Flood susceptibility analysis through remote sensing, GIS and frequency ratio model

Flood susceptibility analysis through remote sensing, GIS and frequency ratio model Papua New Guinea (PNG) is saddled with frequent natural disasters like earthquake, volcanic eruption, landslide, drought, flood etc. Flood, as a hydrological disaster to humankind’s niche brings about a powerful and often sudden, pernicious change in the surface distribution of water on land, while the benevolence of flood manifests in restoring the health of the thalweg from excessive siltation by redistributing the fertile sediments on the riverine floodplains. In respect to social, economic and environmental perspective, flood is one of the most devastating disasters in PNG. This research was conducted to investigate the usefulness of remote sensing, geographic information system and the frequency ratio (FR) for flood susceptibility map - ping. FR model was used to handle different independent variables via weighted-based bivariate probability values to generate a plausible flood susceptibility map. This study was conducted in the Markham riverine precinct under Morobe province in PNG. A historical flood inventory database of PNG resource information system (PNGRIS) was used to generate 143 flood locations based on “create fishnet” analysis. 100 (70%) flood sample locations were selected randomly for model building. Ten independent variables, namely land use/land cover, elevation, slope, topographic wetness index, surface runoff, landform, lithology, distance from the main river, soil texture and soil drainage were used into the FR model for flood vulnerability analysis. Finally, the database was developed for areas vulnerable to flood. The result demonstrated a span of FR values ranging from 2.66 (least flood prone) to 19.02 (most flood prone) for the study area. The developed database was reclassified into five (5) flood vulnerability zones segmenting on the FR values, namely very low (less that 5.0), low (5.0–7.5), moder - ate (7.5–10.0), high (10.0–12.5) and very high susceptibility (more than 12.5). The result indicated that about 19.4% land area as ‘very high’ and 35.8% as ‘high’ flood vulnerable class. The FR model output was validated with remaining 43 (30%) flood points, where 42 points were marked as correct predictions which evinced an accuracy of 97.7% in prediction. A total of 137292 people are living in those vulnerable zones. The flood susceptibility analysis using this model will be very useful and also an efficient tool to the local government administrators, researchers and planners for devising flood mitigation plans. Keywords Remote sensing · GIS · Flood vulnerability analysis · Frequency ratio · Markham river Introduction High intensity downpours in a region often lead to flood - ing in the downstream areas. Floods happen when overland * Sailesh Samanta flow of waters inundates land (Merz et al. 2010). Natural rsgis.sailesh@gmail.com disasters, like floods, are causing massive damages to natural Dilip Kumar Pal and human resources (Du et al. 2013; Youssef et al. 2011). dkpal200090@gmail.com An average of 140 million people is affected per year due Babita Palsamanta to flooding (WHO 2003). In respect of socioeconomic and mundey2@gmail.com environmental consequences, widespread flood analysis is Department of Surveying and Land Studies, The PNG very significant (Markantonis et al. 2013). Control of a flood University of Technology, Private Mail Bag, Lae, Morobe, and prevention measures are necessary to reduce potential Papua New Guinea damages to natural resources, agriculture, infrastructure etc. PNG University of Technology Campus, Private Mail Bag, (Billa et al. 2006; Huang et al. 2008). Therefore, analysis of Lae, Morobe, Papua New Guinea Vol.:(0123456789) 1 3 66 Page 2 of 14 Applied Water Science (2018) 8:66 flood susceptibility is an important task for early warning WofE and FR model are relatively new for flood vulnerabil- system, emergency services towards management strategies ity modelling, and they were widely used in other natural of prevention and mitigation of the future flood episodes hazards like landslide susceptibility mapping (Rahmati et al. (Tehrany et al. 2015). There are several comprehensive tools 2016b). Both models are almost similar and yield compara- available and used by many research organizations world- ble results for flood susceptibility mapping. Flood risk map wide, for example HAZUS, a GIS-based natural hazard through FR model was expected to be used in programmes analysis tool developed for assessing flood hazard; HEC- to reduce the flood and its damage (Lee et al. 2012). FDA, a computer program to assist crop engineers through Modelling of flood is a complex exercise where a lot of vulnerability analysis of flood risk reduction policy. RS and factors are supposed to be considered. RS technique pro- GIS techniques offer a suitable platform to manipulate and vides a significant contribution in flood mapping and risk analyse all relevant information in order to easily demarcate assessment. RS and GIS are quick and more efficient, which suitable hazard zones (Khan et al. 2008; Saha et al. 2005; can provide the best opportunity to capture, store, combine, Wang et al. 2013; Pourghasemi et al. 2014). RS and GIS manipulate, retrieve, analyse and display the information for techniques are useful widely to demarcate and assess flood- the determination of potential hazard areas. This research is related damages (Pradhan et al. 2014; Patel and Srivastava an ensemble method, which proves the efficiency in GIS- 2013) caused by excessive rain in a catchment area or sea based flood modelling. To estimate flood probabilities, the wave surges in the coastal regions. During last decade, sev- frequency ratio (FR) approach was combined with RS and eral models were applied for estimation of flood by many GIS. researcher and scientists (Kisi et al. 2012; Talei et al. 2010). The main goal of the present research is to examine the Alternatively, the conventional flood modelling methods usefulness of Remote Sensing (RS), GIS and the frequency were not reliable for accurate prediction (Tehrany et  al. ratio (FR) models for flood susceptibility analysis and map- 2014b; Li et al. 2012). Currently, geospatial techniques pro- ping in the Markham river basin under Morobe province, vide a wide range of data sources for the modelling of flood Papua New Guinea. The main aim of this study is to identify (Wanders et al. 2014). Analytical hierarchy process (AHP) and map out flood risk zones in the Markham river basin. is one of the most popular and satisfactory methods in dis- The objectives of this research are to create wall-to-wall data aster modelling like flood monitoring, mapping and analys- sets that are considered as input into the FR model to catego- ing complex problems (Billa et al. 2006; Yalcin 2008; Chen rize potential flood prone area, create a flood hazard map of et al. 2011). Apart from AHP, multi-criteria decision sup- the Markham river basin and also to carry out impact analy- port approach (MCDA) (Samanta et al. 2016a), weights of sis which can be useful to local government administrators, evidence (WofE) (Rahmati et al. 2016a), logistic regression researchers and planners for devising flood mitigation plans. (LR) (Tehrany et al. 2014a, b, 2015), adaptive neuro-fuzzy interface system (Sezer et al. 2011), artificial neural net- works (ANN) (Pradhan and Buchroithner 2010; Tiwari and Chatterjee 2010) and FR model (Lee et al. 2012; Rahmati Study location and materials used et al. 2016b; Liao and Carin 2009) were the other known acceptable models for hazard analysis. ANN method was The study was carried out in the final basin (basin-14) of Markham, which is located in the Morobe province of PNG used in prediction of flood, where researchers attempted to show relationship between conditioning parameters and the and encloses an area of 1806.85 km . The longitudinal and latitudinal extensions of the study area are 146.09–147.04° outcome (Pradhan and Buchroithner 2010). It was reported that ANN method can handle all inputs which are uncer- east and 6.23–6.78° south, respectively (Fig. 1). The climate is tropical humid with about 4200 mm of average annual tain to extract meaningful information (Lohani et al. 2012). MCDA, RS and GIS techniques are extremely useful in reli- rainfall in the study area. Markham river emanates from the runoff contributed by 12,450 km catchment (Fig. 1a) with able and accurate analysis and mapping of plausible flood prone zones. MCDA approach is suitable for flood analysis huge bed load (Tilley et al. 2006). This is the fourth longest river in PNG born at Finisterre range (approximately 457 m and mapping in the no-data regions and could be practi- cal for local planners in mitigation of flood. AHP model altitude) and flows into the Huon Gulf near to the downtown of Lae, which is the second largest city of PNG after 180 km was applied in China for flood diversion (Zou et al. 2013). According to Chen et  al. (2011), the weakness of AHP of checkered path (Samanta et al. 2016b). The upper basin is dominated by natural forests, steep slopes and rugged ter- model was correlated to its enslavement on the information provided by experts, which is the main source of uncertainty. rain (Pal et al. 2012; Solin 2012). Mining activities for allu- vial gold extraction on the river path and logging activities FR model may be considered as an important method which is easily understandable and can be used to produce accept- in the surroundings areas result in accelerated soil erosion within the basin area. Final 125 km downstream flow of the able flood risk analysis and mapping (Liao and Carin 2009). 1 3 Applied Water Science (2018) 8:66 Page 3 of 14 66 Fig. 1 Location map of the study area: (a) Entire Markham river basin and (b) The study area-sub-basin 14 1 3 66 Page 4 of 14 Applied Water Science (2018) 8:66 Markham river which is covered by sub-basin number 14 of Landsat 8, operational land imager (OLI). A flood inven- was selected for this study (Fig. 1b). tory database was prepared using PNGRIS national data- The Markham river catchment in Morobe province of base, field investigation and by examining remote sensing PNG is exposed to flood due to intensive rainfall, peak dis- data. Every detail of the data sets used in FR model is given charge and physiographic conditions (Tilley et al. 2006; in Table 1. Samanta et al. 2016a). During 2012, higher magnitude of floods was measured by the global flood detection system (Kugler et al. 2007). On March of 2004, Markham riverine Methodology zones experienced an overwhelming flood on the spate of Markham river. Earlier on, during the month of February It is essential to analyse the occurrence of historical flood (2004), 120-m-long road on Lae-Bulolo road was washed events to estimate future flood (Manandhar 2010). So the out just 1.5  km away towards upstream from Markham flood inventory database is the essential factor for flood bridge. The estimated peak discharge was about 2600 m /s susceptibility mapping. Flood inventory map was prepared to 3200 m /s with an average velocity as 3.4 m/s (Tilley after generating 143 flooded points through “create fishnet” et al. 2006). analysis of PNGRIS national database, field investigation To perform flood susceptibility analysis and risk assess- and satellite data before and after flood events. Seventy ments, researcher suggested different factors which are not percent (70%) of total, i.e. 100 flood points, were selected fixed (Tehrany et al. 2015). There are some common condi- randomly as the training data set for flood modelling and tioning factors which indicate their role in flood mapping. the rest 30% or 43 points were used for validating the flood Recent studies by Rahmati et al. (2016a, b) have achieved model (Rahmati et al. 2016b) (Fig. 1b). high accurate results, where