall only mention the Normalized Difference Vegetation Index (NDVI) proposed by Rouse et al. , (1973), as this was the method implemented in the experiment. NDVI has become a commonly used vegetation index to assess vegetation condition (Barton, 2012; Wallace et al. , 2004) for it allows to measure the state of the vegetation based on how it reflects light at certain frequencies. It’s designed to evidence photosynthetic activity from a surface, taking advantage of the strong contrast in vegetation reflectance observed between the red spectral and NIR spectral domain (Alleaume et al. , 2018). However, it has to be taken into account that the validity of NDVI values are influenced by many factors, such as: surface properties, anisotropic effects (position of the sun and observer) and atmospheric conditions (Alleaume et al. , 2018). 1.4. Study Objectives and Hypothesis With all these facts in mind, the main objectives for this study are: 1) Evaluate the accuracy of satellite data by comparing to UAV data; 2) Evaluate which rescale methods assure best fitness among drones and satellite; 3) Evaluate the influence of land cover on the discrepancy among drone and satellite data. Native data source is expected to have higher dispersion from corrected (rescaled) data since there’s a bigger difference in pixel area from both platforms. 6 Higher dispersion in NDVI is also expected in natural land cover compared to anthropogenic land cover. Considering that NDVI is a vegetation index (meaning higher values for green vegetation), the lowest NDVI values are expected to belong to anthropogenic land cover, regardless the platforms used. In addition, it is expected that discrepancies (dispersion of data) in NDVI values for man-made buildings will be the lowest and not influenced by the spatial scale. On the other hand, it’s expected that the higher NDVI values will be found in natural land cover. However, in this case, it is also expected a major miss match between platforms due to the heterogeneity of forest and bush vegetation (i.e the presence of riparian trees, shrubs and mixed vegetation can be spatial scale dependent) is also predictable. 2. MATERIALS AND METHODS 2.1. Study sites Four stream reaches in two river basins across the Northwest of Portugal were selected: the Lima and the Cávado River basins, draining to the Atlantic Ocean (Fig. 2). Figure 2. Catchment site’s location within NW Portugal. 7 4 study sites within the study area were chosen for UAV image capturing: Ribeira de Cabril, Rabagão and two sites in Rio Vez (Fig. 3). The sites were chosen based on the location of the flights previous to this study. Vez had two different sites (one before and other after the village of Arcos de Valdevez), as one of the original objectives was to evaluate the influence of human settlements in riparian ecosystem health. Although this objective was later abandoned, the sites chosen remained the same. 8 Figure 3. UAV orthophotos of study sites. (A), (B), (C) and (D) refer to Ribeira do Cabril, Rabagão and Rio Vez, respectively. (see ANNEX I) 9 In a general way, all sites are characterized by a corridor of riparian vegetation immediately adjacent to the river, with either one or both margins occupied mainly by small plots of farmland and variable areas with grass and bush vegetation cover. Manmade structures such as houses, sheds and barns are present in all sites, with most of the buildings having road accesses from main roadways. Study sites were named based on the tributary where the images were captured (Table 1) and from this point forward they will be addressed by their respective code. Table 1. Catchment sites specifications and coordinates. 