You are on page 1of 6

Quantitative Data Analysis: CRASAR Small Unmanned Aerial

Systems at Hurricane Harvey


Odair Fernandes1, Robin Murphy1, Justin Adams2, and David Merrick3

Abstract— This paper summarizes the 112* flights by nine managed all air asset requests and allocated either manned or
models of small unmanned aerial systems at Hurricane Harvey unmanned assets.
and presents four recommendations for further research and
development. This was the sixth reported deployment of sUAS This paper discusses only the portion of the quantitative
for a hurricane or flooding disaster by public officials during data collected and analyzed for small UAS at Harvey. It does
the immediate response and recovery phases. This deployment not cover the equally important ethnographic data collected
is particularly interesting because it was the largest number of about crewing and airspace management. This paper will be
flights up to that time, was conducted by a county government organized as follows: first, related work in Sec. II, followed
that already had sUAS and a sUAS policy, had a suite of 14 by overview of the hurricane in Sec. III. Second, mission
models of sUAS to choose from, and used sUAS for four types and sUAS usage in Sec. IV. Then, platforms in Sec. V.
different mission types. This paper analyzes the quantitative Sec. VI presents the preparedness and response phase
data from the flights in terms of missions, objectives, choice of recommendations and concludes the paper.
sUAS, choice of control style, manpower, flight duration,
operations tempo, and data products. The data is also II. RELATED WORK
considered by phase, showing that First Person View was used Hurricane Harvey is the sixth known deployment of
exclusively for the preparedness and response phases (the first sUAS for a federally declared hurricane or flooding disaster
8 days) and preprogrammed mapping flights were requested
by public officials in the US during the immediate response
only during the infrastructure recovery phase. In addition to
the findings, the paper makes four recommendations for
and recovery phases. Note, the immediate response and
additional research. The Harvey datasets are available to the recovery phases are different than humanitarian recovery;
scientific community for further analysis. the missions focus on the public good in terms of life safety,
mitigation of hazards, and restoration of critical
I. INTRODUCTION infrastructure. CRASAR has responded with small UAS to
two prior hurricanes: Katrina (2005, State of Florida
Hurricane Harvey is estimated to be the second most
Emergency Response Team)[1] and Wilma (2005, Fort
costly Hurricane in the history of the United States. The
majority of that damage stemmed from its second landfall in Myers Harbor)[2]. It has responded to four other flooding
events: the Oso Mudslides (2014, Snohomish County Office
the Houston area over August 27-28, 2018, which flooded
Harris and Fort Bend counties. On August 25, The Fort Bend of Emergency Management)[3, 4], Texas Memorial Day
Floods (2015, Wemberly County Office of Emergency
County Office of Emergency Management (FBCOEM)
requested the assistance of the Center for Robot-Assisted Management)[4], Louisiana April Floods (2016, Tangipahoa
and Washington Parishes Office of Emergency
Search and Rescue (CRASAR). On August 26, CRASAR
Management)[5], and Texas 2015 Floods (2015, Fort Bend
provided two types of robots. One was an Hydronalix
County Office of Emergency Management).
EMILY marine surface vehicle carrying a side scan sonar to
attempt to estimate the volume of water in key rivers and Numerous other volunteer teams deployed small UAS
streams. This was not successful. The other was a fleet of to Texas for Harvey and the metro-Houston region but had
small unmanned aerial systems (sUAS), which were not deployed at the request of an agency with incident
successfully used from August 26 through September 4 when command authority. Volunteer flights were stopped in Harris
the infrastructure recovery phase ended and citizens were County due to violations of Texas privacy laws and in Fort
allowed to return home. The flights satisfied all mission Bend County due to violations of the Temporary Flight
requests from FBCOEM incident command, which formally Restriction. Note that Fort Bend County had obtained
sufficient resources with CRASAR and did not need self-
deployed assets.
This work was supported in part by the NSF grant 1762137. Small UAS were also used two weeks later by the State of
1
Odair Fernandes and Robin Murphy are with the Department of Florida for the Hurricane Irma disaster in Florida. The
Computer Science and Engineering, Texas A&M University, College damage from that hurricane was primarily due to wind with
Station, Texas 77843, USA odair_fernandes@tamu.edu,
murphy@cse.tamu.edu little flooding or storm surge. CRASAR also participated in
2
Justin Adams is with CRASAR, College Station, Texas 77843, USA the Hurricane Irma response and recovery.
jadams@crasar.org
3
David Merrick is with Florida State University, 644 Bellamy, Main III. OVERVIEW OF HURRICANE HARVEY
Campus, dmerrick@em.fsu.edu
*Includes the aborted sortie on August 29th The CRASAR team was deployed to Fort Bend County
from August 25 to September 4, therefore they used the
following Response Timeline: Before (August 25-26),
Incident (August 27-28), Response (August 29 – September IV. MISSION TYPES AND SUAS USAGE
1), and Recovery (September 2-4). Although, the hurricane
Harvey hit the Houston area on August 27-28, the first This section discusses the 7 basic mission types that are
CRASAR flight for the Fort Bend County Office of performed in natural disasters. In addition, it describes the 4
Emergency Management (FBCOEM) was on August 26 mission types executed in Hurricane Harvey deployment
(Before or Preparedness Phase). along with their respective primary objectives.

