Mapping of Flood Impacts Caused by the September 2023 Storm Daniel in Thessaly’s Plain (Greece) with the Use of Remote Sensing Satellite Data

Abstract

Floods caused by extreme weather events critically impact human and natural systems. Remote sensing can be a very useful tool in mapping these impacts. However, processing and analyzing satellite imagery covering extensive periods is computationally intensive and time-consuming, especially when data from different sensors need to be integrated, hampering its operational use. To address this issue, the present study focuses on mapping flooded areas and analyzing the impacts of the 2023 Storm Daniel flood in the Thessaly region (Greece), utilizing Earth Observation and GIS methods. The study uses multiple Sentinel-1, Sentinel-2, and Landsat 8/9 satellite images based on backscatter histogram statistics thresholding for SAR and Modified Normalized Difference Water Index (MNDWI) for multispectral images to delineate the extent of flooded areas triggered by the 2023 Storm Daniel in Thessaly region (Greece). Cloud computing on the Google Earth Engine (GEE) platform is utilized to process satellite image acquisitions and track floodwater evolution dynamics until the complete drainage of the area, making the process significantly faster. The study examines the usability and transferability of the approach to evaluate flood impact through land cover, linear infrastructure, buildings, and population-related geospatial datasets. The results highlight the vital role of the proposed approach of integrating remote sensing and geospatial analysis for effective emergency response, disaster management, and recovery planning.

1. Introduction

Climate change is increasingly affecting the daily lives of people worldwide. In Europe specifically, its impacts are expected to intensify, with projections indicating increased precipitation in some regions and rising temperatures in others. Ecosystems change gradually but in some cases, they experience rapid changes if we consider historical patterns and future outlooks [1,2]. Extreme weather phenomena, such as Storm Daniel, which affected Thessaly’s plain in 2023, are occurring more and more frequently and intensely, such as floods, heatwaves, wildfires, and droughts. These are just some examples of natural disasters that cause environmental disruption, loss of human lives, economic losses, and many other effects. Over the last forty years, human losses from natural disasters in Europe have amounted to 85,000 to 145,000, which, together with other disasters, makes it imperative to have immediate mitigation, precautionary, and efficient protection measures for residents and the environment globally [1,2].

In 2023, Europe experienced above-average, frequent, and prolonged precipitation, which caused significant flooding across the continent. The amount of water in European rivers was significantly high in the last two months of 2023, while it was close to average in the previous ten months [3]. As recorded in the annual European State of the Climate (ESOTC) report by the Copernicus Climate Change Service (C3S) for 2023 [3], “one-third of the European river network saw flooding exceeding the ‘high’ flood threshold (five-year return period), and 16% saw river flows exceeding the ‘severe’ flood threshold (20-year return period)”. More specifically, in early September 2023, Greece was struck by the exceptionally intense storm Daniel, which set a rainfall record and was the most intense the country has ever received [3]. The highest daily amount of rainfall ever documented in Greece was observed in the settlement of Zagora in Thessaly which received 528 mm in 10 h and 754 mm in 21 h on that day. For comparison, the average annual rainfall that the region receives is approximately 750 mm. At the height of the flood phenomenon on 5 September 2023, it was observed that the ‘severe’ flood of 2980 m³/s along the Pineios River far exceeded the 20-year return period [3].

In the 2025 document on global risks from the World Economic Forum, extreme weather events’ immediate and long-term impact is easily discernible [4]. Among immediate global risks, extreme weather events take second place. Similarly, in the ranking based on risk severity for the next two years, they come in second. Meanwhile, in the risks projected over the next ten years, extreme weather events will continue to cause concern as they occupy first place, with the next three risks also being environmental. At the same time, for the next ten years, all stakeholders, including international organizations, governments, the private sector, academia, and civil society, consider extreme weather events the most significant global threat. The same is proven by the Executive Opinion Survey (EOS), which was explicitly conducted in Greece and ranked extreme weather conditions as the second most portentous challenge [4].

Despite the significant technological advances [5,6] and the numerous flood risk management initiatives of the last decades, flooding continues to have devastating effects across Europe [7]. In particular, the Mediterranean region has suffered extreme flooding events that have caused substantial impacts on human life, the environment, property, and infrastructure [8,9]. Moreover, evidence shows that such events have increased in recent decades [10], while the Mediterranean region shows a particular sensitivity to climate change [11].

Particularly, catastrophic floods have a range of impacts beyond the actual flooded area. From direct impacts on property and infrastructure to taking a significant toll on human life, flooding affects a range of socioeconomic activities and processes of the natural environment, prolonging the impacts of such events in months to years to come. Especially given the interconnectedness of modern infrastructure and socioeconomic activities, the effects of catastrophic floods can expand in various sectors, increasing the overall risk. In light of this increase in risk, it is important to assess and understand the impacts of such devastating floods more rapidly and concisely.

Remote sensing has demonstrated its capabilities in mapping and recording flood events [12,13,14] with multispectral optical and synthetic aperture radar (SAR) satellite images that have been used in the past to map flood extent [15]. In SAR systems, which are active ones, microwave energy is headed towards objects angularly, and these objects scatter an amount of this energy. The criteria for scattering depend on many parameters, such as the form and the unevenness of each item [16,17]. SAR sensors, like the one on board the Sentinel-1 satellite, can provide images regardless of the weather conditions and cloud cover prevailing in the area, as opposed to satellite images obtained from multispectral optical satellites [18]. The weather conditions in flooded areas are typically characterized by clouds and intense rain, making the use of SAR data the preferred option for mapping the disaster [19].

Despite the limitations posed by weather conditions and cloud coverage, multispectral optical data, like the Sentinel-2 MSI (Multispectral Imager) and Landsat 8/9 OLI (Operational Land Imager), are also utilized to detect flooded areas [20,21]. Multispectral optical imagery is usually easier to interpret, and different band combinations can highlight distinct structures on the Earth’s surface [21]. The Copernicus Sentinel-1 and Sentinel-2 satellites and the Landsat 8/9 offer frequent and continuous coverage with sufficient spatial and spectral resolution and are openly available, making them valuable tools for flood mapping applications.

There are multiple approaches for differentiating between water and non-water pixels using both SAR and multispectral (optical) satellite data. For the SAR data, the delineation of flooded areas can be performed using different methods, such as the triangle, standard deviation, threshold minimum [22], histogram-based thresholding [23], and change detection [24], while the Otsu-thresholding method, either dynamically adjusted or automatic, is one of the most commonly used ones [22,25,26]. For the optical data, one of the most commonly applied methods is water indices, like the Normalized Difference Water Index (NDWI) [27], and threshold setting to extract affected areas [20]. The Modified Normalized Difference Water Index (MNDWI) [28] has also been successfully used to map flooded areas using scene-adjusted thresholds [29,30]. Indices are a quick and effective method for detecting flooded areas, as no additional data are required for their calculation. However, they may be impacted by different land cover types, shadows, and artificial surfaces [21]. To avoid over- or underestimation of the flooded areas, the abovementioned methods can be combined with Digital Elevation Models (DEM) and their derivatives to produce more accurate results [21]. Sentinel-2 data are also often used to validate the Sentinel-1-derived water masks [23,25]. Sentinel-1 and Sentinel-2 data have also been previously used to study the flooded area in the Thessaly Plain after the medicane “Ianos” in September 2020 [31,32].

In the case of floods and other disasters, the role of geospatial intelligence (GEOINT) is vital. GEOINT integrates GIS (Geographic Information Systems) and remote sensing with all the available geospatial and satellite data for an area of interest (AOI) in order to extract valuable products with the respective techniques used in disaster management, emergency response, and decision-making [33,34,35].

Processing and analyzing satellite imagery over extensive periods is computationally intensive and time-consuming, especially when data from different sensors need to be integrated. These challenges often limit the timely delivery of actionable insights when they are urgently needed during disaster events. The lack of efficient workflows for managing large volumes of multi-temporal and multi-source satellite data hinders the accuracy and scalability of flood monitoring and impact assessment.

This work aims to extensively map the flood extent and the impacts caused by Storm Daniel in Thessaly using ESA (European Space Agency, Paris, France) Copernicus Sentinel-1 and Sentinel-2 and USGS (United States Geological Survey, Reston, VA, USA)/NASA (National Aeronautics and Space Administration, Washington, DC, USA) Landsat 8/9 imagery processed in the Google Earth Engine (GEE) platform to map the flooded areas. The flood mapping extends from the onset of the event to the area’s complete drainage. This study explores the applicability of GEE for efficient processing and analyzing large volumes of multi-temporal, multi-sensor satellite imagery for flood monitoring and impact assessment. Additionally, it focuses on assessing the impact of the flood by combining open geospatial datasets related to land cover, linear infrastructure, buildings, and population. In this context, this work aims, through the case of Storm Daniel in Thessaly, to contribute to the improvement of operational mapping of extensively flooded areas as well as impact assessment by addressing the problem of the high computational demands and time constraints associated with processing and integrating time series of multi-sensor satellite imagery. It demonstrates how GEE can support scalable and rapid flood mapping workflows with the broader goal of improving the operational response to large-scale flood events. To this end, the current study employs processing workflows for each imagery type developed in GEE to manage the large volume of satellite images. These are also usable for similar flood mapping cases and aim to make the process faster and more efficient using widely adopted methods.