they used the least number of Selection of effectual parameters is vital to produce a independent parameters. Total of 15 parameters are exam- flood hazard map in any catchment (Kia et al. 2012). It is ined as a preliminary analysis. Demarcation of flood suscep- always tricky to choose factors unanimously for use in flood tibility zones was carried out using ten (10)-folds of geospa- susceptibility mapping (Tehrany et al. 2014a, b). Flood- tial data sets, viz. land use and land cover (LULC), elevation, related geospatial database like land use/land cover, eleva- slope, topographic wetness index (TWI), surface runoff, tion, slope, topographic wetness index (TWI), surface run- landform, lithology, distance from the main river, soil texture off, landform, lithology, distance from the main river, soil and soil drainage. These geospatial layers are selected after texture and soil drainage were prepared using ArcGIS and consulting local hydrological and natural disaster experts Erdas Imagine software. based on their effectiveness in creating a flood. Wall-to-wall LULC directly or indirectly influence infiltration, geospatial database was developed from remotely sensed evapotranspiration and surface runoff generation. LULC data sets like satellite image and digital elevation model and map was prepared (Fig. 2a) from the Landsat-8 OLI satel- National GIS database of PNG. All data sets were rectified lite imagery through supervised classification technique carefully using Erdas imagine software. UTM projection, (Samanta et al. 2011). The altitude and slope are important zone 55S and WGS-84 datum were selected for the image parameters in flood risk and vulnerability mapping. Vari- and map registration purpose. Advanced space-borne ther- ations of elevations have a definitive impact on climate mal emission and reflection radiometer (ASTER) provides characteristics (Samanta et al. 2012). Such different rain- digital elevation model (DEM) with the spatial resolution fall and temperature regimes engendered varied vegetation of 30 m that was used in this research. Elevation and slope and soil forms (Aniya 1985). Slope controls the surface database were derived from the ASTER DEM. LULC was runoff, the ferocity of water flow aggravating soil erosion derived from optical bands with false colour combination (Adiat et al. 2012) as well as vertical percolation (Youssef Table 1 Data sets used for flood vulnerability mapping Sl. No. Data description Data type/resolution Year Source 1 LANDSAT-8, OLI 15 m Pan-Sharpen 2013 Department of Surveying and Land Studies, The PNG University of Technology 2 ASTER-DEM 30 m 2001 3 Rainfall, Polygon/shape file 2009 PNGRIS—PNG Resource Information System 4 Soil texture and drainage 5 Lithology 6 Landforms 7 Historical flood inventory database Point location 1 3 Applied Water Science (2018) 8:66 Page 5 of 14 66 Fig. 2 Parameters used for FR modelling: (a) land use/land cover, (b) altitude, (c) slope, (d) TWI index, (e) surface runoff, (f) landforms, (g) lithology and (h) soil texture characteristics of the study area 1 3 66 Page 6 of 14 Applied Water Science (2018) 8:66 et al. 2011). Both altitude and slope map were prepared and soil drainage was prepared based on PNGRIS data sets using ASTER data in ArcGIS 3-D analysis algorithms after verifying them based on field observations. (Fig. 2b, c). It is essential to analyse past flood record to estimate the Topographic wetness index (TWI) refers to spatial future flood event in any area (Manandhar 2010). Therefore, distribution of wetness and controls the overland flow of mapping the flood locations from past episodes in the study water. TWI has significant impact on flood mapping. TWI area is instrumental in elucidating the correlation among was calculated based on Eq. 1 (Beven and Kirkby 1979; the flooding and the condition factors. It is essential to pre- Regmi et al. 2010; Qin et al. 2011), and spatial distribution pare an inventory database with high precision (Jebur et al. map of TWI was prepared using ArcGIS (Fig. 2d). 2013). FR model was adopted for this study chosen from a plethora of bivariate statistical techniques. The approach TWI = Ln , offers a quantitative relationship between the ‘frequency of (1) tan B flood episodes’ and various conditioning parameters. The where a is the specific catchment area [a = A/L, total basin FR index was calculated using the following Eq. 5 (Tehrany area (A) divided by length of contour (L)], and B is referred et al. 2014a, b). to the slope in degree. Overland flow of water, called surface runoff, occurs in FSI = FR, (5) full throttle in the aftermath of saturated infiltration when where FSI is the flood susceptibility index and FR is the the surplus water minus saturated infiltration component, frequency ratio for each factor. emanating from the storm, melt water, etc. keep flowing The FR can be expressed as the proportion of flooded area over the Earth’s surface (Pal and Samanta 2011). Surface in the total study area (Eq. 6); it is the ratio of the probabili- runoff in urban areas is normally reinforced owing to ties of the ‘flooded’ to ‘not flooded’ area for a given attribute lack of infiltration surface, which gives rise to pernicious (Bonham-Carter 1994). urban flooding. Surface runoff due to storm rainfall is very significant for forecasting floods which are very sudden, ∕ ∕ ∕ FR = (E F) (M L), (6) flashy and of short duration (Pal and Samanta 2011). Sur - where E is the number of flood episodes for each factor; F face runoff database was generated (Fig.  2e) based on soil is the total number of flood episodes; M is the histogram of conservation service (SCS) model (Eqs. 2–4). a class; L is the total histogram of the study area. The frequency ratio model was used to establish the cor- Q = (P−Ia) (P−Ia + S), (2) relation between historical flood locations and the probable supporting factors. FR value indicates (Table 2) the types where Q is actual surface runoff in mm; P is storm rainfall of correlation between factors and floods. A FR value lower (mm); S is the potential maximum retention (mm), and Ia is than 1 indicates weak correlation; on the other hand a value 0.4S [Initial abstraction (mm)]. of more than 1 refers to strong correlations. The overall So the modified form of Eq.  2 can be expressed in Eq. 3. methodology that was applied for flood susceptibility map- ping is shown as a flowchart in Fig.  3. Q = (P−0.4S) (P + 0.6S). (3) To calculate the value of potential maximum retention (S) of Eq. 3, another simple Eq. 4 was used. Results and discussion S = (25400∕CN) − 254, (4) There are many independent variables, called condition- where CN is curve number of hydrologic soil cover com- ing factors that play specific role in order to perform flood plex, which happens to be a function of soil type, land cover susceptibility mapping (Pradhan 2010; Pourghasemi et al. and antecedent moisture condition (SCS 1972; Kumar et al. 2012; Kia et al. 2012). Spatial distribution and statistical 1991; Rao et al.1996). database for all ten (10) conditioning factors namely, LULC, Flood generally occurs near to the bank of the river and elevation, slope, topographic wetness index (TWI), surface inundates low-lying flood plain areas. Distance from the runoff, landform, lithology, distance from the main river, soil river has significant impact on the flood and its magnitude. texture and soil drainage were constructed with their sub- Distance from main river was considered as one parameter classes (Fig. 2 and Table 2). Classification of satellite data which was developed through proximity analysis in Arc- was done (Samanta et al. 2012) considering nine (9) land GIS. Infiltration varies based on the spatial distribution of use/land cover categories, namely dense forest, low dense soil texture and it controls overland flows and inundation. forest, shrub land, outcrop/cleared/burnt lands, mountain/ Database on landform, lithology, soil texture (Fig. 2f–h) upland grassland, settlement, inland water, river water and 1 3 Applied Water Science (2018) 8:66 Page 7 of 14 66 Table 2 Conditioning parameters used for flood vulnerability mapping through FR model Value Class name or description Histogram % of Histogram Flood numbers % flood numbers Frequency Ratio Land use/land cover 1 Dense forest 50700 2.5 1 1.0 0.40 2 Low dense forest 703420 35.0 33 33.0 0.94 3 Shrub land 533068 26.6 26 26.0 0.98 4 Outcrop/cleared/burnt lands 37806 1.9 0 0.0 0.00 5 Mountain/upland grassland 433316 21.6 28 28.0 1.30 6 Settlement 18728 0.9 0 0.0 0.00 7 Inland water 5406 0.3 0 0.0 0.00 8 River water 133552 6.7 11 11.0 1.65 9 Agriculture 91612 4.6 1 1.0 0.22 Elevation (in m) 1 Up to 100 m 746583 37.2 59 59.0 1.59 2 100–200 m 544304 27.1 25 25.0 0.92 3 200–300 m 301652 15.0 11 11.0 0.73 4 300–400 m 120706 6.0 5 5.0 0.83 5 400–500 m 69673 3.5 0 0.0 0.00 6 500–1000 m 177557 8.8 0 0.0 0.00 7 More than 1000 m 47133 2.3 0 0.0 0.00 Slope (in degree) 1 up to 2.5° 336363 16.8 21 21.0 1.25 2 2.5–5.0° 513214 25.6 40 40.0 1.56 3 5.0–10.0° 563975 28.1 29 29.0 1.03 4 10.0–15.0° 209103 10.4 2 2.0 0.19 5 15.0–20.0° 131856 6.6 5 5.0 0.76 6 20.0–25.0° 100073 5.0 2 2.0 0.40 7 More than 25° 153024 7.6 1 1.0 0.13 Topographic wetness index (TWI) 1 up to 6.0 129577 6.5 1 1.0 0.15 2 6.0–6.5 226488 11.3 6 6.0 0.53 3 6.5–7.0 250979 12.5 3 3.0 0.24 4 7.0–7.5 387731 19.3 16 16.0 0.83 5 7.5–8.0 449732 22.4 32 32.0 1.43 6 8.0–8.5 299489 14.9 28 28.0 1.88 7 More than 8.5 263612 13.1 14 14.0 1.07 Surface runoff (in mm) 1 Up to 50 mm 682498 34.0 41 41.0 1.21 2 50–100 mm 281239 14.0 8 8.0 0.57 3 100–150 mm 486606 24.2 28 28.0 1.16 4 150–200 mm 406985 20.3 13 13.0 0.64 5 More than 200 mm 150280 7.5 10 10.0 1.34 Landform 1 Dissected relict alluvial, colluvial mudflow and fans 75020 3.7 0 0.0 0.00 2 Mountains and hills with weak or no structural 476893 23.8 0 0.0 0.00 control 3 Braided floodplains 296938 14.8 36 36.0 2.43 4 Composite alluvial plains 94495 4.7 10 10.0 2.12 5 Composite bar plain and alluvial fan complex 46996 2.3 1 1.0 0.43 6 Little dissected recent alluvial fans 779505 38.8 41 41.0 1.06 7 Homoclinal ridges and cuestas 77006 3.8 0 0.0 0.00 8 Back plains 46935 2.3 4 4.0 1.71 1 3 66 Page 8 of 14 Applied Water Science (2018) 8:66 Table 2 (continued) Value Class name or description Histogram % of Histogram Flood numbers % flood numbers Frequency Ratio 9 Back swamps 40223 2.0 5 5.0 2.50 10 Hilly terrain with weak or no structural control 27365 1.4 0 0.0 0.00 11 Lake 6416 0.3 0 0.0 0.00 12 Meander floodplains 8914 0.4 1 1.0 2.25 13 Undifferentiated swamps 30902 1.5 2 2.0 1.30 Lithology 1 Pleistocene sediments 75020 3.7 0 0.0 0.00 2 Coarse grained sedimentary 290647 14.5 0 0.0 0.00 3 Alluvial deposits 1344906 67.0 100 100.0 1.49 4 Mixed or undifferentiated igneous 19162 1.0 0 0.0 0.00 5 Mixed or undifferentiated sedimentary 61629 3.1 0 0.0 0.00 6 Limestone 74653 3.7 0 0.0 0.00 7 Mixed sedimentary and limestone 14059 0.7 0 0.0 0.00 8 Mixed sedimentary and volcanic 16871 0.8 0 0.0 0.00 9 Acid to intermediate igneous 27385 1.4 0 0.0 0.00 10 Lake 6416 0.3 0 0.0 0.00 11 Low grade metamorphic 76860 3.8 0 0.0 0.00 Distance from river in m 1 up to 1000 m 393500 19.6 26 26.0 1.33 2 1000–2000 m 243382 12.1 19 19.0 1.57 3 2000–3000 m 206855 10.3 9 9.0 0.87 4 3000–4000 m 175027 8.7 13 13.0 1.49 5 4000–5000 m 149073 7.4 4 4.0 0.54 6 5000–6000 m 133660 6.7 4 4.0 0.60 7 More than 6000 m 706111 35.2 25 25.0 0.71 Soil texture 1 Sandy clay loam 439768 21.9 20 20.0 0.91 2 Sandy loam 129127 6.4 13 13.0 2.02 3 Silty clay loam 377624 18.8 12 12.0 0.64 4 Peat 22672 1.1 1 1.0 0.89 5 Silty loam 22699 1.1 4 4.0 3.54 6 Sandy clay 634345 31.6 28 28.0 0.89 7 River course-gravel 216519 10.8 19 19.0 1.76 8 Sand 66418 3.3 2 2.0 0.60 9 Lake 6567 0.3 0 0.0 0.00 10 Silty clay 91869 4.6 1 1.0 0.22 Soil drainage 1 Well drained 1631997 81.3 62 62.0 0.76 2 Waterlogged (swampy) 45371 2.3 5 5.0 2.21 3 Poorly to very poorly drained 105842 5.3 14 14.0 2.66 4 River course 