Modeling the Impacts of Conservation Agriculture with a Drip Irrigation System on the Hydrology and Water Management in Sub-Saharan Africa

: The agricultural system in Sub-Saharan Africa (SSA) is dominated by traditional farming practices with poor soil and water management, which contributes to soil degradation and low crop productivity. This study integrated ﬁeld experiments and a ﬁeld-scale biophysical model (Agricultural Policy Environmental Extender, APEX) to investigate the impacts of conservation agriculture (CA) with a drip irrigation system on the hydrology and water management as compared to the conventional tillage (CT) practice. Field data were collected from four study sites; Dangishita and Robit (Ethiopia), Yemu (Ghana), and Mkindo (Tanzania) to validate APEX for hydrology and crop yield simulation. Each study site consisted of 100 m 2 plots divided equally between CA and CT practices and both had a drip irrigation setup. Cropping pattern, management practices, and irrigation scheduling were monitored for each experimental plot. Signiﬁcant water savings ( α = 0.05) were observed under CA practice; evapotranspiration and runoff were reduced by up to 49% and 62%, respectively, whereas percolation increased up to three-fold. Consequently, irrigation water need was reduced in CA plots by about 14–35% for various crops. CA coupled with drip irrigation was found to be an efﬁcient water saving technology and has substantial potential to sustain and intensify crop production in the region. A m digital elevation watershed characteristics. Soil map were used for the Dangishita and Robit sites. A harmonized world soil map prepared by the Food and Agricultural Organization (FAO) were used for the case of Yemu and Mkindo. Ground climate stations were used for Dangishita and Robit sites, whereas bias-corrected CFSR data were used for Yemu and Mkindo sites. Various management activities and cropping patterns were monitored at each site. The management activities include tillage and planting details; population density; mulch application rate and date; fertilizer and pesticide details, irrigation application rate, and the amount and harvesting date. Groundwater wells were used in Dangishita and Robit sites as a source of drip irrigation, whereas surface water (river) was used for Yemu and Mkindo sites. Farmers used a manual pulley system to lift water from groundwater wells to a water tank. Farmers in Yemu used motorized pumps with pipes to deliver the water from the river pond to the water tanker, whereas farmers in Mkindo used a faucet (installed their home) to ﬁll the water tank. The sizes of the water tanks were 500 L (in Dangishita, Robit, and Mkindo) and 200 L (in Yemu), which were placed near vegetable gardens about one and a half meter above the ground. The water dripped from a storage tank to the plots until the tank was empty (i.e., ﬁxed amount of water per single irrigation). Farmers decide the frequency of irrigation through ﬁeld observation (i.e., by touching the soil using their hand and observing soil moisture). The date of irrigation application and the number of water tanks used were monitored. A management ﬁle was developed, which included irrigation the amount and date of application, and integrated into the model separately for (CA and CT management).