2. Storm Daniel in Thessaly

The AOI, shown in Figure 1, is located in the region of Thessaly in Greece (NUTS2). It includes the Regional Units (NUTS3) of Larissa, Karditsa, Trikala, and Magnissia, while a small part belongs to the Regional Unit of Fthiotida of the Central Greece Region. The main cities in the area are Larissa, Volos, Trikala, and Karditsa, while other important towns also exist. Regarding the population, considering the municipalities within the AOI according to the latest 2021 census [36], it is 662,390. This area includes the biggest plain in Greece, the Thessaly Plain, a large agricultural area mainly consisting of permanently irrigated land. This plain is split into two parts (sub-basins), the eastern in Larissa and the western in Trikala-Karditsa. In terms of hydrology (Figure 2), it has high water demands for irrigation, and it belongs to the Pineios River Basin, which is the largest and the most important river in the region that drains into the Aegean Sea. The Pineios River has many major tributary rivers that flow into it at western Thessaly, including, e.g., the Enipeas, Kalentzis, and Lithaios. At the same time, the area has three important reservoirs: Lake Plastiras, Lake Smokovo, and Lake Karla, with the latter being in the southeastern part of the plain. Lastly, the area is a tectonic sink characterized by a flat topography mainly consisting of fluvial and alluvial deposits, and mountains surround it [37,38,39,40].

On 5 September 2023, Storm Daniel struck Greece, marking an unprecedented natural disaster that had a catastrophic impact on the Greek mainland, particularly in the Thessaly region in central Greece. Over a span of four days, the region experienced unprecedented rainfall, exceeding 600 mm in extensive areas [41]—more than twice the annual rainfall typically recorded in Athens. This extreme precipitation led to widespread flooding, landslides, especially around Pelion Mountain, and severe erosion, causing massive destruction to infrastructure, agricultural land, and property, with significant repercussions for the local population and ecosystems [39,42]. The aftermath involved environmental degradation and several cascade events, such as outbreaks of epidemics due to polluted waters, mass death of fish population and critical shortages in water and electricity that lasted for weeks to months [43,44].

The storm’s death toll was also catastrophic, with 17 fatalities, accompanied by approximately 1900 individuals rescued from submerged areas [43]. Hundreds of homes and businesses were flooded, and the region experienced the loss of nearly 100,000 livestock. Moreover, around 730 km² of Thessaly’s agricultural land—a critical hub of Greece’s agricultural output—was inundated by floodwaters [43]. The scale of destruction sent shockwaves through the country, causing significant economic disruption [45]. With agriculture forming the backbone of Thessaly’s economy, the devastation left a void extending beyond local borders, posing challenges to economic recovery and political cohesion.

In late September 2023, Storm Elias struck the Thessaly region, bringing severe flooding and compounding damage from Storm Daniel earlier that month. High-intensity rainfall led rivers to overflow, submerging towns, villages, and farmland. Thessaly, particularly the area of Volos and the Pelion Peninsula, as well as North Euboea Island in Central Greece, faced catastrophic impacts, with homes, infrastructure, and roads heavily damaged or destroyed. The storm left the local population stranded while emergency services conducted widespread rescues. This back-to-back flooding worsened conditions for residents and delayed recovery efforts, heightening the region’s need for immediate aid and long-term support for rebuilding.

The area has been severely affected by floods in the past, with at least 90 flood events in the period 1510–2024 [43,46], some of which were catastrophic, like the events of 1647, 1799, 1811, 1883, 1907, 1955, 1987, 1994, and 2020. The most recent major event occurred in September 2020, triggered by medicane “Ianos”, inducing significant impacts on property and infrastructure in the western part of the Thessaly plain and the surrounding slopes [32,47].

3. Materials and Methods

3.1. Data

In

Table 1. Utilized datasets.

Datasets	Format	Resolution Spatial | Temporal	Source	Purpose/Use
Sentinel-1 imagery 1	Synthetic Aperture Radar GRD IW	10 m and 20 m (Used) | 2023	ESA Copernicus Google Earth Engine	Flood Mapping
Sentinel-2 imagery 1	Multispectral Level-2A	10 m (Used) | 2023	ESA Copernicus Google Earth Engine	Flood Mapping
Landsat 8/9 imagery 1	Multispectral Collection 2 Tier 1 Level-2	30 m (Used) | 2023	NASA/USGS Google Earth Engine	Flood Mapping
SRTM DEM 1 Arc-Sec V3	Raster	30 m | 2000	Earth Explorer Google Earth Engine	Masking—Floodwaters—Study area
Corine Land Cover 2018	Vector (Polygon)	- | 2018	Copernicus Land Monitoring Service	Impact Assessment
Meteorological Data	Text	- | 2023	Meteo.gr of National Observatory of Athens Network	Meteorological Event Data
Hydrology	Vector (Lines, Polygons)	- | 2018 - | 2024 - | 2012	Special Secretariat for Water OpenStreetMap | Geofabrik EU-Hydro	Area of Interest
Infrastructure	Vector (Lines)	- | 2023	OpenStreetMap Geofabrik	Impact Assessment
Building Footprints	Vector (GeoJSON Polygons)	- | 2014−	Microsoft Bing Maps Global ML Buildings Footprints	Impact Assessment
Population Grid	Raster	100 m | 2020	Global Human Settlement Layer (R2023)	Impact Assessment
EMSR692—Flood in Greece	Multiple (Vector, Maps, etc.)	- | 2023	Copernicus Emergency Management Service	Validation
Administrative Boundaries and Population	Vector (Points, Polygons) | Spreadsheets	- | 2021	Hellenic Statistical Authority	Area of Interest

¹ The used satellite image acquisitions are presented in detail in Table 2.

, a brief presentation of the utilized datasets is provided. This study is based on the open and freely available datasets described in the following paragraphs.

The mapping of flooded areas was conducted using optical/multispectral and radar/SAR data from the Sentinel-1, Sentinel-2, and Landsat 8/9 satellites, briefly described below.

Table 2. List of used satellite image acquisitions.

Event	Satellite Image Acquisition Date and Time (UTC +03:00)	Satellite | Type	Result	Number of Images
Pre-Event	22 July 2023 12:10	Landsat 8 C02 T1 L2	Landsat 8 Pre-Event Water Mask	2
	21 August 2023 12:25	Sentinel-2 L2A	S2 Pre-Event Water Mask	4
	25 August 2023 07:40	Sentinel-1 GRDH IW DES	S1 Pre-Event Water Mask	2
	26 August 2023 19:24	Sentinel-1 GRDH IW ASC		2
Event	6 September 2023 07:40	Sentinel-1 GRDH IW DES	Flooded Area 1	2
Post-Event	7 September 2023 19:24	Sentinel-1 GRDH IW ASC	Flooded Area 1	2
	8 September 2023 12:10	Landsat 8 C02 T1 L2	Flooded Area 1	2
	10 September 2023 12:15	Sentinel-2 L2A	Flooded Area 1	7
	12 September 2023 19:32	Sentinel-1 GRDH IW ASC	Flooded Area 2	2
	13 September 2023 07:32	Sentinel-1 GRDH IW DES	Flooded Area 2	2
	16 September 2023 12:10	Landsat 9 C02 T1 L2	Flooded Area 3	2
	18 September 2023 07:40	Sentinel-1 GRDH IW DES	Flooded Area 4	2
	19 September 2023 19:24	Sentinel-1 GRDH IW ASC	Flooded Area 4	2
	24 September 2023 19:32	Sentinel-1 GRDH IW ASC	Flooded Area 5	2
	25 September 2023 07:32	Sentinel-1 GRDH IW DES	Flooded Area 5	2
Post-Event and Storm Elias	30 September 2023 12:17	Sentinel-2 L2A	Flooded Area 6	4
	30 September 2023 07:40	Sentinel-1 GRDH IW DES	Flooded Area 6	2
	1 October 2023 19:24	Sentinel-1 GRDH IW ASC	Flooded Area 6	2
Post-Event	6 October 2023 19:32	Sentinel-1 GRDH IW ASC	Flooded Area 7	2
	7 October 2023 07:32	Sentinel-1 GRDH IW DES	Flooded Area 7	2
	10 October 2023 12:18	Sentinel-2 L2A	Flooded Area 8	4
	15 October 2023 12:20	Sentinel-2 L2A	Flooded Area 9	4
	18 October 2023 19:32	Sentinel-1 GRDH IW ASC	Flooded Area 10	2
	19 October 2023 07:32	Sentinel-1 GRDH IW DES	Flooded Area 10	2
	20 October 2023 12:19	Sentinel-2 L2A	Flooded Area 11	4
	24 October 2023 07:40	Sentinel-1 GRDH IW DES	Flooded Area 12	2
	25 October 2023 19:24	Sentinel-1 GRDH IW ASC	Flooded Area 12	2
	30 October 2023 12:20	Sentinel-2 L2A	Flooded Area 13	4
Total Images		Sentinel-1 GRDH IW		36
		Sentinel-2 L2A		31
		Landsat 8 C02 T1 L2		6
		Total		73

presents the list of all used satellite acquisitions. All the images utilized are open and available directly from the GEE Data Catalog [48].