216519 10.8 19 19.0 1.76 5 Imperfectly drained 1312 0.1 0 0.0 0.00 6 Lake 6567 0.3 0 0.0 0.00 agriculture land (Fig. 2a). The predominant land cover in the study area, but the FR value was calculated as 1.65 which study area is ‘low density forest’ (35%) in the eastern region is highly correlated to flood (Table  2). The elevation ranged and some pockets of north-west and southern region of the from 0 m to 1790 m of the study area. The spatial distribu- study area. The ‘river class’ along the centre (main river) and tion of the elevation map was prepared after reclassifying north (Erap river) of the study area covers only 6.7% of the the elevation layer into seven (7) classes (< 100, 100–200, 1 3 Applied Water Science (2018) 8:66 Page 9 of 14 66 Fig. 3 Methodological flow chart of flood susceptibility mapping 200–300, 300–400, 400–500, 500–1000 and > 1000  m) 7.5–8.0, 8.0–8.5 and > 8.5) as shown in Fig. 2d. The high- where brown colour indicates maximum altitude found in est TWI was recorded in the middle part of the study area north, north-west, and blue indicated < 100 m altitude zone represented with blue colour. Higher TWI value refers in the east and south-east part in the study area (Fig. 2b). higher chances of flooding in a watershed (Rahmati et al. Up to 100 m altitude zone is the predominant class cov- 2016b). The FR value was calculated as 1.88 where TWI ering 37.2% of the total area, which refers to a higher FR varied from 8.0 to 8.5, and 0.15 for the class with TWI value (1.59) than other classes (Table 2). Slope (in degree) < 6.0, respectively (Table 2). Surface runoff map was gen- was calculated from DEM data and reclassified into seven erated based on storm rainfall of 229.6 mm during 21–23 (7) classes, namely < 2.5°, 2.5–5°, 5–10°, 10–15°, 15–20°, October 2012. Maximum surface runoff was calculated as 20–25° and more than 25°. The flat or lower slope gradient 229 mm along the river class. High surface runoff during a (< 5°) area is situated on both sides of the river as shown storm indicates high probability of flood (Pal and Samanta with blue colour (Fig.  2c). Lower the slope gradient, the 2011). The surface runoff database was reclassified into five more is the possibility of flooding and flood events (Rah- (5) classes, namely < 50, 50–100, 100–150, 150–200, and mati et al. 2016a). Two lower slope gradient classes, namely > 200  mm (Fig. 2e). FR value was calculated as 1.34 for < 2.5° and 2.5–5°, indicates higher FR value of 1.25 and highest runoff category (> 200 mm) (Table 2). Entire study 1.56, respectively, whereas > 25° slope area indicated lower area has been categorized into thirteen (13) landform units, FR value of 0.13 (Table 2). namely (i) dissected relict alluvial, colluvial mudflow and Spatial database on TWI was calculated and categorized fans, (ii) mountains and hills with weak or no structural con- into seven (7) classes (< 6.0, 6.0–6.5, 6.5–7.0, 7.0–7.5, trol, (iii) braided floodplains, (iv) composite alluvial plains, 1 3 66 Page 10 of 14 Applied Water Science (2018) 8:66 Fig. 4 Output map through FR model: (a) flood susceptibility index with all flood points and (b) overlay of villages and road on flood suscepti- bility zones (v) composite bar plain and alluvial fan complex, (vi) lit- The lithology of the study area has been categorized into tle dissected recent alluvial fans, (vii) homoclinal ridges (i) Pleistocene sediments, (ii) coarse grained sedimentary, and cuestas, (viii) back plains, (ix) back swamps, (x) hilly (iii) alluvial deposits, (iv) mixed or undifferentiated igneous, terrain with weak or no structural control, (xi) lake, (xii) (v) mixed or undifferentiated sedimentary, (vi) limestone, meander floodplains and (xiii) undifferentiated swamps (vii) mixed sedimentary and limestone, (vii) mixed sedi- (Fig. 2f). The back swamps and meander floodplains have mentary and volcanic, (ix) acid to intermediate igneous, (x) higher FR value of 2.50 and 2.43, respectively (Table 2). lake and (xi) low grade metamorphic (Fig. 2g). Lithology is 1 3 Applied Water Science (2018) 8:66 Page 11 of 14 66 Table 3 Spatial distribution (statistics) of flood vulnerability classes imperfectly drained and (vi) lake. Within the soil drainage in the study area factor, poorly to very poorly drained class had the highest FR value of 2.66, followed by waterlogged (swampy) area with Sl. no. Flood vulner- FR value range Histogram % area able class the FR value of 2.21. Finally, the soil texture map was gen- erated for the study area where ten (10) soil texture classes 1 Very low < 5.0 206164 10.3 could be found (Samanta et al. 2016b), namely sandy clay 2 Low 5.0–7.5 353062 17.6 loam, sandy loam, silty clay loam, peat, silty loam, sandy 3 Medium 7.5–10.0 340554 17.0 clay, river course-gravel, sand, lake and silty clay (Fig. 2h). 4 High 10.0–12.5 718863 35.8 The FR value was calculated as 3.54 for sandy loam and 0.22 5 Very high > 12.5 388965 19.4 for silty clay (Table 2). Thus, the rating of each subclass of all conditioning param- eter was generated based on the FR values as shown in Table 2. an important conditioning parameter in flooding because it FR value varied from 0 to 3.54 in the study area. Calculated has a direct influence on land permeability and thus surface FR values were indicated as weak (< 1) to strong (> 1) correla- runoff (Haghizadeh et al. 2017). The maximum FR value tions with flood occurrence (Lee et al. 2012). Finally, based on of 1.49 was recorded in alluvial deposits which covers 67% FR model in Eq. 5, the flood probability database was devel- of the total study area (Table  2). Flood intensity became oped (Fig. 4a). FR value in the model output varied from 2.66 less in those locations far away from the river and the risk to 19.02. Greater FR value indicated the higher probability was higher in areas near to the river bank. Distances from to flood occurrences. The developed database was reclassi- the river in the range from 4000 to 5000, 5000 to 6000 and fied into five (5) different flood susceptibility zones, namely > 6000 m have a low probability of flooding, whereas dis- very low (less that 5.0), low (5.0–7.5), moderate (7.5–10.0), tances in the range of < 1000 and 1000–2000 m together high (10.0–12.5) and very high susceptibility (more than 12.5) indicate highest FR values (1.42), which demonstrates the (Fig. 4b). The result indicates that about 19.4% land area are highest flood event (Haghizadeh et al. 2017). Soil texture demarcated as very high, 35.8% as high, 17.0% as moderate, and soil drainage are very important factors in flood sus- 17.6% as low and 10.3% as very low flood vulnerable class ceptibility mapping. Well-drained soils produce less surface (Table 3). High to very high vulnerable classes are mostly runoff than poorly drained soil group (Pal et al. 2012). The located along the middle part of the study area (Fig.  4b). soil drainage database was generated with six (6) catego- These high to very high flood susceptibility zones are charac- ries, namely (i) well drained, (ii) waterlogged (swampy), terized with higher runo p ff otentiality, poorly to very poorly (iii) poorly to very poorly drained, (iv) river course, (v) drained soil, alluvial deposits, braided flood plain, lower slope Table 4 Calculation of prediction accuracy and success rate for the flood susceptibility analysis Sl. no. Susceptibility class Verification Accurate (high to very Prediction accuracy Training Success (high to very Success rate (30% flood high class) (70% flood high class) points) points) 1 Very low 0 42 0.977 0 94 0.94 or or 2 Low 0 0 97.7% 94.0% 3 Medium 1 6 4 High 20 45 5 Very high 22 49 Total 43 100 Table 5 Flood susceptibility Susceptibility class FR value range Village % of village under Total population zone and the risk factor flood vulnerability 1 Very low < 5.0 1 0.3 733 2 Low 5.0–7.5 16 4.5 5021 3 Medium 7.5–10.0 83 23.2 30466 4 High 10.0–12.5 183 51.1 93860 5 Very high > 12.5 75 20.9 43432 1 3 66 Page 12 of 14 Applied Water Science (2018) 8:66 gradient, lower elevation and closer to the main river, which calculated based on the 100 flood training points (70%). The are the important conditioning factors for flood susceptibility validation report indicated a higher prediction accuracy of mapping using the FR model. 97.7% which had been enough to validate the FR model There are many models as proposed by different research- that was used for this study. It is obvious that a higher num- ers, but it is very important to evaluate the accuracy and suc- ber of input data sets generate higher accuracy. FR model cess rate to validate the model used for flood susceptibility requires a large number of flood points as training like 70% analysis (Chung and Fabbri 2003; Tien Bui et al. 2012). The (or 100 points) to generate Frequency Ratio, whereas a less modelled output through FR model is validated in regard of number of input data do not fall under all classes of every success rate and prediction accuracy. The value of 1.0 repre- parameter. In this point of view, we used 70% flood points sented the highest accuracy, which indicates the capability of for flood map development and 30% (70–30) for the valida- the model in predicting natural hazards without any bias (Prad- tion process. In case we used 60–40 or 50–50, the results han and Buchroithner 2010). Success rate was calculated using are not same. The accuracy varied 5–10% lower than 70–30 100 training flood locations and prediction of accuracy, using selection method. To validate the superiority, we sought to remaining 43 flood locations which were not used during the compare the method used in this study (FR model) with model building. Class ranges from ‘high’ to ‘very high’ sus- another multi-criteria decision support approach (MCDA) ceptibility are considered as potential o fl od that might occur in (Samanta et al. 2016a) which was conducted in the same future. Success rate and prediction rate are calculated as 0.94 river basin. As per our assessment, FR model produced bet- and 0.977, respectively (Table 4). So the prediction accuracy ter results compared to MCDA. This FR model can be used corroborates to 97.7% which validates the FR model used in in any other geographical area to develop a flood risk map this flood susceptibility analysis. which can help planners and decision makers to perform Finally, an impact analysis was done to assess the risk fac- proper flood management in future. tor by probable flood occurrences on the local community and Acknowledgements The authors are thankful to the PNGUNITECH population (Table 5). About 75 villages (20.9%) are situated (Papua New Guinea University of Technology) and to the Department in the very high flood vulnerable zone and 183 (51.1%) in the of Surveying and Land Studies for all the facilities made available high flood vulnerable zone (Fig.  4b). A total of 137292 peo- and availed for the work as a researcher. Satellite digital data avail- able from USGS Global Land Cover Facility and used in this study ple are living in those vulnerable zones which require special are also duly acknowledged. The authors gratefully acknowledge the attention from various levels of governments to take appropri- anonymous reviewers for providing their critical comments to improve ate actions to prevent and mitigate future flood occurrence. the quality of this article. Compliance with ethical standards Conclusion Conflict of interest The authors declare that there is no conflict of in- In the current research, the FR model is used to analyse terest for the publication of this article. flood susceptibility zone in the lower part of the Markham Open Access This article is distributed under the terms of the Crea- river basin (Sub-basin 14). Ten independent conditioning tive Commons Attribution 4.0 International License (http://creat iveco factors, like LULC, elevation, slope, TWI, surface runoff, mmons.or g/licenses/b y/4.0/), which permits unrestricted use, distribu- landform, lithology, distance from the main river, soil tex- tion, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the ture and soil drainage were derived from the geospatial data Creative Commons license, and indicate if changes were made. sets and used as input into the FR model towards flood prone area mapping. As the result suggests, these ten variables are likely to be major factors to map flood-affected zone. The FR model was used to establish the relationship between References past and future flood occurrences. 