Mid-Term Monitoring of Suspended Sediment Plumes of Greek Rivers Using Moderate Resolution Imaging Spectroradiometer (MODIS) Imagery

Abstract

This study focuses on the suspended sediment delivery of 17 rivers and streams of various sizes to the sea over a wide geographical area covering most of the Greek peninsula, utilizing two Moderate Resolution Imaging Spectroradiometer (MODIS) products. Equal-area polygons (“plume” polygons), were delineated at the mouths of each selected river. These polygons were utilized to estimate the suspended sediment load of each river through the application of suspended sediment indices, ratios, and masks. To achieve this, 669 Level 1B MODIS images (MOD02) and their corresponding MODIS cloud products (MOD35) were downloaded and processed for a 10-water-year period (2004–2014). During this period of 669 days, there were 58 flood events (episodes) ranging in duration from 5 to 45 days. Relative atmospheric correction was applied to the images based on four selected bright invariant areas (PIFs) scattered along mainland Greece. The second product used in this study was MOD09Q1, an atmospherically corrected 8-day composite processed for the entire record period (2000–2019). Suspended sediment indices, ratios, and masks were developed using all three visible channels and near-infrared (NIR) for the MOD02 dataset, while only Red and Near-InfraRed (NIR) channels were available from the MOD09Q dataset. The resulting rankings from the remote sensing analysis were compared with the predictions of soil loss models, and the outcomes were largely consistent. While the remote sensing results can be considered as a type of experimental data or measurements, they come with inherent limitations. These include infrequent access to cloud-free data on stormy days, the influence of wind and currents, and the potential impact of dust storms originating from Africa, among others. On the other hand, soil loss models are sensitive to the parameter values used, and in some cases, the uncertainties are significant. Hence, the ranking derived from remote sensing can serve as a calibration of the models, particularly for the BQART model, which provides information on the catchment’s sink capacity. An index of “sediment productivity per square kilometer and mm of rainfall” was developed. This index can be considered a “sediment delivery ratio” and is crucial for accurately quantifying the phenomenon.

1. Introduction

Sediment discharge refers to the amount of eroded material that is transported to a particular location in a river system within a specific timeframe, typically measured in M/T. Sediment yield refers to the accumulated sediment discharge per unit area of the upstream catchment over a specific time period, typically one year, and it is commonly expressed as t/km² [1]. River sediment transport is typically classified into three main types: suspended load, bedload, and dissolved load [2,3]. Among the two components of load transport, suspended particulate and dissolved material, suspended particulate material, which includes hard soil particles (sand, silt and clay), soft particles (vegetal debris) and colloidal particles (organic or inorganic), is generally considered much more significant than dissolved material [4]. This is particularly true for rivers and streams in the Mediterranean region. The concentration of suspended sediment in large catchments is influenced by various factors, including climatic, morphological, mineralogical and tectonic conditions, the size of the floodplains, and the hydrodynamic characteristics of the river and its tributaries [5,6,7,8,9]. Additionally, other factors such as biogeochemical processes, flash floods, and sediment heterogeneity due to grain size distribution, sedimentation, and tributary input can further complicate the distribution patterns of suspended sediment concentrations in a particular catchment [10].

Monitoring the variability of sediment flux in various river systems contributes to the understanding of landscape formation, the quantification of biogeochemical processes, and the evaluation of the impacts of environmental changes of both natural and anthropogenic origins [11,12]. Knowledge and understanding of the transport of total suspended sediments in large fluvial systems are essential for evaluating land use, tectonic regimes, and the hydrophysical, geomorphological, and ecological functioning of catchments [13] including the evolution of soil formation, river bank stabilization, hillslopes, channel-floodplain systems, biogeochemical cycles of organic and inorganic compounds, human intervention and sedimentation rates [14,15,16]. Moreover, the dynamics of suspended sediment production and transport in large rivers are crucial geomorphological processes that can significantly impact biodiversity [17].

Multiple efforts have been made to propose a forecasting framework for suspended sediment yield for specific regions of the Mediterranean basin or other areas around the world. Many of these models fall under the category of scoring models, including the widely recognized Pacific Southwest Inter Agency Committee (PSIAC) [18] and encompass the widely used and internationally applied Universal Soil Loss Equation (USLE) formula and its modifications [19,20,21,22]. An additional group is the global sediment yield models, which aggregate information from extensive databases encompassing numerous monitoring stations across the globe [23,24,25,26,27]. Among these models, the BQART model devised by [28] stands out as one of the most comprehensive examples. A significant advantage of the BQART model, when compared to its earlier versions (ART), is its capacity to incorporate variables like human disruptions affecting the catchments and the efficiency of sediment trapping. It also permits the inclusion of factors such as the lithology and the glacial characteristics of the catchment. Socio-economic information, with regards to human presence, which is now quite pronounced on the majority of the rivers worldwide [29,30], is represented in the model by the combination of two common indices, GDP per capita and population density, for the specific catchment area.

Several key land and water quality (vegetation, temperature, concentrations of suspended particulate matter and polychlorinated biphenyl, aquatic plants) and morphological (shorelines, mudbanks, wetlands) parameters, in different types of coastal environments (bays, estuaries, sandy and muddy systems), can be remotely sensed at various spatial and temporal scales, using innovative methods and providing validated products [31]. The availability of repetitive, synoptic and multi-spectral data from various satellite platforms, viz. IRS, LANDSAT, and SPOT has helped to generate information on varied aspects of the coastal and marine environment. Ocean color data from OCANSAT I, OCM, SeaWiFS, and MODIS provide information on biological aspects useful for fisheries and coastal ecosystems [32]. Regional-scale data are generally obtained from moderate-resolution remote sensing images, such as MODIS sensors and Landsat satellites [33].

Rivers and streams are the primary agents delivering erosional sediment materials to the oceans through sediment transport [34,35,36,37]. Suspended sediment concentration can directly influence the turbidity and color of water bodies [38,39]. Although the temporal–spatial distribution of suspended sediment concentration in estuarine and coastal waters is quite complicated and strongly related to the variations of seasonal riverine discharges, and oceanographic conditions in the receiving basin [40,41], remote sensing is a well-established technology for quantifying, mapping, and monitoring total suspended material concentrations in both oceanic and coastal waters [42]. Satellite remote sensing technology, which emerged in the 1960s as a method of gathering information from a distance, is immensely valuable for delivering synoptic and cost-effective estimates of water quality and suspended sediment concentration [43]. Satellite-based ocean color observation can offer nearly daily comprehensive perspectives on the dispersion of suspended sediments, aquatic substances, and concentrations across vast spatial and temporal coverage, a capability not attainable through other sources [12]. Hence, the utilization of satellite imagery for monitoring suspending sediment concentration in coastal and ocean waters is particularly significant in distance settings and complex environments where the presence of monitoring stations is limited [44]. In addition, satellite remote sensing can serve as a swift and cost-effective alternative for evaluating suspended sediment concentration not just in rivers and river deltas, but also in oceans and seas [45,46].

