**Scaling-Up Conservation Agriculture Production System with Drip Irrigation by Integrating MCE Technique and the APEX Model**

#### **Tewodros Assefa 1,\*, Manoj Jha 2, Abeyou W. Worqlul 3, Manuel Reyes <sup>4</sup> and Seifu Tilahun <sup>1</sup>**


Received: 23 August 2019; Accepted: 18 September 2019; Published: 27 September 2019

**Abstract:** The conservation agriculture production system (CAPS) approach with drip irrigation has proven to have the potential to improve water management and food production in Ethiopia. A method of scaling-up crop yield under CAPS with drip irrigation is developed by integrating a biophysical model: APEX (agricultural policy environmental eXtender), and a Geographic Information System (GIS)-based multi-criteria evaluation (MCE) technique. Topography, land use, proximity to road networks, and population density were considered in identifying potentially irrigable land. Weather and soil texture data were used to delineate unique climate zones with similar soil properties for crop yield simulation using well-calibrated crop model parameters. Crops water demand for the cropping periods was used to determine groundwater potential for irrigation. The calibrated APEX crop model was then used to predict crop yield across the different climatic and soil zones. The MCE technique identified about 18.7 Mha of land (16.7% of the total landmass) as irrigable land in Ethiopia. Oromia has the highest irrigable land in the nation (35.4% of the irrigable land) when compared to other regional states. Groundwater could supply a significant amount of the irrigable land for dry season production under CAPS with drip irrigation for the various vegetables tested at the experimental sites with about 2.3 Mha, 3.5 Mha, 1.6 Mha, and 1.4 Mha of the irrigable land available to produce garlic, onion, cabbage, and tomato, respectively. When comparing regional states, Oromia had the highest groundwater potential (40.9% of total potential) followed by Amhara (20%) and Southern Nations, Nationalities, and Peoples (16%). CAPS with drip irrigation significantly increased groundwater potential for irrigation when compared to CTPS (conventional tillage production system) with traditional irrigation practice (i.e., 0.6 Mha under CTPS versus 2.2 Mha under CAPS on average). Similarly, CAPS with drip irrigation depicted significant improvement in crop productivity when compared to CTPS. APEX simulation of the average fresh vegetable yield on the irrigable land under CAPS with drip irrigation ranged from 1.8–2.8 t/ha, 1.4–2.2 t/ha, 5.5–15.7 t/ha, and 8.3–12.9 t/ha for garlic, onion, tomato, and cabbage, respectively. CAPS with drip irrigation technology could improve groundwater potential for irrigation up to five folds and intensify crop productivity by up to three to four folds across the nation.

**Keywords:** scaling-up conservation agriculture; drip irrigation; groundwater potential; sustainable intensification; Ethiopia

#### **1. Introduction**

Crop production in Ethiopia is constrained with several challenges that cause low productivity and economic growth in the region. Soil degradation in the form of soil erosion and decline of soil fertility is the major constraint for crop production [1]. The alarming rise in population caused the exploitation of the rainforest and grasslands in the region to increase cultivated lands, which resulted in soil degradation and deterioration of the environment [2–4]. Crop production in the nation is mainly a rainfed system using traditional farming practices [2]. The expansion of cultivated land at the expense of forest, bushes, and grassland is not a feasible option to sustain crop production let alone increased productivity. Instead, with the current poor soil and water management practices, it contributes to lower production efficiency [3].

On the other hand, rainfall variability is a great concern for a rainfed agricultural system in Ethiopia [5]. Customized local strategies are needed to maximize food supply and enhance the ecosystem at the same time. One such strategy is to enable dry season production to address the adverse effects of rainfall variability. The conservation agriculture production system (CAPS), which promotes no-till, mulching, and diverse cropping, has been shown to provide higher water use efficiency in addition to improving soil fertility [6–9]. Similarly, adoption efficient water application technologies can increase water use efficiency [10]. In relation to irrigation technology, drip irrigation is considered the most efficient water application technology [11,12]. CAPS combined with drip irrigation constitutes efficient soil and water management technology, which helps to maximize the potential of water resources and consequently increase productivity in the region [8]. Another concern is the lack of knowledge of potential to expand irrigated agriculture and maximize production. Worqlul et al., [3] indicated that less than 5% of the potentially irrigable lands are currently under irrigation.

While the positive impacts of CAPS with drip irrigation have been identified, expanding the impact to a large-scale adaptation on a country level and linking it with water resources availability would provide substantial and very useful information to policymakers in a decision-making process to improve the agriculture systems in the nation. Assessments of potentially irrigable land, corresponding crop productivity, and availability of water resources are essential components for the scale-up of CAPS with drip irrigation technology. There are few quantitative studies [3], that provided country-level irrigation potential assessment under the conventional tillage production system (CTPS) with traditional irrigation practice. However, no literature was found for a large-scale adaptation of CAPS with drip irrigation. This study attempts to examine the country-level adaptation of CAPS with drip irrigation technology for its impact on groundwater potential and crop productivity based on experimental results presented in Assefa et al. [8]. The specific objectives were to (1) assess potentially irrigable land using the multi-criteria evaluation (MCE) technique, (2) scale-up crop yields by integrating the MCE technique and a biophysical model field-scale prediction, and (3) asses groundwater irrigation potential for dry season production. The analyses were made for garlic, onion, tomato, and cabbage which are commonly grown vegetables in Ethiopia [13].

#### **2. Materials and Methods**

#### *2.1. Study Area*

This study was conducted in Ethiopia, the second-largest populated country in the entire continent of Africa, next to Nigeria (Figure 1). The landmass of the country is approximately 110 million ha, and elevation ranges from 160 m to 4530 m above mean sea level [3]. Climate variability (as it pertains to variability in rainfall and temperature) was observed to be very high in Ethiopia (i.e., 15% to 50% coefficient of variation for rainfall and 1.6◦C annual average rise) based on the long-term (1955–2015) evaluation of climate data. This poses major risks to rainfed crop production [14,15] which is the dominant agriculture practice in Ethiopia [16]. The southwestern portion of the country receives about 2400 mm of rainfall, whereas northeastern and southeastern lowland receives less than 500 mm per

year [17]. There are three seasons in the year locally known as Kiremt (main rainfall season), Belg (small rainfall season), and Bega (dry season) [18].

**Figure 1.** Ethiopia and its administrative regions with water bodies.

#### *2.2. Scale-Up of Field-Scale Parameters*

The MCE technique and agricultural policy environmental eXtender (APEX) model was used to scale-up the field-scale CAPS with drip irrigation (Figure 2). MCE was used to identify potentially irrigable lands in the country based on various factors that affect irrigated agriculture. MCE is an emerging approach that involves combining multiple variables to produce a single evaluation index for an intended purpose [2,19–26]. The MCE technique has been used for various applications including crop agriculture, water resource management, and other environmental studies [2,3,26–33].

