An Assessment of the Integrated Multi-Criteria and New Models Efficiency in Watershed Flood Mapping

Iran is one of the flood-prone areas in the world with inappropriate climatic patterns. In this study, flood risk maps in three scenarios by combining Analytic Hierarchy Process (AHP), Analytical Network Process (ANP) and Fuzzy Analytic Hierarchy Process (FAHP) models with Ordered Weighted Average (OWA), Weighted Linear Combination (WLC) models., Local Weighted Linear Combination (LWLC) and two new models, Weighted Multi-Criteria Analysis (WMCA) and Geo Technique for Order of Preference by Similarity to Ideal Solution (Geo TOPSIS) were prepared from Heraz watershed in northern Iran. The analysis of the results of the AHP model in the first scenario, the ANP model in the second scenario and the FAHP model in the third scenario show that the criteria of precipitation, slope, land use, elevation, drainage density and distance to river are the most important criteria for the occurrence of floods in Haraz basin. Evaluation of flood risk models show that on average, about 70%, 20%, 8%, and 2% of Haraz basin are in medium, low, high, and no flood risk situations, respectively. Geographically, the southeastern and central parts are in high and low flood risk, respectively, and other parts of the basin are in medium risk. In this basin, many forest lands, pastures, agriculture and population centers are in medium and high risk of flooding. Generally, based on the obtained results, WMCA and Geo TOPSIS models along with WLC, LWLC and OWA models are effective methods for flood risk studies and based on the obtained results, Haraz basin needs necessary planning for flood risk management.


Introduction
Natural hazards such as flood have always been a major concern in the world due to their destructive effects on natural resources, human and animal life (Shrestha et al. 2019). In the last few decades, due to climate changes (Silva et al. 2022) and the destructive effects of humans on the natural environment (Zope et al. 2016), floods have occurred about 40% more than other natural hazards in the world, including Iran. The damage caused by flood in 20 years  was 75mil.USD per year (Hirabayashi et al. 2013;Shahabi et al. 2021. Dueto the geographical location of Iran in arid and semi-arid latitudes, it has unfavorable climatic conditions, so that the northern parts have high rainfall and the southern parts have low rainfall. On the other hand, due to the large changes in land cover in the northern parts and the low density of land use in the southern parts, the occurrence of floods has been growing in recent years (Sadeghi-Pouya et al. 2016;Amini et al. 2022). During the last few decades, 3,700 floods have occurred in Iran and caused damages of about 3.6 billion dollars (Arabameri et al. 2020). The most recent floods in Iran occurred in 2019 and 2022, which caused a lot of damage in 300 cities and 5148 villages, most of which were in the northern parts (Khosravi et al. 2020;Rajabi et al. 2020). Haraz watershed is one of the flood-prone areas due to the irregular distribution of rainfall, topography and high density of drainage, which requires flood studies (Pirnia et al. 2019;Sharifi 2020). So far, several models such as; Machine learning (Costache et al. 2022), hydrological (Nogherotto et al. 2022), statistical (Akay 2021), Multi Criteria Analysis (MCA) (Chen 2022), (Vasconcellos et al. 2021Dahri et al. 2022Jang et al. 2021;Mahmoody Vanolya and Jelokhani-Niaraki 2021;Arabameri et al. 2018;Uddin et al. 2019;Pathan et al. 2022;Mubeen et al. 2021;Afsari et al. 2022;Chauhan et al. 2016;Saha et al. 2022;Ogato et al. 2020;Xiao et al. 2018;Yariyan et al. 2020;Wei et al. 2022;Doorga et al. 2022;Li et al. 2022;Fernandez et al. 2016;Hidayah et al. 2022;Mudashiru et al. 2022;Duan et al. 2022;Dandapat and Panda 2017;Wang et al. 2022) have been used for flood risk studies. In this study, two new methods Geo Technique for Order of Preference by Similarity to Ideal Solution (GeoTOPSIS) (QGIS Python Plugins Repository 2020) and Weighted Multi-Criteria Analysis (WMCA) (QGIS Python Plugins Repository 2020) and the Local Weighted Linear Combination (LWLC) model were examined to prepare Haraz basin flood risk maps, and finally the results were compared with the Weighted Linear Combination (WLC) and Ordered Weighted Average (OWA) models.
For this purpose, first the important criteria in flood events were identified through interviews with natural hazards experts. Then, the weight of each criterion is determined using the Analytic Hierarchy Process (AHP) model, the Analytical Network Process (ANP) model, the Fuzzy Analytic Hierarchy Process (FAHP) model, and finally, each of the AHP, ANP, and FAHP models is used with the GeoTOPSIS, WMCA models, WLC, OWA and LWLC were combined, and finally the flood risk maps of Haraz basin were prepared in the form of three scenarios and in four classes of High, Medium, Low and No hazard of flood. flank of the Alborz mountain range and Damavand volcano, as the highest point in the Middle East, is located in the south of this basin. The elevation of the basin in the south of Amol city is 202 to 5671 m in the southern part at Damavand volcano and its area is 4068.75km 2 . Climatically, the southern regions of the basin have a mountainous climate and the northern regions of the basin have a moderate and the average annual rainfall in the basin is 430 mm (Iran Meteorological Organization 2019).