Introduction
Agricultural production continues to face several challenges in Sub-Saharan Africa (SSA) leading to an insufficient food supply. The population significantly increased from 180 million to 962 million from 1950 to 2015 in SSA [1]. This rapid increase in population imposes a pressure on the already stressed food production system. Insufficient food supply leads to malnutrition, which accounts for more than one-third of all children's death in the region [2]. Another challenge is the rainfall-dependent farming system, which makes it susceptible to climate variability such as drought [3]. Also, the expansion of traditional farming practices aiming to increase in food supply resulted in environmental deterioration due to conventional tillage practices [4,5]. These challenges call for a sustainable growth in food production system that may come from (1) growing high value and nutritious food types, such as fruits and vegetables; (2) using efficient water use strategies (irrigation technologies) that can maximize production and support multiple cropping seasons; (3) enabling of Ghana and Tanzania, respectively ( Figure 1). A total of 43 experimental plots (Robit-6 plots, Dangishita-7 plots, Yemu-15 plots, and Mkindo-15 plots) were established on a 100 m 2 (paired "t" design), in which half of this site was assigned randomly to CA and another half to CT ( Figure 2). Low-cost drip irrigation was installed for both cases. Simple water-lifting technologies were introduced to extract water from groundwater wells and deliver it into water storage tanks (usually 500 L in size). Irrigation water was distributed to the fields using gravity flow from these tanks, installed about 1.5 m above the ground. Farmers could use their intrinsic knowledge to decide the frequency of irrigation (i.e., depending on vegetable water need and on-site observation of soil moisture). Dangishita and Robit sites were situated on Chromic Luvisols soil (hydrologic group C) whereas Yemu and Mkindo sites were on Ferric Luvisols soil (hydrologic group A) and Ferallic Cambisols soil (hydrologic group D), respectively. The infiltration and water transmission rate decrease from hydrologic soil group A to D. Table 1 shows detailed soil characteristics of experimental sites derived using a soil-plant-atmosphere-water (SPAW) field and pond hydrology program. Inputs for the SPAW hydrology program were provided from a harmonized world soil database [37].
Watershed and plot level parametrization were made for Dangishita, whereas plot level parametrization was made for the other sites (due to streamflow data limitation). Streamflow gauging station records in Dangishita were used to verify APEX model simulation at the watershed scale. Figure 3 shows the Dangishita watershed extracted from a 30 m resolution digital elevation model at the outlet, which had a streamflow gauging station, and the experimental plots were close to the watershed outlet. The size of Dangishita watershed was 57.5 km 2 and the majority of the landscape, about 80%, was had less than a 10% slope. Climatic data for the study sites were obtained from nearby weather stations (Dangila for Dangishita sites; and Bahir Dar for Robit sites) ( Figure 3) and nearby climate forecast system reanalysis (CFSR) data for Yemu (Ghana) and Mkindo (Tanzania) due to lack of ground weather data close to the study sites. The CFSR data for Yemu  and Mkindo (1980Mkindo ( -2010 obtained from Texas A&M was bias-corrected with a linear bias correction as indicated in Reference [38]. The mean monthly rainfall of the study sites for Dangishita and Robit Ghana and Tanzania, respectively ( Figure 1). A total of 43 experimental plots (Robit-6 plots, Dangishita-7 plots, Yemu-15 plots, and Mkindo-15 plots) were established on a 100 m 2 (paired "t" design), in which half of this site was assigned randomly to CA and another half to CT ( Figure 2). Low-cost drip irrigation was installed for both cases. Simple water-lifting technologies were introduced to extract water from groundwater wells and deliver it into water storage tanks (usually 500 L in size). Irrigation water was distributed to the fields using gravity flow from these tanks, installed about 1.5 m above the ground. Farmers could use their intrinsic knowledge to decide the frequency of irrigation (i.e., depending on vegetable water need and on-site observation of soil moisture). Dangishita and Robit sites were situated on Chromic Luvisols soil (hydrologic group C) whereas Yemu and Mkindo sites were on Ferric Luvisols soil (hydrologic group A) and Ferallic Cambisols soil (hydrologic group D), respectively. The infiltration and water transmission rate decrease from hydrologic soil group A to D. Table 1 shows detailed soil characteristics of experimental sites derived using a soil-plant-atmosphere-water (SPAW) field and pond hydrology program. Inputs for the SPAW hydrology program were provided from a harmonized world soil database [37]. Watershed and plot level parametrization were made for Dangishita, whereas plot level parametrization was made for the other sites (due to streamflow data limitation). Streamflow gauging station records in Dangishita were used to verify APEX model simulation at the watershed scale. Figure 3 shows the Dangishita watershed extracted from a 30 m resolution digital elevation model at the outlet, which had a streamflow gauging station, and the experimental plots were close to the watershed outlet. The size of Dangishita watershed was 57.5 km 2 and the majority of the landscape, about 80%, was had less than a 10% slope. Climatic data for the study sites were obtained from nearby weather stations (Dangila for Dangishita sites; and Bahir Dar for Robit sites) ( Figure 3) and nearby climate forecast system reanalysis (CFSR) data for Yemu (Ghana) and Mkindo (Tanzania) due to lack of ground weather data close to the study sites. The CFSR data for Yemu  and Mkindo (1980-2010) obtained from Texas A&M was bias-corrected with a linear bias correction as indicated in Reference [38]. The mean monthly rainfall of the study sites for Dangishita and Robit

