Throughfall characteristics in three non-native Hawaiian forest stands

Agricultural and Forest Meteorology 150 (2010) 1453–1466

Contents lists available at ScienceDirect

Agricultural and Forest Meteorology journal homepage: www.elsevier.com/locate/agrformet

Throughfall characteristics in three non-native Hawaiian forest stands Alan Mair ∗ , Ali Fares ¯ Natural Resources and Environmental Management, University of Hawai‘i at Manoa, 1910 East-West Road, Sherman Lab 101, Honolulu, HI 96822, United States

a r t i c l e

i n f o

Article history: Received 9 July 2009 Received in revised form 26 June 2010 Accepted 20 July 2010 Keywords: Throughfall Canopy interception Non-native forest ¯ Makaha valley Psidium cattleianum

a b s t r a c t The effect of non-native trees on hydrological processes in Hawaiian forests remains largely unquantified. In this study, throughfall was determined by stationary methods over 18–23-month study periods at three locations each dominated by one of three non-native tree species including Schinus terebinthifolius, Coffea arabica, and Psidium cattleianum. The lowest mean throughfall percentage of 44% occurred under a monotypic stand of P. cattleianum. Mean throughfall under canopies dominated by C. arabica and S. terebinthifolius were 59 and 60%, respectively. Annualized throughfall rates were computed as 62, 60, and 45% under canopies dominated by S. terebinthifolius, C. arabica, and P. cattleianum, respectively. The low mean throughfall at each location is likely due to high wet-canopy evaporation combined with frequent low-intensity/low magnitude rainfall. The exceptionally low throughfall under P. cattleianum may be the result of larger amounts of intercepted rainfall being diverted to stemflow and/or trunk storage. Observations suggest that stemflow is a substantial component of the canopy water balance in stands of P. cattleianum. Plant area index (PAI) was the lowest (3.66) under P. cattleianum and highest (4.64) under C. arabica. Mean direct throughfall coefficient p was computed from individual storm events at each location and ranged from a low of 0.16 under the canopy dominated by P. cattleianum to a high of 0.31 under the canopy dominated by S. terebinthifolius. A similarly computed mean canopy storage capacity S varied from 2.5 mm under primarily S. terebinthifolius to 4.0 mm at the location dominated by C. arabica. © 2010 Elsevier B.V. All rights reserved.

1. Introduction Rainfall interception by forest canopies and its evaporation to the atmosphere is a major component of the water balance in rain forest ecosystems. Assessing the role that forests play in controlling the delivery of water to rivers, streams, and groundwater aquifers is crucial for successfully managing watersheds that provide fresh water for drinking and flood protection (Levia and Frost, 2006). Forest interception losses in maritime climates can be much greater than those arising from transpiration due to the utilization of advected energy (Calder, 1998). Several studies suggest that interception in tropical forests located at continental edges and islands is higher in maritime climates due to one or more of these factors: (1) frequent low-intensity rain, (2) advection of sensible heat from the nearby ocean, and (3) high epiphyte loading (Scatena, 1990; Cavalier et al., 1997; Dykes, 1997; Schellekens et al., 2000). An understanding of the partitioning of incident gross rainfall into throughfall is essential in areas that depend on water resources derived from maritime forests with high evaporative losses.

∗ Corresponding author. Present address: Water Resources Research Center, Uni¯ versity of Hawai‘i at Manoa, 2540 Dole Street, Holmes Hall 283, Honolulu, HI 96822, United States. E-mail address: [email protected] (A. Mair). 0168-1923/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.agrformet.2010.07.007

Throughfall is the portion of incident gross precipitation that penetrates or drips through a plant canopy, while stemflow is the portion that flows down the trunks of trees. Throughfall can be categorized into two types: free and release. Free throughfall passes directly through a plant canopy without contacting any vegetative surface, whereas release throughfall is initially intercepted and subsequently drips from the plant. The majority of water transfer from the forest canopy to soil occurs via throughfall (Levia and Frost, 2006). While stemflow does not typically represent a large portion of the annual forest water balance, it can be an important mechanism of replenishing soil moisture (Levia and Frost, 2003). Interception in forest ecosystems is a dynamic process and the proliferation of non-native tree species can increase or decrease interception and evapotranspiration (ET) loss, which can alter the availability of fresh water resources. Forest ET is comprised of both interception (i.e., evaporation) and transpiration losses. Tropical island ecosystems are especially vulnerable to non-native species invasions because of the high net resource availability and the poor ability of native species to pre-empt those resources (Denslow, 2003). Non-native trees and shrubs have already created macroscale water scarcity problems, such as streamflow decline, in arid and semi-arid regions (Zavaleta, 2000; Le Maitre et al., 2002). While invasions of non-native tree species have many ecological impacts to Hawaiian forests (Vitousek et al., 1987), they also threaten to