There were 112* flights (sorties) for 56* mission There is an online class called Small Unmanned Aerial
objectives and targets. All flights generally composed of a 2- Systems for Emergency Managers where the instructors go
3 people team (CRASAR pilot team and a County expert). In over the 7 general mission types that have appeared and
total, there were 10 pilots that served as pilot in command documented in the use of small UAS around the world since
(PIC), 3 backup pilots (not PIC), and 2 data managers from 2005. To note that each of these types have different flying
Florida State, Kovar Associates LLC, Lone Star UASC, style, data products, post-processes, and set of expectations.
Texas A&M, and USAA. Fig. 1 shows the mission types generally used in the different
types of natural disasters. These 7 basic mission types are the
Although there were 14 different UAS models available, following:
only 9 models were used due to the personal choice of the
PIC. These were Air Robot 200 (AR200), DJI Mavic Pro, 1. Strategic Situation Awareness (SA), Survey, and
DJI M600 Pro, DJI Phantom 4 Pro, Insitu Scan Eagle, Intel Reconnaissance: It consists of multi-purpose scan of
Falcon 8, Parrot Disco, PrecisionHawk Lancaster 5, and 3DR the area. It could be either Remote Presence or
Solo. In addition, the Air Robot 180, DJI Inspire, DJI Mapping. In the case of Remote Presence, video and
Phantom 3 Pro, Precision Hawk M100, and UAUSA stills are taken, otherwise, if Mapping, 360 views of
Tempest as well as thermal and multispectral packages area (images à master ortho) are taken and does not
complete the 14 different models available. Therefore, the have to be geo-referenced. This mission type will
pilots were not forced to use any particular platform, on the typically use pre-programmed flights and
contrary, they had a large variety of UAS well equipped to photogrammetrics.
choose from. 2. Detailed or Structural Inspection: Examining built
This CRASAR UAS deployment provided innumerous structures for more than general awareness, which are
benefits to the county rescue teams. It not only allowed the often viewpoint dependent and require more elevation
rescue operations to be conducted faster, but also more views than plane views. This type of mission is
efficiently and safer. The ability to use these small UAS also generally performed 10 feet (3.048 meters) away from
allowed to get ahead in the damage and debris assessment, the structure due to poor GPS signal, wind shear, for
which enabled a faster restore of services, contract cleanup, the fact that some DJI platforms have (large) obstacle
and overall faster economy recovery. In addition, this avoidance mechanisms. In this case, it is better to have
deployment permitted monitorization of critical an expert on the team in FPV to re-direct as needed.
infrastructures under flood load, positive economic impact 3. Ground Search: Generally, it is a search for a
from opening bridges sooner, and planning support from data missing person, object, or scene that is on the ground.
that showed the river level. The UAS videos recorded helped For this mission type, high resolution images (not
to deter citizens from trying to re-enter neighborhoods video) are needed, thermal may not be a silver bullet,
flooded and filled with sewage. Finally, it is very important but must be geotagged. In addition, the altitude and
to mention that it helped establish a broad county government Field of View (FOV) are required.
and citizen support of the use of sUAS.
4. Water Search: Used in marine incident (though may
The FBCOEM Harvey data is useful for first responders be combined with ground search in a boating accident
for multiple reasons. One is that it was a record setting on a lake). It requires high resolution images (not
number of flights and diversity of missions for all ESFs over video) and thermal essential and may be the most
suburban and rural areas. In addition, the flights occurred reliable means to find a survivor in water. As in
before, during incident, and then through the response as well Ground Search, it must be geotagged and the altitude
as recovery phases. To note that, the pilots had access to a and FOV are also required.
wide variety of fixed-wing platforms (Parrot Disco, Scan
Eagle, and PrecisionHawk Lancaster 5) and rotorcrafts 5. Debris, flood estimation, and damage: It is more
platforms (AR200, DJI Mavic Pro, DJI M600 Pro, DJI detailed than general recon and more focused on
Phantom 4 Pro, Intel Falcon 8, and 3DR Solo). Last but not documented boundaries of event, counting houses,
least, it is important to mention that all requests for air relating current view to prior condition. Consider
resources went through FBCOEM, which already was fixed-wing platforms or manned assets.
knowledgeable about sUAS, owned drones, had used drones 6. Tactical Situation Awareness: This mission type
in 2 prior federally declared disaster floods, had thought helps the rescue team asses the condition of roads or
about manpower, and already had a drone policy in place. bridges ahead and decide where to go from their
All in all, the missions, the use of external pilots, the current location. sUAS provides on demand SA when
OEM manpower allocation, and a choice on platforms were a team needs.
informed decisions and the positive results were by planning
and design, not by luck.
7. Delivery: Facilitate delivery of material to any phase), there was only debris/damage/flood estimation as
location. Generally, only used in flooding situation. well as during and after the incident, then it starts moving to
The ad hoc delivery is dangerous. Inspection, with a little bit of Strategic and Tactical SA. Note
that the evacuation order started on August 26th and started to
be released around September 1st and 2nd. In addition, note
that there is a big gap from August 27th to 30th due to the
heavy rain and clouds in the area, which made flying very
difficult.