APEX Model Description, Inputs, and Data Monitoring
APEX is an extension of the Environmental Policy Integrated Climate (EPIC) model [39]. APEX, a biophysical model [39], is capable of evaluating the effects of various soil and water management practices on the hydrology of the system, crop growth, and other environmental factors [40]. It has the capability of modeling wide ranges of conservation practices [23,35,[41][42][43][44]. APEX simulates watershed processes based on weather data, soils characteristics, topography, vegetation, and management practices [40]. Multiple options are available in the APEX model to estimate evapotranspiration, surface runoff, peak runoff rate, and available soil water capacity to derive hydrology of the system [45]. Recently, the "ADDMULCH" management was included in the APEX model (2017 release) to simulate organic mulch cover conservation practices on the soil surface. Detailed description of the APEX model major components ( Figure 5) can be found in Reference [46]. The APEX model has been applied to evaluate the effects of conservation practices [35,41,43,44].
The APEX model inputs include Geographic Information System (GIS) data layers, climatic data, and management practices. The GIS data layers are digital elevation model, soil and land use or crop covers. A 30 m digital elevation model (DEM) was obtained from the United States Geographical

APEX Model Description, Inputs, and Data Monitoring
APEX is an extension of the Environmental Policy Integrated Climate (EPIC) model [39]. APEX, a biophysical model [39], is capable of evaluating the effects of various soil and water management practices on the hydrology of the system, crop growth, and other environmental factors [40]. It has the capability of modeling wide ranges of conservation practices [23,35,[41][42][43][44]. APEX simulates watershed processes based on weather data, soils characteristics, topography, vegetation, and management practices [40]. Multiple options are available in the APEX model to estimate evapotranspiration, surface runoff, peak runoff rate, and available soil water capacity to derive hydrology of the system [45]. Recently, the "ADDMULCH" management was included in the APEX model (2017 release) to simulate organic mulch cover conservation practices on the soil surface. Detailed description of the APEX model major components ( Figure 5) can be found in Reference [46]. The APEX model has been applied to evaluate the effects of conservation practices [35,41,43,44].
The APEX model inputs include Geographic Information System (GIS) data layers, climatic data, and management practices. The GIS data layers are digital elevation model, soil and land use or crop covers. A 30 m digital elevation model (DEM) was obtained from the United States Geographical Survey (USGS) website and used to extract watershed characteristics. Soil map prepared by the Ministry of Water, Irrigation and Energy (MoWIE) of Ethiopia were used for the Dangishita and Robit sites. A harmonized world soil map prepared by the Food and Agricultural Organization (FAO) were used for the case of Yemu and Mkindo. Ground climate stations were used for Dangishita and Robit sites, whereas bias-corrected CFSR data were used for Yemu and Mkindo sites. Various management activities and cropping patterns were monitored at each site. The management activities include tillage and planting details; population density; mulch application rate and date; fertilizer and pesticide details, irrigation application rate, and the amount and harvesting date. Groundwater wells were used in Dangishita and Robit sites as a source of drip irrigation, whereas surface water (river) was used for Yemu and Mkindo sites. Farmers used a manual pulley system to lift water from groundwater wells to a water tank. Farmers in Yemu used motorized pumps with pipes to deliver the water from the river pond to the water tanker, whereas farmers in Mkindo used a faucet (installed at their home) to fill the water tank. The sizes of the water tanks were 500 L (in Dangishita, Robit, and Mkindo) and 200 L (in Yemu), which were placed near vegetable gardens about one and a half meter above the ground. The water dripped from a storage tank to the plots until the tank was empty (i.e., fixed amount of water per single irrigation). Farmers decide the frequency of irrigation through field observation (i.e., by touching the soil using their hand and observing soil moisture). The date of irrigation application and the number of water tanks used were monitored. A management file was developed, which included irrigation the amount and date of application, and integrated into the model separately for (CA and CT management). Survey (USGS) website and used to extract watershed characteristics. Soil map prepared by the Ministry of Water, Irrigation and Energy (MoWIE) of Ethiopia were used for the Dangishita and Robit sites. A harmonized world soil map prepared by the Food and Agricultural Organization (FAO) were used for the case of Yemu and Mkindo. Ground climate stations were used for Dangishita and Robit sites, whereas bias-corrected CFSR data were used for Yemu and Mkindo sites. Various management activities and cropping patterns were monitored at each site. The management activities include tillage and planting details; population density; mulch application rate and date; fertilizer and pesticide details, irrigation application rate, and the amount and harvesting date. Groundwater wells were used in Dangishita and Robit sites as a source of drip irrigation, whereas surface water (river) was used for Yemu and Mkindo sites. Farmers used a manual pulley system to lift water from groundwater wells to a water tank. Farmers in Yemu used motorized pumps with pipes to deliver the water from the river pond to the water tanker, whereas farmers in Mkindo used a faucet (installed at their home) to fill the water tank. The sizes of the water tanks were 500 L (in Dangishita, Robit, and Mkindo) and 200 L (in Yemu), which were placed near vegetable gardens about one and a half meter above the ground. The water dripped from a storage tank to the plots until the tank was empty (i.e., fixed amount of water per single irrigation). Farmers decide the frequency of irrigation through field observation (i.e., by touching the soil using their hand and observing soil moisture). The date of irrigation application and the number of water tanks used were monitored. A management file was developed, which included irrigation the amount and date of application, and integrated into the model separately for (CA and CT management).