The study utilized the Sentinel-2 mission’s multispectral satellite images of the ESA Copernicus program. Sentinel-2 satellites are equipped with the MSI sensor and have a 5-day frequency of acquisitions over an AOI. The products used are the Level-2A products of Surface Reflectance (SR) (formerly Bottom-Of-Atmosphere (BOA) reflectance), which are atmospherically corrected and include scene classification. The study utilized the Sentinel-2 mission’s multispectral satellite images of the ESA Copernicus program. Sentinel-2 satellites are equipped with the MSI sensor and have a 5-day frequency of acquisitions over an AOI. The products used are the Level-2A products of Surface Reflectance (SR) (formerly Bottom-Of-Atmosphere (BOA) reflectance), which are atmospherically corrected and include scene classification. Their 10 m resolution in their most commonly used bands, along with the high acquisition frequency, makes them an essential asset for mapping events like floods, and they are also available just a few hours after their acquisition. As they are optical products, their selection depended on the cloud coverage over the study area. As they are optical products, their selection depended on the cloud coverage over the study area [49,50,51].

The ESA’s Copernicus Sentinel-1 mission SAR images and, more specifically, the Level-1 Interferometric Wide-Swath (IW) Ground Range Detected (GRD) products were utilized. The Sentinel-1 mission operates in the microwave area of the 5.405 GHz C-band of dual polarization. GRD products are georeferenced, focused on the ground range projection, and contain only amplitude information. Regarding their resolution, these products have a spatial resolution of 20 m × 22 m with an image resolution of 10 m × 10 m, and they are 5 × 1 multi-looked. At the time of the event, the temporal resolution was 12 days, as it required double the amount of days to capture the Earth since Sentinel-1B stopped operating in December 2021. The GRD IW products the GEE Data Catalog provides are already preprocessed, including the following steps: 1. Thermal noise removal, 2. Radiometric calibration, and 3. Terrain correction. These products are important to flood area mapping due to their all-weather capabilities and fast availability, while the Sentinel-1 mission offered a revisit over an area every 6 days before the loss of the Sentinel-1B satellite [52,53].

The NASA/USGS Landsat Program’s images from the Landsat 8 and 9 satellites were also used in this study, which began acquiring images in 2013 and 2021, respectively. Those satellites are almost identical, carrying the OLI and TIRS (Thermal Infrared Sensor) instruments (Landsat 9 improved OLI-2/TIRS-2), and have a revisit period of 8 days combined (16 days individually). Their images have 11 spectral bands (visible, near-infrared, short-wavelength infrared, and thermal infrared) of 30 m resolution, which were used in this study. Regarding the products that are part of Collection 2 and are Level-2 Tier-1 processed, these are the Landsat science products of the highest geometric and radiometric quality due to their atmospheric and other corrections, containing surface reflectance information [54,55,56,57,58].

The Shuttle Radar Topography Mission (SRTM) DEM 1 Arc-Second Global v3 was obtained freely from the USGS’s Earth Explorer platform and GEE [59,60]. This dataset was utilized for slope masking to correct flood extent results. The SRTM 1 Arc-Second Global has a resolution of 30 m, is void-filled corrected, and offers global coverage [61,62].

Land cover information for the impact assessment of the flooded areas was retrieved from Corine Land Cover 2018 (CLC 2018) in vector polygon format from the vector geodatabase openly available from the Copernicus Land Monitoring Service (CLMS) [63]. The CLC program dates back to 1990, with the most recent available dataset in a six-year update, the CLC 2018, while it retains the same standards across this timespan. The CLC provides sufficient detail, with three levels of detail reaching the 44 land cover classes at Level 3, the most analytical one. It also has an accuracy of ≥85% with a 25-hectare Minimum Mapping Unit (MMU) and a 100 m Mapping Minimum Width (MMW) [63,64,65]. In this study, it is used to map the land cover of the affected flooded areas.

Geospatial vector data regarding the transportation linear infrastructure of the flood-affected area and the streams and rivers were obtained from OpenStreetMap (OSM). OSM is a community-built database that maintains geospatial information contributed around the world, with the data being open and freely available for everyone to use. The OSM data were retrieved from the Geofabrik download server, utilizing the available historical archive for the infrastructure [66,67].

The meteorological data used to understand the extreme meteorological events that caused the floods were retrieved from the National Observatory of Athens Network [68,69], where they are openly available and accessible. Data for the examined precipitation period were obtained from several active meteorological stations of the mentioned network within the study area.

Hydrological data for the area of interest were obtained freely from the Special Secretariat for Water and its geoportal [70]. These vector datasets include rivers, lakes, and river basins derived from the Approved River Basin Management Plans—1st Update for the River Basin District of Thessaly [38]. Also, the EU-Hydro River Network Database 2006–2012 v1.3 was utilized and obtained freely from the Copernicus Land Monitoring Service [71]. This detailed vector dataset includes rivers, streams, and lakes derived from very high-resolution images and a combination of various related datasets, offering an MMU of 1 ha [71,72].

The data related to building footprints were freely downloaded from the Microsoft (Redmond, WA, USA) Bing Maps Global ML Buildings Footprints. This global vector polygon dataset, which has been in production since 2014, is derived using high-resolution satellite images and applying machine learning and deep neural networks techniques. The dataset is characterized by good quality, with a 94.3% precision and 85.9% recall rate in Europe, and only 1.4% of the 5000 building sample was falsely classified as a building [73].

As validation data for this study, the results of Copernicus Emergency Management Service (CEMS) were utilized regarding the EMSR 692—Flood in Greece activation of the service after a request from the authorities. The CEMS component is one of the six Copernicus services, and its role is to provide (a) early warning and monitoring, (b) on-demand mapping, and (c) exposure mapping. On-demand mapping is usually activated under official requests to map and assess the impacts of disasters such as floods and wildfires. In this case, it provided rapid mapping of that emergency, giving near real-time information about the flood event caused by Storm Daniel, utilizing satellite images and geospatial data [74,75].

Lastly, datasets openly available from the Hellenic Statistical Authority—ELSTAT [36] regarding the area of interest were utilized, including the administrative boundaries and the recent population data of the 2021 Census. Additionally, the 2020 population grid from the Copernicus Global Human Settlement Layer [76] (GHS-POP R2023A) was used to assess the population affected by the flood. This 100 m × 100 m raster grid provides the estimated number of people per cell, representing the spatial distribution of the population, as derived from a combination of census and other related datasets [77].

3.2. Platforms and Software

The methodological part of this study utilized both a freely accessible open platform and commercial software.

The GEE platform [78] was primarily employed to access, process, and produce results derived from the multispectral and SAR satellite images. It is a cloud-based platform, developed by Google (Mountain View, CA, USA) enabling geospatial analysis and visualization with high-performance processing capabilities that also provides an extensive data catalog consisting of openly available remotely sensed imagery and geospatial datasets. The platform offers everything users need, from a user interface to data tutorials and examples, while also allowing them to develop their scripts. The platform is free to use for non-profit research and academic purposes, among other purposes. For this study, it proved to be an ideal option due to the large processing volume of satellite image data required [78,79].

The commercial GIS software of ESRI (Redlands, CA, USA) ArcGIS Pro v3.3 was utilized to manage the geospatial datasets, conduct geospatial analysis, and produce maps.

3.3. Methodology

The methods applied for this study’s purposes are described in the following subsections. In the next flowchart, Figure 3, each step followed in the respective methodological part is presented. In the GEE, the scripts used were developed in JavaScript. At the same time, for the data selection and extraction in GEE and GIS, an AOI rectangle was created as geometry and shapefile accordingly containing the affected and the broader area.

