Conference PaperPDF Available

Assessment of an Extreme Rainfall Detection System for Flood Prediction over Queensland (Australia)

Authors:

Abstract and Figures

Flood events represent some of the most catastrophic natural disasters, especially in localities where appropriate measurement instruments and early warning system are not available. Remotely sensed data can often help to obtain near real-time rainfall information with a global spatial coverage without the limitations that characterize other instruments. In order to achieve this goal, a freely accessible Extreme Rainfall Detection System (ERDS – erds.ithacaweb.org) was developed and implemented by ITHACA with the aim of monitoring and forecasting exceptional rainfall events and providing information in an understandable way also for non-specialized users. The near real-time rainfall monitoring is performed taking advantages of NASA GPM IMERG half-hourly data (one of the most advanced rainfall measurements provided by satellite). This study aims to evaluate ERDS performance in the detection of the extreme rainfall that led to a massive flood event in Queensland (Australia) between January and February 2019. Due to the impressive amount of rainfall that affected the area, Flinders River (one of the longest Australian river) overflowed, expanding to a width of tens of kilometres. Several cities were also partially affected and Copernicus Emergency Management Service was activated with the aim of providing an assessment of the impact of the event. In this research, ERDS outputs were validated using both in-situ and open source remotely sensed data. Specifically, taking advantage of both NASA MODIS (Moderate-resolution Imaging Spectroradiometer) and Copernicus Sentinel datasets it was possible to have a clear look of the full extent of the flood event. GPM data proved to be a reliable source of rainfall information for the evaluation of areas affected by heavy rainfall. By merging these data, it was possible to recreate the dynamics of the event.
Content may be subject to copyright.
The 3rd International Electronic Conference on Remote Sensing (ECRS 2019), 22 May5 June 2019;
Sciforum Electronic Conference Series, Vol. 3, 2019
Conference Proceedings Paper
1
Assessment of an Extreme Rainfall Detection System
2
for flood prediction over Queensland (Australia)
3
Paola Mazzoglio 1,*, Francesco Laio 2, Constantin Sandu 1 and Piero Boccardo 3
4
1 ITHACA - Information Technology for Humanitarian Assistance, Cooperation and Action, Torino, 10138,
5
Italy
6
2 Politecnico di Torino, Dipartimento di Ingegneria dell’Ambiente, del Territorio e delle Infrastrutture,
7
Torino, 10129, Italy
8
3 Politecnico di Torino, Dipartimento Interateneo di Scienze, Progetto e Politiche del Territorio, Torino, 10125,
9
Italy
10
* Correspondence: mazzoglio.paola@gmail.com
11
Published: date
12
Academic Editor: name
13
Abstract: Flood events represent some of the most catastrophic natural disasters, especially in
14
localities where appropriate measurement instruments and early warning system are not available.
15
Remotely sensed data can often help to obtain near real-time rainfall information with a global
16
spatial coverage without the limitations that characterize other instruments. In order to achieve this
17
goal, a freely accessible Extreme Rainfall Detection System (ERDS erds.ithacaweb.org) was
18
developed and implemented by ITHACA with the aim of monitoring and forecasting exceptional
19
rainfall events and providing information in an understandable way also for non-specialized users.
20
The near real-time rainfall monitoring is performed taking advantages of NASA GPM IMERG
21
half-hourly data (one of the most advanced rainfall measurements provided by satellite). This
22
study aims to evaluate ERDS performance in the detection of the extreme rainfall that led to a
23
massive flood event in Queensland (Australia) between January and February 2019. Due to the
24
impressive amount of rainfall that affected the area, Flinders River (one of the longest Australian
25
river) overflowed, expanding to a width of tens of kilometres. Several cities were also partially
26
affected and Copernicus Emergency Management Service was activated with the aim of providing
27
an assessment of the impact of the event. In this research, ERDS outputs were validated using both
28
in-situ and open source remotely sensed data. Specifically, taking advantage of both NASA MODIS
29
(Moderate-resolution Imaging Spectroradiometer) and Copernicus Sentinel datasets it was possible
30
to have a clear look of the full extent of the flood event. GPM data proved to be a reliable source of
31
rainfall information for the evaluation of areas affected by heavy rainfall. By merging these data, it
32
was possible to recreate the dynamics of the event.
33
Keywords: early warning system; extreme events; flood monitoring; GPM; hydrology; rainfall
34
35
1. Introduction
36
According to the Australian Government Bureau of Meteorology (BOM), heavy rainfall affected
37
Queensland (Australia) from 26th January 2019 until 9th February 2019 [1]. Several localities
38
received more than four times their February average rainfall [1]. The massive amount of rainfall led
39
to moderate to major flooding.