Model Setup and Prediction of Hydrology
APEX version 1501 (Texas A&M AgriLife Research, Temple, Texas, USA) was used for the model setup. A user-defined watershed with 50 m 2 (each for CA and CT) and stream shapefiles were created to represent an experimental commercial vegetable home garden site. An ArcGIS interface, Arc-APEX (Texas A&M AgriLife Research, Temple, Texas, USA), was used to integrate user-defined vegetable garden, process GIS data layers, and input climatic data. Distinct farm-scale model setups (CA and CT) were carried out for each study site (Dangishita, Robit, Yemu, and Mkindo) which have different weather, topography, vegetation, and management activities. Also, the watershed-based model was established for Dangishita watershed depending on the location of streamflow gauging site. The first step in setting up the APEX model began with the processing of GIS data layers to

Model Setup and Prediction of Hydrology
APEX version 1501 (Texas A&M AgriLife Research, Temple, Texas, USA) was used for the model setup. A user-defined watershed with 50 m 2 (each for CA and CT) and stream shapefiles were created to represent an experimental commercial vegetable home garden site. An ArcGIS interface, Arc-APEX (Texas A&M AgriLife Research, Temple, Texas, USA), was used to integrate user-defined vegetable garden, process GIS data layers, and input climatic data. Distinct farm-scale model setups (CA and CT) were carried out for each study site (Dangishita, Robit, Yemu, and Mkindo) which have different weather, topography, vegetation, and management activities. Also, the watershed-based model was established for Dangishita watershed depending on the location of streamflow gauging site. The first step in setting up the APEX model began with the processing of GIS data layers to delineate the watershed boundary, subareas, and derive watershed characteristics. Watershed characteristics for Dangishita watershed and experimental plots were derived from the Digital Elevation Model (DEM). The second step was to integrate weather data using the Arc-APEX model interface. Moreover, the third step was to perform an initial model run and complete model setup procedures to create APEX model output files for further analysis. All monitored management activities ( Table 2) were integrated into the operation file. The "ADDMULCH" operation was recently included for APEX version 1501 (2017 release) in the fertilizer database to account for the impacts of adding organic mulch cover; input data required the application date and amount of mulch in kg ha −1 . A fixed irrigation application rate (drip irrigation) was provided in the management files. APEX model outputs were updated by re-running the model with modified management files.
The APEX hydrology model simulates all the key water balance components of the system. Precipitation, snow melts, and irrigation are the main inputs to the system, which are then disseminated into various components: surface runoff, subsurface/tile drainage flow, soil water, percolation, and evapotranspiration [46]. The routing phase of hydrology includes the water balance and nutrients. In APEX, the key landscape processes we considered across hydrologically connected units called subareas [45]. Subareas are the smallest unit in APEX with homogenous watershed characteristics, such as soil types, land use/crop cover, slope, and management.
APEX provides two options to estimate the runoff volume [46]: the modified soil conservation service (SCS) [47] curve number (CN) and Green and Ampt infiltration [48] methods. The SCS-CN runoff estimate method (Equations (1a) and (1b)) is a function of rainfall and retention parameter (Equation (2)). CN is a function of land use, management practice, and hydrologic soil group. The subsurface flow is a function of the vertical and horizontal flow and simulated as a simultaneous process [45]. The horizontal flow consists of a lateral flow, whereas the vertical flow (percolation) adds to groundwater storage, which is then subjected to return flow or deep percolation. The vertical component of percolation is calculated as a function of soil water content, field capacity, and travel time (Equation (3)). Five options are available to estimate the potential evapotranspiration: Penman, Penman-Monteith, Baier and Robertson, Priestly and Taylor, and Hargreaves methods [46]. The Hargreaves method is dynamic with a lower data requirement, and is a function of solar radiation, latent heat of vaporization, and temperature (Equation (4)). The methods for computing various hydrological components were selected based on the better simulation of variables before conducting a rigorous sensitivity calibration. The equations of each hydrological component and their descriptions can be found in Reference [39].
where Qi, Ri, ETi, Pi, S, CN, RMXi, HV, SW, FCi, TX, TMXi, TMNi, are runoff, precipitation, evapotranspiration, percolation, return flow, retention parameter, curve number (no unit), maximum daily solar radiation at mid-month (MJ/m 2 /d), latent heat of vaporization (MJ/kg), soil water, field capacity, average, and minimum and maximum temperature ( • C) on day i, respectively. All other unspecified units are in mm. X 1 , X 2 , and X 3 are travel functions of vertical, horizontal, and both vertical and horizontal travel time.