3.3.1. Monitoring Planning

The monitoring of the flooded areas was planned according to the study aims and objectives, as well as the availability of satellite images regarding both the acquisition time and the cloud coverage over the study area. The target was to map the maximum flood extent in the most accurate way possible and then to track the area until its complete drainage from floodwaters, excluding the area of Lake Karla reservoir, which drained several months later. That enabled an estimation of the duration of inundation. The time series approach is similar to the case of medicane “Ianos” in 2020, followed by Falaras et al. [31], Psomiadis et al. [15] for the Sperchios River, and Rättich et al. [80]. After conducting searches in GEE and interpreting the results, the time frame for the satellite image search was set as (a) pre-event in July and August 2023, (b) event 6 September 2023, and (c) post-event 7 September to 30 October 2023. For the pre-event basis, images with the least cloud cover percentage and closest to the event were selected to ensure temporal coherence. For the event and post-event basis, from 6 to 10 September 2023, every available Sentinel-1, Sentinel-2, and Landsat 8/9 image was chosen to estimate the maximum flood extent. This decision was made because of Sentinel-1’s different coverage and the clouds’ presence in the Sentinel-2 and Landsat 8/9 scenes. Next, for monitoring the draining of the area, priority was given to the Sentinel-1 and Sentinel-2 images due to their high frequency of revisit and higher resolution compared to Landsat 8/9, which was used only to fill temporal gaps of rejected Sentinel-2 images due to cloud cover. Also, on the post-event draining monitoring basis, the Storm Elias-caused flooded area is mapped at 30 September–1 October 2023. The coverage of Sentinel-1 passes, Ascending (ASC) and Descending (DES), differs. For this reason, their results were combined into one dataset to be comparable and representative of the AOI, considering their 12 and 36 h temporal gaps. In the cases of Flooded Area 1 and Flooded Area 6, the multiple available acquisitions were combined into a single result to give the most representative outcome, considering the process dynamics of the phenomena, Daniel and Elias, respectively, and the existence of clouds in the multispectral images. Initially, more images and dates were available in the image collections created in GEE. However, they were removed from the final results due to poor coverage caused by cloud cover in the multispectral images and a missing image in Sentinel-1. Table 2 presents the complete list of the images that were ultimately used.

3.3.2. Multispectral Data Flood Mapping

The flood event was mapped using multispectral Sentinel-2 L2A and Landsat 8/9 C2 T1 L2 images, as listed in Table 2. A time series of Sentinel-2 and Landsat 8/9 images was selected to track the floodwaters until they completely drained from the area. Cloud cover was an important issue during the image selection, and the related parameters were carefully selected to ensure proper coverage and accuracy.

The flooded area mapping is based on a water spectral index, the MNDWI [28]. It is a spectral index derived from the NDWI [27] and is widely used in water mapping. This index relies on spectral bands, specifically the visible green (green) and the short-wavelength infrared (SWIR) band, instead of NDWI’s near-infrared (NIR), as it is presented in Equation (1). These spectral bands allow us to better distinguish land from other features and water due to the maximized typical reflectance of water in green and the lower typical reflectance of it in SWIR, which, on the other side, is high on terrestrial vegetation and soils. This modification of the NDWI enables better water extraction than the NDWI due to the use of SWIR, resulting in enhanced contrast and fewer false classifications as water buildup and land features [28]. As a normalized index, MNDWI has values ranging from −1 to 1, with the positive values (≥0) representing water and the negative ones representing land. In Sentinel-2, the spectral bands corresponding to green and SWIR are B3 and B11, and in Landsat 8/9, B3 and B6, respectively.

$$\mathrm{M}\mathrm{N}\mathrm{D}\mathrm{W}\mathrm{I}={\displaystyle \frac{\mathrm{G}\mathrm{R}\mathrm{E}\mathrm{E}\mathrm{N}-\mathrm{S}\mathrm{W}\mathrm{I}\mathrm{R}}{\mathrm{G}\mathrm{R}\mathrm{E}\mathrm{E}\mathrm{N}+\mathrm{S}\mathrm{W}\mathrm{I}\mathrm{R}}}$$

(1)

In GEE, separate scripts were developed using the same steps for both Sentinel-2 L2A and Landsat 8/9 C2 T1 L2 images with the proper modifications.

Firstly, the AOI was imported, and then two collections of multispectral images were created based on the AOI, date range, specific tiles, orbit, path, and row, which cover the area, cloud cover percentage, and the needed spectral bands. The first collection refers to cloud-free tiles of the pre-event state (22 July and 21 August 2023), while the second applies to the post-event state covering a period from the first available post-event image acquisition to when the area drained (8 September 2023–30 October 2023). Also, a list of dates and images was created to record the size of the collections. The next step involved the preprocessing of the collections, which included applying scaling factors for Landsat 8/9, applying a cloud mask and subset, and then resampling to 10 m with the bicubic method. After the preprocessing, a visual inspection of the collections was conducted.

The following step is based on MNDWI’s estimation of the pre- and post-event collections with the respective functions. The extraction of water and the flooded areas was achieved with a thresholding that was based on the application of the MNDWI ≥ 0 values corresponding to water as MNDWI’s concept (Equation (2)) [28].

Flooded: MNDWI ≥ 0

(2)

Before the flooded area extraction, the pre-event water mask was created to mask the existing permanent water based on a concept similar to the subtraction of pre-event NDWI minus the post-event NDWI method of difference in NDWI (DNDWI) [81]. Then, the slope mask using SRTM 1 Arc Sec v3 DEM was estimated to retain areas with a slope of less than 10 degrees, which was selected based on the AOI’s topography. That was performed to improve the accuracy of the results and prevent the misclassification of slopes as water due to shadows [21].

In this case, with the proper function, the post-event MNDWI layers were cleared of the permanent water and high-slope areas, thus enhancing the accuracy of tracing the true inundated areas.

The final steps included the application of the threshold to the later MNDWI images, their binarization to flooded and non-flooded areas, and the generation of mosaics for each acquisition date. The final products were created by removing falsely classified pixels, filtering redundant pixels, narrowing the extent, and calculating the flooded area in square kilometers (km²). Then, the extraction for each date was followed by applying the corresponding functions, which were used to extract the flooded area rasters and vectors, MNDWI rasters, and the original natural and false-color images.

3.3.3. SAR Data Flood Mapping

In the flood mapping with SAR data, Sentinel-1 IW GRD images were utilized, as presented in Table 2. The SAR data are usually preferred for use in flooded area mapping, especially when weather conditions do not permit the use of multispectral data due to cloud cover. On a SAR image, water surfaces have low backscatter values and appear darker than other surfaces because they reflect the radiation of the radar away from the source (mirror-like behavior), so it is possible to detect flooded areas. At the same time, caution should be given to avoid errors and misinterpretations led by the rest of smooth surfaces like roofs and roads, as well as the roughness of water bodies during intense weather conditions [16,17].

The methods used for flood area mapping and water delineation rely on the above-mentioned backscatter characteristics of the water surfaces. For this purpose, the most popular methods are the threshold-based image segmentation methods, which aim to differentiate objects from their background by selecting the optimal gray-level value to separate the pixels of interest, in this case, inundated from non-inundated areas. In this study, the Otsu approach [82], one of the most common and efficient thresholding techniques for flood delineation, is mainly applied to SAR images, but it has also been used for multispectral imagery in other studies [25]. After plotting the pixel values in a histogram, the method maximizes the between-class variance of the two histogram segments and minimizes the within-class variance [26]. The optimal threshold is calculated iteratively and applied to each scene. Then, the features are divided into binary classes, in this case, “water (flooded)” and “non-water (not flooded)” [83].

In GEE, two collections of Sentinel-1 GRD IW images were generated according to the defined AOI, date range, orbit pass, and selected polarization. These collections comprised both ASC and DES orbit pass imagery, while the selected polarization was the Vertical-Vertical—VV. This polarization was determined to be appropriate for the analysis after conducting tests and considering prior relevant research in the region [31]. The pre-event collection consists of the ASC and DES images of (25–26 August 2023), while the post-event collection spans the period from the first available image, when the event was taking place, to the last available image, when the area had drained from the floodwaters (6 September 2023–1 November 2023).

The preprocessing of the collections, an important step for the image quality and the accuracy of the results, followed, taking into account that in GEE, the S1 GRD IW images already have accurate orbits and that they are cleared of thermal noise, radiometrically calibrated, and terrain corrected [84]. To this extent, they were speckle-filtered to enhance the image quality and remove the inherent image speckle noise, selecting the focal median method with a 50 m circle window [85,86]. The images were then subset to the extent of the AOI’s that were applied.

The next step included the slope mask application in the same manner as described for the multispectral data. Due to the multiple image coverage per date for each collection, a mosaic was generated for each date. In the case of the pre-event collection, these mosaics of 25 and 26/08 were combined into one mosaic, taking the median of both dates. That was performed to ensure complete coverage of the whole AOI in one pre-event image, combining the ASC and DES orbit passes since one pass does not entirely cover the eastern part and the other the western part.

The image thresholds were calculated and applied in the following steps with the appropriate functions. The Otsu method was used to determine the threshold value for each pre- and post-event image mosaic. However, due to the area’s large rural extent, the Otsu threshold’s accuracy, after several tests, was not optimal, leading to an overestimation of the flooded area extent.

