Synergistic Use of Earth Observation Driven Techniques to Support the Implementation of Water Framework Directive in Europe: A Review

Abstract

The development of a sustainable water quality monitoring system at national scale remains a big challenge until today, acting as a hindrance for the efficient implementation of the Water Framework Directive (WFD). This work provides valuable insights into the current state-of-the-art Earth Observation (EO) tools and services, proposing a synergistic use of innovative remote sensing technologies, in situ sensors, and databases, with the ultimate goal to support the European Member States in effective WFD implementation. The proposed approach is based on a recent research and scientific analysis for a six-year period (2017–2022) after reviewing 71 peer-reviewed articles in international journals coupled with the scientific results of 11 European-founded research projects related to EO and WFD. Special focus is placed on the EO data sources (spaceborne, in situ, etc.), the sensors in use, the observed water Quality Elements as well as on the computer science techniques (machine/deep learning, artificial intelligence, etc.). The combination of the different technologies can offer, among other things, low-cost monitoring, an increase in the monitored Quality Elements per water body, and a minimization of the percentage of water bodies with unknown ecological status.

1. Introduction

Water, energy, and food security are inextricably linked, especially now that we are at a crossroads in history to ensure water security for the planet. In that regard, the increasing demand by citizens, environmental organizations, and many other vital importance sectors for cleaner inland and coastal waters has been evident.

The Water Framework Directive (WFD) (2000/60/EEC) [1] has been trying since 2000 until today to push the European Union (EU) Member States to improve their water monitoring networks, to restore their sensitive water bodies, as well as to respond to the pressures that the water systems face in a timely manner before they reach levels of total degradation. This overall effort of the Directive has inspired the following very apt phrase, that the WFD is “the most ambitious and complex piece of legislation on the environment ever enacted in the EU” [2], having simultaneously a multifaceted beneficial character for the nexus among socio-economic and environmental dimensions. These growing demands for continuous information and data on surface water quality are impossible to achieved using only conventional in situ techniques [3]. In general, these techniques are time consuming and costly, slowing down the Member State obligations to the WFD, with consequential penalties. Thus, researchers focused on the development of alternative approaches that can be implemented at different water bodies and conditions [4,5,6,7]. Earth Observation (EO) has proved as a valuable tool to address the challenges related to the provision of reliable, timely, and accurate information of water resources [8]. Hence, significant opportunities arise from the new EO means paving the way for an ambitious and realistic future in terms of monitoring capacities.

The main objective of this paper is to provide an update about the far-reaching opportunities of EO-driven information services by reviewing recent research works (2017 to 2022), to understand the current state-of-the-art, the limitations, and the way to address the existing challenges towards the provision of informed services for water resources management. At the same time, the current review did not avoid highlighting the limitations presented by the Directive itself, expecting perhaps more than what the majority of Member States can offer. With a positive outlook and vision, the end result of this effort is really promising, allowing the stakeholders to take meaningful actions to maximize the benefits of the new EO techniques in the general WFD framework.

This paper is divided into four main sections. Section 2 provides a short description of the methodological approach adopted in this work. Section 3 presents the main findings of the review analysis, exploring the current situation related to the WFD and the state-of-the-art EO capacities while at the same time highlighting their limitations. Section 4 proposes synergies for the WFD efficient implementation, and the paper is concluded in Section 5.

2. Methodological Approach

Before moving on, we present the methodological approach (Figure 1) that we followed to provide the necessary information around the following three key thematic domains that guide the current work: EO capacities and limitations; WFD needs and current limitations; and the key for WFD efficient implementation.

2.1. The Neccessity to Improve the Monitoring in the Framework of the WFD

The WFD is implemented at regional and/or national levels, but its improvement requires efforts beyond the European borders and support to the new EU entrant countries. The transition from local to continental level is still a significant shift. Currently, there is a growing realization among the policymakers (see Blueprint to safeguard Europe’s Water Resources) that a systemic approach is key to resolving challenges in the water domain. In this context, the first fitness check from the European Commission [9], shows that the upcoming implementation of the WFD and its integration into sectoral policies such as the Common Agricultural Policy (CAP) could be improved [10].

Today, a total number of 111,062 European surface water bodies are under consideration within the scope of the WFD, where 46% of them are to be sufficiently monitored as to the ecological status, while 23% are lacking in situ water sampling [11]. However, 4442 water bodies (4%) still remain with unknown ecological status [12]. This is mainly the result of the strengthening of the existing Member State national water monitoring networks with conventional techniques. Nevertheless, this has consequently increased the cost of monitoring, preventing significant investments that could potentially support perhaps the heaviest pillar of the WFD at the moment, which is the implementation of the Program of Measures (PoMs).

According to the 2nd cycle of the River Basin Management Plans (RBMPs), the EU Member States have a sufficient level of maturity in identifying the pressures that are responsible for the poor or bad ecological status of the problematic water systems. Despite the progress in this domain, the Member States face significant hindrances to efficiently tackle the aforementioned challenges. For this reason, they present slow restoration actions mainly due to insufficient financial funds. Few Member States are leveraging remote sensing technologies for the WFD monitoring, such as Sweden and Finland, showcasing the potential of a hybrid (novel digital and conventional techniques) approach towards an operation and low-cost surface water quality monitoring framework. In this context, water managers increasingly exploit the Copernicus Global land services or relevant thematic spatial products offered by European Space Agency’s (ESA) exploitation platform to plan their activities. However, the use of coordinated EO, and particularly its in situ component, as a key tool in the systematic provision of EO information services is still not adequately taken up or is completely lacking, especially in less developed countries.

2.2. Developing a thorough Picture of the Current State of Surface Water Monitoring EO Capacities

To develop a thorough picture of the current state-of-the-art EO capacities for surface water monitoring, a review was conducted for the period of 2017–2022 (up to August) for journal publications. The selection period coincides with the 1st Review of the RBMPs, enabling us to consider problems and failures in the Union’s effort to achieve good ecological status, as well as the state-of-the-art digital and novel water monitoring systems developed to meet the required needs. Moreover, previous reviews have provided excellent summaries [3,13,14,15,16], not directly related to WFD, and therefore the time chosen is in general a follow up and supplementary of them. In our analysis, we also include relevant research projects (2014–2020) to take into account best practices and valuable insights derived from an operational point of view. Here, we only focus on finished projects.

By using Elsevier’s Scopus and Web of Science citation databases, we carried out our main search in the international literature regarding the journal papers. Furthermore, conference works that were significantly deemed relevant were also considered in this review. Considering the research projects, we restricted our search to projects implemented within Horizon 2020, Interreg, and Life+ funding mechanisms to emphasize the transnational character of these funding programs by using their official searching website engines (https://cordis.europa.eu/projects, https://www.interregeurope.eu/search, https://webgate.ec.europa.eu/life/publicWebsite/search, accessed on 15 March 2022), while for the Horizon projects, an official documentation which covers the specific searching period and our thematic zone was also used [17].

An overview of our review process and what we tried to write down is presented in Figure 2. For the article selection process, we based our analysis on the preferred reporting items for systematic reviews and meta-analyses methodology [18]. The main search found n = 602 possibly suitable studies from the initial n = 1988 studies, after removing the duplicates. After evaluating each study’s abstract for relevance, we identified n = 133 suitable studies. The eligibility followed the full text of each study assessed, resulting in n = 58 studies that were excluded as they did not adhere to the requirements of this review (i.e., did not examine WFD QEs but other water quality indicators). Finally, we selected n = 71 studies for the main analysis (see Appendix A) related to journal papers while n = 11 for research projects.

This work aims to present the interlinkages between the existing EO capacities with the WFD. In this context, the selection of the final papers to be analyzed was not carried out strictly such that all the papers should meet all of the criteria. Moreover, relevant European Commission documents and reports [2,8,15,16,17] were taken into consideration in the analysis.

3. Exploring the Current Situation to Find the Key for the WFD Efficient Implementation

This section lays out the analysis of the review findings, where WFD needs and current limitations are presented in Section 3.1, while an overview of EO capacities and current limitations is given in Section 3.2.

3.1. WFD Needs and Current Limitations

3.1.1. WFD Overview in the EU

The need for policy without borders in EU, considering also the candidate countries, is prioritized to focus not only on water regulations within each nation but also on preventing negative footprints on each other’s water bodies. The willingness of the Union to strengthen its dynamics has a direct impact on the implementation of existing European legislation and directives such as the WFD, the green agreement, the Common Agricultural Policy (CAP), etc.

The WFD establishes a strategic framework for managing the water environment and sets out a common approach to protect, and setting environmental objectives for all inland, transitional, coastal, and groundwaters.

The Directive sets as a central idea their integrated management at the geographical scale of River Basin Districts (RBDs) and uses the term ‘Quality Elements (QE)’ to describe the different indicators (biological, physico-chemical, hydromorphological) of ecological quality that make up the water systems ecological status classification schemes [19]. The overall ecological status of a water body could be determined by the results for the biological or physicochemical QE with the lowest status, i.e., the element worst impacted by human activity. The Directive refers to this as the ‘one out–all out’ (OOAO) principle and until today, this principle has proven to be safe and ideal [20].

A priority and necessary step, for all EU Member States, for the WFD implementation was the preparation of the RBMPs of each country’s RBDs and their submission every six years to describe how each river basin’s environmental goals will be fulfilled. The RBMPs that are going to be submitted in the next period 2022–2027 will consist of the third cycle and are of vital importance because it is the last window for EU Member States to implement policies that will allow them to meet the WFD’s high goals for surface and groundwater quality by the 2027 target for good water health [21]. In light of the above, the needs to address the well-reported gaps as well as the high fragmentation (considering the candidate countries) of in situ infrastructure in Europe are highlighted.