2.2. Materials and Methods 2.2.1. UAV-Based Imagery UAV image acquisition took place at 3 different dates: the 10th of July 2018, the 5th of August 2019 and the 11th of October 2019. The chosen images were randomly picked from a database of UAV flights made for the CLIMALERT initiative before the investigation. Spatial resolution of captured UAV images is 0.08 m at ground level (Fig. 3). The multispectral sensor consists of five spectral cameras collecting blue, green, red, red edge and near infrared (NIR) imagery (Table 2). The duration of each flight was approximately 5-10 min, and the images were collected between 10:45 and 12:00, in clear sky conditions for all rivers. Flights were carried out at 100 m height from starting point. Mapping took part with an average overlap 80% forward and sideways. In flight triggers were: 233 at CAB1; 287 at RAB2; 351 at VEZ2; 409 at VEZ3. Table 2. Bands and their wavelengths for both Sentinel-2 (adapted from Bertini et al. , 2012) and Micasense Rededge™ (RedEdge User Manual (PDF)) camera mounted on the DJI Phantom 4 RTK. Code Stream name River Basin Latitude Longitude CAB1 Ribeira de Cabril Cávado 41.721487 -8.032822 RAB2 Rabagão Cávado 41.71956 -7.903389 VEZ2 Rio Vez Lima 41.896635 -8.438900 VEZ3 Rio Vez Lima 41.815080 -8.426453 Sensing Platform Band Number Band Central Wavelength (nm) Bandwidth (nm) Spatial Resolution (m) Sentinel -2 1 Violet 443 20 60 10 2.2.2. Satellite time series imagery The Sentinel-2 mission is a two satellite (Sentinel-2A and Sentinel-2B) constellation launched by the Copernicus European Program for Earth observation, providing high-resolution, multispectral images (European Space Agency, 2015). The captured data of Sentinel-2 ranges from the visible to the shortwave infrared parts of the electromagnetic spectrum with 13 spectral bands at 3 different spatial resolutions. Satellite data was downloaded via https://scihub.copernicus.eu (accessed on 5 June 2021) from both the S2A (CAB1 and RAB1) and S2B (VEZ2 and VEZ3) satellites with the spatial resolution of 10 m at ground level. All acquired images are located within the 29TNG tile (UTM tiling grid) from a Level-2A product (atmospherically corrected). Images for RAB2 are from 9 July 2018 at 14:23:05 UTC (10.59% cloud cover), for CAB1 from 3 August 2019 at 14:18:08 UTC (0.89% cloud cover) and for both VEZ2 and VEZ3 from 10 October 2019 at 14:13:58 UTC (0.17% cloud cover) and the images were chosen based on the temporal proximity (1-2 days) to the previously made UAV flight dates to allow for a more viable correlation between the data, as temporal differences account for different stages of vegetation growth, and as such, different NDVI values. 2.3. Earth observation data For this study, all image post-processing was done using QGIS v3.16.15 “Hannover” (long term release) software. UAV orthophotos (Fig 4) were used to classify land cover in all sites for 2 Blue 490 65 10 3 Green 560 35 10 4 Red 665 30 10 5 Rededge 705 15 20 6 Near Infrared 740 15 20 7 783 20 20 8 842 115 10 8b 865 20 20 9 945 20 60 10 1380 30 60 11 Short Wavelength Infrared 1610 90 20 12 2190 180 20 Micasense Rededge™ 1 Blue 475 20 0.08 2 Green 560 20 3 Red 668 10 4 Near Infrared 840 40 5 Rededge 717 10 11 they allow by far the best perception of the study site, enabling a more trustworthy description and classification of land cover. Classification was achieved by manually “drawing” each different polygon and labelling them accordingly. The reason for manually doing this instead of using an object-based machine learning image classification software was that most of the software’s tried and easily available didn’t have enough precision to correctly classify different land covers at the scale of the UAV images (some polygons have areas of <0.2m2). 7 distinct categories were attributed to the polygons based on their different characteristics: River; Road; Bush Vegetation; Forest Vegetation; Grass; Farm; Man Made Structure (see ANNEX IV). 2.4. Normalized Difference Vegetation Index The normalized difference vegetation index (NDVI) as proposed by Rouse et al. , (1973) was designed to evidence photosynthetic activity from a surface, taking advantage of the strong contrast in vegetation reflectance observed between the red spectral and NIR spectral domain. Taking only into account these two spectral bands, NDVI is calculated as seen bellow in Equation 1: NDVI = NIR-RED NIR+RED (1) where NIR stands for near infrared band reflectance and RED for the red band reflectance (see Table 2 for more information regarding the bands). 