1454

A. Mair, A. Fares / Agricultural and Forest Meteorology 150 (2010) 1453–1466

negatively affect the hydrological services Hawaiian forests provide. In the Hawaiian Islands, groundwater aquifers supply 99% of all drinking water and 50% of all fresh water used statewide (Gingerich and Oki, 2000). Roughly 30% of annual rainfall statewide is converted into groundwater recharge with an even higher percentage of 36% on the island of O‘ahu (Lau and Mink, 2006). On O‘ahu, much of the recharge occurs in high rainfall areas such as upper ¯ ¯ Makaha valley. Streamflow in Makaha valley has declined significantly in terms of extent and volume of flow due, in part, to diversions of water, groundwater pumping, and rainfall decline (Mair, 2009). For example, the extent of perennial flow used to extend to the sea but is now confined to elevations above 424 m and mean annual streamflow since 1994 has declined 46% compared to the period from 1960 to 1993. Despite the documented effects of water resource development and climate variability on streamflow, little is known about the effects of non-native tree species ¯ on ET losses (interception, transpiration) in upper Makaha valley. The influence of non-native tree species on ET may represent an additional contributing factor to streamflow decline. Researchers have speculated about the hydrological effects of replacing native forest tree species with non-native trees in Hawai‘i but very few measurements have been made to test their hypotheses. Giambelluca et al. (2007) compared ET losses in a tropical montane cloud forest (TMCF) of native Metrosideros polymorpha with that of a site heavily invaded by Psidium cattleianum on the island of Hawai‘i. They found that ET as a function of available energy was 27% higher at the invaded site than the native site. During dry canopy periods, ET at the invaded site was 53% higher than that of the native site. Kagawa et al. (2009) compared species and stand level water use in a native old-growth M. polymorpha forest with timber plantation dominated by non-native Eucalyptus saligna and Fraxinus uhdei on the island of Hawai‘i. They found that whole-tree daily water use in E. saligna and F. uhdei was more than three times higher than M. polymorpha. Total water use in stands dominated by F. uhdei was more than twice as much as stands dominated by E. saligna and M. polymorpha; water use in stands of M. polymorpha was the lowest among the three species. Thus, these results confirm that non-native P. cattleianum, E. saligna, and F. uhdei are capable of altering the hydrological cycle of Hawaiian forests. ¯ In upper Makaha valley on the island of O‘ahu, Harman (2006) documented the presence of P. cattleianum as well as the following other non-native tree species including Aleurites moluccana, Coffea arabica, Schinus terebinthifolius, and Syzygium cumini. The highly invasive P. cattleianum was found throughout the upper valley in a variety of terrains. Monotypic stands of P. cattleianum, S. terebinthifolius, and C. arabica were found at numerous locations. Remnant native tree species including M. polymorpha, Dodonaea viscosa, Acacia koa, and Diospyros sandwicensis were interspersed with the non-native tree species (Harman, 2006). The non-native P. cattleianum and S. terebinthifolius are listed among the most invasive horticultural plants by the State Division of Forestry and Wildlife, Department of Land and Natural Resources (available at http://www.state.hi.us/dlnr/dofaw/hortweeds/specieslist.htm). Suzuki (2006) measured the plant area index (PAI), leaf angle distribution (LAD), and non-photosynthetic vegetation angle distribution (NPVAD) for three native (A. koa, D. sandwicensis, M. polymorpha) and three non-native (C. arabica, P. cattleianum, S. ¯ terebinthifolius) tree species in upper Makaha valley. Mean PAI, a measure of the one-sided leaf area and non-photosynthetic vegetation area, was higher in the non-native tree species (4.07–4.29) than in the native tree species (3.15–3.53). The LAD and NPVAD were estimated as the mean of direct measurements of leaf and branch angles of each species, respectively. Zenith angles range from 0◦ (horizontal) to 90◦ (vertical). Mean LAD was consistently lower in all three non-native species (21–22◦ ) than the native tree species

¯ Fig. 1. Location of upper Makaha valley; throughfall plots 1, 4, and 5; weather stations 1 and 6; and PAI above canopy monitoring sites OC-1 and OC-2. Aerial image from U.S. Geological Survey (USDI-USGS, 2009).

(34–70◦ ). The highest mean NPVAD was found in P. cattleianum (53◦ ) and S. terebinthifolius (49◦ ). The significant hydrological effects of alien vegetation, including higher ET and reduced groundwater recharge, documented in just a few affected Hawaiian forest sites stress the need for characterizing water uses of non-native vegetative covers in other areas and their effect on water resources (Gaskill, 2004; Kagawa et al., 2009). In addition, recent findings have revealed that wet-canopy evaporation in Hawaiian forests may be significantly higher than previous estimates and that groundwater recharge estimates may need to be ¯ revised (Giambelluca et al., 2009). In upper Makaha valley, investigating the effect of non-native tree species on interception rates is of interest due to the significant decline in valley streamflow (Oki, 2004; Mair et al., 2007; Mair, 2009). This study examines the variability of throughfall at three loca¯ tions dominated by non-native tree species in upper Makaha valley. Specific objectives of the study were: (1) to quantify and compare throughfall at three different locations and (2) to estimate the corresponding canopy interception parameters, i.e., canopy storage capacity and free throughfall coefficient. 2. Study area ¯ Upper Makaha valley is located on the dry leeward coast of O‘ahu, Hawai‘i, along the western side of the Wai‘anae mountain range (Fig. 1). The upper valley encompasses a total area of 5.5 km2 and consists of a deeply eroded valley along the northwestern remnants of the Wai‘anae volcano. Topography is rugged and elevation varies from 300 to 1226 m at the top of Mt. Ka‘ala (Fig. 1). Average annual rainfall varies from 1500 mm in the western portion to more than 2000 mm around Mt. Ka‘ala (Giambelluca et al., 1986). Rainfall is largely dictated by topography of the Wai‘anae mountains and the prevailing northeast trade winds. The upper valley has a rainy season extending from October to March, which accounts for 65–70% of the annual rainfall total. Moisture contributed by fog drip generally occurs above 750 m in Hawai‘i, and varies in frequency and amount due to the interaction between trade winds, and local land and sea breezes (Lau and Mink, 2006). Approximately 22% or 1.2 km2 of the upper valley subwatershed area has an elevation equal to or greater than 750 m. Thus, a potentially significant but unknown amount of moisture may be contributed by fog drip intercepted by vegetation above an elevation of 750 m.

A. Mair, A. Fares / Agricultural and Forest Meteorology 150 (2010) 1453–1466

1455

Table 1 Plot characteristics. Parameter

Units

Elevation Mean annual rainfalla Hillslope Aspect Distance to ridgelineb Number (trough gauges) Gauge areac Canopy aread Plant area index (PAI)e Standard error (PAI) Number (PAI measurements) Vegetation height and numberf Schinus terebinthifolius Psidium cattleianum Syzygium cumini Aleurites moluccana Coffea arabica Metrosideros polymorpha Cordyline fruticosa

m mm %

Station 1

a b c d e f g

◦

km – m2 m2 m2 /m2 m2 /m2 – m m m m m m m

4 344 1574 6–10 200 1.56 4 3.39 150 3.99 0.03 260

7–9 (44)g 2–11 (26) 9–13 (18) 2 (1) – – –

5 609 1598 20–24 10 1.25 2 1.66 75 4.64 0.04 123

– 2–10 (12) 7–9 (6) 6–10 (5) 1–5 (150) 10 (3) 4 (1)

605 1847 21–29 225 0.44 2 1.69 75 3.66 0.06 126 – 2–19 (142) – – – – –

From gridded rainfall database based on Giambelluca et al. (1986). Measured as linear distance along angle of N.45◦ E. Total collection area of N gages. Estimated forest area sampled by N gages. Measured from May 19–22, 2009. Inventory of trees within estimated forest area sampled by N gages. Range in tree height (number of trees).