3.1.2. Deeper Understanding of the WFD Requirements

In this work, we analyzed relevant EU documentation and reports [1,12,22,23,24] in order to shape a better understanding of the current situation and the progress of the WFD implementation. Therefore, around 40% of surface waters (rivers, lakes, transitional, and coastal waters) are in good or potential good ecological status while only 38% of them are in good chemical status (i.e., not polluted). Analyzing the 1st RBMPs to the 2nd cycle, we observed that a small percentage of their status had changed. The main significant pressures on surface water bodies are presented in Figure 3. Considering the diffuse sources, we can notice that the main pressure comes from agriculture, while for atmospheric deposition, it comes particularly from mercury.

It should be noted that the proportion of water bodies with unknown status has decreased while confidence in status assessment has simultaneously grown. In particular, the percentage of surface water bodies with unknown ecological status decreased from 16% to 4%, while in chemical status it decreased from 39% to 16%.

Since the 1st cycle of the RBMPs, the general ecological status or potential of water bodies has not improved. The percentage of water bodies with good or better ecological status or potential for all categories slightly decreased. However, since the implementation of the first RBMPs, the ecological status/potential class of about 20% (16,000 surface water bodies) has improved, often by one class but occasionally by two or three classes.

It should be noted that several EU member countries have improved their monitoring networks (e.g., adding stations). However, the overall number of water bodies with unknown status may have decreased, but it is clearly that the overall description of the water systems ecological status is based on the observation and measurement of limited water QEs (Figure 4) [12]. Therefore, it would be misleading and ambiguous to assert that water bodies ecological status could be classified as good or moderate.

In order to provide enough information for a trustworthy assessment of the state of the relevant QE, Member States shall define the monitoring frequency for any parameter during operational monitoring. As a general rule, monitoring should occur at intervals that do not exceed those presented in Table 1 unless greater intervals would be justified based on technical expertise and professional judgment. The frequency selection must result in a respectable degree of accuracy and confidence. The RBMP must provide estimates of the confidence and precision gained by the monitoring method utilized.

Monitoring frequencies shall be selected taking into account the parameters variability resulting from both natural and anthropogenic conditions. The times at which monitoring is undertaken shall be selected to minimize the impact of seasonal variation on the results, and thus ensure that the results reflect changes in the water body as a result of changes due to anthropogenic pressure.

Table 1 presents WFD QEs together with the corresponding monitoring frequencies defined by the Directive and at the same time provides the percentage of water bodies (per water body type) as well as a description of their overall condition according to RBMPs, including information about the key QEs that have not been reported. Considering the percentages from Table 1 as well as Figure 4, we can conclude that many water bodies are classified according to limited or non-representative WFD QEs. For example, 54% of lakes have unknown status level regarding the chlorophyll a (Chl-a) which is recognized as a mandatory QE by the WFD.

Table 1. The percentage of water bodies (per type) with unknown status level for WFD mandatory QEs and their monitoring frequencies (SF means sampling frequency while USL is the Unknown Status Level in percentage).

	Water Body Type
	Rivers		Lakes		Transitional		Coastal
WFD Key QE	SF	USL	SF	USL	SF	USL	SF	USL
Chl-a (phyto.)	monthly/quarterly	97%	monthly/quarterly	54%	seasonal	57%	15 days	27%
Depth var.	annual	50%	every 15 y.	24%	every 5 y.	68%	every 5/6	64%
Oxygenation	fortnightly/monthly	62%	daily/monthly	85%	monthly	47%	15–30 d	51%
Acidification	fortnightly/monthly	62%	monthly/quarterly	72%	-	-	-	-
Salinity	fortnightly/monthly	86%	monthly/quarterly	88%	monthly	89%	15–30 d	95%
Nutrients	fortnightly/monthly	51%	monthly/quarterly	36%	monthly	42%	15–30 d	34%
Transparency	-		monthly/quarterly	83%	monthly	83%	15–30 d	62%
Thermal	fortnightly/monthly	74%	monthly/quarterly	97%	monthly	90%	15–30 d	93%

Table 1. The percentage of water bodies (per type) with unknown status level for WFD mandatory QEs and their monitoring frequencies (SF means sampling frequency while USL is the Unknown Status Level in percentage).

	Water Body Type
	Rivers		Lakes		Transitional		Coastal
WFD Key QE	SF	USL	SF	USL	SF	USL	SF	USL
Chl-a (phyto.)	monthly/quarterly	97%	monthly/quarterly	54%	seasonal	57%	15 days	27%
Depth var.	annual	50%	every 15 y.	24%	every 5 y.	68%	every 5/6	64%
Oxygenation	fortnightly/monthly	62%	daily/monthly	85%	monthly	47%	15–30 d	51%
Acidification	fortnightly/monthly	62%	monthly/quarterly	72%	-	-	-	-
Salinity	fortnightly/monthly	86%	monthly/quarterly	88%	monthly	89%	15–30 d	95%
Nutrients	fortnightly/monthly	51%	monthly/quarterly	36%	monthly	42%	15–30 d	34%
Transparency	-		monthly/quarterly	83%	monthly	83%	15–30 d	62%
Thermal	fortnightly/monthly	74%	monthly/quarterly	97%	monthly	90%	15–30 d	93%

3.1.3. WFD Current Gaps and Limitations

The increasing demands of WFD in many cases is not consistent with the ability of all EU Member States to respond to them appropriately. Many countries are unable to cope either due to financial difficulties, due to delays (i.e., delays in RBMPs drafting phase, unexpected planning delays for the implementation of the POMs etc.), or even due to insufficient mechanisms [12]. Therefore, it is crucial to underline some of the general WFD current limitations recognized through this work.

It should be noted that during the 2nd RBMP cycle, the directive has increased the quality of observation. Considering the entrant countries, their low or even non-existent know-how to meet the needs of the directive as well as their low economic dynamics creates a big challenge for the future WFD implementation. To date, 27 countries are the playmakers while crucial Member State especially from Balkan region such as Macedonia and Albania are trying to enter the EU community (as candidate countries). Such countries could be game changers because they are excellent examples for transboundary river basins while the synergistic management remains a challenge.

The WFD requires river basin level monitoring and management which is a procedure that also requires managing the land and water systems as one with an interdisciplinary, integrated, and holistic approach to be crucial in order to succeed. This overall framework has great advantages, but many EU countries are far from satisfactorily adopting it. River basins are made up of complicated networks of interlinked natural and human processes, therefore implementing the WFD based on catchment management was never going to be simple process [25].

The current analysis indicates that several mandatory QEs have not been monitored while the ecological status classification of the water system is given based on limited QEs number. Even worse, in several EU countries, such as Greece, water bodies are frequently classified based on the know status of their nearest neighbors or water bodies with similar characteristics. This situation also affects the above statement regarding river basin management leading to wrong decisions with bad consequences for the water bodies as well as economic effects. In that regard, technological maturity, and cost restrictions to perform continuous monitoring could be considered significant causes. Nevertheless, the insufficient monitoring of the water QEs will be evident in the foreground as a vitally important limitation for the future of the WFD.

In the framework of the WFD, the overall ecological status of a water body is determined by the OOAO principle where the lowest score of the biological or physico-chemical or hydromorphological QEs classify the water system in the five-level WFD scale (WFD Annex V, 1.4.2 (i)). This combination rule is appropriate and justified if several stressors contribute to the degradation of the various biological QEs. However, it might be problematic if different stressors have an impact on different biological QEs because this can make it more difficult to examine each biological QE individually [26], leading to over-precautionary results with more sites failing than should. However, the fundamental cause of this issue is the inclusion of QEs with a high level of uncertainty in the assessment [10]. Furthermore, Kats et al. (2022) [27], through an online survey with the participation of two water regional authorities from Netherlands, showed that 63% agree and 32% strongly agree that the OOAO principle does not provide an accurate picture of the water quality within their water authorities.

Following a thorough review, we concluded some significant points:

• The drafting of RBMPs during the COVID-19 pandemic has been challenging, and several EU member states are facing issues submitting their drafts on time.
• In some cases, the draft RBMPs anticipate that objectives will not be achieved before 2050 [22].
• According to the 2019 Fitness Check, slow implementation, a lack of funding, and a failure to integrate environmental objectives into sectoral policies are the main obstacles to preserving and restoring water bodies.
• Lack of budget allocation for RBMPs is a main constraint due to the failure to recover environmental and resource costs from strong economic sectors.
• According to Zingraff-Hamed et al. (2020) [28], the implementation problems are caused by a lack of horizontal cooperation and communication, not by a lack of adequate policy integration as suggested by earlier studies.
• Conflicts between the water policy and other key policies such as those in the agricultural domain, CAP, have arisen [24]. The ambitious WFD PoMs could not be completed unless the gap between these policies is bridged.

Additionally, with conventional techniques, the data management is often time-consuming and usually the data are not easily accessible by competent and local management authorities, making difficult to truly assess the situation and subsequently manage many of the surface water bodies with social, environmental and economic costs. Advanced digital infrastructures for data handling, processing, and storage are missing from the majority of the Member States while data fusion seems to be a non-mature term.

3.2. Overview of EO Capacities for Water Quality Monitoring and Limitations

In this section, we provide the main findings of our analysis to address the potential of the EO resources for WFD support. Special focus is placed on the sensors in use, the observed water QEs, as well as the computer science techniques (machine/deep learning, artificial intelligence, etc.). At the end, we address the current limitations and we try to put all the gained information in a “box” highlighting the suitability of the different existing sensors and technologies for their potential use in the framework of WFD implementation.