143 flooded points were generated through “create fishnet” analysis of the PNGRIS Adiat KAN, Nawawi MNM, Abdullah K (2012) Assessing the accu- national database, field investigation and satellite data before racy of GIS-based elementary multi criteria decision analysis as a spatial prediction tool–a case of predicting potential zones of and after oo fl d events. Random sampling technique was used sustainable groundwater resources. J Hydrol 440–41:75–89 to select 70% input flood data and to calculate frequency Aniya M (1985) Landslide-susceptibility mapping in the Amahata river ratio for each parameters and their subclasses (Rahmati basin, Japan. Ann Assoc Am Geogr 75(1):102–114 et al. 2016b). The use of stratified random sampling would Beven and Kirkby (1979) A physically based variable contributing area model of basin hydrology. Hydrol Sci Bull 24:43–69 be difficult to generate frequency ratio for all classes under Billa L, Shattri M, Mahmud AR, Ghazali AH (2006) Comprehensive each parameter. So random sampling has been mooted by the planning and the role of SDSS in flood disaster management in authors as the prudent choice in this type of studies. FR val- Malaysia. Dis Prev Manag 15:233–240 ues for different categories of all conditioning factors were 1 3 Applied Water Science (2018) 8:66 Page 13 of 14 66 Bonham-Carter GF (1994) Geographic information systems for geo- Pal B, Samanta S, Pal DK (2012) Morphometric and Hydrological scientists: modeling with GIS. In: Bonham-Carter F (ed) Com- analysis and mapping for Watut watershed using Remote Sensing puter methods in the geosciences. Pergamon, Oxford and GIS techniques. Int J Adv Eng Technol 2(1):357–368 Chen YR, Yeh CH, Yu B (2011) Integrated application of the ana- Patel DP, Srivastava PK (2013) Flood hazards mitigation analysis lytic hierarchy process and the geographic information system using remote sensing and GIS: correspondence with town plan- for flood risk assessment and flood plain management in Tai- ning scheme. Water Resour Manag 27:2353–2368 wan. Nat Hazards 59(3):1261–1276 Pourghasemi HR, Pradhan B, Gokceoglu C (2012) Application of fuzzy Chung CF, Fabbri AG (2003) Validation of spatial prediction models logic and analytical hierarchy process (AHP) to landslide suscepti- for landslide hazard mapping. Nat Hazard 30:451–472 bility mapping at Haraz watershed, Iran. Nat Hazards 63:965–996 Du J, Fang J, Xu W, Shi P (2013) Analysis of dry/wet conditions Pourghasemi HR, Moradi HR, Aghda SMF, Gokceoglu C, Pradhan B using the standardized precipitation index and its potential use- (2014) GIS-based landslide susceptibility mapping with probabil- fulness for drought/flood monitoring in Hunan Province China. istic likelihood ratio and spatial multi-criteria evaluation models Stoch Env Res Risk Assess 27(2):377–387 (North of Tehran, Iran). Arab J Geosci 7(5):1857–1878 Haghizadeh A, Siahkamari S, Haghiabi AH, Rahmati O (2017) Pradhan B (2010) Flood susceptible mapping and risk area delinea- Forecasting flood-prone areas using Shannon’s entropy tion using logistic regression, GIS and remote sensing. J Spatial model. J Earth Syst Sci 126:39. https ://doi.org/10.1007/s1204 Hydrol 9:1–18 0-017-0819-x Pradhan B, Buchroithner MF (2010) Comparison and validation of Huang X, Tan H, Zhou J, Yang T, Benjamin A, Wen SW, Li S, Liu landslide susceptibility maps using an artificial neural network A, Li X, Fen S, Li X (2008) Flood hazard in Hunan province of model for three test areas in Malaysia. Environ Eng Geosci China: an economic loss analysis. Nat Hazards 47:65–73 16:107–126 Jebur MN, Mohd Shafri HZ, Pradhan B, Tehrany MS (2013) Per- Pradhan B, Hagemann U, Tehrany MS, Prechtel N (2014) An easy pixel and object oriented classification methods for mapping to use ArcMap based texture analysis program for extraction of urban land cover extraction using SPOT 5 imagery. Geocarto flooded areas from TerraSAR-X satellite image. Comput Geosci Int. https ://doi.org/10.1080/10106 049.2013.84894 4 63:34–43 Khan H, Haider S, Saeed K, Ali N (2008) Assessment of potable Qin et  al (2011) An approach to computing topographic wetness water quality of Kohat division and its impact on health. J Chem index based on maximum downslope gradient. Precision Agric Soc Pak 30:246–250 12:32–43 Kia MB, Pirasteh S, Pradhan B, Rodzi MA, Sulaiman WNA, Moradi Rahmati O, Zeinivand H, Besharat M (2016a) Flood hazard zoning in A (2012) An artificial neural network model for flood simula- Yasooj region, Iran, using GIS and multi-criteria decision analy- tion using GIS: Johor River Basin, Malaysia. Environ Earth Sci sis. Geomat Nat Hazard Risk. https ://doi.or g/10.1080/19475 67:251–264705.2015.10450 43 Kisi O, Nia AM, Gosheh MG, Tajabadi MRJ, Ahmadi A (2012) Rahmati O, Pourghasemi HR, Zeinivand H (2016b) Flood susceptibil- Intermittent streamflow forecasting by using several data driven ity mapping using frequency ratio and weights-of-evidence mod- techniques. Water Resour Manag 26(2):457–474 els in the Golastan Province, Iran. Geocarto International. https Kugler Z, Groeve TD (2007) The Global Flood Detection system; ://doi.org/10.1080/10106 049.2015.10415 59 EUR 23303 EN. Office for Official Publications of the European Rao KV, Bhattacharya AK, Mishra K (1996) Runoff estimation by Communities, Luxembourg, pp 1–45 curve number method- case studies. J Soil Water Conserv 40:1–7 Kumar P, Tiwari KN, Pal DK (1991) Establishing SCS runoff curve Regmi NR, Giardino JR, Vitek JD (2010) Modeling susceptibility to number from IRS digital database. J Indian Soc Remote Sens landslides using the weight of evidence approach: Western Colo- 19(4):245–251 rado, USA. Geomorphology 115:172–187 Lee MJ, Kang JE, Jeon S (2012) Application of frequency ratio Saha AK, Gupta RP, Sarkar I, Arora KM, Csaplovics E (2005) An model and validation for predictive flooded area susceptibility approach for GIS-based statistical landslide susceptibility zona- mapping using GIS. In: Proceedings of Geoscience and Remote tion with a case study in the Himalayas. Landslides 2(1):61–69 Sensing Symposium (IGARSS), 2012 IEEE International. Samanta S, Pal DK, Lohar D, Pal B (2011) Preparation of digital data Munich, 895–898 sets on land use/land cover, soil and digital elevation model for Li XH, Zhang Q, Shao M, Li YL (2012) A comparison of parameter temperature modelling using Remote Sensing and GIS techniques. estimation for distributed hydrological modelling using auto- Indian J Sci Technol 4(6):636–642 matic and manual methods. Adv Mater Res 356–360:2372–2375 Samanta S, Pal DK, Lohar D, Pal B (2012) Interpolation of climate var- Liao X, Carin L (2009) Migratory logistic regression for learning iables and temperature modeling. Theor Appl Climatol 107(1):35– concept drift between two data sets with application to UXO 45. https ://doi.org/10.1007/s0070 4-011-0455-3 sensing. IEEE Trans Geosci Remote Sens 47:1454–1466 Samanta S, Koloa C, Pal DK, Palsamanta B (2016a) Flood risk analysis Lohani A, Kumar R, Singh R (2012) Hydrological time series mod- in lower part of Markham River based on multi-criteria decision eling: a comparison between adaptive neuro-fuzzy, neural net- approach (MCDA). Hydrology 3(3):29. https ://doi.org/10.3390/ work and autoregressive techniques. J Hydrol 442:23–35hydro logy3 03002 9 Manandhar B (2010) Flood Plain Analysis and Risk Assessment Samanta S, Koloa C, Pal DK, Palsamanta B (2016b) Estimation of of Lothar Khola, MSc Thesis, Tribhuvan University, Phokara, potential soil erosion rate using RUSLE and E30 model. J Model Nepal, pp. 64 Earth Syst Environ 2(3):1–11. https ://doi.or g/10.1007/s4080 Markantonis V, Meyer V, Lienhoop N (2013) Evaluation of the envi- 8-016-0206-7 ronmental impacts of extreme floods in the Evros River basin SCS (1972) Soil Conservation Department, Handbook of Hydrology. using contingent valuation method. Nat Hazard 69:1535–1549 Ministry of Agriculture, New Delhi Merz B, Kreibich H, Schwarze R, Thieken A (2010) Assess- Sezer EA, Pradhan B, Gokceoglu C (2011) Manifestation of an adap- ment of economic flood damage. Nat Hazard Earth Syst Sci tive neuro-fuzzy model on landslide susceptibility mapping: 10:1697–1724 Klang Valley Malaysia. Expert Syst Appl 38(7):8208–8219 Pal B, Samanta S (2011) Surface runoff estimation and mapping using remote sensing and geographic information system. Int J Adv Sci Technol 3(2):106–114 1 3 66 Page 14 of 14 Applied Water Science (2018) 8:66 Solin L (2012) Spatial variability in the flood vulnerability of urban improving operational flood forecasting. Hydrol Earth Syst Sci areas in the headwater basins of Slovakia. Flood Risk Manag 18:2343–2357 5:303–320 Wang HB, Wu SR, Shi JS, Li B (2013) Qualitative hazard and risk Talei A, Chua LHC, Quek C (2010) A novel application of a neu- assessment of landslides: a practical framework for a case study rofuzzy computational technique in event-based rainfall–runoff in China. Nat Hazard 69(3):1281–1294 modeling. Expert Syst Appl 37(12):7456–7468 WHO (2003) World Health Organization. Disaster Data-key Trends Tehrany MS, Pradhan B, Jebur MN (2014a) Flood susceptibility map- and Statistics in World Disasters Report; WHO: Geneva, Switzer- ping using a novel ensemble weights-of-evidence and support land. http://www.ifrc.org/PageF iles/89755 /2003/43800 -WDR20 vector machine models in GIS. J Hydrol 512:332–343 03_En.pdf (Accessed 5 April 2017) Tehrany MS, Lee MJ, Pradhan B, Jebur MN, Lee S (2014b) Flood Yalcin A (2008) GIS-based landslide susceptibility mapping using susceptibility mapping using integrated bivariate and multivariate analytical hierarchy process and bivariate statistics in Ardesen statistical models. Environ. Earth Sci 72:4001–4015 (Turkey): comparisons of results and confirmations. CATENA Tehrany MS, Pradhan B, Jebur MN (2015) Flood susceptibility analysis 72:1–12 and its verification using a novel ensemble support vector machine Youssef AM, Pradhan B, Hassan AM (2011) Flash flood risk esti- and frequency ratio method. Stoch Environ Res Risk Assess mation along the St.Katherine road, southern Sinai, Egypt using 29:1149–1165. https ://doi.org/10.1007/s0047 7-015-1021-9 GIS based morphometry and satellite imagery. Environ Earth Sci Tien Bui D, Pradhan B, Lofman O, Revhaug I, Dick OB (2012) Land- 62:611–623 slide susceptibility mapping at Hoa Binh province (Vietnam) Zou Q, Zhou J, Zhou C, Song L, Guo J (2013) Comprehensive flood using an adaptive neuro-fuzzy inference system and GIS. Comput risk assessment based on set pair analysis-variable fuzzy sets Geosci 45:199–211 model and fuzzy AHP. Stoch Env Res Risk Assess 27(2):525–546 Tilley JR, Kell RAM, Haro A (2006) Protecting the Markham Bridge, Morobe Province, PNG. In: Proceedings of the 9th International Publisher’s Note Springer Nature remains neutral with regard to River Symposium, Brisbane, Australia, 4–7 September jurisdictional claims in published maps and institutional affiliations. Tiwari MK, Chatterjee C (2010) Uncertainty assessment and ensemble flood forecasting using bootstrap based artificial neural networks (BANNs). J Hydrol 382(1):20–33 Wanders N, Karssenberg D, de Roo A, de Jong SM, Bierkens MFP (2014) The suitability of remotely sensed soil moisture for 1 3 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Applied Water Science Springer Journals

Flood susceptibility analysis through remote sensing, GIS and frequency ratio model

Free
14 pages
Loading next page...
