Flood Early Warning IoT

Flood Early Warning IoT — Topics for IoT Students

EARLY FLOOD WARNING USING SATELLITE-DERIVED CONVECTIVE SYSTEM AND

PRECIPITATION DATA - A RETROSPECTIVE CASE STUDY OF CENTRAL VIETNAM

T.-V. La∗ , T. H. Nguyen∗ , P. Matgen, and M. Chini

Luxembourg Institute of Science and Technology, 4362 Esch-sur-Alzette, Luxembourg

arXiv:2403.14395v1 [eess.IV] 21 Mar 2024

ABSTRACT progress has been made in addressing this challenge, thanks to

This paper addresses the challenges of an early flood the arrival of many advanced GEOstationary (GEO) satellites, warning caused by complex convective systems (CSs), by including Meteosat, GOES, Himawari, and Gaofen, cover- using Low-Earth Orbit and Geostationary satellite data. We ing Europe/Africa, Americas, and Asia-Pacific, respectively. focus on a sequence of extreme events that took place in cen- Nevertheless, a more comprehensive understanding of the in- tral Vietnam during October 2020, with a specific emphasis tricate relationships between deep convective clouds and ex- on the events leading up to the floods, i.e., those occurring treme weather events is essential to enhance the prediction of before October 10th , 2020.

In this critical phase, several natural disasters such as destructive floods caused by heavy hydrometeorological indicators could be identified thanks rainfall. to an increasingly advanced and dense observation network Mesoscale CS tracking and identification are mainly composed of Earth Observation satellites, in particular those based on the analyses of brightness temperature and pre- enabling the characterization and monitoring of a CS, in terms cipitation. [1, 2, 3] indicated the strong relationship between of low-temperature clouds and heavy rainfall.

Himawari-8 deep convective clouds and strong surface winds via intense images, both individually and in time-series, allow identify- downdrafts, based on the collocation of GEO (Meteosat) and ing and tracking convective clouds. This is complemented Low-Earth Orbit (LEO) Sentinel-1 and Aeolus satellite im- by the observation of heavy/violent rainfall through GPM agery. Furthermore, [2, 3] showed that the high-resolution IMERG data, as well as the detection of strong winds using (HR) surface wind patterns, derived from Sentinel-1 (S1) radiometers and scatterometers.

Collectively, these datasets, images, moved (horizontally) in the same direction as the along with the estimated intensity and duration of the event deep convective clouds detected by Meteosat. In particular, from each source, form a comprehensive dataset detailing the [3] discussed the relationship between strong surface winds, intricate behaviors of CSs. All of these factors are significant heavy rainfall, and deep convection through the collocation contributors to the magnitude of flooding and the short-term of S1 SAR, WindSat scatterometer, and Meteosat data.

They dynamics anticipated in the studied region. concluded that that intense surface winds were followed by heavy rainfall associated with deep convection approximately Index Terms— Convective systems, Floods, Sentinel-1, 30 minutes later. Regarding the relationship between deep Himawari-8, GPM IMERG, WindSat, ASCAT, SMAP. convective clouds, heavy rainfall, and floods, [4] utilized radar data to characterize convective storms and connected 1.

INTRODUCTION them to the flood events that occurred on the northwestern Mediterranean coast. The results showed that the area was Severe convective systems (CSs) are regularly observed in mostly affected by shallow but efficient convection, associ- sub-tropical and tropical regions, namely the Gulf of Guinea, ated with high rainfall rates, which tend to exceed infiltration the Gulf of Mexico, as well as the whole Southeast Asia, Aus- capacities and/or saturate catchments quickly and thus lead tralia, Indian subcontinent, etc.

They are typically associated to fast run-off and flooding. [5] investigated the contribu- with extreme weather events such as intense thunderstorms, tion of mesoscale CSs to the selected flood cases in southern heavy rainfall, and strong surface wind gusts. The develop- West Africa based on the analysis of rainfall rates derived ment of a CS often brings heavy precipitation over a small from brightness temperature obtained from the NOAA Cli- area in a short period, leading to flash floods.

These events mate Data Record of Gridded Satellite Data. It indicated that pose significant hazards due to their rapid onset, not allow- floods in this area had between 31% and 60% of contribution ing sufficient time for warnings and evacuations. However, from mesoscale CSs. accurately nowcasting and forecasting CSs remains a major This paper tackles the challenges of issuing an early flood challenge for meteorologists due to their sudden ignition and warning based on the analyses of multi-source LEO (SAR, fast development.