3.2.1. Quality Elements (QEs) and Water Body Types

We prioritized the indicators that are relevant to the WFD and considered as mandatory water QEs (biological, physico-chemical, hydromorphological) for determining the overall status of water bodies. For more technical details (e.g., valuable wavelengths, resolution etc.), we refer the readers to previous reports [3,14,15,16]. Based on our analysis, a significant number of WFD QEs appeared to be predictable and could be monitored via several EO resources delivering near-real time monitoring at low cost. In this context, a detailed ranking is illustrated in Figure 5.

Biological QEs

Among the biological QEs, the Chl-a is extensively studied (43%) and this could be explained due to its importance for describing the water quality condition and the relative ease of observing it with innovative EO technologies. In this context, three studies examined the Chl-a QE, referring also to its importance for the WFD, using various data streams. Particularly, Attila et al. (2018) [4] derived services for Chl-a estimation to assess the status of coastal waters. Dörnhöfer et al. (2018) [5] integrated remotely sensed Chl-a into the WFD trophic state assessment for a lake case study, while Asper and Alikas (2019) [29] derived the ecological status of small lakes based on Chl-a for WFD reporting purposes in Estonia. Continuing in terms of the biological QEs, Elhag et al. (2019) [30] utilized Sentinel-2 data to monitor Chl-a coupled with regression analysis in a lake case study while two studies by Zoffoli et al. (2020) [31] and Yashira et al. (2021) [32] used Sentinel-2 imagery data to provide high spatial resolution maps of seagrass at coastal zones; the first verified the results with USV data with an overall accuracy value of 62.2% and the second used one very high-resolution scene from WorldView-02 in order to evaluate the representativeness of the Sentinel-2 pixel size.

Pyo et al. (2019) [33] used hyperspectral imagery data from airborne flight (ASIA Aero Survey) coupled with in situ measurements by a handheld spectroradiometer to estimate phycocyanin and Chl-a in a river case study. Pokrinwiskie et al. (2022) [34] explores the synergy of hyperspectral UAV (Unmanned Aerial Vehicle), ground sampling and laboratory analyses to estimate Chl-a, turbidity, and phycocyanin in a pond-controlled environment.

In addition, according to Topp et al. (2020) [15], the algal biomass of a waterbody also affects its overall biological productivity, called the trophic state. In the same context, to monitor the trophic state as required by the WFD, Submerged Aquatic Vegetation (SAV) needs to be monitored regularly [35,36]. With the description, others (6%) are mostly biological water quality indicators (e.g., DIN, DIP, macroalgal etc.) that are not considered mandatory QEs in WFD, but the studies presented strong correlation with our work in general.

Physico-Chemical QEs

Very high score in the ranking was achieved by the transparency key feature where the turbidity ranked second in the row with 16%, Secchi depth with 11%, while the TSS follows with 8.5%. The aforementioned parameters are considered as mandatory QEs while Delegido et al. (2019) studied the combination of the turbidity and Secchi depth provided an operational method for estimating turbidity with Sentinel-2.

Interesting findings resulted for the TP (Total Phosphorous) and TN (Total Nitrogen) where five studies [7,37,38,39,40] from the total number of 71 examined the TP while four studies [7,38,39,41] the TN. The impressive thing is that for both QEs, the UAVs’ contribution was noteworthy reporting 40% for TP [38,39] and 60% for TN [38,39,41]. These findings show the great potential of the UAV platforms in the domain of water monitoring.

Lastly, Suspended Particular Matter (SPM) was also studied in estuary case studies by Shang et al. (2018) [42] and Sent et al. (2021) [43].

Hydromorphological QEs

We also have notable findings regarding the hydromorphological QEs, where the surface water extent leads in the ranking followed by bathymetry and water flow. Huang et al. (2018) [44] used a fully automated approach for mapping surface water extent in a coastal zone using satellite-based Sentinel-1 C-band Synthetic Aperture Radar (SAR) data. Sun et al. (2020) [45] used GAOFEN-1 satellite to study a river water body extraction. Markret et al. (2020) [46] used SAR data from Sentinel-1 satellite to generate surface water extent maps. In a river case study, they validated the results using manually analyzed high spatial resolution Planet Scope data.

Basith and Prastyani (2020) [47] and Parente and Pepe (2018) [48] used high resolution WorldView-03 imagery data to obtain bathymetry map in sea and river case studies accordingly while Goraj et al. (2018) [6] used an observer plane with LiDAR scanning system to identify bathymetry indicator in river.

Madeo et al. (2020) [49] developed a custom made USV (Unmanned Surface Vehicle) platform with multiple sensors on board that could measure the water flow, among other indicators (DO, salinity, pH), but the water flow sensor requires the vehicle to be motionless when measuring the value via the water flow sensor.

The overall trend of the researchers appeared in the study of the lake environment with 40% of the total 71 studies, while river and coastal zones followed in the ranking with 20% for each (Figure 6). Reservoirs, estuaries, sea, lagoons, and ponds were found to be studied by the researchers but with significantly lower percentages than the three aforementioned water systems.

Figure 7 provides an illustration of the distribution of the four predominant indicators (Chl-a, Turbidity, Secchi depth, TSS) per water body type, as resulted from Figure 5. Only Chl-a has been studied in all water bodies while the same results come for turbidity excluding only the sea environment. In contrast, the Secchi depth and TSS indicators were studied mostly in lake and river water bodies.

3.2.2. The Spaceborne Domain

An overview of the various EO technologies in the selected studies is illustrated in Figure 8 in which with the description other, we have two works related to airborne flights [6,33] and one work that used a spectrum measurement fixed platform [50]. Based on the analysis, the two most common satellite sensors used for water quality monitoring are Sentinel-2 from the European Copernicus Space component and National Aeronautics and Space Administration (NASA) Landsat-8 archive (43% of the studies for Sentinel-2 and 27% for Landsat). In that regard, the free and open data policy of spaceborne data (e.g., Landsat and Sentinel) is key to the ongoing success of EO domain.

Furthermore, several studies used synergies among different satellite sensors, some of them for comparison reasons [7,51] while some others as supplementary data sources [52]. Dörnhöfer et al. (2018) [5] includes a series of remote sensing data from MODIS, Landsat 7/8 and Sentinel-2 as well as in situ measurements to analyze the mutual inter-comparability of satellite products and in situ data, for Chl-a estimations, in a lake case study. Gohin et al. (2019) [53] explores the same spaceborne sensing systems to estimate Chl-a concentrations in a coastal zone case study. Erena et al. (2019) [54] explores the combination of Landsat-8 with SPOT satellite to overcome the cloud problem prevented in the Landsat images with special attention to Chl-a and turbidity predictions in lagoon environment. Wang et al. (2022) [50] presents a methodological framework by using the Google Earth Engine (GEE) cloud computing platform, tested in a lake case study, in which surface reflectance values from multi-sensor satellite observations (Landsat-5/7/8 and Sentinel-2) are pairing automatically, with ground water quality samples in time and space to form match-up points. A recent work was carried out by Hakimdavar et al. (2020) [55] where the combination of Landsat-5/7/8, MODIS, and Sentinels-1/2 were used to measure surface water extent for various water bodies (lakes, river, estuary) at 250 m and 30 m spatial resolution, concluding that statistical comparisons between different surface water data products can help provide some degree of confidence for countries during their validation process. As an additional tool for the monitoring of macroalgal blooms in estuaries, where the presence of increased cloud cover presents an extra restriction, Karki et al. (2021) [56] analyzed Landsat-5/8 imagery and Sentinels-1/2. In situ mapping was compared to the results for verification and validation. Li et al. (2022) [57] take the advantages offered by the combination of Landsat-5/8, Sentinel-2, and GAOFEN-2 and they propose a new framework for accurate water extraction by using unsupervised deep learning and Normalized Difference Water Index (NDWI) of multispectral images.

In addition, the fact that the most common satellite sensor synergy is Landsat-8 with Sentinel-2 (7 studies) gains our attention. Bresciani et al. (2018) [58] used the advantage of Landsat-8 and Sentinel-2 optical sensors to increase the information collected from in situ data on algal blooms dynamics in deep subalpine lakes. Similarly, Bonansea et al. (2019) [59] assessed the suitability of these multispectral systems for estimating and mapping of Secchi disk transparency. Govedarica et al. (2019) [7] leverages the Landsat-8 data along with machine learning algorithms to extract EO based information for turbidity, TSS, TN, and TP. Despite the Landsat-8 capabilities, Sentinel-2 higher spatial and temporal resolution being a better alternative for monitoring water quality explains the significant increase of its use in relevant studies. The effectiveness of Sentinel-2 and Landsat-8 sensors for assessing Chl-a, Secchi disk depth and turbidity was examined by Pizani et al. (2020) [51] by developing multiple regression models with adequate predictive performance. Ghuvita et al. (2021) [60] calculated the concentration of suspended sediment in a river case study by using Landast-8 and Sentinel-2 satellite imagery. Hafeez et al. (2022) [61] analyzed if the consistency between the Landsat-8 and Sentinel-2 products were evaluated through an extensive evaluation of spectral consistency with case studies in estuary, coast, and lake for algae monitoring. The analysis reveals that the difference in the algae area increases with the time difference between the same-day overpass. Cabarello et al. (2022) [62] examined the evolution of the key indicators of water quality during the most recent ecological crisis in 2021, which resulted in a significant loss of benthic vegetation and strange death rate events affecting various aquatic species.