Since 2002, the advent of the new era of freely available, daily, medium-resolution imagery from the MODIS instrument onboard the TERRA and AQUA platforms of NASA’s EOS program certainly had a great impact on the spread of coastal remote sensing applications. MODIS multispectral data, despite their relatively low resolution, can reflect information on land surface condition, atmospheric water vapor, aerosol and surface temperature, atmospheric temperature, ocean color, phytoplankton, sea surface temperature, sea surface salinity, chlorophyll-a and many other properties [33,47]. A daily time resolution is essential for the monitoring of rapidly evolving, dynamic phenomena such as the movement of sediment-laden plumes from rivers into the sea during and after floods.

Numerous total suspended matter applications of MODIS imagery have been carried out in diverse locations, including the Mississippi River and the Pontchantrain Lake estuary in the northern Gulf of Mexico [38], the Bahmasheer River estuary in the south Persian Gulf [48], Mayaguez Bay in Puerto Rico [49], Tampa Bay, Florida [50], Adur River, southern Gulf of Biscay [51], Lower Yangtze River, China [52]. River plumes have also been monitored with MODIS imagery for a variety of purposes in many places around the world such as San Diego, California [53], the Pearl River estuary, China, [54], Durno Estuary, Portugal [55], Guadalquivir River, Spain [56], Central Chile rivers [57], South California rivers [58], Danube river in the Black Sea [59], Amazon River [60], Mississippi River [61,62], Taihu Lake, China [63], the Great Barrier Reef [64] and Greenland fiords [65,66]. Ports have also been monitored in numerous studies [67]. Some of the above studies used L1A/L1B products (MOD02), while others made use of the atmospherically corrected L3 (MOD09Q) product.

Total suspended matter identification algorithms for Remote Sensing, either empirical or semi-analytical, have been developed by various researchers [43,68,69,70]. Semi-analytical models decouple reflectance to the inherent optical properties (IOPs) of absorption and backscattering, while most of the empirical models make use of regression calibrations between water sample analyses collected in situ and remotely sensed data (match-ups). Sometimes this process is mediated by additional calibration of in situ spectroradiometer measurements [71].

The aim of this study is to compare 17 rivers, located in Greece, and rank them in terms of their suspended sediment delivery to the sea using indices from two MODIS products, MOD02 for the period 2004–2014 and MOD09Q for the entire record period (2000–2019). The remote sensing analysis expanded in scope for the MOD09Q product to include the two most productive rivers in the Mediterranean, Seman and Vjose (Aoos) flowing into the Adriatic Sea from Albania, along with four more rivers draining in the Greek coastline, for a total of 23 rivers. The rationale for including the Albanian rivers was that, given their status, they could serve as a confirmation of the study, as indeed they did. MODIS products were chosen for the analysis since they have a long record of operation (more than 20 years to date), an adequate revisit time (almost daily), and a satisfactory spatial resolution (250 m for channels red and near-infrared and 500–1000 m for the others). To assess the effectiveness of the employed remote sensing technique, the outcomes of the remote sensing analysis were subsequently compared with the predictions from soil loss models including the Revised Universal Soil Loss Equation (RUSLE) [19,20,21,22], BQART [28] and the empirical model introduced by Karalis et al. [72] for mountainous catchments in Greece.

The significance of this research is that it suggests a simple intuitive method for monitoring the activity of riverine sediment carried to the coasts (the plume polygons as regions of interest). We applied it in numerous catchments, covering almost the entire Greek Peninsula. In addition, a new index stemming from the remote sensing analysis is proposed. This index can be considered as a sediment delivery ratio and proves to be useful in hydrological studies. The proposed method is certainly a step that dedicated environmental protection institutions may adopt easily. In addition, total suspended matter is a critical water quality parameter related to land-use practices and water resource conservation and can serve as a detector of harmful algal blooms. Sediment exchange also plays a critical role in the global carbon cycle, since half of the terrestrial organic carbon exported by rivers is ultimately buried in marine sediment [73].

It was found that the MOD09Q product can be sufficient for this application, and this is greatly beneficial since no atmospheric correction is needed and the acquisition and manipulation of this product are much simpler.

2. Study Area

This study compares the suspended sediment delivery of 17 rivers, which drain a wide geographical area covering the majority of mainland Greece and streams, to the sea. The remote sensing analysis also considers the Seman and Vjosa (Aoos) Rivers, which flow into the Adriatic Sea, and are the two most sediment-laden rivers in the Mediterranean. Figure 1 illustrates the rivers included in this study and their associated catchments, while

Table 1. Primary characteristics of the rivers included in this study, such as the length of the main channel, catchment area, and the name of the receiving basin. The catchment area of the rivers does not include the area upstream of the dams. For Axios, the catchment area is the one within the Greek borders (its total area is approximately 20,000 km²). Selinous’s catchment area includes also Kerynitis’s, as they both flow into the same plume polygon.

No	River	Length of the Main Channel (km)	Catchment Area (km2)	Receiving Basin
1	Acheloos	220	420	Ionian Sea
2	Acheron	52	718	Ionian Sea
3	Alfios	110	2800	Ionian Sea
4	Axios	380	23,747	Thermaikos Gulf
5	Evinos	80	813	Gulf of Patras
6	Evrotas	82	1736	Lakonikos Gulf
7	Kalamas	115	1898	Ionian Sea
8	Krathis	33	155	Gulf of Corinth
9	Krios	20	130	Gulf of Corinth
10	Mornos	70	389	Gulf of Corinth
11	Pamissos	44	794	Messiniakos Gulf
12	Pinios	205	10,840	Thermaikos Gulf
13	Piros	43	576	Gulf of Patras
14	Selinous	48	450	Gulf of Corinth
15	Sperchios	80	1661	Maliakos Gulf
16	Strymon	392	5141	Strymonikos Gulf
17	Vouraikos	38	273	Gulf of Corinth
18	Aliakmon	322	7517	Thermaikos Gulf
19	Arachthos	135	1900	Amvrakikos Gulf
20	Evros	528	53,000	Thracian Sea
21	Nestos	243	2975	Thracian Sea
22	Seman	281	5649	Adriatic Sea
23	Vjosa (Aoos)	272	6706	Adriatic Sea

Rivers from 18 to 23 were only included in the remote sensing analysis.

provides their primary characteristics, like the length of the main channel, catchment area, and the name of the receiving basin.

The Greek mainland, situated at the southernmost tip of the Balkan Peninsula, consists predominantly of mountains or hills and is deeply dissected, making the country one of the most mountainous and rugged regions in Europe. The Pindus Mountain range runs across the center of the country in a northwest-southeast orientation, with its highest peak reaching an elevation of 2637 m. The southern extensions of this mountain range continue across the Peloponnese. The geomorphological diversity of the landscape is strongly influenced by the regional geotectonic context, which is primarily dominated by the subduction of the African plate beneath the Eurasian plate [74].

According to the Climate Atlas of Greece, published by the Hellenic National Meteorological Service (H.N.M.S.), the prevailing climate in Greece is mainly Mediterranean (Köppen classification: Csa) [75]. However, owing to its greatly diverse topography, Greece exhibits local climate deviations and a wide range of microclimates. The climate to the western side of the Pindus Mountain range is typically wetter and exhibits certain maritime characteristics, whereas the region situated to the east of the Pindus Mountain range tends to be drier and windier, particularly in the summer months.