 
/lp/springer_journal/flood-susceptibility-analysis-through-remote-sensing-gis-and-frequency-pfNu76rf1g
Publisher
Springer Berlin Heidelberg
Copyright
Copyright © 2018 by The Author(s)
Subject
Earth Sciences; Hydrogeology; Water Industry/Water Technologies; Industrial and Production Engineering; Waste Water Technology / Water Pollution Control / Water Management / Aquatic Pollution; Nanotechnology; Private International Law, International & Foreign Law, Comparative Law
ISSN
2190-5487
eISSN
2190-5495
D.O.I.
10.1007/s13201-018-0710-1
Publisher site
See Article on Publisher Site

Abstract

Papua New Guinea (PNG) is saddled with frequent natural disasters like earthquake, volcanic eruption, landslide, drought, flood etc. Flood, as a hydrological disaster to humankind’s niche brings about a powerful and often sudden, pernicious change in the surface distribution of water on land, while the benevolence of flood manifests in restoring the health of the thalweg from excessive siltation by redistributing the fertile sediments on the riverine floodplains. In respect to social, economic and environmental perspective, flood is one of the most devastating disasters in PNG. This research was conducted to investigate the usefulness of remote sensing, geographic information system and the frequency ratio (FR) for flood susceptibility map - ping. FR model was used to handle different independent variables via weighted-based bivariate probability values to generate a plausible flood susceptibility map. This study was conducted in the Markham riverine precinct under Morobe province in PNG. A historical flood inventory database of PNG resource information system (PNGRIS) was used to generate 143 flood locations based on “create fishnet” analysis. 100 (70%) flood sample locations were selected randomly for model building. Ten independent variables, namely land use/land cover, elevation, slope, topographic wetness index, surface runoff, landform, lithology, distance from the main river, soil texture and soil drainage were used into the FR model for flood vulnerability analysis. Finally, the database was developed for areas vulnerable to flood. The result demonstrated a span of FR values ranging from 2.66 (least flood prone) to 19.02 (most flood prone) for the study area. The developed database was reclassified into five (5) flood vulnerability zones segmenting on the FR values, namely very low (less that 5.0), low (5.0–7.5), moder - ate (7.5–10.0), high (10.0–12.5) and very high susceptibility (more than 12.5). The result indicated that about 19.4% land area as ‘very high’ and 35.8% as ‘high’ flood vulnerable class. The FR model output was validated with remaining 43 (30%) flood points, where 42 points were marked as correct predictions which evinced an accuracy of 97.7% in prediction. A total of 137292 people are living in those vulnerable zones. The flood susceptibility analysis using this model will be very useful and also an efficient tool to the local government administrators, researchers and planners for devising flood mitigation plans. Keywords Remote sensing · GIS · Flood vulnerability analysis · Frequency ratio · Markham river Introduction High intensity downpours in a region often lead to flood - ing in the downstream areas. Floods happen when overland * Sailesh Samanta flow of waters inundates land (Merz et al. 2010). Natural rsgis.sailesh@gmail.com disasters, like floods, are causing massive damages to natural Dilip Kumar Pal and human resources (Du et al. 2013; Youssef et al. 2011). dkpal200090@gmail.com An average of 140 million people is affected per year due Babita Palsamanta to flooding (WHO 2003). In respect of socioeconomic and mundey2@gmail.com environmental consequences, widespread flood analysis is Department of Surveying and Land Studies, The PNG very significant (Markantonis et al. 2013). Control of a flood University of Technology, Private Mail Bag, Lae, Morobe, and prevention measures are necessary to reduce potential Papua New Guinea damages to natural resources, agriculture, infrastructure etc. PNG University of Technology Campus, Private Mail Bag, (Billa et al. 2006; Huang et al. 2008). Therefore, analysis of Lae, Morobe, Papua New Guinea Vol.:(0123456789) 1 3 66 Page 2 of 14 Applied Water Science (2018) 8:66 flood susceptibility is an important task for early warning WofE and FR model are relatively new for flood vulnerabil- system, emergency services towards management strategies ity modelling, and they were widely used in other natural of prevention and mitigation of the future flood episodes hazards like landslide susceptibility mapping (Rahmati et al. (Tehrany et al. 2015). There are several comprehensive tools 2016b). Both models are almost similar and yield compara- available and used by many research organizations world- ble results for flood susceptibility mapping. Flood risk map wide, for example HAZUS, a GIS-based natural hazard through FR model was expected to be used in programmes analysis tool developed for assessing flood hazard; HEC- to reduce the flood and its damage (Lee et al. 2012). FDA, a computer program to assist crop engineers through Modelling of flood is a complex exercise where a lot of vulnerability analysis of flood risk reduction policy. RS and factors are supposed to be considered. RS technique pro- GIS techniques offer a suitable platform to manipulate and vides a significant contribution in flood mapping and risk analyse all relevant information in order to easily demarcate assessment. RS and GIS are quick and more efficient, which suitable hazard zones (Khan et al. 2008; Saha et al. 2005; can provide the best opportunity to capture, store, combine, Wang et al. 2013; Pourghasemi et al. 2014). RS and GIS manipulate, retrieve, analyse and display the information for techniques are useful widely to demarcate and assess flood- the determination of potential hazard areas. This research is related damages (Pradhan et al. 2014; Patel and Srivastava an ensemble method, which proves the efficiency in GIS- 2013) caused by excessive rain in a catchment area or sea based flood modelling. To estimate flood probabilities, the wave surges in the coastal regions. During last decade, sev- frequency ratio (FR) approach was combined with RS and eral models were applied for estimation of flood by many GIS. researcher and scientists (Kisi et al. 2012; Talei et al. 2010). The main goal of the present research is to examine the Alternatively, the conventional flood modelling methods usefulness of Remote Sensing (RS), GIS and the frequency were not reliable for accurate prediction (Tehrany et  al. ratio (FR) models for flood susceptibility analysis and map- 2014b; Li et al. 2012). Currently, geospatial techniques pro- ping in the Markham river basin under Morobe province, vide a wide range of data sources for the modelling of flood Papua New Guinea. The main aim of this study is to identify (Wanders et al. 2014). Analytical hierarchy process (AHP) and map out flood risk zones in the Markham river basin. is one of the most popular and satisfactory methods in dis- The objectives of this research are to create wall-to-wall data aster modelling like flood monitoring, mapping and analys- sets that are considered as input into the FR model to catego- ing complex problems (Billa et al. 2006; Yalcin 2008; Chen rize potential flood prone area, create a flood hazard map of et al. 2011). Apart from AHP, multi-criteria decision sup- the Markham river basin and also to carry out impact analy- port approach (MCDA) (Samanta et al. 2016a), weights of sis which can be useful to local government administrators, evidence (WofE) (Rahmati et al. 2016a), logistic regression researchers and planners for devising flood mitigation plans. (LR) (Tehrany et al. 2014a, b, 2015), adaptive neuro-fuzzy interface system (Sezer et al. 2011), artificial neural net- works (ANN) (Pradhan and Buchroithner 2010; Tiwari and Chatterjee 2010) and FR model (Lee et al. 2012; Rahmati Study location and materials used et al. 2016b; Liao and Carin 2009) were the other known acceptable models for hazard analysis. ANN method was The study was carried out in the final basin (basin-14) of Markham, which is located in the Morobe province of PNG used in prediction of flood, where researchers attempted to show relationship between conditioning parameters and the and encloses an area of 1806.85 km . The longitudinal and latitudinal extensions of the study area are 146.09–147.04° outcome (Pradhan and Buchroithner 2010). It was reported that ANN method can handle all inputs which are uncer- east and 6.23–6.78° south, respectively (Fig. 1). The climate is tropical humid with about 4200 mm of average annual tain to extract meaningful information (Lohani et al. 2012). MCDA, RS and GIS techniques are extremely useful in reli- rainfall in the study area. Markham river emanates from the runoff contributed by 12,450 km catchment (Fig. 1a) with able and accurate analysis and mapping of plausible flood prone zones. MCDA approach is suitable for flood analysis huge bed load (Tilley et al. 2006). This is the fourth longest river in PNG born at Finisterre range (approximately 457 m and mapping in the no-data regions and could be practi- cal for local planners in mitigation of flood. AHP model altitude) and flows into the Huon Gulf near to the downtown of Lae, which is the second largest city of PNG after 180 km was applied in China for flood diversion (Zou et al. 2013). According to Chen et  al. (2011), the weakness of AHP of checkered path (Samanta et al. 2016b). The upper basin is dominated by natural forests, steep slopes and rugged ter- model was correlated to its enslavement on the information provided by experts, which is the main source of uncertainty. rain (Pal et al. 2012; Solin 2012). Mining activities for allu- vial gold extraction on the river path and logging activities FR model may be considered as an important method which is easily understandable and can be used to produce accept- in the surroundings areas result in accelerated soil erosion within the basin area. Final 125 km downstream flow of the able flood risk analysis and mapping (Liao and Carin 2009). 1 3 Applied Water Science (2018) 8:66 Page 3 of 14 66 Fig. 1 Location map of the study area: (a) Entire Markham river basin and (b) The study area-sub-basin 14 1 3 66 Page 4 of 14 Applied Water Science (2018) 8:66 Markham river which is covered by sub-basin number 14 of Landsat 8, operational land imager (OLI). A flood inven- was selected for this study (Fig. 1b). tory database was prepared using PNGRIS national data- The Markham river catchment in Morobe province of base, field investigation and by examining remote sensing PNG is exposed to flood due to intensive rainfall, peak dis- data. Every detail of the data sets used in FR model is given charge and physiographic conditions (Tilley et al. 2006; in Table 1. Samanta et al. 2016a). During 2012, higher magnitude of floods was measured by the global flood detection system (Kugler et al. 2007). On March of 2004, Markham riverine Methodology zones experienced an overwhelming flood on the spate of Markham river. Earlier on, during the month of February It is essential to analyse the occurrence of historical flood (2004), 120-m-long road on Lae-Bulolo road was washed events to estimate future flood (Manandhar 2010). So the out just 1.5  km away towards upstream from Markham flood inventory database is the essential factor for flood bridge. The estimated peak discharge was about 2600 m /s susceptibility mapping. Flood inventory map was prepared to 3200 m /s with an average velocity as 3.4 m/s (Tilley after generating 143 flooded points through “create fishnet” et al. 2006). analysis of PNGRIS national database, field investigation To perform flood susceptibility analysis and risk assess- and satellite data before and after flood events. Seventy ments, researcher suggested different factors which are not percent (70%) of total, i.e. 100 flood points, were selected fixed (Tehrany et al. 