Based on our analysis, a significant but smaller number of studies explore coarser resolution (>250 m) sensors such as MODIS (Moderate Resolution Imaging Spectroradiometer) (13%) [5,52,55,63,64,65,66,67] and MERIS (MEdium Resolution Imaging Spectrometer) (8.6%) [4,52,53,68,69,70]. For example, Zheng and DiGiacomo (2017) [52] collected match ups for four sensors including the SeaWiFS, the MODIS onboard Aqua, the MERIS, and VIIRS to develop a new Chl-a model based on the semi-analytical approach, in a coastal zone case study. However, their spatial resolution set them as not proper monitoring systems for water quality mapping in small surface water systems.

Recently, new EO technologies offering higher spatial (<3 m) and spectral resolution (>50 bands) will likely revolutionize surface water monitoring and reveal dynamics at unprecedented levels of detail. Very high resolution data (e.g., Planet, PRISMA, Worldview etc.) allow for an accurate monitoring of water parameters, mainly focusing on Chl-a, even for small-scale water bodies [35,46,71]. However, they offer limited spectral bands in the VNIR region. On the other hand, hyperspectral remote sensing systems, such as DESIS and PRISMA, can capture subtle differences in the properties of reflectance spectra. For instance, Bresciani et al. (2022) [72] evaluate these hyperspectral systems to retrieve water quality parameters over four Italian lakes. It should be noted that the capabilities of hyperspectral systems have already been explored in previous studies making use of airborne systems [6,33] or fixed platforms [50] equipped with such kinds of sensing instruments.

Recent advancements in unmanned vehicle (both UAV and USV) capacities have brought them to the forefront of innovation, becoming a crucial component in operational data-driven water surface monitoring. In this context, UAV platforms follow in the ranking (14% of the studies) making use of various sensors able to collect high spatial and spectral data. Thus, UAVs have proven to be valuable tools for monitoring several water QEs such as Chl-a, TN, TSS and pH. A detailed review regarding UAVs in the aquatic environment is provided by Kislik et al. (2018) [73].

Special attention was given also to USV moving platforms as a water monitoring tool where it is noteworthy that four studies [32,49,74,75] used USV platforms with several sensors on board. It should be noted that 75% are custom-made solutions (see Table 2). These findings highlight the need for providing commercial incentives for the development and expansion of in situ networks.

In order to provide the strength of the in situ moving platforms (UAVs, USVs, airborne), it was considered appropriate and interesting to collect the variety of sensors in the studies under examination. The sensors on board in the different moving platforms with the observed parameter and the corresponding studies are presented in Table 2.

It was also considered necessary and useful for the analysis to focus on the distribution of the four predominant indicators, in terms of the EO resources and the following graph emerged (Figure 9). The Sentinel-2 satellite has been extensively used (29 studies) compared to the rest EO resources for the predominant QEs. Landsat-8 has a significantly lower number of studies than Sentinel-2, especially for the Chl-a and Secchi depth indicators but similar numbers for turbidity and TP. MERIS and MODIS satellites are also used for Chl-a, turbidity, and Secchi depth. It should be noted that a significant number of studies make use of UAVs for Chl-a, turbidity, and Secchi depth estimations.

Table 2. The variety of sensors on board Unmanned Aerial Vehicles (UAVs), Unmanned Surface Vehicles (USVs) and airborne platforms and their potential to monitor different water quality parameters in various water bodies.

	Platform	Sensor/Equipment	Parameter (s)	WB	Ref.
UAV	Quadcopter md4-1000	Sony Alpha ILCE-5100 camera with a 24.3 MPix res.	Aquatic vegetation cover	Lake	[76]
	Multirotor G4 SkyCrane UAS	Headwall Photonics Nano-Hyperspec sensor, 270 spectral bands across a spectral range 400–1000 nm	Chl-a, Turbidity, Phycocyanin	Pond	[34]
	DJI P4 Multispectral	Six 1/2.9” CMOS, including 1 RGB sensor for visible light imaging and 5 monochrome sensors for mult. imaging	Chl-a, TN, TP, COD	River	[39]
	DJI Matrice 600 PRO®	Nano-Hyperspec®, 400–1000 nm with 272 spectral bands	TN	Lake	[41]
	Tholeg THO-R-PX8	Senop Oy ’Rikola’ Hyperspectral (HS) Camera, VNIR spectral range between 504 and 900 nm	pH	River	[77]
	Multi-rotor UAV	Rededge-MX multi-spectral camera	Chl-a, TN, TP, NH3-N, Turbidity	River	[38]
	Remo-M	Parrot Sequoia camera	Chl-a	River	[74]
	SenseFly, Swinglet CAM model	Canon ELPH 110HS	TSS	Lake	[78]
	S800 EVO Hexacopter	Canon EOS 5DS R and Headwall Nano-Hyperspec® 274 bands and 400–1000 nm	Reef monitoring	Sea	[79]
	Custom hexacopter	Custom off-the-marker sensors	Temperature, EC, DO, pH	Pond	[80]
USV	Custom USV	Van Veen grab sampler	Sediment sampling	Lake	[75]
	Custom USV	AlgaeChek Ultra fluorometer	Chl-a	River	[81]
	Data provided by Satria MGA, 2019	Not specified-underwater camera	Seagrass	Coast	[32]
	Custom USV	Vernier pH, oxidation, salinity, DO, flow rate sensor	DO, salinity, water flow, pH	River	[49]
Airborne	Vulcanair P68 TC Observer plane	Parrot Sequoia camera and Riegl VQ-1560i-DW LiDAR system	Bathymetry	River	[6]
	ASIA Aero Survey	ASIA Eagle-SPECIM, 400–970 nm	Chl-a, Phycocyanin	River	[33]

Table 2. The variety of sensors on board Unmanned Aerial Vehicles (UAVs), Unmanned Surface Vehicles (USVs) and airborne platforms and their potential to monitor different water quality parameters in various water bodies.

	Platform	Sensor/Equipment	Parameter (s)	WB	Ref.
UAV	Quadcopter md4-1000	Sony Alpha ILCE-5100 camera with a 24.3 MPix res.	Aquatic vegetation cover	Lake	[76]
	Multirotor G4 SkyCrane UAS	Headwall Photonics Nano-Hyperspec sensor, 270 spectral bands across a spectral range 400–1000 nm	Chl-a, Turbidity, Phycocyanin	Pond	[34]
	DJI P4 Multispectral	Six 1/2.9” CMOS, including 1 RGB sensor for visible light imaging and 5 monochrome sensors for mult. imaging	Chl-a, TN, TP, COD	River	[39]
	DJI Matrice 600 PRO®	Nano-Hyperspec®, 400–1000 nm with 272 spectral bands	TN	Lake	[41]
	Tholeg THO-R-PX8	Senop Oy ’Rikola’ Hyperspectral (HS) Camera, VNIR spectral range between 504 and 900 nm	pH	River	[77]
	Multi-rotor UAV	Rededge-MX multi-spectral camera	Chl-a, TN, TP, NH3-N, Turbidity	River	[38]
	Remo-M	Parrot Sequoia camera	Chl-a	River	[74]
	SenseFly, Swinglet CAM model	Canon ELPH 110HS	TSS	Lake	[78]
	S800 EVO Hexacopter	Canon EOS 5DS R and Headwall Nano-Hyperspec® 274 bands and 400–1000 nm	Reef monitoring	Sea	[79]
	Custom hexacopter	Custom off-the-marker sensors	Temperature, EC, DO, pH	Pond	[80]
USV	Custom USV	Van Veen grab sampler	Sediment sampling	Lake	[75]
	Custom USV	AlgaeChek Ultra fluorometer	Chl-a	River	[81]
	Data provided by Satria MGA, 2019	Not specified-underwater camera	Seagrass	Coast	[32]
	Custom USV	Vernier pH, oxidation, salinity, DO, flow rate sensor	DO, salinity, water flow, pH	River	[49]
Airborne	Vulcanair P68 TC Observer plane	Parrot Sequoia camera and Riegl VQ-1560i-DW LiDAR system	Bathymetry	River	[6]
	ASIA Aero Survey	ASIA Eagle-SPECIM, 400–970 nm	Chl-a, Phycocyanin	River	[33]

3.2.3. Modeling, Processing, and Data Handling Digital Infrastructures

It is important to note that 28 of the final 71 papers that were analyzed used Artificial Intelligence (AI) modeling techniques to estimate the QEs (Figure 10). This finding indicates a widespread use of AI technology coupled with various EO technologies. For more technical details, we refer the readers to recent works [15,16,82,83,84].

The general technological progress, in recent years, gave the opportunity to the scientific and research community to explore new computational methods for water QEs monitoring and prediction, taking advantages from different EO data sources along with existing open databases related to water ecosystems. The results from Figure 10 could be summarized as following:

• The most used algorithms for this analysis come from the group of AI Neural Networks (NN), with 54% of studies utilizing them.
• The Random Forest (RF) algorithm follows with 46%, and Boosting algorithms and Support Vector Machines (SVM) are also high in the ranking with 25% and 21.4%, respectively.
• PLSR and DT are used less, due to the increasing interest in AI methods, such as LSTM which can be more effective at capturing the linear and non-linear correlations for water quality parameters estimation.
• The majority of studies utilizing AI algorithms focused on Chl-a (around 46%) and TN (14.28%).

Parameters and AI models per water body type are provided in Table 3. In the absence of ground data, Faizi and Mahmood (2022) [85] suggest an AI-driven approach to comprehend the spatio-temporal dynamics of water clarity patterns in a reservoir case study.

It should be noted that nine out of the ten studies utilized data derived from UAVs in the modeling analysis. It is also very important to emphasize that most applications using prediction models appeared in 2020 with five studies and with nine studies each for 2021 and 2022. Therefore, it is accepted that the use of prediction models is completely in line with the evolution of EO technologies and the wider idea of open data.