During the last four decades, considerable scatterometers, and radiometers) and GEO satellite data. The ∗ These authors contributed equally. different sensors provide complementary observations re-

garding hydrometeorological factors contributing to a flood Table 1: Satellite Images Used for Early Flood Warning - event. This study explores the correlation between deep con- Case Study of the Central Region of Vietnam, Oct. 2020. vective clouds identified by GEO satellites, patterns of strong Satellite Type/Mode Variable Availability surface winds, and heavy rainfall observed by LEO satellites, LEO/ Wind, Oct. 03, 05, 06 all of which are associated with severe flooding.

The findings Sentinel-1 C-SAR Flood extent & 10 are validated through the mapping of flood extents using SAR LEO/ Wind/ Oct. 07 & 08 WindSat technology. A specific case study of the flood events from Radiometer Precipitation Oct. 3rd to 10th , 2020 that occurred in the central region of ASCAT- LEO/ Oct. 03-10 Wind A/B/C Scatterometer Vietnam is carried out in this paper. As reported by the Red LEO/ Oct. 04, 06, 07 Cross [6], this area experienced prolonged, heavy rains that SMAP Wind Radiometer & 09 caused severe and widespread flooding and landslides in eight Himawari-8 GEO Brightness Oct. 03-10, provinces between Oct. 6th and 14th , 2020.

This was due temperature every 10 min Oct. 03-10, to the combination of numerous weather systems, including GPM IMERG Multi-satellite Precipitation every 30 min Inter Tropical Convergence Zones (ICZ) combining with cold air, tropical storms, and typhoons. coast of the QB and QT provinces, Vietnam, on Oct. 10th , 2. METHODOLOGY 2020. They are mapped using the S1 image acquired on the Oct. 10th with respect to a reference non-flood S1 image ac- 2.1.

Data selection and preparation quired on Aug. 11th , 2020. Recent research works [8, 9] on flood forecasting relied on S1-derived flood extent map Fig. 1 illustrates the region of interest (ROI), i.e., 14◦ − as validation and assimilation data. The radiometers Wind- 20◦ N, 103◦ − 110◦ E, including eight provinces of Vietnam, Sat and SMAP also provide ocean surface wind speed and where flood events were observed starting from Oct. 6th , direction for 2 to 4 days, at a coarser resolution.

Similarly, 2020 [7]. It also consists of the coastal areas from which using scatterometer data (ASCAT-A/B/C), we can extract the CSs are formed and developed before arriving on land and wind speed over the ocean surfaces. WindSat data can also of- producing strong surface winds and heavy rainfall.

The ROI fer precipitation data in addition to half-hourly GPM IMERG include the provinces Nghe An (NA), Ha Tinh (HT), Quang precipitation accumulations [10]. Binh (QB), Quang Tri (QT), Thua Thien Hue (TT), Quang Nam (QN1) and Quang Ngai (QN2), as well as the city of Da Nang (DN).

Fig. 2: S1-derived flooded areas observed along the coast of

QB and QT provinces, Vietnam on Oct. 10th , 2020. Fig. 1: Region of interest (ROI), including eight provinces and ocean areas of Vietnam.

The satellite data, utilized to investigate the relationship 2.2. CS Identification from LEO and GEO Images between deep convective clouds, strong surface winds, heavy 2.2.1. Deep convective clouds rainfall, and flood events for the period of Oct. 3rd to 10th , 2020, are shown in Table 1.

To detect deep convective clouds H8 images observe deep convective clouds that typically that can produce strong surface winds and heavy rainfall, we exhibit a low brightness temperature between 200 and 220 rely on the GEO Himawari-8 (H8) images provided every 10 K. They correspond to convective wind patterns observed minutes. We use S1 SAR images (available on four different on LEO images, namely S1, WindSat, ASCAT-A/B/C, and days) to retrieve the HR surface wind speed and map flood SMAP.

The time-series of H8 images are also taken into ac- extent. Fig. 2 presents the flooded areas observed along the count to detect cloud movements (i.e., general direction and

velocity) during extreme weather, and track those with low temperature . This enables a broad assessment of the area affected by the investigated CS, particularly in the presence of concurrent heavy rainfall.