For this reason, the Otsu threshold was adjusted by estimating each image’s backscatter histogram standard deviation (StDev). Standard deviation is a statistical measure that indicates the distance of an observation from the data’s average. The higher the standard deviation, the greater the data dispersal from the mean. Therefore, it can be helpful as a measure of the spread of the data distribution. Once the standard deviation was estimated, the value was subtracted from the Otsu threshold to modify it. After testing, the selection of the proper multiplier for the standard deviation to be subtracted from the Otsu threshold was proved ideal. More analytically, the multiplier input modifies the standard deviation, resulting in the Modified Standard Deviation (MStDev), as presented in Equation (3).

MStDev = StDev × Multiplier

(3)

Then, the MStDev was subtracted from the original Otsu threshold, leading to an adjusted Otsu threshold that generates better results based on the characteristics of each image acquisition (Equation (4)). This semi-manual approach utilizes each image’s statistical measures derived from the Otsu and Standard Deviation and the manual adjustment in terms of multiplier selection, thus improving the accuracy in a less subjective and faster way. The calculated threshold was applied based on Equation (5), creating a binary image of flooded and non-flooded areas. The pre-event water mask was subtracted from the post-event, resulting in the flood extent.

Threshold = Otsu Threshold − MStDev

(4)

Flooded: Backscatter < Threshold

(5)

The final steps include the same process as in the multispectral images with the export of the results and the relevant S1 image rasters.

3.3.4. Impact Assessment, Results Production, and Comparisons

The results of the flooded area derived from the above-mentioned procedure were processed using the GIS software ESRI ArcGIS Pro v3.3. As described in Section 3.3.1, the multiple extracted results were combined through cartographic overlay techniques (e.g., merge, dissolve) into one as the total flooded area, in which four flooded area vector polygons were combined into one. Also, before combining, a visual check was performed comparing with the utilized satellite images, and where necessary, cleaning of falsely classified results took place to improve the accuracy of the results. After these procedures, the final flood extent and the time series results were produced.

Then, the geospatial data for the impact assessment were imported to GIS and processed to be suitable for analysis. With cartographic overlay (intersect), various calculations were performed using the total flood extent vector layer. These steps resulted in the production of impact assessment layers, maps, statistics, and graphs.

It is important to note that the GHS raster layer was converted to vector polygon format for the calculations. The Microsoft Bing ML Global building footprints were edited, excluding footprints with areas less than 15 m², which were filtered out because they mostly represent small structures or are falsely classified as buildings. Also, structures classified as buildings, including greenhouses and solar panels, were removed after visual checking to improve the accuracy of the results. In its final form, this dataset includes the affected houses, buildings, and structures; warehouses; barns; and stables that are considered building footprints. For the linear infrastructure, the polyline vectors were used based on the category and type of the infrastructure. To accurately represent reality, buffers were created to intersect with the flooded area and estimate the affected road and railway length. More analytically, in the case of railways, which are standard gauge (1.435 m) [87] in the AOI, the buffer has an estimated width of 3 m, while for the road network, after advising the Road Works Design Directives of Greece (RWDD) [88] and basemap measurements, the buffers were set: (a) motorways 16 m (8 m per direction), (b) primary 8 m (4 per lane), (c) secondary 7 m (3.5 per lane), and (d) tertiary 6 m (3 m per lane).

The maximum flooded area extent results were compared with the CEMS EMSR 692 results [74,75] based on the date and production version between 6 and 10 September 2023. The respective results of each AOI’s maximum water extent were combined after removing permanent waters. A visual inspection and quantitative analysis were performed after applying a cartographic overlay between the layers to calculate their symmetrical difference.

4. Results

4.1. Flooded Area and Drainage

The total flooded area caused by the storm “Daniel” in Thessaly reached 949.52 km², with the floodwaters inundating a large part of the Thessaly plain on both the eastern and western sub-basins, as illustrated in Figure 4. This flat agricultural area was flooded along the Pineios River and its major tributaries, as well as to the axis from the city of Larissa to Lake Karla reservoir. This area is characterized by high potential flood risk [39,89], and it was predominantly flooded.

The three main concentrations of floodwaters are shown analytically in Figure A1, Figure A2 and Figure A3. The first one, shown in Figure A1, is in Western Thessaly within a triangle of the towns Megala Kalyvia–Farkadona–Palamas, where many rivers, e.g., Enipeas and Kalentzis, drain into the Pineios River, which was almost completely inundated with many settlements, e.g., Metamorphosi and Vlochos being submerged, especially in the axis of Palamas-Farkadona. The Pineios River floodplain in that region was completely flooded, as were the Enipeas and Kalentzis rivers’ floodplains.

The second concentration (Figure A2), in the broader area of the city of Larissa, shows that the northern districts of the city were mainly impacted by the flood, including Giannouli, with the northeastern region of the E-75 motorway to the entrance of the Pineios River being flooded to a large extent, including settlements. A considerable amount of floodwater has accumulated in the eastern part, close to Platykampos and Eleftherio.

The third one (Figure A3) is the area around and to an axis from Lake Karla reservoir to the north. A large agricultural area, important infrastructure, and settlements were flooded to the west of the reservoir. The floodwaters inundated an area that was the original Lake Karla [90].

To further understand the distribution of flooded areas, which were existent across five regional units, as illustrated in

Table 3. Flooded area per regional unit (NUTS 3).

No.	Regional Unit	Flooded Area (km2)	Flooded Area Distribution (%)	Flooded Area to Regional Unit Area (%)
1	Regional Unit of Karditsa	398.37	41.96%	15.11%
2	Regional Unit of Larissa	355.21	37.41%	6.59%
3	Regional Unit of Trikala	153.24	16.14%	4.53%
4	Regional Unit of Magnissia	27.62	2.91%	1.17%
5	Regional Unit of Fthiotida	15.08	1.59%	0.34%
	TOTAL	949.52	100.00%	-

, the most affected are the regional units of Karditsa with 398.37 km² or 41.96% and Larissa with 355.21 km² or 37.41%. It is also important to examine the impact at the municipal level, as presented in

Table 4. Flooded area per municipality.

No.	Municipality	Flooded Area (km2)	Flooded Area Distribution (%)	Flooded Area to Municipality Area (%)
1	M. of Palamas	209.62	22.08%	54.86%
2	M. of Kileler	135.31	14.25%	13.86%
3	M. of Sofades	97.79	10.30%	13.58%
4	M. of Farkadona	94.43	9.94%	25.62%
5	M. of Larissa	81.77	8.61%	24.32%
6	M. of Karditsa	54.01	5.69%	8.31%
7	M. of Trikala	53.44	5.63%	8.77%
8	M. of Aghia	42.12	4.44%	6.34%
9	M. of Farsala	39.59	4.17%	5.35%
10	M. of Mouzaki	36.95	3.89%	11.80%
11	M. of Tempi	31.60	3.33%	5.48%
12	M. of Rigas Fereos	26.85	2.83%	4.88%
13	M. of Tyrnavos	24.74	2.61%	4.71%
14	M. of Domokos	15.08	1.59%	2.13%
15	M. of Pyli	3.41	0.36%	0.45%
16	M. of Meteora	1.97	0.21%	0.12%
17	M. of Almyros	0.40	0.04%	0.04%
18	M. of Volos	0.36	0.04%	0.09%
19	M. of Elassona	0.08	0.01%	0.00%
	TOTAL	949.52	100.00%	-

and Figure 5. A total of nineteen municipalities were affected, with the municipality of Palamas being heavily struck by the floodwaters because the Pineios, Kalentzis, and Enipeas rivers are within the municipality’s boundaries, resulting in 209.62 km² and over half of its total area (54.86%) being flooded. The municipality of Kileler is the second most affected (135.31 km²) and is located in the area of Lake Karla reservoir, which was extensively flooded. The municipalities of Sofades and Farkadona follow, with 94–97 km² flooded, being significantly affected due to the Enipeas and Pineios rivers accordingly.

The evolution of the flooded area was examined, and the results are presented in

Table 5. Flooded area per acquisition dates and changes.

No.	Satellite Image Acquisition Date	Satellite	Flooded Area (km2)	Change (%)
1	6–7–8–10 September 2023	Sentinel-1, Sentinel-2, and Landsat 8	949.52	-
2	12–13 September 2023	Sentinel-1	262.21	−72.38%
3	16 September 2023	Landsat 9	244.35	−6.81%
4	18–19 September 2023	Sentinel-1	77.30	−68.37%
5	24–25 September 2023	Sentinel-1	58.35	−24.51%
6	30 September–1 October 2023	Sentinel-1 and Sentinel-2	257.77	341.78%
7	6–7 October 2023	Sentinel-1	126.32	−50.99%
8	10 October 2023	Sentinel-2	132.88	5.19%
9	15 October 2023	Sentinel-2	129.87	−2.26%
10	18–19 October 2023	Sentinel-1	113.65	−12.49%
11	20 October 2023	Sentinel-2	119.64	5.27%
12	24–25 October 2023	Sentinel-1	105.87	−11.51%
13	30 October 2023	Sentinel-2	117.70	11.17%

and Figure 6 and Figure 7, while detailed maps for each date are displayed in Figure A4, Figure A5, Figure A6 and Figure A7. Starting with the quantitative analysis of the flood duration, which lasted from 5 September 2023 to 30 October 2023, excluding, as mentioned in the previous section, the case of the region of Lake Karla reservoir, in which the floodwaters drained after an extended period. From the initial flood extent, significant drainage took place, reducing the inundated area by 72.38% in 12–13 September. That reduction, which reached almost 94% of the initial flooded area, lasted until 30 September 2023, when Storm Elias struck. This storm increased the flooded areas to 257.77 km² or 27.14% of the initial flood extent, and then a reduction to half (51%) followed at 6–7 October. Then, slight decreases followed, including slight increases. Those differences are caused by the different characteristics of the Sentinel-1 and Sentinel-2 images, in which the Sentinel-2 attributes a larger water area than Sentinel-1 images, as the study of Falaras [91] confirms on parallel multitemporal monitoring. A primary difference is also spotted on the riverbanks where Sentinel-1 fails to capture overflowed water and the remaining water. On 30 October, the flooded area was 117.70 km², which corresponds to the area of Lake Karla reservoir’s remaining floodwaters.