The "ADDMULCH" Subroutine
The new subroutine named "ADDMULCH" was developed and integrated in the APEX model to simulate the effect of adding organic mulch-cover on the soil surface. The subroutine adds the partial weight of the organic mulch (kg/ha) as carbon and nitrogen to the soil composition via two litter pools: metabolic and structural. Both pools receive equal amount of carbon (21% of the mulch weight) and nitrogen (0.42% of the mulch weight). Additionally, 3.7% of the weight as carbon and 0.35% of the weight as nitrogen are also added to the lignin component of the structural litter.

Sensitivity Analysis, Model Calibration, and Validation
Model sensitivity analysis is a method of identifying key parameters that affect model performance and are essential for model parametrization. The APEX model has large sets of parameters related to hydrology, sediment, nutrients, crops, and other environmental factors. Sensitivity analysis is the first step for hydrological models, which helps to diagnose and narrow down the large sets of parameters for calibration. Model calibration is a process in which model parameters are modified so that a model output mimics observed data, whereas validation is the use of modified parameters to simulate another set of observed data. The APEX auto-calibration and uncertainty estimator (APEX-CUTE) was used to perform sensitivity analysis and auto-calibration for the APEX hydrology model [40], followed by manual adjustment of a few parameters.
The first step was to examine the APEX hydrology model outputs for modifications. Some default methods and input parameters might need modification to get better simulation prior to sensitivity analysis and calibration [49]. Accordingly, the default values of some parameters from the APEX control, parameter, and subarea files were modified as shown in Table 3. APEX has various methods for linking CN and soil water (SW) [49]. The variable CN non-linear (CN/SW) with depth soil water weighting method (Non-Varying CN, NVCN = 0) was used in this study, as it can perform well in various situations [23,50]. Similarly, APEX provides many ways of estimating the field capacity/wilting point. The Rawls method is a dynamic method and is suggested for cropland modeling [45]. Thus, the Rawls method (Field Capacity/Wilting Point, ISW = 3) was used for this study. On the other hand, the Hargreaves potential evapotranspiration (PET) estimation method (Potential Evapotranspiration, IET = 4) was selected for this study among the five options available in the APEX model.   Table 2) [51]. Most of the parameters considered during calibration were related to soil properties. Annual values of hydrological components for nearby the watershed/station were obtained from the literature and used to adjust the APEX hydrology model for Yemu and Mkindo sites (no streamflow records were found close to the sites), following the suggestion from Reference [45], which explained the need for examining the APEX hydrology model based on literature when no or insufficient flow record exists. The third step includes evaluating the APEX model in simulating the plot level vegetable yields as a check-up for APEX hydrology components. Various crops were grown at each study site ( Table 2) for a period of 2 to 3 years to validate the APEX model predictions under CA and CT practices.

Model Performance Statistical Measures
The APEX model performance in predicting hydrology of the system was evaluated using commonly used statistical measures. Reference [52] listed and described the most commonly used statistical measures for APEX hydrology and crop modeling, which are: Nash-Sutcliffe efficiency (NSE), percent bias (PBIAS), and root mean squared error-observation standard deviation ratio (RSR). Also, Wang, et al. [53] used percent error (PE) to evaluate the performance of the APEX model. NSE (Equation (5)) is a normalized statistical measure that was proposed in Reference [54]. PBIAS (Equation (6)) measures the deviation of model prediction as an under-or overestimation from observation [55]. RSR (Equation (7)) is the normalized error index measure, which is used to evaluate hydrological components of the model [55]. Percent error (PE) (Equation (8)) is used to evaluate systematic over-or underprediction [53].
where Y oi and Y si are the i th observation and simulated value for the constituent being evaluated respectively; and Y m is the mean of the observed data, for the constituent being evaluated, and n is the total number of observations.