2.2.2. Surface convective wind patterns

Surface convective wind patterns can be derived from the im- ages acquired by SARs, scatterometers, and radiometers. The wind speed observed over sea surfaces, potentially associated with the coldest convective clouds, can be estimated using S1 Fig. 3: Data and variables of interest. images. Such wind speed can be up to 25 m/s.

It has been shown that the intense downdrafts associated with deep con- vective clouds can induce surface wind gusts when they hit cal operational questions can be answered based on decision- the surface of the sea [1, 2, 3]. In other words, vertical winds level data fusion taking into account anecdotal evidence from assert the active level of a CS, as well as being an element each dataset. Except brightness temperature, every variable that induces heavier rainfall through downdrafts.

The strong of interest can be estimated or measured from at least two wind gusts significantly impact sea surface roughness, lead- satellites. This enables enhanced spatial coverage and tem- ing to observable high-intensity radar backscattering or areas poral sampling, particularly for areas with restricted cover- with increased Normalized Radar Cross Section (NRCS) in age or infrequent revisits. Surface wind speed estimates are the images captured by SARs and scatterometers.

Using Geo- provided in LR by WindSat/ASCAT/SMAP and in HR by S1 physical Model Functions, e.g. CMOD for C-band, LMOD data. They are to be analyzed in conjunction with the location for L-band, and XMOD for X-band, we can retrieve surface of low-temperature clouds and presence of intense precipi- wind speed from NRCS, as well as wind direction, and other tation, constituting deep convections.

Precipitation measure- radar parameters. ments can also be either from GPM IMERG or from WindSat, Regarding the radiometer data, surface wind speed is re- offering complementary rainfall observations over ocean sur- trieved from the sea surface brightness temperature measured face regions. by SMAP and WindSat. To this end, [8] took into account the differences between measured sea surface brightness tem- peratures and those of a flat ocean surface and matched these 3.

EARLY FLOOD WARNING DECISION SYSTEM differences to the Radiative Transfer Model (RTM). Once the differences in brightness temperature and other parameters This section unveils the outcomes of satellite data analysis are determined, one can invert the RTM to estimate surface and integration, focusing on the identification and surveil- wind speed. lance of intense CSs and the corresponding extreme weather events responsible for the floods witnessed in the central re- gion of Vietnam in October 2020.

(Fig. 2). The identification 2.2.3. Rainfall of the relationship between deep convective clouds and ex- In addition to strong surface winds, deep convective clouds treme weather events is facilitated in recent years, thanks to have the potential to generate substantial rainfall, a crucial GEO H8 data acquired every 10 minutes, and continuous ob- factor contributing significantly to the initiation of flash flood servations every 30 minutes from GPM IMERG Early Run. events, as discussed in [5].

In this study, we utilized the pre- Fig. 4 shows the collocation of H8, S1, and GPM IMERG cipitation accumulation data provided by the GPM IMERG images acquired on the Oct 5th , 2020, i.e., before the reported [9] and by WindSat (over the ocean surface). As demon- flood events [7]. They show, respectively, (a) cloud brightness strated in [3], convective rainfall may occur subsequent to the temperature, (b) strong surface wind patterns, and (c) heavy onset of strong surface winds, aligning with the intensity of rainfall occurring in the region.

The deepest (coldest) con- the deep convective clouds. vective clouds depicted by a brightness temperature of around 200 K (Fig. 4a), align with strong surface wind zones ranging from 18 to 25 m/s along the coast (Fig. 4b), coupled with es- 2.3. Data Fusion calating rainfall accumulation (Fig. 4c). Such a combination The combination of precipitation and brightness temperature can be found in convection lines, typically squalls [1, 2, 3] has been leveraged to identify, locate, and track mesoscale over the coastal areas.

This substantiates the presence of a CSs. Fig. 3 summarizes this fusion framework between the CS that developed along the coast (effectively an Intertropical input data. Using H8 and GPM IMERG, in addition to S1 Convergence Zone or ICZ), attributed to the S1-derived HR HR and WindSat/ASCAT/SMAP low-resolution (LR) identi- wind patterns and the frequent temporal sampling provided fication of surface wind direction and speed, a series of typi- by H8 and GPM IMERG data.