Spatially, the main floodwater concentrations that persisted were in the three areas mentioned above, specifically between Palamas and Farkadona, the north of Larissa, and in the eastern Thessaly plain strip in the Lake Karla reservoir area. On 12 September 2023, when an embankment on the Pineios River in the area of Gyrtoni failed, floodwaters drained from the western basin, flowed southeast following the streams, and inundated the areas around Lake Karla reservoir, increasing the water levels in the region [43]. Storm Elias caused flooding, mainly affecting the western Thessaly plain at the region of Megala Kalyvia–Farkadona–Palamas–Sofades, while at the same time, the eastern part, mainly the area of Lake Karla reservoir, was filled with more water.

4.2. Impact Assessment

4.2.1. Land Cover

Based on CLC 2018 and as presented in Figure 8 and Figure 9 and

Table 6. Affected land cover based on Corine Land Cover 2018.

CLC Category	Flooded Area (km2)	Flooded Area (%)
111 Continuous urban fabric	0.003	0.00%
112 Discontinuous urban fabric	13.270	1.40%
121 Industrial or commercial units	4.342	0.46%
122 Road and rail networks and associated land	1.315	0.14%
124 Airports	2.335	0.25%
131 Mineral extraction sites	0.436	0.05%
133 Construction sites	0.905	0.10%
141 Green urban areas	0.613	0.06%
142 Sport and leisure facilities	0.019	0.00%
211 Non-irrigated arable land	50.366	5.30%
212 Permanently irrigated land	794.152	83.64%
221 Vineyards	0.275	0.03%
222 Fruit trees and berry plantations	0.682	0.07%
223 Olive groves	0.117	0.01%
231 Pastures	29.048	3.06%
242 Complex cultivation patterns	11.227	1.18%
243 Land principally occupied by agriculture, with significant areas of natural vegetation	1.366	0.14%
311 Broad-leaved forest	1.050	0.11%
312 Coniferous forest	0.000	0.00%
313 Mixed forest	0.001	0.00%
321 Natural grasslands	1.707	0.18%
323 Sclerophyllous vegetation	0.633	0.07%
324 Transitional woodland-shrub	0.721	0.08%
331 Beaches, dunes, sands	2.512	0.26%
333 Sparsely vegetated areas	0.043	0.00%
411 Inland marshes	5.216	0.55%
421 Salt marshes	1.060	0.11%
511 Watercourses	22.194	2.34%
512 Water bodies	3.887	0.41%
523 Sea and ocean	0.012	0.00%
TOTAL	949.52	100.00%

, the affected land cover from the flood shows that agricultural land, irrigated and non-irrigated, was predominantly impacted, reaching 88.94%, or 844.51 km². When considering only the second class of agricultural areas, the impact reaches 93.44%, or 887.23 km². The impact on settlements is evident by the 13.27 km² of discontinuous urban fabric that was inundated. In addition, a large area of important transportation infrastructure, as well as industrial and commercial facilities, was flooded, reaching almost 8 km². Regarding the watercourses, which take up 2.94% of the flooded area, it is evident that the rivers overflowed, flooding their neighboring areas.

4.2.2. Buildings

The affected buildings, as estimated using the Microsoft Bing Maps Global ML Building Footprints dataset, are presented in

Table 7. Estimated affected buildings based on the Microsoft Bing Maps Global ML Building Footprints.

No.	Municipality	Affected Building Footprint Area (m2)	Number of Affected Building Footprints
1	M. of Palamas	761,124.74	5541
2	M. of Larissa	621,797.46	2333
3	M. of Kileler	497,034.62	915
4	M. of Farkadona	375,235.72	2349
5	M. of Trikala	374,791.13	2085
6	M. of Mouzaki	228,248.44	1448
7	M. of Karditsa	95,687.62	565
8	M. of Pyli	92,671.39	57
9	M. of Tyrnavos	88,671.44	323
10	M. of Sofades	88,540.48	265
11	M. of Volos	78,301.26	78
12	M. of Rigas Fereos	70,617.37	173
13	M. of Tempi	51,108.65	236
14	M. of Farsala	46,639.60	161
15	M. of Aghia	20,965.91	137
16	M. of Elassona	4755.74	2
17	M. of Domokos	2512.53	33
18	M. of Meteora	600.29	6
19	M. of Almyros	-	-
	TOTAL	3,499,304.40	16,707

and Figure 10. The results indicate that 16,707 buildings or structures were affected or potentially affected by the flood, covering an area of 3,499,304 m². The three most affected municipalities in terms of building area are Palamas, Larissa, and Killeler, and many buildings in Palamas, Larissa, and Farkadona. More precisely, in the municipality of Palamas, an important number of settlements were inundated, leading to 5541 buildings being impacted, which totals an area of 761,124 m², while in the neighboring Farkadona, 2349 buildings with an area of 375,235 m² were influenced. In the Municipality of Larissa, 2333 buildings were struck, especially in the northern districts, equal to a 621,797 m² building area. Lastly, Kileler has an affected number of 915 buildings, and their area is considerable, reaching 497,034 m². As previously mentioned, the building footprints include residential buildings and other facilities, such as industrial buildings, warehouses, and barns.

4.2.3. Linear Infrastructure

The affected or potentially affected linear infrastructure, as estimated using the OpenStreetMap dataset, is presented in

Table 8. Affected linear infrastructure based on the OpenStreetMap datasets.

Linear Infrastructure	Affected Length (km)	Distribution (%)
Motorway	30.155	6.71%
Primary	49.566	11.04%
Secondary	92.313	20.55%
Tertiary	260.125	57.91%
Railway	17.002	3.79%
Total	449.161	100.00%

,

Table 9. Affected linear infrastructure based on the OpenStreetMap datasets per municipality and type.

No.	Municipality	Motorway (Km)	Primary (Km)	Secondary (Km)	Tertiary (Km)	Railway (Km)	TOTAL (Km)
1	M. of Palamas	10.73	0.65	45.92	64.86	0.70	122.86
2	M. of Larissa	7.03	10.80	1.46	36.79	5.90	61.98
3	M. of Farkadona	-	19.41	6.13	35.64	-	61.18
4	M. of Kileler	6.82	3.67	9.74	26.98	1.39	48.59
5	M. of Trikala	2.03	1.60	6.24	19.95	0.75	30.58
6	M. of Sofades	0.15	0.56	4.60	20.56	2.28	28.15
7	M. of Tempi	2.04	1.66	3.79	9.40	3.45	20.34
8	M. of Mouzaki	0.20	7.48	2.60	9.63	0.38	20.29
9	M. of Karditsa	1.05	2.50	0.91	11.70	0.44	16.61
10	M. of Farsala	-	0.18	1.83	11.34	1.09	14.45
11	M. of Aghia	0.08	-	1.76	7.73	-	9.58
12	M. of Tyrnavos	-	0.01	6.24	0.03	-	6.28
13	M. of Domokos	-	0.14	-	3.91	0.62	4.67
14	M. of Rigas Fereos	0.02	0.16	0.97	0.15	-	1.30
15	M. of Volos	-	0.72	0.13	0.31	-	1.15
16	M. of Pyli	-	-	-	0.97	-	0.97
17	M. of Meteora	-	0.03	-	0.15	-	0.18
	TOTAL	30.16	49.57	92.31	260.12	17.00	449.16

and

Table 10. Affected road and railway bridges based on the OpenStreetMap datasets per municipality and type.