Model Performance Statistical Measures
After evaluating the model performance, the effects of CA on hydrology and crop growth/yields were evaluated per cropping season (planting to harvesting period). Evapotranspiration (ET), runoff (Q), percolation (PRK), and root zone soil water (RZSW) were the main hydrological components used to evaluate the effects of CA on hydrology (agricultural water management). The percent decrease in water loss through ET and Q, and percent increase in water saving through PRK and RZSW were computed for each vegetable and cropping season to determine the impacts of the CA management on agricultural water savings as compared to the CT practice.

APEX Model Sensitivity Analysis, Calibration, and Validation for Hydrology and Crop Yield
All relevant parameters for APEX hydrology components were included in the sensitivity analysis based on Reference [40]. The results of the sensitivity analysis in the Dangishita watershed depicted that streamflow was sensitive to the following parameters: Hargreaves potential evapotranspiration (PET) equation exponent (PARM-34), runoff CN residue adjustment parameter (PARM-15), runoff volume adjustment factor (PARM-92), runoff CN initial abstraction (PARM-20), soil water lower limit (PARM-5), and soil evaporation coefficient (PARM-12), in order of decreasing influence ( Table 4). The most sensitive parameters were associated with soil characteristics and climatic conditions. The parameter PARM-34 was found to be the most sensitive parameter for streamflow followed by PARM-15, possibly because ET was the second most-dominant hydrological process after rainfall. Reference [56] also found ET affecting the water yield of the catchment in their scenario analysis. Similarly, the sensitivity analysis results were found to be consistent with the findings of References [57]. Variable CN nonlinear CN/SW with depth soil water weighting method (NVCN = 0) was used, which is a function of soil water content and is directly linked with ET. Parameters PARM-92 and PARM-20 were found to be the third and fourth most sensitive parameters for streamflow prediction. Parameters APM and PARM-90 were less sensitive and thus not used for calibration.   Table 4. The APEX model water yield simulation showed very good agreement with the observed monthly streamflow (both calibration and validation) based on statistical performance measure ratings proposed in Reference [58] for a monthly time step; NSE > 0.75 (very good), RSR ≤ 0.5 (very good), and PBIAS ≤ ±10% (very good). Also, the determination coefficient showed very good performance rating (R 2 > 0.80) as based on Reference [59] for monthly streamflow comparison. The performance of the APEX hydrology model (monthly streamflow) for Dangishita watershed is shown in Table 5. The model slightly underestimated (Figures 6 and 7) when the rainy season started to cease (October to January), perhaps due to the gauging station conditions (some stagnant water was observed at the gauging station and could affect the rating curve development for this specific period).

Dangishita and Robit Plot Level Model Parameters
Dangishita experimental plots were located within and in the proximity of the Dangishita watershed outlet. Calibrated watershed hydrology parameters (Table 4) were transferred to Dangishita experimental plots. No streamflow record was available for the Robit study site. The Dangishita and Robit sites have a similar agro-ecological zone [51] and have the same soil type (Chromic Luvisols), which are categorized under hydrologic soil group C. Similarly, calibrated parameters from the Dangishita watershed were transferred to the Robit site as well and used for hydrological analysis. Thus, the transferred parameters were used to evaluate the impacts of CA with drip irrigation on the hydrology of the system.

Dangishita and Robit Plot Level Model Parameters
Dangishita experimental plots were located within and in the proximity of the Dangishita watershed outlet. Calibrated watershed hydrology parameters (Table 4) were transferred to Dangishita experimental plots. No streamflow record was available for the Robit study site. The Dangishita and Robit sites have a similar agro-ecological zone [51] and have the same soil type (Chromic Luvisols), which are categorized under hydrologic soil group C. Similarly, calibrated parameters from the Dangishita watershed were transferred to the Robit site as well and used for hydrological analysis. Thus, the transferred parameters were used to evaluate the impacts of CA with drip irrigation on the hydrology of the system.