A range of variables (Figure 2) were considered in this study to identify potentially irrigable land in the country. These factors include topography (slope), land use/cover, proximity to road networks, and population density. Topography affects the choice of irrigable land as it affects irrigation practices [3]. Digital elevation model (DEM) data with 30 m resolution, was used to derive the landscape slope for the entire nation. Land use/cover data also provide a vital figure in the selection of economically productive lands for irrigated agriculture. Similarly, population density and proximity to road networks were used to account market accessibility to support irrigated agriculture. Euclidean distance was calculated to establish the proximity criteria to road networks. Factors were reclassified into various suitability classes depending on the Food and Agricultural Organization [34] guidelines: Highly suitable (S1), moderately suitable (S2), marginally suitable (S3), currently unsuitable (S4), and permanently unsuitable or constraint (N1). The equal interval ranging technique was used to reclassify population density and proximity to road networks based on Worqlul et al., [3]. The pairwise method was used to compare each factor one-to-one and weights were scaled using works of Saaty [35] and Worqlul et al., [3]. The pairwise method is a relatively unbiased ranking technique [2,27] and applied to weigh each factor considered in this analysis. The pairwise technique, Saaty [35], makes use of a scale broken down from 1 to 9 indicating the equal and absolute importance of a factor when compared

to a one-to-one basis, respectively. Consistency ratio was used to check the consistency of the pairwise matrix using Equation (1) described in Saaty [36]. The equal interval ranging technique was used to distribute the weight of each factor into the suitability classes. Factors were then combined using "weighted sum overlay" to produce a single evaluation index (0% to 100%) map, and constraints (permanently unsuitable lands) were excluded from the analysis. Combined weights of greater than 85% were considered to be potentially irrigable lands [2,3].

$$\text{CR} = \text{Cl/RI},\tag{1}$$

where CR = consistency ratio, CI = consistency index, and RI = consistency index of randomly generated matrix.

**Figure 2.** Method of scaling-up crop production under conservation agriculture production system (CAPS) with drip irrigation technology.

Soil and climate data were used to define unique areas for biophysical model development. Spatial variability of parameters and the effect of scaling needs to be carefully considered while upscaling modeling results [37–39]. Table 1 shows the type and source of data used to scale-up field-scale crop production to the national level. The mean annual rainfall point data from observed ground weather stations were used to compute spatial annual rainfall and classify the nation into different rainfall regimes. Similarly, soil texture data was used to classify the nation into various soil classes. Climate regimes and soil texture data were combined using the intersection of the ArcGIS overlay function to identify geographical equivalence zones of similar climate and soil (i.e., 39 zones) where each zone has the same soil texture and rainfall regimes. The APEX [40], a biophysical model, was set up on the unique climate and soil combinations to simulate crop yield. Weather data (rainfall, maximum and minimum temperature, wind speed, relative humidity, and solar radiation), soil characteristics, as well as vegetation and management practices are the main inputs to the APEX model [41]. The model is capable of evaluating the effects of soil and water conservation practices on hydrology, crop yield, and other environmental variables such as sediment, nutrient load, and soil organic carbon [42–44]. Proper calibration and validation of model parameters are essential steps for reliable predictions [45]. The APEX model was calibrated and validated for a few sites in Ethiopia using adequate field data from 13 experimental plots. The calibration results (i.e., model performances) are presented in Table 5 and Table 9 of Assefa et al., [9] for hydrology and crop yield, respectively. Based on efficiency measures suggested in Moriasi et al., [46] and Wang et al. [47], the model performance was found within the range of acceptable to very good. Satisfactory model performance during calibration and validation provides greater confidence in the modeling results when evaluating various plausible scenarios for modeling prediction. The present study was built on the same model for its application in up-scaling the impact. Input data were changed based on the unique climate-soil combinations across Ethiopia, but the same model parameters were used which was established during the calibration. Heat unit scheduling (OPV7 = 1) was used with shortening cropping period compared to the experimental plots to capture crop growth variability across the unique regions.


**Table 1.** Data and sources for upscaling crop production to the county level.

Irrigation requirement of crops is mainly a function of reference evapotranspiration and rainfall, which are variable in space and time. Therefore, variable irrigation water volumes were applied in the model for each climate zone depending on the type of vegetables grown and weather conditions. The net irrigation requirement (NIR) for each of the vegetables was calculated depending on reference evapotranspiration (ETo), crop coefficients of each vegetable at mid-stage (Kc), irrigation application inefficiency, and effective rainfall amount (ER). Crop coefficients at the mid-stage of crop growth (Kc-mid) were obtained from Allen et al., [48] of the Food and Agricultural Organization (FAO) for various vegetables. The net irrigation requirement equation derived by Worqlul et al., [3] for the country using conventional irrigation inefficacy in Equation (2) was modified in this study (Equation (3)) to account for drip irrigation inefficiency. Howell [49] indicated that 95% efficiency can be attainable whereas 90% is the average efficiency for drip irrigation. Thus, 10% application inefficiency was considered for irrigation and some minor losses such as leaching [50]. Assefa et al. [8] showed that significant (*p* ≤ 0.05) reduction of irrigation volume was observed in the conservation of agriculture (CA) experimental sites when compared to conventional tillage (CT) practice. Therefore, Equation (3) was further modified using linear coefficient (Cf) to account for the reduction of irrigation volume in CA practice by comparing the irrigation data from the experimental sites for CA and CT managements (Equation (4)) (net irrigation requirement for conservation practice, NIRc). The contribution of rainfall to soil moisture, effective rainfall (ER), in the growing season of vegetables was estimated using the United States Department of Agriculture Soil Conservation Service (USDA-SCS) method [51], which is a function of precipitation (P), see Equation (5a) and Equation (5b).

$$\text{NIR} = 1.6 \times \text{Kc} \times \text{ETo} - \text{ER} \tag{2}$$

$$\text{NIR} = 1.1 \times \text{Kc} \times \text{ETo} - \text{ER} \tag{3}$$

$$\text{NIRRc} = 1.1 \times \text{Cf} \times \text{Kc} \times \text{ETo} - \text{ER} \tag{4}$$

$$\text{ER} = \text{P} \times \frac{125 - 0.2 \times \text{P}}{125}; \text{ For P} \le 250 \text{ mm/m} \tag{5a}$$

$$\text{ER} = 125 + 0.1 \times \text{P}; \text{ For P} > 250 \text{ mm/m} \tag{5b}$$

where NIR, NIRc, Cf, Kc, ETo, P, and ER are the net irrigation requirement for the tilled system, net irrigation requirement for conservation agriculture, the coefficient of conservation agriculture, crop coefficient, reference evapotranspiration, precipitation, and effective rainfall, respectively.

Soil properties, weather data, net irrigation requirements, and cropping details were supplied to the well-calibrated APEX model in each unique zone. Then, crop yield simulation was integrated with the irrigable land to limit crop yield estimation only on the potentially irrigable land. Groundwater source of irrigation with a depth less than 30 m from the surface was considered in this study. The potential borehole yields and the potential numbers of wells that could be installed were used to estimate groundwater availability in the regions. Maintaining a one-kilometer clear distance between wells (i.e., the radius of influence) is suggested by Howsam and Carter [52] to estimate the potential numbers of wells that could be installed. Maintaining the radius of influence helps to avoid the groundwater drawdown effect of one well on another. The net irrigation requirement for CA practice, groundwater availability, and depth to groundwater were considered to determine the potential of groundwater wells in unique zones. Vegetable yields on the irrigable land were further constrained based on groundwater availability to identify the potential scale-up areas for CAPS with drip irrigation technology.

#### **3. Results and Discussion**

The results of scaling-up crop yield under CAPS with drip irrigation technology to country-level were presented into three categories: (1) Assessment of potentially irrigable land in the country using the MCE technique, (2) simulation of potential crop production under CAPS with drip irrigation using a well-calibrated APEX model, and (3) assessment of groundwater potential for dry season crop production.

