Innovatory rainfall simulator design – A concept of moving storm automation

We developed an advanced design programmable rainfall simulator (RS) to simulate a moving storm rainfall condition. The RS consists of an automated nozzle control system coupled with a pressure regulator mechanism for an operating range of 50 kPa to 180 kPa at a drop height of 2000 mm above the soil flume surface. Additionally, a programmable mobile application was developed to regulate all RS valves. Near natural rainfall conditions were simulated at varying spatial and temporal resolutions in a controlled environment. A soil flume of 2500 mm × 1400 mm × 500 mm was fabricated to conduct 5 different hydrological experiments. The flume was designed to record overland, subsurface, and base flows simultaneously. This study focused on a detailed analysis of moving storms and their impact on hydrograph characteristics. Experimental results showed a considerable difference in terms of time to peak (tp), peak discharge (Qp), and hydrograph recession for two different storm movement directions (upstream and downstream). Two multiple regression models indicate a statistically significant relationship between the dependent variable (tp or Qp) and the independent variables (i.e. storm movement direction, 10 storm velocity, and bed slope gradient) at a 5% level of significance. Further, the impact of these moving storm phenomena reduces with the increase in the storm movement velocity.

5%); and (b) Three different storm movement velocities (2 m min −1 , 3 m min −1 and 6 m min −1 ). A multiple regression model was used to test the statistical significance of the relationship between storm direction and the hydrograph characteristics.

60
This section provides a detailed description of the soil flume and moving storm design along with the circuit diagram of the Bluetooth module.

Structural design
The rainfall simulator (RS) used in this study was designed at the Department of Hydrology, IIT Roorkee, India. The schematic diagram of RS is presented in Figure 1. The instrument consists of a 3 m x 2 m frame connected with a pipe attached to a header 65 (supporting 11 nozzles) and a pressure gauge. The frame was supported by four telescopic legs of 6 m each ( Figure 1). The rainfall regulating structure connects to a centrifugal pump capable for controlling the water pressure and lifting the water from a feeder tank. The main components of the simulator are the frame, header for nozzle mounting, nozzles, and pumping station.
A feeder tank was located near the RS to maintain a sufficient water supply. Water pressure in the system was adjusted by a pressure regulator, and a "shut-off" valve was used to apply back pressure at the outflow end of the simulator system. Another 70 valve was used to facilitate the accurate control of water pressure to the nozzles. Six full-cone nozzles manufactured by Spraying Systems Co. were used to simulate low to high-intensity rainfall. These nozzles produced a solid cone-shaped spray pattern with a circular impact area (Figure 1). A uniform spray coverage and distribution over a wide range of flow rates and pressure is possible using these nozzles (B1/88G-SS4.4W). The nozzles also had removable caps and vanes for easy inspection and cleaning. A transparent acrylate wall at one vertical side of the soil flume facilitated easy visual observation. A base frame of 500 mm height was designed for stability and to support the jack system. A manually operated worm wheel gear jack setup was installed to change the slope of the flume (0 % -7.5 %). The flume has three sub-partitions to accommodate three different soil types 80 at a single simulation. Outlets for surface flow, sub-surface flow, and base flow gauging were provided at the downstream end of the flume. The surface and the subsurface flow outlets were placed at the height of 500 mm and 250 mm from the bottom of the flume, respectively. The outlet for the baseflow measurement was located at the bottom edge of the flume. Additionally, ten release/ seep slots (5 mm each) were provided at each sub-partition to analyze the change in the piezometric head. These slots can also be used for leachate studies.

85
The soil flume was filled with gravel up to 50 mm depth to prevent the washout of the soil. Above gravel bed, sand was added to a depth of 25 mm, and the remaining 425 mm flume space was filled with sandy loam soil ( Figure 2b). The sand, silt, and clay composition of the soil used was 66 %, 29 %, and 5 %, respectively, measured using a mechanical sieve analyzer.

Moving storm design
Detailed descriptions of components used to generate moving storm conditions are presented in Table 1 . A set of 11 nozzles 90 were used for simulating the moving storm condition. Electrically operated flow control valves were used to control these nozzles through an Arduino Mega (AM) microcontroller board to simultaneously regulate the flow through nozzles. A nozzle control system was incorporated using three components: servo-operated valve, AM microcontroller, and Bluetooth module (BM) (Appendix A). This system serves two purposes; communication with the user interface in the handled device and control of the opening and closing of nozzles. The detailed operational flowchart of the moving storm system is shown in Figure 3.

95
The nozzles were grouped into four clusters for this experiment to ease the flow regulation operation (Figure 4). Three clusters (NC1, NC2, and NC3) consisted of three nozzles, and one cluster (NC4) consisted of two nozzles. These clusters were activated and deactivated with a specific time gap to simulate the moving storm over the plot area. If needed, full control can be given to each nozzle to regulate individually. To obtain variable rainfall intensities, a servo motor operated flow control valve was inserted into the pipe openings just before the nozzle. Further, an android mobile application was developed for regulating the 100 valves through the BM. The pressure regulating system (PRS) (Appendix B) comprises of a motorized globe valve, pressure transmitter, and Proportionalintegral-derivative controller. The PRS was designed to maintain constant pressure throughout the simulation to achieve a constant rainfall intensity regardless of the opening and closing of the various number of nozzles.

