Assessment of Alluvial Aquifer Intrinsic Vulnerability by a Generic Drastic Model; A Discussion on Data Adequacy and Pragmatic Results

DRASTIC is a model that is commonly used to assess vulnerability to groundwater 16 contamination at the landscape scale. When sparse data are available to populate the layers of 17 the model, it can be difficult to ascertain the true usefulness of the model produced map. In 18 this research an alluvial aquifer, the Sahneh aquifer in Kermanshah province of western Iran, 19 was mapped using the generic DRASTIC model. The data available for populating the model 20 layers were generally sparse. The model was validated using a nitrate concentration map 21 constructed from well water measurements within the DRASTIC map area. A Receiver 22 Operating Curve (ROC) analysis was conducted by placing 500 random points in the 23 DRASTIC generated map compared to the nitrate concentration map. The area under the curve 24 was compared and yielded a value of 0.72 or 72% concordance, which means it has good 25 validity. This investigation demonstrates that a generic DRASTIC model can yield acceptable 26 results without modification or increasing its complexity. If the ROC analysis had yielded a 27 value <50%, then the DRASTIC would have been considered to not be useful. A common mistake in the use of DRASTIC is to modify the method to greatly increase its complexity, which may actually decrease, not increase the resultant model usefulness. This comparison was used to evaluate the real variability in the DRASTIC model outputs to illustrate the potential errors with the assessment and how the output should be used within the context of groundwater management. A test of the model validity was used to ascertain its usefulness by using nitrate data with the context of a Receiver Operating Curve analysis (ROC). This research provides a method to review the generic DRASTIC method and validation of its outputs to create a practical guide to water managers on the usefulness of the model and what restrictions should be placed on the output use.


35
The DRASTIC model has been evaluated to assess both its advantages and disadvantages for assess intrinsic vulnerability is interesting and thought-provoking, because the word itself has 46 a strong meaning and indicates a situation that should be considered (Dictionary, C 2008). 47 Despite the apparent simplicity of the seven factors in the DRASTIC model, development 48 and presentation of thematic and practical maps with accurate data (or at least sufficient) is 49 not so easy. While DRASTIC is a simple and restrained method that gives results in any case, 50 the adequacy of the data and the realism of the results can make the output of this model 51 sufficiently reliable for use by local planners and decision makers. However, the use of the 52 model results must be constrained within the famous statement that "All models are wrong, 53 but some are useful; the practical question is how wrong they have to be to not be useful." 54 (Box and Draper 1987). The statement by Box and Draper (1987) applies not only to the 55 accuracy of the model, which is commonly controlled by the input parameters (data), but also 56 applies to how the model is applied with consideration of the error framework. 57 The simplicity of the DRASTIC model is the evaluation methodology and use of specific 58 weights and rates as suggested by Aller et al. (1987). Advancements in software development 59 now make it is possible to easily overlap the seven layers within a GIS framework and obtain 60 a digital output that can be categorized in different ways. In this regard, there are numerous  Box and Draper (1987) "All models are wrong, but some are 89 useful; the practical question is how wrong do they have to be to not be useful." 90 The answer to all of these questions lies in the ability of the researcher(s) to extract or 91 provide adequate data and to perform sufficient analyses on the model output results to provide 92 a reasonable degree of assurance that the model is useful. To produce an accurate vulnerability 93 assessment, the researcher must be familiar with the basic principles of hydrogeology. It is 94 very important to understand the nature of various units constituting and affecting the aquifer 95 from the unsaturated zone to the base of the aquifer. In addition, there are fundamental 96 relationships between some of the seven DRASTIC parameters that need to be considered, 97 such as the soil media and the aquifer media may be the same when the water table occurs at 98 or near land surface. There is also a close relationship between impacts occurring within the 99 vadose zone and the soil media. For example, it is possible to provide a single layer based a 100 few water drilling logs for the impact of the aquifer media and the vadose zone. If the nature 101 of the upper part of aquifer system is over-simplified by ignoring the occurrence of a clay or 102 gravel unit, the results of the DRASTIC assessment could be inaccurate. This issue is quite 103 extreme in karst environments, where flow conduits could provide a bypass through which 104 contaminants could directly enter deep into an aquifer (Taheri, et al. 2017).

