2.1. Objectives of the Study

A part of the HyDEPP-SATREPS project objectives is to assess flood risk in the context of future climate change. This paper aims to examine the use of freely available satellite data and the GEE platform for early flood detection and information sharing under the constraints imposed by the COVID-19 pandemic, which restricted field surveys. During the project period, we attempted to visualize flood damage caused by typhoons by constructing a web platform to share municipal disaster data and flood information.

We focused on the daily municipal evacuee counts as reliable reference data. Flooded areas extracted from the satellite images were combined with population distribution data to estimate the “potentially affected population.” Previous studies ^9,10 similarly used inundation extent data from satellite images in conjunction with population data to estimate affected populations but did not compare these estimates with actual evacuee counts. This study analyzes the relationship with actual evacuee numbers and examines the challenges of using satellite observations for early disaster impact assessment.

2.2. Utilization of Satellite Data

In typhoon disasters, local governments report damage information, which the NDRRMC compiles into daily reports ¹¹. During Typhoon Ulysses (international name: Vamco) in November 2020, the number of evacuees recorded by the NDRRMC increased daily and peaked several days after the disaster occurred. This time lag can be attributed not only to delays in people entering evacuation centers but also to delays in reporting and communication within local areas. Such delays may hinder timely disaster response, highlighting the potential utility of satellite-based indirect assessments as supplementary or alternative measures. With the recent entry of private companies, high-resolution optical and synthetic aperture radar (SAR) satellite images have become increasingly accessible. Therefore, it is beneficial to verify the advantages and challenges of using satellite data through case studies.

2.3. Google Earth Engine

We used GEE to share the overall flood situation caused by Typhoon Ulysses. GEE is a geospatial analysis platform provided by Google that includes remote sensing datasets, freely available for non-commercial use. Various satellite and geographic information system datasets are pre-stored on the server, and algorithms created in the browser-based code editor are executed server-side, generating visualized outputs in real time. Moreover, the developed algorithms can be published as web applications, allowing any user to run the same analysis and confirm the results.

A big advantage of GEE is its ability to rapidly establish globally accessible systems for data aggregation and sharing, regardless of location or resource availability. Information can be shared interactively with high flexibility. In this study, the use of GEE had the following implications for disaster response:

1. 1)

   Provided an overview of the affected distribution even when the full extent of damage was unclear, supporting members less familiar with satellite data analysis and enabling the identification of areas with high support needs.

2. 2)

   Facilitated discussions on disaster response and recovery under COVID-19 restrictions, where researchers could not meet in person.

3. 3)

   Enabled damage distribution assessment by overlaying inundation maps with land-use data, thereby providing more actionable information for response and recovery.

4. 4)

   Allowed expansion of layers, supporting not only immediate disaster response but also monitoring of recovery and reconstruction progress.

3.1. Target Typhoon and Damage Situation

The target disaster is Typhoon Ulysses, which occurred in November 2020. Ulysses formed as a tropical depression east of Mindanao Island on November 8, intensified into a typhoon on November 9, and crossed Luzon Island from east to west between the night of November 11 and the morning of November 12. Its minimum central pressure was 955 hPa, with a maximum wind speed of 43 m/s. In some locations, the 24-hour accumulated rainfall exceeded 300 mm, causing extensive flooding. According to the final 29th report of NDRRMC ¹¹, 5.18 million people (1.26 million families) were affected, with 101 deaths, 85 injuries, and 10 missing. Agricultural damage amounted to 7.3 billion pesos (USD 150 million), infrastructure damage to 12.9 billion pesos (USD 260 million), and approximately 210,000 houses were reported damaged.

3.2. Target Basins

This study focused on three basins: the Cagayan River Basin (27,000 km\(^2\)), discharging into the Philippine Sea at the northern tip of Luzon; the Pampanga River Basin (11,000 km\(^2\)), discharging into Manila Bay; and the Pasig–Marikina–Laguna (PML) Basin (4,000 km\(^2\), including Laguna Lake), comprising 21 tributaries flowing into Laguna Lake and the Pasig and Marikina Rivers, which connect to Manila Bay (Fig. 1).

3.3. Typhoon Track Data

Typhoon track and gale area data were obtained from the RSMC Best Track Data ¹³ and Digital Typhoon archive ¹⁴.

3.4. Precipitation Data

Precipitation conditions were visualized using Japan Aerospace Exploration Agency’s Global Satellite Mapping of Precipitation (GSMaP) ¹⁵, which integrates multiple satellite observations to provide global rainfall data on a 0.1° grid at hourly intervals. Near-real-time data (GSMaP_NRT, \(\sim\)4-hour latency) and standard products (GSMaP_MVK, \(\sim\)3-day latency) are archived in GEE. We used these to calculate and visualize accumulated rainfall during the typhoon passage.