Figure 1. Mission Types vs. Natural Disasters

Four of the seven mission types were requested by


FBCOEM: Figure 3. Number of Missions by Mission Types vs Day
1. Debris/Damage/Flood Estimation: In this mission It is also important to state the importance of the real-time
type there was the classic need to map the affected experts in these missions. Their knowledge helps speed up
areas and a lot more visual assessment where the the response process since the they know the affected areas
experts were looking to see the current state in very well. In addition, when compared with the UAV pilots,
order to assess it very quickly. In addition, they the experts know exactly what to look for and inspect in
assessed the tornado impact. natural disasters like this. Therefore, although the pilots have
2. Inspection: Inspections were performed mainly by the experience of flying in different disasters, it is important
looking at bridges, buildings, levees, and dams. to have a local expert in the team in order to facilitate the
response and make real-time decisions. As observed in Fig. 3,
3. Strategic SA/Recon/Survey Mapping: This type the first 7 days needed real-time experts (the missions were
of mission is used for public information. For mainly debris/damage/flood estimation – visual assessment,
instance, the county judge used the sUAS Inspection – bridge and building inspections, and Tactical
information in the daily reports to explain to the SA – route and transportation survey), then back-in-the-
citizens why the evacuation orders were still in office experts and no experts (Strategic SA) since the
place. missions were more focused on mapping, public information,
4. Tactical SA: This mission was executed not only and inspection of levees or dams. Note that only public
for overwatch by the teams that were going to information missions were general (for situation awareness),
neighborhoods to prepare, but also to help rescue the other were very specific and focused.
people and for route or transportation survey in The number one mission type used was Inspection (64
order to ease help get the responders to the right sorties), followed by debris/damage/flood estimation (39
locations. sorties), then Strategic SA (5 sorties), and finally Tactical SA
Fig. 2 shows these 4 mission types performed at Harvey, (4 sorties). Table I shows more in detail the number of sorties
with the specific objectives that required real-time expertise for each mission type and respective primary objective.
(mission specialist) in bold. Some of these missions require a Concerning the Inspection mission type, the Levee or Dam
specialist because there are aspects that a regular sUAS pilot primary objective had a larger number of sorties due to
can do. For instance, in Visual Assessment objective type, it mapping missions where there were many flights and images.
was crucial to have an expert that knew the map of the area However, bridge inspection was mainly to know the state of
and what to look for specifically (landmark). This gives the the river because of the transportation going into the Fort
ability to compare the current state of the area to how it Bend suburban county. For debris/damage/flood estimation,
looked before the disaster. The same thing goes for bridge visual assessment was very important, in particular for the
and building inspection and rout or transportation survey. ability to look into the flood inundation maps and checking
the status of the area. Finally, there were a few sorties for
Strategic and Tactical SA.

