Adaptability of rainfall simulators as a research tool on urban sealed surfaces – a review

ABSTRACT

Rainfall simulators can be a useful research tool for some purposes but are quite unsuitable for others. They have been useful in soil erosion and infiltration studies for over eight decades, but the possibility of using a rainfall simulator in urban nonpoint source pollution involving urban non-erodible surfaces has not been fully explored. In this review, the versatility of different rainfall simulators of varying sizes, configurations and styles used in the past two decades are appraised for possible adaptation to urban sealed surfaces. Recommended criteria for detailed rainfall simulator reporting are also outlined.

EDITOR M.C. Acreman

ASSOCIATE EDITOR not assigned

KEYWORDS:

• intensity
• kinetic energy
• non-erodible surface
• plot size
• rainfall simulator
• raindrop diameter

Introduction

Rainfall simulators (RSs) are specially made devices designed to simulate the natural characteristics of rainfall as closely as possible. Comparison of RSs is a complex subject, because of their diversity in application and purpose. RS technology has evolved from water sprinkler cans, stationary nozzles, irrigation sprinklers, capillary tubes, rotating discs and booms to the more recent oscillating nozzles (Norton and Savabi 2010). The first spray-type RS was designed and reported by Meyer and McCune (1958) using the pressurized nozzles recommended by Meyer (1958). One notable shortcoming of this RS was the production of an excessive rain volume in a short period of time. The past four decades have witnessed the development of a variety of RSs to address the shortcomings observed in the earlier models (e.g. Grace and Eagleson 1966, Hall and Wolf 1967, Grierson and Oades 1977, Pall et al. 1983, Guevara-Escobar et al. 2007, Resso et al. 2007, Scherrer et al. 2007). For example, Neibling et al. (1981) developed an automated RS that took into account the temporal and spatial variability of the rainfall, reduced water application cycles and faster data collection. Since then, other improved forms of RSs have been used successfully in modelling temporal variations of rainfall.

For over eight decades, RSs have been a useful research tool in the study of soil loss from rill and inter-rill erosion in the agricultural field (Agassi and Bradford 1999, Feng et al. 2013, Salem et al. 2014). The development of the Universal Soil Loss Equation (USLE) and Water Erosion and Prediction Project (WEPP) models all utilized RSs (Grismer 2012).

Rainfall has been recognized as the most important player in the hydrology of nonpoint source (NPS) pollution, and the uncertainty associated with its natural occurrence has been seen as a major challenge in conceptualizing urban wet weather models (Shen et al. 2012). Rainfall simulators have been used in the study and modelling of NPS pollution from urban sealed surfaces in recent times (Herngren et al. 2010, Miguntanna et al. 2013, Zhao and Li 2013). Gilbert and Clausen (2006) found that non-erodible surfaces respond to higher generation of runoff and contribute higher pollution loading to streams than erodible surfaces. It was also well documented by Leecaster et al. (2002) and Tiefenthaler and Schiff (2003) that non-erodible surfaces have direct correlation with increases of NPS pollution generation; and highly urbanized area are more susceptible to NPS than less developed areas due to anthropogenic activities (Bronstert 2004, Stein et al. 2006). Therefore, by the virtue of their differences in responding to rainfall, transportation of pollutants and constraints of space, erodible and the non-erodible surfaces naturally require different criteria for rainfall simulation.

Some researchers have outlined general design criteria for all RSs (e.g. Bowyer-Bower and Burt 1989, Iserloh et al. 2012) and their evaluation of cost based on utility and labour has been reviewed by Kainz et al. (1992). Simulator publications can be divided into three categories:

1. technical design and construction protected by copyright;

2. calibration reports; and

3. practical applications.

The first involves innovation, which is normally documented in the form of protected copyrights (Coody and Lawrence 1994, Rodriguez and Rodriguez 2005, Ahn et al. 2013). The second gives details on the RS performance and its rainfall characteristics (Tossell et al. 1987, Humphry et al. 2002, Ries et al. 2009). The third is aimed at establishing the effect of the rain characteristics on the behaviour of a certain phenomenon (Tayfur and Kavvas 1998, Sharpley and Kleinman 2003, Moriwaki et al. 2004, Badia and Marti 2008).

The most important aspects of studies using RSs are twofold: the method of simulating the rainfall and the measurement of the runoff from the catchment. The former involves careful selection of the RS components and a configuration that will ensure spatial and temporal control of application rates, reproduction of natural rainfall characteristics as closely as possible, and reproduction of kinetic energies of drops as in natural rainfall. The latter involves instant measurement of runoff rates from the catchments to a reasonably comparable measure and comparison of the extent to which rainfall intensities could be produced during a given time interval.

Researchers are often faced with the need to make a compromise in the selection criteria between various requirements of RS because the need to satisfy one normally conflicts with another. Kuhn et al. (2014) emphasized specific compromises that are often needed in order to overcome the many constraints when undertaking field experiments using RS. The objective of this review is to examine the design, configurations and simulated rain characteristics of the RSs used in the past two decades and determine how the RSs could best be suited for event investigations on urban non-erodible surfaces. To achieve this objective, the published literature considered were selected based on their comprehensive calibration reports and were judged to represent various types of RSs used in the past two decades.

Discussion

Rainfall simulator classifications

Rainfall simulators can be upward (Bryan and Poesen 1989, Assouline et al. 1997, Sporre and Lanyon 2004) or downward (Navas et al. 1990, Regmi and Thompson 2000, de Lima and Singh 2003) in terms of their spray configuration; small (Munn and Huntington 1976, Cerdà et al. 1997) or large (Esteves et al. 2000, Sharpley and Kleinman 2003, Adams and Elliott 2006) based on their size; and simple (Wilcox et al. 1986, Miller 1987, Humphry et al. 2002) or complex (Lascano et al. 1997, Keim et al. 2006, Norton and Savabi 2010, Pérez-Latorre et al. 2010) based on their operation.