#### *3.1. Potentially Irrigable Land*

Four basic factors (topography, land use, proximity to road networks, and population density) were considered in the MCE technique to identify potentially irrigable land in the nation. Topography in the nation ranges from 0% (flat land) to greater than 100% (steepest land which is about 0.07% of the landmass) (Figure 3a). The slope was reclassified into five categories based on Worqlul et al., [27]: Highly suitable (0%–2%), moderately suitable (2%–8%), marginally suitable (8%–12%), less suitable (12%–30%), and unsuitable (above 30%). The various land use classes in the nation (Figure 3b) were reclassified into four suitability classes based on Assefa et al. [2], Worqlul et al., [3], and FAO [53]: Highly suitable (agricultural land), moderately suitable (grassland), marginally suitable class (shrubs, bare land), and unsuitable class (forest, urban lands, wetlands, and water). Population density ranges from 0 to 69,350 persons per square kilometer (Figure 3c), whereas proximity to road network ranges from 0 to 118 km (Figure 3d).

**Figure 3.** Irrigation suitability factors map: (**a**) Topography in terms of landscape slope, (**b**) land use, (**c**) population density (PD), and (**d**) distance to road networks.

The Eigenvector was computed as the nth root of individual factors' weight and then normalized with the cumulative Eigenvector to derive the final weights of factors (Table 2). Topography was found to be relatively the most influential factor in irrigated agriculture, which was consistent with Worqlul et al.,'s [3] result. Proximity to road networks and land use were found to be the second and third most influential factors in determining potentially irrigable land in the nation. The consistency ratio was found to be trustworthy (CR = 0.03 ≤ 0.2) based on Chen et al., [54] and Koczkodaj et al., [55]. The final weights of factors were distributed to the various suitability classes and factors were combined using a weighted sum overlay. An 85% threshold was used to obtain potentially irrigable land (Figure 4). About 18.7 Mha of land, 16.7% of the total landmass, was found to be potentially irrigable in the nation without considering soil and weather. The suitability ranges in Figure 4 cover different portions of the irrigable land: 85%–88% (76% of the irrigable land), 88%–91% (11% of the irrigable land), 91%–94% (1.5 of the irrigable land), 94%–97% (11.5% of the irrigable land), and 97%–100% (0.4% of the irrigable land).



**Figure 4.** Potentially irrigable land (all green, weight ≥ 85); S4 (currently unsuitable areas)- weight < 85%; N1—constraint (permanently unsuitable areas); abbreviations in the map are administrative regions (TG—Tigray, AM—Amhara, AF—Afar, BG—Benshangul Gumaz, AD—Addis Ababa, DD—Dire Dawa, GP—Gambela Peoples, SNNP—Southern Nations, Nationalities and Peoples, and SM—Somali).

Irrigation demand of each vegetable was computed by considering the conservation agriculture principles, drip irrigation technology, water use of different vegetables, and weather conditions. Oromia regional state has the highest irrigable land (35.4%) when compared with other states. Figure 5 illustrates the degree of irrigation suitability (i.e., marginal, satisfactory, medium, high, and very high) for potentially irrigable lands: 18.7 Mha, 4.5 Mha, 2.5 Mha, 2.2 Mha, and 0.082 Mha at 85%, 88%, 91%, 94%, and 97% suitability classes, respectively.

**Figure 5.** Degree of irrigation suitability for the potentially irrigable land.

#### *3.2. Potential Crop Production under Conservation Agriculture*

Figure 6a shows the various soil texture classes in the nation. The mean annual rainfall was computed spatially using inverse distance weighting interpolation from weather stations point data (Figure 6b), and the spatial rainfall was reclassified using natural breaks into eight rainfall zones (Figure 6c). Soil textures and rainfall zones were combined, which resulted in 39 unique regions for further analyses of crop yields. The APEX model was developed for each of 39 unique zones, which was defined using the soil texture classes and climate zones. Results were then aggregated as per administrative boundaries of Ethiopia (11 regional states) to provide input for decisionmakers in developing policy and implementation strategies for water resource and agriculture-related projects.

**Figure 6.** Soil textures (**a**), locations of rainfall gauging stations (**b**), and rainfall regions regimes (**c**).

Crop coefficients at the mid-stage of crop growth (Kc-mid) were obtained from Allen et al., [48] of the Food and Agricultural Organization (FAO) for various vegetables. These data indicate that more irrigation is needed for tomato during the mid-stage of growth followed by cabbage when compared to garlic and onion. Moderate resolution image spectrum (MODIS) potential evapotranspiration data (2000–2010) was used to estimate the net irrigation demand in the region during the dry season. The growing period used for garlic, onion, and cabbage was December through February whereas December through March was the growing season for tomato.

This study used these water use data for the various vegetables under CA and CT practices from Assefa et al., [8] and developed a linear irrigation coefficient, Cf, for each vegetable to account for irrigation volume reduction under CA during the calculation for the irrigation requirement. The value of Cf obtained from CA and CT comparison was 0.58, 0.54, 0.80, and 0.81 for garlic, onion, tomato, and cabbage, respectively. These coefficients explain the advantage of conservation practices over conventional tillage systems for irrigation water savings mainly due to mulch cover and no-till practice in CAPS plots minimized water loss through evaporation and runoff. Additionally, the water-saving in garlic and onions was higher when compared with cabbage and tomato. This could be due to the

less leaf area of garlic and onion, which made the impact of CAPS significant in reducing water loss when compared with tomato and cabbage.

Net irrigation demand was computed for each vegetable over the irrigable land considering drip irrigation efficiency, crop coefficient, effective rainfall during the growing period, and irrigation coefficient. These data along with other inputs such as soils, weather data, cropping details, irrigation application rate, and crop water demand were supplied to the calibrated APEX model to estimate crop yield. Crop yield results were averaged for the simulation period (2000–2010) and limited to potentially irrigable land in the nation. The average fresh vegetable yield under CAPS ranged from 1.8–2.8 t ha−<sup>1</sup> for garlic (Figure 7a), 1.4–2.2 t ha−<sup>1</sup> for onions (Figure 7b), 5.5–15.7 t ha−<sup>1</sup> for tomato (Figure 7c), and 8.3–12.9 t ha−<sup>1</sup> for cabbage (Figure 7d). Crop productivity was found to be higher in Oromia and Amhara regions due to the combined effects of the weather and soil condition. The variation of yields for tomato was found relatively high when compared to other vegetables, possibly due to weather variations and the fact that tomato is more sensitive to cold weather. The maximum and minimum allowable temperature for tomato is 27◦C and 10◦C, respectively, for optimal crop growth.

**Figure 7.** Potential crop yields over the irrigable lands (**a**) garlic, (**b**) onion, (**c**) tomato, and (**d**) cabbage. Abbreviations in the map are administrative regions (TG—Tigray, AM—Amhara, AF—Afar, BG—Benshangul Gumaz, AD—Addis Ababa, DD—Dire Dawa, GP—Gambela Peoples, SNNP—Southern Nations, Nationalities and Peoples, and SM—Somali).