Land cover in the upper valley subwatershed is largely dominated by non-native trees and shrubs interspersed with remnant native species (Harman, 2006). Surveys from the late-1960s (Hommon, 1970; Yen et al., 1972) describe many of the same

dominant tree species documented by Harman (2006) including P. cattleianum, S. terebinthifolius, S. cumini, and A. moluccana. These data suggest that the dominant vegetation in the upper valley has not changed over at least the last 40 years. The upper valley is part of

Fig. 2. Arrangement of stationary throughfall gauges: (a) station 1, (b) station 4, and (c) station 5. The numbered collectors are fed by two to five collection troughs. The shading corresponds to the estimated canopy sampling area.

1456

A. Mair, A. Fares / Agricultural and Forest Meteorology 150 (2010) 1453–1466

Fig. 3. (a) Throughfall gauge system comprised of stationary collector and four collection troughs (gauge 1B). (b) Collector funnel connected with two collection troughs. (c) Tipping bucket mechanism equipped with actuator, switch and event-based data logger.

a much larger preservation district in the Wai‘anae mountains with critical importance for groundwater recharge, irrigation water supply, and protected habitat of endangered native plant and animal species (Townscape, 2009).

3. Methodology 3.1. Experimental setup The study used data from three throughfall plots in upper ¯ Makaha valley (Fig. 1). Elevation, aspect, gross rainfall, and vegetation type were used as site selection criteria with the goal of conducting the study at three diverse locations (Table 1). Due to a lack of stands of native tree species suitable for this type of investigation, our study plots were confined to areas where non-native tree species were completely dominant. Measurements of throughfall were recorded using four stationary gauges from 8 September 2006 to 10 August 2008 at station 1. At station 4, measurements were recorded using two stationary gauges from 19 January 2007 to 11 August 2008, while at station 5 measurements were recorded using two stationary gauges from 26 January 2007 to 12 August 2008. The gauges at each plot were clustered together as close as practical (Fig. 2). Information on stemflow in similar Hawaiian forest ecosystems is scarce. Gaskill (2004) found stemflow varied from 0.6 to 4% of gross rainfall in tree plantations comprised of Eucalyptus robusta, F. uhdei, Casuarina glauca, and Grevillea robusta in a forest preserve in the southern Wai‘anae mountains. DeLay (2005) found stemflow ranged from 0.6 to 1.9% of gross rainfall in an elfin cloud forest comprised of Cheirodendron trigynum and M. polymorpha on the island of Hawai‘i. In studies conducted in tropical dry forests and tropical rainforests outside of Hawai‘i, stemflow was typically found to range from 0.6 to 1.8% of gross rainfall (Levia and Frost, 2003). Therefore, we assumed stemflow to be negligible and no measurements were collected for this study. Gross rainfall was measured at each site using a 203-mm diameter tipping bucket rain gauge with a tip resolution of 0.25 mm and an accuracy of ±2% at an intensity of 25.4 mm h−1 (Model 3665R, Spectrum Technologies, Inc., Plainfield, IL). Each gauge was fitted with an electronic data logger for recording cumulative tips at intervals that varied from 15 to 60 min during the monitoring period. Due to restrictions on installing instrumentation towers within the forest canopy, each gauge was installed in a nearby forest clearing at a height of 1.5–3 m depending on the height of surrounding vegetation. At stations 1, 4, and 5, the gauges were located approximately 40, 165, and 20 m, respectively, from the throughfall study plots. Gridded normal monthly and annual rainfall data based on the rainfall atlas by Giambelluca et al. (1986) were used to obtain average monthly and annual rainfall at each gauge. The database was provided by Giambelluca and Cao. Long-term average rainfall

at each of the monitoring sites was computed using the gridded data. Pro-rated values of average rainfall were used for periods less than an entire month. Temperature, humidity, solar radiation, and wind speed were monitored at opposite ends of the subwatershed using two weather stations (Model 700ET, Spectrum Technologies) equipped with data loggers for recording at an interval of 60 min or less (Fig. 1). The weather station at station 1 was installed at a height of 2 m, while the station at station 6 was installed at an elevated height of 3 m to avoid obstruction from nearby vegetation. Missing daily rainfall data were estimated from surrounding gauges using the normal ratio method (Mair and Fares, 2010). Daily rainfall and throughfall data were used to compute the cumulative rainfall and throughfall totals.

3.2. Throughfall measurements Stationary measurement gauges were constructed of fabricated galvanized steel tipping bucket gauges and received throughfall water captured in troughs that extended under the forest canopy (Fig. 3). Stationary trough-type gauges have been used to measure throughfall in other studies (McJannet et al., 2007; Sraj et al., 2008; Ziegler et al., 2009). Ziegler et al. (2009) used a very similar stationary trough gauge to measure throughfall in an evergreendominated forest stand in northern Thailand. They found that a stationary gauge offers several advantages over roving gauges because they integrate a sampling area equivalent to several rovers, do not require daily visits, and do not have to be periodically repositioned. Stationary trough gauges are well-suited to reliably monitor throughfall in remote and difficult terrain, such as that found in ¯ upper Makaha valley. Each tipping bucket was equipped with a reed switch and actuator for monitoring each tip using an event-based PendantTM data logger (Onset Computer Corporation, Bourne, MA). A static calibration was conducted for each tipping bucket. The static volume of throughfall required to produce one tip was 150 cm3 at stations 1 and 4 (0.17–0.18 mm), and 200 cm3 at station 5 (0.24 mm). A dynamic calibration correction computed for this type of gauge was later applied to account for differences in tip volume over a range of tipping rates. The dynamic calibration correction was provided by T.W. Giambelluca. The width of each collection trough was measured at 30 cm intervals to the nearest 0.01 cm, and the average width along each trough length ranged from 4.56 to 4.70 cm (Table 2). Each trough had a triangular-shaped channel with 5-cm rails (station 1) or 3cm rails (stations 4 and 5) to reduce rain splash loss. The collective length of troughs for each gauge was approximately 18 m. The average trough angle for each collector ranged from 4.8 to 10.5◦ and was varied to match the steepness of the surrounding terrain. However, the angle within each section was constant over each

A. Mair, A. Fares / Agricultural and Forest Meteorology 150 (2010) 1453–1466

1457

Table 2 Trough area computations. Average angle (◦ )

