Anda di halaman 1dari 25

A.I.J.M. van Dijk (2002) Water and Sediment Dynamics in Bench-terraced Agricultural Steeplands in West Java, Indonesia.

PhD Thesis, Vrije Universiteit Amsterdam

Chapter 7 Rainfall intensity - kinetic energy relationships: a critical literature appraisal

Abstract Knowledge of the relationship between rainfall intensity and kinetic energy and its variations in time and space is important for erosion prediction. However, between studies considerable variations exist in the reported shape and coefficients of this relationship. Some differences can be explained by methods of measurement and interpretation and sample size, range and bias, while part of the variability corresponds to actual differences in rainfall generating mechanisms. In this chapter we critically review published studies of rainfall intensity and kinetic energy with a view to derive a general predictive equation of an exponential form. The performance of this general equation is compared with that of existing equations using measured rainfall intensity and kinetic energy data for a site in southeastern Australia. It appeared that the energy of individual storms could only be predicted with limited accuracy because of natural variations in rainfall characteristics. By and large, the general equation produced energy estimates that were within 10% of predictions by a range of parameterisations of the exponential model fitted to specific data sets. Re-calculation of rainfall erosivity factors as obtained by the older and revised USLE approaches does not seem warranted for most locations. However, in regions experiencing strong oceanic influence or at high elevations, overall rainfall energy appears to be considerably lower than predicted by the general or USLE equations. Conversely, data collected at semi-arid to sub-humid locations suggest that rainfall energy may be higher than expected under those conditions. Standardised measurements are needed to evaluate rainfall intensity - kinetic energy relationships for such areas. Published as: Van Dijk, A.I.J.M., Bruijnzeel, L.A., Rosewell, C.J, 2002. Rainfall intensity - kinetic energy relationships: a critical literature appraisal. Journal of Hydrology 261: 1-23.

7.1. Introduction It is well-established that the amount of soil that is detached by a particular depth of rain is related to the intensity at which this rain falls. The results of various studies further suggest that soil splash rate is a combined function of rainfall intensity and some measure of raindrop fall velocity (Ellison, 1944; Bisal, 1960 and many later workers). In particular, rainfall kinetic energy EK (the product of mass and fall velocity squared) has often been suggested as an indicator of rainfall erosivity, i.e. the ability of rainfall to detach soil particles (e.g. Mihara, 1951; Free, 1960). Rose (1960) concluded that rainfall

107

CHAPTER 7 - RAINFALL INTENSITY - KINETIC ENERGY RELATIONSHIPS momentum is a slightly better predictor of soil detachment than kinetic energy, but Hudson (1971) demonstrated that for natural rainfall, momentum and kinetic energy exhibit very similar relationships with intensity. Other widely used indices of rainfall erosivity also include kinetic energy but allow for the fact that the use of kinetic energy alone tends to over-estimate the erosivity of low intensity rainfall. Attempts to correct for this include the use of a threshold value for rainfall intensity (e.g. Hudson, 1961; Morgan, 1977) or multiplying kinetic energy with a measure of drop size (Govers, 1991). Alternatively, storm kinetic energy has been multiplied with a measure of maximum rainfall intensity (Wischmeier and Smith, 1958; Lal, 1976), sometimes in combination with the use of a threshold value for storm depth (Renard et al., 1997), although these latter erosivity indices relate to gross erosion and also include soil detachment and transport by runoff. The rationale for such adaptations relates to the finding that smaller drops (dominating low intensity rainfall) are less efficient at detaching soil (Sharma and Gupta, 1989; Salles and Poesen, 2000), in addition to the fact that at high rainfall intensities saturation and ponding of the soil (at least at low depths) may increase the efficiency of detachment (e.g. Sloneker et al., 1976; Torri et al., 1987). Despite these complications, rainfall kinetic energy still represents the total energy available for detachment and transport by rainsplash. Consequently, knowledge of the relationship between rainfall intensity and kinetic energy is important for the prediction of erosion hazard. Drop terminal velocity, and thus kinetic energy and erosivity, are strongly influenced by drop size (e.g. Laws, 1941). Wiesner (1895) was among the first to report on the distribution of drop sizes in natural rainfall arriving at the earths surface using a piece of absorbent paper dusted with a water-soluble dye. Since Wiesners pioneering efforts, numerous researchers have used this filter paper or stain technique, reviewed by Hall (1970). Other widely used methods include the flour pellet technique (Laws and Parsons, 1943), the use of electromechanical distrometers (Joss and Waldvogel, 1967), cameras (Kinnell, 1980; McIsaac, 1990) and, more recently, optical pluviospectrometers (e.g. Wang et al., 1979; Illingworth and Stevens, 1987; Lavergnat and Gol, 1998; Salles and Poesen, 1999). However, when the results of different studies of raindrop size distribution are compared, there appears to be considerable variability, both in terms of observed ranges and of spatial and temporal patterns. Possible reasons for this include methodological differences in measurements and subsequent calculations, data sets that encompass only a small range of rainfall intensities, as well as true differences between geographical locations or weather systems (Kinnell, 1980; McIsaac, 1990). Unfortunately, a comprehensive assessment of the importance of these various factors is still lacking. On a related note, the widely used (Revised) USLE approach to predict soil loss involves the calculation of storm kinetic energy using equations that are based on measurements at a single location only, i.e. Washington D.C. in the case of USLE (Wischmeier and Smith, 1958; 1978) and Holly Springs, Mississippi, U.S.A in the case of RUSLE (Renard et al., 1997), respectively (although the latter to some extent also draws on measurements made at other locations as explained in Section 7.3.2). The general applicability of these equations has not been evaluated systematically as yet, even though a considerable volume of additional information has become available in the last two decades. Thus, there is a need for a critical appraisal of the literature on the rainfall intensity kinetic energy (R-eK) relationship. The current paper aims to critically review published studies of rainfall intensity versus drop size distribution or kinetic energy and to evaluate their applicability for predictive purposes. To enable the use of data from different studies in the derivation of a general predictive equation, data were re-interpreted in
108

CHAPTER 7 - RAINFALL INTENSITY - KINETIC ENERGY RELATIONSHIPS some cases. Next, the ability of the newly found general equation to satisfactorily predict kinetic energy was examined by comparing its predictions with those of equations found for other locations. In case of major discrepancies, the causes were highlighted and the errors involved were compared with temporal variations in the R-eK relationship. Finally, the implications of using a general versus a site specific R-eK relationship for erosion hazard prediction and modelling are discussed.

7.2. Theory Rainfall represents a distribution of differently sized drops that attain corresponding different terminal velocities in stable air. Therefore, drop size distributions as measured in a volume of air and upon arrival at the earths surface will be different (Marshall et al., 1955; Uijlenhoet and Stricker, 1999). Because the current paper is concerned with processes at the earths surface, the term drop size distribution here refers to that measured at the ground surface only. The actual form of this size distribution is determined by the interaction of competing microphysical processes which lead to the growth (coalescence, condensation) or decay (break-up, evaporation) of raindrops as they fall to the ground (Uijlenhoet and Stricker, 1999). Raindrop size distributions are generally described as unimodal and positively skewed distributions that can be described with less than three parameters (Uijlenhoet and Stricker, 1999), although multimodal distributions have also been proposed (e.g. Asselin de Beauville et al., 1988). One of the most widely used parameterisations of raindrop size distribution is the twoparameter exponential distribution function proposed by Marshall and Palmer (1948). Their measurements were made under temperate continental conditions in Canada using the filter paper technique in rainfall of up to 23 mm h-1. The two parameters involved in -1 ur HhuhyyQhyr qvviv spv hr h hhrr  ) that is related to rainfall intensity by a power law equation and an (approximately constant) parameter N0 (mm-1 m-3uhtrurvu qrrvrurhyirsqvhyrshv Combining the Marshall-Palmer parameterisation with subsequently derived power law relationships between rainfall intensity and radar reflectivity (Marshall et al., 1955), Uijlenhoet and Stricker (1999) derived six different, consistent sets of power law equations relating different combinations of rainfall intensity (R), the two drop size qvvivhhrr hqN0), drop terminal velocity (v) and radar reflectivity (Z, cf. Table 7.1). From these equations other power law relationships, such as those relating rainfall intensity to median drop diameter, momentum or kinetic energy, are readily obtained (Uijlenhoet and Stricker, 1999). For example, the relationship between rainfall intensity, R (mm h-1) and median drop size, D50 (mm) can be expressed in the following general form:
D50 = R

[7.1]

Table 7.1 lists the values of the coefficients (in h) and in Eq. [7.1] as obtained by different studies. Reported values for range from 0.80 to 1.28 h against 0.123 to 0.292 for . Values of and derived by Uijlenhoet and Stricker (1999) varied considerably, depending on the particular set of equations used. In particular, the inclusion of radar reflectivity Z as a parameter led to coefficients that were notably different (Table 7.1).