No.	Municipality	Number of Affected Bridges	Affected Bridges Distribution (%)
1	M. of Palamas	23	16.20%
2	M. of Farkadona	20	14.08%
3	M. of Larissa	17	11.97%
4	M. of Tempi	14	9.86%
5	M. of Trikala	14	9.86%
6	M. of Kileler	12	8.45%
7	M. of Sofades	10	7.04%
8	M. of Mouzaki	6	4.23%
9	M. of Karditsa	5	3.52%
10	M. of Tyrnavos	5	3.52%
11	M. of Farsala	4	2.82%
12	M. of Aghia	3	2.11%
13	M. of Domokos	3	2.11%
14	M. of Meteora	2	1.41%
15	M. of Pyli	2	1.41%
16	M. of Volos	2	1.41%
	TOTAL	142	100.00%

, as well as Figure 11 and Figure 12. A total of 449.16 km of road and rail network was affected, with the tertiary having the most affected length out of the other categories, totaling 260.12 km. However, the critical length of the primary and secondary road network that connects major towns and cities was 49.56 km and 92.31 km, respectively. Regarding the railway, with an estimated 17 km and a reported longer damaged length, the resolution, deposits, and time of acquisition may result in that difference [43].

Considering the affected linear infrastructure per municipality, as shown in Table 9, the most affected linear infrastructure exists in the most inundated areas where most of the waters are spatially concentrated. Considering this, the Palamas, Larisa, Farkadona, and Kileler Municipalities have the most impacted linear infrastructure. In the Municipality of Palamas, the impacted length is two times longer (122.86 km) than in Larissa (61.98 km), but in the second region, more primary roads and railways were affected. Also, it is evident that the tertiary network, which is of lower standards, is at greater risk than the higher categories.

Proceeding to the affected bridges (Table 10), the number affected or potentially affected, based on the OpenStreetMap, is 142 bridges, which refer only to the road and railway ones. Based on the literature, important pedestrian bridges [43,92] were severely damaged in the area, though the analysis here examines the road and rail network. Consulting the literature, the results of the affected bridges include pinpointed by inspections heavily damaged or collapsed, such as in Ampelia, close to the town of Farsala, and in Palaiopyrgos, close to the Pineios River delta. Spatially, the bridges intersect with the rivers and streams, which significantly overflowed.

4.2.4. Population

The affected population (

Table 11. Estimated affected population per municipality according to the GHS POP 2020 dataset.

No.	Municipality	Affected Population	Affected Population Distribution (%)
1	M. of Larissa	11,526	27.108%
2	M. of Palamas	8475	19.931%
3	M. of Trikala	5308	12.483%
4	M. of Farkadona	3739	8.793%
5	M. of Karditsa	2439	5.736%
6	M. of Kileler	2310	5.432%
7	M. of Mouzaki	2122	4.990%
8	M. of Volos	1490	3.505%
9	M. of Sofades	1266	2.977%
10	M. of Farsala	822	1.934%
11	M. of Rigas Fereos	813	1.912%
12	M. of Tyrnavos	581	1.366%
13	M. of Aghia	481	1.131%
14	M. of Tempi	455	1.069%
15	M. of Pyli	437	1.028%
16	M. of Domokos	199	0.469%
17	M. of Meteora	32	0.074%
18	M. of Elassona	26	0.062%
19	M. of Almyros	1	0.002%
20	TOTAL	42,520	100.000%

, Figure 13), according to the GHS POP 2020 dataset of 100 m resolution, reveals that 42,520 people were either directly or possibly affected. This number makes evident the massive impact of such extreme events on the population, which, in the case of Storm Daniel, was even greater, considering that other areas not studied in this work were also affected.

As expected, in the most populated municipality of Larissa, 27.10%, or 12,526, of the affected population is located, especially in the northern districts and suburbs of the city. Within the municipality of Palamas, the next almost 20% (8475) of the estimated population is located, which was also expected since over 50% of the municipality’s area was inundated, including many settlements with parts of the town of Palamas also affected. Within the municipality of Trikala, 5308 people were affected, corresponding to 12.48% of the distribution, primarily located in the town of Megala Kalyvia.

4.3. Comparative Analysis

The quantitative comparison with the cartographic overlay between the flood extent of this study and the CEMS results [75] showed that the results are identical. The primary differences are due to the different image acquisitions regarding dates, image type and resolution, and the defined AOI that differentiates the northern and northwestern parts.

The calculated flooded area, taking into consideration the results of maximum water extent from 6 September to 10 September 2023 from the CEMS dataset, is 895.22 km². That represents a difference of 5.72% from the study’s maximum flood extent. As shown on the following map, 121.90 km² of the study’s results and 67.60 km² of the CEMS results differ after their spatial comparison. More analytically, in Figure 14, the cyan color on the map represents the overlapping flooded area between the two flood extents, while the orange and green areas indicate where only one or the other layer exists, as calculated by their symmetrical difference.

5. Discussion

This study employed remote sensing satellite data to map and analyze the extent of the impact of the flooding caused by Storm Daniel in Thessaly’s agricultural plain in Greece. By utilizing Sentinel-1, Sentinel-2, and Landsat 8/9 imagery processed through the GEE, we provided a detailed temporal and spatial assessment of flood extent and drainage dynamics, and we demonstrated the capabilities of the approach to identify an important part of the direct impacts of the flood. The results illustrate the significant impact of the flood event, affecting extensive agricultural areas, settlements, and infrastructure. Furthermore, the study compared its findings with those from the CEMS to validate the derived flood maps. Using geospatial datasets to assess the impacts helped to gain geospatial intelligence over the large-scale flood events caused by extreme weather phenomena. The study’s highlight is the use of cloud computing in extracting flooded areas and their duration, using widely applied methods and multiple types of satellite imagery, and then integrating geospatial data to gain a holistic insight into the impacts of flood events.

The primary contribution of this study lies in its detailed and high-resolution flood mapping approach, which provides an extensive spatial and temporal assessment of the flooding caused by Storm Daniel. By integrating multispectral and SAR satellite data, the study ensures a more accurate depiction of the flood extent, effectively addressing challenges related to cloud cover that often hinder multispectral sensors. The methodology allows for a rapid assessment of the disaster, which is crucial for emergency response, resource allocation, and post-flood recovery planning. Moreover, utilizing the flood mapping results combined with geospatial datasets, the study offers a quantitative analysis of the impact on land cover, infrastructure, and population, aiding policymakers and disaster management agencies in understanding the full scope of flood damage.

This research presents several novel aspects. Firstly, it applies a comprehensive multi-sensor approach, leveraging multispectral and SAR data, to generate a more robust and validated flood map. Secondly, the study employs an optimized Otsu-based thresholding method with standard deviation modification, enhancing the accuracy of SAR-based flood delineation. Harnessing cloud computing capabilities within GEE further streamlines the workflow, demonstrating a scalable and efficient approach for flood monitoring utilizing well-known methods. Additionally, the study tracks the floodwaters over an extended period, providing a time series analysis of flood dynamics that highlights the persistence of floodwaters in specific areas. Lastly, comparing the results with CEMS data adds a validation component, ensuring the reliability of the results.

In this work, unlike studies by Johary et al. [24] and Moharrami et al. [26], satellite images from various multispectral optical and radar satellites were used to ensure a comprehensive and accurate depiction of the flooded area, regardless of cloud cover or the lack of timely image acquisition from a specific satellite. In addition, the impact of the flood was studied not only on buildings and the population, as Eudaric et al. [22] did, but also on other infrastructures, such as bridges, the railway network, and the road network by type, as well as on land use covers, analyzed per municipality across the affected area. This study applies the Otsu threshold method and the MStDev, similar to Tran et al. [25], in contrast to other studies, such as those by Eudaric et al. [22], Moharrami et al. [26], Samela et al. [30], and Kulk et al. [29], that used additional or alternative methods for broader result comparisons. Another effective method for flood mapping is the examination of change detection in the area, as utilized by Johary et al. [24] and UN-SPIDER [93]. However, it was not preferred in the present study. At the same time, data processing and results extraction were implemented in GEE to reduce processing time and data volume compared to Tran et al. [25] and Samela et al. [30], who used the SNAP software. Additionally, this study used a standard MNDWI value to automate processing, such as Caballero et al. [20], though the latter study differed because it defined a 0.1 threshold value. Furthermore, the current study employed a manual approach for thresholding, similar to the method used by Samela et al. [30], in contrast to the histogram analysis and varying thresholds applied by Kulk et al. [29]. The satellite image time series used in this study provided a more accurate depiction of the flooded area, similar to Caballero et al. [20] and Tran et al. [25], whereas Eudaric et al. [22] chose to explore the Storm Daniel phenomenon in Thessaly by single-image acquisitions. The flood duration was explored using multi-sensor data also in other studied areas, speaking of studies like those of Rättich et al. [80] and Psomiadis et al. [15], while O’Hara et al. [94] examined inundation duration by using only radar images.

Storm Daniel in September 2023 was one of the most severe extreme weather events that struck Greece and affected the region of Thessaly the most. This study focused on estimating the effects of this phenomenon caused by the subsequent flooding in Thessaly’s plain, which reached a maximum extent of 949.52 km², as estimated using multi-sensor satellite imagery. The floodwater drainage was slow, and the area drained completely almost two months from the event’s onset after the flooding rebounded from Storm Elias in late September 2023. The waters persisted in some areas along the Pineios River and the Lake Karla reservoir region, with the latter being drained in the summer of 2024. The regional units of Karditsa and Larissa were predominantly flooded, with the most inundated municipality being Palamas, containing over one-fifth of the floodwaters.