Area (m2 )

TF depth per tip (mm)

4.80 5.25 5.00 5.00

0.865 0.833 0.845 0.852

0.17 0.18 0.18 0.18

4.65 4.57

10.50 7.17

0.822 0.835

0.18 0.18

4.62 4.63

7.83 7.00

0.844 0.843

0.24 0.24

Gauge

Trough length (m)

Average width (cm)

1A 1B 1C 1D

18.662 18.362 18.094 18.158

4.65 4.56 4.68 4.70

4A 4B

18.098 18.434

5A 5B

18.439 18.453

continuous length of trough. The area of each continuous trough length was corrected for the trough angle. The corrected areas of each trough length were then summed to compute the trough area for each gauge. Field testing was conducted to confirm the trough angle allowed for adequate transport to the gauge. The trough area for each gauge ranged from 0.822 to 0.865 m2 after correcting for the trough angle. The troughs were positioned 0.1–0.8 m above ground. Each set of gauges were visited every 6–10 weeks in order to download data and remove accumulated litter in the troughs and collectors. Wetting losses associated with the collection troughs were estimated in the laboratory by taking a 44-cm long section (3- and 5-cm rails) and running water through the trough. The trough was allowed to drain before it was swabbed with absorbent material. The mass of water was then measured to the nearest 0.01 g. The process was repeated four times and the average value used to estimate wetting loss for the trough section. The loss was then scaled to the collective length of troughs for each gauge and reported as the wetting loss per rainfall event, which implies that the trough completely dries between each event. The wetting loss was computed separately for gauges with 3- and 5-cm rails. To quantify the range of effects of wetting loss on throughfall measurements, we then examined rainfall events under two event scenarios for each plot. For this analysis, rainfall events were defined as occasions where cumulative rainfall was 0.25 mm or more provided that there was a minimum of 3 h without rainfall between events. Under the first scenario, all rainfall events were evaluated and wetting loss (%) was computed as the number of rainfall events multiplied by the lab-computed wetting loss per event divided by the total event throughfall. Accounting for wetting

loss using all rainfall events produces an upper bound estimate of measurement error since as little as only 3 h separates each event and the troughs may not have been completely dry before the start of the following rainfall event. Under the second scenario, only those events that met the following two conditions were counted to compute wetting loss: (1) a minimum of 24 h of dry conditions preceded each event and (2) measurable amounts of throughfall were recorded (i.e., at least one tip at one throughfall gauge). The second scenario provides a lower bound estimate because of a high likelihood the troughs were dry before the start of each rainfall event and the events where only partial wetting of the trough may have occurred were excluded. In this study, the canopy area describes the area roughly sampled by the trough gages. Based on observations made during multiple throughfall events, we estimated the canopy area roughly as extending a distance of one meter on each side of a trough. We then computed a sampling area of 37.5 m2 per gauge. Tree species located within the canopy area footprint or with foliage/branches that extended into the canopy sample area were inventoried at each plot. Thus, we only inventoried trees with potential to impact throughfall. 3.3. Calculation of interception parameters Throughfall components are described by the Gash analytical model (Gash, 1979). Assuming stemflow is negligible, the throughfall over the duration of an event is given by:



Pnwet =

pPG + Tr pPG + Tr

PT < PG PT ≥ PG

(1)

Table 3 Rainfall and throughfall descriptive statistics. Statistic

Station 1

4

PT (mm)

Pn (mm)

TF (%)

PT (mm)

5 Pn (mm)

TF (%)

PT (mm)

Pn (mm)

TF (%)

2322 – – – – – –

1011 959 1063 – – – 2

43.6 41.3 45.8 – – – 2

a,b

Cumulative summary Total/meanc Minc Maxc Std. dev. CV (unitless) Std. error N (gauges) Event-based summarya , d Mean Min Max Std. dev. CV (unitless) N (rainfall events) a b c d

1684 – – – – – – 7.5 0.3 111 16 2.1 177

1012 883 1113 96 0.10 48 4

60.1 52.4 66.1 5.7 0.10 2.8 4

4.7 0.04 84 12 2.6 –

33 3.2 90.6 22.9 0.69 –

1998 – – – – – – 9.6 0.3 129 20 2.1 167

1179 1154 1205 – – – 2 5.8 0.09 98 15 2.5 –

Concurrent monitoring period from 26 January 2007 to 10 August 2008. Computed using daily data. Missing daily rainfall data estimated using normal ratio method. Throughfall depth adjusted for wetting loss error at station 1 (1.5%), station 4 (1.3%), and station 5 (1.6%). Excludes events with periods of missing hourly data and no recorded throughfall.

59.0 57.8 60.3 – – – 2 35.8 4.6 124 23 0.64 –

10 0.3 116 18 1.8 177

4.5 0.12 62.1 9.51 2.1 –

32.1 7.1 121 14.6 0.45 –

1458

A. Mair, A. Fares / Agricultural and Forest Meteorology 150 (2010) 1453–1466

 Pnsat =

0  ¯ R)](P ¯ [1 − (E/ G − PG )

Pn = Pnwet + Pnsat

PT < PG PT ≥ PG

(2) (3)

Pnwet

is the net throughfall before canopy saturation, p is where the free throughfall coefficient, PG is the cumulative gross rainfall at a particular time during an event, Tr is the cumulative release or redirected throughfall before canopy saturation, PT is the total gross rainfall for a particular event, PG is the gross rainfall required to saturate the canopy, Pnsat is the net throughfall after canopy saturation, E/R is the ratio of the evaporation rate to the rainfall rate during saturated canopy conditions, and Pn is the total net throughfall for the entire event. The free throughfall coefficient p was estimated using the slope of the linear regression relating PG to Pnwet constrained through the origin and during the initial portion of the pre-saturation curve to minimize the effect of redirected throughfall on Pnwet in Eq. (1). The canopy saturation point PG was estimated by identifying the inflection point in the PG − Pn relationship subjectively. The ratio of the evaporation rate to the rainfall rate during saturated canopy conditions, E/R, was estimated using the slope of the linear regression relating (PG − PG ) to Pnsat in Eq. (2). The canopy capacity, S, is defined as the amount of water left on the canopy in zero evaporation conditions when rainfall and throughfall have ceased (Gash and Morton, 1978), and is computed as: S = (1 − p)PG − Iw