However, RSs can be classified broadly into two larger groups: drop formers (DFs) and pressurized nozzle (PN) simulators. The variability of RSs in the replication of natural rainfall has been acknowledged (Bowyer-Bower and Burt 1989). The DFs, as the name implies, drip water from specially made equipment consisting mostly of drilled holes or needles of a known gauge under gravitational water pressure. The drilled holes or needles determine the simulated raindrop diameter, which in combination with their spacing and the height of the plot, determines the intensity and the kinetic energy (KE) (Hignett et al. 1995, Dimoyiannis et al. 2001). In the case of PN simulators, water is pumped to nozzle(s) under high pressure. The pressure, nozzle type and arrangements determine the intensity, the raindrop diameter and the KE (de Lima and Singh 2003, Pérez-Latorre et al. 2010, Corona et al. 2013). Grismer (2012) reported that 80% of all RSs in use are the PN type. These simulators suffer from production of excessive intensity and KE (Horton 1941, Hammad et al. 2006).

To mediate the high KE in a PN simulator, Assouline et al. (1997), Esteves et al. (2000), Kato et al. (2009) and Abudi et al. (2012) jetted the water upwards, while Blanquies et al. (2003), Vahabi and Nikkami (2008), Kim et al. (2010), Sangüesa et al. (2010), Aksoy et al. (2012) and Dunkerley (2012) used an oscillating motor to avoid the generation of large volumes of water within a short period of time.

Due to the diverse advantages of the DFs and PN simulators in replicating natural rainfall characteristics, Wildhaber et al. (2012) developed a hybrid RS for field soil erosion studies. These investigators placed a screen beneath the nozzle but found that it neither improved the KE nor enhanced the drop size distribution (DSD) of the simulated rain. Egodawatta (2007) also used a similar method to reduce the excessive KE but found that the technique reduced the KE at the expense of rain uniformity.

Dynamics of simulated rain characteristics

Irrespective of the way the RS simulates rain, the most important evaluator is the simulated rain characteristics in relation to the physical characteristics of natural rainfall (Hall and Wolf 1967, Pérez-Latorre et al. 2010). Natural raindrop diameters generally range between 0.5 mm and 5 mm (Kincaid et al. 1996, Lal 1998, Jayawardena and Rezaur 2000, Yakubu et al. 2016) and attain a maximum threshold of 7–10 mm (Hudson 1993, Pruppacher and Klett 1997, Campos 1999, Blanquies et al. 2003, Brodie and Rosewell 2007, Lynch and Lommatsch 2011), beyond which they become hydrodynamically unstable and the raindrops tend to split into smaller drops (Boxel 1997, Campos 1999). Van Dijk et al. (2002) reviewed the existing literature on the intensity–KE relationship and they found that the KE of natural rainfall ranges from an average lower value of 26.4 J m⁻² mm⁻¹ to an average uppermost value of 28.3 J m⁻² mm⁻¹. They reported the highest recorded KE of 35.9 J m⁻² mm⁻¹ for southern Portugal and the lowest value of 24.6 J m⁻² mm⁻¹ in North Carolina, USA.

Simulated rainfall can be representative of natural rainfall (Salem et al. 2014), especially at smaller raindrop sizes (Assouline et al. 1997). Artificial rainfall can attain up to 94% average terminal velocity (Mckenzie et al. 2002, Júnior and Siqueira 2011, Salem et al. 2014), more than 90% of the uniformity of rain (Miller 1987, Herngren et al. 2005a), and a maximum of 10% of the variation in KE (Assouline et al. 1997, Alves Sobrinho et al. 2008, Júnior and Siqueira 2011). Brodie and Dunn (2010) investigated the commonality of rainfall variables in the suspended solids washoff process from impervious surfaces. They concluded that the total suspended solids (TSS) runoff and event mean concentration (EMC) achieved using simulated rainfall bear a strong correlation with the TSS and EMC of natural rainfall events.

Rainfall characteristics and plot sizes are the two most important criteria for RS evaluation (Hall et al. 1989, Corona et al. 2013). The choice of RS concept and components should ensure that selection of the correct operating pressure will produce the highest rain uniformity, reasonable drop diameter and distribution of the rain at an intensity that will produce a KE comparable with that of natural rainfall (Hall et al. 1989). These requirements appear to be interwoven and may, in large part, inherently conflict with each other (Hall and Wolf 1967, Pérez-Latorre et al. 2010). Therefore, achieving the optimal balance of all requirements may not be feasible without some compromise (Hall et al. 1989).

Plot size

Plot size refers to a pre-defined confinement upon which parameters are isolated for the purpose of investigation using simulated rain. The size of the plot plays a significant role in determining the uniformity of the simulated rain and is one of the most important factors that determine the size of the RS. Reported plot size varies from as small as 0.24 m² (Cerdà et al. 1997) to as large as 99 m² (Moore et al. 1983). Researchers have seldom related the spatial area covered by their RSs and the plot sizes upon which simulation took place. This lack of information often makes comparison of RSs extremely difficult. Some researchers reported the plot size used but only limited details on the corresponding rain characteristics (e.g. Kim et al. 2010, Sangüesa et al. 2010, Delaune and Moore 2013), while some clearly distinguished between the size of their RS and the plot size used (e.g. Cerdà et al. 1997, Wildhaber et al. 2012).

The results of Sangüesa et al. (2010), as presented in Table 1, offer an opportunity to understand the influence of chosen plot area and the nozzle arrangements on uniformity. Sangüesa et al. (2010) used one nozzle capable of wetting an area of 10 m² on a 4 m² (2 m × 2 m) plot and achieved a uniformity of 86%. Decreasing this area to a quarter (1 m × 1 m), the uniformity increased to 91.4%. When four nozzles were arranged in straight line, a metre apart, on a 1.25 m × 3.25 m plot, this arrangement attained a uniformity of 90%. The results they obtained when four nozzles were used on a 2.5 m² plot are comparable with the results obtained when a single nozzle was used, indicating the influence of plot size on increased uniformity. To further elucidate the effect of plot size, the 18 plots in Table 1 were categorized as small, medium or large, as presented in Table 2. The results indicate a consistent decrease of uniformity with increasing plot size. A plot size of 2 m² or less ensures more than 90% uniformity of rain. A plot size of more than 4 m² is categorized as large and, on average, has a spatial consistency of 85%, while the spatial uniformity of a medium plot size is an average of 87%.

Table 1. Characteristics and performance of different RS systems.