109

CHAPTER 7 - RAINFALL INTENSITY - KINETIC ENERGY RELATIONSHIPS

Authors

Location

Coefficients (Eq. [7.1]) 1.28 0.92 0.92 0.80 0.98 b 1.15 1.08 1.03 1.06 0.93 0.93 0.95 0.173 0.210 0.123 0.270 0.292 0.211 0.210 0.214 0.210 0.229 0.210 0.258

Method

r2

Laws and Parsons (1943) Atlas (1953) a Brandt (1988) Kelkar (1959) Zanchi and Torri (1980) Van Dijk (unpubl.) Uijlenhoet and Stricker (1999)c

Washington D.C. Ottawa, Canada Amazonia, Brazil Poona, India Firenze, Italy West Java Canada c: N0 N0-v -v N0-Z -Z v-Z

flour filter filter filter flour filter theory

mm h-1 0.51 95 0.4-144 ? ? <23 ? 134 ? ? 107 0.1-41 0.88 ? 1-140 0.71 21 0.3-124 n.a. n.a. <23

Based on data of Marshall and Palmer (1948). " was found to be related to air temperature (in EC) as =0.499T 0.225; given value of is for T=20EC. c Theoretical values based on data of Marshall et al. (1947) and Marshall and Palmer (1948). See text for explanation.
b

Table 7.1. Values of the coefficients " and $ in the relationship between rainfall intensity (R) and median drop size (D50, cf. Eq. [7.1]) as reported in selected studies. Also listed, where known, are the method used to measure drop size, the coefficient of determination (r2) corresponding to the fitted relationship, the number of samples (N) and the rainfall intensity (R) range represented by them. n.a.= not applicable.

The relationship expressed by Eq. [7.1] implies that the median drop size continues to increase indefinitely with rainfall intensity. However, since Hudson (1965), there is substantial evidence to suggest that a maximum median drop size is reached at high rainfall intensities (usually above 70-100 mm h-1), after which the median drop size either stabilises (Kinnell, 1980; Rosewell, 1986; Brown and Foster, 1987) or even decreases (Hudson, 1963; Baruah, 1973; Carter et al., 1974). It is unfortunate in this respect that the measurements of Marshall and Palmer (1948) on which Uijlenhoet and Stricker (1999) based their theoretical derivations, were restricted to rainfall of relatively low intensity (less than 23 mm h-1). Therefore, it remains to be seen to what extent the power law relationship suggested by Eq. [7.1] is still applicable to rain falling at intensities exceeding 70-100 mm h-1. As stated earlier, drop size is related to terminal velocity, and therefore also with kinetic energy. The kinetic energy, EK,D (in J) of a raindrop of size D is the product of its mass, mD (in kg) and the square of its terminal velocity vD (in m s-1):
1 2 E K , D = mD v D 2

[7.2]

Note that D (in mm) represents the equivalent drop size, i.e. the diameter of a sphere that has the same volume as the drop under consideration. Hence, the mass of the drop is related directly to equivalent drop size and the density of water (1A10-6 kg mm3). Under stable atmospheric conditions, a raindrop of size D falling to the earth will attain a certain

110

CHAPTER 7 - RAINFALL INTENSITY - KINETIC ENERGY RELATIONSHIPS

10 8 6 Atlas and Ulbrich (1977) 4 2 0 0 1 2 3


D (mm)

Best fit exponential function

v (m s )

-1

Eq. [7.3]

Fig. 7.1. The relationship between raindrop size (D) and terminal velocity (v) under standard conditions (data from Gunn and Kinzer, 1949). The dashed line represents a best fit exponential function, the interrupted line the power-law function of Atlas and Ulbrich (1977), and the solid line a best fit polynomial function (Eq. [7.3]).

constant terminal velocity vD, at which point gravitational and frictional forces are in equilibrium. Using gradually more refined techniques, Laws (1941), Gunn and Kinzer (1949) and Hinkle et al. (1987) determined the relationship between drop size and terminal velocity. This relationship is not linear because larger drops tend to become more flattened by drag forces during their fall. Also as a consequence of air drag, there is an upper limit to the size of raindrops, above which they become unstable and tend to break up again. This limit appears to be in the order of 6 to 8 mm (e.g. Laws and Parson, 1943; Joss and Waldvogel, 1967). The resulting relationship between drop size and terminal velocity is usually described by an exponential or power law equation (Atlas and Ulbrich, 1977), although the latter, in particular, shows a rather poor fit to experimental data (Fig. 7.1). Beard and Pruppacher (1969) and Beard (1976) derived physical
rrv s ur rvhy rypv s hrq s ( & rvhyr

diameter, which describe the experimental data of Laws and Parsons (1943) and later workers (e.g. Gunn and Kinzer, 1949) very well. However, the practical application of these equations is limited by the fact that it requires calculation of the Reynolds number (NRe), which is a rather complicated function of air and fluid densities, dynamic viscosity and surface tension. Evaluation of the equations proposed by Beard and Pruppacher (1969) and Beard (1976) within the context of this paper showed that under standard conditions of air pressure (1 bar) and air temperature (20EC) and for drop sizes of 0.1 to 7 mm, these equations are approximated very well by the third-order polynomial equation: v D = 0.0561D 3 0.912 D 2 + 5.03D 0.254 [7.3]

111

CHAPTER 7 - RAINFALL INTENSITY - KINETIC ENERGY RELATIONSHIPS

Coefficients (Eq. [7.5]) Authors Laws and Parsons (1943) Carter et al. (1974) Zanchi and Torri (1980, 10 EC) Zanchi and Torri (1980, 20 EC) Van Dijk (unpublished) Blanchard (1953) Uijlenhoet and Stricker (1999) N0-v, -v, v-Z -Z N0-Z N0-
a

N 1.12 1.29 1.20 1.07 1.15 1.14 0.85 0.67 0.80 0.82 0.84 0.88 5a 13 ? ? 21 113

1.11 1.39 1.18 0.92

1.00 0.21 0.55 1.14

drop size distribution presented as histograms.

Table 7.2. Values of the coefficients and in the relationship between median volume-based drop size and effective drop size (Eq. [7.5]) as derived from selected studies. Where known, the number of measurements (N) is also listed.

This equation agrees to within 3% with the more complex equations of Beard (1976) and experimental data of, inter alia, Gunn and Kinzer (1949) (Fig. 7.1). Therefore, Eq. [7.3] can be considered accurate enough for the purpose of this study. However, changes in air temperature and especially in air pressure that are associated with a rise in altitude are known to influence drop terminal velocity (Hinkle et al., 1987). Beard (1977) gave a physical-mathematical description of this effect. The relationship between air temperature, pressure and fall velocity is drop size dependent and, again, rather complicated. Calculations were made by the first author for raindrop sizes between 1 and 6 mm, assuming a temperature gradient of 0.65 EC km-1 and a pressure gradient of 0.105 bar km-1, with initial values of 20EC and 1 bar at sea level, respectively. The results suggest that the kinetic energy of a raindrop at 500 m is about 4-5 % higher than at sea level, the difference increasing with drop size. Similarly, EK is already 8-10% higher at 1000 m and 44-61% higher at 4000 m. For a distribution of different drop sizes, as for instance found in the studies listed in Table 7.2, it appeared that assuming an increase in kinetic energy of 1% per 100 m rise in elevation yielded estimates that were within 1% of calculated amounts at altitudes up to 2000 m. Using Eqs. [7.2] and [7.3], the kinetic energy of a volume of rainfall can be calculated when the associated raindrop size distribution is known. For comparative purposes, the kinetic energy is best expressed as energy contents, i.e. the EK per unit rainfall depth. To distinguish rainfall kinetic energy contents from EK as such, the former will be denoted by eK (in J m-2 mm-1) from hereon. The kinetic energy contents associated with a particular drop size distribution is thus given by:

eK =

1 k 2 f i vi 2 i =1

[7.4]

112

CHAPTER 7 - RAINFALL INTENSITY - KINETIC ENERGY RELATIONSHIPS

4
10oC 1:1

D e (mm)

20oC

2
Van Dijk (unpubl.)

Laws and Parsons (1943) Carter et al. (1974) Blanchard (1953) Zanchi & Torri (1980)

0 0 1 2 3 4

D 50 (mm)

Fig. 7.2. Effective drop size (De), as defined in the text, versus median volume-based drop size (D50) in rainfall based on data presented in selected studies.