(4)

where Iw is the intercepted rainfall that is evaporated during canopy wetting. For this analysis, rainfall events were defined as occasions where Pn was nonzero and PT was 0.25 mm or more provided there was a minimum of 3 h without rainfall between events. Interception parameters, p and S, were estimated from 15-, 30-, or 60-min resolution rainfall and throughfall measurements for each event exceeding 20 mm PT , and preceded by a minimum of 24 h of dry conditions to increase the likelihood that the canopy was dry prior to the event. The 20 mm threshold ensured a distinct inflection point for determining PG . Total event PT and Pn were computed for the duration of each event, provided no more than 3 h elapsed without rainfall. If 3 h or more elapsed without rainfall, then the event computations were terminated. To test the sensitivity of S to estimated Iw , S was computed using Eq. (4) by three methods (Link et al., 2004): (1) Iw assumed negligible; (2) Iw = (E/R)PG , which assumes

that E/R is constant throughout the entire event; (3) Iw estimated using the Penman method (Penman, 1948). Periods with missing unit value measurements of gross rainfall (15-, 30-, 60-min) were excluded from the event-based analyses. We chose not to estimate missing hourly rainfall data to avoid introducing uncertainty in rainfall estimation at unit value time intervals. Thus, a separate concurrent period of hourly rainfall and throughfall data were used to conduct the event-based analyses. The concurrent period for hourly data extends from 7 February to 14 December 2007, and from 7 February to 31 July 2008. 3.4. Plant area index

The PAI of the canopy area sampled at each throughfall plot was estimated using plant canopy analyzers (Model LAI-2000, LI-COR Biosciences, Lincoln, NE). The instrument determines the PAI by comparing light measurements made simultaneously above and below the tree canopy at five angles. Simultaneous measurements above and below the canopy were collected using two instruments oriented in the same direction. Above canopy measurements were collected from one stationary instrument located at one of two forest clearings (Fig. 1). Location OC-1 was used for stations 1 and 5, while OC-2 was used for station 4. Below canopy measurements

Fig. 4. Cumulative rainfall and throughfall over 563-day concurrent monitoring period from 26 January 2007 to 10 August 2008.

were taken along three transects aligned parallel to and slightly above each trough. One transect was located along the centerline of each trough, while the remaining two transects were offset a distance of one meter on either side. Measurements were collected at one meter intervals along the length of each transect. A total of 260, 123, and 126 measurements were collected at stations 1, 4, and 5, respectively. All measurements were collected under a cloudless sky at sunrise or sunset from 19 to 22 May 2009. A 45◦ view cap was used to minimize the contribution of scattered radiation. Computer software was used to reconcile above and below canopy measurements, and compute the PAI for each plot (LI-COR Biosciences, 2007). 4. Results All direct comparisons between the three throughfall plots were made for the concurrent 563-day monitoring period from 26 January 2007 to 10 August 2008, except where noted. Measurements collected for the entire monitoring period at each plot were incorporated for analyzing seasonal changes in throughfall within each plot. 4.1. Rainfall A total of 1684 mm of rainfall was recorded at station 1 (Table 3), which was well below the estimated average rainfall of 2353 mm for the concurrent monitoring period. Rainfall totals of 1998 mm at station 4 and 2322 mm at station 5 were also below their estimated average rainfall of 2385 and 2784 mm, respectively. The number of rain days (≥0.25 mm d−1 ) was 320, 343, and 339 at stations 1, 4, and 5, which corresponds to daily rainfall frequencies ranging from 57 to 61%. Mean daily rainfall per rain day was 5.3, 5.8, and 6.8 mm d−1 at stations 1, 4, and 5, respectively. Total recorded rainfall increased from station 1 to station 5 or from west to east (Fig. 4), resulting in a rainfall gradient of 449 mm km−1 . The increasing rainfall trend from station 1 to 5 is consistent with spatial patterns of monthly and annual rainfall, which indicate greater annual average rainfall near the ridgeline of the Wai‘anae mountain range (Giambelluca et al., 1986). Rainfall at all three locations follows a distinct bimodal diurnal pattern with a secondary maximum in the predawn period between 0300 and 0500 h, a minimum at 1000–1100 h, a primary maximum at 1400–1600 h, and another minimum at 2200–2300 h. Time series rainfall measurements between all three plots were highly correlated at

A. Mair, A. Fares / Agricultural and Forest Meteorology 150 (2010) 1453–1466

1459

hourly (Pearson’s R = 0.87–0.94) and daily (R = 0.96–0.98) time steps. Hourly rainfall intensities range from less than 0.25 mm h−1 at all three stations to a maximum of 49.3, 42.6, and 32.8 mm h−1 at stations 1, 4, and 5, respectively. Mean hourly rainfall intensity for recorded rainfall occurrences ranged from 1.4 mm h−1 at station 1 to 1.5 mm h−1 at stations 4 and 5. The cumulative exceedance probabilities for hourly rainfall are positively skewed at all three plots (Fig. 5a–c). The proportion of hourly records with measurable rainfall varies from 8% at station 1 (Fig. 5a) to 11% at station 5 (Fig. 5c) at each of the three plots, which indicate substantial periods of no measurable rainfall at each plot. A total of 167 to 177 rainfall events were identified at stations 1, 4, and 5 (Table 3). Event magnitudes range from a minimum of 0.25 mm to a maximum of 111, 129, and 116 mm at stations 1, 4, and 5, respectively. Similar to rainfall intensity, the exceedance probabilities for rainfall events were positively skewed (Fig. 6a–c). The rainfall regime at each plot was dominated by frequent low-intensity/low magnitude rainfall events, but occasional high intensity/high magnitude rainfall contributed a significant portion of total rainfall. Low rainfall intensities (≤5 mm h−1 ) accounted for 94–95% of all recorded rainfall measurements, but their sum total comprised only 52–56% of total event rainfall at all three plots (Fig. 5a–c). The proportion of low magnitude events (≤5 mm) comprised roughly 88% of all rainfall events at stations 1 and 4, and 84% at station 5, yet their sum total comprised only 21–23% of total event rainfall (Fig. 6a–c). Conversely, the number of high magnitude events (>20 mm) accounted for only 5–6% of all rainfall events but comprised 57–59% of total event rainfall at each plot. The frequency and mean values of rainfall event magnitude (PG ) increased from station 1 to station 5 and suggest that proximity to the ridge¯ line in upper Makaha valley is a controlling factor for rainfall frequency and event magnitude (Table 1). High frequency lowintensity/magnitude rainfall is characteristic of orographic rainfall in Hawaii. Measurement errors of 15–20% have been documented in tipping bucket rain gauges at high intensities of 300 mm h−1 (Lanza et al., 2005). Because the rain gauges used in this study measured cumulative tips per unit interval (i.e., per 15- to 60-min interval), the range of 1-min intensities is not known. However, hourly rainfall intensities >25 mm h−1 (i.e., factory calibrated value) comprise only 5–8% of cumulative rainfall at the three stations (Fig. 5a–c). Thus, the measurement error due to high intensity rainfall on the total recorded rainfall depth is likely small. 4.2. Throughfall and interception loss Cumulative mean throughfall ranged from 43.6% of cumulative rainfall at station 5 to 60.1% at station 1 (Table 3). Among the individual stationary gauges, the lowest throughfall was recorded at station 5 (41.3%) while the highest throughfall was recorded at station 1 (66.1%). The 563-day monitoring period encompassed both wet and dry cycles despite below average rainfall totals (Fig. 4). The standard error at station 1, equipped with four trough gauges, was 2.8% and within the range of standard errors (≤5–10%) for reliable throughfall estimates (Ziegler et al., 2009). Standard error could not be calculated for stations 4 and 5 because they were only equipped with two gauges at each plot. The range between minimum and maximum observed throughfall rates at each plot was 14.1% (station 1), 2.5% (station 4), and 4.5% (station 5), respectively. The narrower range of measured throughfall at stations 4 and 5 may reflect the greater homogeneity in dominant vegetative canopy type at these two sites: C. arabica at station 4 and P. cattleianum at station 5 (Table 1). The variability in throughfall among individual rainfall events was much greater than cumulative totals. Minimum event-based throughfall Pn ranged from 3.2% of PT at

