Natural Hazards and Earth System Sciences
Open Access
Atmospheric Chemistry and Physics
Open Access
Nat. Hazards Earth Syst. Sci., 13, 669–677, 2013 www.nat-hazards-earth-syst-sci.net/13/669/2013/ doi:10.5194/nhess-13-669-2013 © Author(s) 2013. CC Attribution 3.0 License.
cess
Geosciences
G
Measurement Techniques
E. Schnebele and G. Cervone
Open Access
Improving remote sensing flood assessment using volunteered Atmospheric geographical data
Dept. of Geography and Geoinformation Science, George Mason University, 4400 University Drive, Fairfax, VA, USA Received: 6 September 2012 – Published in Nat. Hazards Earth Syst. Sci. Discuss.: – Biogeosciences Revised: 19 November 2012 – Accepted: 23 January 2013 – Published: 19 March 2013
Open Access Open Access Open Access
The ability to produce accurate and timely flood assessments is a critical safety tool for flood mitigation and response. Several methodologies have been developed to assess the risks associated with flooding by using ground measurements such as precipitation, water flow or level (e.g. Richter et al., 1998; Apel et al., 2006). Satellite remote sensing data have been utilized for flood assessment because of their high spatial resolution and capacity to provide information for areas of poor accessibility or lacking in ground measurements (Smith, 1997). High resolution satellite data is particularly useful for the spatial analysis of water pixels. When data before and after a flood event are available, it is possible to classify land cover change, and thus identify which areas are flooded. The Landsat satellite program has been collecting data about the Earth and its environment since the 1970s, and has been employed to monitor and mitigate the impacts of
Open Access
Introduction
flooding (Sanyal and Lu, 2004). The use of Landsat data Climate for flood assessment can be highly effective. Frazier and Page (2000) employed a supervised maximum-likelihood of the Past classification to map water bodies with Landsat Thematic Mapper (TM), with an overall accuracy of over 97 %. Although effective for detecting water, the satellite’s orbit revisit time can constrain data availability making it difficult System to create a comprehensiveEarth time series of a flood event. Cloud and vegetative cover can obscure surface Dynamicsmeasurements when utilizing optical data, often resulting in partial coverage and incomplete flood assessment. Numerous attempts have been proposed to overcome the Geoscientific limitations of remote sensing data, often by supplementing them with additional Instrumentation data to provide a more accurate and comprehensive flood assessment. MethodsLaura and et al. (1990), Townsend and Walsh (1998) have proposed the use of Data Systems RADAR remote sensing data for the assessment of floods. RADAR has the unique advantage of penetrating through canopy and clouds, and can easily distinguish water bodies from most other land coverGeoscientific types. However, RADAR data is not widely available, and usually have limited swaths with Model Development long revisit times. Efforts have been made toward increasing RADAR’s availability and accessibility. Wang et al. (2002) have proposed the integration of Landsat TM data with a digital elevation model Hydrology (DEM) and riverand gage data to create a comprehensive assessment of flood depth under forest and Earth cloud canopy. Although river gage System data is usually sparse and not universally available, especially in more remote areas; Sciences this methodology proved very robust and is routinely used for flood assessment. The research described in this paper is inspired by Wang’s methodology, where Landsat, DEM, and river gage data are used collectively in an attempt to Ocean Science improve flood analysis. The main difference consists in the Open Access
1
Open Access
Abstract. A new methodology for the generation of flood hazard maps is presented fusing remote sensing and volunteered geographical data. Water pixels are identified utilizing a machine learning classification of two Landsat remote sensing scenes, acquired before and during the flooding event as well as a digital elevation model paired with river gage data. A statistical model computes the probability of flooded areas as a function of the number of adjacent pixels classified as water. Volunteered data obtained through Google news, videos and photos are added to modify the contour regions. It is shown that even a small amount of volunteered ground data can dramatically improve results.
Open Access
Correspondence to: E. Schnebele (
[email protected]) and G. Cervone (
[email protected])
Solid Earth
Open Acces
Published by Copernicus Publications on behalf of the European Geosciences Union.