40
This study aims to evaluate NASA GPM (Global Precipitation Measurement) IMERG
41
(Integrated Multi-satellite Retrievals for GPM) V05B early run half-hourly data [2] in the detection of
42
the extreme rainfall that led to this massive flood event by comparing the weekly accumulated
43
The 3rd International Electronic Conference on Remote Sensing (ECRS 2019), 22 May5 June 2019;
Sciforum Electronic Conference Series, Vol. 3, 2019
2
rainfall with in-situ rainfall measurements. Alerts provided by ITHACA Extreme Rainfall Detection
44
System (ERDS) were analyzed in order to estimate the most affected areas. ERDS outputs were also
45
validated using an automatic flooded areas extraction performed both on Sentinel-3 and on MODIS
46
(Moderate-resolution Imaging Spectroradiometer) optical images acquired after the end of the rainy
47
period.
48
Obtained results highlighted that both IMERG data and ERDS outputs proved to be a reliable
49
source of information for the evaluation of areas affected by heavy rainfall.
50
2. Experiments
51
2.1. ITHACA Extreme Rainfall Detection System
52
The Extreme Rainfall Detection System is a service for the monitoring and forecasting of
53
exceptional rainfall events [3]. This system provides both information on the rainfall amount and
54
heavy rainfall alerts for different aggregation intervals (12, 24, 48, 72 and 96 hours) using NASA
55
GPM IMERG early run half-hourly data as near real-time source of rainfall measurements. Outputs
56
are provided with a 0.1° spatial resolution in the latitude range between 60° N 60° S.
57
The extreme rainfall detection is based on the concept of activation threshold: an event is
58
identified when the rainfall exceeds a given threshold value. An “event-identification threshold”
59
(EIT) represents the amount of rainfall needed to trigger a flood event induced by extreme rainfall
60
[3]. Specifically, an alert is provided if the accumulated rainfall exceeds the EIT. The proper EIT
61
values were assessed in Mazzoglio et al. [3] for every previously mentioned aggregation interval.
62
The proposed threshold methodology is based on threshold values equal to a percentage (pT.R.) of the
63
mean annual precipitation.
64
    
(1)
where
65
T represents the threshold;
66
T.R. represents the total rainfall (i.e., the mean annual rainfall calculated using 10 years of
67
GPCC monthly “Monitoring Product” [4]);
68
pT.R. is a parameter representing the fraction of the total rainfall leading to the extreme event
69
identification.
70
A lower bound was also applied in order to avoid very low thresholds in localities
71
characterized by low mean annual precipitation (Table 1). Conversely, an upper bound was applied
72
to avoid unrealistically high threshold values in localities characterized by high mean annual
73
precipitation.
74
Table 1. Threshold values used for the extreme rainfall detection.
75
Aggregation Interval
(hours)
pT.R.
(%)
Lower Bound
(mm)
12
6
100
24
8
120
48
12
140
72
15
170
96
16
190
2.2. Analysis of Rainfall Measurements and Alerts Datasets
76
In this study, NASA GPM IMERG V05B early run half-hourly data were analyzed in the
77
following three different time periods:
78
from 18th January 2019 23:00 UTC to 25th January 2019 22:59 UTC;
79
The 3rd International Electronic Conference on Remote Sensing (ECRS 2019), 22 May5 June 2019;
Sciforum Electronic Conference Series, Vol. 3, 2019
3
from 25th January 2019 23:00 UTC to 1st February 2019 22:59 UTC;
80
from 1st February 2019 23:00 UTC to 8th February 2019 22:59 UTC.
81
Specifically, a comparison with daily rainfall measurements [5] contained in the Bureau of
82
Meteorology climate database, the Australian Data Archive for Meteorology (ADAM), was
83
performed. The weekly difference between these two gridded products was evaluated in order to
84
detect local underestimations/overestimations.
85
Also the spatial and temporal distribution of the alerts provided by ERDS were analyzed in the
86
same three time periods in order to evaluate the most affected areas.
87
2.3. Automatic Flooded Areas Extraction
88
Freely accessible satellite images were analyzed in order to evaluate the presence and the
89
temporal evolution of the flooded areas. Sentinel-1 and Sentinel-2 images, despite the good spatial
90
resolution, were discarded due to the long revisit time. Sentinel-1 and Sentinel-2 satellite, in fact,
91
were able to cover only a small portion of Queensland in the useful time window. Sentinel-3 is
92
instead characterized by a revisit time of less than two days, allowing to obtain frequent updates.
93
MODIS was instead designed with a daily revisit time.