3.5. Disaster and Evacuee Information

NDRRMC reports commenced on November 11, 2020, at 08:00 (Report No.1) and continued until January 13, 2021 (Report No.29). Table 1 lists the evacuee counts. In the PML basin, 5,879 evacuees were reported in No.2 (Nov. 12, 5:00 PM), which increased to 27,886 in No.3 (Nov. 13, 8:00 AM). In contrast, no evacuees were recorded in Cagayan and Pampanga until No.4 (Nov. 14, 5:00 PM). Peak values were observed several days later: Cagayan 37,992 (Nov. 17, No.7), Pampanga 162,253 (Nov. 18, No.8), and PML 113,753 (Nov. 23, No.13).

3.6. Satellite Data

We used ESA’s Sentinel-2 optical imagery (S2-OPT, 10 m) and Sentinel-1 SAR (S1-SAR) ¹⁶. S2-OPT captured broad coverage of all three basins after 10:00 AM on November 13, revealing inundation in the cloud gaps (Fig. 2, left). S1-SAR observed most of the Cagayan middle–lower basin at approximately 6:00 PM on November 13. Pampanga and PML were imaged before 6:00 AM on November 17 (five days after landfall), and on November 18, both S2-OPT (after 10:00 AM) and S1-SAR (after 6:00 PM) were available. These datasets are archived in GEE and are typically updated within 24 hours of acquisition.

4.1. Sentinel-2 Optical Data and Flood Extraction

Sentinel-2 provides multispectral bands ranging from the visible to the shortwave infrared. Inundation can be delineated at 10 m resolution when cloud-free data are available. We applied the modified normalized difference water index ¹⁷ to post-flood images, classifying pixels with values greater than 0.15 as flooded. Although frequent cloud cover during typhoon season reduces usability, in this case, inundation was confirmed in all three basins on November 13 (Fig. 2, left).

4.2. Sentinel-1 SAR Data and Flood Extraction

SAR penetrates clouds and operates both day and night. Water surfaces appear dark owing to specular reflection, making flood detection straightforward ¹⁸. We compared pre- and post-typhoon imagery, classifying pixels with VH backscatter decreases \(\ge\)5 dB as inundated (value \(=1\); else \(=0\)). Slopes greater than 5° (digital elevation model-based) were masked, and a \(5\times 5\) median filter was applied to reduce speckle. After averaging \(5\times 5\) windows, areas greater than 0.1 were extracted as inundation (Fig. 2, right).

4.3. Estimating Potentially Affected Population

The flooded areas were overlaid with WorldPop 100 m population data ¹⁹. The “potentially affected population” was defined as the number of people within flooded zones aggregated at the municipality level. For consistency, inundation rasters (10 m resolution) were resampled to a 100 m grid, and counts were summed within administrative polygons.

6.1. Municipal Inundation Detection

On November 13, inundation was identified municipality by municipality (Fig. 2, Table 2). In Cagayan, cloud cover limited S2-OPT detection to 11 of 95 municipalities (12%); however, S1-SAR detected flooding in 56 municipalities (\(\sim\)60%) later that day. Pampanga and PML had fewer clouds, with S2-OPT detecting 68 (\(\sim\)90%) and 46 (\(\sim\)90%) municipalities, respectively. The optical and SAR data complemented each other effectively, enabling rapid post-typhoon assessments.

6.2. Estimation of Potentially Affected Population

The estimated population on November 13 was Cagayan (\(\sim\)390,000), Pampanga (\(\sim\)270,000), and PML (\(\sim\)80,000) (Table 2). Compared to peak NDRRMC evacuees, these values were approximately 10 times higher in Cagayan, 1.7 times higher in Pampanga, and 0.7 times lower in PML. Discrepancies arise because the estimates reflect only direct flooding, while evacuation decisions depend on other hazards. Furthermore, dense urban and under-canopy flooding is poorly detected by S2/S1 methods, and WorldPop distributions sometimes allocate population to fields, wetlands, or floodplains. The wide-area integration of such errors likely explains the large discrepancy in Cagayan. In PML, estimates lower than the recorded evacuee numbers may represent more realistic conditions.

6.3. Detection of Evacuation Occurrence

We examined whether the satellite-derived inundation data could detect municipalities with at least one evacuee (Table 3). The correct answer rates were high: S2-OPT achieved 100% (Cagayan), 95% (Pampanga), and 80% (PML). The lowest accuracy was observed in the S1-SAR case for Cagayan (74%), which was likely due to errors in the population dataset. Nevertheless, both optical and SAR imagery effectively identified the municipalities in which evacuations occurred.

6.4. GEE Analysis and Applications

Examples of the GEE app are shown in Figs. 3–5.

6.4.4. Post-Disaster Applications

Satellite data provide objective records of past and present conditions, supporting emergency decisions. In recovery, Build-Back-Better principles emphasize sharing past damage to build a common understanding. Monitoring recovery via GEE apps allows communities to assess changes and make equitable comparisons with others.

Despite the limited accuracy of the tools used, this methodology can provide a useful general overview of the situation in these unmonitored areas, thereby offering relevant stakeholders sufficient information for planning in the affected areas.