M
670
E. Schnebele and G. Cervone: Flood maps using satellite and volunteer data
fusing methodology employed in this study to integrate the different data sources. An emerging and quickly growing data source not yet fully utilized with respect to natural hazards is volunteered geographic information (VGI) (Goodchild, 2007). This general class of data, voluntarily contributed and made available, contain temporal and spatial geographical information. Data sources include pictures, videos, sounds, text messages, etc. Due to the spread of the internet to mobile devices, an unprecedented and massive amount of ground data have become available, often in real-time. Some data are geolocated automatically, while others can be geolocated by analyzing content. Although volunteered data is often published without scientific intent, and usually carry little scientific merit, it is still possible to mine mission critical information. For example, during hurricane Katrina, geolocated pictures and videos searchable through Google provided early emergency response with ground-view information. These data have been used during major events, with the capture in near real-time the evolution and impact of major hazards (De Longueville et al., 2009; Pultar et al., 2009; Heverin and Zach, 2010; Vieweg et al., 2010; Acar and Muraki, 2011; Verma et al., 2011; Earle et al., 2012; Tyshchuk et al., 2012). This work is based on a specific subset of this general class of data, namely photos, videos, and news. Volunteered photos about natural hazards have emerged as a data source during crisis and hazardous events. Liu et al. (2008); Hyv¨arinen and Saltikoff (2010); McDougall (2011); Zhang et al. (2012) show how photos from Flickr have been used to derive local meteorological information, capture and record the physical features of an event, and identify and document flood height. Recently, volunteered data have been evaluated for estimating flood inundation depth and for mapping flood extent (Poser and Dransch, 2010; McDougall, 2011). These potentially valuable, real-time data have yet to be regularly applied in large scale disaster relief situations for multiple reasons, including difficulties of authentication and confirmation, questions of quality and reliability, and difficulties associated with harvesting data from heterogeneous and non-structured sources (Flanagin and Metzger, 2008; Schlieder and Yanenko, 2010; Tapia et al., 2011). This paper proposes a new methodology that leverages data freely obtained from the internet to improve flood hazard estimation. Combining the high spatial resolution and reliability of satellite imagery with the high temporal resolution of ground data takes advantage of the strengths of both data types while allowing for mutual data confirmation. Despite the non-scientific nature of volunteered information, the integration of these data with traditional data sources offers an opportunity to include new and additional information in flood extent mapping. It is assumed that ground truth data is not available, and therefore Nat. Hazards Earth Syst. Sci., 13, 669–677, 2013
the quantitative analysis of the results discusses the changes introduced by fusing the different data layers. The novelty of this study is the development of a methodology that takes advantage of “citizens as sensors”, as discussed in Goodchild (2007), to fuse observations culled from social media with satellite and topographic data for flood assessment. A case study is presented for the May 2011 flooding of the Mississippi River. This was one of the worst floods since the Great Flood of 1927. In Memphis, TN the Mississippi River crested at 14.6 m, the highest crest since 1937, which caused the evacuation of approximately 1300 homes. The methodology was implemented using the R statistical package1 . 2
Data
2.1
Volunteered data
The data used in this study have been downloaded using the Google search engine through their photos, videos and news portal. They included sources from Flickr, YouTube, Weather Underground, Wikipedia, and abc24.com. In particular, videos (n = 6) and photos (n = 8) from the first two weeks of May 2011 which documented the flooding were selected. A list of Memphis road closures on 12 May 2011 (n = 37) was collected from an on-line news source. Some of the data contained geolocation information, while others were geolocated using the Google API. 2.2
Remote sensing data
Full-resolution GeoTIFF Multispectral Landsat ETM+ images for 2 January and 10 May 2011 are used. The data were downloaded from the USGS Hazards Data Distribution System (HDDS). Landsat data are comprised of seven spectral bands: optical (0.45–0.52, 0.52–0.60, 0.63– 0.69 µm), near-IR (0.77–0.90 µm), mid-IR (1.55–1.75, 2.09– 2.35 µm) and thermal-IR (10.40–12.50 µm) with a spatial resolution of 30 m. The images were georeferenced to UTM coordinates in ArcGIS and an area encompassing Memphis and its greater metropolitan area was selected at a scale of 1:145 000. 2.3
Digital elevation model data
A USGS Seamless Data Warehouse DEM with a 30 m resolution was used in this study. The DEM is georeferenced to UTM coordinates in ArcGIS and exported at the same 1:145 000 scale as the Landsat data (Fig. 1). 2.4
Meteorological data
Meteorological data relative to maximum daily precipitation rate and total daily precipitation were obtained from the 1 www.r-project.org
www.nat-hazards-earth-syst-sci.net/13/669/2013/
E. Schnebele and G. Cervone: Flood maps using satellite and volunteer data
671
Figure Maximumdaily dailyprecipitation precipitation rate rate and and and Fig. 2. 2: Maximum and accumulated accumulated precipitation for the period ranging from 1 April to 31 May 2011. Figure 2: Maximum daily ranging precipitation and accumulated precipitation for the period from 1rate April to and 31 May 2011.
precipitation for the period ranging from 1 April to 31 May 2011.
Figure 1: Digital Elevation Model for the region of study. 30
Fig. 1. Digital Model for the region study.of study. Figure 1: Elevation Digital Elevation Model for theofregion 2.5. River Gage Data
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30 32
2 RiverGage gageData data for the Mississippi River in Memphis32 2.5. River 34 was CPC collected from Precipitation the US Army Corps of Engineers NCEP Morphed Model (Joyce et al., River gage 3data for the The Mississippi River Memphis 4 RiverGages website. data used forin this study 2 2004)collected and Weather Underground respectively. 3436 was fromfrom the US MS126 Army (WU), Corps were collected gage located ofat Engineers longitude: 3 Figure 2 shows the NCEP daily precipitation rate (bars)study and RiverGages website. The data used forN. this 6 90.07667000 W, latitude: 35.12306000 Data were the WU accumulated precipitation (solid line) for the period wereselected collected gage(meters) MS126 located at they longitude: in from elevation format so could3638 ranging from 1 April to 31 May 2011. The acquisition time 90.07667000 latitude: 35.12306000 N.DEM. Data were 8 effectively W, be used in conjunction with the for the Figure May data(meters) is shown, and it occurs afterMS126 the 3840 selected in Landsat elevation so they could 3 shows the heightformat information for period of intense rainfall during the end of April. These effectively usedyear in conjunction the DEM. 10 for the be entire 2011. The with acquisition time for the meteorological data used todata identify appropriate dates Figure shows the height information for and MS126 January3and Mayare Landsat are indicated, they4042 the data. The Ittois the desirable a scene is 12 correspond, respectively, almostthat minimum and for the remote entire sensing year 2011. acquisition time for the maximum heights for the entire year. The river gage4244 selected after period of intense rainfall in order toand identify January and awater May Landsat data are indicated, they 14 height information is paired withalmost the DEM to deriveand the correspond, respectively, to the minimum the maximum flood extent. approximate extent. maximum water flood heights for the entire year. The river gage 4446 height information is paired with the DEM to derive the 2.5 River gage data 48 approximate flood extent. 