#### *3.3. Groundwater Potential for Crop Production under CAPS with Drip Irrigation*

Groundwater depth of less than 30 m is considered feasible for irrigation in the nation Gebregziabher [56]. Thus, depth to groundwater less than 30 m were considered in this study for the estimation of groundwater potential. Worqlul et al., [3] validated the British Geological Survey (BGS) groundwater borehole yield estimates in the central part of Ethiopia using actual groundwater recharge data from the Agricultural Transformation Agency (ATA). The net irrigation requirements for crops were deducted from groundwater potential to identify areas where groundwater fully supports to produce vegetables during the dry season. Figure 8 depicts areas where groundwater potential can support to produce garlic, onion, tomato, and cabbage, respectively.

**Figure 8.** Crop yields over the irrigable lands (**a**) garlic, (**b**) onion, (**c**) tomato, and (**d**) cabbage. Abbreviations in the map are administrative regions (TG—Tigray, AM—Amhara, AF—Afar, BG—Benshangul Gumaz, AD—Addis Ababa, DD—Dire Dawa, GP—Gambela Peoples, SNNP—Southern Nations, Nationalities and Peoples, and SM—Somali).

Table 3 presents the potential of groundwater for different vegetables as a percentage of potentially irrigable land over the administrative regions. For instance, considering the Oromia region, groundwater is enough to irrigate 0.95 Mha, 1.5 Mha, 0.6 Mha, or 0.5 Mha if planting garlic, onion, tomato, or cabbage, respectively, from the potentially irrigable land (6.6 Mha) if planting garlic or onion. That means, 8.9% to 14.3% of the potential land in Oromia could be irrigated depending on the type of crop using groundwater if CAPS with drip irrigation is used. Similarly, 8.8% to 30% of the

potential land in Amhara and 11.6% to 29.8% of the potential land in Southern Nations, Nationalities and Peoples (SNNP) could be irrigated using groundwater. Oromia has the highest groundwater potential (40.9% of total potential) followed by Amhara (20% of total potential), and SNNP (16%). At country level (aggregated from administrative regions), groundwater potential was found to support about 2.3 Mha (Figure 8a), 3.5 Mha (Figure 8b), 1.6 Mha (Figure 8c), and 1.4 Mha (Figure 8d) of land to produce garlic, onion, cabbage, and tomato, respectively in the dry season. Onion has relatively the least irrigation demand and thus has the highest production area coverage using groundwater, followed by garlic, whereas tomato and cabbage have relatively high irrigation demands and thus less area coverage for production using groundwater source.


**Table 3.** Irrigable land and potential of groundwater for various vegetables.

Note: SNNP—Southern Nations, Nationalities and Peoples.

#### **4. Conclusions**

This study is the first of its kind in providing insight into the impacts of the large-scale adaptation of CAPS with drip irrigation on groundwater potential and crop productivity for common vegetables grown in Ethiopia. The results from the MCE technique indicated that there was substantial amount of land for irrigation using groundwater source (~17% of the total landmass). A comparison between suitable areas for irrigation and groundwater potential showed that a modest amount of land (up to 19% of the irrigable land) could be irrigated under CAPS and drip irrigation. The potential of groundwater, however, is a limiting factor to expand irrigated agriculture on suitable lands. Oromia and Amhara regional states provided about 61% of the nation's groundwater potential for irrigation, hence it would be a wise choice for policymakers to consider these results in expanding irrigated agriculture for dry season crop productions.

A comparison between groundwater potential results under CAPS with drip irrigation (1.4 to 3.5 Mha) and CTPS [3], showed that CAPS with drip irrigation significantly increased groundwater potential for irrigation (i.e., 0.6 Mha under CTPS versus 2.2 Mha under CAPS on average). Groundwater potential could be further improved if irrigation scheduling was incorporated with the drip application system. Garlic and onion could be produced in relatively larger areas compared to tomatoes and cabbages due to relatively lower irrigation demand. In addition, CAPS with drip irrigation could significantly improve crop productivity in the nation when compared to CTPS with traditional irrigation. Production potential under CAPS with drip irrigation for cabbage (8.3 t ha−<sup>1</sup> to 12.9 t ha−1) was substantially higher than CTPS, [57], which is 7.9 t ha−<sup>1</sup> for the national average. Therefore, CAPS with drip irrigation is a feasible strategy to improve groundwater potential and crop productivity in the nation. Hence, policymakers should consider CAPS with drip irrigation in expanding small-scale irrigated agriculture.

**Author Contributions:** T.A. contributed to the conceptual design, data collection and acquisition, data analysis, and writing the manuscript. M.J. contributed to the conceptual design, data acquisition, data analysis, and revising

the manuscript for scientific content. A.W.W. contributed to the conceptual design, data acquisition, data analysis, and revising the manuscript for scientific content. M.R. contributed to the conceptual design, data acquisition, and analysis. S.T. contributed to the conceptual design, data acquisition, data analysis, and revising the manuscript for scientific content.

**Funding:** This research and publication are made possible by the generous support of the American people through support by the United States Agency for International Development Feed the Future Innovation Labs for Collaborative Research on Small Scale Irrigation (Cooperative Agreement No. AID-OAA-A-13-0005, Texas A&M University) and Sustainable Intensification (Cooperative Agreement No. AID-OAA-L-14-00006, Kansas State University). The opinions expressed herein are those of the author(s) and do not necessarily reflect the views of the U.S. Agency for International Development.

**Acknowledgments:** We would like to acknowledge the International Water Management Institute (IWMI), and Ethiopian National Meteorological Agency (ENMA) for providing quality data for this research.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Flooding Urban Landscapes: Analysis Using Combined Hydrodynamic and Hydrologic Modeling Approaches**

#### **Manoj K. Jha \* and Sayma Afreen**

Civil, Architectural, and Environmental Engineering, North Carolina A&T State University, Greensboro, NC 27411, USA; safreen@aggies.ncat.edu

**\*** Correspondence: mkjha@ncat.edu

Received: 26 May 2020; Accepted: 11 July 2020; Published: 14 July 2020

**Abstract:** The frequency and severity of floods have been found to increase in recent decades, which have adverse effects on the environment, economics, and human lives. The catastrophe of such floods can be confronted with the advance prediction of floods and reliable analyses methods. This study developed a combined flood modeling system for the prediction of floods, and analysis of associated vulnerabilities on urban infrastructures. The application of the method was tested on the Blue River urban watershed in Missouri, USA, a watershed of historical significance for flood impacts and abundance of data availability for such analyses. The combined modeling system included two models: hydrodynamic model HEC-RAS (Hydrologic Engineering Center—River Analysis System) and hydrologic model SWAT (Soil and Water Assessment Tool). The SWAT model was developed for the watershed to predict time-series hydrograph data at desired locations, followed by the setup of HEC-RAS model for the analysis and prediction of flood extent. Both models were calibrated and validated independently using the observed data. The well-calibrated modeling setup was used to assess the extent of impacts of the hazard by identifying the flood risk zones and threatened critical infrastructures in flood zones through inundation mapping. Results demonstrate the usefulness of such combined modeling systems to predict the extent of flood inundation and thus support analyses of management strategies to deal with the risks associated with critical infrastructures in an urban setting. This approach will ultimately help with the integration of flood risk assessment information in the urban planning process.