where fi is the mass fraction and vi the terminal velocity of drops in size class i, and the mass density associated with 1 mm of rainfall (in kg m-2 mm-1/ kg dm-3; taken to be unity). To compare the kinetic energy of a particular drop size distribution with that of its median drop diameter, it is useful to introduce the effective drop size (De), i.e. the drop size that would result in the same kinetic energy if all drops would be of this size, which is not necessarily equal to median volume drop size (D50). Because the terminal velocity of drops of different sizes changes differently with height, standard conditions are again assumed. Application of the various equations linking R  N0, v and Z as derived in the theoretical study of Uijlenhoet and Stricker (1999) results in the following power law relationship:

De = D50

[7.5]

where ( and * are dimensionless coefficients. The values of ( and * derived from the basic equations of Uijlenhoet and Stricker (1999) vary widely, depending on the set of underlying equations that is used (Table 7.2). Naturally, both median and effective drop sizes can also be determined in a more direct manner from measured drop size distributions. This was done using data from a number of published studies (Table 7.2, Fig. 7.2). The resulting values of ( ranged between 1.07 and 1.29, whereas the coefficient * varied between 0.67 and 0.88. There seems to be a weak inverse relationship between ( and *, such that De equals D50 for drop sizes between 1.5 and 3.0 mm (Fig. 7.2). It should be noted that the drop size distribution data presented by Zanchi and Torri (1980) were idealised and represented by equations relating rainfall intensity to volume percentile drop sizes. The resulting relationship appeared to be dependent on air temperature and therefore Eq. [7.5] was evaluated for temperatures of 10EC and 20EC in the present study.

113

CHAPTER 7 - RAINFALL INTENSITY - KINETIC ENERGY RELATIONSHIPS A number of studies have presented direct relationships between rainfall intensity and kinetic energy. Uijlenhoet and Stricker (1999) derived six power law equations relating kinetic energy to rainfall intensity in a theoretical manner. It should be reiterated, however, that the drop size measurements on which these equations were ultimately based (Marshall and Palmer, 1948) represented rainfall intensities of less than 23 mm h-1. Moreover, the power equation that was used by Uijlenhoet and Stricker (1999) to relate drop size to terminal velocity (Atlas and Ulbrich, 1977) must have led to additional errors in the estimates of eK (Fig. 7.2). Similarly, Wischmeier and Smith (1958) used measurements of drop size distribution and terminal velocity made at Washington D.C. under near standard conditions (Laws, 1941; Laws and Parsons, 1943) to derive a relationship between rainfall intensity and kinetic energy contents. Instead of a power law relationship they proposed a logarithmic function having the form:
e K = 11.9 + 8.73 log R

[7.6]

Eq. [7.6] has been used for decades and continues to be used all over the world within the framework of the (R)USLE approach for predicting soil loss (e.g. El-Swaify et al., 1982). Although proven to yield useful estimates of rainfall erosivity, there is a case to be made against the use of a logarithmic relationship. Like a power law equation, Eq. [7.6] implies that there is no upper limit to kinetic energy, whereas research has suggested that a maximum value does exist (Hudson, 1963; Baruah, 1973; Carter et al., 1974; Kinnell, 1980; Rosewell, 1986; Brown and Foster, 1987). Wischmeier and Smith (1978) also acknowledged this phenomenon and proposed an adaptation to their original equation, to the effect that the kinetic energy contents was considered to remain constant at 28.3 J m-2 mm-1 for rainfall intensities exceeding 76 mm h-1 (equalling 3 inch h-1). Other researchers, starting with Kinnell (1980), suggested that a continuous, exponential equation better describes the R-eK relationship. Such an equation has the general form: eK = emax [ a exp( bR )] 1 [7.7]

where emax denotes maximum kinetic energy contents and a and b are empirical constants. The coefficient a, together with emax determines the minimum kinetic energy contents (i.e. the value attained at very low rainfall intensity). The coefficient b on the other hand defines the general shape of the curve. A low value of b will result in a curve that only gradually approaches emax at high rainfall intensities (Kinnell, 1980). Although it often produces curves that are rather similar to the approach proposed by Wischmeier and Smith (1978), a continuous equation like Eq. [7.7] is to be preferred for the sake of consistency. Possibly also for this reason, Renard et al. (1997) suggested the use of an equation proposed by Brown and Foster (1987), having the form of Eq. [7.7], as an alternative to Eq. [7.6] in the Revised USLE approach. In addition, Eq. [7.7] involves one fitting coefficient more than Eq. [7.6] and this potentially increases its flexibility to describe experimental data. The three forms of the R-eK relationship (i.e. power law, logarithmic and exponential equations) have been compared with kinetic energy contents as calculated from the drop size distribution data of Laws and Parsons (1943) using the methods described in Section 7.3.1. The respective curves are shown in Figs. 7.3a and 7.3b, with rainfall intensity

114

CHAPTER 7 - RAINFALL INTENSITY - KINETIC ENERGY RELATIONSHIPS

40
(a)
exponential (Eq. [7.7]) r 2=0.53 e K=13.1R 0.191 r 2=0.53

30 e K (J m mm )
-1 -2

20
logarithmic (USLE; Eq. [7.6]) 2 r =0.52

10

0 0 20 40 60 -1 R (mm h ) 80 100 120

40

(b)
30 e K (Jm -2 mm -1 )

power-law

20

exponential

10
logarithmic

0 0.1 1 10 -1 R (mm h ) 100 1000

Fig. 7.3. (a) linear and (b) semi-log plots of relationships between rainfall intensity (R) and kinetic energy contents (eK), based on the data of Laws and Parsons (1943) and fitted with power law and exponential equations, respectively. The logarithmic equation proposed by Wischmeier and Smith (1958, 1978) is also shown.

plotted on linear and logarithmic scales, respectively. A power law relationship seems to predict kinetic energy contents at lower rainfall intensities well but over-estimates eK at higher intensities (Fig 7.3a). The logarithmic curve of Wischmeier and Smith (1978) describes eK at both low and high rainfall intensities rather well, but, as stated earlier, there are some limitations to its use. Finally, an exponential curve fits the data particularly well at higher intensities (above ca. 50 mm h-1; Fig. 7.3a). However, the form of Eq. [7.7] implies not only a maximum value for kinetic energy contents, but also a minimum, which is reached at very low intensities. Although this does not necessarily

115

CHAPTER 7 - RAINFALL INTENSITY - KINETIC ENERGY RELATIONSHIPS represent reality (Fig. 7.3b), the effect on overall storm kinetic energy EK is small. To compute overall storm EK, rainfall intensity has to be multiplied times the corresponding eK values for fixed time intervals. Consequently, higher rainfall intensities are much more important in determining overall storm energy than lower intensities. As such, it is of prime importance that good estimates are obtained for eK at higher rainfall intensities. Bearing this in mind, an exponential equation is better suited than a power law or logarithmic equation (cf. Kinnell, 1980; Brown and Foster, 1987; Renard et al., 1997). In the following sections, published data on kinetic energy or drop size distributions will be used to parameterise Eq. [7.7] for the respective measuring sites listed in Table 7.3.

7.3. Methods 7.3.1. Analysis of published data In total, 19 studies were identified in the literature that presented observations of raindrop size distribution and/or kinetic energy at different rainfall intensities. These measurements were made at 24 different locations. For the sake of readability, the respective studies are referred to by number from hereon: site and methodological background information for the original studies and details of the methods of data analysis used in the present review are listed in Table 7.3. A number of studies had already interpreted their measurements using an exponential relationship of the form of Eq. [7.7] and these could be cited directly without further data processing. However, the methods that were used originally to obtain and interpret these data varied. For example, the Australian, Portuguese and Chinese (Hong Kong) studies (nos. 6, 8 and 17 in Table 7.3) made direct measurements of kinetic energy with a distrometer, whereas other studies calculated kinetic energy from drop size distribution measurements made in various ways (nos. 3, 4, 5 and 16). Where corrections for altitude had been made in the original study or the effect of altitude had been measured directly using distrometers, the data were normalised to represent standard conditions using a correction factor of 1% per 100 m of rise in elevation (cf. Section 7.2). In cases where an exponential equation (Eq. [7.7]) had yet to be derived, the data presented in the original study often needed to be re-worked. However, the original, individual drop size distribution or kinetic energy measurements themselves were not always listed. Sometimes the data had to be read from graphs (nos. 7 and 14), although in the latter case the graph only showed data pooled by rainfall intensity class rather than individual measurements. In two other studies, only median values of drop size distributions were reported (nos. 1 and 15) and assumptions were needed about the relationship between D50 and De (Table 7.3). When measurements of drop size or kinetic energy had been made at relatively low rainfall intensities only (nos. 9 and 14), a wide range of values for the coefficient emax was possible and an assumption had to be made about its value to constrain the equation. In accordance with the suggestions of Rosewell (1986) and Brown and Foster (1987), a value of 29 J m-2 mm-1 was assumed in these cases. Finally, in one study (no. 13) only a curve linking rainfall intensity to kinetic energy was presented, whereas additional curves were extracted from graphs presented in Hudson (1971; nos. 10, 11 and 19). The respective curves were digitised and exponential equations fitted to them (Table 7.3).