46 16 3. Methodology River gage data for the Mississippi River in Memphis 3.1. Overview were collected from the US Army Corps of Engineers 4850 3. Methodology 18 The 3proposed methodology based on datastudy fusion RiverGages website. The datais used forthethis ofOverview different layers generated from different data sources. 3.1. were collected from gage MS126 located at longitude: 52 20 Figure 4 ◦ illustrates this integration ◦of multiple layers,50 90.07667000 W, methodology latitude: 35.12306000 were The proposed is basedor onN. theData data fusion which may have varying resolutions sparse data. The selected in elevation (m) format so they could effectively be a 254 of different layers generated from different data sources. 22 output is shown in the bottom most layer, where usedflood in conjunction with DEM. Figure 4 hazard illustrates this integration of multiple layers,of52 map isthe generated. The input consists Figure showsvarying the height information for MS126 for 456 which may3 have resolutions or sensing sparse data. The 24 different layers generated using remote data, DEM, output shown in The the bottom most a 54 the ground entireis year 2011. acquisition time layer, for the where January information etc, as normally discussed in the flood hazard map is generated. The input consists of and May Landsat are indicated, and they correspond, is 658 26 literature. Thedata novelty of the proposed methodology, different layers generated using remote sensing data, DEM, respectively, to the of almost minimum maximum water the introduction volunteered dataand as an additional layer,56 ground information etc, The as river normally discussed in the 60 28 and for their inyear. refining the hazard map.information Therefore, heights the use entire gage height although in this paper of we used specific remote sensing and literature. The novelty proposed methodology, is 588 is paired with the DEM to the derive the approximate flood the introduction of volunteered data as an additional layer, extent. 62 and their use in refining the hazard map. Therefore, 60 3 www.rivergages.com although in this paper we used specific remote sensing and
2 www.weatherunderground.com 3 www.rivergages.com
3 www.rivergages.com
www.nat-hazards-earth-syst-sci.net/13/669/2013/
DEM data, the methodology is not limited by these data types and can easily be extended to integrate additional DEM data, the methodology is not limited by these data or different sources. It is crucial for obtaining the types and can easily be extended to integrate additional most accurate measurements that the correct classification or different sources. It is crucial for obtaining the methodology is applied to each data type when creating most accurate measurements that the correct classification the flood extent layer(s). methodology applied eacha verified data type whentruth”, creating The groundis data usedto is not “ground the flood extent layer(s). but can be utilized as reliable information to assess the The ground data used is notina verified presence or absence of water specific “ground locations.truth”, It but can be utilized as reliable information compensates for the potential miss classificationtoofassess remotethe presence or absence of water satellite in specific sensing data due to resolution, orbitlocations. limitation, It compensates fordata the potential miss classification of remote cloud cover, or acquisition problems. Furthermore, sensing dataof due to resolution, the volume the data alone as a satellite function orbit of timelimitation, can be an indication of data the geospatial rate of progression of the cloud cover, or acquisition problems. Furthermore, event, and can helpdata prioritize to specific areas. the volume of the alone response as a function of time can be The methodology consists of arate three process: of the an indication of the geospatial of step progression
event, and can helpofprioritize response to specific areas. 1. Identification Flood Extent. The methodology consists of maps. a three step process: 2. Generation of flood hazard
34
36
38
40
42
on vo we ob th loc po a as
44
46
48
3. Identification Ground data integration. 1. of Flood Extent. 50 Fig. 3. Year 2011 water height profilemaps. for Mississippi River at 2. Generation of hazard Figure 3: Year 2011offlood water 3.2. Identification Flood height Extentprofile for Mississippi River at Memphis, TN. 3. Ground Memphis, TN. data integration.
Different methodologies can be used to identify the 52 extent of water over the geographical region of interest. 3.2. Identification of Flood Extent The each goal of this to stepbe is to generate This one orprobability more maps map pixel flooded. Different methodologies can beregions used where to identify 3for Methodology 54 using the input layers which identify water isthe is generated byover applying a kernel region density smoothing extent of water the geographical of interest. detected. The task is method-independent, and it can use operation 2Dsuited then normalizing 3.1 Overview The goal ofover thisthe isdata, to generate onebyor more mapsthe 56 any method that isstep best forand a particular combination result. Let (wlayers . . , w1 xnregions ) be weighted samples using theand input where water is 1 x1 , wwhich 1 x2 , . identify of data location. The proposed is based onare theemployed data fusion detected. The task is method-independent, and it can drawn from a methodology distribution with an unknown density f , the 58 In this article, two different methods touse of different layers different data sources. any method that is generated best suited for particular combination goal is to estimate the shape of athis function. general identify flood extent. The firstfrom involves the use of The remote of data and location. 4 illustrates this integration of multiple layers, Figure sensingdensity data andestimator machine learning classification, andwhich the kernel is 60 Inhave this article, two different methods areThe employed second thevarying use of aresolutions DEM and or river gagedata. data. may sparse output isto
n identifyinflood extent. Thew(x first−where involves the use of remote shown the bottom most layer, hazard map 1 X xi ) a flood 62 f (x) = K( )classification, 3.3. Generation ofmachine Flood Hazard Maps sensing dataThe and learning and the (1) is generated. input consists of different layers generated nh i=1 h After orofmore floodand extent maps generated,etc., a second theone use a DEM river gageare data. using remote sensing data, DEM, ground information, 64 hazarddiscussed map is created by computing the probability asflood normally in the literature. The novelty of the where K is the kernel function, h is the bandwidth, 3.3. Generation of Flood Hazard Maps is theweighting introduction of volunteered 3 proposed 10 and w ismethodology, a user selected scalar. The weight 66 62 After one or morelayer, flood and extent maps are generated, data as an additional their use in refining the a w describes the importance of a particular observation, flood hazard map is created by computing the probability 12 or the confidence associated with the flood extent map. 68 3 In this Nat. application, usingSyst. a weighted kernel 2013 function Hazards Earth Sci., 13, 669–677, 14 is paramount because ground observations cannot be considered “ground truth” proper, since volunteered 16 geographical data carry intrinsic uncertainties due to their
co reg Ke stu
3.4
ma of th sp flo
me jus so an cla wh Th as vid we pic bo
4.