105
Application of the DRASTIC model could be used in basic educational for students of 106 hydrogeology, groundwater and related disciplines (Rich and Onasch 1997). Because

131
Study area 132 The study was conducted on the Sahneh aquifer that is located mainly in the Sahneh Sub-133 catchment in the Kermanshah province of western Iran (Fig. 1a). The Sahneh aquifer covers 134 a relatively small area that is located in a cold and dry region (Fig. 1b,c). This aquifer is located 135 in the structural zone of Sanandaj-Sirjan and thrust fault zone of Zagros. The behavior of the 136 aquifer geological units is a function of the structural conditions of these two zones (Fig.1e).

137
The geological units of the region can be categorized into karst and non-karst units. Karstic   Hydrogeologic assessment of the Sahneh aquifer was performed using data extracted from 156 the seven monitoring wells. Although this number is insufficient, however, no action has been 157 taken to correct it. Few exploratory studies of the aquifer have been conducted. The primary 158 source of knowledge on the aquifer and its subsurface media comes from the logs of private 159 wells used for water supply. Water levels compiled into hydrographs indicate that the 160 groundwater balance is negative, meaning that the discharge rate is higher than the aquifer 161 recharge rate (declining water levels) (Fig. 1f). According to the results of the water balance 162 analysis, the recharge rates in the alluvium and higher altitudes are 3.36 million cubic   (2) 179 where Wj is the weighting of each parameter and Rj is the corresponding rate. Depth to Water Table (D) 198 It is defined as the vertical distance from land surface to the water table (Fig. 2a). In porous 199 media, a groundwater depth map can be constructed by using the groundwater levels measured Aquifer media is one of the most important parts of the study by the DRASTIC method.

224
Preparing a map with good accuracy can give the final DRASTIC greater credibility and 225 usefulness. The best way to map the aquifer media is to use the results of geophysical surveys 226 and match it with the results of exploratory geologic well logs. In the absence of these two 227 data sources, information was used from public water well drilling logs with a high degree of 228 distribution in the aquifer area.

229
To prepare the map of the aquifer media layer, the type of sediment above and below the 230 water table was determined using statistical analysis and sedimentological diagrams. By 231 assigning numerical codes to each of these types of sediments, it is possible to interpolats and 232 prepare a sediment distribution map. In this study, the sediment diagram of Shepard (1954) 233 was used to classify sediments and prepare a map of the aquifer and vadose zone (Fig. 2f, 3b).

234
Because it is practically difficult to separate clay and silt in the desert sediments, instead of 235 separating these two factors, clay and silt were aggregated into a single class and gravel was 236 added as the third factor. This change in the naming triangle is more applicable to aquifers 237 and unsaturated sediments, although it may not be accurate in terms of sedimentologic 238 laboratory studies.

239
Soil media (S) 240 The soil media occurs at the top of the system and controls the infiltration of water into the 241 unsaturated zone and ultimately the recharge of the aquifer to some degree (Fig. 3a).

242
Permeable soils facilitate higher rates of water infiltration and fine-grained soils produce more 243 runoff. In many places, soil maps are available that were produced by organizations related to 244 agriculture and soil science. In this study, the soil map of Iran was used within the context the into a slope layer (Fig. 1i, 2b, 3a). and Cherry, 1979) (Fig. 1g, h). However, using geophysical test data and the portability and 282 thickness of the aquifer, a hydraulic conductivity layer can be obtained. In this study, average 283 hydraulic conductivity was obtained by dividing the transmissivity layer by the aquifer 284 thickness (Fig. 1g, 1h). The hydraulic conductivity of the aquifer was computed using

289
Groundwater experts in Iranian regional water companies and experienced drilling supervisors 290 have provided useful information on hydraulic conductivity.