2.5. Data processing and analysis A well-defined step-by-step process was created for the conduction of data processing and analysis to achieve all goals of the study (Fig. 5). All images were registered to CRS WGS84/UTM zone 29N with EPSG:32629. NDVI maps were obtained using the Equation 1 in the “Raster Calculator” tool using the respective NIR and Red bands for each platform (Fig. 6). 18 Table 4. Results of linear regression model applied to AV algorithm. Estimate Std. Error t value Intercept 0.122 0.005 24.271 ndvi_sat 0.528 0.006 92.155 RAB2 0.038 0.002 16.117 VEZ2 0.027 0.002 11.919 VEZ3 0.0215 0.002 8.975 Farm -0.028 0.003 -8.645 Forest 0.041 0.003 12.452 Grass -0.042 0.004 -10.511 Manmade -0.1 0.006 -15.608 River -0.098 0.005 -20.322 Road -0.065 0.006 -6.016 R2 0.483 p value < 2.2e-16 Like previous linear regression models applied to total NDVI values, significant relationships (p-value = < 2E-16) were found among all linear regression models applied this time. This tells us that land cover and catchment site are significant parameters in influencing the dispersion of values within upscaling algorithms. GOF measures for 3Q model showing the lowest RMSE (RMSE = 0.12) are reported in Table 5 and Fig. 12. Highest value of RMSE (RMSE = 0.510) is seen in native resolution while the lowest in seen in 3Q rescale algorithm (RMSE= 0.12). Average and median rescaled algorithms show similar RMSE values with median Average=0.225 and median Median=0.235. In contrast, R2 had the lowest value in native resolution (R2 = 0.00) and the highest in average upscaling method (R2 = 0.670). Table 5. Calculated GOF values for 3Q algorithm. LandUse River RMSE R2 Forest CAB1 0.12 0.21 RAB2 0.14 0.26 VEZ2 0.14 0.14 VEZ3 0.14 0.27 Bush CAB1 0.15 0.28 RAB2 0.16 0.15 VEZ2 0.13 0.28 VEZ3 0.12 0.32 Farm CAB1 0.18 0.53 RAB2 0.24 0.01 19 VEZ2 0.14 0.59 VEZ3 0.15 0.54 Road CAB1 0.16 0.39 RAB2 0.16 0.17 VEZ2 0.18 0.28 VEZ3 0.16 0.39 Man CAB1 0.25 0.1 RAB2 0.15 0.3 VEZ2 0.17 0.54 VEZ3 0.18 0.36 River CAB1 0.2 0.16 RAB2 0.16 0.36 VEZ2 0.21 0.06 VEZ3 0.29 0.07 Grass CAB1 0.17 0.1 RAB2 0.12 0.61 VEZ2 0.14 0.26 VEZ3 0.17 0.28 Figure 11. Boxplot of GOF measures resulting from linear regression model to evaluate best fit of rescale to satellite values. Data subsets containing the GOF values of land cover and type within each different method were created to assess the cause of discrepancy within methods (Fig. 13). 20 Figure 12. Boxplot representation of linear regression applied to 3Q algorithm subset of GOF measures within a rescaling method. In Fig. 14 the relationship between Satellite NDVI with UAV NDVI for 3Q Algorithms. Different trend lines are present for each category of land cover. Figure 13. Facet_wrap plot for CAB1. (see ANNEX VII) 21 4. DISCUSSION 4.1. Comparing NDVI values from UAV and Satellite An expected higher degree of dispersion between native resolutions was confirmed, with CAB1 and RAB2 showing the strongest dispersion values of the 4 sites. This is because the biggest discrepancy between pixel area is at native level. While higher NDVI values follow a similar trend both in satellite and UAV (between 0.5 and 1) at native resolution, lower values are much more present in UAV native image capturing. This can be explained by the much higher resolution of the UAV being able to capture more pixels with low values over the same area as satellite. When comparing rescaling methods, NDVI values show less dispersion, but at the same time, because of the loss of information associated with upscaling process (Fig. 7) (Messina et al. , 2020) the homogenization of values is high, especially at lower values. A trend (peak) in higher NDVI values can be seen in every rescale. The same peak is present in satellite image histograms for values between 0,5 and 1 (see ANNEX V). 22 Figure 14. CAB1 NDVI satellite image (A) and UAV rescaled images derived from “Raster Warp” processing of the native resolution. (B), (C) and (D) correspond to the median, average and third quartile rescaling methods, respectively. 