The intense tectonic activity coupled with complex geology and climatic conditions, has resulted in the development of numerous drainage networks (of both perennial rivers and ephemeral streams), with catchments characterized by a wide range of distinct topographic features. Most Greek rivers are relatively short and fast-flowing, with steep gradient channels and rapids in certain regions. River discharges vary seasonally, typically reaching their maxima between late autumn and early spring, while their minima are usually observed during late summer. Periods of peak discharge are associated with high precipitation and/or snowmelt in the adjacent mountains [76].

Rivers in western Greece experience their highest discharges in December, whereas rivers situated in eastern Greece reach their peak levels in January. The rivers draining the mountainous region of northern Greece experience their highest discharges during the spring season. Smaller rivers follow a similar pattern, showcasing a wet season spanning from December to April and minimal discharges during the summer months of July to September [76].

In Greece, there is a lack of comprehensive river hydro-sedimentology monitoring. In situ measurements of suspended sediment loads have been carried out exclusively by the Public Power Corporation S.A [77]. This monitoring has primarily focused on stations located along rivers of the western and northern regions of the country. The data collected span time periods of up to 15–20 years, specifically from the years 1966 to 1983, before the rivers were dammed. The fluvial systems of Greece are mostly small and mountainous, with their catchments lying adjacent to the receiving marine basins resulting in exceptionally high sediment yield [78]. In the 19th century, human activities, especially the construction of multiple dams for electricity generation and irrigation purposes, led to a significant decrease in the supply of river sediment to the coastal zone [76].

Most of the rivers in this study flow into enclosed and semi-enclosed Gulfs, many of which are shallow, and sheltered. However, the Gulf of Corinth and the Gulf of Patras maintain higher depths. Three of the rivers—Evros, Nestos, and Strymon—empty into the North Aegean Sea, specifically the Thracian Sea. The rivers Kalamas, Acheron Acheloos and Alfios discharge into the Ionian Sea, while the receiving basin of Vjosa and Seman is the Adriatic Sea.

The underwater morphology of Greece exhibits diversity. The Aegean Sea, situated between continental Greece and Türkiye, reaches maximum depths of approximately 2500 m whereas the Ionian Sea, located west of continental Greece, hosts the deepest basins in the Mediterranean Sea. The greatest depth in Greek seas reaches 5127 m (Oinouses Well), located offshore southwest of the Peloponnese.

The morphological configuration and evolution of the Greek coastline are influenced by the water/sediment fluxes from numerous rivers and ephemeral streams, the lithology (erodibility) of the coastal zone, the bathymetry of the adjacent continental shelf, prevailing oceanographic conditions (such as wave and current activity), and recent tectonic activity [79,80]. The continental shelf, characterized by slopes <2%, constitutes about 4% of the subaqueous relief in the Ionian Sea and 21% in the Aegean Sea [78]. According to the wave and wind atlas of the Hellenic seas [81], mean annual significant wave heights for Greek seas are less than 1.5 m with sheltered embayments like Amvrakikos Gulf, inner Thermaikos Bay, Maliakos Gulf, Gulf of Corinth experiencing very low wave activity. Conversely, river mouths exposed to the open sea, where the continental shelf is deep, experience higher wave activity. In Greek waters, the astronomical tide is generally less than 10 cm. However, the overall sea-level fluctuation exceeds 0.5 m due to meteorological forcing (differences in barometric pressure, wind and wave setup) [82]. The tide range variation for the Greek seas is generally less than 0.87 m [83].

3. Materials and Methods

3.1. Moderate Resolution Imaging Spectroradiometer (MODIS) Imagery

The activity of the rivers concerning their sediment loads was quantified using the time series of the values of the indices/masks within the plume polygons drawn in front of the rivers’ mouths. All these polygons have an approximate area of ~20 km². This decision was made without specific reasoning other than the necessity for them to fit within areas of adjacent river deltas/fan deltas, where the river mouths are in close proximity to each other. The polygons were delineated at a distance of one to two pixels away from the coastline to prevent interference from shallow waters and infections from the signal of land pixels. In many cases, the prevailing direction of longshore currents was taken into consideration. Greek coasts are micro-tidal (with tidal range ~20 cm), thus no significant impact from tidal activity is expected [84].

3.1.1. Selection, Acquisition, and Processing of Images

Post-processed MOD02 Level 1B images (MOD02QKM / MOD02HKM) (Figure 2) were ordered and downloaded from the Distributed Active Archive Center (DAAC). This is an option provided by the DAAC, where all procedures including cropping (spatial subsetting), spectral subsetting, reprojecting and mosaicking are omitted for the user. The trade-off for this convenience is ending up with thousands of files, which was managed through proper organization within folders.

The cloud mask product, MOD35, was downloaded for the same dates. These 669 days were selected based on the hydrograph of Vouraikos, a small perennial river that is located exactly at the center of the study area [85]. They cover a 10-water-year period (2004–2014) and are grouped into 58 flood events, or “episodes”, varying in duration from 5 to 45 days. Continuous daily rainfall data for these 10 years were also available from three stations operated by the Public Electricity Corporation (DEH) within the Vouraikos catchment, in addition to temperature and discharge data. Daily rainfall data for the entire region were acquired from the Global Precipitation Climatology Project (GPCP). The GPCP 1-degree daily precipitation dataset provides daily rainfall accumulation globally on a one-degree grid of latitude and longitude, covering the period from October 1996 to the present. It relies on the GPCP monthly product for the total monthly rainfall and primarily utilizes geostationary infrared satellite imagery to determine daily rainfall rates [86]. Those precipitation time series were found to be in good agreement between them (local and global), and also confirmed that, indeed, the selected flood events’ “episodes”, were usually common to most of the Greek peninsula.

Relative atmospheric correction was applied to the images. Relative atmospheric correction methods avoid the evaluation of atmospheric components of any type and instead rely on the observation that, for a specific sensor channel, the relationship between radiances at the top-of-atmosphere (TOA) and ground-level follows a linear trend for the variety of Earth features depicted in the image. The method, known as Relative Radiometric Normalization (RRN), relies on the existence of at least two invariant areas within the image (one bright and one dark) that are supposed to maintain their reflective properties over time (Pseudo-Invariant Features or PIFs). This approach serves to establish a uniform comparison basis for the study [87]. Some researchers have claimed its superiority over absolute atmospheric correction methods, particularly for retrieval of total suspended matter concentrations in inland/coastal waters [88].

Several methods for the automatic selection of PIFs have been developed. However, in this study, the manual selection was preferred as it allowed the observer to get acquainted with the territory. Since PIFs are ideally supposed to be at least 2 or 3 times larger than the spatial resolution of the image and should avoid mixed land cover, spotting such areas was a challenging task, even within the extensive land coverage of our images. Efforts were made to scatter these areas geographically, reducing the possibility of them being clouded at the same time, and also to choose sites with low elevation and similar reflectance values. Three abandoned open quarries and a bare rock hilltop were eventually selected as bright PIFs after their “history” was inspected using historical imagery in Google Earth (the small green crosses in Figure 2a). As the dark PIF, the darkest sea pixel was selected. The reason for selecting more than one bright PIF was the intention to transform as many images as possible. The PIFs were prioritized in order. If the first bright PIF was clouded, the program proceeded to the next one, and so forth.