Yemu and Mkindo Plot Level Model Parameters
Parameters for Yemu were calibrated against simulated data for the nearby Tamale site [60]. Accordingly, values of some of the water balance components were obtained from the literature of the region or areas close to the study site (see Table 6). The plot level model was calibrated based on annual water balance components. Calibrated parameters (Table 7) were used to evaluate the impacts of CA with drip irrigation on hydrology for Yemu and Mkindo sites. APEX simulation for annual ET, Q, and PRK were in good agreement for both Yemu and Mkindo with the values from the literature for the Yemu and Mkindo sites. As per Reference [61], the performance rating is considered very good when PE is less than 15%. The Hargreaves PET method and daily CN nonlinear CN/SW with depth soil weighting were used and were found robust to simulate ET and Q, respectively, for the study sites.

Model Validation for Crop Yield
The APEX model was evaluated in predicting crop yield across the sites (Dangishita, Robit, Yemu, and Mkindo) for various vegetables ( Table 2). The model successfully simulated vegetable yields for various climatic, soil, and environmental conditions under CA (with the "ADDMULCH" subroutine) and CT management. Overprediction occurred as often as under prediction for both CT and CA practices; however, model efficiency measures [45] were found to be very good for both calibration and validation across the study sites (Table 8).

Impact of CA on Hydrology and Water Management
The impacts of CA on hydrology were analyzed at a plot level for all sites (Table 9). A one-tailed paired t-test was conducted to examine the significance of CA in improving agricultural water management. Significant (α = 0.05) reductions were observed for water loss through ET, P(T <= t) = 0.007, and Q, P(T <= t) = 0.027, under CA across the sites for different cropping seasons and vegetables. ET was decreased in CA: 44-49%, 28-44%, 1-9%, and 1-11% for various vegetables and cropping seasons grown at Dangishita, Robit, Yemu, and Mkindo sites, respectively. Likewise, Q was reduced for CA: 17-54%, 34-62%, 2-12%, and 20% for different vegetables and cropping seasons at Dangishita, Robit, Yemu, and Mkindo sites, respectively. On the other hand, water saving was significantly (α = 0.05) increased under CA across the sites through increased PRK, P(T <= t) = 0.007, and RZSW, P(T <= t) = 0.0001, for various vegetables for different cropping seasons. PRK increased substantially under CA: 173-231%, 52-312%, 2-21%, and 25-91% in Dangishita, Robit, Yemu, and Mkindo sites, respectively. PRK in the study sites was mainly because of rainfall. Similarly, the average RZSW was increased under CA: 13-18%, 4-25%, 5-40%, and 7-21% for different vegetables at Dangishita, Robit, Yemu, and Mkindo sites, respectively. Farmers in Dangishita and Robit adopted different irrigation use for CA and CT plots based on their intrinsic knowledge of vegetable water need and moisture content of the soil. Meanwhile, farmers in Yemu and Mkindo applied an equal amount of irrigation water to both management system (the drip kits installed for the sites did not have separate switch control), and thus comparison was not made between CA and CT regarding irrigation use. Irrigation water use in Ethiopia (IRGA) was significantly (α = 0.05) reduced under CA, P (T <= t) = 0.0001, for various vegetables (14-35% reduction); however, the model result indicated that CA could not remove the need for irrigation in the region. The amount of rainfall (RF) during the growing period is shown in Table 9 (the rainfall in Dangishita and Robit sites was either at the beginning or at the end of the cropping period). The degree of CA impact varies depending on several factors including weather condition, water input, crop type, soil characteristics, and type and thickness of mulch.