Circuit design 105
The circuit design comprised of AM, BM, and 11 servo motor-operated valves. The detailed design of the circuit diagram is given in Figure 5. The four connections, R x , T x , V cc , and G nd in the BM were connected to pin 11, pin 10, 3.3 V, and G nd recipient pins in the AM, respectively. The 11 servos were clustered into four sets of 3, 3, 3, and 2 each in every cluster. The signal pins of these clusters of servo motors were connected to the digital signal pins of AM, numbered as pin 3, pin 5, pin 6 and pin 9. Each group had a power supply of 5 V -2 A, and these were grounded to AM, which had a power supply of 12 V -110 1 A.
The AM was coded (Appendix C) so that the servo motor could be regulated to control all the motors simultaneously, with any android based phone through a Bluetooth application. Two basic libraries used in this code were "SoftwareSerial.h' and "Servo.h." The software used to write the code is Arduino Editor, free coding software available online (https://auth.arduino.cc).
The android OS application was designed to operate all of the 11 nozzles simultaneously, with one-touch Bluetooth connectivity 115 and four slider bars to control four groups of servos operated valves at any value ranging from 0 to 100 percent. This application was developed using the "MIT app developer" software.

Simulation Uniformity Assessment
The following equation is used to calculate the Christiansen Uniformity Coefficient (UC) where µ is the average of all the measurements, |X i −µ| is the sum of the individual deviations from the mean, and n is the number of measurements. The UC was measured using 66 beakers kept in a square array, 250 mm apart, beneath the simulator, covering the plot area of 2500 mm*1440 mm.

Design of experimentation
The experiments consisted of 12 scenarios (Table 2 ) with three replications. Experiments were performed under fully saturated 125 soil bed conditions to reduce the variability among scenarios and replications.
The simulation results were analyzed using a multiple regression model considering one categorical variable (storm direction) and two numerical variables (velocity and bed slope) as independent variables and, time to peak (t p ) or peak discharge (Q p ) as the dependent variable. An indicator variable was used to include the direction of the storm in the multiple regression model (downstream = 0, upstream=1). An indicator variable allows interpretation of the regression coefficients for storm direc-130 tion as an additive effect on the hydrograph characteristics. By choosing the indicator variable for an upstream storm as 1, the regression coefficient of storm direction indicated the difference in the mean response of time to peak (t p ) or peak discharge (Q p ) of the upstream directional storm. A positive coefficient implies a positive effect upstream strom direction storm on the hydrograph characteristics compared to downstream directional storm. The null hypothesis of regression coefficient for each of the independent variables as equal to zero was tested against alternative hypothesis of significant effect of the independent 135 variable on storm hydrograph characteristics at 5% level of significance.

Results and discussion
After completing the design of the moving storm rainfall simulator (RS) (Figure 6), we checked the feasibility of the RS for generating the moving storm events. Before stepping into the different rainfall scenarios analysis, the rainfall distribution over the plot was analyzed (Figure 7). It can be elucidated from the rainfall distribution graph that 70% of the plot area receives a 140 uniform amount of rainfall, i.e., 30 mm to 36 mm. The lower rainfall (15 mm to 22 mm) amount was recorded at the plot edges.
However, the overall UC was found to be 84.2 %. de Lima and Singh (2003) also conducted their rainfall simulator experiment with an average UC of 88 %.

Results of experimentation conducted at 2.5% slope
The results recorded by the rainfall simulator (RS) of the moving storm clearly exhibits the effect of storm direction, velocity, 145 and slope on the overland flow hydrographs. For example, the hydrograph generated using the both downstream and upstream direction of storm movement with the velocity of 2 m min −1 and 3 m min −1 at a slope of 2.5% is shown in Figure 8. However, very little runoff was generated for velocity of 6 m min −1 , thus not included in the result.
The time to peak (t p ) of the hydrographs generated by the upstream to the downstream storm was less than the downstream to the upstream storm irrespective of their velocities (Figure 8). When the storm moves towards the outlet (i.e., upstream to