The entire workflow (Figure 3) was completely automated through a suite of Interactive Data Language (IDL) programs.

The images were initially “restored”, involving the correction of reflectance values. Another program constructed the cloud masks from the MOD35 cloud-mask product, using bitwise operations. After the correction and construction of cloud masks for all 669 images, a clear image with minimal Near-Infrared (NIR) reflectance and minimal cloud cover was selected as the reference image. This image was further analyzed to determine thresholds for the masks and reflectance values for the PIFs for each band. In the subsequent RRN procedure, a total of 382 images were finally transformed into subject (slave) images. A total of 43% of the data was lost due to cloud coverage affecting all 4 bright PIFs. Indices, ratios and masks (see below) using 4 bands were calculated for the images, masked to the regions of interest and finally masked with the cloud mask. Using the final masked indices’ rasters, complete statistics were calculated for each index, along with the percentages of masks within each region of interest. The resulting database comprised 382 days (the theoretical maximum number of observations since the actual maximum was 282 days due to cloud cover), multiplied by 17 sites (river plumes).

For the MOD09Q product, which consists of an 8-day composite of red and near-infrared bands, the process is notably simpler as the images are already atmospherically corrected (denoted as Analysis Ready Data). The entire time series of images (1840 tiff files, two bands) was acquired from NASA’s AρρEEARS platform (https://lpdaacsvc.cr.usgs.gov/appeears, accessed on 5 March 2021) and were processed primarily with the help of R scripts.

3.1.2. Suspended Sediment/Suspended Material Indices, Ratios and Masks

As mentioned earlier, the assessment of rivers’ activity concerning their sediment loads was quantified using the time series of the values of suspended sediment/suspended material indices and threshold masks within the approximately ~20 km² plume polygons drawn in front of the river mouths.

Suspended sediment indices (or suspended material indices, depending on whether organic material is considered as well) have been developed by various researchers, taking into account the principle that the presence of suspended matter in turbid waters enhances reflectance (and radiance) across the entire visible and near-infrared spectrum range [89]. Turbidity is a measure of the amount of cloudiness or haziness in sea water caused by individual particles that are too small to be seen without magnification. It is an optical property very strongly correlated with suspended sediment concentrations, and is, in effect, a measure of scattering of the light traveling through a water column [89]. This phenomenon arises from both suspended organic elements (including algae) and suspended inorganic particles. Scattering escalates with higher suspended loads (greater concentrations) and depends on the size and composition of the suspended matter, such as clay, silt, colloidal particles, plankton and other microscopic organisms. Variations of sediment type (such as grain size, refractive index) along with changing illumination conditions (whether the sky is clear or overcast) affect the reflectance signal of coastal waters and limit the accuracy of suspended sediment concentration estimations from remote sensing measurements [90].

For low concentrations of suspended sediment, reflectance from nearly any wavelength will exhibit a notably linear correlation with suspended sediment concentrations, while at higher concentrations, a curvilinear trend line with longer wavelengths (especially red) is a more appropriate descriptor [91]. The peak of the reflectance spectra shifts towards longer wavelengths as suspended sediment concentration increases (Figure 4). Many researchers have noted the utility of the near-infrared band in establishing empirical relationships for the quantification of suspended sediment concentration [92,93]. Additionally, the use of ratios between near-infrared and visible bands has been proposed for this purpose [94]. As evident from

Table 2. Indices and ratios used for the identification of suspended sediment/material in coastal waters in this study. The lower part of the table includes the masks and their formulas.

Index/Ratio Name	Formula	Reference
Normalized Difference Suspended Sediment Index (NDSSI)	$\frac{{\rho }_{blue}-{\rho }_{nir}}{{\rho }_{blue}+{\rho }_{nir}}$	A. Hossein et al. (2010) [95]
Normalized Suspended Material Index (NSMI)	$\frac{{\rho }_{red}+{\rho }_{green}-{\rho }_{blue }}{{\rho }_{red}+{\rho }_{green}+{\rho }_{blue }}$	L. Montalvo (2010) [96]
NIR to Red (n2r) and Green to Blue (g2b) ratios	$\frac{{\rho }_{nir}}{{\rho }_{red}},\frac{{\rho }_{green}}{{\rho }_{blue }}$	various
First Principal Component (PC1)	0.15 × ρblue + 0.35 × ρgreen + 0.80 × ρred + 0.43 × ρnir	This study
Mask Name	Formula
Li	ρred > 0.031
Fleuve	ρred > 0.031 and ρnir > 0.020
NSMI	ρred + ρgreen − ρblue > 0

, all 4 bands within the visible and near-infrared range were used for the construction of indices and ratios. The blue band is used as the one least affected by the presence of suspended sediment/suspended material.

In addition to indices and ratios, threshold masks are a common methodology for delineating areas of interest in remote sensing [97]. Three distinct masks were used in this study. The first one is an adaptation based on the algorithm proposed by R.R. Li et al. in 2003 [98]. They investigated methods for masking turbid coastal waters with the aim of enhancing remote sensing of aerosols over oceans using MODIS. In brief, their algorithm operates as follows: they first use the apparent MODIS reflectances at 0.47, 1.24, 1.64 and 2.13 μm (corresponding to bands 3, 5, 6 and 7 of MODIS, with band 3 representing the blue band) to derive an atmospheric spectral power law (a log–log fit between wavelength and reflectance). Subsequently, using this established relationship, they calculate reflectances for the wavelengths 0.55, 0.66 and 0.86 μm (which correspond to bands 4, 1 and 2 of MODIS—representing Green, Red and Near-Infrared—NIR). Any significant (positive) deviation of the measured reflectances in these bands from the values predicted by the power law is attributed to the presence of suspended sediment/suspended material. In designing their mask, they opted to utilize the green band. This choice was made to effectively mask shallow waters with bottom reflections, given that the penetration depth of the green band is considerably greater than that of the red band (ibid). This methodology has been adopted successfully in other studies [48]. In this approach, the red band was employed to define this mask, since extremely shallow waters are not commonly encountered along Greek coasts, at least not at a safe distance from the coastline (more than 500 m), which is the range within which the plume polygons were delineated. This mask was called the Li mask. A more conservative mask based on reflectances in both Red and Near-InfRared bands, called the Fleuve mask, was used to distinguish the fresh vein of sediment-laden river water from the surrounding turbid waters of the plume. Lastly, the Normalized Suspended Material Index (NSMI) was found to vary from −0.2 to around 0.5, with values surpassing −0.1 indicating the presence of suspended material (Figure 5). This provides a threshold at a value of zero, which served as the basis for creating a third mask, aptly named the NSMI mask. To interpret the dataset and reduce its dimensionality in an interpretable manner, while preserving the majority of information within the data, Principal Components Analysis (PCA) [99] was employed. This technique was conducted in a number of images within three distinct areas as identified through visual inspection: clear sea, suspended sediment plumes, and areas of the sea dominated by suspended material (largely of organic origin). This analysis resulted in the creation of the PC1 index (Table 2).