672
E. Schnebele and G. Cervone: Flood maps using satellite and volunteer data
28
Volunteered Data 30
DEM 32
Landsat Band 1 34
Landsat Band 2 Landsat Band 3
36
Landsat Band 4
38
Landsat Band 5
40
Landsat Band 6
42
Landsat Band 7
44
46
48
50
Flood Hazard Map 52
Fig. 4. The proposed methodology fuses several layers to generate Figure 4: The proposed methodology fuses several layers to generate a flood hazard map. 54 a flood hazard map.
56
machine although learning classification, hazard For map.theTherefore, in this paper 4wecontrol used 2 areas of roughly the same size are 2 over the specific remote sensing and DEM data,identified, the methodology is 58 river as examples of water pixelsbeand 2 over not Mississippi limited by these data types and can easily extended different additional regions with water pixels as counter-examples. to4 integrate ornodifferent sources. It is crucial for 60 Landsat multispectral data relative to these regions are obtaining the most accurate measurements that the correct 6 used as training events by the decision tree classifier. The 62 classification methodology is applied to each data type when learned tree is then used to classify the remaining water creating the flood extent layer(s). 8 pixels in the scene. This process is repeated for both the 64 The ground data used is not “ground truth”,5b,d. but January and May scene, anda verified is illustrated in Figure can be utilized as reliable information to assess the presence 10 About 1% of the total number of pixels are used 66 as training pixels in (events), the remaining 99% are or absence of water specificand locations. It compensates 12 classified according to the induction tree generated. for the potential misclassification of remote sensing data 68 due to resolution, satellite orbit limitation, cloud cover, or Flood Hazard Maps Furthermore, and Ground Data Integration data4.3. acquisition problems. the volume of the 70 14 The as methodology is employed data alone a functiondescribed of time in canSection be an3.3 indication of to generaterate flood mapsofusing both the and 72 the geospatial of hazard progression the event, andDEM can help 16 Landsat pixel classifications. The goal is to assign each prioritize response to specific areas. pixel a probability of being part of the flooded area. 74 The methodology consists of a three step process: 18
Figure 6a,c show the probability contour lines for January
May, respectively. 1. and Identification of flood extent.
76
Additional data with ground information is then to refineofthe January May flood hazard maps 78 2. used Generation flood hazard and maps. 22 (Figure 6b,d). The images show both the location and of the information (Video, Photos, News), 80 3. type Ground dataground integration. 24 and the resulting probabilities when these data are taken account. The imposed on both the January 3.2 into Identification ofdata floodare extent 26 and May hazard maps, although all ground data are 20
layers which watergenerated is detected. collected from identify the Mayregions flood. where The image for The task is method independent, and it can use any method that is the May flood, (Figure 6d), shows modifications to the best suited a particular of of data and location. flood hazardformap after thecombination incorporation the ground In this are employed data. The article, ground two datadifferent are also methods incorporated with the to Januray image 6b) to the illustrate identify (non-flood) flood extent. The (Figure first involves use ofhow remote asensing preliminary hazard map could be generated if current data and machine learning classification, and the satellite data arethe notuseavailable. both second involves of a DEMInand riverinstances, gage data.the addition of supplemental information in the form of volunteered groundof data alters the flood map by expanding 3.3 Generation flood hazard maps the area of possible inundation and by adjusting pixel values. After one or more flood extent maps are generated, a flood The pixel summarized in Table 1 and for hazard map classifications is created byarecomputing the probability by a histogram in Figure 7. As expected, when generally each pixel to be flooded. This probability map is generated comparing flood versus non-flood scenes as in Figure 6a,c, by applying a kernel density smoothing operation over more pixels have a higher probability (60-100%) of being the 2-Danddata, and then normalizing result.of Let flooded less pixels have abylower probabilitythe (0-40%) (w1 x1flooded , w1 x2 , .in. .the , w1May xn ) be weighted samples drawn tofrom being (flooded) image as compared a distribution with an unknown the January (non-flooded) image. density f , the goal is to estimate this are function. The general kernel When the the shape groundofdata incorporated into the hazard (Figure 6b,d), a spatial analysis shows density maps estimator is noteworthy changes. of ground data The incorporation n w (x xi ) May flood hazard map 1 X yields enhancements to −the K , (1) f (x) = (Figure 6d) which are hevident by the progression of nh i=1 contour lines and reclassifications of pixels. Higher wherecontour K is thelines, kernel function,a hgreater is the bandwidth, value indicating probability and of w ais region flooded, progress the northeast, a user being selected weighting scalar.toward The weight w describes where the majority of groundobservation, informationorare the importance of a particular thelocated. confidence Examining the differences between the two May scenes associated with the flood extent map. In this application, in Table 1, the percentage of pixels classified as having a using a weighted kernel