Staying in the same vein, it was considered important to examine the mechanisms used by the researchers, behind the data handling and processing, in a framework related to big data management and processing. More than 45% of the studies used advanced tools and software for data handling and processing while it should be noted that data pre-processing takes approximately 70% of the overall data analysis. A recent work provided by Gomez et al. (2020) [86] summarizes most of the tools that are exploited to monitor water quality parameters.

Table 3. A detailed presentation of the sensors, parameters, and models per water body type for the 28 examined studies related to modeling techniques (for EO resource column, S = Sentinel and L = Landsat, while for Water Body column, L = Lake, R = River, C = Coast, S = Sea, Re = Reservoir, La = Lagoon, E = Estuary).

Year	EO Resource	Parameter	Model	Water Body	Ref.
2022	S-2	TSM, Secchi depth	ANN	Re	[85]
2022	MODIS	Temperature	RF	S	[66]
2022	L-8	MNDWI	CNN	L	[87]
2022	S-2, L-8/5, GAOFEN-2	NDWI	RF	R, L	[57]
2022	L-7/8	Chl-a, Transparency, TP	MLR	L	[40]
2022	UAV	Aquatic vegetation cover	ANN	L	[76]
2022	Spectra meas. fixed platform	Chl-a	BP, SVR, RFR	L, Re, R	[50]
2022	MODIS	DIN, DIP	DBN, MPNN, GRNN	C	[64]
2022	UAV	Chl-a, TN, TP, COD	LASSO, BP, RF, XGBOOST	R	[39]
2021	S-2	Chl-a	MDN	L	[88]
2021	UAV	TN	RF, Bagging alg., XGBOOST	L	[41]
2021	S-1/2, L-5/8	Macroalgal	ANN	E	[56]
2021	S-2	Chl-a	OC3, SLR, MLR, GAMs	C	[89]
2021	UAV	pH	RF, SVM, SAM	R	[77]
2021	UAV	Chl-a, TP, TN, NH3-N, Turbidity	GA-BOOST, DNN, RF, GA-RF, AdaBOOST, GA_ADABOOST	R	[38]
2021	S-2	Chl-a	RF, SVM, ANN, DNN	La	[86]
2021	S-2	NDWI, NWI, EWI, AWE-nsh	RF, SVM, PLSR, PLSR-SVM	L	[90]
2021	S-3	Chl-a	NN	S	[91]
2020	GAOFEN-1	Surface water extent	RF, ADABoost, DT	R	[45]
2020	MODIS, S-2	Chl-a	SVR, RFR, LSTM	C	[63]
2020	S-2, 3	Chl-a	MDN	C, L	[92]
2020	S-2	Chl-a, TSS	Cubist	R	[93]
2020	L-5/7/8, S-2	Chl-a	SVM	L	[94]
2019	L-8, S-2	Turbidity, TSS, TP, TN	NN, SVM	R	[7]
2019	UAV	TSS	ANN	L	[78]
2019	Airborne flight	Phycocyanin, Chl-a	CNN	R	[33]
2018	UAV	Reef monitoring	SVM	S	[79]
2018	S-1, L-8	Surface water extent	RF	C	[44]

Table 3. A detailed presentation of the sensors, parameters, and models per water body type for the 28 examined studies related to modeling techniques (for EO resource column, S = Sentinel and L = Landsat, while for Water Body column, L = Lake, R = River, C = Coast, S = Sea, Re = Reservoir, La = Lagoon, E = Estuary).

Year	EO Resource	Parameter	Model	Water Body	Ref.
2022	S-2	TSM, Secchi depth	ANN	Re	[85]
2022	MODIS	Temperature	RF	S	[66]
2022	L-8	MNDWI	CNN	L	[87]
2022	S-2, L-8/5, GAOFEN-2	NDWI	RF	R, L	[57]
2022	L-7/8	Chl-a, Transparency, TP	MLR	L	[40]
2022	UAV	Aquatic vegetation cover	ANN	L	[76]
2022	Spectra meas. fixed platform	Chl-a	BP, SVR, RFR	L, Re, R	[50]
2022	MODIS	DIN, DIP	DBN, MPNN, GRNN	C	[64]
2022	UAV	Chl-a, TN, TP, COD	LASSO, BP, RF, XGBOOST	R	[39]
2021	S-2	Chl-a	MDN	L	[88]
2021	UAV	TN	RF, Bagging alg., XGBOOST	L	[41]
2021	S-1/2, L-5/8	Macroalgal	ANN	E	[56]
2021	S-2	Chl-a	OC3, SLR, MLR, GAMs	C	[89]
2021	UAV	pH	RF, SVM, SAM	R	[77]
2021	UAV	Chl-a, TP, TN, NH3-N, Turbidity	GA-BOOST, DNN, RF, GA-RF, AdaBOOST, GA_ADABOOST	R	[38]
2021	S-2	Chl-a	RF, SVM, ANN, DNN	La	[86]
2021	S-2	NDWI, NWI, EWI, AWE-nsh	RF, SVM, PLSR, PLSR-SVM	L	[90]
2021	S-3	Chl-a	NN	S	[91]
2020	GAOFEN-1	Surface water extent	RF, ADABoost, DT	R	[45]
2020	MODIS, S-2	Chl-a	SVR, RFR, LSTM	C	[63]
2020	S-2, 3	Chl-a	MDN	C, L	[92]
2020	S-2	Chl-a, TSS	Cubist	R	[93]
2020	L-5/7/8, S-2	Chl-a	SVM	L	[94]
2019	L-8, S-2	Turbidity, TSS, TP, TN	NN, SVM	R	[7]
2019	UAV	TSS	ANN	L	[78]
2019	Airborne flight	Phycocyanin, Chl-a	CNN	R	[33]
2018	UAV	Reef monitoring	SVM	S	[79]
2018	S-1, L-8	Surface water extent	RF	C	[44]

With the advances of EO technology (e.g., launch of advanced satellites, extensive use of UAVs, etc.), both the veracity and volume of EO data have significantly increased. Hence, research groups have adapted cloud-based processing platforms to efficiently handle and analyze EO big data. Specifically, around 20% of the studies leverage the unprecedented opportunities offered by GEE [95] to greatly simplify the EO data analysis and generate water related spatial explicit indicators. The applications vary at different scales and considering multiple parameters such as Chl-a [94], turbidity [7], TSS [7,60] and surface water extent [55]. Furthermore, the multiple APIs (Application Programming Interfaces) that are supported by GEE enable the researchers to deploy specific algorithms, making GEE even more popular [96]. In the same context, the Joint-Research-Center-hosted (JRC) platform (JEODPP) [97] was developed for EO big data processing and visualization needs.

For water resources management, the Data Cubes (DC) also provide tremendous capabilities since they allow us to stack heterogeneous datasets along the time dimension, enabling the formation of a sparse EO-based time series collection that can be used to monitor water quality at various times. A very interesting term “think global, cube local” related to the EO DCs was provided recently by Sudmans et al. 2022 [98], which means that open source EO DCs can be self-hosted to serve specific regional or thematic requirements and user groups, while Nextcloud or Owncloud could be used as a self-hosting options. Malthus et al. (2019) [99] develop and deploy a DC, based on the Open Data Cube (ODC) [100], to organize Landsat data as analysis ready data, to analyze big time series in order to develop a protype algal bloom alerting tool in Australia. In addition, a detailed overview regarding the ODCs contribution to national policies and practices was provided by Dhu et al. (2019) [101].

The first action on the national scale comes from the EO DC [102] established in Australia provided the Digital Earth Australia [103]. This technology was adopted by the newer ODC [100], where it formed the basis for many of today’s fundamental operational digital infrastructures on regional and national scales such as Digital Earth Africa, Digital Earth Pacific, Swiss DC, Colombia DC, Vietnam DC, Armenia DC, Catalonia DC, Mexican Geospatial DC, and Virginia DC. In the same context, the xCube technology was implemented in the Euro DC as part of the Euro DC facility as well as for the EODataBee which is a DC service, developed in the framework of H2020 DCS4COP project, generating high-quality information for value-adding industry in the coastal and inland water market.

Table 4. Examples of DC technology adopted at various scales in support of EO-driven water resources management (in sources column, S = Sentinel and L = Landsat).

Initiative	Description	Sources	Water Related Services	Powered by	Access Link (Accessed on 15 June 2022)	Ref.
DE Australia	Embed satellite imagery and data into decisions that support a sustainable Australian environment.	S-2, L-8	Water availability (WOfS),	ODC	https://www.dea.ga.gov.au/	[103]
DE Africa	Apply EO data to address local and national needs as well the objectives of the GEO and the 2030 SDGs Agenda.	L-5/7/8/9, S-1/2	Coastline erosion, WOfS,	ODC	https://www.digitalearthafrica.org/	[104]
Swiss DC	Support the Swiss government in environmental monitoring and reporting.	L-5/7/8, S-1/2	Snow cover evolution	ODC	https://www.swissdatacube.org/	[105]
Virginia’s ODC	Helping to solve Virginian’s environmental and social challenges through the use of satellite data.	Landsat’s	Water quality & supply, coastal resiliency	ODC	https://www.data4va.org/	-
Vietnam DC	Address the needs of satellite data users, giving them a better picture of their land resources and land change.	S-1/2, L, SPOT, VNREDSat 1, ALOS	Water management	ODC	https://vnsc.org.vn/en/	[106]
Armenia DC	A game-changing technology for remote sensing EO and national-level data visualization.	L-5/7/8, S-1	Water detection, water quality, coastal change.	ODC	http://datacube.sci.am/	[107]
Catalonia DC	A DC solution useful for small natural regions such as the Catalonia sub-national region.	S-2	NDWI, water vapor,	ODC	http://datacube.uab.cat/	[108]
Mexican Geospatial DC	Combination of in situ with EO data to produce an ever-expanding range of derived, decision -ready products.	L-4/5/6/7/8	MNDW, WOfS	ODC	http://en.www.inegi.org.mx/default.html	[109]
EODataBee	Generation of high-quality information for value-adding industry in the coastal and inland water market.	S-2	Chl-a, PAR, Turbidity, SPM, Temperature	xCube	https://eodatabee.eu/	-

provides examples of operational DCs initiatives around the globe related to the water ecosystem, proving that the big data management is not an excuse for not using new technologies when there are existing tools that are being applied with great success.