3.1.3. Soil Loss Models

The effectiveness of the employed remote sensing technique was evaluated by comparing the outcomes of the remote sensing analysis with the predictions from three soil loss models. Recent Revised Universal Soil Loss Equation (RUSLE15) model [22] outcomes were obtained from the erosion maps available at the European Soil Data Center (ESDC, 2015). RUSLE15, however, does not account for gully erosion. As it reports local (gross) erosion, a sediment delivery factor has to be implemented to estimate sediment delivery at certain stations or the sea. Calculations were performed for the BQART model [28] and for another model proposed by Karalis et al. in 2018 [72] for 17 of the rivers that were monitored through their plumes in the remote sensing analysis. Karalis et al. (2018) model [72] incorporates slope, precipitation and the proportion of easily erodible geological formations within the catchment and has primarily been calibrated using the available estimations of Suspended Sediment Yield (SSY) within the mountainous, flysch-laden, and precipitation-abundant catchments of the western Greek peninsula. Both the BQART and Karalis et al. [72] models were generally successful in predicting soil delivery in the catchments of Northwestern Greece, with regards to existing measurements estimations.

Table 3. Soil Loss Models used in the study, along with their respective formulas and explanations.

Model	Formula and Explanation	Comments
RUSLE2015 [19,20,21,22]	A = RKLSCP where: A = average annual erosion, R = rainfall-runoff (erosivity) factor, K = soil erodibility factor, LS = length slope factor, C = crop management factor, P = soil conservation factor	An application of a modified version of the Revised Universal Soil Loss Equation (RUSLE) model, specifically RUSLE2015, was employed to estimate soil loss in Europe for the reference year 2010. This estimation involved the use of the latest pan-European datasets to model the input factors. Results and factor values are accessible through maps provided by the European Soil Data Centre (ESDC).
BQART [28]	Qs = ωΒQ0.31A0.50RT  for T > 2 °C Qs = 2ωΒQ0.31A0.50R  for T ≤ 2 °C where: ω = a constant that varies based on whether the results are required in kg/s or MT/y, A = area, R = relief, T = temperature, Q = runoff, and B = IL(1 − Te)Eh, B combines the cumulative effects of lithological (L) and glacial (I) characteristics of the catchment, its sediment trapping capacity (Te), and the anthropogenic pressures applied to it (Eh)	The BQART predicts the long-term flux of sediments delivered by rivers based on geomorphic, tectonic and geographic factors, and is pooling data from an extensive database of rivers and stations. The B factor can be assumed as unity (representing a global average) in the absence of specific information regarding its four distinct encompassing parameters. When solely L parameter is incorporated into the B formula (as predominantly employed in the current study), the model is referred to as LQART.
Karalis et al. (2018) [72]	SSY = 0.0049 S1.51 P0.94 + 102.87 e1.46L where: S = Slope (%), P = mean annual Precipitation (mm), L = Lithology (fraction of easily erodible geological formations in the catchment)	This empirical model has been developed and trained using the available sediment yield estimations from the mountainous catchments of the western Greek peninsula. It can also be employed without the additive term to estimate a minimum.

provides a concise description of each model.

4. Results and Discussion

4.1. Moderate Resolution Imaging Spectroradiometer (MODIS) Imagery

Principal Components Analysis (PCA) was performed within three visually identified regions: clear sea, suspended sediment plumes, and areas of sea dominated by suspended materials (primarily of organic origin) (

Table 4. First principal component loadings for three different areas. In general, the first Principal Component corresponds to the mean value of the reflectance of the channels involved in the transformation.

Channel	Clear Sea	Suspended Sediment-Dominated Sea	Suspended Material-Dominated Sea
Blue (459–479 μm)	0.74 ÷ 0.82	0.12 ÷ 0.17	0.34 ÷ 0.38
Green (545–565 μm)	0.48 ÷ 0.52	0.34 ÷ 0.39	0.72 ÷ 0.86
Red (620–670 μm)	0.22 ÷ 0.34	0.78 ÷ 0.82	0.32 ÷ 0.55
NIR (841–876 μm)	0.15 ÷ 0.25	0.42 ÷ 0.46	−0.15 ÷ 0.03
Percentage of variance explained by the first component	56 ÷ 70	75 ÷ 89	62 ÷ 86

). A corrected image was used for the analysis from which the distinct parts were identified, sampled and then analysed. For the clear water case and the sediment plume case, the chosen area was in the coastal sea, while for the sediment material-dominated waters, an area within a lake was chosen (Lake Koroneia near Thessaloniki). This lake is known for its poor environmental status, including anoxia episodes, since it is the main receiver of both agricultural and civil waste waters.

In the clear sea, the channels exhibit a significantly lower correlation compared to the suspended sediment/suspended material-dominated sea, clearly indicating that their reflectance is enhanced by the presence of suspended sediment/suspended matter. In the clear sea, the dominant loading of PC1 originates from the blue channel and diminishes as the wavelength increases, while the PC1 accounts for only 56% to 70% of the overall variance. In the SS-dominated sea, the primary loading comes from the red channel, followed by the near-infrared channel, with their loadings being significantly increased in comparison to that of the clear sea, while the loading of the green channel slightly decreases and the loading of the blue channel greatly decreases. In an SM-dominated sea, the dominant loading comes from the green channel followed by the red and blue channels, while the near-infrared channel’s loading is eliminated. In the last two cases (suspended sediment/suspended material) the variance explained by the PC1 is notably higher than that in the clear sea. Based on the aforementioned observations, a PC1 suspended sediment index was constructed also, used as a proxy for total suspended sediment concentrations. This approach is supported by the findings of other researchers who have documented a significant positive correlation between total suspended sediment concentrations and PC1 [100].

The MOD02 results were initially visually inspected in ENVI software to assess their range, behavior and characteristics. The histograms of the measured and calculated quantities are presented in Figure 6. The NSMI was found to fall within the range of −0.2 to 0.5 (with positive values signifying higher suspended sediment concentration), and NDSSI was observed to range from −0.9 to 0.9 (with higher values indicating lower suspended sediment concentration). However, due to the peculiar characteristics of the distribution of this index, it was only used in an auxiliary fashion. It was found that images displaying extreme values (maxima) of indices were indicative of cloud-affected cases of regions of interest (where the cloud mask was unsuccessful), or cases of mosaicked/half images. For this reason, in the results, outliers with extreme values were omitted and only meaningful ranges were considered. Furthermore, in a subsequent analysis, a filter was applied to ensure that the cloud-free area of the region of interest exceeded 30%.

Noteworthy is the perceived character of the different indices. The NSMI mask, which includes the green and red bands of MODIS, indicates the presence of suspended material, whether organic or not. In contrast, masks based on the red or red and infrared bands primarily reflect the inorganic materials. It was also found that the green-to-blue ratio shows a strong correlation with the total river discharge.