**Keywords:** flood analysis; hydrologic modeling; hydrodynamic modeling; SWAT; HEC-RAS; flood zone delineation

#### **1. Introduction**

Over the years, the adverse effects of flooding have increased due to changing climate conditions and human interventions [1]. The major factors which lead communities to increased exposure of such flooding risks include urban expansion, changing demographic features within floodplains, changes in flood regime, and human intervention (social and economic developments) in the ecological system [2]. The hydro-meteorological catastrophes of such floods cannot be totally avoided, but the impacts and after-effects can be managed by developing the effective risk reduction and prevention strategies through applications of advanced geospatial tools and decision support systems [3]. Among the most effective ways of assessing the flood risk to people and infrastructures, one approach is the development and application of flood models which identify areas prone to flooding events and support risk analysis and mitigation processes [4]. Flood modeling has provided an indispensable tool to inform the development of the robust flood risk management strategies to avoid or mitigate the adverse impacts of floods on individuals, communities, and critical infrastructures such as transportation routes, hospitals,

and others. A reliable flood model could alert the flood risk areas and warn the vulnerable population to relocate before the hazards take place. This will potentially alleviate the extent of devastation due to flooding and nullify causalities.

Flood modeling alludes to the process of transformation of rainfall into flood hydrographs which are then hydraulically translated into the depths of water at a spatial scale throughout the watershed [5]. Hydraulic models play an important role in determining flood inundation areas using sets of hydrodynamic equations. One of the major input data is the information on flood hydrographs at multiple locations as upstream and/or lateral boundary conditions. While these data can be obtained from observation data at gaging stations, these are often very limited or not available. Hydrologic rainfall-runoff models are thus frequently used, which when calibrated and validated to a reasonable accuracy, provide hydrograph information at desired locations. There are numerous studies that have independently evaluated the performance of hydrologic models [6–10] and hydrodynamic models [11–14] for their ability to perform the tasks they are developed for. New and improved algorithms have been continuously improving and evolving while capturing more robust simulation approaches and improved capabilities. Over the years, research efforts have been made to improve the numerical accuracy and computational efficiency of hydrodynamic flood models. However, the existing models are still computationally prohibitive for large-scale applications, especially in urban environments where high-resolution representation of complicated topographic features is necessary [14,15]. Similarly, hydrologic models can be computationally efficient in simulating hydrological processes but at the price of representing less detailed physical processes.

There have been several attempts combining hydrodynamic model with hydrological model which may compliment and overcome the shortcomings of either type of modeling approach. The integration of these models can be done various ways. External coupling uses the pre-acquired hydrographs from hydrological models as the upstream and/or lateral boundary conditions for the hydrodynamic models in flood routing analysis through complicated river network systems [16–18]. In the internal coupling method, governing equations of the hydraulic models and hydrological models are solved separately, with information at the shared boundaries updated and exchanged at each or several computational time steps [19]. Fully coupling of these models are not very well understood due to the complexities of reformulating and simultaneously solving governing equations in a single code base [14]. Other approaches include a hybrid method where a 2-D hydrodynamic model is combined with simplified unit hydrographs derived using variations of shallow-water equations [20–24] and integrated catchment models, suitable for flash flood modeling that simulate the complete hydrology and flow, generating runoff, leading to discharge, and then to flooding [25].

Combining hydrodynamic and hydrologic models for flood prediction and analysis is not new. However, the continuous modeling advances and the increase in computational resources over the years make it feasible to conduct flood simulations in high spatial resolution for flood risk assessment. In addition, scientific literature in combining of these two modeling approaches for urban flood simulation is limited [14], and thus underscores the need to continuously develop and apply robust models of improved capabilities for more efficient and accurate analysis. This study demonstrates the flood modeling and analysis method using advanced modeling tools of the present time via the combined or external coupling of hydrodynamic and hydrologic models. The hydrologic model, namely the Soil and Water Assessment Tool (SWAT) [26], was used to derive flow hydrographs at designated locations, which then fed into the hydrodynamic model, namely the Hydrologic Engineering Center's River Analysis System (HEC-RAS) [27] for flood prediction. Both models were independently calibrated and validated using sets of input databases, calibration techniques, observation data, and statistical performance evaluation methods. The combined application was used for flood simulations and the identification of the extent of inundation. The analysis provided the assessment of the impact of flood hazards by the identification of flood risk zones and the threatened infrastructures. The approach was applied in the Blue River Watershed in Missouri, USA, which has historic significance with respect to frequent severe floods. The watershed provides a rich database of observation data

developed over the years. The combined modeling system provides crucial flood risk information necessary for the development of an accurate and reliable forecasting system for assist in evacuation, relief operation route, cost estimation of the damaged properties, and other pertinent information.

#### **2. Materials and Methods**

The method included development of two mathematical models: SWAT and HEC-RAS. The SWAT model was developed via Geographic Information System (ArcGIS) interface of the model, called ArcSWAT, using a set of spatial datasets including topography data (digital elevation model), land use data (National Land Cover Database), and soil types and soil characteristics data (State Soil Geography Database), as well as time-series daily dataset on meteorological parameters, including precipitation, maximum and minimum temperature, wind speed, solar radiation, and relative humidity. The model was calibrated for the overall watershed hydrological water balance followed by monthly streamflow at a gaging location at the watershed outlet by comparing model simulated values with the observation data collected at the gaging site. Once the model is calibrated and validated with satisfactory statistical performance measures, it was then used to develop a series of simulated streamflow hydrographs to be used as an input to the HEC-RAS model.

The HEC-RAS model was developed for river segments within the watershed. The geometric data and the Manning's roughness coefficient values (n) were established for the modeling setup using ArcGIS interface of the model, called HEC\_geoRAS. It was then calibrated and validated using the past flood data collected from the USGS gaging stations within the study area.

The flow chart in the Figure 1 portrays different step of the processes performed in this study. Based on the streamflow input from calibrated SWAT model, the calibrated HEC-RAS model predicts flood levels and the extent of the flood in the surrounding landscapes. Further analysis was conducted to identify vulnerabilities of critical infrastructures including hospitals, railroads, airports, and transportation routes by examining the proximities of these infrastructures from the flood zones.

**Figure 1.** Schematic of data and models for flood prediction and analysis.

#### *2.1. Study Area*

The Blue River also known as Big Blue River is a part of tributaries of Missouri River located in Kansas City, Missouri (Figure 2). The Blue River watershed extends from the south of Johnson County in Kansas State into the State of Missouri and drains an area of 658.9 km<sup>2</sup> into the Missouri River in Kansas City, Jackson County, Missouri. The Blue River Watershed spreads over roughly one-half of the Kansas City metropolitan area south of the Missouri River. The watershed course through two states (Missouri and Kansas), four counties (Johnson and Wyandotte in Kansas; Jackson and Cass in Missouri), and 11 municipalities [28]. The Blue River is 39.8 mile (64.1 km) long stream, and the mouth of the river is at 221 feet elevation in the east of Johnson County near the borders of the states of Kansas and Missouri. The percentage of Blue river watershed within the state of Missouri is about 46%, which is within the Kansas City metropolitan area. The area is moderately to highly developed

and contain a mix of residential and commercial structures and is subjected to flooding every year due to urban development, dense soils, and the configuration of the Blue River basin [29]. The lower part of the watershed is primarily industrial, whereas the middle and upper part are rapidly being converted to residential areas [30]. Due to the flood sensitive nature of this river zone, U.S. Geological Survey (USGS) has been studying this area closely and acquired an extensive dataset over the time. The abundance of data in this location was very helpful in accurately calibrating the mathematical models for flood prediction and the analysis objective of this study.