2015). There are some common condi- randomly as the training data set for flood modelling and tioning factors which indicate their role in flood mapping. the rest 30% or 43 points were used for validating the flood Recent studies by Rahmati et al. (2016a, b) have achieved model (Rahmati et al. 2016b) (Fig. 1b). high accurate results, where they used the least number of Selection of effectual parameters is vital to produce a independent parameters. Total of 15 parameters are exam- flood hazard map in any catchment (Kia et al. 2012). It is ined as a preliminary analysis. Demarcation of flood suscep- always tricky to choose factors unanimously for use in flood tibility zones was carried out using ten (10)-folds of geospa- susceptibility mapping (Tehrany et al. 2014a, b). Flood- tial data sets, viz. land use and land cover (LULC), elevation, related geospatial database like land use/land cover, eleva- slope, topographic wetness index (TWI), surface runoff, tion, slope, topographic wetness index (TWI), surface run- landform, lithology, distance from the main river, soil texture off, landform, lithology, distance from the main river, soil and soil drainage. These geospatial layers are selected after texture and soil drainage were prepared using ArcGIS and consulting local hydrological and natural disaster experts Erdas Imagine software. based on their effectiveness in creating a flood. Wall-to-wall LULC directly or indirectly influence infiltration, geospatial database was developed from remotely sensed evapotranspiration and surface runoff generation. LULC data sets like satellite image and digital elevation model and map was prepared (Fig. 2a) from the Landsat-8 OLI satel- National GIS database of PNG. All data sets were rectified lite imagery through supervised classification technique carefully using Erdas imagine software. UTM projection, (Samanta et al. 2011). The altitude and slope are important zone 55S and WGS-84 datum were selected for the image parameters in flood risk and vulnerability mapping. Vari- and map registration purpose. Advanced space-borne ther- ations of elevations have a definitive impact on climate mal emission and reflection radiometer (ASTER) provides characteristics (Samanta et al. 2012). Such different rain- digital elevation model (DEM) with the spatial resolution fall and temperature regimes engendered varied vegetation of 30 m that was used in this research. Elevation and slope and soil forms (Aniya 1985). Slope controls the surface database were derived from the ASTER DEM. LULC was runoff, the ferocity of water flow aggravating soil erosion derived from optical bands with false colour combination (Adiat et al. 2012) as well as vertical percolation (Youssef Table 1 Data sets used for flood vulnerability mapping Sl. No. Data description Data type/resolution Year Source 1 LANDSAT-8, OLI 15 m Pan-Sharpen 2013 Department of Surveying and Land Studies, The PNG University of Technology 2 ASTER-DEM 30 m 2001 3 Rainfall, Polygon/shape file 2009 PNGRIS—PNG Resource Information System 4 Soil texture and drainage 5 Lithology 6 Landforms 7 Historical flood inventory database Point location 1 3 Applied Water Science (2018) 8:66 Page 5 of 14 66 Fig. 2 Parameters used for FR modelling: (a) land use/land cover, (b) altitude, (c) slope, (d) TWI index, (e) surface runoff, (f) landforms, (g) lithology and (h) soil texture characteristics of the study area 1 3 66 Page 6 of 14 Applied Water Science (2018) 8:66 et al. 2011). Both altitude and slope map were prepared and soil drainage was prepared based on PNGRIS data sets using ASTER data in ArcGIS 3-D analysis algorithms after verifying them based on field observations. (Fig. 2b, c). It is essential to analyse past flood record to estimate the Topographic wetness index (TWI) refers to spatial future flood event in any area (Manandhar 2010). Therefore, distribution of wetness and controls the overland flow of mapping the flood locations from past episodes in the study water. TWI has significant impact on flood mapping. TWI area is instrumental in elucidating the correlation among was calculated based on Eq. 1 (Beven and Kirkby 1979; the flooding and the condition factors. It is essential to pre- Regmi et al. 2010; Qin et al. 2011), and spatial distribution pare an inventory database with high precision (Jebur et al. map of TWI was prepared using ArcGIS (Fig. 2d). 2013). FR model was adopted for this study chosen from a plethora of bivariate statistical techniques. The approach TWI = Ln , offers a quantitative relationship between the ‘frequency of (1) tan B flood episodes’ and various conditioning parameters. The where a is the specific catchment area [a = A/L, total basin FR index was calculated using the following Eq. 5 (Tehrany area (A) divided by length of contour (L)], and B is referred et al. 2014a, b). to the slope in degree. Overland flow of water, called surface runoff, occurs in FSI = FR, (5) full throttle in the aftermath of saturated infiltration when where FSI is the flood susceptibility index and FR is the the surplus water minus saturated infiltration component, frequency ratio for each factor. emanating from the storm, melt water, etc. keep flowing The FR can be expressed as the proportion of flooded area over the Earth’s surface (Pal and Samanta 2011). Surface in the total study area (Eq. 6); it is the ratio of the probabili- runoff in urban areas is normally reinforced owing to ties of the ‘flooded’ to ‘not flooded’ area for a given attribute lack of infiltration surface, which gives rise to pernicious (Bonham-Carter 1994). urban flooding. Surface runoff due to storm rainfall is very significant for forecasting floods which are very sudden, ∕ ∕ ∕ FR = (E F) (M L), (6) flashy and of short duration (Pal and Samanta 2011). Sur - where E is the number of flood episodes for each factor; F face runoff database was generated (Fig.  2e) based on soil is the total number of flood episodes; M is the histogram of conservation service (SCS) model (Eqs. 2–4). a class; L is the total histogram of the study area. The frequency ratio model was used to establish the cor- Q = (P−Ia) (P−Ia + S), (2) relation between historical flood locations and the probable supporting factors. FR value indicates (Table 2) the types where Q is actual surface runoff in mm; P is storm rainfall of correlation between factors and floods. A FR value lower (mm); S is the potential maximum retention (mm), and Ia is than 1 indicates weak correlation; on the other hand a value 0.4S [Initial abstraction (mm)]. of more than 1 refers to strong correlations. The overall So the modified form of Eq.  2 can be expressed in Eq. 3. methodology that was applied for flood susceptibility map- ping is shown as a flowchart in Fig.  3. Q = (P−0.4S) (P + 0.6S). (3) To calculate the value of potential maximum retention (S) of Eq. 3, another simple Eq. 4 was used. Results and discussion S = (25400∕CN) − 254, (4) There are many independent variables, called condition- where CN is curve number of hydrologic soil cover com- ing factors that play specific role in order to perform flood plex, which happens to be a function of soil type, land cover susceptibility mapping (Pradhan 2010; Pourghasemi et al. and antecedent moisture condition (SCS 1972; Kumar et al. 2012; Kia et al. 2012). Spatial distribution and statistical 1991; Rao et al.1996). database for all ten (10) conditioning factors namely, LULC, Flood generally occurs near to the bank of the river and elevation, slope, topographic wetness index (TWI), surface inundates low-lying flood plain areas. Distance from the runoff, landform, lithology, distance from the main river, soil river has significant impact on the flood and its magnitude. texture and soil drainage were constructed with their sub- Distance from main river was considered as one parameter classes (Fig. 2 and Table 2). Classification of satellite data which was developed through proximity analysis in Arc- was done (Samanta et al. 2012) considering nine (9) land GIS. Infiltration varies based on the spatial distribution of use/land cover categories, namely dense forest, low dense soil texture and it controls overland flows and inundation. forest, shrub land, outcrop/cleared/burnt lands, mountain/ Database on landform, lithology, soil texture (Fig. 2f–h) upland grassland, settlement, inland water, river water and 1 3 Applied Water Science (2018) 8:66 Page 7 of 14 66 Table 2 Conditioning parameters used for flood vulnerability mapping through FR model Value Class name or description Histogram % of Histogram Flood numbers % flood numbers Frequency Ratio Land use/land cover 1 Dense forest 50700 2.5 1 1.0 0.40 2 Low dense forest 703420 35.0 33 33.0 0.94 3 Shrub land 533068 26.6 26 26.0 0.98 4 Outcrop/cleared/burnt lands 37806 1.9 0 0.0 0.00 5 Mountain/upland grassland 433316 21.6 28 28.0 1.30 6 Settlement 18728 0.9 0 0.0 0.00 7 Inland water 5406 0.3 0 0.0 0.00 8 River water 133552 6.7 11 11.0 1.65 9 Agriculture 91612 4.6 1 1.0 0.22 Elevation (in m) 1 Up to 100 m 746583 37.2 59 59.0 1.59 2 100–200 m 544304 27.1 25 25.0 0.92 3 200–300 m 301652 15.0 11 11.0 0.73 4 300–400 m 120706 6.0 5 5.0 0.83 5 400–500 m 69673 3.5 0 0.0 0.00 6 500–1000 m 177557 8.8 0 0.0 0.00 7 More than 1000 m 47133 2.3 0 0.0 0.00 Slope (in degree) 1 up to 2.5° 336363 16.8 21 21.0 1.25 2 2.5–5.0° 513214 25.6 40 40.0 1.56 3 5.0–10.0° 563975 28.1 29 29.0 1.03 4 10.0–15.0° 209103 10.4 2 2.0 0.19 5 15.0–20.0° 131856 6.6 5 5.0 0.76 6 20.0–25.0° 100073 5.0 2 2.0 0.40 7 More than 25° 153024 7.6 1 1.0 0.13 Topographic wetness index (TWI) 1 up to 6.0 129577 6.5 1 1.0 0.15 2 6.0–6.5 226488 11.3 6 6.0 0.53 3 6.5–7.0 250979 12.5 3 3.0 0.24 4 7.0–7.5 387731 19.3 16 16.0 0.83 5 7.5–8.0 449732 22.4 32 32.0 1.43 6 8.0–8.5 299489 14.9 28 28.0 1.88 7 More than 8.5 263612 13.1 14 14.0 1.07 Surface runoff (in mm) 1 Up to 50 mm 682498 34.0 41 41.0 1.21 2 50–100 mm 281239 14.0 8 8.0 0.57 3 100–150 mm 486606 24.2 28 28.0 1.16 4 150–200 mm 406985 20.3 13 13.0 0.64 5 More than 200 mm 150280 7.5 10 10.0 1.34 Landform 1 Dissected relict alluvial, colluvial mudflow and fans 75020 3.7 0 0.0 0.00 2 Mountains and hills with weak or no structural 476893 23.8 0 0.0 0.00 control 3 Braided floodplains 296938 14.8 36 36.0 2.43 4 Composite alluvial plains 94495 4.7 10 10.0 2.12 5 Composite bar plain and alluvial fan complex 46996 2.3 1 1.0 0.43 6 Little dissected recent alluvial fans 779505 38.8 41 41.0 1.06 7 Homoclinal ridges and cuestas 77006 3.8 0 0.0 0.00 8 Back plains 46935 2.3 4 4.0 1.71 1 3 66 Page 8 of 14 Applied Water Science (2018) 8:66 Table 2 (continued) Value Class name or description Histogram % of Histogram Flood numbers % flood numbers Frequency Ratio 9 Back swamps 40223 2.0 5 5.0 2.50 10 Hilly terrain with weak or no structural control 27365 1.4 0 0.0 0.00 11 Lake 6416 0.3 0 0.0 0.00 12 Meander floodplains 8914 0.4 1 1.0 2.25 13 Undifferentiated swamps 30902 1.5 2 2.0 1.30 Lithology 1 Pleistocene sediments 75020 3.7 0 0.0 0.00 2 Coarse grained sedimentary 290647 14.5 0 0.0 0.00 3 Alluvial deposits 1344906 67.0 100 100.0 1.49 4 Mixed or undifferentiated igneous 19162 1.0 0 0.0 0.00 5 Mixed or undifferentiated sedimentary 61629 3.1 0 0.0 0.00 6 Limestone 74653 3.7 0 0.0 0.00 7 Mixed sedimentary and limestone 14059 0.7 0 0.0 0.00 8 Mixed sedimentary and volcanic 16871 0.8 0 0.0 0.00 9 Acid to intermediate igneous 27385 1.4 0 0.0 0.00 10 Lake 6416 0.3 0 0.0 0.00 11 Low grade metamorphic 76860 3.8 0 0.0 0.00 Distance from river in m 1 up to 1000 m 393500 19.6 26 26.0 1.33 2 1000–2000 m 243382 12.1 19 19.0 1.57 3 2000–3000 m 206855 10.3 9 9.0 0.87 4 3000–4000 m 175027 8.7 13 13.0 1.49 5 4000–5000 m 149073 7.4 4 4.0 0.54 6 5000–6000 m 133660 6.7 4 4.0 0.60 7 More than 6000 m 706111 35.2 25 25.0 0.71 Soil texture 1 Sandy clay loam 439768 21.9 20 20.0 0.91 2 Sandy loam 129127 6.4 13 13.0 2.02 3 Silty clay loam 377624 18.8 12 12.0 0.64 4 Peat 22672 1.1 1 1.0 0.89 5 Silty loam 22699 1.1 4 4.0 3.54 6 Sandy clay 634345 31.6 28 28.0 0.89 7 River course-gravel 216519 10.8 19 19.0 1.76 8 Sand 