Reference	Nozzle type	Nozzle spacing (m)	RS size(m^2)	Plot size(m^2)	Pressure (kPa)	Fall height (m)	D50 (mm)	KE(J m^−2 mm^−1)	Intensity (mm h^−1)	Cu (%)	Remarks
Hignett et al. (1995)	DF	NA	1.60 (1.14 × 1.4)	1.00 (two 0.5 × 1)	NA	Varied: 0.17–2.54	2.70 and 5.10	Varied: 1.6–16.6 (for 2.7 mm D), 1.6–19.9 (for 5.1 mm D)	Varied: 40–100	>95	DF, for soil erosion study, consists of oscillating hypodermic needles (23 gauge), spaced 25 mm both ways (BW). Adjustable height.
Dunkerley (2012)	DF	NA	0.25	1	NA	2.2	2.36	_	10–30	_	DF, for soil infiltration study, consists of 361 oscillating drippers spaced 25 mm BW. Fixed height.
Vahabi and Nikkami (2008)	DF	NA	1.1 (0.89 × 1.2)	1.2 (0.95 × 1.25)	NA	1.5	3.6	_	24.5 and 32.0	_	DF, for soil erosion study, consists of 216 oscillating formers of 0.5 mm diameter, spaced 70 mm BW
Blanquies et al. (2003)	4 flood jet nozzles	0.99	_	3.56 (1 × 3.56)	47.6	2.4	1.7	_	_	>90	Downward PN, using Floodjet 3/8 K SS45 for soil erosion study.
Kim et al. (2010)	4 nozzles (#8005 with screen)	_	_	6.00 (2 × 3)	207	_	_	_	60	_	PN backpack sprayer with four nozzles (#8005 with screen) on a 1.5 m oscillating boom. For surface runoff study.
Aksoy et al. (2012)	4-VeeJet 8030	1.45 and 1.25	_	8.84 (1.36 × 6.50)	40	2.43	2.19	25.9	45	82.1	Downward PN, for soil erosion studies using four different types of oscillating VeeJet nozzles spaced 1.45 m for 45 and 65 mm/h; 1.25 m for 85 and 105 mm/h.
	4-VeeJet 8050				42		2.77	22.3	65	86
	5-VeeJet 8060				33		3.13	20.7	85	83.6
	5-VeeJet 8070				48		2.9	21	105	88.8
Herngren et al. (2004)	3-VeeJet 80100	1	6.24	3 (2 × 1.5)	41	2.4	2.1	25.4	Varied: 65–133	95	Downward PN, for urban water quality research, consists of three VeeJet nozzles oscillating to regulate intensity.
Iserloh et al. (2012)	1-Lechler 460 608	NA	0.2 (0.45 × 0.45)	0.28	25	2	1.00–1.50	5.8	40	90.6	Downward PN, for soil erosion study using single nozzle.
Wildhaber et al. (2012)	1-Lechler 469.448.39	NA	Six 1 (1 × 1)	1 (1 × 1)	_	7.0 for DF and 1.0–1.5 for hybrid RS	1.80 for DF and 3.50 for hybrid RS	12.4 ± 1.3 for DF and 12.2 ± 1.4 for hybrid RS	60	_	Two simulators, DF and hybrid, for soil erosion study. Hybrid made from PN using screen size of 2 mm × 1.7 mm placed 50 cm below nozzle.
Júnior and Siqueira (2011)	2-FullJet ½ SSHH40	1.06	_	3 (2 × 1.5)	50, 80, 110, 140 and 170	2.8	2.12 (for 80 kPa), 2.50 (for 50 kPa) and 1.63 (110 kPa)	22.53 (for 80 kPa	40–182	68.3–82.2	Downward PN, for urban water quality research. No oscillation; intensity was regulated by solenoid valves located at each nozzle.
Sangüesa et al. (2010)	4-UniJet TG-SS14 W	1	Spray area 4 (1.25 × 3.25) for four nozzles or (2 × 2) for single nozzle	2.4 (2.4 × 1)	75 and 100	1.8	_	_	124–119	86, 91.4, 90	Downward PN; RS has no frame, it stands on four iron pillars, each nozzle with 1.8 m radius. Cu = 86% (single nozzle, 2 m × 2 m plot), 91.4% (central 1 × 1 zone), 90% (4 nozzles on 1 m × 1.25 m plot).
Assouline et al. (1997)	1-FullJet	Single	Single nozzle on 3.5 m arm	16 (4 × 4)	95	5	1.17	_	12	87.3	Upward PN, tested three FullJets with 3.6, 5.2 and 6.7 mm nozzle diameters.
					135		1.21		20	83.2
					120		1.34		28	86.3
Esteves et al. (2000)	6- 1H106SQ	Located on 5.5 × 5.5 square grids	Depends on number of nozzles	50 (5 × 10)	41.18	6.53	2.4	23.5	65	78–92	Upward PN, for soil erosion study. Each nozzle sprays an area of 7 m × 7 m. Six nozzles cover 50 m^2. Water jetted to 7.5–8 m.
Abudi et al. (2012)	1–1½″ H 30 nozzle	NA	_	Two (1 × 2) total 4	60	2	1.5	9.9	29	98	Upward PN, for soil erosion study. Intensity was regulated by a pneumatic valve. Water jetted to 4.5 m.
Cerdà et al. (1997)	1-Hardi-1553–10	NA	0.16 (0.4 × 0.4)	0.24	152	2	2.53	7.1	55	93.2	Downward PN, for infiltration studies. Using single nozzle that covers approx. 1 m^2.

Table 2. Average values of rain uniformity with respect to plot size.

Plot size category (m^2)	References	Rain uniformity, Cu (%)
		Mean	SD	Min.	Median	Max.
Small (0.1–2.0)	Hignett et al. (1995), Cerdà et al. (1997), Vahabi and Nikkami (2008), Sangüesa et al. 2010), Dunkerley (2012), Iserloh et al. (2012), Wildhaber et al. 2012)	93	3.90	89	92	98
Medium (2.1–4.0)	Blanquies et al. (2003), Herngren et al. (2004), Júnior and Siqueira (2011), Abudi et al. (2012)	87	11.90	68	90	98
Large (>4.0)	Assouline et al. (1997), Esteves et al. (2000), Kim et al. (2010), Aksoy et al. (2012)	85	2.25	82	86	89

Table 1 reveals an interesting pattern of sizes between DFs and PN simulators. The DFs are generally smaller in size (0.98 ± 0.68 m²) and cover a smaller plot area (1.07 ± 0.12 m²). In contrast, PN simulators are generally larger (5.12 ± 1.58 m²), except for those using single nozzles. PN simulators are generally used on plot sizes 4.47 ± 2.48 m² when two, three or four nozzles are involved.