Next, considering the ob-

served westward movement of these low-temperature clouds ing from the initial Intertropical Convergence Zone (ICZ) and (Fig. 4d), jointly with the coarse surface wind direction pro- tropical storm Linfa, it is evident that factors contributing to vided by WindSat, it can be inferred that the area comprising flooding have accumulated, even though official warnings had four regions (namely TT, DN, QN1, and QN2) was poised to not been issued at that time.

As a matter of fact, the areas im- experience substantial rainfall, potentially leading to floods in pacted by heavy rainfall associated with many observed se- the following days, with DN identified as the most vulnerable vere CSs correspond to the detected flooded areas (Fig. 2). area. Using GPM IMERG data, the heavy precipitation per- sisted in these areas (partly due to the incoming tropical storm 4. CONCLUSIONS Linfa) until Oct. 9th , 2020, when it became less severe in DN, QN1 and QN2.

This paper presents a preliminary concept of an early flood Fig. 5 depicts the integration of H8, WindSat surface wind warning system that is based on the use of multiple satel- speed, and rainfall, along with GPM IMERG images obtained lite data, namely Low-Earth Orbit (LEO) imagery data and on October 7th , 2020. This timeframe falls after the occur- Geostationary (GEO) imagery data. We examine a series of rence of the initial ICZ and before the onset of the series of extreme events that transpired in central Vietnam in October typhoons/tropical storms, which commenced on October 11th , 2020, with a specific focus on the period leading up to it (be- 2020, as documented in [6, 7].

They depict, respectively, ob- fore Oct. 10th , 2020). Within these events, various hydrom- served deep convective clouds, strong surface winds, and in- eteorological indicators could be identified in advance, par- tense precipitation. On the H8 image (Fig. 5a), as shown ticularly those capable of tracking and monitoring extreme earlier in Fig. 4a, we can identify the deep convective clouds events, such as low-temperature clouds and prolonged heavy based on its low brightness temperature of approximately 200 rainfall in specific areas.

If these pertinent indicators had K. It should be noted that like GPM IMERG data, WindSat been provided in a timely manner, they could have been valu- can provide precipitation estimates over sea surfaces, but at a able for forecasting the geographical extent and severity of larger spatial resolution. Similarly, the WindSat-derived sur- the ensuing extreme flood events Therefore, in this research face wind maps are coarser than S1 ones; however, their larger work, we investigated the collocation of different LEO/GEO swath allow to observe a wider offshore wind area as well as datasets in order to determine the causality between intense identify future incoming events. rainfall, low-temperature clouds, strong wind speed, and the At the same time and location, we observe strong surface potential for causing substantial flooding in the same location. wind patterns (12.5–20 m/s) and heavy rainfall rate (8–13 This work demonstrates the feasibility of issuing a warning mm/h) on the WindSat images (Fig. 5b-5c).

GPM IMERG between several hours and one day in advance due to the pres- also confirms the intense precipitation present over the whole ence of high rainfall rates and low-temperature clouds cov- coastal region. Continuous observations in these areas, with ering large coastal regions. In addition, utilizing H8 time- H8, GPM and WindSat/ASCAT/SMAP between Oct. 7th series of brightness temperature images, it becomes viable to and Oct. 10th suggests that these areas are likely to experi- forecast the trajectory and velocity of these convective low- ence prolonged periods of intense winds and heavy rainfall. temperature clouds moving from the sea towards the land.

The matching of these observed variables demonstrates the This allows us to identify the regions susceptible to significant strong association between deep convective clouds and ex- rainfall. Lastly, using different low-resolution scatterometers treme weather events. This further supports the widespread in combination with the HR wind speed estimation from S1 utilization of GEO images in combination with wind and images, local wind patterns and CS effects could be derived, rainfall data for the early detection of flood events in regions showing the severity of wind in the studied areas. or seasons where severe CSs occur regularly.

In addition, concerning wind speed data over the stud- 5. ACKNOWLEDGMENT ied ROI, a summary of satellite observations leading up to Oct 10th , 2020 combining S1, WindSat, ASCAT-A/B/C and This work is supported by the Luxembourg National Re- SMAP data has been analyzed. In order to harmonize the search Fund (FNR) in the framework of the CORE project wind speed estimated from SARs and scatterometers, they are C20/SR114703579. categorized into “none”, “weak”, “moderate”, and “severe”, respectively corresponding to less than 5 m/s, 5-10 m/s, 10- 6.