Fig. 5. Cumulative exceedance curve for hourly rainfall intensity (dashed line) during concurrent monitoring from 26 January 2007 to 10 August 2008. Cumulative rainfall (solid line) and throughfall (dotted line) as percent of total recorded rainfall and throughfall, respectively: (a) station 1, (b) station 4, and (c) station 5.

station 1 to 7.1% of PT at station 5, while maximum throughfall Pn ranged from 91% of PT at station 1 to 124% of PT at station 4 (Table 3). Recorded throughfall tips were aggregated into 1-min intervals and then adjusted using the dynamic calibration correction. Use of the correction resulted in a relatively small increase in estimated throughfall. Had we assumed a constant depth-per-tip relationship,

1460

A. Mair, A. Fares / Agricultural and Forest Meteorology 150 (2010) 1453–1466

Fig. 7. Recorded throughfall for each event plotted versus associated rainfall for all three stations. Includes only rainfall events that were preceded by minimum of 24 h of dry conditions.

Fig. 6. Cumulative exceedance probability curve for rainfall event magnitude (dashed line) during concurrent monitoring from 26 January 2007 to 10 August 2008. Cumulative rainfall (solid line) and throughfall (dotted line) as percent of total event rainfall and throughfall, respectively. Cumulative throughfall rate (dash-dot line) plotted as cumulative throughfall divided by cumulative rainfall: (a) station 1, (b) station 4, and (c) station 5.

mean throughfall percentages would have been underestimated by only 2.3–2.4%. Ziegler et al. (2009) using similar equipment found that throughfall would have been underestimated by 20% without dynamic calibration correction; rainfall would have been underestimated by 15%. The lack of 1-min rainfall intensity measurements

from our study prevents direct comparison of short-interval rainfall intensity between studies. However, we suspect the differences in underestimation are likely due, in part, to the higher rainfall intensity recorded at their study site in Thailand. Thus, the lower daily rainfall intensities imply less underestimation of rainfall and throughfall for our study. Estimated wetting loss associated with each trough gauge was 40–41 cm3 or 0.05 mm event−1 , which assumes the troughs were dry at the start of each event and no evaporation occurred over the duration of the rainfall event. As a result of these wetting losses, total cumulative throughfall depth was underestimated by 0.9–2.1% at station 1, 0.7–1.9% at station 4, and 0.8–2.3% at station 5. We used the average of the lower and upper bounds of wetting loss error at each plot to adjust the total cumulative throughfall at each plot (Table 3). The throughfall regime at each plot suggests that the vegetative canopy can act as a low-pass filter by intercepting high frequency low-intensity/magnitude rainfall and increasing the relative contribution of occasional high intensity rainfall to total throughfall. For example, low magnitude events (≤5 mm) contributed only 8–9% of total throughfall among the three plots, but 21–23% of total rainfall (Fig. 6a–c). Conversely, high magnitude events (>20 mm) contributed 70–72% of total throughfall but only 57–59% of total rainfall. Under changing climate scenarios, these interception processes play a significant role in determining the amount of moisture reaching the ground surface and have important ramifications for groundwater recharge, streamflow generation, and watershed ecology. The observed throughfall rates were highly dependent on the rainfall regime, and further illustrate the low-pass filtering capacity at each plot (Fig. 6a–c). For example, the cumulative throughfall rate at stations 1, 4, and 5 was only 23, 25, and 19%, respectively, for high frequency low magnitude rainfall events (≤5 mm), but increased to 60, 58, and 43% when high magnitude events were included. For rainfall events with maximum initial canopy storage capacity (i.e., ≥24 h of no rainfall before start of event), throughfall rates were highly variable for events <20 mm and averaged 41, 39, and 31% at stations 1, 4, and 5 (Fig. 7). However, throughfall rates at high magnitude events (>20 mm) stabilized at average rates of 78% at station 1 (N = 7), 69% at station 4 (N = 7), and 54% at station 5 (N = 7). The PAI at the three throughfall plots varied from 3.66 at station 5 to 4.64 at station 4 (Table 1). Although throughfall is known to depend on factors in addition to PAI, the lower observed throughfall at station 4 compared to station 1 is consistent with

A. Mair, A. Fares / Agricultural and Forest Meteorology 150 (2010) 1453–1466

1461

Fig. 8. Detailed hourly meteorological data for rainfall events on 3–5 November 2007 at stations 1, 4, and 5: (a) cumulative depth of rainfall and throughfall, (b) rainfall, throughfall, and interception intensity, (c) vapor deficit and temperature, and (d) maximum wind speed and net radiation. Meteorological data at station 6 used for comparison at stations 4 and 5.