In terms of NSMI ranking, Axios River comes first, followed by Alfios, Pinios, Sperchios, and Kalamas. The catchments of these rivers undergo intensive agricultural exploitation, and the suspended material includes not only chlorophyll and other organic materials (such as leaves) but also substances from urban and agricultural effluents, often colored by pesticides. Ranking using the Fleuve mask (red and near-infrared) elevates smaller rivers like Mornos and Selinous to the level of larger rivers such as Pinios or Alfios. Additionally, it places rivers like Axios and Sperchios at the top, highlighting the influence of their sediment’s lithological composition. However, as previously mentioned, doubt arises concerning their signal, primarily due to recirculation when rivers discharge into closed, shallow gulfs.

The time series of the results was plotted and analyzed for the majority of the 58 episodes for each river or cluster of rivers (east, west, south and central). An example of such an analysis for “episode” No. 23 is given in Figure 7. This is the longest episode (lasting 45 days from 9 January to 23 February 2009). However, due to cloud coverage, only a portion of these days was finally available for measurements.

From the map of the first results (Figure 8) and the final ranking (Figure 9a,b and Figure 10) it becomes evident that two rivers of the western flank (Alfios, and Kalamas) emerge as the primary contributors of soil delivery to the sea. On the eastern flank, significant contributions are made by Pinios and Sperchios followed by Axios. Among these top five rivers, the first three discharge into the open sea, and their plumes are rapidly transferred and dissipated by hydrodynamic activity. Hence, the ranking can be deemed reliable. Sperchios and Axios empty into the enclosed Maliakos and Thermaikos Gulfs, respectively, where sediment is entrapped by local circulation within the Gulf. The drainage networks of the rivers and streams that discharge into these two shallow gulfs, particularly Thermaikos, drain heavily populated and intensively cultivated areas. These sites are prone to frequent algal blooms and harmful algal blooms (HABs) due to the eutrophic nature of the basins [101]. Most of the small to medium-sized rivers in central Greece that discharge into the Gulf of Corinth (Mornos, Piros and Selinous, Krathis, Vouraikos, and Krios) along with Pamissos, exhibit very high suspended particulate matter productivity and are clustered in the second position.

In the third cluster, of the less dynamic rivers, we find Evrotas, Acheron, Strymon, and Acheloos (Figure 8), despite their considerable size and discharge volume. Even though the hydrological years 2007–2008 and 2008–2009 were the lowest in precipitation, they ranked highest in sediment productivity as can be seen in Figure 9g. This paradox was resolved when it was found that during these years there were significant dust storms emanating from Africa.

It is worth noting that a more in-depth analysis of the data revealed the impact of significant dust storms triggered by the high temperatures during that time period. It was found that the biennium 2007–2008 represented the most intense period of sand and dust storms [102], accompanied by extreme temperature records in the summer of 2006, while significant sand and dust storm activity was also noted in 2006 [103]. This is further supported by other more localized studies [104,105]. The quantity of material that can be transported by sand and dust storms is substantial: an estimated 25 million tons annually for the eastern Mediterranean basin, as early as 1979 [106] and this trend has shown a continuous increase since then [103,107].

Therefore, the performance during the water years 2006–2007 and 2007–2008 could be attributed to the washing of the dust particles settled by sand and dust storms. For this reason, the water years 2006–2007 and 2007–2008 were excluded from the ranking process for both products. Sand and dust particles were found to be mostly identifiable in the near-infrared channel of MODIS.

Considering the monthly ranking and interannual variability in total suspended matter delivery, it becomes apparent that January and December contribute the highest amounts of material to the sea (Figure 9d). They are followed by February, November, and March. Surprisingly, October ranks relatively low (and September also), trailing behind both May and June, while April along with August also shows diminished contributions. There were no episodes recorded in July.

The results from MOD09Q1 (where the PC1 index was calculated for only its red and NIR contributions) generally reaffirm the findings from MOD02, but not quite. In Figure 8, which also includes the rivers Seman and Vjose (Aoos), as well as Evros, the largest river of the Balkan peninsula, these two rivers were found to be much more productive in sediment delivery than all the others, indirectly confirming the results. Seman River, with an estimated yield of 22 million tons per year and a specific sediment yield of around 4200 t km⁻² yr⁻¹ [108], is considered the leader within the Mediterranean basin. Indeed, as depicted in Figure 10, both Seman and Aoos score more than double the other rivers in the red channel, while being at a similar level with Kalamas regarding the near-infrared channel. This is consistent with the geographical proximity of these rivers and the geological composition of their catchments, which suggests a potential similarity in the mineralogical composition of their suspended solids as well as with the high mean annual precipitation within their catchments.

The calculation of a PC1 per square kilometer and mm of rainfall index for each river, which can be considered a kind of sediment delivery ratio in the Hydrological sense, across the entire monitoring period, showed that smaller mountainous rivers (with rough catchments characterized by steep slopes) are considerably more productive compared to larger ones (with catchments of relatively smooth relief characterized by gentle slopes) (Figure 9). In addition, rivers located in the western part of the Greek peninsula exhibit significantly higher productivity compared to those found in the eastern and southern regions. Overall, based on the PC1 per square kilometer and mm of rainfall index, the most productive, in terms of suspended sediment delivery to the sea, are the four small rivers of the North Peloponnese that drain into the Gulf of Corinth. They achieve a score approximately three times higher than that of the western rivers, which, in turn, score about twice as high as the eastern (including also the southern) rivers (Figure 11). For the majority of the river systems, the formation of plumes in the river mouth is highly controlled by the dams’ presence which retain sediment. The catchment areas of the North Peloponnese rivers are not controlled by dams and therefore are prone to flash floods [109]. In addition, these catchments are made of highly erodible geological formations (mainly marls of the Plio-Plestocene age) leading to high sediment productivity. Another reason is the torrential character of these streams as a result of the intense neotectonic activity in this area. High tectonic uplift has affected the streams forcing their steep channels to deeply incise forming valleys with slopes of high gradient [110].

One of the limitations of the remote sensing method is the fact that sediment is entrapped by local circulation within some of the receiving basins. For instance, within the shallow Gulf of Patras, significant recirculation influenced by inertial effects is observed as illustrated in Figure 12. Under the influence of prevailing wind forcing (predominantly westerly winds) very strong wind-driven currents, more pronounced than the tidal-induced currents, [111] develop in the Rio-Antirio strait, and funnel the sediment plumes into the much deeper Gulf of Corinth, where they dissipate along its southern coastline. This is the reason why the plume of Mornos River, flowing into the Gulf of Corinth immediately east of the Rio-Antirio strait, seldom attains its complete expanse. These patterns allow for suspended sediment loads from certain rivers to be partially recorded in another river to the east. Consequently, these patterns raise uncertainty regarding the accuracy of suspended sediment estimation results for this particular area. The wind climate of each coastal area also has an impact that was not taken into consideration. Amongst the whole oceanographic and morphological conditions in the receiving basin, the fate of the suspended sediment reaching the river mouth is also defined by the depth of the basin. For instance, the Thermaikos Gulf is a shallow enclosed embayment [112] apposite to the Gulf of Corinth or Gulf of Patras [113,114] which maintains higher depths, a parameter that influences turbidity.