REFERENCES 15 m/s, and greater than 15 m/s. From this collection, it can be noted that on the Oct. 6th , 2020, there is an indication of [1] T. V.

La and C. Messager, “Convective system dynamics wind getting stronger (starting from Oct. 5th PM and subse- viewed in 3d over the oceans,” Geophysical Research quently reaching severe levels in the afternoon of October 6th . Letters, vol. 48, no. 5, p. e2021GL092397, 2021.

Over the next three days, severe winds continue to be preva- lent across the entire ROI, In conjunction with the presence [2] ——, “Convective system observations by leo and geo of observed low-temperature clouds and heavy rainfall result- satellites in combination,” IEEE Journal of Selected

(a) Himawari-8, 22:40 UTC (b) S1, 22:36:04 UTC (c) GPM, 22:30-23:00 UTC (d) Extent of clouds

Fig. 4: Combination of satellite data for observing a severe CS and its associated factors, on Oct. 5th , 2020, including (a) deep convective clouds detected by H8 band 10 (central wavelength: 7.35 µm), (b) strong surface wind pattern observed by S1 (c) heavy rainfall extracted from GPM IMERG, and (d) Extent of clouds under 220 K between 22:00-23:50 UTC showing a westward direction.

(a) Himawari-8, 23:10 UTC (b) WindSat, 23:12 UTC (c) WindSat, 23:12 UTC (d) GPM, 23:00-23:30 UTC

Fig. 5: Combination of various satellite data for observing a severe CS and its associated factors, on Oct. 7th , 2020: (a) deep convective clouds detected by H8 band 10 (central wavelength: 7.35 µm), (b) strong surface wind pattern and (c) heavy rainfall observed by WindSat, and (d) heavy rainfall extracted from GPM IMERG data.

Topics in Applied Earth Observations and Remote Sens- nam: Floods - Final Report, Operation n° ing, vol. 14, pp. 11 814–11 823, 2021. MDRVN020 ,” https://reliefweb.int/report/viet-nam/ vietnam-floods-final-report-operation-n-mdrvn020, [3] ——, “Different observations of sea surface wind pat- accessed: 2024-XX-XX. tern under deep convection by sentinel-1 sars, scat- terometers, and radiometers in collocation,” IEEE Jour- [7] Viet Nam Disaster Management Authority (VNDMA), nal of Selected Topics in Applied Earth Observations “Flash update no. 4: Viet Nam floods, land- and Remote Sensing, vol. 15, pp. 3686–3696, 2022. slides and storms (as of 28 October 2020),” https://phongchongthientai.mard.gov.vn/en/Pages/ [4] A. del Moral, M. del Carmen Llasat, and T.

Rigo, “Con- flash-update-no-4-viet-nam-floods-landslides-and-storms-as-of-28-o necting flash flood events with radar-derived convective aspx, accessed: 2024-XX-XX. storm characteristics on the northwestern mediterranean coast: knowing the present for better future scenarios [8] T. H. Nguyen, S.

Ricci, C. Fatras, A. Piacentini, A.

Del- adaptation,” Atmospheric Research, vol. 238, p. 104863, motte, E. Lavergne, and P. Kettig, “Improvement of 2020.

Flood Extent Representation With Remote Sensing Data and Data Assimilation,” IEEE Transactions on Geo- [5] W. A. Atiah, L.

K. Amekudzi, and S. K.

Danuor, science and Remote Sensing, vol. 60, pp. 1–22, 2022. “Mesoscale convective systems and contributions to flood cases in southern west africa (swa): A system- [9] T. H. Nguyen, S.

Ricci, C. Fatras, A. Piacentini, E.

Si- atic review,” Weather and Climate Extremes, p. 100551, mon, R. Rodriguez Suquet, and S. Pena Luque, “Gaus- 2023. sian Anamorphosis for Ensemble Kalman Filter Anal- ysis of SAR-Derived Wet Surface Ratio Observations,” [6] International Federation of Red Cross And IEEE Transactions on Geoscience and Remote Sensing, Red Crescent Societies (IFRC), “Viet- vol. 62, no. 1-21, 2023.

[10] T. Meissner, L. Ricciardulli, and F.

J. Wentz, “Capabil- ity of the smap mission to measure ocean surface winds in storms,” Bulletin of the American Meteorological So- ciety, vol. 98, no. 8, pp. 1660–1677, 2017.