Results of experimentation conducted at 5% slope
At 5% slope condition, rainfall simulation experiments were performed with three different velocities 2 m min −1 , 3 m min −1 , and 6 m min −1 (Figure 9). The storm movement directions were the same as the previous experiments, i.e., upstream and downstream. It can be clearly illustrated from Figure 9 that the recession characteristics, time to peak (t p ), and peak discharge 165 (Q p ) followed the same trend as the hydrographs generated at 2.5% slope. An interesting observation was noticed during the testing of 6 m min −1 storm velocity, i.e., the recession curve and Q p of both hydrographs completely matched with each other during the upstream and downstream directional storm movement (Figure 9c). Only t p varied slightly in these two hydrographs of 6m min −1 velocity storm.
A detailed description of storm characteristics during different test scenarios is presented in Table 3 . It is observed that as 170 the slope increased, the t p value decreased when the storm was moving to the upstream direction. While moving towards the downstream, the Q p value of 5% slope was 53.7 % and 43.3 % higher than 2.5% slope condition at a velocity of 2 m min −1 and 3 m min −1 , respectively. Similarly, for upstream directional storm at a velocity of 2 m min −1 and 3 m min −1 , the Q p of 5% slope was 59% and 42.8% higher than the 2.5% bed slope condition, respectively. These analyses concluded that the Q p value increases significantly with an increase in soil flume slope and decreases with increased storm velocity.

175
Further, observations show that the discharge volume followed the same relationship pattern as Q p with slope and storm velocity. The discharge volume of the 5% slope condition was 67.38% and 82 % higher than the 2.5% slope for the downstream and upstream directional storm, respectively. Similarly, for 3m min −1 velocity, the increase in discharge volume due to slope increment was 58.56% and 55.56% for downstream and upstream storm directions, respectively.
From the above sets of experiments, it can be concluded that rainfall moving upstream results in; slower rise time, lower 180 Q p ; and longer recession time (base time) compared to rainfall moving downstream. de Lima and Singh (2003) found that the high-velocity storm results in a smaller volume of runoff. This phenomenon was also observed in our study for both the slope conditions. For the 2.5 % slope condition, storm velocity of 6 m min −1 barely generated any runoff at the plot outlet; thus, it did not create any runoff hydrographs. Moreover, de Lima and Singh (2003) observed that high storm velocity did not result in significant changes in hydrographs during upstream and downstream storm movement because of the surface tension force.  Similar observations are also noted in the present study during 6 m min -1 storm velocity (Figure 9c). The storm movement was so fast that the whole plot could never generate runoff at the outlet at an instant. The difference between Q p of the upstream and downstream directional storm is the function of storm velocity. Further, Lima et al. (2011) discussed the influence of slope on moving storm runoff hydrograph, i.e., how the runoff volume increases with an increase in slope angle. The effect of storm direction on Q p is also discussed by de Lima and Singh (2003), Isidoro et al. (2012) and Seo and Schmidt (2013). They concluded that the downstream storm produces higher Qp than the upstream directional storm. Similar results were also obtained in the current study (Table 3 ).
To further test the significance of the effect of the experimental variables on the observed differences in time to peak (t p ) and peak discharge (Q p ), multiple regression analysis was performed. Regression analysis with an indicator variable for storm direction results in two regression model equations for each hydrograph characteristic modelled (Eq. (2)).

195
T (2) The model fit details for the regression models are shown in Table 4 . Interpreting the value of β1 in the models, for a storm moving upstream, t p is higher (positive) by 10 ± 4.776 sec and Q p is lower (negative) by 0.00256 ± 0.00086 l sec −1 .
These values are significant at 5% level of significance thus provides a satisfactory hydrological verification for the moving storm rainfall simulator. The multiple regression models for t p and Q p explain 89% and 94% of the variability in t p and Q p  and an LED to display its operational activity [22]. In this experiment, the Bluetooth module was used as a slave set (default configuration). modulation of 4 -20 mA actuator was used. It had an intelligent circuit that sensed hindrance in valve movements. An AC sensor was used for circuit protection that can shut down the valve during an overload condition. It had a robust and compact design to ease the installation.

255
A Mass PT11 pressure transmitter has a very compact design with stainless steel construction. It is highly stable against shock and vibration and also have features such as reverse polarity, limit protection and have high accuracy. This pressure sensor was installed to check the main line pressure. PID controller can sense the change in pressure and can act accordingly to operate bypass and to maintain a constant pressure in the main line for a uniform rainfall intensity.
B3 Proportional-integral-derivative (PID) controller 260 Selec PID500 is a controller which is employed widely in the industrial process controls. PID controller is a control loop feedback system and is used to operate a motorized globe valve on the basis of an input signal from a pressure sensor. Whenever there is a pressure offset from the set value, it sends a signal to the motorized valve to re-attain the set value. The I/O signal from the pressure sensor and PID respectively ranged between 4-20 mA. The controller had a compact square housing with panel mounting facility in its enclosure, powered by a 240 V AC supply [23]. PID controller was used to control the bypass 265 flow by operating a motorised valve to maintain constant pressure in the main line against any pressure drop generated due to the moving storm simulation Figure B1. Components used for pressure regulating system.
Author contributions. The main contributions from each co-author are as follows. RKM contributed to the methodology, software, validation, and draft preparation. SS contributed to the conceptualization, supervision, and funding acquisition. AN and BD contributed to the visualization, writing, and editing of the paper. AM contributed to the review of paper.