It should be noted that the suggested methodology is general, but the thresholds and particulars of the spectral signatures should be locally studied and calibrated, ideally with the aid of in situ measurements, which were not available in this study.

4.2. Comparison with Soil Loss Models

The results from two predictive suspended sediment load models (BQART and Karalis et al. [72]) for the 17 catchments of the Greek rivers included in the remote sensing analysis are presented in

Table 5. Model parameters and results from BQART and Karalis et al. (2018) [72] models for 17 of the catchments included in the remote sensing analysis. The top five rivers are indicated in bold.

River	Area km2	Q km3	L	Eh	TE	B	R km	T °C	BQART t/km2	BQART Ton × 106	LQART t/km2	LQART Ton × 106	S %	P mm	L Fraction	Karalis et al. [72] Ton × 106
Alfios	2907	1.714	1.5	1	0.2	1.8	1.50	17.5	435	1.27	544	1.58	25.42	1100	0.33	1.85
Axios	23,747	5.000	0.5	1	0.2	0.8	1.62	11.0	48	1.14	60	1.43	8.12	553	0.22	4.41
Acheloos	420	4.383	0.5	1		0.8	0.32	16.0	125	0.05	125	0.05	13.06	908	0.10	0.11
Acheron	718	0.315	1.5	1	0.5	0.8	2.00	13.5	334	0.24	668	0.48	29.62	1300	0.15	0.59
Vouraikos	273	0.095	2.0	1		2.0	2.25	15.5	1288	0.35	1288	0.35	36.65	927	0.50	0.25
Evinos	828	0.917	2.0	1		2.0	2.39	15.0	1531	1.27	1531	1.27	21.18	1008	0.59	0.47
Evrotas	1736	0.444	1.3	1	0.3	1.2	2.05	18.0	566	0.98	566	0.98	21.53	836	0.30	0.77
Kalamas	1899	2.048	1.5	1		1.8	2.11	13.0	745	1.42	745	1.42	26.40	1297	0.23	1.37
Krathis	155	0.091	1.7	1		1.7	2.27	15.5	1442	0.22	1442	0.22	42.40	965	0.37	0.17
Krios	130	0.064	2.0	1		2.0	1.75	16.0	1326	0.17	1326	0.17	34.88	900	0.49	0.11
Mornos	399	0.404	2.0	1		2.0	2.42	15.0	1736	0.69	1736	0.69	38.11	998	0.57	0.41
Pamissos	794	0.341	1.5	1		1.7	1.59	18.0	688	0.55	688	0.55	19.06	922	0.31	0.33
Pinios	8184	2.558	1.5	1	0.2	1.8	2.78	14.5	452	3.70	565	4.63	15.60	654	0.29	2.42
Piros	576	0.122	1.5	1		2.0	2.16	16.0	713	0.41	713	0.41	20.34	596	0.48	0.23
Selinous	450	0.155	1.5	1		1.7	2.01	16.0	808	0.36	808	0.36	35.26	938	0.42	0.38
Sperchios	1661	0.693	1.5	1		1.8	2.12	17.0	750	1.24	750	1.24	23.41	908	0.43	0.90
Strymon	4141	0.877	0.5	1	0.5	0.4	2.18	11.0	57	0.23	113	0.47	16.37	550	0.15	1.05

. According to the BQART sediment load model results, Pinios River has the highest potential for suspended sediment yield (3.70 × 10⁶ tons), followed by Kalamas with 1.42 × 10⁶ tons, Alfios and Evinos with 1.27 × 10⁶ tons each, and Sperchios with 1.24 × 10⁶ tons. The findings of the model developed by Karalis et al. (2018) [72] indicate that the top five more productive rivers are Axios, Pinios, Alfios, Kalamas and Strymon with sediment yields of 4.41 × 10⁶ tons, 2.42 × 10⁶ tons, 1.85 × 10⁶ tons, 1.37 × 10⁶ tons, and 1.05 × 10⁶ tons, respectively.

Table 6. The ranking of the rivers in the remote sensing analysis (PC1-MOD09Q) is compared to predictions from soil loss models and hydrological data for the catchments. The catchment areas include all upstream regions of dams, except for Acheloos and Pinios. Top five rivers in each ranking are in bold.

RS Rank (PC1MOD09Q)	RIVER	Karalis et al. (2018) [72] Tons × 106	BQART Tons × 106	RUSLE15 ESDC Tons × 106	Poulos (1996) [78] Tons × 106	P mm	Sl %	Area km2	Vol hm3
1	Kalamas	1.37	1.42	1.56	0.93	1297	26	1899	2048
2	Axios a,b	4.41 *	1.43	2.59	2.01	553	8	23,747	5000
3	Sperchios a	0.90	1.24	0.59	0.40	908	23	1661	693
4	Pinios a	2.42	4.36	2.46	1.10	654	16	8184	2558
5	Alfio s a	1.86	1.58	1.49	0.68	1100	25	2907	1714
6	Evinos b	0.47	1.27	0.65	0.43	1008	21	828	917
7	Strymon a,b,c	1.05 *	0.47	0.75	0.49	550	16	4141	877
8	Morno s b	0.41	0.69	0.23	0.17	998	38	399	404
9	Acheloos b	0.11 *	0.05	0.17	0.10	908	13	420	4383
9	Acheron c	0.59	0.48	0.83	-	1300	30	718	315
9	Pamissos	0.33	0.55	0.32	-	922	19	794	341
9	Selinous	0.38	0.36	0.31	-	938	35	450	155
10	Krios	0.11	0.17	0.09	-	900	35	130	64
10	Evrotas a	0.77	0.98	0.67	-	836	22	1736	444
10	Krathis	0.17	0.22	0.10	-	965	42	155	91
10	Piros	0.23	0.41	0.33	-	596	20	576	122
10	Vouraikos	0.25	0.35	0.15	-	927	37	273	95

^(a) Rivers with catchments with intense agricultural exploitation ^(b) Rivers seriously dammed ^(c) Rivers with extensive wetlands/marshes at their delta (*) Basins with slopes under 20% (Karalis et al.’s model [72] is applicable for slopes greater than 20%).

combines the clustering-based ranking of the remote sensing analysis with the predictions from the three suspended sediment yield models used in this study (BQART-LQART, Karalis et al., 2018 [72] and Rusle15). The results of a power relation of annual sediment flux with the area of the catchment proposed by Poulos et al. (1996) [78] are also presented for nine of the rivers. RUSLE’s predictions may be taken as maxima, since it reports gross erosion. Anthropogenic pressures were not taken into consideration.

Although there are some differences among the results from the two approaches, a fair agreement between the rankings of the rivers from the remote sensing analysis and the predictive models is ascertained and quantified using Kendall’s tau (

Table 7. Correlations of the rankings of the first eight rivers (Table 6) from different indices/products and soil loss models using Kendall’s tau (−1 to 1).