function is paramount because low probability (0 - 20%) of being flooded as well as the ground observations cannot be considered “ground truth” pixels classified as having a high probability (80-100%) proper, since volunteered geographical data carry intrinsic of flooding decreases or increases 6 percentage points, uncertaintiesafter duethe to incorporation their generally non-scientific nature. respectively, of ground data. These Therefore, using different values ofboth w properly assigns levels changes illustrate that although the DEM/river of confidence to the various observations. gage and Landsat classification techniques can be highly accurate in identifying of flooded areas, additionspecific of real- and The identification weight w istheproblem time on the ground data, verifying the presence of water on domain dependent, but most importantly, it is dependent in a specific can augment an to inundation map.concept of data quality.area, A weight is used include the Applying the layer of ground data to the “significance” of the data in the algorithm and theJanuary analysis. It Hazard Map(Figure 6b) illustrates how a small amount is assumed when working with such heterogeneous data, the of real-time volunteered ground data could be integrated information might vary significantly, and therefore decisions with an historical image to identify possible flooded should be The made,number when possible, using better regions. of non-water pixels data. (0-20%) is Quantitative measurespoints to define the weight can be reduced by 7 percentage and reclassified to higher established.classes For example, when using satellite data the pixels probability (Table 1). along the 6b,d center of the swath or thoseground that are cloud Figure shows while volunteered data does free modify the flood maps, Most the amount of modification are preferred in hazard the analysis. satellite products have a isquality limitedindex by the spatial distribution of the ground associated with each pixel that can bedata. used to The evolution of theweight. contour lines, or areas of change, in set the appropriate both images are restricted to regions For volunteered data, the weightwhere may the varyvolunteered depending on data are located. This illustrates while the incorporation the characteristics of the source. For example, the volume of volunteered ground data does affect a change in both of the data can be used to assign higher weight to data flood hazard maps, the areas of change are controlled by withquantity dense and spatial coverageofand numerous observations. the distribution the volunteered data.
Higher weight can also be dependent on the source itself. For example, observations published by local news are assumed to have been validated more than points volunteered Different methodologies can be used to identify the extent 5 anonymously. Finally, there is also a subjective component of water over the geographical region of interest. The goal that can be taken into account, assigning different weight to of this step is to generate one or more maps using the input specific users or regions. Nat. Hazards Earth Syst. Sci., 13, 669–677, 2013
www.nat-hazards-earth-syst-sci.net/13/669/2013/
E. Schnebele and G. Cervone: Flood maps using satellite and volunteer data The output of the kernel smoothing is a map with contour lines illustrating the probability that specific regions are flooded. The specific methodology used for Kernel smoothing and its R implementation used in this study are described by Wand and Jones (1995). 3.4
Ground data integration
The last step consists of modifying the flood hazard map by the integration of the ground data. The nature of this data is different from the data used to generate the previous flood extent maps. It is usually comprised of sparse point data, identifying the presence or absence of flooding in a specific region. For this study, weight values are assigned experimentally. They are first equally assigned and then adjusted based on characteristics and confidence in the data source. By assuming the machine learning tree induction and the DEM/river gage approach are equally adept at classifying water, their weight values are kept constant while values of the volunteered ground data are adjusted. The flooded roads documented by local news sources are assumed more reliable than the sparse, point data of the videos and the pictures. Based on this assumption, the weights were set to 3 for the news data, and to 2 for the pictures and videos. Equal values of 1 were assigned for both the DEM and Landsat data. 4 4.1
Results Flood classification using DEM and river gage data
A DEM and river gage data are used to classify water pixels for 2 January and 10 May 2011. Pixels in the DEM with a height less than or equal to the river gage height are set as water pixels. Specifically, heights of 56 m and 70 m are used for January and May, respectively. Figure 5a, c show the areas identified as water for January and May dates, imposed over the DEM. The scale information is the same as in Fig. 1. 4.2
Flood classification using machine learning tree induction
Water pixels are identified in both the January and May Landsat images by using a machine learning tree induction classifier. Ripley (2008) describes the general rule induction methodology and its implementation in the R statistical package used in this study. For the machine learning classification, 4 control areas of roughly the same size are identified, 2 over the Mississippi river as examples of water pixels and 2 over different regions with no water pixels as counter-examples. Landsat multispectral data relative to these regions are used as training events by the decision tree classifier. The learned tree is then used to classify the remaining water pixels in the scene. This process is repeated for both the January and May scene, and is illustrated in Fig. 5b, d. www.nat-hazards-earth-syst-sci.net/13/669/2013/