The analysis results show that the whole inundated area is an area of high flood risk after consultation of the flood management plans [89]. It is critical that both sub-basins of Thessaly were flooded to a large extent, while during previous events, only one of the two was mainly affected, e.g., in the case of medicane “Ianos” in 2020, the western part of Thessaly’s basin was flooded. Also, the flood duration analysis pinpointed the areas with large floodwater concentrations that are not easily drained, and their soils become considerably saturated. The area between the towns of Palamas and Farkadona is such a case where many major Pineios River tributaries flow. The area’s rivers could not handle the large floodwater volume, leading to critical overflow with the breaking of embankments and water gates, thus flooding areas around them. The analysis of the satellite images made clear that most of the rivers, due to the summer-dry season, had their lowest flow, and for this reason, a large volume of water was considered floodwater between their banks.

The inundated areas were almost entirely agricultural land according to the CLC 2018 dataset. During the part of the year when crop harvesting was taking place, agricultural production losses were severe for the local population, as the area’s economic pillar was damaged, and major crops, like cotton and corn, were lost. Regarding the infrastructure, many buildings and structures were struck and potentially damaged, with an estimated area of about 3.5 million square meters. Most are located within settlements, while others include industrial areas, barns, and warehouses. Many settlements were almost entirely submerged, with the building stock being irreversibly damaged. This impact was significant on the building infrastructure of the region, affecting the lives of many people who lost their residences and belongings, considering that most of the buildings are within residential areas. However, higher-resolution and more detailed datasets, in terms of building usage, could have helped more accurately record the impacts. Proceeding to the linear infrastructure, the road network, on which the region’s transportation is dependent, was severely affected, with key roads being damaged and connections being disrupted, including the E-75 national motorway that connects Athens and Thessaloniki, cutting Greece’s transportation in half, as well as many local roads, which provide regional and local transportation of people and goods, insulating settlements and towns. Also, the railway network, which includes parts of the central axis that connects Athens and Thessaloniki, was impacted. Even though, in some cases, due to image resolution and the characteristics of mud and deposits transferred by floodwaters on the linear infrastructure that were not characterized as flooded, the effects were severe, leading to erosion, displacement of rail tracks and roads, and the damage of multiple associated infrastructure. Regarding the bridges, especially those intersected with the overflowed rivers and streams, some of them collapsed, and others have irreversible damage, while the rest need monitoring for their safety [43].

Considering the population, over 42,520 people were affected by the floods after the estimation using the flood extent and the GHS POP 2020 dataset. However, the impact of the flood was greater because it affected a region of over 600,000 residents, as it paralyzed almost every sector. Also, there were multiple casualties, while thousands of livestock animals died [43].

The study’s results offered sufficient flood extent mapping despite the difficulties with the multispectral imagery affected by cloud cover, leading to the exclusion of acquisitions. Compared to the results of CEMS [75], which used not only the same but also different sensors of higher resolution and different methods, the estimated deviation of 5.72% in the maximum flood extent with the differences spatially being a result of differently defined AOIs, satellite images, and methods used.

Despite its contributions, this study has certain limitations. Although sufficient for large-scale mapping, the resolution of Sentinel-1, Sentinel-2, and Landsat 8/9 may not capture finer details in highly urbanized or complex landscapes. For instance, flood mapping of settlements, buildings, and linear infrastructure requires satellite imagery of less than 10 m resolution to be more accurate, as proved by consulting reports based on in situ data [43]. Temporal gaps, as well as the weather conditions of the AOI for multispectral images, can lead to the loss of critical information regarding the floodwater dynamics and a loss in mapping the actual flood extent and its impacts. Furthermore, the floodwaters have different spectral and backscattering characteristics compared to most water surfaces, and both multispectral and SAR-based flood detection may be affected, resulting in potential misclassifications. The reliance on open-source geospatial data, such as OpenStreetMap, for infrastructure assessment introduces uncertainties in the accuracy and completeness of the dataset. Official and detailed datasets, open and available by the authorities, would have improved the accuracy of the impact assessment. However, the AOI’s large extent and rural characteristics were challenging not only in the satellite data volume but also in the flood extent extraction despite the use of GEE, which made the application of methods like Otsu needing modifications to produce better results due to overestimation of flooded areas. Also, the geospatial data volume was large and more challenging to inspect, validate, and process. Additionally, the lack of extensive ground truth validation poses a limitation in confirming the precise flood extents, as field data collection was not possible. Finally, the study does not incorporate hydrodynamic modeling, which could offer further insights into flood propagation mechanisms along with the thorough temporal analysis that it provides.

Based on this study, several key observations can be made regarding improvements and potential future work. A major step forward in the future evolution is performing the complete processing on Google Earth Engine and producing the results only with cloud computing in a web app. That will make the processing time and resource usage more efficient, enabling the delivery of results in near real-time and being available within a web app environment. In this step, integrating other contemporary methodological approaches, such as machine learning for floodwater extraction and flood depth calculation, could lead to more accurate and objective results, with faster processing following training on various flood scenarios. Additionally, utilizing satellite images of higher resolution that are available in a timely manner can enhance the accuracy of the results. Another step is to compare methods and improve the utilized approaches, leading to enhanced accuracy and more extensive validation. For the validation part, near-real-time collaboration with emergency responders, civil protection authorities, and scientific personnel can give us helpful information for the impact assessment and validation of the performed analysis. A future step is using more parameters and geospatial datasets in the flood impact assessment. Also, expanding the study to incorporate socio-economic impact assessments would provide a more comprehensive understanding of flood consequences and help develop resilience-building strategies for affected communities. In addition, updating and correlating the flood risk plans and the related modeling can also contribute significantly to managing extreme flood events. Finally, implementing this approach to other future flood events can also accommodate improvements and contribute significantly to the stages from monitoring a flood event to the response, impact assessment, and planning.

From the above analysis, it is evident that some areas suffered more than others and that the Municipality of Palamas is the most affected, as is evident by the results. The analysis highlighted that many affected areas need urgent attention to recover and take the necessary measures to protect them from a potential future flood. The zone in the Pineios River floodplain is an area that needs better management, as well as the major tributaries, with the required works, including embankments, water gates, cleaning of the banks, and drainage canals that need to be made or improved. Especially the areas flooded in western Thessaly are almost the same as those flooded three years before Storm Daniel by the medicane “Ianos” with heavy impact on the above-mentioned triangle of towns Megala Kalyvia–Farkadona–Palamas. That is an evident example of the area’s vulnerability, which is also confirmed by the October 1994 flood event [32,43].

The scale of the disaster exposed the limitations of the country’s existing infrastructure, preparedness mechanisms, and response systems and revealed the inadequacies in forecasting and planning for extreme weather events. This event is a stark reminder of the escalating severity of climate-induced disasters and their ability to reshape entire regions. In particular, it underscored how the convergence of environmental, economic, and social vulnerabilities can transform a natural event into a multidimensional crisis with far-reaching impacts on the nation’s stability. The unprecedented nature of Storm Daniel has challenged conventional understandings of climate risks, making it clear that extreme weather events pose existential threats even to regions considered well-equipped to manage them.

6. Conclusions

The study achieved the aim of mapping the flooding caused by Storm Daniel in September 2023 and assessing the impact in the Thessalian Plain, utilizing a combination of remote sensing with cloud computing and GIS techniques, with publicly available satellite images and geospatial datasets contributing valuable knowledge regarding the impacts. The results, which include the maximum flood extent, the evolution of floodwaters, and the multiple geospatial data that cover different impact aspects, offered added intelligence on flood impact. In the case of Thessaly, this study’s results showed that Storm Daniel led to the inundation of a large agricultural plain, resulting in severe and wide-ranging consequences.

The approach presented in this study can contribute significantly to the mapping and assessment of flood impacts and can be replicated and modified accordingly for every flood event. The use of cloud computing in GEE, in particular, makes the processing of multiple and multi-sensor satellite images efficient and usable for any flood case. Additionally, by improving the processing speed, this approach can help provide timely available near-real-time flood inundation maps and track flood evolution. Combining the flood extent with geospatial datasets helps gain geospatial intelligence over a flood event. All this information can contribute significantly to assessing, managing, and mitigating the impact, as well as planning and decision-making regarding restoration and preparation for possible future extreme events. Another aspect is that in flood-prone areas like Thessaly’s Plain, the applied approach enables the efficient mapping of past flood events and their evolution. More specifically, past and current flood events can be mapped depending on the image availability, thus producing a comprehensive flood impact archive. That archive can be an invaluable tool in pinpointing the repetitiveness of the flood phenomena in the area and can be used to produce the respective flood risk models. Also, areas highly vulnerable to floods can be traced, helping to take the proper measures usable in flood risk management and policy-making.