Our work highlighted the recent trend to use EO big data processing platforms that are also able to execute user-defined algorithms by the various research groups which was difficult to deploy on the existing geographical information systems as a simple plugin, such as the Semi-automatic Classification Plugin [88].

However, we should emphasize that research groups typically have legacy algorithms (e.g., deep learning) that are difficult to re-write and deploy on GEE or DC platforms, or researchers often find it difficult to deploy specific pre-processing routines (e.g., radiometric calibration, mosaicking). Hence, there are many software available for these needs such as the SNAP (Sentinel Application Platform) tool [110,111] and the ENVI image analysis software [32,54] that couple deep learning with all types of data including multi-spectral, hyperspectral, thermal, LiDAR and SAR and the PIX4D for UAV data processing [39,74,78] and the MEPHySto offering a hyperspectral pre-processed library [73].

3.2.4. Project Initiatives—Lessons Learned and Impact

EO in support of water-related information services provision has been tested in a wide range of operational activities. These range from transnational projects targeting representative water bodies to large-scale demonstrators at a wider scale, aiming to bridge the gap between science and user needs; and provide the basis for an EO-driven water resources monitoring framework of WFD’s QEs. These activities have made great strides toward the development of technological solutions. For instance, the majority of Horizon projects (e.g., MONOCLE, EOMORES etc.) include activities whereby RS data from existing multispectral spaceborne systems (e.g., Sentinels and Landsat) and spectral indices (e.g., NDWI, EWI, etc.) have been used with, and validated by, in situ sensor systems (buoys, ships, etc.), pointed spectroscopy (high spectral resolution), and UAVs (e.g., high spatial and spectral image spectroscopy). Further, by combining a host of data, large scale demonstration projects monitored essential variables in support of Sustainable Development Goal (SDG) implementation and provided services to make decisions about when, where, and how to relevant communities using friendly online decision support software. Moreover, these projects foster organizational capacity building focusing on three pillars: infrastructure, human capital, and organizational development, especially in less developed countries. It should be noted that a co-design approach has been prioritized by various projects (e.g., SWOS, Aqua3s, etc.) in order to gather initial top-level requirements from relevant consortium partners and end-user groups, by interviewing project members and through targeted questionnaires, to produce a reference framework of requirements. This guides the development of applications and services (e.g., MONOCLE FreshWater Watch platform) that meet exactly the user’s needs, having a sustainable effect beyond the project’s lifetime.

Table 5. Project overview analysis (for study area column: L = Lake, R = River, C = Coast, S = Sea, Re = Reservoir, La = Lagoon, E = Estuary; while for spaceborne column: S = Sentinel, L = Landsat, WV = Worldview, QB = Quickbird).

Project	Objective	Study Area	Parameters	Sources
				Spaceborne	In-Situ
MONOCLE (H2020)	Improve sensors on autonomous platforms and incorporate EO to fill the gaps of the gathered in-situ information	C, L, and E	Chl-a, Turbidity, SPM, Temperature, Transparency	S-1/2/3, Suomi-NPP VIIRS, MODIS	Buoys, ships, drones, hyperspectral radiometer, WISPstation, mini-Secchi disk, smartphones
EOMORES (H2020)	Integrate cutting-edge, optical in situ instrument with free and open satellite data and sophisticated numerical models.	R, L, and C	Chl-a, TSS, Turbidity, Transparency, CDOM, Cyanobacteria biomass, Temperature	S-2/3, L-8, MODIS, MERIS, VHR	In situ point measurements (WISPstation)
CoastObs (H2020)	Develop a service platform for coastal water monitoring with validated products derived from EO	C	Algal blooms, chl-a, SPM, TSS, Temperature, SPM, turbidity, seagrass, phytoplankton- macrophytes,	S-2, L-8	-
INTCATCH (H2020)	Bring the lab on the field through the development and application of Novel, integrated Tools for monitoring and management Catchments	L, R, E, and Re	Microbiological analyses, Metagenomics analysis, E. coli, Heavy metals	-	Eco-innovative autonomous and radio-controlled boats, sensors, DNA test kits and run-off treatment technologies
Aqua3s (H2020)	Combined novel technologies in water safety and security, aiming to standardize existing sensor technologies complemented by state-of-the-art detection mechanisms.	R, Re	Area classification, Oil spill detection, flood delineation	S-1/2	UAVs, social media observations from citizens
SWOS (H2020)	User friendly wetland monitoring and information service by taking full advantage of satellite imagery data	Wetlands	-	S-1/2, L-8	In situ databases, results from the ESA Globwetland projects.
EUGENIUS (H2020)	Develop viable market based EO services by involving the end users in the whole life cycle of the project.	Marine env.	Chl-a, TSS, Turbidity, transparency, Temperature	L-7/8, S-2, SPOT, Pleiades, WV, QB	-
DCS4COP (H2020)	Addresses the downstream challenges of big data integrating Copernicus services	L, R, and C	Chl-a, PAR, Turbidity, SPM, Temperature	S-1/2/3	-
AquaNEX (Interreg)	Develop and test integrated solutions for the most effective monitoring of the aquatic and terrestrial ecosystem	R, L	Chl-a, NO3-N, sediments, pH, transparency, bathymetry	S-2, L-8	Autonomous Water Telemetry Sensing System (AWTSS-1), UAV
AG_UAS (Life+)	Develop a cost-effective, spatial tool for more efficient, sustainable, water monitoring and management, to bridge the gap between traditional satellite remote sensing and airborne remote sensing	L and R	Detection of discharges into river basins, monitoring of the ecological and chemical status of surface waters,	-	UAV with thermal infrared and multispectral camera

provides a detailed project overview analysis of the total 11 research projects under examination.

3.2.5. Current EO Dynamics and Limitations

Primarily, it should be mentioned that we will try to identify the main strengths and weaknesses of the new technological waves always in relation to the needs of the WFD.

Spaceborne Dynamics and Limitations

Based on our analysis, remote sensing technologies could provide an alternative cost-effective solution for surface water monitoring with higher spatial and temporal coverage. Furthermore, the innovative EO technologies could cover larger areas and locations where data are missing or are limited while it can increase the total number of water bodies to be monitored and with the potential of monitoring more QEs per water system. The spatial and temporal resolution of those technologies is at a satisfactory level and is constantly improving with the help of new generation satellite sensors [46,72] and with the worldwide general technological evolution. As an example, in the framework of WFD, lakes larger than 50 ha should be monitored and classified according to the WFD 5-level scale. Furthermore, Sentinel-2 and Landsat-8 satellites have suitable spatial resolution from 10 to 30 m, which could enable the creation of new lake-based applications to support WFD monitoring requirements [29] outperforming the coarse resolution of MODIS and MERIS [15]. Also according to our analysis, many researchers used a combination of sensors for various reasons, among others, to cover missing time series data or to improve the satellite imagery spatial resolution. The fact that the dominant combination of sensors was Sentinel-2 and Landsat-8 was no coincidence as most researchers wanted to take advantage of the fact that Landsat-8 predated Sentinel-2 and thus could enhance their time series before 2016 (Sentinel-2 launch year), but at the same time to gain better spatial information from Sentinel-2 (10 m) and better revisit time. In addition, sensors such as IKONOS, Pleiades, and SPOT 6/7 were widely used mainly for hydromorphological QEs while the arrival of Wolrdview-2/3 offers high-resolution multispectral satellite imageries thanks to the newly added spectral bands. Furthermore, Worldview-3 is also equipped with the CAVIS (Computer Aided Verification of Information Systems) sensor which measures the atmospheric parameters required for atmospheric correction of Worldview-3 imagery [47].

A new generation of spaceborne hyperspectral sensors, such as PRISMA, GaoFen-5, DESIS, and HISUI, is now available for adding value to the overall monitoring of the water ecosystem, and future missions are also in the development phase. This is in addition to the launch of Hyperion, which was followed a few years later by Chris-PROBA and HICO (e.g., EnMap, CHIME, SBG) [72]. However, at the present, hyperspectral sensors are mostly mounted in airborne systems with limited sensors being in space course, and the use of the sensors in future satellite missions is also under planning [112]. In Figure 11, we summarize the sensors that have been utilized for WFD purposes.

Following the above, it should be stated that apart from the advantages remote sensing technologies can offer, several limitations regarding their use still exist. Cloud coverage is mainly hindering the data collection considering the optical systems and simultaneously the cloud shadows, which may decrease the image information, significantly affecting the overall water body monitoring [44]. In addition, it has already been mentioned that it is difficult to separate water QEs with remote sensing techniques when sediments, dissolved matter, and Chl-a are all present (https://appliedsciences.nasa.gov/sites/default/files/S2P1SDG6.pdf, accessed on 20 April 2022).