673
About 1 % of the total number of pixels are used as training pixels (events), and the remaining 99 % are classified according to the induction tree generated. 4.3
Flood hazard maps and ground data integration
The methodology described in Sect. 3.3 is employed to generate flood hazard maps using both the DEM and Landsat pixel classifications. The goal is to assign each pixel a probability of being part of the flooded area. Figure 6a, c show the probability contour lines for January and May, respectively. Additional data with ground information is then used to refine the January and May flood hazard maps (Fig. 6b, d). The images show both the location and type of the ground information (video, photos, news), and the resulting probabilities when these data are taken into account. The data are imposed on both the January and May hazard maps, although all ground data are collected from the May flood. The image generated for the May flood, (Fig. 6d), shows modifications to the flood hazard map after the incorporation of the ground data. The ground data are also incorporated with the January (non-flood) image (Fig. 6b) to illustrate how a preliminary hazard map could be generated if current satellite data are not available. In both instances, the addition of supplemental information in the form of volunteered ground data alters the flood map by expanding the area of possible inundation and by adjusting pixel values. The pixel classifications are summarized in Table 1 and by a histogram in Fig. 7. As expected, when generally comparing flood versus non-flood scenes as in Fig. 6a, c, more pixels have a higher probability (60–100 %) of being flooded and less pixels have a lower probability (0–40 %) of being flooded in the May (flooded) image as compared to the January (non-flooded) image. When the ground data are incorporated into the hazard maps (Fig. 6b, d), a spatial analysis shows noteworthy changes. The incorporation of ground data yields enhancements to the May flood hazard map (Fig. 6d) which are evident by the progression of contour lines and reclassifications of pixels. Higher value contour lines, indicating a greater probability of a region being flooded, progress toward the northeast, where the majority of ground information are located. Examining the differences between the two May scenes in Table 1, the percentage of pixels classified as having a low probability (0–20 %) of being flooded as well as the pixels classified as having a high probability (80–100 %) of flooding decreases or increases 6 percentage points, respectively, after the incorporation of ground data. These changes illustrate that although both the DEM/river gage and Landsat classification techniques can be highly accurate in identifying flooded areas, the addition of real-time on the ground data, verifying the presence of water in a specific area, can augment an inundation map.
Nat. Hazards Earth Syst. Sci., 13, 669–677, 2013
674
E. Schnebele and G. Cervone: Flood maps using satellite and volunteer data
Table 1. Number of pixels classified as water. P(w)
Jan
Jan + Ground
May
May + Ground
0–20 % 20–40 % 40–60 % 60–80 % 80–100 %
126 567 (51 %) 53 718 (21 %) 21 136 (08 %) 12 660 (05 %) 35 919 (14 %)
108 932 (44 %) 57 475 (23 %) 27 601 (11 %) 18 969 (08 %) 37 023 (15 %)
114 104 (46 %) 40 190 (16 %) 21 235 (08 %) 23 260 (09 %) 51 211 (20 %)
99 368 (40 %) 30 373 (12 %) 26 997 (11 %) 29 368 (12 %) 63 894 (26 %)
(a) Water Classification using DEM for Jan 2011
(b) Water Classification using Landsat for Jan 2011
(c) Water Classification using DEM for May 2011
(d) Water Classification using Landsat for May 2011
Fig. 5. Water pixel classification using the DEM (a and c) and Landsat (b and d) and for January (top) and May (bottom) data. The background (b and d) is from Landsat band 3.
Nat. Hazards Earth Syst. Sci., 13, 669–677, 2013
www.nat-hazards-earth-syst-sci.net/13/669/2013/
E. Schnebele and G. Cervone: Flood maps using satellite and volunteer data
(a) DEM + Landsat Jan 2011
(b) DEM + Landsat + Ground Jan 2011
(c) DEM + Landsat May 2011
(d) DEM + Landsat + Ground May 2011
675
Fig. 6. Flood hazard map indicating the probability of flood in percentage using DEM, Landsat, and ground data for January 2011 (a and b) and for May 2011 (c and d).
Applying the layer of ground data to the January Hazard Map (Fig. 6b) illustrates how a small amount of realtime volunteered ground data could be integrated with an historical image to identify possible flooded regions. The number of non-water pixels (0–20 %) is reduced by 7 percentage points and reclassified to higher probability classes (Table 1).
www.nat-hazards-earth-syst-sci.net/13/669/2013/
Figure 6b, d show while volunteered ground data does modify the flood hazard maps, the amount of modification is limited by the spatial distribution of the ground data. The evolution of the contour lines, or areas of change, in both images are restricted to regions where the volunteered data are located. This illustrates while the incorporation of volunteered ground data does affect a change in both flood hazard maps, the areas of change are controlled by the quantity and distribution of the volunteered data. Nat. Hazards Earth Syst. Sci., 13, 669–677, 2013