23 4.2. Finding the best rescaling method and influence factors. The proposal for “best method” is based on the one that differs less from the satellite image NDVI values, in other words, the one which produces less error during the upscaling process. Because RMSE measures how far apart the predicted values are from the observed values in a dataset, the choice for best rescaling method was based upon the RMSE values present at Table 5 and Fig. 12. It was concluded that out of the three methods, upscaling via third quartile (3Q) had the best fit with satellite values (RMSE median values (0,16); p-values < 2e-16). By creating a subset of GOF measures by land cover and river within the best method (3Q) we can find an explanation for dispersion within the group. R2 is used to measure how statistically similar values in the two datasets are (using a simple linear regression model). This explains the variation within the model or in other words, how land cover or river site affects these values. As confirmed in Fig. 13, river sites have very similar median R2 values, which means they all contribute with about the same level to the dispersion of values, hence, it’s concluded they’re not the main reason for value differences. On the contrary, by examining land cover, it is evident that there are big differences between R2 values. Fig. 13 accounts for catchment site in the land cover values, so what we see is the total R2 for that land cover within all rivers. Greatest disparity is confirmed in Farm, and by cross referencing this data with a clustered column chart of GOF measures in the subset (see ANNEX VIII) we can see that the lowest value comes from RAB2. Revisiting the map in QGIS (Fig. 16) gives insight regarding the lowest R2 value (0.01). RAB2 has the whole east margin covered in farmland. Although there was no in situ validation to check if farms were monoculture, at the time the images were captured, some fields were uncultivated, as fallow ground is still widely used in agriculture during crop rotation (Collins et al. , 1992) and/or presented different growth stages. Low overall average R2 (0.22) for “Forest Vegetation” can be associated with different plants that make up the riparian forest community, as different species tend to have different NDVI values. “Forest Vegetation” polygons were drawn based on the tree canopies visible by the UAV orthophotos, ignoring bushes and grasses that make up the entire forest per se , which in turn, have different NDVI values than those of the tree canopies. These NDVI values can give us a very coarse idea of plant species richness (Fairbanks & McGwire, 2004), but because there was no ground validation for the study, further research is needed to validate. Lowest overall R2 belongs to the land class River, probably because the thin and very shallow water column of the analysed streams allowed satellite and UAV to capture NDVI values from river substrate, 24 algae and macrophytes. Although not accounted for during land classification, algae and macrophyte communities do exist in these rivers (Dodkins et al. , 2012; Ribeiro & Torgo, 2008) and account for different NDVI values. Figure 15. RAB2 farmland examination where (A) is the NDVI map with correspondent randomised points and (B) is the division of land cover with the same points. 4.3. Land cover and Heterogeneity The highest value accounted for % of area coverage in all sites was that of farmland in RAB2 with almost half the map (48,55%), while in the same site we get the lowest value of all, with just 0,98% coverage by manmade structures. On average, throughout all sites, farmland is the type of land cover with more % of area covered, with an average of 41,32 and following the same order, manmade structures only account for 2,12% of total area coverage. Regarding the number of polygons, forest vegetation shows the greater number of polygons per map in CAB1 with 167 polygons of the class present (or 29,98% of the map) being, at the same time, the most polygon rich class across all sites, with 23,58% (or 107,25) of total polygons. On the low end is river polygons with 8,02% (32,75) polygon coverage over all sites, and with the lowest value of all in VEZ3 with only 2 polygons to account for the entire river sections (see ANNEX IX). 