TABLE I. SORTIES BY MISSION TYPE

Figure 2. Mission Types and Specific Objectives

Fig. 3 shows the number of mission by mission type by


day and the respective response phase (before, incident,
response, and recovery phases). In the beginning (before
* Includes the aborted sortie

V. PLATFORMS
FBCOEM used 9 different UAS models in the Hurricane
Harvey deployment and these were chosen personally by the
PIC. These were the Air Robot 200 (AR200), DJI Mavic Pro,
DJI M600 Pro, DJI Phantom 4 Pro, Insitu Scan Eagle, Intel
Falcon 8, Parrot Disco, PrecisionHawk Lancaster 5, and 3DR
Solo. Note that the Scan Eagle is a bit bigger than the other
platforms, providing a bigger area coverage, higher altitudes,
and hours in air. The other 8 platforms (120 m AGL or less)
have a smaller area of coverage, lower altitudes, and 0.5-1.5 Figure 5. Platform Type vs Day
hours in air.
Fig. 4 shows the number of flights flown by each
platform by day and with the respective response phases. The
DJI Mavic Pro was by far the preferred UAS to use with a
total of 78 of 112 sorties. In addition, the little $1000 DJI
Mavic Pro was used in every mission type and for every data
product (images, videos, and map making).
One interesting point about the platforms is that rotorcraft
platforms (AR200, Falcon 8, M600 Pro, Mavic, Phantom 4,
and Solo) were the exclusive platform type used in the first 3 Figure 6. Control Type vs Day
phases (before, during the incident, and response) of the
deployment. The fixed-wing platforms (Disco, Lancaster 5, As mentioned earlier, FPV and rotorcraft flights were
and Scan Eagle) were not used until September 2, even preferred during the response time with an average sortie
though they were available at all times. Fig. 5 shows the time of 8min:58sec and an average mission time of
usage of each platform type by day. Another remarkable 13min:52sec (average of 1.2 sorties per mission).
aspect was the fact that all the rotorcraft flights in the first 3 Unfortunately, some flight times were not consistently
phases were flown in FPV even tough PP controls and
recorded, so they were estimated by means timestamp of the
mapping were available. Therefore, it was all about the
experts wanting to make the decisions in real time, so FPV first and last data product taken. In addition, these
was used until the recovery operations requested mapping. estimations made the average time off by approximately 2-3
Fig. 6 shows the relationship between control type and day. minutes for take-off and landing, pushing it up to
All in all, the FPV and the rotorcraft were preferred during approximately 12-minute average sortie time. Therefore, it is
the response because the experts needed the sUAS for rapid safe to say that the platforms do not have to be in the air for
decision making (not for mapping), need of short flights due more than 15 minutes or stay in the station for 1.5 hours.
to “I will know it when I see it” purposeful perception, the Based on the average sortie time, the sorties are fast, short,
operations tempo are intense that spending extra 20 minutes and to the point. Unfortunately, it is not possible to rate the
for mapping is not time effective, the ability to hover and flight times since the data for the average area covered by
watch the video in order to estimate flow rates, extremely each flight was not collected, but all flights were within the
limited landing zones, generally 60 meters AGL, and visual line of sight. Table II shows the average sortie times
orthomosaics are large capacity files, which makes it hard to (known, estimated, and know + estimated sortie times) by
email or transfer. day for rotorcraft using FPV control. In addition, having the
experts also helped on the op-tempo because they would be
precise on what needs to be recorded, capturing less than
75% of the sortie. The experts not only were helpful when
the pilots needed to record the sorties, but their knowledge
of the area also helped to reach the different mission
locations, which would take 2-4 hours in the 32 Km radius
area that had to be analyzed. Fig. 7 shows the location of the
56 missions deployed in the Houston area.

TABLE II. AVERAGE SORTIE TIME BY DAY

Figure 4. Platform vs Day


TABLE IV. NUMBER OF VIDEOS FOR MISSION TYPE BY DAY

VI. CONCLUSIONS AND RECOMMENDATIONS


Based on the data presented above, there are four
recommendations for robotics developers to consider in
optimizing the use of sUAS for hurricanes and flood events.