309
Various organizations in Iran charged with the collection of groundwater quality data may 310 also have available nitrate data, but the availability of these data is limited, because they are 311 considered to be confidential. However, the nitrate concentrations have been measured in 312 water pumped from public water wells in the area. Thereby, it was used to evaluate the model

364
Based on the data available in Kermanshah Regional Water Authority (KRWA) and its 365 affiliated departments as well as general soil maps of Agriculture Organization and 366 Climatology office of the Meteorological Organization, seven maps required for the 367 DRASTIC model were prepared (Figures 3a,b,c,d and 4). These maps provide the most  (Fig. 5b).

380
The results show that only 8.6% and 6.7% of the study area fall under very low and low 381 vulnerability class, respectively. The moderate and high vulnerability classes cover 26%, and 382 17.2% of the total study area. Most of the area, consisting of 42%, is classified as having high 383 vulnerability (Fig. 5a,b). The southern part of the study area falls under the very high In this study, using the analysis of drinking water wells, the nitrate concentration in water 391 produced from these wells was mapped in the GIS environment (Fig. 5b). There are several

399
The ROC method is commonly used to assess validity of modeling results in water 400 resources management (Khosravi et al. 2018). ROC is considered to be an evaluation of a 401 binary classification system whose detection threshold is also variable. In this method, the of the created DRASTIC model. To do this, 500 random points were placed within the map 405 grid using GIS (Fig. 5c). The data corresponding to these points for both the nitrate 406 concentration map and the final DRASTIC vulnerability map were assesed. For more nitrate 407 concentrations greater than the standard limit of 50 mg/L a value of 1 was assigned and a 408 value 0 of was assigned to the points below the nitrate limit. Using SPSS software, the 409 characteristic curve of the graph was created. In this method, a curve showing above 50% 410 indicates the validity of the model, and the higher the value, the more valid the model. 411 Yesilnacar and Topal (2005) classified AUC values with respect to prediction accuracy into 5 412 quantitative ranges, which are 0.5-0.6 for low/poor, 0.6-0.7 for moderate/average, 0.7-0.8 for 413 good, 0.8-0.9 for very good, and 0.9-1 for excellent. Based on this approach in this study 414 AUC value obtained was 0.72 or 72%, which is considered to be good validity (Fig. 5d).   Research .

559
All authors contributed to the study conception and design. Material preparation, data 560 collection and analysis were performed by Kamal Taheri, Amjad Maleki and Reza Omidipour.

561
Amjad Maleki and Jamil Bahrami supported field visits superintendence for data collection.  The location of studied area in Iran (a), Kermanshah province (b), and Sahneh Sub-catchment (c), isohytal map (d), simpli ed geological map and distribution of carbonate, non-carbonate formations and porous media (plain) and alluvial aquifer boundary (e), water budget summary(f), sediment thickness map (g), transmissivity map (h) and digital elevation model (i). Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.

Figure 3
Seven layers of DRASTIC model; a) soil media, and net recharge, b) vadose zone media, and aquifer depth, c) aquifer media, and hydraulic conductivity, and d) slope map. Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.

Figure 4
Flowchart of the nal output model of the general DRASTIC model in the study area. Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors. Nitrate analysis results (a) and its adaptation to general DRASTIC model (b) and DRASTIC with expert weights (b) linear regression (c) ROC curve with 500 random points (d) and Idrisi output (e). Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.

Figure 6
The fuzzy-DRASTIC vulnerability of the Sahneh aquifer (a) owchart of GIS tools application and build fuzzy DRASTIC map (b) fuzzy Drastic compared to the generic DRASTIC map (c), normalized DRASTIC weights for fuzzy method (d). Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.