Although the actual sizes of the PN simulators were seldom reported, they were commonly of two styles: those on a suspended boom and those on elevated pipes. The designs of Blanquies et al. (2003), Herngren et al. (2004), Kim et al. (2010), Sangüesa et al. (2010), Júnior and Siqueira (2011) and Aksoy et al. (2012) belong to the former category, while the designs of Assouline et al. (1997), Esteves et al. (2000) and Abudi et al. (2012) belong to the latter category. Their basic difference is in the method of spraying water. Nozzles placed pointing upward are meant to enable water drops to fall at terminal velocities, similarly to natural rainfall. Their spray area usually covers a minimum of 8 m² and a maximum of 16 m², depending upon the nozzle type and the applied pressure.

Plot areas used by researchers may be smaller or larger than the RS surface area. For instance, Hignett et al. (1995), Herngren et al. (2004) and Sangüesa et al. (2010) had their RS surface areas bigger than the plot areas; while, Cerdà et al. (1997), Vahabi and Nikkami (2008), Dunkerley (2012) and Iserloh et al. (2012) chose their plot size bigger than the RS. This variability may be dependent on the research objective or its scale of interest. Thus, plot sizes are dictated by the size of the RS, the parameter being investigated, the accessibility and the slope (Loch and Donnollan 1983, Battany and Grismer 2000, Sharpley and Kleinman 2003).

Many methods have been employed by researchers to measure discharge from the plot surface when using RSs. However, there has been no consensus about the best method for measuring discharge, which makes comparison between RS studies extremely difficult (Grismer 2011). Although efforts were made to standardize the use of plots (e.g. Agassi and Bradford 1999, Kinnell 2005, 2006, Wildhaber et al. 2012), to the best of our knowledge, no standardized method has yet been developed.

Pressure

The correct choice of a nozzle is fundamental for the success of PN simulators in replicating natural rainfall as closely as possible (Tossell et al. 1987). Setting the pressure at which a balance could be achieved between rain intensity, uniformity, raindrop diameter and KE is the keystone of selecting operating pressure. The approach to setting the pressure differs among researchers. Cerdà et al. (1997) found the most homogeneous rain at 152 kPa using a HARDI-1553–10 nozzle; above this value there was a higher rain concentration at the border, and below this value there was a higher rain concentration at the centre of the plot. They found increased rain uniformity with increasing pressure up to a maximum value, after which the uniformity began to decrease. Júnior and Siqueira (2011) used a different nozzle (FullJet 1/2SSHH40) and found a similar trend, but the uniformity tended to remain constant at pressures above 150 kPa and they achieved a simulated rain uniformity of 86% at 170 kPa. To determine whether any trend exists, a regression analysis was performed on both relationships and a general exponential equation (Equation (1)) was obtained for the two datasets, with R² = 0.81 and R² = 0.99, respectively:

(1)

where C_u and p are uniformity coefficient (%) and pressure (kPa), respectively, and x = 89, y = 76 × 10, z = 0.85(Cerdà et al. 1997)

x = 85, y = 386, z = 0.94 (Júnior and Siqueira 2011)

Although the C_u–p relationships are generally similar, it is difficult to tell whether the monitored pressures referred to in the Cerdà et al. (1997) and Júnior and Siqueira (2011) designs are the nozzle pressure or the pump pressure. This missing information has fundamental implications for understanding and modelling interdependent RS components.

The effect of pressure across similar nozzles of different apertures can be appreciated from the work of Aksoy et al. (2012), which is presented in Table 1. Except for the VeeJet 8060, the influence of nozzle aperture on C_u shows an increased uniformity with increasing pressure and aperture size. Thus, the bigger the aperture, the more uniform the rain distribution; however, more pressure would be needed to achieve higher rain intensity. Pall et al. (1983) studied the effect of the nozzle aperture angle on the intensity and uniformity of simulated rain. They posited that the uniformity and intensity of the simulated rainfall are affected mostly by the nozzle pressure and the nozzle aperture angle, with aperture angle having the greatest impact on intensity and nozzle pressure on uniformity. Further work on the uniformity, KE, impacting angle, fall height and drop size can be found in Lovell et al. (2002).

The effect of pressure on the same nozzle can be appreciated from the work of Júnior and Siqueira (2011), which is presented in Table 1, with the C_u–p relationship presented in Figure 1. Sangüesa et al. (2010) operated their RS at two pressures, 75 kPa and 100 kPa, to investigate soil loss in Chile. They obtained a mean rain uniformity of 86% on a 2 m × 2 m plot using one nozzle and 90% using four nozzles spaced 1 m apart on a 1.25 m × 3.25 m plot.

Figure 1. Simulated rain uniformity and operating pressure relationship.

Nozzle spacing and oscillation

The spacing between nozzles plays a significant role in ensuring uniformity of rain. Uniformity of rain is expected to differ between overlapping areas and non-overlapping areas of spray. Unfortunately, information on this very important aspect is scarce in the literature, making comparison between studies difficult. Aksoy et al. (2012) adjusted the nozzle spacing between 1.25 m and 1.45 m for different simulated intensities using different nozzles to maintain nearly the same simulated rain uniformity, leading to the use of different plot sizes ranging between 0.25 m² and 12 m², depending upon the type of nozzle used and operating pressure.

Oscillation of the boom ensures spatial variation of rain but reduces the uniformity, as a significant portion of the rain is sprayed outside the plot area. Herngren et al. (2004) and Júnior and Siqueira (2011) incorporated a catch tray to recover this out-of-plot water and minimize the use of water within the RS system. Sangüesa et al. (2010) used cut-off valves to reduce the water cut-off time. They also reported two instances of increased rain uniformity: at the central zone of the single nozzle and at the intersecting zone between nozzles. Aksoy et al. (2012) also reported a higher proportion of larger drop sizes in the central zone and higher intensity at the intersecting zones. As shown in Table 1, for downward sprayers, on average, nozzles are spaced 1.1 m apart. For upward sprayers, nozzles are spaced no less than 5 m apart when more than one nozzle is required.