116

CHAPTER 7 - RAINFALL INTENSITY - KINETIC ENERGY RELATIONSHIPS

Location and characteristics No. Location 1 Washington D.C., U.S.A. 2 Mississippi / Louisiana, U.S.A. 3 Holly Springs, Mississ., U.S.A. 4 Miami, Florida, U.S.A. 5 Mazowe, Zimbabwe 6a Brisbane, Qld., Australia 6b Gunnedah, NSW, Australia 7a Franklin, N-Car., U.S.A. 7b Majuro, Marshall Islands 7c Pina Range / Ft. Sherman, Panama 7d Island Beach, New Jersey, U.S.A. 7e Bogor, Java, Indonesia 8 Mrtola, Algarve, Portugal 9 Ottawa, Ont., Canada 10 Japan 11 Trinidad 12 Firenze, Central Italy 13 Southeast Arizona, U.S.A. 14 Gembloux, Belgium 15 Reserva Ducke, Manaus, Brazil 16 Malangbong, West Java, Indonesia 17 Hong Kong 18a,b Honolulu, Hawaii, U.S.A. 19a,b Poona, Maharashtra, India
a) b)

Study details (see Table 7.4) M Reference M F Laws and Parsons (1943) F Carter et al. (1974) F Brown and Foster (1987) P Kinnell (1980) F Kinnell (1980) D Rosewell (1986)

Method of analysis M Remarks A R D50 converted to Dea pooled data from R two locationsb data of McGregor O and Mutchler (1976) O original data from O Hudson (1961) O r2 relates to intensity class averages only

Alt. Clim (m) ate 8 Cf 180Cf 18 180 Cf 3 Cf 1230 Cw 25 Cf 305 Cf 1360 Cf 3 Af 50 Af 3 Cf 260 Af 21 Cs 75 Df ? Cf ? Af 90 Cs 900BS 1800 150 Cf 35 Af 670 Af 50 Cw 670Af 2250 570 BS

P McIsaac (1990)

data collected by Mueller and Sims R (1967-1968); r2 relates to intensity class averages only.

D Countiho and Toms (1995) O S Marshall and Palmer (1948) C ? Mihara (in Hudson, 1971) ? Ker (in Hudson, 1971) F Zanchi and Torri (1980) D Tracy et al. (1984) S Bolline et al. (1984) S Brandt (1988) S authors, unpublished Jayawardena and Rezaur D (2000) S Blanchard (1953) ? Kelkar (in Hudson, 1971) C C C max. eK assumed c C R R D50 converted to Ded O O storm averaged values used measurements on R volumetric basis e differentiated into C storm type f

c)

d) e)

f)

Using D50-De relation based on five reported drop distributions (Eq.[7.1], Table 7.1) Drop size distribution for two locations (Baton Rouge, Louisiana and Holly Springs, Mississippi) listed as average for 13 rainfall intensity classes. Therefore r2 unknown, number of observations in each class was used as a weighing factor. Data presented as power-law relationships between distribution percentiles and rainfall rate. Maximum eK was assumed to be reached at R=76 mm h-1 Using Eq. [7.5] with all empirical parameter combinations listed in Table 7.2 to calculate De values from D50 Measurements made on a volumetric basis in mid-air were converted to distributions arriving at the ground (cf. Uijlenhoet and Stricker, 1999); differentiated into measurements made within or just below clouds (17a) and measurements in orogaphic rainfall well below the clouds (17b). Measurements differentiated into storms in general (18a) and thunderstorms (18b)

Table 7.3. Details of the studies and methods of data analysis used in the present review. (Climates according to Kppen classification, approximate altitudes estimated in some cases. Measurement methods (MM): Fflour pellet technique; Pphotographical methods; Ddistrometer; Sstain or filter paper method. Method of data analysis (MA): Ooriginal equations; R (re-)analysed data; Cequation fitted to curve)

117

CHAPTER 7 - RAINFALL INTENSITY - KINETIC ENERGY RELATIONSHIPS Next, the various studies were sorted into three categories. Studies with data sets consisting of more than 75 R-eK data pairs and including at least four measurements exceeding a (somewhat arbitrarily chosen) rainfall intensity of 80 mm h-1 were considered to have yielded accurate coefficient values. Data sets that did not meet these criteria or for which the required information was not available were pooled in a second group for which the accuracy of derived equation coefficients was deemed uncertain. Finally, three studies in which the characteristics of different types of rainfall were examined separately (nos. 18 and 19) were placed together in a third group.

7.3.2. Development and testing of a general equation for the prediction of storm kinetic energy Values for emax, a and b that were considered to be accurate were obtained for data collected at 12 different locations, presented in seven separate papers (see Tables 7.3 and 7.4). These values were averaged to derive a single general predictive relationship having the exponential form of Eq. [7.7]. The applicability of this general equation was tested in two ways. First, it was used to predict the kinetic energy of individual storms for which direct measurements of rainfall intensity and kinetic energy were available at Gunnedah, New South Wales, Australia, as reported by Rosewell (1983, 1986). Both rainfall kinetic energy and intensity at Gunnedah (study no. 6b in Table 7.3) were measured at 1-minute intervals using a Joss-Waldvogel distrometer. Comparison with measurements using a tipping bucket raingauge (0.2 mm nominal rainfall depth per tip) suggested that on average, rainfall intensity measurements by the distrometer were within 4% of the tipping bucket measurements (Rosewell, 1983). From the data set, 24 storms were selected at random, representing 3767 individual measurements with a total duration of 63 hours and 332 mm of rainfall. This constituted 29% and 25% of the entire data set in terms of rainfall duration and depth, respectively. Further details of storm characteristics are given in Section 7.4.2. Secondly, it was investigated to what extent storm kinetic energy predictions by the newly developed general equation differed from estimates produced by the equations of the same exponential form derived for the individual locations. In addition, predictions by the general equation were compared with estimates obtained using the USLE approach of Wischmeier and Smith (1978) and the equation of Brown and Foster (1987) suggested by Renard et al. (1997) for use in the Revised USLE. Values of the coefficients a and b for the latter equation were derived from measurements made in Mississippi, U.S.A. (study no. 3 in Table 7.3), with a preset value for maximum kinetic energy contents (emax=29 J m2 mm-1), based on a comparison of earlier studies (i.e., nos. 1, 2, 4, 5 and 6). The same 24 storms recorded at Gunnedah were used throughout the analysis. A correction was made for site elevation (ca. 300 m a.s.l.) by increasing the calculated storm energy values by 3% (Section 7.2). The results of the respective predictions were compared with the direct measurements of storm energy as made with the distrometer (Rosewell, 1986).