higher observed PAI at station 4. However, the lower PAI and lower throughfall at station 5 clearly suggest that other factors (i.e., phenological properties of P. cattleianum) may be causing a proportionally greater reduction in release throughfall. Anecdotal evidence collected over the course of this study suggests that a notable amount of intercepted rainfall was instead diverted to stemflow. The higher NPVAD in P. cattleianum (Suzuki, 2006) and its characteristically smooth bark may be physical attributes that reduce release throughfall but enhance stemflow production. The greater proportion of P. cattleianum at station 1 as compared to station 4 (Table 1) may also be resulting in less release throughfall at station 1 per unit rainfall. However, it is difficult to distinguish the

effects of P. cattleianum at sites with heterogeneous canopies, such as station 1 and 4, without additional stemflow measurements. The interception capacity of the different plots may also vary because of changes in vegetative cover due to season or storm damage. A comparison of detailed rainfall and weather data illustrates the varying effect of forest canopy on throughfall intensity as well as the variability in conditions that control interception loss during a high rainfall period (combination of two rainfall events) (Fig. 8a–d). The period began at 2300 h on 3 November 2007 and lasted 24 h at all three plots. During this period, between 110 to 130 mm of rain fell at stations 1, 4, and 5. Weather data from station 6 was used as a proxy for meteorological conditions at stations 4 and 5.

Table 4 Interception parameter estimation summary. TF (%)

E/R

Sa (mm)

Sb (mm)

Sc (mm)

p

23.4 17.0 6.9 22.0 6.2 6.2 12.7 94.4 13.5

75.0 73.5 82.5 77.8 74.1 82.1 81.7 77.8 78.1

0.23 0.22 0.07 0.20 0.17 0.14 0.13

2.6 2.1 2.4 na 3.1 2.3 na

1.7 1.5 2.1 na 2.3 1.8 na

2.5 2.1 2.4 na 2.9 2.4 na

0.29 0.19 0.30 na 0.32 0.43 na

0.17

2.5

1.9

2.5

0.31

24.9 47.3 97.7 41.2 19.8 46.4 20.6 297.9 42.6

21.1 23.5 31.7 14.3 13.4 17.6 13.0 134.6 19.2

54.1 66.9 75.5 74.2 59.5 72.5 61.5 68.9 66.3

0.44 0.23 0.23 0.23 0.31 0.18 0.30

1.6 na 2.5 na 5.0 8.0 5.1

0.6 na 1.8 na 3.0 6.3 3.4

1.6 na 2.5 na 5.0 6.6 4.4

0.26 na 0.15 na 0.21 0.13 0.13

0.27

4.4

3.0

4.0

0.18

62.1 15.6 19.5 20 30.3 43.4 14 204.9 29.3

51.7 17.7 13.9 16.8 23.1 40.6 12.8 176.6 25.2

54.6 46.8 58.4 54.3 56.7 51.7 52.2 53.7 53.5

0.43 0.48 0.39 0.44 0.37 0.49 0.45

3.9 4.0 1.6 1.7 na na na

2.1 1.8 0.9 0.7 na na na

3.8 3.5 1.5 1.7 na na na

0.08 0.12 0.18 0.24 na na na

0.44

2.8

1.4

2.6

0.16

Station

Start

Duration (h)

PT (mm)

Pn (mm)

1

10/14/06 7:00 PM 12/28/06 2:15 AM 1/23/07 4:30 AM 11/4/07 12:00 AM 2/16/08 3:00 PM 2/24/08 5:00 AM 3/21/08 11:00 AM Total Mean

30.75 12.75 17.75 13 15 10 8

93.6 64.3 39.3 99.2 23.9 34.6 69.5 424.4 60.6

70.2 47.3 32.4 77.2 17.7 28.4 56.8 330.0 47.1

4

1/23/07 7:30 AM 3/9/07 2:30 AM 11/3/07 11:00 PM 12/30/07 7:00 PM 2/24/08 5:00 AM 3/21/08 10:00 AM 7/9/08 10:30 AM Total Mean

46 70.8 129.4 55.5 33.2 64 33.6 432.5 61.8 113.8 33.3 33.4 36.8 53.4 84 26.8 381.5 54.5

5

a b c

11/3/07 11:00 PM 11/29/07 4:30 PM 2/16/08 3:00 PM 2/24/08 5:00 AM 3/21/08 10:00 AM 5/21/08 11:30 AM 7/9/08 10:30 AM Total Mean

15.3 15.75 29.25 23 12.5 14.5 7 10.5 16.1 11.5 14.5 17.5 13 7 23.5 9 13.7

Iw = 0. Iw = (E/R)PG . Iw calculated from Penman method.

Inet (mm)

1462

A. Mair, A. Fares / Agricultural and Forest Meteorology 150 (2010) 1453–1466

Fig. 9. Example plots of data used to estimate canopy direct throughfall proportion and saturation storage capacity, assuming no evaporation during canopy wetting. Linear regressions are fit to a scatterplot of cumulative throughfall vs. cumulative rainfall: (a) station 1 during 14 October 2006 event, (b) station 4 during 23 January 2007 event, (c) station 5 during 24 February 2008 event.

The event was characterized by relatively high rainfall intensity, low net radiation, and variable wind speed and vapor deficit. Total interception loss, assuming negligible stemflow, at stations 1, 4, and 5 was 26.2, 31.7, and 58.3 mm, respectively. All three plots show the same pattern of cumulative rainfall and throughfall development suggesting a similar temporal pattern across the entire upper valley (Fig. 8a). The hourly interception storage rate, computed as the difference between gross rainfall and throughfall, indicates the canopy was gaining water (positive values) throughout much of the event and that the greatest rate of interception coincided with the highest rainfall intensity at 0300 h on 4 November (Fig. 8b). In just a few cases, the interception storage rate was slightly negative which indicates that the canopy was only briefly losing water. At stations 4 and 5, the interception storage rate dipped to near zero at 0400 h on 4 November following the hour of highest rainfall intensity, probably as a result of a combination of high winds and the temporal lag between periods of canopy supersaturation and drainage under varying rainfall intensity. A similar yet less pronounced pattern also occurred at station 1. The microclimate variations of vapor deficit, temperature, maximum wind speed, and net radiation during the same rainfall period suggest low evaporation across the upper valley. Maximum wind speeds were used because mechanical dislodging of intercepted water is likely to be a function of the maximum, rather than the mean wind speed (Link et al., 2004). The vapor deficit at station 1 decreased at the beginning of the event and remained close to zero

