Predicting Fault Locations based on Morphometric Features of Alluvial Fans and Basins using Articial Neural Networks

The aim of this study is to investigate the morphometry of alluvial fans located in the vicinity of the 18 Sabzevar and Sang-Sefid faults in northeastern Iran to determine their influence on erosion Principal 19 component analysis (PCA) was used to select the most important morphometric factors affecting erosion. 20 The data regarding the important parameters were input into adaptive neural-fuzzy networks (ANFIS) to 21 predict erosion rates. The asymmetric factor ( Af ) , hypsometric integral (Hi), and basin shape (BS) indicate 22 that most of the sub-basins are tectonically active. The results of the PCA revealed that the most important 23 parameters affecting erosion were A f , P f , L f , R f , V f , P b , A b , L C , L b , Dd, and the geological unit. The ANFIS 24 method showed that among the soil erosion prediction models, the FCM hybrid model had the highest 25 accuracy. It is concluded that morphometric features can be used to predict the erosion processes in the 26 basin. The results show that A f , P f , L f , R f , V f , P b , A b , L C , L b , Dd, and formation material had the greatest influence on erosion rates in the study area. PCA was used to identify the most important parameters influencing erosion. Using ANFIS, the soil erosion in the study area was predicted using these parameters. The effects of two large Sabzevar faults and the Sang-e-Sefid fault on the morphometric characteristics of alluvial fans and their watersheds were also investigated. The results showed that tectonic activity was the main factor in the formation, development, and evolution of alluvial fans in the study area. The Sabzevar and Sang-e-Sefid faults have been more influential on morphometry than other tectonic factors. The results 351 show that faults are active in the study area and affect the morphometry of the watershed. One of the most 352 important outcomes of this study is the confirmation of the ability to identify and predict the tectonic 353 activities of the watershed quantitatively


tief@txstate.edu 11
Introduction 32 As a river flows out of a mountainous region and enters a plain with a low slope, the capacity for carrying  The sediment of alluvial fans includes sand, gravel, silt, and clay, which increase the particle size from

89
The ANN is a predictive model that has been used in many geomorphological studies but has not yet 90 been applied to the study and prediction of erosion based on alluvial fan morphometry. This study aims to 91 predict erosion from alluvial fan morphometry using the adaptive network-based fuzzy inference system 92 (ANFIS) method. The alluvial fans in the vicinity of the Sabzevar and Sang-Sefid faults in northeastern 93 Iran are the objects of study. This is among the few articles that have employed the ANN method to 94 investigate and predict alluvial morphometries and their relationships to erosion in upstream watersheds 95 (Lucà and Robustelli 2020). The PCA method was used to determine the most important morphometric 96 parameters affecting the fault activity and soil erosion. This study is also innovative in that it strives to 97 predict fault activity in the region based on the watershed's morphometric characteristics.

98
The remainder of this paper is organized as follows. section 2 explains the case study. In Section 3, the 99 method of extracting alluvial fans is described. In addition, the formulation of the proposed method to select

106
The study area is located in the Central Desert watershed, located at 35°2'2" to 35°33'00" N and 107 57°38'24" to 59°06'24" E (  morphometries of an alluvial fan can be a semi-conical surface. In the GIS algorithm, a conical surface is 143 created by joining a series of profiles radiating from the fan apex. The channels were mapped, the radial 144 slopes were mapped, and the semi-conical surface was interpolated ( Fig. 2(b)).

145
Radial profile analysis is mainly based on a fixed or variable minimum slope threshold that examines 146 slope changes along each fan (slope threshold is defined by trial and error or training on a representative 147 alluvial fan). The semi-conical surface of the alluvial fan was used to cut the radial profile. The apex is the 148 location of the input of sediment input to the alluvial fan ( Fig. 2(b)). In the next step, the topographic surface 149 was placed on the radial profile to determine the shape of the alluvial fan. Profiles for all of the alluvial fans 150 in the watershed were prepared from the DEM using a stepped process (Fig 3).

151
After extracting the alluvial fans, the morphometric parameters of both the fans and watersheds were 152 determined (Table S1).

169
To determine the extent of geological activity, hypsometry was analyzed with Hi. The altimeter curve 170 is the ratio of the total height of the basin to the total area of the basin (Strahler 1952

212
ANFIS was used in this study to predict the erosion rates. The ANFIS was introduced by Jang (1993).

213
This method is based on the first-order Sugeno-fuzzy method. Because the fuzzy system is a very efficient 214 modeling method, it has been widely used. Empirical knowledge is transformed into a mathematical map 215 using linguistic rules. In systems where the knowledge of the expert is either unavailable or inaccurate, the 216 neural network method can be used to create membership functions and rules for the system. For example, 217 the two laws are defined by Eqs. Six and 7 (Bisht and Jangid 2011).
245 where x and y are the crisp inputs, and Ai and Bi are the language membership functions. Pi, qi, and ri are 246 the sugar output parameters. The ANFIS also has a structure ( Fig. S3 (b)) (Bisht and Jangid, 2011). it is 247 operated in steps (Fig. S4). The alluvial fans of the study area were extracted using a semi-automatic method and a DEM (Fig. 4).

261
There were 54 alluvial fans in the study basin, most of which were affected by the Sabzevar faults and the 262 Sang-Sefid fault. The morphometric properties of the 54 alluvial fans were determined using GIS (Table   263 1).

264
The mean values for each of the morphometric features of the alluvial fan and its upstream watershed 265 were determined (Table 1)

296
Grid partitioning, subtractive, and FCM models were used to predict soil erosion using the ANFIS.

297
Hybrid and backpropagation modes were used for each model (run in MATLAB). The results showed that 298 modeling soil erosion in the study area using the subtractive method had the lowest error ( Fig. S6 and Table   299 4). Two radii of 0.01 and 0.03, were used. The hybrid method with radii of 0.01 and 0.03 had R 2 = 0.99, 300 MSE=0, and RMSE= 0.03, and high accuracy. This method requires four rules (Fig. S7).

301
The relationships between the parameters Af, Ab, Pf, Rf, Lf, and soil erosion in the three dimensions are 302 shown (Fig. S8)

304
In general, several features are found in a fuzzy neural network, such as learning power, as well 305 as costing, classifying, writing, and compiling. Another advantage is that it allows the extraction 306 of fuzzy rules from a variety of information and calculates the basic rules proportionally. Fuzzy 307 neural networks have been proven to have the ability to model multiple processes in recent studies 308 to predict the erosion rate (Nguyen et al. 2020). The artificial neural network (ANN) model 309 performs better when there is sufficient information and data. Observational data are used to train 310 the network, so the system's performance is reduced when there is a lack of data. in the fuzzy 311 inference system, the input and output variables in this model are described linguistically. Because 312 there is no formal method for doing this, the fuzzy system uses innovative approaches when the 313 information is incomplete and contradictory. This is usually time-consuming and error-prone. 314 Nauck and Kruse (1999)

328
Hi was calculated for the sub-basins using the GIS software. Watersheds 21-34 and 43-54 have the 329 highest Hi (> 0.5), indicating that they are areas of high tectonic activity. Watersheds 1-20 and 35-41 had 330 the lowest Hi (< 0.4), indicating less tectonic activity in these sub-basins (Fig. 7).

331
The BS values indicate that most watersheds have a coefficient higher than 1, indicating elongated 332 basins and high tectonic activity in these areas (Fig. 8).

333
The tectonic state of a region can be determined using morphometric features. Bahrami