Uniformity

Uniformity of simulated rainfall on a plot is a very important measure of how rainfall is spatially distributed on the plot to avoid localized saturation of one area under the plot (Tossell et al. 1987). The most important aspect of ensuring spatial distribution of drops over the plot is the consistency of uniform KE (Lascelles et al. 2000, Ries et al. 2009). The uniformity of rainfall is normally estimated using the uniformity coefficient (C_u) (Christiansen 1942):

(2)

where SD is the standard deviation of the rain over the plot and I_m is the mean simulated rain intensity.

Different nozzles exhibit different spray patterns. Nozzles used in RSs are generally of two types (cone and flat spray), based on their mould. Uniformity for cone spray nozzles decreases from the centre of the plot outwards (Cerdà et al. 1997, Lovell et al. 2002, Sangüesa et al. 2010, Abudi et al. 2012, Iserloh et al. 2012). For flat-spray nozzles, the uniformity of the rain depends upon the width or length but generally decreases from the edges inwards (Herngren et al. 2004). This phenomenon of localized saturation is usually referred to as the “edge effect.” Another dependent factor for PN simulators is the operating water pressure and simulated rain intensity. As shown in Table 1, uniformity increases with increasing pressure and intensity. Drop formers demonstrate that their intensities remain fairly uniform across the area. Spatial rainfall distribution from PN simulators can vary significantly between different test experiments by the mere changing of nozzles (Ries et al. 2009). Miguntanna (2009) and Egodawatta (2007) used RS developed by Herngren (2005) for environmental studies. They achieved different drop diameters and DSDs using the same simulator, suggesting the effect of factors such as wear and tear of the parts, changing nozzle angle and run-time error.

Methods of measurement play a significant role in achieving correct data on rain uniformity over the plot area. Júnior and Siqueira (2011) measured intensity over the entire plot using a 1.5 m × 2.0 m × 0.1 m reservoir in one instance and 63 cups in 250 mm grids in another. They noted that the commonly used method of using cups underestimates the intensity at lower pressures. This method is laborious where a plot of more than a square meter is involved, but has been used for small plots involving not more than one nozzle (Iserloh et al. 2012).

As noted by Júnior and Siqueira (2011) and Ries et al. (2009), it is difficult to compare the uniformity based on RS types because of the different methods used to measure simulated rain uniformity and different reporting of where the samples were taken within the plot. However, DF simulators achieved higher uniformity than PN simulators at lower rain intensity. Irrespective of RS type, researchers achieved an average of 83% simulated rain uniformity within the range of 10–182 mm h⁻¹ intensity.

Drop size and drop size distribution (DSD)

The reported literature on DSD has yet to establish a standard method for obtaining raindrop diameter size and DSD using RS. However, the most widely used parameters are the volumetric median (D₅₀) and mean drop (D_m) diameters. The D₅₀, D_m and DSD in the literature were found to differ depending upon which method was used. The most commonly used methods were the flour pellet method (Herngren et al. 2004, Júnior and Siqueira 2011), the stain method (Assouline et al. 1997, Lovell et al. 2002, Salem et al. 2014), the oil immersion method (Blanquies et al. 2003), the piezoelectric disdrometer (Wildhaber et al. 2012) and the laser method (Iserloh et al. 2012, Sanchez-Moreno et al. 2012). The laser and piezoelectric methods tend to overestimate larger drops (Kincaid et al. 1996), while the stain and flour pellet methods tend to underestimate the percentage volume of smaller drops (Kincaid et al. 1996, Campos 1999). However, the oil immersion method is laborious. Kincaid et al. (1996) compared these methods using different nozzles and concluded that different measurement methods give different raindrop diameters and DSDs. Ries et al. (2009) also evaluated these methods using RS consisting of a circular plot area of 0.3 m² and a single Hardi Syntal 1553–10 nozzle placed 2 m above the plot surface and operated at a pressure of 200 kPa. A constant intensity of 40 mm h⁻¹ was simulated and its DSD was measured. They noticed variability of both drop size and the DSD for different methods. The study concluded that different DSD patterns were dependent on the method used in their measurement. Lascelles et al. (2000) compared the rainfall spatial variation of two different RSs, the DF and the PN; they concluded that the two types produced similar and consistent DSDs.

The raindrop diameter values of DFs obtained in this review (Herngren et al. 2004, Vahabi and Nikkami 2008, Dunkerley 2012) and presented in Table 1 suffer from narrow DSDs compared with PN simulators. The DF simulators produce an average raindrop diameter of 3.45 ± 1.2 mm. Thus, DF simulators could not model the spatial variation of raindrop diameters smaller than 2.2 mm, which are found in natural rainfall. This inability to model the spatial variation of small raindrops is quite understandable, as the drops depend on the size of the dripper and the dripper is often made to allow drops to fall under gravitational force. However, DFs are still being used as a tool in the study of physical processes with noteworthy outcomes (e.g. Vahabi and Nikkami 2008, Dunkerley 2012).

For PN simulators, the DSD depends primarily on the nozzle type, operating water pressure, height of fall and intensity of the simulated rain (Cerdà et al. 1997, Ries et al. 2009). The DSD may also depend upon the nozzle aperture, nozzle pressure, spacing of the nozzles and the periodic nozzle movement. Raindrop diameters of DF simulators range from 2.4 mm (Dunkerley 2012) to 5.1 mm (Hignett et al. 1995), while PN simulators produce smaller raindrops of 0.8 mm (Lascelles et al. 2000) to 3.1 mm (VeeJet nozzles). The raindrop diameter of the upward PN simulators is typical of any nozzle-type RS and depends on factors stated above rather than its spray style. Placement of nozzle(s) pointing upward at the end of the water-supply pipe is best suited for this type of RS to reduce obstruction of falling drops. The choice of RS will therefore depend on the D₅₀ to be simulated rather than the method of simulating the rain.