I. C., Management, 2010. Microblogging for Crisis tion: Examination of Twitter Use in Response to a 38 Crisis in the Seattle-Tacoma, Washington, Area. ISC Table 1: Number of pixels classified as water. Hyv¨ arinen, O., Saltikoff, E., 2010. Social media as 676 E. Schnebele and G. Cervone: Flood maps using satellite and volunteer data 40 meteorological observations. Monthly Weather Rev 3175–3184. De Longueville, B., Smith, R., and Luraschi, “OMG,P., fromXie, here, P., 2004. 42 Joyce, R., Janowiak, J., G.: Arkin, I can see the flames!”: a use case of global mining location based estimates method that produces precipitation social networks to acquire and spatio-temporal on forest fires, spatial a 44 microwave infrared data data at high in: Proceedings of the 2009 International Workshop on Location resolution. Journal of Hydrometeorology 5 (3), 487– Based Social Networks, 73–80, 2009. 46 Laura, L., Melack, J., David, S., 1990. Radar detectio Earle, P., Bowden, D., and Guy, M.: Twitter earthquake detection: beneath the forest canopy: A review. Internation earthquake monitoring in a social world, Ann. Geophys., 54, 48 Remote Sensing 11 (7), 1313–1325. 708–715, 2012, Liu, S., Palen, L., Sutton, J., Hughes, A., Vieweg, S., 2 http://www.ann-geophys.net/54/708/2012/. 50 of biggerM.: picture: The emergent role of on-line p Flanagin, A. andthe Metzger, The credibility of volunteered in times ofGeoJournal, disaster. 72, Proceedings of ISCRAM 8. geographic information, 137–148, 2008. 52 P. and McDougall, K.,body 2011. Using Frazier, Page, K.: Water detection andvolunteered delineation withinformation floods. Surveying & Spatial Scie Landsat TMqueensland data, Photogramm. Eng.Proc. Rem. S., 66, 1461–1467, 54 Conference. 2000. Goodchild, Poser, M.: Citizens as sensors:D., the 2010. world Volunteered of volunteered geographi K., Dransch, geography, GeoJournal, 69,management 211–221, 2007. with application to rapid 56 for disaster Heverin, T. andestimation. Zach, L.: Microblogging for 64 crisis communication: Geomatica (1), 89–98. examination of Twitter use in response a 2009T., violent crisis 58 Pultar, E., Raubal, M., to Cova, Goodchild, M., 20 in Seattle-Tacoma, Washington in: Proceedings of and the volunteer GIS case studies: area, Wildfire evacuation Seventh International Information Systems for 60 information. Transactions inCrisis GIS Response 13 (1), 85–104. and Management Conference (ISCRAM), Seattle, Washington, Richter, B., Baumgartner, J., Braun, D., Powell, Fig. 7. Number of pixels classified as water. 2010. 62 spatial assessment of hydrologic alteration within a r Hyv¨arinen, O. and Saltikoff, E.: Social media as a source of Figure 7: Number of pixels classified as water. Regulated Rivers: Research & Management 14 (4), meteorological observations, Mon. Weather Rev., 138, 3175– 64 Ripley, B., 2008. Pattern recognition and neural net 3184, 2010. bridge J., Univ Pr.P., and Xie, P.: CMORPH: A 5 Conclusions Joyce, R., Janowiak, Arkin, 5. Conclusions 66 Sanyal, J.,global Lu, precipitation X., 2004.estimates Application of remote sen method that produces from passive with special reference to monsoon as This paper presents a new methodology fusing volunteered microwave management and infrared data at high spatial and temporal 68 Hazards 33 (2), 2004. 283–301. geographical with remote DEM data for theto fuse This paper data presents a sensing new and methodology resolution, J.Natural Hydrometeorol., 5, 487–503, Schlieder, C., Yanenko, O., 2010.ofSpatio-temporal p Laura, L., Melack, J., and David, S.: Radar detection flooding creation of hazard maps. It is shown that even a small amount volunteered geographical data with remote sensing and beneath the social forest canopy – A review, Int. J. Remote framework Sens., 11, 70 distance: a confirmation for soc of ground data points can change the flood assessment when DEM compared data fortothe creation of hazard maps. It is shown 1313–1325, In: 1990.Proceedings of the 2nd ACM SIGSPATIAL satellite and DEM data alone. The spatial Liu, S.,72Palen, L., Sutton, J., on Hughes, A., andBased Vieweg,Social S.: In search that even a small amount of ground data can change Workshop Location Networks. ACM distribution of the volunteered data limits thepoints areas where of the bigger picture: emergent role of on-line Smith, L.,the 1997. Satellite remotephoto-sharing sensing of river inu the flood assessment change is detected. when compared to satellite and DEM in times of disaster, Proceedings of the Systems 74 stage, in: and discharge: A Information review. Hydrological proc The methodology is particularly useful when data data alone. The spatial distribution of satellite the volunteered for Crisis Response and Management Conference (ISCRAM), 1427–1439. is limited or of poor quality. Additionally, the ground data 2008. data limits the areas where change is detected. 76 Tapia, A., Bajpai, K., Jansen, B., Yen, J., Giles, L., can add a time component which can help determine the McDougall, K.: Using volunteered information to map the The methodology is particularly useful when satellite the trustworthy tweet: Can microblogged data fit th change occurring between remote sensing observations. Queensland floods, in: Proceedings of the Surveying and Spatial 78 needs of disaster response and humanitarian relief o data is limited or of poor quality. Additionally, the ground Sciences Biennial Conference, 2011. Proceedings of the geographic 8th International ISCRAM Co Poser, K. and In: Dransch, D.: Volunteered information data can add a time component which can help determine Acknowledgements. Work performed under this project has been 80 1–10. for disaster management with application to rapid flood damage the change between remote sensing observations. partiallyoccurring supported by the US Department of Transportation award Townsend, P., 89–98, Walsh, estimation, Geomatica, 64, 2010.S., 1998. Modeling floodplai 202717 (RITARS-12-H-GMU, CFDA). 