66418 3.3 2 2.0 0.60 9 Lake 6567 0.3 0 0.0 0.00 10 Silty clay 91869 4.6 1 1.0 0.22 Soil drainage 1 Well drained 1631997 81.3 62 62.0 0.76 2 Waterlogged (swampy) 45371 2.3 5 5.0 2.21 3 Poorly to very poorly drained 105842 5.3 14 14.0 2.66 4 River course 216519 10.8 19 19.0 1.76 5 Imperfectly drained 1312 0.1 0 0.0 0.00 6 Lake 6567 0.3 0 0.0 0.00 agriculture land (Fig. 2a). The predominant land cover in the study area, but the FR value was calculated as 1.65 which study area is ‘low density forest’ (35%) in the eastern region is highly correlated to flood (Table  2). The elevation ranged and some pockets of north-west and southern region of the from 0 m to 1790 m of the study area. The spatial distribu- study area. The ‘river class’ along the centre (main river) and tion of the elevation map was prepared after reclassifying north (Erap river) of the study area covers only 6.7% of the the elevation layer into seven (7) classes (< 100, 100–200, 1 3 Applied Water Science (2018) 8:66 Page 9 of 14 66 Fig. 3 Methodological flow chart of flood susceptibility mapping 200–300, 300–400, 400–500, 500–1000 and > 1000  m) 7.5–8.0, 8.0–8.5 and > 8.5) as shown in Fig. 2d. The high- where brown colour indicates maximum altitude found in est TWI was recorded in the middle part of the study area north, north-west, and blue indicated < 100 m altitude zone represented with blue colour. Higher TWI value refers in the east and south-east part in the study area (Fig. 2b). higher chances of flooding in a watershed (Rahmati et al. Up to 100 m altitude zone is the predominant class cov- 2016b). The FR value was calculated as 1.88 where TWI ering 37.2% of the total area, which refers to a higher FR varied from 8.0 to 8.5, and 0.15 for the class with TWI value (1.59) than other classes (Table 2). Slope (in degree) < 6.0, respectively (Table 2). Surface runoff map was gen- was calculated from DEM data and reclassified into seven erated based on storm rainfall of 229.6 mm during 21–23 (7) classes, namely < 2.5°, 2.5–5°, 5–10°, 10–15°, 15–20°, October 2012. Maximum surface runoff was calculated as 20–25° and more than 25°. The flat or lower slope gradient 229 mm along the river class. High surface runoff during a (< 5°) area is situated on both sides of the river as shown storm indicates high probability of flood (Pal and Samanta with blue colour (Fig.  2c). Lower the slope gradient, the 2011). The surface runoff database was reclassified into five more is the possibility of flooding and flood events (Rah- (5) classes, namely < 50, 50–100, 100–150, 150–200, and mati et al. 2016a). Two lower slope gradient classes, namely > 200  mm (Fig. 2e). FR value was calculated as 1.34 for < 2.5° and 2.5–5°, indicates higher FR value of 1.25 and highest runoff category (> 200 mm) (Table 2). Entire study 1.56, respectively, whereas > 25° slope area indicated lower area has been categorized into thirteen (13) landform units, FR value of 0.13 (Table 2). namely (i) dissected relict alluvial, colluvial mudflow and Spatial database on TWI was calculated and categorized fans, (ii) mountains and hills with weak or no structural con- into seven (7) classes (< 6.0, 6.0–6.5, 6.5–7.0, 7.0–7.5, trol, (iii) braided floodplains, (iv) composite alluvial plains, 1 3 66 Page 10 of 14 Applied Water Science (2018) 8:66 Fig. 4 Output map through FR model: (a) flood susceptibility index with all flood points and (b) overlay of villages and road on flood suscepti- bility zones (v) composite bar plain and alluvial fan complex, (vi) lit- The lithology of the study area has been categorized into tle dissected recent alluvial fans, (vii) homoclinal ridges (i) Pleistocene sediments, (ii) coarse grained sedimentary, and cuestas, (viii) back plains, (ix) back swamps, (x) hilly (iii) alluvial deposits, (iv) mixed or undifferentiated igneous, terrain with weak or no structural control, (xi) lake, (xii) (v) mixed or undifferentiated sedimentary, (vi) limestone, meander floodplains and (xiii) undifferentiated swamps (vii) mixed sedimentary and limestone, (vii) mixed sedi- (Fig. 2f). The back swamps and meander floodplains have mentary and volcanic, (ix) acid to intermediate igneous, (x) higher FR value of 2.50 and 2.43, respectively (Table 2). lake and (xi) low grade metamorphic (Fig. 2g). Lithology is 1 3 Applied Water Science (2018) 8:66 Page 11 of 14 66 Table 3 Spatial distribution (statistics) of flood vulnerability classes imperfectly drained and (vi) lake. Within the soil drainage in the study area factor, poorly to very poorly drained class had the highest FR value of 2.66, followed by waterlogged (swampy) area with Sl. no. Flood vulner- FR value range Histogram % area able class the FR value of 2.21. Finally, the soil texture map was gen- erated for the study area where ten (10) soil texture classes 1 Very low < 5.0 206164 10.3 could be found (Samanta et al. 2016b), namely sandy clay 2 Low 5.0–7.5 353062 17.6 loam, sandy loam, silty clay loam, peat, silty loam, sandy 3 Medium 7.5–10.0 340554 17.0 clay, river course-gravel, sand, lake and silty clay (Fig. 2h). 4 High 10.0–12.5 718863 35.8 The FR value was calculated as 3.54 for sandy loam and 0.22 5 Very high > 12.5 388965 19.4 for silty clay (Table 2). Thus, the rating of each subclass of all conditioning param- eter was generated based on the FR values as shown in Table 2. an important conditioning parameter in flooding because it FR value varied from 0 to 3.54 in the study area. Calculated has a direct influence on land permeability and thus surface FR values were indicated as weak (< 1) to strong (> 1) correla- runoff (Haghizadeh et al. 2017). The maximum FR value tions with flood occurrence (Lee et al. 2012). Finally, based on of 1.49 was recorded in alluvial deposits which covers 67% FR model in Eq. 5, the flood probability database was devel- of the total study area (Table  2). Flood intensity became oped (Fig. 4a). FR value in the model output varied from 2.66 less in those locations far away from the river and the risk to 19.02. Greater FR value indicated the higher probability was higher in areas near to the river bank. Distances from to flood occurrences. The developed database was reclassi- the river in the range from 4000 to 5000, 5000 to 6000 and fied into five (5) different flood susceptibility zones, namely > 6000 m have a low probability of flooding, whereas dis- very low (less that 5.0), low (5.0–7.5), moderate (7.5–10.0), tances in the range of < 1000 and 1000–2000 m together high (10.0–12.5) and very high susceptibility (more than 12.5) indicate highest FR values (1.42), which demonstrates the (Fig. 4b). The result indicates that about 19.4% land area are highest flood event (Haghizadeh et al. 2017). Soil texture demarcated as very high, 35.8% as high, 17.0% as moderate, and soil drainage are very important factors in flood sus- 17.6% as low and 10.3% as very low flood vulnerable class ceptibility mapping. Well-drained soils produce less surface (Table 3). High to very high vulnerable classes are mostly runoff than poorly drained soil group (Pal et al. 2012). The located along the middle part of the study area (Fig.  4b). soil drainage database was generated with six (6) catego- These high to very high flood susceptibility zones are charac- ries, namely (i) well drained, (ii) waterlogged (swampy), terized with higher runo p ff otentiality, poorly to very poorly (iii) poorly to very poorly drained, (iv) river course, (v) drained soil, alluvial deposits, braided flood plain, lower slope Table 4 Calculation of prediction accuracy and success rate for the flood susceptibility analysis Sl. no. Susceptibility class Verification Accurate (high to very Prediction accuracy Training Success (high to very Success rate (30% flood high class) (70% flood high class) points) points) 1 Very low 0 42 0.977 0 94 0.94 or or 2 Low 0 0 97.7% 94.0% 3 Medium 1 6 4 High 20 45 5 Very high 22 49 Total 43 100 Table 5 Flood susceptibility Susceptibility class FR value range Village % of village under Total population zone and the risk factor flood vulnerability 1 Very low < 5.0 1 0.3 733 2 Low 5.0–7.5 16 4.5 5021 3 Medium 7.5–10.0 83 23.2 30466 4 High 10.0–12.5 183 51.1 93860 5 Very high > 12.5 75 20.9 43432 1 3 66 Page 12 of 14 Applied Water Science (2018) 8:66 gradient, lower elevation and closer to the main river, which calculated based on the 100 flood training points (70%). The are the important conditioning factors for flood susceptibility validation report indicated a higher prediction accuracy of mapping using the FR model. 97.7% which had been enough to validate the FR model There are many models as proposed by different research- that was used for this study. It is obvious that a higher num- ers, but it is very important to evaluate the accuracy and suc- ber of input data sets generate higher accuracy. FR model cess rate to validate the model used for flood susceptibility requires a large number of flood points as training like 70% analysis (Chung and Fabbri 2003; Tien Bui et al. 2012). The (or 100 points) to generate Frequency Ratio, whereas a less modelled output through FR model is validated in regard of number of input data do not fall under all classes of every success rate and prediction accuracy. The value of 1.0 repre- parameter. In this point of view, we used 70% flood points sented the highest accuracy, which indicates the capability of for flood map development and 30% (70–30) for the valida- the model in predicting natural hazards without any bias (Prad- tion process. In case we used 60–40 or 50–50, the results han and Buchroithner 2010). Success rate was calculated using are not same. The accuracy varied 5–10% lower than 70–30 100 training flood locations and prediction of accuracy, using selection method. To validate the superiority, we sought to remaining 43 flood locations which were not used during the compare the method used in this study (FR model) with model building. Class ranges from ‘high’ to ‘very high’ sus- another multi-criteria decision support approach (MCDA) ceptibility are considered as potential o fl od that might occur in (Samanta et al. 2016a) which was conducted in the same future. Success rate and prediction rate are calculated as 0.94 river basin. As per our assessment, FR model produced bet- and 0.977, respectively (Table 4). So the prediction accuracy ter results compared to MCDA. This FR model can be used corroborates to 97.7% which validates the FR model used in in any other geographical area to develop a flood risk map this flood susceptibility analysis. which can help planners and decision makers to perform Finally, an impact analysis was done to assess the risk fac- proper flood management in future. tor by probable flood occurrences on the local community and Acknowledgements The authors are thankful to the PNGUNITECH population (Table 5). About 75 villages (20.9%) are situated (Papua New Guinea University of Technology) and to the Department in the very high flood vulnerable zone and 183 (51.1%) in the of Surveying and Land Studies for all the facilities made available high flood vulnerable zone (Fig.  4b). A total of 137292 peo- and availed for the work as a researcher. Satellite digital data avail- able from USGS Global Land Cover Facility and used in this study ple are living in those vulnerable zones which require special are also duly acknowledged. The authors gratefully acknowledge the attention from various levels of governments to take appropri- anonymous reviewers for providing their critical comments to improve ate actions to prevent and mitigate future flood occurrence. the quality of this article. Compliance with ethical standards Conclusion Conflict of interest The authors declare that there is no conflict of in- In the current research, the FR model is used to analyse terest for the publication of this article. flood susceptibility zone in the lower part of the Markham Open Access This article is distributed under the terms of the Crea- river basin (Sub-basin 14). Ten independent conditioning tive Commons Attribution 4.0 International License (http://creat iveco factors, like LULC, elevation, slope, TWI, surface runoff, mmons.or g/licenses/b y/4.0/), which permits unrestricted use, distribu- landform, lithology, distance from the main river, soil tex- tion, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the ture and soil drainage were derived from the geospatial data Creative Commons license, and indicate if changes were made. sets and used as input into the FR model towards flood prone area mapping. As the result suggests, these ten variables are likely to be major factors to map flood-affected zone. The FR model was used to establish the relationship between References past and future flood occurrences. 