118

CHAPTER 7 - RAINFALL INTENSITY - KINETIC ENERGY RELATIONSHIPS 7.4. Rainfall intensity - kinetic energy relationships 7.4.1. Variation in rainfall intensity-kinetic energy relationships between different studies The characteristics of the exponential equations that were fitted to the results of the 19 studies under consideration are listed in Table 7.4. When individual data points were used to derive the equations, the coefficient of determination (r2) has been added, whereas in the case of fitting an equation to a curve, the maximum relative difference (in %) between the two curves within the measured range of rainfall intensities is given instead (Table 7.4). Arguably, among the best data sets available are those collected with a high-speed camera at five locations around the world (nos. 7a-e; McIsaac, 1990) and the measurements made by Rosewell (1986) and Coutinho and Toms (1995), using distrometers in Australia (no. 6) and Portugal (no. 8), respectively. Other studies that were considered to have yielded accurate parameter estimates include those of Kinnell (1980) in Florida using photography (no. 4), of Carter et al. (1974) and Brown and Foster (1987) in the Southern United States using the flour pellet method (nos. 2 and 3, excluding the preset value for emax) and, to a lesser extent, the measurements made by Laws and Parsons (1943) using the flour pellet technique at Washington D.C. (no. 1). The average values (and their standard deviations) of the coefficients derived for the respective data sets were determined separately for the groups of good and uncertain parameter accuracy (Table 7.4). The average maximum kinetic energy contents, emax, for the first group was 28.3 J m-2 mm-1 ("2.9 J m-2 mm-1 or 10%). At nine out of twelve locations, emax was between 26.4 and 29.3 J m-2 mm-1. A single distinctively higher value of 35.9 J m-2 mm-1 was obtained in Southern Portugal, whereas somewhat lower values were reported for North Carolina (no. 7a) and New Jersey (no. 7d) at 24.6 and 25.1 J m-2 mm-1, respectively (McIsaac, 1990). The average value for emax derived from ten additional studies with less certain accuracy was not significantly different (29.9"4.3 J m-2 mm-1) and also showed a similar range of values (23.7-36.8 J m-2 mm-1). For this group, at four out of ten locations did emax differ by more than one standard deviation from the average of the emax values found for comprehensive data sets (Table 7.4). The coefficient a, together with emax, determines the value of minimal kinetic energy contents (Section 7.2, Eq. [7.7]). The variation in derived emin values proved considerably greater than that for maximum values (Table 7.4). Accurate values ranged from 6.7 J m2 mm-1 in the Central Southern U.S.A. (no. 2) to 21.1 J m-2 mm-1 in Florida (no. 4). The average value for the coefficient a for the first group was 0.52 ("0.15 or 29%), corresponding to an emin of 13.5 J m-2 mm-1 ("4.3 or 32%). The average values of a and emin for the second (lower accuracy) group were 0.62 ("0.14 or 23%) and 11.1 J m-2 mm-1 ("3.7 or 33%), respectively. Values of minimum kinetic energy derived for studies of the second group were within the cited extremes, with the exception of the studies in Zimbabwe and Canada, which yielded values of 3.2 and 7.5 J m-2 mm-1, respectively. The coefficient b determines the general shape of the curve between minimum and maximum kinetic energy contents. The average value of b based on the twelve most comprehensive data sets was 0.042 ("0.020 or 48%) and 0.044 ("0.013 or 30%) for the group of less comprehensive studies (see Table 7.3). Low values of b, indicating a gradual transition from minimum to maximum kinetic energy levels, were found for Florida (no. 4) and the Marshall Islands (no. 7b) (0.018 and 0.011, respectively). A low value of 0.019 was also inferred from a curve based on unspecified measurements made in Japan (no. 10). Higher than average values of b, indicating a relatively fast increase of

119

CHAPTER 7 - RAINFALL INTENSITY - KINETIC ENERGY RELATIONSHIPS

Location

Approx. altitude (m)

Coefficients of Eq. [7.7] emax (J m-2 mm-1) 28.9 28.0 (29) 29.3 26.4 28.1 24.6 29.2 28.8 25.1 26.8 35.9 a b emin (J m-2 mm-1) 13.3 6.7 8.1 21.1 8.7 11.6 13.3 14.3 15.8 15.1 19.0 15.8

r2 / max. % diff.

M R (mm h-1) Aa

Comprehensive data sets Washington D.C., U.S.A. (1) Southern U.S.A. (2) Mississippi, U.S.A. (3) Florida, U.S.A. (4) Queensland, Australia (6a) New South Wales, Australia (6b) North Carolina, U.S.A. (7a) Marshall Islands (7b) Panama (7c) New Jersey, U.S.A. (7d) Lowland W-Java, Indonesia (7e) South Portugal (8) Average values used in Eq. [7.8] (" standard deviation) Unknown coefficient accuracy Ontario, Canada (9) Japan (10) Trinidad (11) Zimbabwe (5) Central Italy (12) Arizona, U.S.A. (13) Belgium (14) Amazonia, Brazil (15) Upland West Java, Indonesia (16) Hong Kong (17) Average values (" standard deviation)

8 180 / 18 180 3 25 305 1360 3 50 3 260 21

0.54 0.76 0.72 0.28 0.67 0.60 0.46 0.51 0.45 0.40 0.29 0.56

0.059 0.090 0.050 0.018 0.035 0.040 0.037 0.011 0.033 0.045 0.049 0.034

0.49 ? ? 0.50 0.98 0.94 0.85 0.97 0.96 0.86 0.66 0.64

95 496 315 200 6,360 13,438 33,871 19,000 18,021 22,393 13,371 8,190

0.4-114 <260 0-372 2-309 1-161 1-146 2-193 2-170 0.5-194 14-148 13-180 4-103

R R O O O O R R R R R O

28.3 0.52 0.042 13.5 ("2.9) ("0.15) ("0.020) ("4.3) 75 ? ? 1230 90 900/1800 150 35 670 50 (29) 23.7 24.7 29.2 31.1 33.6 (29) 30 30.4 36.8 0.74 0.51 0.37 0.89 0.59 0.55 0.60 0.56 0.69 0.69 0.039 0.019 0.030 0.048 0.044 0.052 0.061 0.044 0.060 0.038 7.5 11.6 15.6 3.2 12.8 15.1 11.6 13 9.4 11.4 5% 2% 6% 0.71 8% 5% 0.63 0.47 0.73 0.75 ? ? ? 50 ? ? 89 92 21 18 <23 8-135 1-155 19-229 1-140 <76 0.3-39 0-105 0.3-124 12-120 C C C O C C R R O O

29.9 0.62 0.044 11.1 ("4.3) ("0.14) ("0.013) ("3.7) R R C C

Differentiated data 0.3 cloud- Hawaii, U.S.A. (18) 670-2250 12.7 0.98 0.011 0.66 64 0-25 6.9 orographic- Hawaii, U.S.A. (18) 670-2250 23.7 0.71 0.031 0.88 29 0-127 general - Central India (19) 570 20.2 0.50 0.067 10.1 2% ? 4-40 thunder - Central India (19) 570 35.8 0.60 0.022 14.3 5% ? 4-100 a MAMethod of data analysis: Ooriginal equations; R (re-)analysed data; Cequation fitted to curve.

Table 7.4. Summary of coefficients for the exponential relationships (cf. Eq. [7.7]) fitted to selected R-eK data identified in the literature. Values of emax between brackets were not derived by regression but assumed a priori and not used in the calculation of average values and standard deviations. Values printed in bold deviate from the respective average values for the comprehensive data set group by more than one standard deviation. Also listed are number (N) and rainfall intensity range (R) of measurements; further details are given in Table 7.3.

kinetic energy contents with rainfall intensity, were found in parts of the southern U.S.A. (up to 0.090, no. 2) and, to a lesser extent, in Washington D.C. (0.059, no. 1), Belgium (0.061, no. 14) and upland West-Java (0.060, no. 16).

120

CHAPTER 7 - RAINFALL INTENSITY - KINETIC ENERGY RELATIONSHIPS

Storm size class

Rainfall characteristics N Total Avg P R


mm mmh-1

Observed Avg storm EK


J m-2

Predicted (Rosewell, 1986) Total Avg. Total EK storm EK EK


J m-2 J m-2 J m-2

Storm based difference Abs. Rel.

Cumulative difference Class Perc. total of total


J m-2 %

J m-2

0-2.5 mm 2.5-5 mm 5-25 mm 25-50 mm >50 mm Total

10 2 8 3 1 24

12.4 6.2 104.5 115.6 93.0 331.7

3.1 4.7 6.2 4.4 6.3 5.3

31 79 322 730 1770 292

311 158 2580 2190 1770 7008

20 51 280 729 1798 272

202 102 2241 2187 1798 6529

11 28 43 64 28 30

25% 36% 13% 8% 2% 19%

-109 -56 -339 -3 28 -479

-2% -1% -5% 0% 0% -7%

Table. 7.5. Characteristics of 24 storms at Gunnedah, New South Wales, Australia that were used for testing the performance of the predictive equation derived for the different locations in Table 7.3. Listed are the number of storms (N) in each size class and the corresponding total rainfall depth (P), rainfall intensity (R), observed and predicted kinetic energy (EK) and the difference between these.