25 5. CONCLUSIONS AND FUTURE PERSPECTIVES The ecological importance of this thesis resides in the ability to correlate satellite and UAV imagery, hence, being able to get the best of both methods in conservation efforts regarding plant community health in riparian zones. In this study, a detailed analysis and comparison of multispectral imagery of riparian zones in NW Portugal, is presented with the aim to calculate NDVI disparity between both methods, at native as well as rescaled (for UAV) resolutions and based on these results find the upscale method that is closer to satellite imagery. Statistical comparison between NDVI values at native resolution show, as expected, a bigger difference in values than those found in UAV upscaled version, due to the higher spatial resolution of the UAV’s sensor. Upscaling via third quartile seems to be the closest rescale method to satellite (RMSE < 0.2) when it comes to measuring NDVI values in riparian zones with similar characteristics than those in this study. However, we concluded that homogenization of values tends to be highest at lower NDVI values, because UAV image upscaling tends to attribute smaller pixel value than those found in satellite images, which can contribute to reading errors. Land cover was proven to be the main factor influencing value dispersion within each method, primarily caused by farmland, most of which, at the time of image capturing, had a mix of fallow and cultivated and/or different crop growth stages. Further studies within a time frame that represents a full vegetative season should be considered to greatly reduce or even eliminate this error. Forest and bush vegetation data should also be complemented in further studies with ground validation, for a more accurate classification of riparian ecosystems. In conclusion, both platforms provide important information for the vegetation cover and land cover on riparian zones, and they are proven again to be important tools for conservation work. The choice for the most appropriate platform depends mainly on the use and the aim of the intended data, as they have different spatial resolutions, cost, and requirements. Sentinel-2 is a valuable platform when information for areas of large extent is needed and therefore is not the optimal method to evaluate ecosystems as complex as riparian zones or as spatially small as the farms of the chosen sites, as these farms resource to farming practices which are typical of this region of Portugal (small, highly fragmented plots of cultivated land, usually with a mix of crops in a small area). As such, UAV platforms are a better choice when detailed information is required. The lack of studies on the comparison of multispectral data retrieved between Sentinel-2 and UAV imagery was what motivated the explorative approach taken in the present work. Most of 26 the times it proved to be an obstacle, since no term of comparison and further discussion could be traced between the results found here and the ones described in the literature. As remote sensing techniques are gradually picking up the pace to become the standard method for evaluation of ecosystem health throughout the globe. Although further studies and techniques in correlating data between UAV and satellite need to be developed, this study demonstrated the potential for the comparison of multispectral data to interpret riparian zone ecosystems. 27 SUPPLEMENTARY MATERIAL ANNEX I – Orthophotos of study sites Figure I A. CAB1 stream reach orthophoto. 34 Figure II D. VEZ3 stream reach UAV NDVI map. 35 ANNEX III – NDVI maps of study sites based on Sentinel-2 images Figure III A. CAB1 stream reach Sentinel-2 NDVI map. 36 Figure III B. RAB2 stream reach Sentinel-2 NDVI map. 37 Figure III C. VEZ2 stream reach Sentinel-2 NDVI map. 38 Figure III D. VEZ3 stream reach Sentinel-2 NDVI map. 39 ANNEX IV – Land cover maps of study sites Figure IV A. Land cover classification map of CAB1. 