Intensity and kinetic energy

Intensity control in drop formers and pressurized simulators

Intensity is one of the measures by which RSs are appraised. Most other natural characteristics of rainfall are defined by rain intensity. The most closely related to rain intensity are KE and the resulting momentum (Salles et al. 2002, Fornis et al. 2005, Brodie and Rosewell 2007, Sanchez-Moreno et al. 2012). Most RSs are operated at constant rain intensity, though methods vary to achieve the temporal changes between DF and PN simulators. The control of the simulated rain intensity is central to rainfall simulation on any surface.

Control of rain intensity can be quite tasking and time consuming using DF simulators. Control of rain intensity involves manual movement of the casing of the drippers vertically upward or downward (Dimoyiannis et al. 2001), or changing the drippers manually (Hignett et al. 1995). For PN simulators, both the intensity and the raindrop diameters can be controlled easily by controlling the pressure or the swing motion. The use of DF simulators is best suited to studies where uniformity of rain at a lower application rate is desired.

Kinetic energy regulation in drop formers and pressurized simulators

The KE of rain, a measure by which rain energy is quantified, could be expressed as a function of the rain volume (J m⁻² mm⁻¹) or duration (J m⁻² h⁻¹). The method of regulating KE differs among RS types and research objectives. Achieving a higher KE with DF simulators is synonymous with a RS that is less portable due to the fall height required. Using different modules to vary the raindrop sizes from 2.7 mm to 5.1 mm and fall height from 0.17 m to 2.54 m, Hignett et al. (1995) achieved KE of 17 J m⁻² mm⁻¹ for 40 mm h⁻¹ rain and a maximum KE of 20 J m⁻² mm⁻¹ for 100 mm h⁻¹intensity. Aksoy et al. (2012) achieved a similar result using downward VeeJet nozzles at a height of 2.4 m, but at reduced rain uniformity over the plot. Their RS attained a KE of 21 J m⁻² mm⁻¹ at 45 mm h⁻¹ rain and of 26 J m⁻² mm⁻¹ at 105 mm h⁻¹. Wind can also cause a significant sway of raindrops outside the plot area (Yu et al. 2003) possibly due to low KE, which is undesirable on non-erodible surfaces because of limited plot size.

The PN simulators with upward sprays achieve a KE that is similar to natural rainfall at a lower rain intensity, because the drops are allowed to accelerate upward at high pressure until their initial exit velocity in the nozzle equals zero. The drops will then begin to accelerate downward. With sufficient height of fall, the drops will attain and fall at terminal velocities. The RS developed by Abudi et al. (2012) achieved a low KE despite its high intensity of rain because the height of 2 m was not sufficient for a 1.5 mm drop size to have reached terminal velocity. This observation would also suggest that the height of the RS correlates negatively with rain uniformity because of the projectile nature of the raindrops. A short height may suggest better uniformity of rain intensity but will conflict with achieving a higher KE. The results presented in Table 1 indicate that, irrespective of how an isolated single nozzle sprays water, the nozzle consistently records KE as low as 5.8 J m⁻² mm⁻¹ (Iserloh et al. 2012) and a maximum of just 12.2 J m⁻² mm⁻¹ (Wildhaber et al. 2012). On average, the investigators could only achieve a KE of 8.7 ± 2.9 J m⁻² mm⁻¹. The low value of KE could perhaps be attributed to a limited raindrop population compared with combined nozzles.

Generally, the data presented in Table 1 indicate linearity of rainfall uniformity and KE with increasing rain intensity. This linearity remains one of the primary shortcomings of PN simulators to achieve comparable KE of rain at lower intensity with natural rainfall because most of the RSs are often operated at excess intensity to achieve reasonable uniformity of rain.

Influence of fall height in achieving kinetic energy

The distance between a rain simulation device (drip hole or nozzle face) and the surface of the plot plays a dominant role in determining the KE of the simulated rain. The DF simulators require a “fall height” of approximately 8 m to reach terminal velocity (Ries et al. 2009, Wildhaber et al. 2012). Iserloh et al. (2012) reported that their RS achieved a very low KE of 5.8 J m⁻² mm⁻¹ due to the small fall height of 2 m. However, close examination of Table 1 reveals that the raindrop diameter has a proportionate influence on KE, as much as the fall height. Thus, Wildhaber et al. (2012) used a similar nozzle with an increased drop diameter and achieved a higher KE of 12.2 ± 1.4 J m⁻² mm⁻¹ at a lower fall height (1.0–1.5 m). The results presented by Hignett et al. (1995) suggest that the influence of raindrop diameter on KE increases with height and that the raindrop diameter tends to have no effect at lower fall heights. Thus, raindrops with diameters of 2.7 mm and 5.1 mm achieved the same KE of 1.6 J m⁻² mm⁻¹ at a lower fall height of 0.17 m, but KE was 16.69 J m⁻² mm⁻¹ and 19.9 J m⁻² mm⁻¹, respectively, at a height of 2.54 m. These observations underline one of the fundamental differences between DF and PN simulators in terms of the fall height required to achieve a reasonable KE.

The PN simulators tend to achieve higher KE at a shorter height than DF simulators due to the action of pressure. For a fall height of 2.4 m and a D₅₀ of 2.19 mm, Aksoy et al. (2012) and Herngren et al. (2004) achieved KEs of approximately 25 J m⁻² mm⁻¹ compared with the results of Hignett et al. (1995), where a height of 2.5 m and a bigger D₅₀ of 2.7 mm achieved a KE of only 16.6 J m⁻² mm⁻¹. Comparing the result of DF and PN downward sprayers with upward sprayers shows that DF simulators underestimate the KE at a shorter height. However, PN downward sprayers overestimate the KE at a similar height. The PN upward sprayers achieved KE similar to that of natural rainfall more easily than the two types presented above because the jetted water travelled high enough to fall at terminal velocity. Although Abudi et al. (2012) used PN upward RS, they did not achieve a KE higher than 9.89 J m⁻²mm⁻¹, despite the small D₅₀ of 1.5 mm for the RS, which was less affected by drag forces due to the smaller surface area. Small drops thus attained terminal velocities much faster than larger drops because the fall height of 4.5 m was not sufficient for the raindrops to have reached terminal velocities, as compared with Esteves et al. (2000), who jetted the water to a fall height of 7.5–8.0 m.

The interdependency of nozzle pressure and spacing on intensity, uniformity of KE and drop size and the influence of plot size on uniformity are summarized in Table 3.