Satillate Image and Ancillary Data
In this study, a Landsat-8 image scene was used to prepare the 2020 land use map. The date of the images was selected according to the natural, topographic and climatic conditions of the study area with minimum cloud cover (0.17) and maximum density of land uses. Google Earth images were also used for the river flow layer, soil map, geological map, Digital elevation model (DEM) layer from the Shuttle Radar Topography Mission (SRTM) data set, Global Positioning System data and meteorological data (Tables 1 and 2).

Reference Data
The reference data in this study was through field visit using Garmin GPS device MAP 65 s. This device is a product of Taiwan with a 3-inch color touch screen and a precise QUAD HELIX antenna capable of tracking GPS, GLONASS and Galileo satellites. The accuracy of this device is ideally less than 2 m. GPS points were taken from May 20, 2020 to June 10, 2020 to prepare and validate the 2020 land use map from agricultural, forest, pasture, water  bodies, barren, and residential lands. The reference data were randomly divided into two parts: the first part included 70% of the points to prepare the land use map and the second part included 30% of the points were used for the validation of the land use map.

Image Prepressing
For the pre-processing operation, Landsat-8 image was first referenced using ground control points (collected using GPS) with root mean square error (RMSE) less than 0.5 pixels (Pal and Ziaul 2017), then atmospheric and radiometric corrections to The arrangement was done by converting Digital number (DN) to Radiance and applying the Atmospheric & Topographic Correction (ATCOR) algorithm (Wang et al. 2021) in ENVI5.3 and Erdas imagine 2015 software. The ATCOR algorithm is one of the radiation transfer models that are used to model electromagnetic waves to information such as longitude and latitude of the studied area, date and time of measurements, height of the sensor at the time of imaging, average height of the area, atmospheric models such as mid-latitude summer, mid-latitude winter and tropical (Fig. 2). Satellite images in the form of Randince require information about the wavelength and the atmospheric field of view of the studied area at the time of imaging (Jensen 2015).

LULC Map
The 2020 land use map was prepared using GPS points and training data by applying Support vector machine (SVM) algorithm in ENVI5.3 software in 6 classes (Table 3). Training

Accuracy Assessment
The accuracy of the 2020 land use map was evaluated using GPS data. In this study, the parameters of Overall Accuracy, User Accuracy, Producer Accuracy and Kappa Coefficient were calculated to validate the 2020 land use map (Das and Angadi 2022).

Flood Hazard Maps
Flood risk maps were prepared in three scenarios. In the first scenario, the AHP model was combined with the WLC, LWLC, OWA, GeoTOPSIS and WMCA models, in the second scenario, the ANP model was combined with the WLC, LWLC, OWA, GeoTOPSIS and WMCA models, and in the third scenario, the FAHP model was combined with the WLC, LWLC, OWA models. GeoTOPSIS and WMCA were combined and finally hazard maps were prepared in four classes of High, Medium, Low and No hazards.

Flood Conditioning Criterions
In the present study, the effective criteria for investigating the flood risk of Haraz basin were identified through field works, interviews with experts and the review of related studies. These criteria include elevation, slope, drainage density, distance to river, flow accumulation, precipitation, land use, Lithology, soil, Topographic Wetness Index (TWI), curvature, Stream Transport Index (STI), Stream Power Index (SPI) and Topographic Ruggedness Index (TRI). Then, all the criteria were processed and linearly normalized (Eq. (1)) in ARCGIS 10.4 environment (Msabi and Makonyo 2021).
Here X i is the normalized factor, Ci is the raw standard values, C min is the minimum standard value, C max is the maximum standard value, and R is the standard normal range (0-1) (Li et al. 2020). (1)