40 Figure IV B. Land cover classification map of RAB2. 41 Figure IV C. Land cover classification map of VEZ2. 42 Figure IV D. Land cover classification map of VEZ3. 43 ANNEX V – Histograms of NDVI values Figure V A. Histograms of UAV native resolution NDVI values. (A), (B), (C) and (D) correspond to CAB1, RAB2, VEZ2 and VEZ3, respectively. 50 Table VI C. Results of linear regression model applied to MN algorithm. Estimate Std. Error t value Intercept 0.114 0.007 16.722 ndvi_sat 0.561 0.008 71.891 RAB2 0.047 0.003 14.703 VEZ2 0.036 0.003 11.697 VEZ3 0.029 0.003 8.997 Farm -0.046 0.004 -10.673 Forest 0.028 0.005 6.299 Grass -0.062 0.005 -11.459 Manmade -0.134 0.009 -15.404 River -0.108 0.007 -16.562 Road -0.076 0.008 -9.933 R2 0.367 p value < 2.2e-16 51 Table VI D. Results of linear regression model applied to AV algorithm. Estimate Std. Error t value Intercept 0.122 0.005 24.271 ndvi_sat 0.528 0.006 92.155 RAB2 0.038 0.002 16.117 VEZ2 0.027 0.002 11.919 VEZ3 0.0215 0.002 8.975 Farm -0.028 0.003 -8.645 Forest 0.041 0.003 12.452 Grass -0.042 0.004 -10.511 Manmade -0.1 0.006 -15.608 River -0.098 0.005 -20.322 Road -0.065 0.006 -6.016 R2 0.483 p value < 2.2e-16 52 ANNEX VII – Point dispersion plots for UAV rescaled images Figure VII A. Comparison between UAV and satellite NDVI values at native and rescaled resolutions in CAB1. 53 Figure VII B. Comparison between UAV and satellite NDVI values at native and rescaled resolutions in RAB2. 54 Figure VII C. Comparison between UAV and satellite NDVI values at native and rescaled resolutions in VEZ2. 55 Figure VII D. Comparison between UAV and satellite NDVI values at native and rescaled resolutions in RAB2. 56 ANNEX VIII - Stacked column chart for land cover based on R2 measurements Figure VIII A. Stacked column chart for land cover based on R2 measurements. 0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 R2 57 ANNEX IX – Land cover occupancy by polygon n.º and area Table IX A. Land cover occupancy values in % and n.º of polygons at all river reaches. 58 ANNEX X – Facet wraps plots of value dispersion taking land cover into account. Figure X A. Facet graph plot for CAB1. 59 Figure X B. Facet graph plot for CAB1. 66 Martin, R. V. (2008). Satellite remote sensing of surface air quality. Atmospheric Environment , 42 (34), 7823–7843. https://doi.org/10.1016/j.atmosenv.2008.07.018 Matese, A., Toscano, P., Di Gennaro, S. F., Genesio, L., Vaccari, F. P., Primicerio, J., Belli, C., Zaldei, A., Bianconi, R., & Gioli, B. (2015). Intercomparison of UAV, Aircraft and Satellite Remote Sensing Platforms for Precision Viticulture. Remote Sensing , 7 (3), 2971–2990. https://doi.org/10.3390/rs70302971 Messina, G., Peña, J. M., Vizzari, M., & Modica, G. (2020). A Comparison of UAV and Satellites Multispectral Imagery in Monitoring Onion Crop. An Application in the ‘Cipolla Rossa di Tropea’ (Italy). Remote Sensing , 12 (20), 3424. https://doi.org/10.3390/rs12203424 Mligo, C. (2017). Diversity and distribution pattern of riparian plant species in the Wami River system, Tanzania. Journal of Plant Ecology , 10 (2), 259–270. https://doi.org/10.1093/jpe/rtw021 Nezhad, M. M., Groppi, D., Laneve, G., Marzialetti, P., & Piras, G. (2018, April). Oil Spill Detection Analyzing “Sentinel 2” Satellite Images: A PersianGulf Case Study . The 3rd World Congress on Civil, Structural, and Environmental Engineering. https://doi.org/10.11159/awspt18.134 Nezhad, M. M., Heydari, A., Fusilli, L., & Laneve, G. (2019, April). Land Cover Classification by using Sentinel-2 Images: A case study in the city of Rome . The 4th World Congress on Civil, Structural, and Environmental Engineering. https://doi.org/10.11159/iceptp19.158 Oliveira, D., Fernandes, A., Rapazote, J., Teixeira, L., Brioso, R., Valente, C., Teixeira, F., & Medeiros, M. (2004). A UTILIZAÇÃO DO “RIVER HABITAT SURVEY” NA GESTÃO AMBIENTAL DE ECOSSISTEMAS FLUVIAIS. POTENCIALIDADES E LIMITAÇÕES. 