7.4.2. Testing of a general predictive equation of rainfall kinetic energy Based on the average parameter values that were derived in the previous section for the sites with the best data sets, the following general equation to predict storm kinetic energy from rainfall intensity data is proposed (in metric units):

e K = 28.3[ 0.52 exp( 0.042 R )] 1

[7.8]

This equation is rather similar to that obtained by Rosewell (1986) (no. 6b, Table 7.4) for Gunnedah, Southeast Australia on the basis of the comprehensive data set from which 24 storms were selected to evaluate the various equations obtained for the respective sites (see below). The basic characteristics of these 24 storms in terms of measured rainfall intensity and kinetic energy are listed in Table 7.5. The selected storms varied in depth between 0.3 and 93 mm, with average rainfall intensities between 0.8 and 52 mm h-1. Measured storm energy ranged between 4 and 1770 J m-2, with a total of 7008 J m-2 for all 24 storms. In addition, estimates of storm EK using the equation of Rosewell (1986; no. 6b) are listed and compared with measured amounts. From Table 7.5, a number of observations can be made. First, it appears that differences between predicted and measured storm EK can be considerable. The relative difference generally decreases with storm depth, from an average "26-36% for storms of less than 5 mm, to "8% for storms of 25-50 mm and only 2% for the largest storm (93 mm; Table 7.5). Apparently, considerable natural variation exists between individual storms, which cannot be accounted for by variations in rainfall intensity alone. Different synoptic conditions and wind effects may well be responsible for most of this variability, as will be discussed in Section 7.6.4. Interestingly, the decrease of the relative prediction
121

CHAPTER 7 - RAINFALL INTENSITY - KINETIC ENERGY RELATIONSHIPS error was caused by the fact that the absolute difference between measured and predicted storm kinetic energy did not increase as much as storm size itself, from an average "11 J m-2 for storms less than 2.5 mm, to "64 J m-2 for storms of 25-50 mm. These findings suggest that the relative accuracy of storm kinetic energy predictions is greatest for large storms, which is good news in terms of erosion hazard prediction and modelling as these large storms are likely to contribute most to total soil loss. The largest absolute difference between measured and predicted EK occurred for a storm of 21.8 mm, for which the predicted energy was lower by 126 J m-2 (27%) than the measured amount (462 J m-2). Overall, predicted kinetic energy was 7% less than the observed EK for the 24 storms. The largest part of this under-estimation, i.e. 5% of overall kinetic energy, was related to storms of 5-25 mm, although these made up less than a third of total rainfall (Table 7.5). The Rosewell (1986) equation used here to predict storm kinetic energy was derived from a data set that was larger than the 24 storms that were used in the present study. It may be assumed that the overall predicted EK for the entire data set would be nearly equal to the measured amount. The total kinetic energy for the 24 storms at Gunnedah as predicted by the equations derived for the different locations listed in Table 7.3 was compared with the measured amounts as well as to the predictions made by the general equation proposed in this study (Eq. [7.8]). The same was done for the predictions by the USLE approach of Wischmeier and Smith (1978). In Table 7.6, the differences between predicted and observed total EK are listed for the equations derived from the respective studies. In addition, in Table 7.7 the average difference per storm is listed for selected equations, together with the coefficient of determination of the relationship between estimated and observed EK, and the difference in r2-values, compared with the r2 associated with the original equation of Rosewell (1986). Most equations produced total kinetic energy estimates that were similar to the measured amount (Table 7.6). Fourteen out of the 23 equations predicted total kinetic energy within 10% of the measurements. These included predictions by the USLE approach (Wischmeier and Smith, 1978; +5%), the Revised USLE equation (Brown and Foster, 1987; -8%), the general equation proposed in this study (-0.3%) and that of Rosewell (1986; -7%). Equations under-estimating total kinetic energy by more than 10% include those developed for Japan, Zimbabwe, Queensland, the Marshall Islands and Canada, whereas the ones developed for Hong Kong, lowland West Java (no. 7e), Florida, Portugal and Arizona over-estimated kinetic energy by more than 10% (Table 7.6). Surprisingly, the general equation proposed in this study predicted a total energy amount that was almost equal to the measured amount (Table 7.7). This fact should not be overvalued given the indicated variability between storms. It seems likely that Rosewells equation would have given a better prediction if the entire data set had been used. Nevertheless, the fact that the new (Eq. [7.8]) and original equation (no. 6b) show very similar coefficients of determination (r2=0.98-0.99, respectively) and average differences between observed and estimated storm EK ("17% and "19%, respectively), suggests that the proposed general equation predicts storm kinetic energy with very similar accuracy as the original site equation of Rosewell (1986) (Table 7.7). The exponential RUSLE equation (Brown and Foster, 1987) also produces reasonable estimates, although the average difference in storm EK is somewhat higher at 24%. The equation suggested by Wischmeier and Smith (1978) produces estimates that are similar, albeit slightly lower, than the equation based on the data of Laws and Parsons (1943; Table 7.6). This similarity was expected, because the former authors used the same data to derive their equation. However, there appears to be a significant difference between
122

CHAPTER 7 - RAINFALL INTENSITY - KINETIC ENERGY RELATIONSHIPS

Location / equation Japan (10) Zimbabwe (5) Queensland (6b) Marshall Islands (7b) Canada (9) N-Carolina (7a) Mississippi (RUSLE) (3) New South Wales (6b) Trinidad (11) New Jersey (7d) Southern U.S.A. (2) General equation (Eq. [7.8]) Malangbong, West-Java (16) Belgium (14) Panama (7c) Brazil (15) USLE (Eq. [7.6]) Italy (12) Washington D.C. (1) Hong Kong (17) Bogor, West-Java (7e) Florida (4) Portugal (8) SE-Arizona (13)

Rel. difference with total measured EK (%) -26 -20 -20 -16 -15 -10 -8 -7 -6 -3 -1 0.3 3 4 4 4 5 6 7 12 12 15 19 21

Table 7.6. Relative agreement between measured storm energy during 24 storms at Gunnedah, New South Wales, Australia and predicted values using the parameterisations of Eq. [7.7] and listed in Table 7.4. Equations in italics represent the original equation derived for the site (Rosewell, 1986), as well as the general equations proposed by this study (Eq. [7.8]) the Revised USLE (Renard et al., 1997) and the USLE (Wischmeier and Smith, 1978) (Table 7.7).

predictions by the older USLE approach and the more recent RUSLE equation; the former producing an estimate of total kinetic energy that was 13% higher than predicted by the latter. Although the magnitude of this difference can be expected to increase with overall rainfall intensity, it agrees well with the findings of Yu (1998). He compared estimates of the rainfall erosivity factor R (which includes kinetic energy) for the tropical region of Australia using both equations and found that the RUSLE equation yielded a factor that was 5.7-18% higher than the value resulting from the USLE approach. When the general equation proposed in this study is used, an intermediate value is likely to result. As mentioned, most other equations produced overall EK estimates that were within 10% of both measured kinetic energy and predictions by the general equation. Some of the exceptions may well be related to dominating rainfall generation mechanisms and this possibility will be discussed in Section 7.6.4.

123

CHAPTER 7 - RAINFALL INTENSITY - KINETIC ENERGY RELATIONSHIPS

Equation

Agreement between predicted and measured EK Total diff. (%) Average diff. per storm (%) r2 Rel.diff. with best r2 (%) -0.8 0.0 -0.5 -2.0

RUSLE (3) Original site equation (6b) Universal equation (Eq. [7.8]) USLE (Eq. [7.7])

-8.1 -6.8 -0.3 5.0

24 19 17 15

0.98 0.99 0.98 0.97

Table 7.7. Indicators of the performance of several equations for the prediction of the kinetic energy of 24 storms at Gunnedah, New South Wales, Australia. Tested equations include the original equation derived for the site (Rosewell, 1986, no. 6 in Table 7.4), the universal equations proposed in this study (Eq. [7.8]), the Revised USLE (Renard et al., 1997; no. 3 in Table 7.4) and the older USLE (Wischmeier and Smith, 1978; Eq. [7.7]) equations.

7.5. Discussion: causes of the observed differences between locations A number of causes may underlie the differences in parameter values obtained for the various studies listed in Table 7.4. These may be categorised as being related to differences in measuring techniques and procedures, limited and/or biased sampling, errors or uncertainties in the interpretation of the data and, finally, to actual differences in raindrop size distributions associated with different synoptic conditions and rainfall generating mechanisms. Below, these various sources of discrepancy are discussed in more detail.

7.5.1. Measurement technique and procedure Different authors have used different methods to measure drop size distributions or kinetic energy (Table 7.3). Every method has its own drawbacks and sources of error and this will undoubtedly have caused some of the discrepancies between studies. For instance, Hall (1970) pointed out that the calibration of the filter paper method is not as straightforward as is often assumed; also, the results of such measurements may be affected by the production of small droplets by splashing large drops, as well as by changes in the moisture content of the medium used. On the other hand, distrometer measurements can be seriously affected by wind (Kinnell, 1973; Illingworth and Stevens, 1987). Needless to say, Eq. [7.2] refers to terminal velocities in stable air. The actual striking velocity of raindrops is increased by side wind, whereas turbulence may slow raindrops down (Rowland, 1976; Kinnell, 1976). Joss and Waldvogel (1977) inferred a deviation of 10% from terminal velocity for natural rain at wind speeds of 6 m s-1 at 2 m above the ground, whereas the observations of Donnadieu (1980) and Stow and Jones (1981) suggested similar changes occur already at much lower wind speeds, the effect of side wind also depending on drop size. Rosewell (1983) reported that the kinetic energy of rain falling at an intensity of 100 mm h-1 in Southeast Australia was sometimes increased by as much as 39% because of wind.