94
An automatic flooded areas extraction was performed both on Sentinel-3 and on MODIS
95
images acquired on 10th, 13th, 15th and 21st February. Images acquired before 10th February cannot
96
be used due to the high cloud coverage. Satellite images were downloaded using Sentinel-hub
97
EO-Browser [6].
98
This water extraction was performed taking advantage of the peculiarity of the Normalized
99
Difference Water Index (NDWI) in the identification of water features. This index makes use of
100
near-infrared radiation and visible green light to enhance the presence of such features while
101
eliminating the presence of soil and terrestrial vegetation features [7].
102
    
  
(2)
where:
103
G is the green band;
104
NIR is the near infrared band.
105
NDWI was generated from Sentinel-3 using B06 (λ centre equal to 560 nm, 300 m spatial
106
resolution) as green band and B19 centre equal to 900 nm, 300 m spatial resolution) as NIR band
107
[8] while for MODIS it was used B04 (545 565 nm, 500 m spatial resolution) as green band and B05
108
(1230 1250 nm, 500 m spatial resolution) as NIR band [9].
109
Pixels characterized by a NDWI equal to or greater than 0.1 were classified as water. Pixels
110
characterized by NDWI lower than 0.1 were classified as no-water.
111
This approach, unfortunately, suffers from drawbacks induced by false alarms in cloudy zones.
112
A manual refinement of the water mask by means of visual interpretation proved to be necessary in
113
order to remove false alarms induced by the presence of clouds in some portion of the images.
114
3. Results
115
Figure 1 compares the accumulated rainfall for three different time period evaluated both from
116
in-situ and from GPM IMERG products. The first row shows the results obtained using daily rainfall
117
measurements contained in the Bureau of Meteorology climate database while the second row was
118
obtained using NASA GPM IMERG early run half-hourly data. The third row shows the difference
119
between these two products.
120
Both in-situ and satellite data confirmed that the maximum accumulated rainfall was recorded
121
in the second week (maximum value is equal to 1064 mm according to BOM and 773 mm according
122
to GPM IMERG data).
123
The 3rd International Electronic Conference on Remote Sensing (ECRS 2019), 22 May5 June 2019;
Sciforum Electronic Conference Series, Vol. 3, 2019
4
During the second week, a modest overestimation in the weekly accumulated rainfall obtained
124
using IMERG data emerged in the southern part of Queensland (red zones reported in the third row
125
of Figure 1). This underestimation could be partly induced by the absence of measurements in the
126
daily in-situ rainfall totals. While the original analysis shows zero accumulated rainfall in some parts
127
of that area, the recalibrated dataset is characterized by the absence of data.
128
During the third week, instead, a considerable underestimation is recorded in the central part of
129
Queensland (blue zones reported in the third row of Figure 1). Also the location of the maximum
130
accumulated rainfall is subject to a modest spatial shift.
131
The maximum positive weekly difference was recorded during the third week (741 mm) near
132
Townsville (one of the most affected cities). The maximum negative weekly difference was instead
133
recorded during the second week (- 599 mm) near Normanton.
134
135
Figure 1. Accumulated rainfall obtained using in-situ and satellite measurements. In the graphs
136
related to the differences between in-situ and satellite products (third row), negative values highlight
137
places were IMERG overestimates the weekly rainfall amount. The reference system is WGS84.
138
Figure 2 provides an overview of the areas affected by heavy rainfall (according to ERDS near
139
real-time alerts) in the three previously mentioned weeks. In the first week, alerts were provided
140
only in the northern areas of Queensland. During the second and third weeks, instead, alerts were
141
issued also in the central part of Queensland.
142
The 3rd International Electronic Conference on Remote Sensing (ECRS 2019), 22 May5 June 2019;
Sciforum Electronic Conference Series, Vol. 3, 2019
5
For a proper understanding of the results, it is important to highlight that ERDS analyzes 48
143
half-hourly rainfall measurements every day. In other words, every week ERDS could provide a
144
maximum of 336 half-hourly alerts.
145
Figure 3 shows the automatic flooded areas extraction performed both on Sentinel-3 and on
146
MODIS images. This water extraction includes also the reference water (a clip was performed only
147
over the sea). Red features represent the Areas of Interest analyzed by Copernicus Emergency
148
Management Service [10]. A modest underestimation is visible in the flooded areas extracted using
149
MODIS images.
150
151
Figure 2. Number of extreme rainfall alerts provided by ERDS. The reference system is WGS84.
152
153
Figure 3. Flooded areas extraction performed both on Sentinel-3 and on MODIS images. Dark grey
154
zones represents areas non covered by the images. The first figure shows both Copernicus
155
Emergency Management Service AOIs (Areas of Interest) and STRM (Shuttle Radar Topography
156
Mission) DEM (Digital Elevation Model). The reference system is WGS84.