15. Pace, G., Gutiérrez-Cánovas, C., Henriques, R., Boeing, F., Cássio, F., & Pascoal, C. (2021). Remote sensing depicts riparian vegetation responses to water stress in a humid Atlantic region. The Science of the Total Environment , 772 , 145526. https://doi.org/10.1016/j.scitotenv.2021.145526 Piégay, H., Arnaud, F., Belletti, B., Bertrand, M., Bizzi, S., Carbonneau, P., Dufour, S., Liébault, F., Ruiz‐Villanueva, V., & Slater, L. (2020). Remotely sensed rivers in the Anthropocene: State of the art and prospects. Earth Surface Processes and Landforms , 45 (1), 157–188. https://doi.org/10.1002/esp.4787 Pievani, T. (2014). The sixth mass extinction: Anthropocene and the human impact on biodiversity. Rendiconti Lincei , 25 (1), 85–93. https://doi.org/10.1007/s12210-013-0258-9 Pinter, Jr., Paul J., Hatfield, J. L., Schepers, J. S., Barnes, E. M., Moran, M. S., Daughtry, C. S. T., & Upchurch, D. R. (2003). Remote Sensing for Crop Management. Photogrammetric Engineering & Remote Sensing , 69 (6), 647–664. https://doi.org/10.14358/PERS.69.6.647 Pontoglio, E., Dabove, P., Grasso, N., & Lingua, A. M. (2021). Automatic Features Detection in a Fluvial Environment through Machine Learning Techniques Based on UAVs Multispectral Data. Remote Sensing , 13 (19), 3983. https://doi.org/10.3390/rs13193983 Postel, S., & Richter, B. (2003). Rivers for Life: Managing Water For People And Nature . Island Press. Pu, R., Ge, S., Kelly, N. M., & Gong, P. (2003). Spectral absorption features as indicators of water status in coast live oak ( Quercus agrifolia ) leaves. International Journal of Remote Sensing , 24 (9), 1799–1810. https://doi.org/10.1080/01431160210155965 Quinn, J. M., Brown, P. M., Boyce, W., Mackay, S., Taylor, A., & Fenton, T. (2001). Riparian Zone Classification for Management of Stream Water Quality and Ecosystem Health1. JAWRA 67 Journal of the American Water Resources Association , 37 (6), 1509–1515. https://doi.org/10.1111/j.1752-1688.2001.tb03656.x RedEdge User Manual (PDF)—Legacy . (n.d.). MicaSense Knowledge Base. Retrieved 24 February 2022, from https://support.micasense.com/hc/en-us/articles/215261448-RedEdgeUser-Manual-PDF-Legacy Revill, A., Florence, A., MacArthur, A., Hoad, S., Rees, R., & Williams, M. (2020). Quantifying Uncertainty and Bridging the Scaling Gap in the Retrieval of Leaf Area Index by Coupling Sentinel-2 and UAV Observations. Remote Sensing , 12 (11), 1843. https://doi.org/10.3390/rs12111843 Ribeiro, R., & Torgo, L. (2008). A comparative study on predicting algae blooms in Douro River, Portugal. Ecological Modelling , 212 (1–2), 86–91. https://doi.org/10.1016/j.ecolmodel.2007.10.018 Rocchini, D., Hernández-Stefanoni, J. L., & He, K. S. (2015). Advancing species diversity estimate by remotely sensed proxies: A conceptual review. Ecological Informatics , 25 , 22–28. https://doi.org/10.1016/j.ecoinf.2014.10.006 Rouse, J., Haas, R. H., Schell, J. A., & Deering, D. (1973). Monitoring vegetation systems in the great plains with ERTS. Undefined . https://www.semanticscholar.org/paper/Monitoringvegetation-systems-in-the-great-plains-RouseHaas/fb2f60fe0fe2874e5cbf927a2556d719c32eac29 Rundle, S. (2002). Threats to the Running Water Ecosystems of the World. Environmental Conservation , 29 , 134–153. https://doi.org/10.1017/S0376892902000097 Sabater, S. (2008). Alterations of the Global Water Cycle and their Effects on River Structure, Function and Services. Freshwater Reviews , 1 (1), 75–88. https://doi.org/10.1608/FRJ1.1.5 Santos, J. M., Godinho, F., Ferreira, M. T., & Cortes, R. (2004). The organisation of fish assemblages in the regulated Lima basin, Northern Portugal. Limnologica , 34 (3), 224– 235. https://doi.org/10.1016/S0075-9511(04)80047-1 Shanmugapriya, P., Rathika, S., Ramesh, T., & Janaki, P. (2019). Applications of Remote Sensing in Agriculture—A Review. International Journal of Current Microbiology and Applied Sciences , 8 (01), 2270–2283. https://doi.org/10.20546/ijcmas.2019.801.238 Stimson, H. C., Breshears, D. D., Ustin, S. L., & Kefauver, S. C. (2005). Spectral sensing of foliar water conditions in two co-occurring conifer species: Pinus edulis and Juniperus monosperma. Remote Sensing of Environment , 96 (1), 108–118. https://doi.org/10.1016/j.rse.2004.12.007 ten Harkel, J., Bartholomeus, H., & Kooistra, L. (2019). Biomass and Crop Height Estimation of Different Crops Using UAV-Based Lidar. Remote Sensing , 12 (1), 17. https://doi.org/10.3390/rs12010017 Themistocleous, K. (2014). The Use of UAV Platforms for Remote Sensing Applications: Case Studies In Cyprus . 