	RS MOD09Q	PC1 MOD02	Red MOD02	Li MOD02	NSMI MOD02	Karalis et al. [72]	BQART	Rusle15	Poulos
RS (MOD09Q)	1	0.29	0.36	0.36	0.43	0.50	0.36	0.50	0.50
PC1 MOD02		1	0.21	0.21	0.21	0.50	0.36	0.21	0.21
red MOD02			1	0.85	0.36	014	0.14	0.14	0.14
Li MOD02				1	0.50	0.14	0.29	0.14	0.14
NSMI MOD02					1	0.21	0.36	0.21	0.21
Karalis et al. [72]						1	0.57	0.29	0.29
BQART							1	0.29	0.29
Rusle15								1	1
Poulos									1

). This serves as verification for both aspects. The contradiction between the two approaches (remote sensing and models) serves to highlight the weaknesses of each method. For instance, which part of the area of each catchment should enter the yield calculations, what is the role of sediment traps in or near the river mouth, such as estuaries, deltas, and lakes, and what are the impacts of significant water withdrawals for agricultural use on sediment discharge? On the other hand, what is the impact of the winds and currents in the receiving basins on the rapid dispersal of the sediment plumes? Additionally, to what extent do these factors influence the likelihood of the plumes being detected by overpassing satellites? Some of these questions can be addressed through a joint examination of the outcomes from both methods.

Useful conclusions can be drawn from the differences or perceived inconsistencies between the two approaches. For instance, medium to large-sized rivers such as Strymon, Acheron and Evrotas, despite their size and characteristics, do not yield a strong remote sensing signal. This is because Strymon and Acheron are characterized by sediment deposition in the broader area around their mouths in the form of delta plain deposits, wetlands and estuaries, while distinctive karst processes are present in Evrotas [115] along with intensive agricultural exploitation.

Finally, the rankings for the top eight rivers as depicted by the remote sensing analysis for the PC1 index of MOD09Q (for MOD09Q the indices PC1, Red and NIR, are in almost perfect agreement as can be seen in the diagrams of Figure 9, were compared with the rankings of the indices PC1, Li, Red and NSMI of MOD02, as well as with the rankings of the four models, and are presented in Table 6.

In general, the rankings derived from the various indices present fair to moderate correlations. Models are in moderate correlation to the ’measurements’, while a good correlation is found between the models of Karalis et al. [72] and BQART indicating that they are locally well-adopted. The final MOD09Q RS ranking is in moderate to good correlation with three out of four models (τ = 0.5). From the indices derived from MOD02 the one that relates best with the MOD09Q indices is NSMI. Of interest is the only fair correlation (0.36) of the Red channel for MOD09Q and MOD02, indicating the differences in the distributions of the original band data that were mentioned earlier. It is also notable that Poulos’s ranking is identical to that of Rusle15.

5. Conclusions

This study aimed originally at the monitoring of 17 river plumes over the Greek peninsula for a 10-water-year period from 2004–2005 to 2013–2014, using two MODIS products. In a later stage, six more rivers were added for the second product (MOD09Q). The rivers were ranked according to their suspended sediment delivery to the sea, and the findings from the remote sensing approach were compared to the results obtained from the RUSLE, BQART, and Karalis et al. [72] soil erosion predictive models.

The final remote sensing ranking indicates that Kalamas and Alfios, two fluvial systems of the western flank, are primary contributors of soil delivery into the sea, while significant suspended sediment contributions on the eastern flank are made by the fluvial systems of Pinios, Sperchios and Axios. The results of the soil erosion predictive models show that despite the slight differences, the group of the five most productive rivers in terms of suspended sediment includes Kalamas, Axios, Sperchios, Pinios and Alfios. Hence, there is a strong agreement between the two methods regarding the identification of the most productive rivers. Regarding the interannual variability in the delivery of total suspended matter to the sea, January and December contribute the highest amounts of suspended material. These months are followed by February, November, and March, whereas October ranks lower, falling behind both May and June, while April along with August also showed reduced contributions.

The remote sensing approach led to the development of an Index of “sediment productivity per square kilometer and mm of rainfall” (PC1 per km² and mm of rainfall index) for the 10-year study period. This index can be considered a remotely sensed sediment delivery ratio (SDR) and is crucial for accurately quantifying the phenomenon of sediment transport in hydrological basins. The calculation of this index for each river throughout the entire monitoring period revealed that smaller mountainous rivers, with rugged catchments featuring steep slopes, exhibit considerably higher productivity compared to larger rivers with catchments of relatively smooth relief characterized by gentle slopes. Furthermore, according to this index, the four small rivers of the north Peloponnese (Piros, Krathis, Selinous, and Vouraikos) attain a score roughly three times higher than that of the western rivers, which, in turn, score about twice as high as the eastern (including also the southern) rivers. An albeit expected finding is that rivers located in the western part of the Greek peninsula exhibit significantly higher productivity in comparison to those situated in the eastern and southern regions of the country.

From a remote sensing/spectral signatures point of view, the fact that the two series (MOD02 and MOD09Q) had distinctive spectral signatures, not in very good agreement between them, was of interest. Reflectance values of the second product were found smaller than that of the first. MOD09Q only provides reflectances for two bands, for which the actual distributions per site differed from those of MOD02, but it was discovered that their ratio, NIR to Red, was in moderate to good agreement (Kendall’s tau = 0.52). This suggests a problem of scaling, and it is likely that the different methods used for the atmospheric correction of the images must have played a role. Another interesting challenge was the interpretation of the different indices and masks since they were not in absolute agreement between them, as evident from the figures presented thus far.

Regarding the comparison between the two products for the purposes of the study, it became clear that MOD09Q, being an 8-day composite, is superior not only in coverage and ease of manipulation (since no atmospheric correction is needed), but also in consistency. On the other hand, the contribution of other bands, notably the green band, to the analysis cannot be underestimated. It is possible, in a future study, to combine these two products with a program of in-situ measurements, at least for the main rivers. This will allow the establishment of quantitative regression algorithms which will be very beneficial in any monitoring environmental protection program. To resolve matters for smaller and/or adjacent rivers, a finer resolution would be advantageous.

While the results obtained from this remote sensing approach can be considered as a type of experimental data or measurements, they come with inherent limitations. These limitations include, among others, infrequent access to cloud-free data during stormy days, the influence of wind climate at the receiving basins, and the effects of river discharge, water stratification, surface layer mixing, tides and currents, etc. The potential impact of dust storms originating from Africa or the Middle East is another concern.

It is shown that the present remote sensing approach is robust, and even though it cannot provide a quantitative estimate of the amount of sediment that is transported to the sea, it can verify or discredit estimations coming from other sources (models, measurements, etc.). Moreover, the ranking derived from this remote sensing method can serve as a calibration of the models and can introduce valid skepticism about model results, prompting researchers to develop an in-depth understanding of the conditions within the catchments. In addition, the study’s outcomes encompass valuable conclusions about the lithological composition of each river’s sediment load and the influence of sand and dust storms, underscoring the nature of open systems.