Discussion
CA was found to significantly reduce water loss via evapotranspiration and surface runoff, and thus improved water saving through increased soil moisture and percolation. As a result, less irrigation needs were observed (14-35% reduced) for various vegetables under CA management ( Table 9). As a result of soil disturbance (no till) and continuous soil cover using organic mulch, soil quality and soil structure must have been improved due to the fundamental principles of the CA practice. When an organic mulch material gets decomposed, it adds organic matter to the soil, which invites microorganisms (worms, bacteria, fungi, etc.). Microorganisms digest the organic matter and produce glue material that helps to stick small soil particles together and form aggregates; as such, soil pore space gets increased. Similarly, mulch cover avoids soil crusting, i.e., the breakdown and movement of small soil particles to the soil pore from the energy of dropping water, which creates a thin layer on the soil surface. Soil crusting greatly reduces infiltration. Thus, the soil structure, water holding capacity, and drainage was improved under CA resulting in higher water savings through higher percolation and soil moisture compared to CT ( Table 9). The model considered the effects of mulch application on water management using "ADDMULCH" subroutine. This option changed the soil albedo and soil cover factor, which directly affected the computation of evapotranspiration, which further affected other hydrological variables. The no-till practice was captured in the model by providing zero-tillage depth in the tillage database of the model. Also, the rotation of vegetable production was integrated into the model through a management database.
CA reduced water loss through evaporation from the soil by providing shields from the sun and reducing soil heat, decreasing ET (Table 9). Reference [62] reported higher ET under CT in China when compared with conservation practice. The decrease in ET helps the soil to keep its moisture and consequently reduce irrigation water needs as seen in the Dangishita and Robit sites. Similarly, Reference [52] found increased infiltration and soil moisture under CA in Zimbabwe. Mulch application reduces water loss through reduction in runoff as it provides an obstacle on the soil surface and slows the movement of water, which provides extra time for the water to infiltrate into the soil [54,63,64]. In general, the degree of reduction in evapotranspiration and surface runoff varies depending on vegetable types, cropping seasons, water supply (precipitation and irrigation), weather conditions for the cropping periods, and other site-specific conditions. Furthermore, CA was found to increase water percolation significantly. Reference [65] reported a similar observation of significant increase in infiltration under CA in Australia.
The degree of improvements in agricultural water management was relatively less in Yemu and Mkindo when compared to the Dangishita and Robit sites. One reason could be due to the soil condition since the soil in Yemu is hydrologic group A, whereas the soil in Mkindo is hydrologic soil group D; the other two sites are hydrologic soil group C (Table 1). Hydrologic soil group A soils have good soil structure and drainage, which is evidenced by the low runoff and high percolation even under CT (Table 9). Hydrologic soil group D soils have poor soil structure and conditions, which are characterized by a very low infiltration rate and high runoff (Table 9). Therefore, the rate of improvement in soil structure for hydrologic soil groups A and D is expected to be relatively less since the soils were in very good and poor conditions, respectively, as compared to soil hydrologic group C. Another reason could be related to the vegetable types. For instance, farmers in Yemu grew sweet potato during the first cropping season in 2016, which has an extensive root system with higher water demand. Though soil evaporation was less under CA due to mulch cover, transpiration could have a significant contribution to ET due to sufficient water availability and the extensive root system of sweet potato. In addition, the amount of water input could be another reason for the relatively lower rate of water savings in Yemu and Mkindo. Reference [66] found a higher positive effect on CA with lower water input. Farmers in Dangishita and Robit applied less irrigation water to CA plots as compared to CT plots, whereas, farmers in Yemu and Mkindo applied the same amount of water for both CA and CT plots.

Conclusions
The impacts of CA on agricultural water management was analyzed through the integration of a field experiment and a biophysical model, APEX, at four study sites (Dangishita, Robit, Yemu, and Mkindo) in SSA. The APEX model was validated with a reasonable model performance using field-scale measurements (stream flow and crop yield) and the established literature. Once the APEX model was validated for hydrology and crop yield, the impacts of CA practices on water management was evaluated and compared with CT practice. Evapotranspiration, surface runoff, irrigation water, soil moisture in the root zone, and percolation below the root zone were compared between CA and CT systems under drip irrigation. Both evapotranspiration and surface runoff were found to decrease significantly (α = 0.05) under CA by up to 49% and 62%, respectively. Also, irrigation water use based on farmers' practice was decreased significantly (α = 0.05) at the Dangishita and Robit sites for various seasons and vegetables (14-35%). In contrast, percolation and soil moisture were increased significantly (α = 0.05) under CA (up to 231% and 28%), respectively, as compared to CT. The increase in water savings was high for the Dangishita, Robit, and Mkindo sites compared to Yemu depending on soil conditions, selected vegetable types, and cropping periods. The results depicted significant improvement in soil structure and water-holding capacity across the sites. CA with drip irrigation was found to be a promising approach to improve water and soil management, and as a result, improve food production. It is essential and recommended to (1) provide thick mulch cover on the soil surface, and (2) introduce an irrigation scheduling approach based on actual evapotranspiration loss for the substantial improvement in soil quality and irrigation water management.
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 the scientific content. M.R. contributed to the conceptual design, data acquisition, and analysis. A.W.W. contributed to the conceptual design, data acquisition, data analysis, and revising the manuscript for the scientific content.