82 using an integrated GIS withM.:radar and optical re Pultar, E., Raubal, M., Cova, T., and Goodchild, DDynamic Geomorphology 21 and (3),volunteered 295–312.geographic The publication of this article was funded in part by the George GIS case studies: wildfire evacuation References Mason University Libraries Open Access Publishing Fund. 84 Tyshchuk, Y.,inHui, C., Grabowski, M., Wallace, W., information, Transactions GIS,13, 85–104, 2009. media andJ.,warning impacts in extreme ev Richter, B., Baumgartner, Braun, D.,response and Powell, J.: A spatial Acar, A., Muraki, Y., 2011. Twitter for crisis communication: lessons Edited by: N. Kerle assessment of hydrologic alteration within a experiment. river network, In: Proce 86 from a naturally occurring learned fromby:Japan’s tsunami disaster. International Journal Regul. of Reviewed three anonymous referees River., 14, 329–340, 45th Annual 1998. Hawaii International Conference on Sy Web Based Communities 7 (3), 392–402. Ripley,88 B.: Pattern recognition andpp. neural networks, Cambridge (HICSS). IEEE, 818–827. Apel, H., Thieken, A., Merz, B., Bl¨ oschl, G., 2006. A probabilistic University Press, 2008. Verma, S., Vieweg, S., Corvey, W., Palen, L., Martin, J Lu, X.: Application of remote in flood language modelling system for assessing flood risks. Natural Hazards 38Sanyal, (1), 90 J. andSchram, A., Anderson, K., sensing 2011. Natural References managementthe withrescue? special reference to monsoon Asia: a awareness review, 79–100. extracting situational tweets Nat. Hazards, 33, 283–301, 2004. De Longueville, B., Smith, R., Luraschi, G., 2009. OMG, from 92 emergency. Proc. ICWSM. A. and Y.: Twitter crisis communication: lessons based Schlieder, C. and Yanenko, O.: Spatio-temporal proximity and here,Acar, I can seeMuraki, the flames!: Afor use case of mining location Vieweg, S., Hughes, A., Starbird, K., Palen, L., 2010. M from to Japan’s tsunami disaster, International of social distance: a confirmation framework for social reporting, social learned networks acquire spatio-temporal dataJournal on forest fires. 94 during two natural hazards events: what twitter m Web Based Communities, 7, 392–402, 2011. in: Proceedings of the 2nd ACM SIGSPATIAL International In: Proceedings of the 2009 International Workshop on Location toLocation situational In: Proceedings Apel, H., Thieken, A., Merz, B., and Bl¨oschl, G.: A probabilistic Workshop on Basedawareness. Social Networks, ACM, 60–67,of the 28th Based Social Networks. ACM, pp. 73–80. 96 conference on Human factors in computing system modelling system for assessing flood risks, Nat. Hazards, 38, 79– 2010. Earle, P.,100, Bowden, D., Guy, M., 2012. Twitter earthquake detection: 1079–1088. 2006. earthquake monitoring in a social world. Annals of Geophysics 98 Wand, M., Jones, M., 1995. Kernel smoothing. Vol. 60 54 (6). Hall/CRC. Nat. A., Hazards Earth M., Syst. 2008. Sci., 13, 669–677, 2013 of volunteered100 www.nat-hazards-earth-syst-sci.net/13/669/2013/ Flanagin, Metzger, The credibility Wang, Y., Colby, J., Mulcahy, K., 2002. An efficien geographic information. GeoJournal 72 (3), 137–148. mapping flood extent in a coastal floodplain using Frazier, P., Page, K., 2000. Water body detection and delineation102 and DEM data. International Journal of Remote Se with Landsat TM data. PE & RS- Photogrammetric Engineering 3681–3696. 36
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30
E. Schnebele and G. Cervone: Flood maps using satellite and volunteer data Smith, L.: Satellite remote sensing of river inundation area, stage, and discharge: A review, Hydrol. Process., 11, 1427–1439, 1997. Tapia, A., Bajpai, K., Jansen, B., Yen, J., and Giles, L.: Seeking the trustworthy Tweet: can microblogged data fit the information needs of disaster response and humanitarian relief organizations, in: Proceedings of the 8th International Conference on Information Systems for Crisis Response and Management (ISCRAM), 10 pp., 2011. Townsend, P. and Walsh, S.: Modeling floodplain inundation using an integrated GIS with radar and optical remote sensing, Geomorphology, 21, 295–312, 1998. Tyshchuk, Y., Hui, C., Grabowski, M., and Wallace, W.: Social media and warning response impacts in extreme events: results from a naturally occurring experiment, in: Proceedings of the 45th Annual Hawaii International Conference on System Sciences (HICSS), IEEE, 818–827, 2012. Verma, S., Vieweg, S., Corvey, W., Palen, L., Martin, J., Palmer, M., Schram, A., and Anderson, K.: Natural language processing to the rescue? Extracting “situational awareness” Tweets during mass emergency, in: Proceedings of the ACM 2012 conference on Computer Supported Cooperative Work Companion, 227– 230, 2011.
www.nat-hazards-earth-syst-sci.net/13/669/2013/
677
Vieweg, S., Hughes, A., Starbird, K., and Palen, L.: Microblogging during two natural hazards events: what twitter may contribute to situational awareness, in: Proceedings of the 28th international Conference on Human Factors in Computing Systems, ACM SIGCHI, 1079–1088, 2010. Wand, M. and Jones, M.: Kernel smoothing, Vol. 60, Chapman & Hall/CRC, 1995. Wang, Y., Colby, J., and Mulcahy, K.: An efficient method for mapping flood extent in a coastal oodplain using Landsat TM and DEM data, Int. J. Remote Sens., 23, 3681–3696, 2002. Zhang, H., Korayem, M., Crandall, D., and LeBuhn, G.: Mining photo-sharing websites to study ecological phenomena, in: Proceedings of the 21st international conference on World Wide Web, 749–758, 2012.
Nat. Hazards Earth Syst. Sci., 13, 669–677, 2013