**Figure 2.** Blue River Watershed in Kansas City, Missouri, USA.

#### *2.2. Data Collection*

The development of the combined modeling system required extensive data collection from various sources. Table 1 lists some of the major data types and their sources. Subsections below provided more specific information on data collection efforts.

**Table 1.** Datasets and their sources used for creating the combined modeling system.


#### *2.3. Hydrologic Model Overview–SWAT*

SWAT is a river basin scale model developed to predict the impact of land management practices on water, sediment and agricultural chemical yields in large complex watershed with varying soils, land use and management conditions over extended periods of time [6,20]. SWAT is a long-term yield model extensively used to simulate watersheds on multiple spatial–temporal scales including hydrological processes [7,9,31,32], fate and transport of sediment and nutrients [33–35], land use change [36], climate change [37–43], and others.

The major inputs required to develop a SWAT model are topographical data which are used to define stream network and delineate a number of subwatersheds; land use data, soil data, and slope information to delineate each subwatershed into hydrologic response units (HRUs) which represents unique combination of land use, soil types, and slope; (3) the daily time-series information on meteorological parameters, and (4) the model's inbuilt databases and initialization assumptions. The outputs include spatiotemporal time-series data on water balance components, streamflow, sediment and nutrient loadings, and others.

#### 2.3.1. Modeling Setup and Watershed Delineation

The Blue River Watershed was delineated using stream generation functionality in ArcGIS based on the supplied 10 m resolution DEM projected in Northern America Datum NAD\_1983 UTM zone 15N. The delineated subwatersheds (Figure 3) were further subdivided into multiple lumped units within each subwatershed. These lumped units are called HRUs, a unique combination of land use, slope, and soil types. An HRU represents a percentage of a sub-watershed area and not spatially identified within a subwatershed. All water balance calculations and modeling simulations are conducted at the HRU level. Outputs from each HRU within a subwatershed are aggregated at the subwatershed level which are then routed through the streams leading to the next downstream subwatershed. Outputs from each subwatershed are subsequently routed all the way to the watershed outlet on a daily basis. Muskinghum method was used in the hydrologic routing process. Other methods include Curve-Number approach for flow generation, and Penman-Moneith method for the estimation of evapotranpiration.

**Figure 3.** Delineation of Blue River Watershed and location of NOAA weather stations.

The land use data was obtained from National Land Cover Dataset (NLCD) from 2011 (https: //www.mrlc.gov/nlcd11\_data.php). Classification of the land use data was found to cover dense urban areas (48%), developed open area (20%), pasture/hay (13.5%), forest (9%), cultivated crops (7%), grassland (1%), open water (0.6%), wetland (0.5%), shrub (0.2%), and barren land (0.2%). The soil data source was State Soil Geographic (STATSGO) database (https://catalog.data.gov/dataset/statsgo) which was already included in the ArcSWAT inbuilt datasets. Figure 4 presents reclassified land use, soil types and slope categories used in HRU delineation. The time-series meteorological information was obtained for 9 weather stations located in and around the watershed (Figure 3) using data download function at the NOAA-NCDC website (https://www.ncdc.noaa.gov/cdo-web/datatools/findstation).

**Figure 4.** Reclassified land use, soil and slop data used in the HRU delineation.

#### 2.3.2. SWAT Simulation and Calibration/Validation Approach

The SWAT modeling setup was executed on a daily time step for 8 years of simulation from 2010–2017 (2-year warm-up, 4-year calibration, and 2-year validation period). Calibration and validation of the SWAT model was performed using SWAT's Calibration and Uncertainty Program, SWAT-CUP [44]. This autocalibration tool can perform sensitivity analysis, calibration, validation, and uncertainty analysis. Sensitivity analysis of the model's hydrologic parameters were conducted and ten parameters were identified as the most sensitive. There are runoff curve number (CN2), soil evaporation compensation factor (ESCO), water holding capacity of the soil (SOL\_AWC), plant uptake compensation factor (EPCO), groundwater revap coefficient (GW\_REVAP), base flow alpha factor (ALPHA\_BF), threshold depth of water in the shallow aquifer required for return flow (REVAPMN), groundwater delay (GW\_DELAY), surface runoff lag coefficient (SURLAG), and threshold depth of water in the shallow aquifer required for return flow to occur (GWQMN). The details of these model parameters can be found in the User's Manual. In the calibration process, defaulted values of these parameters were adjusted within their permissible ranges to a final calibrated value after comparing simulated results with the observations with acceptable performance measures tested through statistical procedures [39]. The autocalibration tool identified the best fitted values of all ten parameters while fitting the monthly comparison of simulated flow values with the observations from gaging stations at the watershed outlet. Statistical evaluation was conducted using four indicators: coefficient of determination (R2), Nash–Sutcliff's efficiency (NSE), percentage bias (PBIAS), and RMSE standard deviation ratio (RSR). Calibration process concluded with satisfactory performance in visual comparison and acceptable statistical comparisons. During the validation process, the model was executed with already defined value of calibration parameters (no further), followed by the same statistical evaluation as that of the calibration duration.

#### *2.4. Hydrodynamic Model Overview–HEC-RAS*

HEC-RAS [21] can perform one and two-dimensional hydrodynamic calculations for a full network of natural and constructed channels. The major capabilities of HEC-RAS are user interface, hydraulic analysis components, data storage and management, graphics and reporting, and RAS Mapper. The HEC-RAS system accommodates several river analysis components for steady and unsteady flow water surface profile computations, movable boundary sediment transport computations and water quality analysis. Hydrodynamic equations calculate water surface elevations at all locations of interest for a given peak flood. The major data inputs are river geometric cross-section data, river floodplain data (length, elevation), the distance between successive river cross-sections, manning roughness coefficient values (n) for the land use type covering the river and the floodplain area, and boundary conditions (flow hydrograph and normal depth). Under steady flow, the boundary conditions are a discharge

upstream and a stage downstream. The model proceeds to calculate stages throughout the interior points, keeping the discharge constant in space. Under unsteady flow, a discharge hydrograph is the upstream boundary and a discharge-stage rating at the downstream boundary. The model calculates discharges and stages throughout the interior points. Unsteady flow simulation uses the Saint-Venant equations or the diffusion wave equations using an implicit finite volume algorithm. The outputs from the HEC-RAS model include water surface elevations, rating curves, hydraulic properties (energy grade line slope, elevation, flow area, velocity), and visualization of the extent of flooding.

The steady flow simulation based on a peak flow discharge throughout the river line represents the water flow without any change over time. It consists of flow regime, discharge information and boundary condition. Multiple profiles can be created with different discharge values. The unsteady flow simulation is developed with a series of discharge data with respect to the time of occurrence. The data required for the unsteady flow simulations include boundary conditions (external and internal) and initial conditions.

The calibration of the model was initiated by calibrating for the steady flow simulation followed by the calibration of the unsteady flow simulations. The model was calibrated for a peak flow event at five USGS gaging stations (Figure 5) with adjustments in the parameters such as Manning's n value and required boundary conditions [21]. The upstream boundary condition was a flow hydrograph and the downstream boundary condition was normal depth for steady state simulation. The HEC-RAS was executed to develop water level data which were compared with observed water elevations. After the calibration, the model was validated for two other flood events at all five stations based on the calibrated parameters.