9229 , 92290S. https://doi.org/10.1117/12.2069514 Tilman, D., Clark, M., Williams, D. R., Kimmel, K., Polasky, S., & Packer, C. (2017). Future threats to biodiversity and pathways to their prevention. Nature , 546 (7656), 73–81. https://doi.org/10.1038/nature22900 Tilman, D., Reich, P. B., & Isbell, F. (2012). Biodiversity impacts ecosystem productivity as much as resources, disturbance, or herbivory. Proceedings of the National Academy of Sciences of the United States of America , 109 (26), 10394. https://doi.org/10.1073/pnas.1208240109 68 Tomsett, C., & Leyland, J. (2019). Remote sensing of river corridors: A review of current trends and future directions. River Research and Applications , 35 (7), 779–803. https://doi.org/10.1002/rra.3479 Torresani, M., Rocchini, D., Sonnenschein, R., Zebisch, M., Marcantonio, M., Ricotta, C., & Tonon, G. (2019). Estimating tree species diversity from space in an alpine conifer forest: The Rao’s Q diversity index meets the spectral variation hypothesis. Ecological Informatics , 52 , 26–34. https://doi.org/10.1016/j.ecoinf.2019.04.001 Tucker, C. J., & Sellers, P. J. (1986). Satellite remote sensing of primary production. International Journal of Remote Sensing , 7 (11), 1395–1416. https://doi.org/10.1080/01431168608948944 van der Meij, B., Kooistra, L., Suomalainen, J., Barel, J. M., & De Deyn, G. B. (2017). Remote sensing of plant trait responses to field-based plant–soil feedback using UAV-based optical sensors. Biogeosciences , 14 (3), 733–749. https://doi.org/10.5194/bg-14-733-2017 Verbyla, D. L. (1995). Satellite Remote Sensing of Natural Resources . CRC Press. Vieira, J. M. P., Duarte, A. A. L. S., & Pinho, J. L. S. (1998). APLICAÇÃO DE UM MODELO DE QUALIDADE DA ÁGUA DO RIO CÁVADO AO ESTUDO DO IMPACTO DA CAPTAÇÃO DE ÁGUA DE AREIAS DE VILAR . 17. Vihervaara, P., Auvinen, A.-P., Poikolainen, L., Törmä, M., Ahlroth, P., Anttila, S., Böttcher, K., Forsius, M., Heino, J., Heliölä, J., Koskelainen, M., Kuussaari, M., Meissner, K., Ojala, O., Tuominen, S., Viitasalo, M., & Virkkala, R. (2017). How Essential Biodiversity Variables and remote sensing can help national biodiversity monitoring. Global Ecology and Conservation , 10 , 43–59. https://doi.org/10.1016/j.gecco.2017.01.007 Vitousek, P. M., Mooney, H. A., Lubchenco, J., & Melillo, J. M. (1997). Human Domination of Earth’s Ecosystems. Science , 277 (5325), 494–499. https://doi.org/10.1126/science.277.5325.494 Wall, D. H., Palmer, M. A., & Snelgrove, P. V. R. (2001). Biodiversity in Critical Transition Zones between Terrestrial, Freshwater, and Marine Soils and Sediments: Processes, Linkages, and Management Implications. Ecosystems , 4 (5), 418–420. Wallace, J. F., Caccetta, P. A., & Kiiveri, H. T. (2004). Recent developments in analysis of spatial and temporal data for landscape qualities and monitoring. Austral Ecology , 29 (1), 100– 107. https://doi.org/10.1111/j.1442-9993.2004.01356.x Westoby, M. J., Brasington, J., Glasser, N. F., Hambrey, M. J., & Reynolds, J. M. (2012). ‘Structurefrom-Motion’ photogrammetry: A low-cost, effective tool for geoscience applications. Geomorphology , 179 , 300–314. https://doi.org/10.1016/j.geomorph.2012.08.021 Xu, J., Zhao, J., Wang, F., Chen, Y., & Lee, Z. (2021). Detection of Coral Reef Bleaching Based on Sentinel-2 Multi-Temporal Imagery: Simulation and Case Study. Frontiers in Marine Science , 8 . https://www.frontiersin.org/article/10.3389/fmars.2021.584263 Yadav, N. S., Kumar, A., & Sharma, M. P. (2014). Ecological Health Assessment of Chambal River using Water Quality Parameters. Journal of Integrated Science and Technology , 2 (2), 52– 56.