124

CHAPTER 7 - RAINFALL INTENSITY - KINETIC ENERGY RELATIONSHIPS 7.5.2. Limited or biased sampling Some of the studies analysed in this review included only limited observations under conditions of high rainfall intensities, whereas the study in Zimbabwe (no. 5) was limited to intensities higher than 19 mm h-1. Some other studies were based on a relatively small number of observations. In particular, when arduous methods such as the filter paper or flour pellet techniques are used, the number of observations for individual storms, as well as on aggregate, is often low and this may affect the results. For example, the value of maximum kinetic energy contents is strongly dependent on observations made during high rainfall intensities and if these are few, they will exert a correspondingly strong influence on the resulting equation. The studies conducted in Canada (no. 9 in Table 7.4), Arizona U.S.A. (no. 13) and Belgium (no. 14) provide a case in point. Similarly, the coefficient a, and therefore the value of minimum kinetic energy contents, is strongly influenced by observations made at low rainfall intensities. The low minimum value of 3.2 J m-2 mm-1 derived for Zimbabwe (no. 5) probably results from the fact that rainfall intensities below 19 mm h-1 were not sampled in this case (Hudson, 1961). In addition to a limited sample range, sampling strategy may also influence the results if measurements are not made throughout a storm. Biased sampling, e.g. by concentrating observations at the beginning or end of a storm, can have a profound effect because of the hysteresis in drop size distribution that occurs during many storms or cloudbursts within storms. Naturally, this phenomenon is difficult to demonstrate with only a limited number of observations, but it becomes apparent when (semi-)continuous measurements are made. An example from Gunnedah, Southeast Australia is shown in Fig. 7.4, which represents some of the data collected by C.J. Rosewell that have been used for evaluating the various equations in this study (Section 7.5.2). The general pattern that becomes apparent from Fig. 7.4 is that of kinetic energy contents being higher than predicted by the site equation (no. 6b) during the beginning of a storm or cloudburst, and lower than average values at the end of the storm. Although Fig. 7.4 represents a particularly clear example of this hysteresis effect, similar patterns were found during many other storms and cloudbursts. Indeed, this phenomenon is often seen in the form of a few large drops heralding the onset of a storm. While wind may play a role in creating this effect, its main reason is probably related to the fact that larger drops tend to fall faster than smaller ones, which result in a sorting of raindrop sizes with progressively smaller drops falling in the course of the storm (Blanchard, 1953).

7.5.3. Method of interpretation The way in which exponential curves were fitted to various literature data or curves during the present exercise may also have had an influence on the resulting parameter values listed in Table 7.4. Among the best data sets this may apply in particular to the study conducted at Washington D.C. (no. 1), which only presented a graph of rainfall intensity versus median drop size. The median drop sizes were converted to kinetic energy contents on the basis of five additional graphs showing (presumably idealised) drop size distributions (Laws and Parsons, 1943). A comparable problem occurred with the data collected by Carter et al. (1974) in Louisiana and Mississippi (no. 2). Their data were presented as averages for different rainfall intensity classes and pertained to two different locations. Similarly, the data reported by McIsaac (1990) for a range of locations (nos. 7a-e) only represented average kinetic energies of data pooled into rainfall
125

CHAPTER 7 - RAINFALL INTENSITY - KINETIC ENERGY RELATIONSHIPS

35

(a)
30

mm ) e K (J m

25 20 15 10 5 0 0 20 40 60
-1

-2

-1

80

R (mm h )
80 (b) R 40 60 30 40 20 eK 20 10 e K (J m -2 mm -1 ) R (mm h -1 ) 50

0 17:55

18:02

18:10 Time (hrs)

18:17

0 18:24

Fig. 7.4. Example of hysteresis in the rainfall intensity - kinetic energy relationship at Gunnedah, New South Wales, Australia on 3 April 1977. (a) Measurements of kinetic energy contents (eK) versus rainfall intensity (R) (open circles) and values predicted by the equation of Rosewell (1986; line), (b) course of rainfall intensity and kinetic energy load during the onset of the storm.

intensity classes. Similar problems occurred with some other studies (see Table 7.3 for details). In those cases where equations had to be fitted to digitised curves, the scatter and range of the original observations remained unknown and this must have introduced a certain error as well. Finally, values for maximum kinetic energy had to be assumed for the data collected in Canada (no. 9) and Belgium (no. 14) because the maximum observed rainfall intensities for these locations were too low to enable an estimation of emax to be made within acceptable margins of error during curve fitting (Section 7.4.1).

126

CHAPTER 7 - RAINFALL INTENSITY - KINETIC ENERGY RELATIONSHIPS Brown and Foster (1987) had sufficient data for rainfall intensities up to ca. 125 mm h-1, but enforced an emax value of 29 J m2 mm-1, which was based on a comparison of earlier studies (nos. 1, 2, 4, 5 and 6 in Table 7.4).

7.5.4. Storm type Despite the errors introduced during measurements and interpretation as discussed in the previous sections, some of the observed differences in R-eK relationships between locations are caused by real differences in rainfall characteristics. For instance, Rosewell (1986) used the same electromechanical distrometer to measure rainfall kinetic energy at two locations in eastern Australia (nos. 6a and 6b) and obtained two statistically different relationships. Similarly, McIsaac (1990) reported drop size distributions measured at various locations around the world using the same high-speed camera and also found notable differences between locations (nos. 7a-7e). Blanchard (1953) published drop size distributions for a range of altitudes up to 2250 m a.s.l. in Hawaii. Three different rainfall types were distinguished: drizzle within clouds, drizzle just below clouds and thunderstorms of orographic origin. The corresponding drop size distributions, converted to kinetic energy contents under standard conditions (using Eq. [7.4]), are shown in Fig. 7.5a. There appears to be a clear contrast in the kinetic energy contents associated with orographic storms and those associated with drizzle. An important reason for this is the fact that many of the smaller droplets that constitute drizzle evaporate as they fall over progressively larger distances, leaving larger drops and so decreasing rainfall intensity as well as increasing kinetic energy contents (Blanchard, 1953). Similarly, Kobayashi (1955, cited in Mason and Andrews, 1960), reported notable differences between drop size distributions associated with individual (convective) showers, continuous warm frontal rain and drizzle from low stratus clouds in Japan, whereas Mason and Andrews (1960) themselves found median volume drop diameters to increase when going from rainfall associated with warm fronts via coalescence showers to thunderstorms. Kinnell (1973) was also able to separate drop size distributions at Island Beach, New Jersey, U.S.A. during storms associated with pre-cold fronts, air masses (i.e. convection) and (cold-type) occlusions (Fig. 7.5b). In general, at a particular rainfall intensity, kinetic energy also increased in this order, although the single highest value for kinetic energy (41 J m-2 mm-1) was associated with pre-cold frontal rain (Fig. 7.5b). Zanchi and Torri (1980) suggested that rainfall drop size distribution was also positively related to air temperature. They obtained an unknown number of raindrop size distributions at varying rainfall intensities and air temperatures using the flour pellet technique in Central Italy and found that median drop size could be expressed as a power law function of temperature. For example, their equation suggested a median drop size of 2.6 mm at an intensity of 50 mm h-1 when air temperature was 10EC, increasing to 3.1 mm at 20EC (Table 7.1). However, this temperature effect may also reflect differences in storm type, rather than air temperature alone. For example, rainfall in summer may well have fallen as thunderstorms in many cases, which are associated with larger drop sizes. No synoptic conditions were given by Zanchi and Torri (1980). Overall, it appears that the raindrop size distribution at a particular rainfall intensity decreases with altitude and increases when going from warm frontal rain and drizzle to more energetic rainfall associated with cold fronts and thunderstorms. Taking the kinetic energy data presented in Table 7.6 at face value, the largest differences appear to be found between measurements made at locations experiencing a strong oceanic influence
127

CHAPTER 7 - RAINFALL INTENSITY - KINETIC ENERGY RELATIONSHIPS

30

(a)
25

e K (Jm mm )

-1 -2

20 15 10 5 0 0 10 20 30
-1

drizzle just below clouds drizzle in clouds orographic rainfall 40

R (mm h )
50

40

e K (Jm mm )

30

-2

-1

20

10

air mass (convective) pre-cold front cold-type occlusion


0 20 40
-1

0 60 80 100

R (mm h )

Fig. 7.5. Examples of the effect of storm type on rainfall kinetic energy contents based on measurements at (a) Honolulu, Hawaii, with lines representing equations fitted to measurements in drizzle and orographic rainfall (no. 17a, b in Table 7.4; data from Blanchard, 1953) and (b) Island Beach, New Jersey, U.S.A. (data from Kinnell, 1973), with the line representing the R-EK relationship found for this location (no. 7d in Table 7.4).