143 flooded points were generated through “create fishnet” analysis of the PNGRIS Adiat KAN, Nawawi MNM, Abdullah K (2012) Assessing the accu- national database, field investigation and satellite data before racy of GIS-based elementary multi criteria decision analysis as a spatial prediction tool–a case of predicting potential zones of and after oo fl d events. Random sampling technique was used sustainable groundwater resources. J Hydrol 440–41:75–89 to select 70% input flood data and to calculate frequency Aniya M (1985) Landslide-susceptibility mapping in the Amahata river ratio for each parameters and their subclasses (Rahmati basin, Japan. Ann Assoc Am Geogr 75(1):102–114 et al. 2016b). The use of stratified random sampling would Beven and Kirkby (1979) A physically based variable contributing area model of basin hydrology. Hydrol Sci Bull 24:43–69 be difficult to generate frequency ratio for all classes under Billa L, Shattri M, Mahmud AR, Ghazali AH (2006) Comprehensive each parameter. So random sampling has been mooted by the planning and the role of SDSS in flood disaster management in authors as the prudent choice in this type of studies. FR val- Malaysia. Dis Prev Manag 15:233–240 ues for different categories of all conditioning factors were 1 3 Applied Water Science (2018) 8:66 Page 13 of 14 66 Bonham-Carter GF (1994) Geographic information systems for geo- Pal B, Samanta S, Pal DK (2012) Morphometric and Hydrological scientists: modeling with GIS. In: Bonham-Carter F (ed) Com- analysis and mapping for Watut watershed using Remote Sensing puter methods in the geosciences. Pergamon, Oxford and GIS techniques. Int J Adv Eng Technol 2(1):357–368 Chen YR, Yeh CH, Yu B (2011) Integrated application of the ana- Patel DP, Srivastava PK (2013) Flood hazards mitigation analysis lytic hierarchy process and the geographic information system using remote sensing and GIS: correspondence with town plan- for flood risk assessment and flood plain management in Tai- ning scheme. Water Resour Manag 27:2353–2368 wan. Nat Hazards 59(3):1261–1276 Pourghasemi HR, Pradhan B, Gokceoglu C (2012) Application of fuzzy Chung CF, Fabbri AG (2003) Validation of spatial prediction models logic and analytical hierarchy process (AHP) to landslide suscepti- for landslide hazard mapping. Nat Hazard 30:451–472 bility mapping at Haraz watershed, Iran. Nat Hazards 63:965–996 Du J, Fang J, Xu W, Shi P (2013) Analysis of dry/wet conditions Pourghasemi HR, Moradi HR, Aghda SMF, Gokceoglu C, Pradhan B using the standardized precipitation index and its potential use- (2014) GIS-based landslide susceptibility mapping with probabil- fulness for drought/flood monitoring in Hunan Province China. istic likelihood ratio and spatial multi-criteria evaluation models Stoch Env Res Risk Assess 27(2):377–387 (North of Tehran, Iran). Arab J Geosci 7(5):1857–1878 Haghizadeh A, Siahkamari S, Haghiabi AH, Rahmati O (2017) Pradhan B (2010) Flood susceptible mapping and risk area delinea- Forecasting flood-prone areas using Shannon’s entropy tion using logistic regression, GIS and remote sensing. J Spatial model. J Earth Syst Sci 126:39. https ://doi.org/10.1007/s1204 Hydrol 9:1–18 0-017-0819-x Pradhan B, Buchroithner MF (2010) Comparison and validation of Huang X, Tan H, Zhou J, Yang T, Benjamin A, Wen SW, Li S, Liu landslide susceptibility maps using an artificial neural network A, Li X, Fen S, Li X (2008) Flood hazard in Hunan province of model for three test areas in Malaysia. Environ Eng Geosci China: an economic loss analysis. Nat Hazards 47:65–73 16:107–126 Jebur MN, Mohd Shafri HZ, Pradhan B, Tehrany MS (2013) Per- Pradhan B, Hagemann U, Tehrany MS, Prechtel N (2014) An easy pixel and object oriented classification methods for mapping to use ArcMap based texture analysis program for extraction of urban land cover extraction using SPOT 5 imagery. Geocarto flooded areas from TerraSAR-X satellite image. Comput Geosci Int. https ://doi.org/10.1080/10106 049.2013.84894 4 63:34–43 Khan H, Haider S, Saeed K, Ali N (2008) Assessment of potable Qin et  al (2011) An approach to computing topographic wetness water quality of Kohat division and its impact on health. J Chem index based on maximum downslope gradient. Precision Agric Soc Pak 30:246–250 12:32–43 Kia MB, Pirasteh S, Pradhan B, Rodzi MA, Sulaiman WNA, Moradi Rahmati O, Zeinivand H, Besharat M (2016a) Flood hazard zoning in A (2012) An artificial neural network model for flood simula- Yasooj region, Iran, using GIS and multi-criteria decision analy- tion using GIS: Johor River Basin, Malaysia. Environ Earth Sci sis. Geomat Nat Hazard Risk. https ://doi.or g/10.1080/19475 67:251–264705.2015.10450 43 Kisi O, Nia AM, Gosheh MG, Tajabadi MRJ, Ahmadi A (2012) Rahmati O, Pourghasemi HR, Zeinivand H (2016b) Flood susceptibil- Intermittent streamflow forecasting by using several data driven ity mapping using frequency ratio and weights-of-evidence mod- techniques. Water Resour Manag 26(2):457–474 els in the Golastan Province, Iran. Geocarto International. https Kugler Z, Groeve TD (2007) The Global Flood Detection system; ://doi.org/10.1080/10106 049.2015.10415 59 EUR 23303 EN. Office for Official Publications of the European Rao KV, Bhattacharya AK, Mishra K (1996) Runoff estimation by Communities, Luxembourg, pp 1–45 curve number method- case studies. J Soil Water Conserv 40:1–7 Kumar P, Tiwari KN, Pal DK (1991) Establishing SCS runoff curve Regmi NR, Giardino JR, Vitek JD (2010) Modeling susceptibility to number from IRS digital database. J Indian Soc Remote Sens landslides using the weight of evidence approach: Western Colo- 19(4):245–251 rado, USA. Geomorphology 115:172–187 Lee MJ, Kang JE, Jeon S (2012) Application of frequency ratio Saha AK, Gupta RP, Sarkar I, Arora KM, Csaplovics E (2005) An model and validation for predictive flooded area susceptibility approach for GIS-based statistical landslide susceptibility zona- mapping using GIS. In: Proceedings of Geoscience and Remote tion with a case study in the Himalayas. Landslides 2(1):61–69 Sensing Symposium (IGARSS), 2012 IEEE International. Samanta S, Pal DK, Lohar D, Pal B (2011) Preparation of digital data Munich, 895–898 sets on land use/land cover, soil and digital elevation model for Li XH, Zhang Q, Shao M, Li YL (2012) A comparison of parameter temperature modelling using Remote Sensing and GIS techniques. estimation for distributed hydrological modelling using auto- Indian J Sci Technol 4(6):636–642 matic and manual methods. Adv Mater Res 356–360:2372–2375 Samanta S, Pal DK, Lohar D, Pal B (2012) Interpolation of climate var- Liao X, Carin L (2009) Migratory logistic regression for learning iables and temperature modeling. Theor Appl Climatol 107(1):35– concept drift between two data sets with application to UXO 45. https ://doi.org/10.1007/s0070 4-011-0455-3 sensing. IEEE Trans Geosci Remote Sens 47:1454–1466 Samanta S, Koloa C, Pal DK, Palsamanta B (2016a) Flood risk analysis Lohani A, Kumar R, Singh R (2012) Hydrological time series mod- in lower part of Markham River based on multi-criteria decision eling: a comparison between adaptive neuro-fuzzy, neural net- approach (MCDA). Hydrology 3(3):29. https ://doi.org/10.3390/ work and autoregressive techniques. J Hydrol 442:23–35hydro logy3 03002 9 Manandhar B (2010) Flood Plain Analysis and Risk Assessment Samanta S, Koloa C, Pal DK, Palsamanta B (2016b) Estimation of of Lothar Khola, MSc Thesis, Tribhuvan University, Phokara, potential soil erosion rate using RUSLE and E30 model. J Model Nepal, pp. 64 Earth Syst Environ 2(3):1–11. https ://doi.or g/10.1007/s4080 Markantonis V, Meyer V, Lienhoop N (2013) Evaluation of the envi- 8-016-0206-7 ronmental impacts of extreme floods in the Evros River basin SCS (1972) Soil Conservation Department, Handbook of Hydrology. using contingent valuation method. Nat Hazard 69:1535–1549 Ministry of Agriculture, New Delhi Merz B, Kreibich H, Schwarze R, Thieken A (2010) Assess- Sezer EA, Pradhan B, Gokceoglu C (2011) Manifestation of an adap- ment of economic flood damage. Nat Hazard Earth Syst Sci tive neuro-fuzzy model on landslide susceptibility mapping: 10:1697–1724 Klang Valley Malaysia. Expert Syst Appl 38(7):8208–8219 Pal B, Samanta S (2011) Surface runoff estimation and mapping using remote sensing and geographic information system. Int J Adv Sci Technol 3(2):106–114 1 3 66 Page 14 of 14 Applied Water Science (2018) 8:66 Solin L (2012) Spatial variability in the flood vulnerability of urban improving operational flood forecasting. Hydrol Earth Syst Sci areas in the headwater basins of Slovakia. Flood Risk Manag 18:2343–2357 5:303–320 Wang HB, Wu SR, Shi JS, Li B (2013) Qualitative hazard and risk Talei A, Chua LHC, Quek C (2010) A novel application of a neu- assessment of landslides: a practical framework for a case study rofuzzy computational technique in event-based rainfall–runoff in China. Nat Hazard 69(3):1281–1294 modeling. Expert Syst Appl 37(12):7456–7468 WHO (2003) World Health Organization. Disaster Data-key Trends Tehrany MS, Pradhan B, Jebur MN (2014a) Flood susceptibility map- and Statistics in World Disasters Report; WHO: Geneva, Switzer- ping using a novel ensemble weights-of-evidence and support land. http://www.ifrc.org/PageF iles/89755 /2003/43800 -WDR20 vector machine models in GIS. J Hydrol 512:332–343 03_En.pdf (Accessed 5 April 2017) Tehrany MS, Lee MJ, Pradhan B, Jebur MN, Lee S (2014b) Flood Yalcin A (2008) GIS-based landslide susceptibility mapping using susceptibility mapping using integrated bivariate and multivariate analytical hierarchy process and bivariate statistics in Ardesen statistical models. Environ. Earth Sci 72:4001–4015 (Turkey): comparisons of results and confirmations. CATENA Tehrany MS, Pradhan B, Jebur MN (2015) Flood susceptibility analysis 72:1–12 and its verification using a novel ensemble support vector machine Youssef AM, Pradhan B, Hassan AM (2011) Flash flood risk esti- and frequency ratio method. Stoch Environ Res Risk Assess mation along the St.Katherine road, southern Sinai, Egypt using 29:1149–1165. https ://doi.org/10.1007/s0047 7-015-1021-9 GIS based morphometry and satellite imagery. Environ Earth Sci Tien Bui D, Pradhan B, Lofman O, Revhaug I, Dick OB (2012) Land- 62:611–623 slide susceptibility mapping at Hoa Binh province (Vietnam) Zou Q, Zhou J, Zhou C, Song L, Guo J (2013) Comprehensive flood using an adaptive neuro-fuzzy inference system and GIS. Comput risk assessment based on set pair analysis-variable fuzzy sets Geosci 45:199–211 model and fuzzy AHP. Stoch Env Res Risk Assess 27(2):525–546 Tilley JR, Kell RAM, Haro A (2006) Protecting the Markham Bridge, Morobe Province, PNG. In: Proceedings of the 9th International Publisher’s Note Springer Nature remains neutral with regard to River Symposium, Brisbane, Australia, 4–7 September jurisdictional claims in published maps and institutional affiliations. Tiwari MK, Chatterjee C (2010) Uncertainty assessment and ensemble flood forecasting using bootstrap based artificial neural networks (BANNs). J Hydrol 382(1):20–33 Wanders N, Karssenberg D, de Roo A, de Jong SM, Bierkens MFP (2014) The suitability of remotely sensed soil moisture for 1 3

Journal

Applied Water ScienceSpringer Journals

Published: Apr 21, 2018

References

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off