**Figure 5.** Location of USGS Gaging Stations used in the calibration.

#### **3. Results and Discussion**

#### *3.1. Clibration and Validation of the Hydrolgic Model*

The SWAT model developed for the Blue River Watershed was calibrated using the automated calibration technique (SUFI-2) for flow by comparing simulated values with the observations at the watershed outlet (USGS 06893500, Blue river at Kansas City, MO, USA). Table 2 lists all parameters used in the calibration process with their permissible ranges and the final fitted values after the calibration.


**Table 2.** List of parameters used for calibration with their ranges and the fitted value.

Figure 6 shows the comparison of simulated versus observed streamflow at the watershed outlet using monthly data. The comparison seems to match well except for slight underprediction of peaks. The hydrograph seems to follow very close for its recession, baseflow and other patterns. Table 3 provides values of statistical measures for both calibration and validation periods. Overall, these values show a strong correlation of the simulated streamflow with the observation. Thus, it can be concluded that the SWAT model was well-calibrated to simulate streamflow with reliable performance in the Blue River Watershed. The calibrated model output was used to generate discharge (streamflow) data at several locations within the watershed to be used as input for the HEC-RAS model.

**Figure 6.** Monthly comparison of simulated and observed streamflow data for the calibration (2012–2015) and validation (2016–2017) periods along with precipitation data for the entire range (cms: m3/s).


**Table 3.** Statistical Evaluation of the calibration & validation of the Blue River Watershed.

#### *3.2. Clibration and Validation of the Hydrodynamic Model*

The hydrodynamic model developed for the Blue River by HEC-RAS was calibrated and validated at the five USGS gaging stations located on the River (Figure 5). Model simulated water surface elevations were compared with the observed water surface elevations at the USGS gages. The Manning's roughness coefficient (n) values were adjusted until the simulated values match closely with the values at USGS gages. The calibration was performed for the flood event of 17 May 2015 and the results are presented in Table 4. The results show that the difference between the observed and simulated values were very minimal and thus justify the model's ability to simulate water surface levels. The statistical evaluation using two performance measures NSE and R2 yielded a strong correlation with value of 0.989 and 0.98 respectively. The validation process was conducted for two peak events: the floods on 27 April 2016 and 22 September 2017. The difference in observed and simulated water surface elevations were small and therefore the results are considered satisfactory which is portrayed in Table 5. It can be concluded that the HEC-RAS model developed for the Blue River performed very well to simulate water surface elevations.


**Table 4.** HEC-RAS Model Calibration for the flood event of 17 May 2015.



#### *3.3. Flood Inundation Mapping*

Accurate prediction of the flood inundation area for a given flood event is necessary for risk mitigation strategies. Over the last few decades, there have been vast improvements in flood inundation modeling [46]. While empirical methods are considered adequate for flood monitoring and post-disaster assessment, hydrodynamic models are critical to represent detailed flow dynamics to investigate impacts of management strategies such as dam break, flash floods, etc. Simplified conceptual models are usually adopted for probabilistic flood risk assessment on a large floodplain with well-defined channels. Different modeling approaches produce different predictions highlighting the uncertainty associated with the modeling practices, which is mainly generated by uncertainty in the design flow, terrain elevations, water surface elevations, and accuracy of the techniques used for mapping the inundation area [47].

In this study, the flood inundation area was developed using ArcGIS based on the HEC-RAS simulation of desired flood event. The pseudo-validation of the developed inundation map was conducted by comparing it with inundation maps already developed by the USGS which was available to view/download from the USGS Flood Inundation Mapper (https://wimcloud.usgs.gov/apps/FIM/ FloodInundationMapper.html). Figures 7 and 8 show the comparison between inundation maps created by HEC-RAS simulation (right figures) with the USGS inundation maps (left figures) at two separate locations. The comparison was done visually by comparing important features along the Blue River. The maps fairly show comparable zones of inundated area in both cases. It is important to note that the discharge data used by HEC-RAS in developing the inundation extent was based on the "simulated" discharge data from the hydrologic model, which may have contributed greatly to the disagreements between the two maps. Moreover, the comparison was against the another simulated map as explained in the disclaimer by the USGS (https://fim.wim.usgs.gov/fim/) which states that "the flood boundaries shown were estimated based on water stages (water-surface elevations) and streamflows at selected USGS streamgages. Water-surface elevations along the stream reaches were

estimated by steady-state hydraulic modeling, assuming unobstructed flow, and using streamflows and hydrologic conditions anticipated at the USGS streamgage(s)".

**Figure 7.** Comparison of the Inundation Map created by the HEC-RAS simulation (**right-side**) with the USGS (**left-side**) generated Inundation Map beside the Winner Park in Kansas City, MO.

**Figure 8.** Comparison of the Inundation Map created by the HEC-RAS simulation (**right-side**) with the USGS (**left-side**) generated Inundation Map beside the Truman Sports Complex in Kansas City, MO.

#### *3.4. Vulnerability Assessment on Infrastructures*

Vulnerability assessment is an essential part of the flood management and preparedness process to reduce the impact. It requires an in-depth analysis of many factors including location of critical infrastructures such as hospitals, transportation routes and density of the population in order to increase the effectiveness of emergency plans. Indicators of flood hazard generally include the flood extent, water depth, flow velocity, duration, propagation of waterfront, and the rate at which the water rises [48]. These parameters are then linked with the economic damages and other vulnerability assessment. There are many studies linking inundation extent to determine economic losses or risks for planning purposes such as insurance, etc. [49]. The vulnerability criterion focused on human stability (not economic values) has also been analyzed using slipping, toppling, and drowning as indicators of human stability [50]. A flood modeling simulation in an urban area used inundation maps to analyze transport accessibility and human safety on pedestrians and drivers for its implications on emergency routes and service areas [51,52]. A comparative study of hydraulic models evaluated their capabilities for estimations of the vulnerability assessment to capture the uncertainties in the prediction [53].

In this study, we present the vulnerability assessment in terms of critical infrastructures being exposed to floodwaters by proximity to flood inundation extent over the study area. The analysis was based on flood event of May 2015. The infrastructures selected for vulnerability assessment purposes were local hospitals, transportation routes, airport facilities, and railroad networks. These infrastructures are very crucial for emergency responses such as for mitigation, preparedness, recovery, and response. For example, emergency response teams could use the inundation maps to optimize their routes to the flood affected locations, avoiding the inundated transportation routes. The inundation maps could also assist in the allocation of recovery resources from the high-risk zones following a flood event. Inundation maps could be created assuming a future storm event causing a flood, and therefore highly threatened flood zones could be alarmed ahead of time, thereby saving lives and resources.

#### 3.4.1. Impact of Inundation on Local Hospitals

Hospitals are one of the major locations highly prioritized in the disaster mitigation process. Figure 9 depicts four hospitals that could be threatened due to similar flood situation like as May 2015. One of the four risked hospitals was identified to be almost under inundation and the rest could be impacted with an increase of a few units of water level caused by a more hazardous flood. The surrounding hospitals could be indirectly affected due to the closure of the nearby transportation routes. This vulnerability identification could help the management authorities warn the hospitals listed under the adverse impact, ahead of any upcoming hazards. The vulnerable hospitals showed in Figure 9 are listed in Table 6 with their distance from the inundation area. To understand the different levels of flood vulnerability, a ranking is given to the hospitals with respect to the distance of the hospitals from the flood extent at their respective locations.