Table 3. Relationships of RS interdependent components (R+: positively correlated; R–: negatively correlated; N: not related).

	Pressure	Intensity	Cu	Plot Area	D	Nozzle space	KE*
Pressure
Intensity	R+
Cu	R+	R+
Plot Area	N	N	R-
D	R-	R-	N	N
Nozzle space	N	R-	R-	N	R-
KE	R+	R+	R+	N	R+	R+

*Pressure increase with KE remains valid when certain threshold of pressure is reached.

Increased pressure increases intensity, uniformity and KE but decreases drop size. However, the increased pressure with intensity holds true when a certain threshold value of pressure is reached and the drops become too small and mist is produced. A plot area different from the total spray area is not related to any of the parameters except uniformity. Nozzle pressure does not seem to influence nozzle spacing, but nozzle spray angle does influence nozzle spacing, as discussed inter alia. In RS design, the nozzle spacing is often kept constant. The nozzle pressure and angle therefore influence only the spray area based on fixed nozzle spacing. Irrespective of the RS type, the most influential parameters to be considered in the success of RSs are the intensity, kinetic energy and uniformity of the simulated rain.

Requirements of simulation on erodible and non-erodible surfaces

Investigation on either erodible or non-erodible surfaces requires a RS that can achieve rainfall characteristics as close as possible to those of natural rainfall at affordable cost, and is portable, easy to control and operate. But, as pointed out earlier, the simulated characteristics vary greatly between the diverse RSs and a balance could not be reached between all the simulated rainfall parameters without a compromise. This section summarizes the differences between the requirements of simulation on erodible and non-erodible surfaces with emphasis on the most important factors that need to be achieved when considering simulation on non-erodible surfaces.

Most of the investigations on erodible surfaces have involved erosion, tillage and infiltration studies (Coutinho and Tomás 1995, Agassi and Bradford 1999, Dimoyiannis et al. 2001, Badia and Marti 2008, Abudi et al. 2012, Dunkerley 2012). In contrast, the processes concerning urban wet weather studies involved non-erodible surfaces and were defined based on the volume of pollutants and the corresponding volume of discharge (Corbett et al. 1997, Bertrand-Krajewski et al. 1998, Deletic 1998, Sansalone and Cristina 2004). For instance, Corbett et al. (1997) measured storm water runoff volumes, flow rates and sediment loads from an erodible watershed and a non-erodible watershed using 10 simulated rainfall events. The simulation results indicated runoff volume was on average 5.5× (±2.7) and sediment yield 5.5× (±2.3) greater from the non-erodible surface than from the erodible surface. They also found that the proportion of rainfall to runoff volume was on average 14.5% higher on the urban non-erodible surface compared to the erodible surface. The study concluded that runoff volumes were governed by the total non-erodible area and were independent of other impervious surface spatial characteristics (e.g. shape, contiguity, location, size). Hall et al. (1989) also concluded that the drop diameters and the fall velocities are of primary consideration in erosion and infiltration simulation involving erodible surfaces, and further posited that simulation involving non-erodible surfaces requires rainfall–runoff as a principal criterion. Many researchers involved with simulation on non-erodible surfaces have confirmed the importance of the rainfall–runoff relationship (Nazahiyah et al. 2007, Chow et al. 2012, Zhao and Li 2013, Yakubu 2015). The simulation on non-erodible surfaces increased runoff volumes linearly and peak flow rates exponentially; flow rates and sediment loads were also controlled by non-erodible surface spatial characteristics (Corbett et al. 1997). From this review the following are emphasized:

(1) The methods for collecting runoff on erodible and non-erodible surfaces are not the same. The generation of simulated rain on non-erodible surfaces is more challenging in both the method of simulating the rain and collecting the runoff (Herngren et al. 2005b) because non-erodible surfaces are mostly public paved surfaces where digging and excavation are restricted. Recovery of all the generated runoff from plots with non-erodible surface is a desired priority. This generated runoff could give rise to difficulties when it comes to storage (Herngren 2005, Miguntanna 2009). Thus, an increased proportion of imperviousness brings with it shorter lag times between commencement of simulated rain and subsequent higher runoff peaks and total volume of runoff in receiving waters (Shuster et al. 2005). Therefore, the recommended plot on an urban non-erodible surface should be big enough to handle the runoff without overflow, but small enough not to generate excessive runoff volume. The recommended plot size should be medium, based the classification in Table 2. In NPS studies involving non-erodible surfaces, researchers have employed vacuum cleaners to ensure the entire runoff was recovered. Detailed criteria for choosing vacuum cleaners in sampling on non-erodible surface can be found in Yakubu and Yusop (2015).

(2) In the field of soil erosion and infiltration, the length and slope of the plot are important requirements for simulation (Clarke and Walsh 2007, Schindewolf and Schmidt 2012, Wildhaber et al. 2012, Ries et al. 2014). In contrast, the slope of the plot in the study of NPS on impervious surfaces has not been given much significance, except in simulation experiments that aimed at understanding the effect of storm movement and the influence of runoff responses on impervious surfaces (e.g. de Lima and Singh 2003).

(3) For urban non-erodible surfaces, it is important to recognize the mobility of pedestrians and vehicles. In addition, portability could play a more significant role in non-erodible surfaces because of space constraints. For instance, Egodawatta et al. (2007, 2009) and Herngren et al. (2006, 2010) used 2.5 × 1.5 m plots to simulate artificial rain on road surfaces. The width of 1.5 m is less than 40% of the standard lane width of a road considered sufficient to allow for continuous passage of vehicles.

(4) The use of elevated pipes with a single nozzle could be more suitable on erodible surfaces where the entire wetting area could be of interest and there is seldom a space constraint. Similarly, upward spray PNs, considering their covered area of spray of 8–16 m², also may not be suitable for simulation on non-erodible surfaces.

(5) The requirement for simulating reasonable intensity, runoff and rain depth in the case of non-erodible surfaces would make the PN more suitable for simulation in NPS investigations on non-erodible surfaces. The ease with which the rain parameters can be controlled/operated and the wide range of choices available for suitable nozzles to ensure a certain intensity is simulated is unprecedented in the PN simulators. The difficulty in ensuring temporal control of intensity for the DF simulators will limit their use in NPS investigations on non-erodible surfaces.