157
The 3rd International Electronic Conference on Remote Sensing (ECRS 2019), 22 May5 June 2019;
Sciforum Electronic Conference Series, Vol. 3, 2019
6
4. Discussion
158
GPM IMERG early run half-hourly data proved to be a good source of information for rainfall
159
monitoring at the regional scale. Due to the coarse spatial resolution (0.1°), local scale validation is
160
recommended.
161
Obtained results indicated that ERDS was able to detect the most affected areas. The
162
discrepancies between flooded areas and ERDS alerts location are mainly induced by the
163
characteristics of the early warning system. The system, in fact, provides alerts about heavy rainfall.
164
No analysis regarding the areas that will be affected by flood events is performed. Further studies
165
could be conducted in order to implement information about the morphology of the territory with
166
the aim of providing information about where the flood events will occur.
167
5. Conclusions
168
The evidence presented in this study suggested that both GPM IMERG early run data and
169
ERDS outputs proved to be a reliable source of near real-time rainfall information for the monitoring
170
of heavy rainfall events. These findings have important applications for countries were an
171
appropriate network of measurement instruments is still missing.
172
However, a number of significant limitation must be highlighted. Modest
173
underestimations/overestimations are reported in different localities, especially when high rainfall
174
rate is measured. Local-scale validation is recommended due to the native spatial resolution.
175
Acknowledgments: We express our gratitude to Simone Balbo and Paolo Pasquali, that helped us in the
176
development of the web app.
177
Author Contributions: Formal analysis, P.M. and C.S.; Methodology, P.M., F.L. and C.S; Software, P.M. and
178
C.S.; Writing original draft, P.M.; Writing review & editing, P.M., F.L., C.S., P.B.
179
Conflicts of Interest: The authors declare no conflict of interest.
180
References
181
1. Australian Government Bureau of Meteorology. Special Climate Statement 69an extended period of
182
heavy rainfall and flooding in tropical Queensland.
183
www.bom.gov.au/climate/current/statements/scs69.pdf (accessed on 8 March 2019).
184
2. Huffman, G.J. GPM IMERG Early Precipitation L3 Half Hourly 0.1 Degree x 0.1 Degree V05; Goddard Earth
185
Sciences Data and Information Services Center (GES DISC): Greenbelt, MD, USA, 2015.
186
3. Mazzoglio, P.; Laio, F.; Balbo, S.; Boccardo, P.; Disabato, F. Improving an Extreme Rainfall Detection
187
System with GPM IMERG data. Remote Sens. 2019, 11, 677. doi.org/10.3390/rs11060677
188
4. Schneider, U.; Becker, A.; Finger, P.; Meyer-Christoffer, A.; Rudolf, B.; Ziese, M. GPCC Monitoring
189
Product: Near Real-Time Monthly Land-Surface Precipitation from Rain-Gauges based on SYNOP and
190
CLIMAT data. 2011. doi.org/10.5676/DWD_GPCC/MP_M_V4_100
191
5. http://www.bom.gov.au/jsp/awap/rain/index.jsp (accessed on 17 February 2019).
192
6. Sentinel-hub EO-Browser. https://apps.sentinel-hub.com/eo-browser/ (accessed on 09 April 2019).
193
7. McFeeters, S.K. The use of the Normalized Difference Water Index (NDWI) in the delineation of open
194
water features. Int J Remote Sens. 1996, 17:7, 1425-1432. doi.org/10.1080/01431169608948714
195
8. User Guides - Sentinel-3 OLCI Radiometric Resolution Sentinel Online.
196
https://sentinel.esa.int/web/sentinel/user-guides/sentinel-3-olci/resolutions/radiometric (accessed on 09
197
April 2019).
198
9. MODIS Specifications. https://modis.gsfc.nasa.gov/about/specifications.php (accessed on 09 April 2019).
199
10. Copernicus Emergency Management Service 2019 European Union), EMSR342.
200
https://emergency.copernicus.eu/mapping/list-of-components/EMSR342 (accessed on 09 April 2019).
201
© 2019 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access
202
article distributed under the terms and conditions of the Creative Commons Attribution
203
(CC-BY) license (http://creativecommons.org/licenses/by/4.0/).
204
... GPM data are used to provide information regarding the accumulated rainfall over the past 12, 24, 48, 72 and 96 h, while GFS data are used to provide the forecasts for the upcoming 12, 24, 48, 72 and 96 h. The system was tested and validated over several case studies, highlighting good alerting capabilities [7][8][9][10][11]. However, some types of rainfall events (such as short-duration, very localized convective events) can undermine its capacity to detect extreme rainfall events [7]. ...