Ordered Weighted Average (OWA)
The OWA model (Eq. (2)) is one of the multi-criteria methods that is not limited to fuzzy sets and in which a combination of Boolean overlapping rules and weighted linear combination is available (Malczewski and Liu 2014). This concept includes ordinal weights that are different from standard weights. Criterion weights are assigned to the used criteria and ordinal weights are assigned to the criteria values location by location (Rinner and Malczewski 2002). With this model, a wide range of results (maps) can be obtained by determining and applying a suitable set of sequential weights in spatial decisions (Ferretti and Pomarico 2013).
Here zi1 ≤ …… ≤ zin is obtained by sorting the values of a criterion. vj is the ordinal weight and wj is the criterion weight (Malczewski 2006). This model includes two main characteristics of degree of Orness (Eq. (3)) and the amount of Trade off relationship (Eq. (4)). The degree of Orness is related to the position of the OWA operator in and (Minimum) and or (Maximum) relationships and indicates the presence of risk (value of 1) and the absence of risk (value of 0) in the decision. Trade off relationship also shows the effectiveness of one index from other indices (Ebrahimian Ghajari et al. 2017).
v j is the ordinal weight of the criterion with the i-th rank and n is the number of criteria.

Ordered Weighted
Ordinal weights are a method to control the criteria that are assigned to the location of the cells and allow the decision maker to choose the criteria that are more important in his opinion; Consider with the same weight and importance. Also, sequential weights provide the possibility of controlling the risk level for the decision maker (Yager 1988).

Weighted Linear Combination (WLC)
WLC is based on the concept of weighted average. In this model, the analyst gives weights to each criterion based on the relative importance of each criterion, then the final weight for each criterion is obtained by multiplying the relative weight by the value of that criterion (Eq. (5)). After the final value of each alternative is determined, the alternative with the highest value will be the most suitable for the desired goal (Hwang and Yoon 1981).
Here n is the number of criteria, x ij is the fuzzified value of criterion j, and w j is the weight of criterion j (Ghosh and Lepcha 2019). (2)

Local Weighted Linear Combination (LWLC)
LWLC is a local form of the WLC model introduced by Malczewski (2011). By changing the weight of the criteria using different normalization methods such as Identity normalization, Maximum score normalization and Score range normalization, this model estimates the value of the criteria in local situations and thus compensates for the lack of uniformity in the evaluation process in decision making (Eq. (6)).
Here V(S jc ) is the overall score of j in neighbourhood of c, w ic is the local weight of criterion i in neighbourhood of c, v i S jc is the value of criterion i at location of j when criterion i is standardized by all values of i in neighbourhood of c.
Here, w ic is the local weight of criterion i in the neighbourhood c, Wi is the total (global) weight of criterion i, r ic is the value of criterion i in the neighbourhood c, r i is the total (global) value of criterion i (Eq. (7)). The local weight of each criterion (S) must be less than or equal to one and greater than or equal to 0, and the sum of the weights of all criteria for location j must be equal to 1 (Carter and Rinner 2014).

Weighted Multi-Criteria Analysis (WMCA)
The WMCA model was created in 2020 for studies of utility, sustainability and zoning such as flood risk in QGIS software. In this model, the weight of each criterion is combined with the classified criteria and Grad value and the final map is prepared. Grad values in this model are between 0 and 10, which are assigned to criteria classes based on importance (relative to the purpose of the study). So that high values of Grad (such as 10) are assigned to classes of criteria that are very important in the subject under study (QGIS Python Plugins Repository 2020).

Geo Technique for Order of Preference by Similarity to Ideal Solution (GeoTOPSIS)
The logic of this method defines the positive and negative ideal solution. The ideal (positive) solution is the solution that increases the benefit criterion and decreases the cost criterion. The optimal option is the option that has the smallest and largest distance from the positive and negative ideal solution, respectively. In other words, in the ranking of options using the TOPSIS method, the options that are most similar to the ideal solution are ranked higher (Najafabadi et al. 2016). During the last few years, the mentioned model has been developed based on the use of GeoTOPSIS vector data in QGIS software (QGIS Python Plugins Repository 2020). In this model, the weight of the criteria is calculated based on AHP, ANP and FAHP models, and the user can prepare the final map after defining the weight of the criteria. This model includes five stages (Behzadian et al. 2012): w s r sc ∕r 1 Step 1. Creating a matrix based on m criteria and n options: Step 2. Creating a normalized matrix (n ij ) Here, i = 1,2,3,…,n and j = 1,2,3,…,m.
Step 3. Calculation of the normalized decision weight matrix (Vij).
Here, w i is the weight of the i-th index or criterion of Step 4. Determining positive and negative ideal solution: Here, i depends on profit criteria and j depends on cost criteria.
Step 5. Calculation of distance criteria: Here, d + j is ideal and d − j is minimal alternative respectively. Step 6. Determining the coefficient of relative proximity with the i-th option (Ci) to the ideal solution: Step 7. Ranking the alternatives based on C i , which varies between 0 and 1. C i = 1 indicates the highest rank and Ci = 0 indicates the lowest rank.