Furthermore, the importance of shallow waters is indisputable for various reasons [113] in the framework of the WFD as well as in a more general environmental context. However, remote sensing of shallow waters can be challenging due to a number of factors, including:

• Bottom influence

Bottom influence refers to the way that light interacts with the sea floor in shallow water, which can affect the spectral characteristics of the water column. This can make it difficult to accurately measure water properties such as Chl-a concentration or water depth. The bottom reflectance is not directly observable [114] and substantially affects the upwelling optical signals in shallow coastal areas [115]. According to that, various algorithms for the bottom depth and composition mapping were proposed. Some algorithms are suitable for shallow waters and some others for greater depths [115]. Data collected from EO platforms can be used to infer relative water attenuation coefficients in shallow water, provided that the water is well mixed, and a relatively homogeneous bottom exists over a range of water depths. Many satellite sensors cannot penetrate to depths larger than about 5 m and that still remains a main constraint. Kutser et al. (2020) [113] studied in detail, the above issues and a 50-year retrospective and future directions are given. After all these years of sensors’ improvement, even medium-resolution data, such as Sentinel-2 imagery with 10 m resolution, can map benthic habitats that were not possible before. It has also found that density of microalgae on sand in coral reef lagoons is highly dependable on weather conditions [113,116,117]. Satellites with revisit times <5 days (Sentinel-2, Planet, etc.) with a 2–5 day revisit could evaluate those dynamic processes [118,119].

• Shoreline effect

The shoreline effect refers to the way that water properties can change near the shore due to factors such as waves, tides, and sediment transport. Toure et al. (2019) [120] studied the shoreline detection using optical remote sensing and state that the shoreline detection problem has still not been adequately solved since the algorithms are sensitive to the type of image, and every method is often adapted to a particular application. Tajima et al. (2021) [121] studied the shoreline detection using an ANN and SAR images. The model first classifies the pixels into land and sea and then makes a classification in four layers: an input layer, two hidden layers, and an output layer, where the input layer is based on the pixel values of the SAR image. Further tests using different SAR scenes are necessary for the evaluation of the above methodology, but Tajima et al. (2021) finally suggest that the proposed method for shoreline detection and monitoring has the advantage of a relatively high accuracy and low computational cost.

• Dark waters

Dark waters refer to water with high levels of dissolved organic matter, which can make it difficult to measure water properties using remote sensing techniques. This is because the organic matter can absorb light in the visible and near-infrared regions of the spectrum, which are commonly used in remote sensing. In these situations, the only water-leaving signal detectable by remote sensing sensors occurs at two peaks—near 710 nm and 810 nm [122]. The first peak has been widely used in remote sensing of eutrophic waters for more than two decades. Kutser et al. 2016 [122] showed that the 810 nm is in correlation with those parameters that describe phytoplankton biomass such Chl-a, TSS, and SPOM (suspended particulate organic matter) in most cases. In black lakes, the 700–720 nm wavelength is normally used for the retrieval of Chl-a, but it is still affected by CDOM absorption, and for that case, the 810 nm seems to be very useful. Landsat 8 bands are not very suitable for detecting the two peaks, but Sentinel-2 band 5 (705 nm) is very suitable for mapping phytoplankton biomass (Chl-a) and band 7 (783 nm) can also detect the 810 nm peak in the water reflectance spectrum, which makes Sentinel imagery a very useful tool especially for the case of black lakes [122].

• Spatial and spectral resolution

In shallow waters, high spatial resolution is often needed to accurately map features such as coral reefs or seagrass beds while high spectral resolution is needed to accurately measure water properties such as Chl-a concentration or water depth. Nevertheless, while high spectral resolution can already be available for the study of water quality, the spatial resolution limits the suitability of the imagery, because small-scale spatial heterogeneity is in the focus of investigation [123]. However, in recent years, things have changed, and the newer multispectral satellites such as Sentinel-2A and 2B and hyperspectral sensors can offer advanced opportunities for water quality monitoring. Hyperspectral remote sensing is now an established tool for monitoring purposes usually by inverting a semi-analytical model of water reflectance. Jay et al. (2017) [124] propose a realistic probabilistic model of shallow water reflectance variability based on the semi-analytical model.

• Atmospheric correction

Atmospheric correction (AC) issues refer to the way that atmospheric conditions can affect the signal received by a remote sensing sensor. Especially in coastal areas where the atmosphere is often more variable, this can make it difficult to accurately measure water properties using remote sensing techniques. Accurate AC is essential for making reliable measurements of water properties in shallow waters. In this context, Warren et al. (2019) [125] evaluated six AC algorithms with Sentinel-2 data against in situ measurements, in five different EU countries. They concluded that the red to NIR bands have low performance with high uncertainties which is crucial if we consider that these bands are required to determine QEs such as Chl-a and turbidity. Sandoval et al. (2019) [126] determined that, for Chl-a and Secchi depth monitoring, the most appropriate AC processors to be applied to Sentinel-2 data are Polymer and C2RCC. Pahlevan et al. (2021) [127] attempted a global assessment of AC methods for Landsat-8 and Sentinel-2 over surface water bodies. The study describes the Atmospheric Correction Intercomparison Exercise (ACIX-Aqua), a joint NASA–ESA activity, which was initiated in order to enable an evaluation of eight state-of-the-art AC methodologies available for Landsat-8 and Sentinel-2 imagery. For that reason, reflectance values from 1000 radiometric matchups from both freshwaters (rivers, lakes, reservoirs) and coastal waters were utilized. Significant performance differences between inland and coastal waters were found, but overall, the uncertainties were lower in the coastal environments. Pan et al. (2022) [128] carried out an evaluation of AC algorithms over lakes using high-resolution multispectral imagery. According to this study, there are a lot of factors which can affect the optical imagery such as the complexity of water optical properties, the surface glint, the heterogeneous nature of atmospheric aerosols, or the proximity of bright land surfaces. The study proposed ten AC algorithms which were then evaluated for high-resolution multispectral imagery of Landsat-8 and Sentinel-2 multispectral images. The evaluation was made using in situ optical measurements from almost 300 lakes across Canada. The results of the validation process showed that the performance of the algorithms varied according to each spectral band and evaluation metrics. Windle et al. (2022) [129] tried to evaluate AC algorithms applied to OLCI Sentinel-3 data of Chesapeake Bay Waters. This sensor takes advantage of higher spatial (300 m), spectral (21 bands), and temporal (2-day) resolutions in coastal water bodies than in any other operational ocean color satellite sensor. This study conducted a radiometric evaluation of four different AC algorithms compared to a regional in situ dataset from Chesapeake Bay waters.

In Situ Moving Platforms Dynamics and Limitations

Moving forward, a wide range of innovative in situ solutions are becoming available that enable better utilization of EO capabilities to understand aquatic ecosystems. Based on this review, a host of novel in situ observational platforms such as UAVs and USVs and small-sized multiple (hyperspectral, multispectral) sensors mounted on them have been shown to be rapidly maturing and becoming valuable alternatives to more expensive conventional solutions. Because of their affordability, flexibility, and susceptibility to interference from clouds, UAVs are excellent choices for surface water monitoring, especially in small water bodies [39] while they can contribute very well in shallow water applications [130]. The sensors on board could offer reliable and near-real time information having higher spectral and spatial resolution providing continuous ground feature spectral data. The captured drone data can be inserted in a semi-automated post-processing workflow to create tailored services and maps. A comparative example of the different spatial resolutions is presented in Figure 12. The difference in the spatial resolution of the sensor on board to UAV platform related to the satellite data from Sentinel-2 and Planet is evident. However, strong expertise for handling and navigating such platforms is required, while until today mostly research teams, private sector, and individual institutions applied this solution with the result of the data not being open and being costly as well. In addition, special permission is also required in regions close to airports or to military bases. Furthermore, for cost-affordable UAVs, a challenge remains the ability to lift more weight, having longer flight durations as well as to reach higher altitudes. The above-mentioned also creates the main UAV limitations for their use in large water bodies, keeping the monitoring costs at high levels.

In the same context, USVs have proven to be flexible and efficient in situ solutions moving with relative ease in all water body types as well as to extreme environments. A big advantage is that USVs could carry a variety of different sensors including spectrolysers, physicochemical sensors, active samplers, cameras, etc., offering a multiparameter water quality monitoring which is valuable in the framework of WFD. Mostly they work under telemetry technology and servo-electric mechanisms providing the opportunity to take measurements both at the water surface as well as in the water column profile. Τheir energy autonomy compared to UAVs is much better as they can accommodate batteries with a larger capacity as well as photovoltaic panels to recharge them, thus providing longer mission durations. Moreover, no great expertise is required for their use, but at this point, it should be highlighted that the automated and reliable guidance and navigation for USV remains challenging [14]. As with the UAVs regarding their use from mostly private and institutional sectors, the same is valid also for USVs.

The General Outlook—Synergistic Use of EO Driven Techniques

In general, satellite-based remote sensing methods and images are beneficial because they frequently gather uniform standardized data covering wide areas [14]. In addition, the continuous real-time data and cost-effective services from UAVs and USVs are perfect to feed the purposes for the satellite imagery data calibration and validation. However, non-satellite remote sensing data sources based on in situ measurement or aerial measurement are more expensive and have a limited ability to observe in detail wide areas of lakes and rivers [16]. In this framework and based both on the findings of this review as well as the actual potentials of the EO technologies, we provide an overview of the general dynamics (

Table 6. The general dynamics of the main spaceborne and in situ resources for selected mandatory WFD QEs based on this review. (In moving platforms columns, USV = Unmanned Surface Vehicles and UAV = Unmanned Aerial Vehicles, while in Satellite Sensor System columns, S = Sentinel, L = Landsat, RE = RapidEye, PR = PRISMA, WV = WorldView, MD = MODIS, and ME = MERIS).