Article
Full-text available
In this work, we describe the integration of Weather and Research Forecasting (WRF) forecasts produced by CIMA Research Foundation within ITHACA Extreme Rainfall Detection System (ERDS) to increase the forecasting skills of the overall early warning system. The entire workflow is applied to the heavy rainfall event that affected the city of Palermo on 15 July 2020, causing urban flooding due to an exceptional rainfall amount of more than 130 mm recorded in about 2.5 h. This rainfall event was not properly forecasted by meteorological models operational at the time of the event, thus not allowing to issue an adequate alert over that area. The results highlight that the improvement in the quantitative precipitation scenario forecast skills, supported by the adoption of the H2020 LEXIS computing facilities and by the assimilation of in situ observations, allowed the ERDS system to improve the prediction of the peak rainfall depths, thus paving the way to the potential issuing of an alert over the Palermo area.
Chapter
The Extreme Rainfall Detection System (ERDS) is an early warning system (EWS) developed for the monitoring and forecasting of rainfall events on a global scale. Within ERDS the near real-time rainfall monitoring is performed using the Global Precipitation Measurement data, while rainfall forecasts are provided by the Global Forecast System model. Rainfall depths determined on the basis of these data are then compared with a set of rainfall thresholds to evaluate the presence of heavy rainfall events: in places where the rainfall depth is higher than a rainfall threshold, an alert of a severe rainfall event is issued. The information provided by ERDS is accessible through a WebGIS application (http://erds.ithacaweb.org) in the form of maps of rainfall depths and related alerts to provide immediate and intuitive information also for nonspecialized users. This chapter is intended to describe the input data and the extreme rainfall detection methodology currently implemented in ERDS. Furthermore, several case studies (2019 Queensland flood event, 2017 Atlantic hurricane season, and 2017 Eastern Pacific hurricane season) are included to highlight the strengths and weaknesses of this EWS based on global-scale rainfall datasets.
Chapter
A timely identification and monitoring of flood events by means of Earth Observation (EO) data is, nowadays, increasingly feasible thanks to recent advances achieved in remote sensing and hydrological process simulations. Despite the notable progress in these fields, a considerable effort will still be required to reduce the intrinsic inaccuracies of these types of approaches. The coarse spatial and temporal resolution of satellite measurements (compared to the one that characterizes in-situ instruments), in fact, continues to require a local-scale validation. Taking into account pros and cons of the approaches based on remotely-sensed data, this chapter reviews some of the most relevant open-access techniques, products, and services that research and academic institutes are currently providing for the detection and the near real-time monitoring of extreme hydrometeorological events.
Article
The Normalized Difference Water Index (NDWI) is a new method that has been developed to delineate open water features and enhance their presence in remotely-sensed digital imagery. The NDWI makes use of reflected near-infrared radiation and visible green light to enhance the presence of such features while eliminating the presence of soil and terrestrial vegetation features. It is suggested that the NDWI may also provide researchers with turbidity estimations of water bodies using remotely-sensed digital data.
GPM IMERG Early Precipitation L3 Half Hourly 0.1 Degree x 0.1 Degree V05
  • G J Huffman
Huffman, G.J. GPM IMERG Early Precipitation L3 Half Hourly 0.1 Degree x 0.1 Degree V05; Goddard Earth 185
Improving an Extreme Rainfall Detection 187 System with GPM IMERG data. Remote Sens
  • P Mazzoglio
  • F Laio
  • S Balbo
  • P Boccardo
  • F Disabato
Mazzoglio, P.; Laio, F.; Balbo, S.; Boccardo, P.; Disabato, F. Improving an Extreme Rainfall Detection 187 System with GPM IMERG data. Remote Sens. 2019, 11, 677. doi.org/10.3390/rs11060677
  • U Schneider
  • A Becker
  • P Finger
  • A Meyer-Christoffer
  • B Rudolf
  • M Ziese
Schneider, U.; Becker, A.; Finger, P.; Meyer-Christoffer, A.; Rudolf, B.; Ziese, M. GPCC Monitoring 5. http://www.bom.gov.au/jsp/awap/rain/index.jsp (accessed on 17 February 2019).
The use of the Normalized Difference Water Index (NDWI) in the delineation of open 194 water features
  • S K Mcfeeters
McFeeters, S.K. The use of the Normalized Difference Water Index (NDWI) in the delineation of open 194 water features. Int J Remote Sens. 1996, 17:7, 1425-1432. doi.org/10.1080/01431169608948714