Figure 7. The 56 Mission Locations in the Houston Area


The data indicates that First Person View (FPV) control
is essential for the initial life saving and disaster mitigation
One aspect to consider when talking about the large phase of an event rather than mapping. Therefore, sUAS
areas that need to be analyzed is the wireless streaming. should always provide FPV.
Fortunately, most of the wireless infrastructure of the area Related to the First Person View (FPV) control finding,
was intact, so there was no problem with the incorporation of experts is critical for the response phase.
communication. However, it is important to know that if Response is all about the expert seeing at a distance; the use
streaming is needed, this would be the kind of coverage the of sUAS is not a reduction in manpower or substitution, it is
wireless network would have to support. an amplifier and accelerator. As a result the pilot may not be
There were 3 types of data products (images, videos, and the expert. However, the expert may have to be co-located
orthomosaics) generated in the flights since they were with the pilot because high bandwidth wireless streaming
crucial for experts to make the necessary rapid decisions. over a 32 km radius is unlikely. Therefore the human-robot
interaction between the pilot-expert and specialized
Tables III and IV shows the number pictures and videos
interfaces for expert, such as [6], are needed.
taken in each type of mission and primary objective. Based
on the tables, there were mainly pictures (getting still image The data showed that inexpensive rotorcraft platforms
of how things were before the hurricane hit) in the Before were able to conduct all missions and with flight durations
Phase, pictures and videos (get the natural viewpoints of on the order of 12 minutes. The safety constraints of
what’s going on in the affected areas) for the Response maintaining visual line of sight in densely trafficked
Phase, and mapping (get the master images and work over airspaces suggest that beyond visual line of sight will be
longer distances) for the Recovery Phase. Approximately, rare. Therefore, research for flooding applications should
150GB of data was generated by the missions with 85GB not focus on creating long endurance platforms but rather
belonging to the DJI Mavic Pro. Since there was a large on the data management.
quantity of data generated, it is important to have dedicated The 112 flights produced over 150GB of data products,
data managers to handle this work because the GIS branch with a daily average of 3.10GB. The CRASAR team had a
or manager will not be able to absorb extra work. In the person devoted to data management and still could not keep
Hurricane Harvey deployment there were dedicated data up with maintaining chain of custody, backups, archiving,
managers to handle the backups, renaming files, annotating, and curation. Therefore, research into the informatics
transferring to agency, editing, snipping, uploading, associated with the unmanned systems is warranted.
monitoring, and download mapping. Therefore, keep in The Harvey datasets are available to the scientific
mind that data handling will impact manpower. community for further analysis at <http://hrail.crasar.org>.
TABLE III. NUMBER OF PICTURES FOR MISSION TYPE BY DAY
ACKNOWLEDGMENT
The CRASAR deployment was all volunteer and the
analysis was supported in part by a grant from the National
Science Foundation 1762137. The authors would like to thank the Fort
Bend County Office of Emergency Management, especially Jeff Braun,
Lachlan Mullen, and Adam Wright.

REFERENCES

[1] K. Pratt, R. Murphy, S. Stover, and C. Griffin, "CONOPS and


Autonomy Rcommendations for VTOL SUASs Based on
Hurricane Katrina Operations," Journal of Field Robotics, vol.
26, pp. 636-650, 2009.
[2] R. Murphy, E. Steimle, E. Griffin, C. Cullins, M. Hall, and K.
Pratt, "Cooperative Use of Unmanned Sea Surface and Micro
Aerial Vehicle at Hurricane Wilma," Journal of Field Robotics,
vol. 25, pp. 164-180, 2008.
[3] R. R. Murphy, B. A. Duncan, T. Collins, J. Kendrick, P.
Lohman, T. Palmer, et al., "Use of a Small Unmanned Aerial
System for the SR-530 Mudslide Incident near Oso,
Washington," Journal of Field Robotics, vol. online view, pp.
n/a-n/a, 2015.
[4] R. Murphy, J. Dufek, T. Sarmiento, G. Wilde, X. Xiao, J. Braun,
et al., "Two case studies and gaps analysis of flood assessment
for emergency management with small unmanned aerial
systems," in Safety, Security, and Rescue Robotics (SSRR), 2016
IEEE International Symposium on, 2016, pp. 54-61.
[5] R. R. Murphy, "Emergency Informatics: Using Computing to
Improve Disaster Management," IEEE Computer Magazine,
special issue on Emergency Response, vol. 49, pp. 19-27, 2016.
[6] J. M. Peschel and R. R. Murphy, "On the human–machine
interaction of unmanned aerial system mission specialists," IEEE
Transactions on Human-Machine Systems, vol. 43, pp. 53-62,
2013.

You might also like