(6) Another factor that may affect drop diameter and its distribution is the simulated water quality. Unlike investigations of soil erosion and soil infiltration where the initial water quality is of low significance, urban water quality research often involves measurements of pollution levels and modelling of pollutant transport on non-erodible surfaces (e.g. Warnemuende et al. 2003, Herngren 2005, Egodawatta 2007, Yakubu 2015). For instance, Herngren (2005), used PN RS to study the kinetics of PAHs and heavy metals on urban non-erodible surface using deionized water. Egodawatta (2007) also used deionized water to study the buildup and washoff processes of heavy metals on urban paved surfaces. On the other hand, Yakubu (2015) used PN RS to study the transportation of heavy metals in urban areas from different paved surfaces using stored natural rainfall water. Water quality was found to present a challenge in DF simulators and could significantly affect the simulated raindrop diameter, DSD and other parameters such as uniformity and intensity (Clarke and Walsh 2007). Clogging may be caused by water impurities and algal and other microbial organisms that could develop with time in and around the drippers. This concern is less likely in the case of PN simulators because the pressure would assist in reducing blockages.

(7) Time to complete sampling is one of the most important requirements in experiments involving non-erodible surfaces using RSs, as most of the sampling locations are in urban areas. In this regard, the use of PN simulators would offer a better benefit than DFs, considering the time it would take to generate an appreciable amount of runoff using DFs. On the other hand, the use of DFs on erodible surfaces would be more desirable, especially in soil infiltration studies that require slow percolation of the simulated rainfall.

(8) The DFs achieved higher rainfall uniformity; this is considered a good requirement for erodible surfaces involving infiltration studies where the interest is in measuring water filtered downward at each point within the plot. Therefore, the samples at each location should have a minimum deviation. However, for non-erodible surfaces the interest is in collecting runoff after a certain event, which has few restrictions within each point on the plot. Thus, saturation of the entire plot would be achieved within a much shorter period of time than for erodible surfaces.

(9) The design of a RS requires flexibility of mounting and dismounting. The flexibility of a RS to select different rainfall intensities with ease is a general requirement for both erodible and non-erodible surfaces. However, it is more vital for non-erodible surfaces, because most of the experiments would require changes of intensity within a simulated storm and/or between storms in a very short period of time to control the generation of runoff volume.

Conclusion and recommendations

This review has brought to the fore the diversity and versatility of different RSs based on their design, configuration and fashion and how these parameters influence simulated rain characteristics. Largely, RSs are conceived, designed and fabricated based on research objectives. Drop former simulators ensure uniformity of rain but suffer from narrow DSD and low KE. The upward PN simulators present a good opportunity to replicate natural rainfall, but the requirement of a large plot size and the inability of the upward PN simulators to model the spatial variation of raindrop diameter may make their use on an urban non-erodible surface with constrained space unattractive. The downward PN simulators suffer from producing excessive KE with the accumulation of large volumes of water within a short period of time, but the ease of operation and flexible control of rain parameters in the downward PN simulators offer greater advantages to undertake parametric investigation on non-erodible surfaces. Large RSs are generally expensive and may require additional skilled labour above the basic requirements of small RSs, which would make them suitable for studies close to field conditions with full access. To ensure uniformity of results, a moderate-size RS would offer objective advantages when the investigation involves an urban non-erodible surface. No standard method of collecting runoff from a non-erodible catchment is yet available. Researchers involved in an urban non-erodible study may therefore need to give further thought to the method of collecting runoff.

This review also demonstrates that the versatility of different RSs is as intrinsic as natural rainfall characteristics. Even within the PN type, the RSs are as diverse as the available nozzles. The configuration, size and operation of the RSs are as diverse as the research objectives. To help standardize RS reporting to a comparative measure, we recommend that the following be taken into consideration when detailing an RS report:

1. The configuration and dimensions of RS size and the plot size should be distinguished clearly and precisely.

2. To help understand the referred operating pressure, the suction pressure and the nozzle pressure, the position of the pressure monitor in the RS should be clearly reported.

3. The unit of pressure should be standardized to kPa, while other seldom-used unconventional units should be avoided. For instance, Cerdà et al. (1997) put forward 2 kg cm⁻² as a low working pressure, but translating 2 kg cm⁻² into the widely reported units of kPa would indicate the high working pressure of 196 kPa.

4. It is crucial to provide the location where samples were taken within the plot—at the centre, at the intersecting zone(s) of nozzle sprays, or at the edges—with the corresponding pressure at which the samples were taken for both uniformity and raindrop diameters.

5. The mention of a nozzle should be accompanied by its technical catalogue number. Some researchers barely mention the name of the nozzle, while some include their order number instead of the catalogue number, and such omissions deny other researchers vital information for their work.

6. The method of measuring natural and simulated rainfall should also be standardized. As stated in this text, various methods exist for the measurement of drop diameters, which give variable results. Recently, Iserloh et al. (2013) compared different methods for measuring simulated and natural drop diameters, and recommended the use of a laser precipitation monitor over other methods because it allows for detailed comparisons between natural and simulated raindrop diameters. The use of a laser precipitation monitor for measuring natural drop diameters was also recommended by Gopinath et al. (2016).

In addition to the above, Grismer (2012) has recommended the standardization of methodology for RS design, runoff frame installation and the analysis of results, which needs to be developed and applied to all studies. Further recommendations can be drawn from the work of Iserloh et al. (2013). The study compared the simulated rainfall characteristics of 13 different small RSs. They found that identical measurement methods for simulated rainfall characteristics provided a means of achieving comparable results. The study also suggested that future RS should give adequate attention to water efficiency, drop size distribution and spatial rainfall distribution, as well as reproducibility, handling and control of test conditions.

Acknowledgements

Valuable comments by Professor Gustaff Olson and Professor Helmut Krois during the UTM and International Water Association (IWA) Publication Workshop in January 2013 are highly appreciated.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

The authors wish to acknowledge the Research Management Centre of Universiti Teknologi Malaysia (UTM) for facilitating this research under Research University Grant (vote no. Q.J130000.2509.01H72). This study was also supported by the Asian Core Program of the Japanese Society for the Promotion of Science (JSPS), the Ministry of Higher Education (MOHE) in Malaysia and Nineteen Consult Ltd, Nigeria.