● Highly Suited	EARTH OBSERVATION RESOURCES
● Suitable	IN SITU RESOURCES		SPACEBORNE RESOURCES
	Moving Platforms		Satellite Sensor System
Quality Elements	USV	UAV	S-1	S-2	S-3	L-8	RE	PR	WV-3	MD	ME
Biological
Chl-a	●	●	●	●	●	●	●	●	●	●	●
Seagrass			●	●		●			●
SAV	●			●		●	●
Physico-chemical
Turbidity	●	●		●		●	●	●			●
Secchi depth		●		●	●	●	●			●	●
TSS	●	●		●	●	●		●		●	●
TP	●	●		●	●	●			●
TN	●	●		●	●	●			●
CDOM				●	●	●	●	●			●
pH	●	●		●		●
Temperature	●	●		●	●	●			●	●
DO	●	●		●		●		●		●
Hydro/gical
Surf. Water Ext.		●	●	●		●	●		●	●
Bathymetry	●	●	●	●		●		●	●		●

) of the main spaceborne and in situ resources, for selected mandatory WFD QEs. We classified the EO resources based on a binary logic (highly suited, suitable) according to their potential to monitor the different QEs.

The synergistic use of all these EO data tools could provide better monitoring frequencies for all water bodies. The dynamics of UAV and USV to cover cloudy periods can offer valuable missing information in satellite timeseries data. Furthermore, it should be highlighted that the combination of these technologies could increase the number of monitored QEs per water body offering a more representative ecological status of the water systems. Undoubtedly, this increase, at the same time, will cause a noticeable decrease in the percentage of water bodies with unknown status level related to WFD QEs as already presented in Table 1.

The new era of EO technologies provides more and large amounts of free and open data and their handling and storage are still challenging. Specialized technical knowledge and trained human resources are required, as well as advanced infrastructures for storing, debugging, and processing the data with powerful processors and GPUs (Graphical Processing Units). However, referring to country budgets, the investment for the development of such infrastructures is much less than the expenses required for measurements and analysis with conventional techniques and with unbeatable benefits for the WFD requirements and implementation.

The data fusion from different EO technologies is challenging if we consider the different measuring standards and protocols facing the general data harmonization problem. Lack of cooperation and communication between different data holders and users, lack of homogeneous protocols, and dissemination of the data are also still challenges. These obstacles are holding back the application of new technologies to address EU policies such as WFD, especially if we refer to newly entrant countries and countries with limited economic resources.

4. What Is the Key for the Efficient Implementation of WFD?

European nations today depend on monitoring networks, with unprecedented and increasing complexity and several times forming a fragmented in situ component. Vitally, a series of novel monitoring systems have brought insight in the reporting and verification of WFD. In this new reality, if we are to ensure continued sustainment of Europe’s societies and vital natural and industrial ecosystems, comprehensive, yet installation-specific methodologies and tools are essential to allow an efficient and harmonized monitoring framework (common protocols). With collaboration of organizations from throughout Europe and worldwide, highlighted as a crucial step to respond to the recognized challenges with an innovative, user-facing AI-enhanced EO-based monitoring framework, enabling a step-change in RBMPs from all the Member States. Hence, the research community should focus on the development, integration, and utilization of best-in-class technological capabilities to strengthen Europe’s capacity through the detection and monitoring of surface water quality. In that regard, important capabilities and priorities are presented below.

4.1. The Upcoming EO Contribution

While for multispectral imagery data the methods to predict water parameters are already matured substantially, the coupling of these methods to hyperspectral data streams is still largely lacking. Nonetheless, it is widely recognized that imaging spectroscopy can greatly increase the reliability of water-related indicators monitoring. Future studies can focus on the exploitation of PRISMA, DESIS, and EnMAP archives to develop and test new proxies for water WFD-related indicators. Further, improvement of the hyperspectral imagery data will be achieved through the introduction of deep learning-based super-resolution modeling (e.g., SR-GAN) aiming to augment the spatial resolution of hyperspectral data (30 m). This could be achieved by establishing a relationship between the predictor high spatial resolution Sentinel-2 data (10 m) aggregated to low-resolution data pixels and the low-resolution variables that need to be sharpened, and thus allow enhanced monitoring of low-scale features and relevant water bodies. The utilization of multi-infrastructure monitoring is recommended to improve the spatial and temporal resolution by leveraging the emerging technologies of in situ low-cost platforms, as well as the recently advanced spaceborne platforms. For instance, the deployment of UAVs seems to address key tasks (cloud coverage, mapping of small water bodies, etc.). This is imperative to pave the way for the synergistic use of UAVs and spaceborne capacities to optimize advanced sensors (e.g., thermal) and imaging capabilities as new ways for stakeholders to increase the comprehensiveness and credibility of the monitoring process. Moreover, data fusion with unlimited scaling between EO systems: Exploitation of multiple datasets for water bodies monitoring, including EO and terrestrial data is necessary in order to enhance the predictive performance of the model (e.g., active learning) and the spatial and temporal resolution. Furthermore, AI-enabled services and distributed actionable analytics can also be considered to streamline the execution of distributed analytics workflows through semi-automatic orchestration of distributed execution engines and utilization of rich datasets residing in federated computing infrastructures. This will eliminate the hesitation of the relevant authorities to exchange sensitive data, allowing the operation of advanced decision support tools by applying AI to deploy fit for purpose algorithms. All the aforementioned can be achieved following an interdisciplinary approach and by exploiting technological components and results produced by a series of research projects that focus on a variety of aspects related to the WFD concept and objectives. However, the proposed methodologies and tools should be validated in more operational environments, along with the active involvement of the relevant stakeholders. Hence, a framework for better communication and cooperation between the actual actors involved for the WFD implementation can be achieved through a co-design approach in order to reflect and solve their real needs.

4.2. The Citizen Science and IoT Contribution

To address the challenges in water resources monitoring, several novel approaches and methodologies, utilizing diverse yet complementary technological platforms, can be adopted and combined with the EO component. Among others, the potential of ICT (Information and Communication Technology)-enabled citizen observatories have been recognized [131]. Herein, citizen observatories refer to an environment and infrastructure supporting an information environment for WFD-relevant communities and decision makers to discuss, monitor, and intervene in situations, places, and events. Despite their increasing popularity and acclaimed potential, citizen observatories are not “plug and play” solutions or simple technical fixes for citizen-based in situ data collection, stakeholder engagement, or participation in the decision-making process [132]. Kelly-Quinn et al. (2022) [133] proposed a unique framework to leverage the enabling technologies for citizen observatories in order to fill small water body data gaps. This may foster tremendous opportunities considering that a significant number of surface water bodies are not characterized due to their small coverage. Similarly, Hegarty et al. (2021) [134] utilized citizen science to evaluate water quality in river bodies and simultaneously fill the recognized data gaps in an effort to support United Nations Sustainable Development Goal 6 objectives. The need to embed the enabling technologies for citizen observatories into their social dimensions should be highlighted, ensuring a continuous uptake of resulting solutions, always respecting privacy of personal data.

Furthermore, the recent advances, during the last decade, in key technologies for the development of Internet of Things (IoT) have made them a valuable solution for water quality monitoring. The main advantage of IoT sensor networks is that they provide continuous measurements; however, they are able to geographically cover a limited area. This is well recognized by the EO community, using them for calibration/validation of satellite-based data and for covering the gap of data streams between two EO data observations. Wide-scale networks have been put in place for water quality monitoring in rivers [135], lakes [136], and marine environments [137].

5. Conclusions

Without doubt, the WFD gave a great boost to EU countries to improve the ecological status of their water bodies, to identify the driving forces of pressures, and to provide continuous water with high quality to the European community. Nevertheless, as the demand for more and better water quality increases, the requirements of the Directive increase at the same time.

This work summarizes the current efforts of the research community, international organizations, and space operators, to provide services and tools in order to deliver EO-based information related to monitoring and reporting of water quality parameters in support of WFD.

The increasing number of peer reviewed studies (>70%) highlights that spaceborne and in situ EO components have reached an adequate level of maturity. This is continuously driven by a set of advancements in spaceborne sensing systems such as the improvements in spatial, spectral, and temporal resolution. This results in:

(i).

better monitoring frequencies of surface water bodies,

(ii).

a significant increase of the number of monitored water QEs per water body type,

(iii).

a minimization of the percentage of the water bodies with unknown ecological status,

(iv).

a decrease in the overall monitoring cost supporting indirectly the PoMs implementation, and

(v).

the RBMPS from all the Members states being delivered on time.

By increasing the ability to monitor more QEs per water body, the percentage of the unknown status level for each QE will decrease. However, how much this unknown percentage will decrease is very difficult to estimate at this stage. For a reliable answer, further research is required, as it will be necessary to determine precisely what types of water bodies present the highest unknown percentages and why (e.g., size, location, policy etc.), and then to determine and document to what extent the new EO technologies can reduce these percentages and how.

Furthermore, progress is still needed in providing water resources management services that can support informed decision making at various scales. Despite recognition and analysis of novel approaches and tools and well-established practices in the field of EO-driven water quality monitoring, a common methodological and monitoring framework for transboundary river basins considering a common data collection and harmonization protocol is still challenging. In that regard, water quality monitoring for WFD using EO means obviously requires interdisciplinary research in the domains of sensing systems, AI, and data handling. The percentage of AI models in water quality monitoring has also increased the last three years; however, models’ transferability and interoperability of the proposed approaches could be the bridge for an efficient transfer of know-how and a smooth integration of novel techniques from already developed to developing countries into the European community.