Criterions Weight
The weights of criteria were calculated using ANP, AHP and FAHP models with compatibility rates of 0.4, 0.2 and 0.6, respectively. The weights of the criteria in these models are very close to each other and have little differences. This problem has caused that the types of important and non-important criteria for Haraz watershed flood events are the same. According to the results of ANP, AHP and FAHP models, the parameters of precipitation, slope, land use, height, drainage density and distance from the river have the most weight compared to other criteria and these criteria are important criteria for the occurrence of floods. On the other hand, flow accumulation, Lithology, Soil, Curvature, TRI, STI, SPI criteria with less weight are less important for flood occurrence.

Flood Hazard Assessment
The  (Table 4). In addition to medium and high flood risk classes, low flood risk and no risk also exist in Haraz basin. The largest area of the low flood risk class can be seen in the central parts (Fig. 4). In addition to the central parts, there are very few in other parts of the basin, such as the northern, western and eastern parts of the Haraz basin. A large area of low flood risk was determined in the three scenarios of the WMCA model and the AHP-ANP-OWA model, and the lowest area of this class was also determined in the ANP-WLC model (7823.44 hectares), AHP-LWLC (38,921.6 hectares) and FAHP-AHP-OWA model (52,894.3 hectares). Comparison of low and high flood risk states shows that high risk is more than low risk in AHP-WLC, LWLC and ANP-WLC models, and low risk is higher than high flood risk in other models. Also, flood-free areas have the smallest size in Haraz basin in all models, except FAHP-WMCA model (23,206.2 hectares). Spatial monitoring of risk-free areas shows that these areas in AHP-WLC, AHP-ANP-OWA, ANP-WLC, ANP-FAHP-OWA, FAHP-WLC and FAHP-AHP-OWA models are mostly in the southwestern half and in the models AHP-LWLC, ANP-LWLC and FAHP-LWLC have been determined in the eastern parts of Haraz basin. Also, in the new models AHP-WMCA, ANP-WMCA, FAHP-WMCA, AHP-GeoTOPSIS, ANP-Geo-TOPSIS and FAHP-GeoTOPSIS, there are areas without flood risk in the central and northern parts of Haraz basin, and the area of the northern areas without flood risk is less than the central areas (Fig. 5).

Investigating the Land Use Conditions
Investigating the effect of land use on flood risk is an important step for flood occurrence studies. In this study, a land use map of Haraz basin was prepared with an overall accuracy of 93% and a Kappa coefficient of 90% (Table 5). Examining the land use map shows that many parts of the basin, except the northern parts, have pasture lands. There are forest lands in the northern parts of the watershed and barren lands in the southern parts. There are also agricultural lands in the northern and southeastern parts, residential lands in the southeastern, northern, and northwestern parts, and water areas in the southern parts (Fig. 6). Comparison of land use and flood risk maps shows that a large part of pasture, forest, agricultural, residential and water areas in the southeastern parts are in high flood risk conditions. The high flood risk areas of these lands are more in AHP-WLC and ANP-WLC models and less in AHP-ANP-OWA, ANP-WMCA, ANP-GeoTOPSIS and FAHP-WMCA models (Tables 6, 7 and 8).
Also, the forest and agricultural lands in the northern parts are in medium and high flood risk, but the agricultural lands in the north-western parts and pasture lands in these areas are in medium risk in all flood risk models, except for the FAHP-WMCA model.
Due to the large area of low flood risk in the FAHP-WMCA model, agricultural lands in the northwestern parts are at low flood risk in this model. Also, the western and southwestern parts, which mostly have pasture and barren lands, are also with medium and low flood risk. The low flood risk in FAHP-WMCA and ANP-WMCA models is higher than other models.
The largest area in the central part that is in low flood risk is shown in the FAHP-WMCA model.
In addition to natural land uses, residential lands are also in an unfavorable situation in terms of flood occurrence, because a large part of the population living in the Haraz basin, including urban and rural centers, are in medium, high, and low flood risk situations (Tables 9, 10 and 11). These results can be shown in all scenario models. The population centers on the banks of the rivers in the south-east and north-west parts are at high and medium risk of flooding. The highest number of population is shown in high flood risk areas in LWLC-FAHP, WLC-ANP and LWLC-AHP models.
The population centers of the central, southern, northwestern and western parts are facing moderate and low flood risk. But a large part of the population of this area is facing the risk of flooding.