**Figure 9.** Location of hospitals in and around the inundated area due to Flood of May 2015.


**Table 6.** List of hospitals vulnerable to the flood of May 2017 in Kansas City, Missouri.

**Figure 10.** Location of major transportation routes under inundation due to the flood of May 2015.

#### 3.4.2. Impact of Inundation on Transportation Routes

Intense precipitation is the foremost cause of weather-related disruption to the transportation sector [39]. It can cause severe damage to an area by obstructing the movement of people and goods, hampering social and economic functionality. The flooding on major transportation routes, like interstates and state highways, cut off the flooded zone's communication with the surrounding area which also delays the emergency management processes. Figure 10 shows the transportation routes that are directly affected due to the flooding scenario modeled for May 2015 flood. Parts of the interstate I70, Blue Parkway, Highway I435, Highway US 40, and Independence Avenue are found to be under the impact of inundation caused by the flood of May 2015 as simulated by HEC-RAS.

#### 3.4.3. Impact of Inundation on Airport

Figure 11 shows the threatened location of Airports due to flood of May 2015. One of the airport facilities will be directly affected by the flood and the other one is very close by the inundated regions. A ranking is given to airports for flood vulnerability with respect to the distance of the facility from the inundation map at their respective locations (Table 7).

**Figure 11.** Location of airport under inundation due to the Flood of May 2015.


**Figure 12.** Location of railroad routes under inundation due to the Flood of May 2015.

3.4.4. Impact of Inundation on Railroad Facilities

The railroad is one of the most used routes in big cities and metropolitan areas. Inundated railroads could cause fatal accidents that would affect huge numbers of people travelling in the trains. Figure 12 shows the location of railroad routes under inundation due to the flood of May 2015 as simulated by HEC-RAS. Railroads that are subjected to inundation will result in obstruction of the whole rail route within and surrounding the city. Table 8 provides the coverage of railroads under the flood.


**Table 8.** List of railroads vulnerable to the flood of May 2017 in Kansas City, Missouri.

#### **4. Discussion and Conclusion**

This study presents a systematic approach of combining hydrodynamic model HEC-RAS with hydrologic model SWAT in delineating flood inundation zones and subsequently assessing the vulnerability of critical infrastructures in the Blue River Watershed in Kansas City, Missouri. Both models were independently calibrated and validated using various datasets and proven strategies. The HEC-RAS flood simulation model was found to be suitable in simulating flood events and spatially depicting the vulnerability of the region towards a hazard event in terms of inundation extent, whereas SWAT was proven to be a powerful tool in generating simulated flood hydrographs at desired locations. The models developed can be said to have generated reliable quantified output based on the statistical evaluation results. This study approach provides quantified information on the hydrologic modeling, hydrodynamic modeling, and flood prediction and analysis for flood management strategies.

The catastrophic possessions of flood disaster could be mitigated by integrating scientifically reliable information with the flood inundation map developed using this study approach. Vulnerability assessment approach used in this study for identifying and providing a vulnerability rank based on proximity to flood area is a simple yet powerful approach. It not only identified most to least vulnerable critical infrastructures, but also provided enough information for flood preparedness processes that could significantly reduce the impact. The approach could easily be extended for the vulnerability evaluation of other infrastructures in order to estimate economic losses, navigation route of people including high density area, and other region-specific important factors. Moreover, futuristic higher magnitude flood events can be simulated to assess magnified vulnerability and associated risks. Land use planning decisions could be made based on the flood inundation map which indicates the floodplains. Following such approaches will help save lives and resources at the same time, and provide a proven and more accurate way to contest the uncertainties of the natural events causing flood.

The flood modeling system presented in this study is an integrated system to stakeholders to investigate potential mitigation options and strategies in response to expected flooding scenarios. The use of hydrologic model in flood modeling proves very useful in studying alternative "what if" scenarios such as impacts of projected land use changes, climate variabilities, urban planning strategies and others. For all plausible scenarios, a well-calibrated hydrologic model of the region can easily simulate new conditions and yield changes in flow hydrographs at desired locations, which can then be translated into flood depths over the region using hydrodynamic models. A previous flood modeling attempt in Kansas River basin, close to the study watershed, used hydrologic model HEC-HMS (Hydrologic Modeling System) to generate estimates of peak flows for design storm for different land use scenarios [4]. The output was then used to execute the HEC-RAS model for estimates of water elevations and flood inundation extents for those design storms and land use scenarios. The results provided useful information, however the study was designed at a macro scale of change which does not necessarily reflect the flooding impacts at smaller scale.

Added benefits of this combined modeling presented in this study system also includes the flexibility of hydrodynamic modeling for testing flood reduction or mitigation strategies through channel modifications and other best management practices within the floodplains areas. Such a modeling system also enables the assessment and determination of vulnerable areas that will not be able to receive effective adaptation solutions, which then calls for drastic measures to mitigate flood-prone impacts.

Such modeling application also comes with several limitations, including the availability, resolution, and accuracy of the data for the development, calibration, and validation of the models, the integration methods such as external coupling approach used in this study, flexibility provided by the statistical performance measures for the approval of a robust model, and ability to replicate/simulate best management practices with a degree of accuracy to support flood mitigation and adaptation options. As such, in the application presented in this study, major limitations in using hydrologic model may include (a) limited accuracy in model calibration and validation: resolution of the input data and limited set of observation data, e.g., calibrating only for the monthly flow and only at the watershed outlet, and (b) simulated data to be exported as input to another model: calibrated models produced simulated hydrographs to be used as input boundary conditions in hydrodynamic modeling. Similarly, sources of uncertainties in using hydrodynamic modeling may include input data quality of topography and surface roughness characterization as it affect both flow area and velocity [54,55]. The role of topography on flood studies has been discussed in many past studies [56], but the role of surface roughness has received less attention. A recent study exhibits the sensitiveness of surface roughness and highlights the source of uncertainties in flood modeling studies [57]. Reducing the uncertainty in surface roughness will greatly enhance the calculation of flood extent on landscapes. It is also noteworthy to mention that, while surface roughness plays an important role in simulating accurate flow hydrodynamics in both the channel and floodplain, Manning's n is not viewed as important (less-sensitive parameter) in simplified hydrological modeling.

Moreover, the errors in the simulation results from the combined modeling system of flood analysis arise from various sources of uncertainties, as discussed above, which probably propagates in an unknown and non-linear fashion. The next level of analysis should shed light on the assessment and quantification of these errors and how these propagate through the modeling system.

**Author Contributions:** Conceptualization, M.K.J.; methodology, M.K.J. and S.A.; software, M.K.J.; validation, M.K.J. and S.A.; formal analysis, S.A.; investigation, S.A. and M.K.J.; resources, S.A.; data curation, S.A.; writing—original draft preparation, S.A. and M.K.J.; writing—review and editing, M.K.J. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

MDPI St. Alban-Anlage 66 4052 Basel Switzerland Tel. +41 61 683 77 34 Fax +41 61 302 89 18 www.mdpi.com

*Water* Editorial Office E-mail: water@mdpi.com www.mdpi.com/journal/water

MDPI St. Alban-Anlage 66 4052 Basel Switzerland

Tel: +41 61 683 77 34 Fax: +41 61 302 89 18