on the one hand, and at continental (especially sub-humid) locations on the other. This contrast presumably reflects a difference in synoptic conditions and rainfall generating mechanisms. For example, a value of 26.4 J m-2 mm-1 was found for the maximum kinetic energy emax at Brisbane on the tropical east coast of Australia (no. 6a in Table 7.4). Rosewell (1986) ascribed the difference with the value of 28.2 J m-2 mm-1 (converted to standard conditions) determined for Gunnedah, situated more than 200 km from the coast behind a mountain range (no. 6b), to the oceanic influence prevailing at

128

CHAPTER 7 - RAINFALL INTENSITY - KINETIC ENERGY RELATIONSHIPS Brisbane, which was thought to result in lower maximum temperatures and consequently less energetic storm cells than at Gunnedah. Even lower values (24.6 and 25.1 J m-2 mm-1) were found in North Carolina and New Jersey, respectively (McIsaac, 1990; nos. 7a and 7d in Table 7.4). Similarly, low maximum kinetic energy values were found in Japan (23.7 J m-2 mm-1 ; no. 10) and Trinidad (24.7 J m-2 mm-1; no. 11). The exact locations of these studies are unknown (Table 7.3) but they may very well represent coastal locations. Drop sizes often appear smaller at high altitudes as well (McIsaac, 1990; Blanchard, 1953), although the effect on kinetic energy is offset to a certain extent by the greater fall velocities associated with higher elevations (Beard, 1977). At the other end of the scale, sub-humid to semi-arid continental locations may experience more frequent storms with relatively high energy contents during which part of the precipitation may even fall as (melted) hail (Rosewell, 1986). For example, high maximum kinetic energy values have been reported for Southern Portugal (35.9 J m-2 mm-1) and Arizona (33.6 J m-2 mm-1). In the former study, which was conducted during a period of prolonged drought, most rainfall (in particular that falling at high intensities) was associated with non-frontal convergence of air into low pressure points (Coutinho and Toms, 1995). The high estimates of EK by the equations derived for lowland West-Java (no. 7e) and subtropical humid Florida are primarily related to the high kinetic energy contents predicted at low rainfall intensities (Table 7.4). Other humid tropical locations, both oceanic (upland West Java (no.16), Panama and Trinidad) as well as continental (Brazil), do not demonstrate this phenomenon and resulted in more moderate EK estimates. The measurement and interpretation methods of the Florida study in particular seem to be sound and the cause of the discrepancy is unclear. The study by Kelkar in Central India (cited in Hudson, 1971) provides further evidence of the larger drop sizes associated with thunderstorms, compared to less energetic rainfall generation mechanisms (Table 7.4). From Kelkar's study, a maximum kinetic energy of 35.8 J m-2 mm-1 resulted for thunderstorms. Such high values come close to the single peak energy values of about 38 J m-2 mm-1 measured in Southeastern Australia (Rosewell, 1986) and sub-humid northern Nigeria (Kowal and Kassam, 1977). The 41 J m-2 mm-1 recorded on one occasion in New Jersey (Kinnell, 1973; Fig. 7.5b) does not fit the overall pattern of high emax values during thunderstorms in that this all time maximum occurred during the passage of a pre-cold front. Drops larger than 6-8 mm are unlikely to occur in natural rainfall (Laws and Parsons, 1943; Joss and Waldvogel, 1977). It follows that the absolute upper limit to kinetic energy contents, reached if all raindrops have this size, would be in the order of 42-44 J m-2 mm-1. Therefore, like the occasionally recorded extreme values exceeding 40 J m-2 mm-1 in rainfall at Gunnedah (no. 6b), the high value measured at New Jersey is more likely to be the result of (melted) hailstones (Rosewell, 1986).

7.6. Concluding remarks The present review of the literature on the relationship between rainfall intensity, drop size distribution and kinetic energy has resulted in a new general predictive equation. Although having the same general form as the continuous exponential function first proposed by Kinnell (1980) and used with different coefficient values in RUSLE technology (Renard et al., 1997), its parameterisation is based on data from a range of locations for which good data sets are available. The derived coefficients also compared
129

CHAPTER 7 - RAINFALL INTENSITY - KINETIC ENERGY RELATIONSHIPS favourably with coefficients based on less comprehensive studies in many other locations. Arguably, therefore, the proposed equation comes closer to a general equation than the equation suggested in the Revised USLE, which, like its predecessor in the USLE (Wischmeier and Smith, 1978), was partially developed from measurements made at a single location in the U.S.A. only (Brown and Foster, 1987). The predictive capacity of the individual parameterisations of the exponential equation derived from data sets from all over the world (N=21) was compared with that of the still widely used USLE approach (Wischmeier and Smith, 1978), of the Revised USLE approach (Renard et al., 1997) and of the newly developed general equation (Eq. [7.8]). To this end, 24 storms during which rainfall intensity and kinetic energy had been measured continuously, were selected from a detailed data set collected by Rosewell (1986) in Southeastern Australia. Storm kinetic energy predicted by the various equations was compared with measured amounts. It appeared that there was substantial variability in the measured kinetic energy of the 24 storms that could not be accounted for by any single R-eK equation. Reasons for this are probably partly related to synoptic conditions, whereas side wind may also have exerted an influence on the measurements. As a consequence, an average difference of 30 J m-2 occurred between observed storm energy and EK predicted by the equation related to the entire Gunnedah data set (Rosewell, 1986), while the difference for large storms was not much greater than that for small storms. Such findings suggest that storm kinetic energy can be predicted relatively accurately for large storms, which is good news for erosion modelling. Overall, equations for most locations resulted in estimates of total storm kinetic energy that were within 10% of the amounts that were measured or predicted by the newly developed general equation (Eq. [7.8]), the latter two being almost equal. Comparison with the USLE and Revised USLE confirmed the notion of Yu (1998), who found that the rainfall erosivity factor as calculated within the Revised USLE was 5.718% lower than values produced by the older USLE approach. A difference of 13% was found for the 24 sample storms that were used for equation testing in this study. The newly proposed general equation produced an intermediate result, being 8% higher and 5% lower than estimates produced by the Revised and older USLE, respectively, but coming much closer to measured values (Table 7.6). Differences of such magnitude are rather modest compared with the natural variations that were found to exist between storms and that are essentially unpredictable, at least without prior information on storm type or air mass. Also, in the context of erosion hazard prediction using the (R)USLE approach, the cited differences are much smaller than the uncertainties involved in, for example, estimation of the soil erodibility factor K (Renard et al., 1997) or those arising from the general difficulty that the (R)USLE has in describing the actual erosion processes rather than their overall approximate result. As such, a re-examination of estimates of the rainfall erosivity factor by (R)USLE does not seem warranted for most locations. However, at locations experiencing strong coastal influence, use of the (R)USLE or the proposed general equation may lead to a significant over-estimation of rainfall erosivity, whereas an under-estimation is likely to result for sub-humid to semi-arid continental locations. Similarly, it is questionable to what extent the general equation will be applicable in montane areas, where currently available rainfall erosivity maps often show the highest values (e.g. Renard et al., 1997). While the effect of altitude on kinetic energy can and should be included in the calculations, it seems that drop size distributions may well be generally smaller in these areas (Blanchard, 1953). More research on the nature of the R-eK relationship is needed under such conditions. A
130

CHAPTER 7 - RAINFALL INTENSITY - KINETIC ENERGY RELATIONSHIPS promising instrument in this respect is the optical spectropluviometer that is able to measure drop size and terminal velocity distributions simultaneously (e.g. Lavergnat and Gol, 1998; Salles and Poesen, 1999). A comparative study of the two variables at contrasting locations, involving standardised high-frequency measurements will further improve our understanding of the relationships between rainfall intensity and kinetic energy. In terms of process-based research, it appears that our knowledge of the distributions of drop size and terminal velocity in natural rainfall is well ahead of our understanding of the way in which these interact to detach and transport soil particles by splash. If rain falling at high intensities is compared with that falling at low intensities, the former appears to be considerably more effective in detaching soil than is to be expected from the difference in eK alone (Hudson, 1961; Morgan, 1977). Although results from laboratory studies of the effect of drop size and fall velocity (Sharma and Gupta, 1989; Salles and Poesen, 2000) and saturation or ponding of the soil on detachment (Sloneker et al., 1976; Torri et al., 1987) go some way to explain this phenomenon, such experiments have often been fraught with interpretational difficulties (Chapter 8). Moreover, the translation of laboratory results to field situations is not straight-forward because of the fundamental differences between the drop size distributions and fall velocities of artificial and natural rainfall. More work, preferably involving measurements of drop size and fall velocity distributions along with soil detachment, in natural rainfall and at high time resolution, is needed to elucidate the underlying processes.

131

Anda mungkin juga menyukai