Discussion
According to the results obtained in this study, Haraz basin is one of the flood-prone areas in Mazandaran province and northern Iran. Various criteria have a significant impact on the occurrence of floods in this area, so that according to the AHP, ANP and FAHP models,  the parameters of precipitation, slope, land use, height, drainage density and distance from river flows are the main factors of occurrence in Haraz basin. Examining the flood risk maps and criteria shows that the southeastern parts of Haraz basin, which have high flood risk, have more rainfall, lower slope and height, and higher drainage density than other parts. Also, the central parts with low flood risk status have less rainfall, higher altitude and slope, lower drainage density than other parts, which shows the impact of different criteria in the occurrence of floods in this area. This situation is shown in all the models presented in the three scenarios with a slight difference. The results of different scenario models were closely related to each other, so that the flood risk classes in terms of location in the models of Ordered Weighted Average (OWA), Weighted Linear Combination (WLC), Local Weighted Linear Combination (LWLC), Weighted Multi-Criteria Analysis (WMCA) and Geo Technique for Order of Preference by Similarity to Ideal Solution (GeoTOPSIS) are very similar in the three scenarios, but in terms of the area of the flood risk classes, slight differences can be seen in the models. The biggest difference is related to the high risk and low risk classes, and the lowest difference is related to the medium risk and no flood risk classes. One of the main reasons for the similarity of the results of the flood risk models is the closeness of the weights of the criteria in the AHP, ANP and FAHP models. In all three  Doorga et al. (2022) and Mudashiru et al. (2022) had good results like the present study. The difference between this study and the previous studies is that in the previous studies, each of the OWA and WLC models were combined with only one of the AHP, ANP, and FAHP models, but in the present study, each of the OWA and WLC models were combined separately with the AHP, ANP, and FAHP were combined and flood hazard maps were prepared in three scenarios. This work shows the application of OWA and WLC models in combination with different models. Also, in this research, the effectiveness of two new models, GeoTOPSIS and WMCA, and the LWLC model in combination with ANP, ANP and FAHP models were evaluated to prepare the flood risk map. So far, GeoTOPSIS, WMCA and LWLC models have not been used for flood risk studies. Examining and comparing these models with OWA and WLC models shows that the results of GeoTOPSIS, WMCA and LWLC models are aligned with WLC and OWA models in three scenarios. Considering the closeness of the results of the flood risk maps in three scenarios, the combination of these maps with the land use map and population centers showed similar results. As the surveys showed that a large part of pasture, forest and agricultural lands are in conditions of medium and high flood risk and compared to barren lands (most of these lands are at low flood risk) they need proper management and planning. Also, a large part of the residential lands is at high risk (southeastern parts) and moderate flood (river flow margins in the northwestern and southern parts). This situation has resulted in a large part of the population of these areas being exposed to flood risks, and the intensity of these risks is much higher in the southeastern parts of the Haraz basin than in other parts. Therefore, the southeastern parts of Haraz basin need special attention in crisis management planning.

Conclusion
The results of this study showed that a large part of the Haraz basin is in medium risk conditions, the southeastern parts are in high flood risk, the central parts are in low flood risk, and the areas without flood risk are very low, which was shown in all three scenarios models. Also, the comparison of GeoTOPSIS, WMCA and LWLC models with WLC and OWA models showed that the results of these models were very close to each other in three scenarios. Therefore, GeoTOPSIS, WMCA, and LWLC models, like WLC and OWA models, are efficient methods for flood risk studies. In general, following the obtained results, Haraz basin needs necessary planning to manage the risk of possible floods and it should be prioritized in the southeastern parts of the basin. Also, in order to develop the methodology of flood risk studies, it is suggested to use Geo-TOPSIS, WMCA and LWLC models for similar studies.