through 0800 h on 4 November before increasing, fluctuating, and finally decreasing through the remainder of the period (Fig. 8c). At station 6, the deficit also decreased at the start of the event but remained close to zero through the entire period. Once rainfall ceased, the vapor deficit remained close to zero at station 1 but increased dramatically at station 6 suggesting the potential for more rapid drying at station 6. Low wind speed and low net radiation at station 1 (Fig. 8d) combined with an elevated vapor deficit during only a portion of the period indicates that only minimal evaporation could occur. At station 6, relatively high winds were prevalent throughout much of the event which likely enhanced throughfall (Fig. 8d). However, the low vapor deficit combined with the very low net radiation (Fig. 8d) implies that only negligible evaporation could occur. Total potential evaporation (Penman method) during the period of rainfall was only 0.9 mm at station 1 and 0.3 mm at station 6. These results suggest that nearly all of the intercepted rainfall was diverted to canopy storage and stemflow during this particular rainfall period. 4.3. Interception parameters Rainfall events were examined to better understand throughfall behavior under initially dry canopy conditions. A total of seven events were identified each at stations 1, 4, and 5 (Table 4). Values for p, S, and E/R were estimated for each event from a plot of cumulative rainfall vs. cumulative throughfall (Fig. 9a–c, one example

A. Mair, A. Fares / Agricultural and Forest Meteorology 150 (2010) 1453–1466

1463

for each plot). Canopy interception parameters were estimated for storms with PT > 20 mm to ensure a distinct inflection point. Analyses of storms 10 mm < PT < 20 mm were also conducted; however, a distinct inflection point could not be determined for these lesser magnitude events. The value for p ranged from 0.19 to 0.43 with a mean of 0.16 at station 1, 0.13–0.26 with a mean of 0.18 at station 4, and 0.08–0.24 with a mean of 0.18 at station 5 (Table 4). In addition to differences in canopy vegetation at each plot, the temporal and spatial variability of p could also be the result of differences in raindrop sizes (Calder, 1998), greater dislodging of intercepted drops by higher wind gusts (Hormann et al., 1996), or differences in canopy defoliation resulting from storm damage. Despite similar temporal rainfall patterns at all three plots during the events of 3–5 November 2007, observed wind gusts were as much as 8 m s−1 higher in the upper valley (station 6) than the lower valley (station 1) over the same period (Fig. 8d). In December 2007, high wind events caused a significant amount of canopy defoliation, particularly at stations 4 and 5, which likely altered canopy interception until regrowth was complete. From April 2007 to August 2008, only 30- and 60-min resolution gross rainfall data were available for analysis. Values for p and S for a total of six events during this period could not be determined from the 30- and 60-min resolution data (Table 4). An estimate for p and S was also not possible during the 9 March 2007 event at station 4 due to lack of unit value data during canopy pre-saturation. Canopy storage capacity S values were computed from Eq. (4) using three values for Iw (Table 4). Comparison between the S values computed for Iw = 0 and Iw using the Penman method indicate that the assumption of negligible Iw is valid for most events due to very small vapor deficits during canopy wetting. The mean S at stations 1, 4, and 5 was 2.5, 4.0, and 2.6 mm, respectively (Table 4, Iw computed from Penman method). The relatively high S computed at station 4 is consistent with the highest observed PAI of the three plots (Table 1). 4.4. Spatial and temporal variability Differences in canopy phenology can influence throughfall variability among the different plots. However, other factors such as rainfall intensity, event magnitudes, wind speed and direction, and evaporation can also affect throughfall (Sraj et al., 2008; Ziegler et al., 2009). Monthly values of rainfall, throughfall, and throughfall rate were computed from daily data for the duration of monitoring at each plot. The greatest variability occurred at station 1 where the throughfall rate varied from a low of 14% in October 2007 to a high of 78% in March 2008 (Fig. 10a). Similar but less pronounced patterns of variability occurred at the other plots with throughfall rates ranging from 21 to 71% at station 4 (Fig. 10b) and 23–53% at station 5 (Fig. 10c). The monthly values were used to compute annualized throughfall rates of 62% at station 1, 60% at station 4, and 45% at station 5. A significant increase in rainfall coupled with high winds in December 2007 produced a noticeable rise in throughfall at all three plots (Fig. 10a–c). A large amount of defoliation was observed at stations 4 and 5 during visits in December 2007 immediately after the high wind event. The defoliation reduced canopy interception for a period of time as evidenced by the elevated and consistent throughfall rate at stations 4 and 5 from January to March 2008. The distribution of event throughfall rates and rainfall at each plot was assessed before conducting statistical tests. The Shapiro–Wilk test was used to determine whether the assumption of a normal distribution was valid for original, log-transformed, and the square-root transformed observations of throughfall rates and rainfall at each plot (Shapiro and Wilk, 1965). The results indicate non-normal distributions of throughfall rate and rain-

Fig. 10. Monthly rainfall, throughfall, and throughfall rate from October 2006 to July 2008: (a) station 1, (b) station 4, and (c) station 5.

fall at all three plots for both original and transformed datasets (p < 0.01). To investigate temporal and spatial differences in rainfall and throughfall, concurrent observations of throughfall and rainfall were grouped according to rainfall event magnitudes: (1) low: PT ≤ 5 mm; (2) medium: 5 mm < PT ≤ 20 mm; (3) high: PT > 20 mm. These events were also classified according to quarter (i.e., Q1 = January–March, Q2 = April–June, etc.) to investigate for seasonal differences in throughfall and rainfall. A non-parametric one-way ANOVA (Kruskal–Wallis test) on ranks was then applied

1464

A. Mair, A. Fares / Agricultural and Forest Meteorology 150 (2010) 1453–1466

Table 5 Summary of non-parametric ANOVA testing for differences in throughfall rate and rainfall. Magnitude (mm)

≤5 5–20 >20 All

Pr > 2

N ST1

ST4

ST5

TF (%)

PT

132 28 17 177

118 32 17 167

112 43 22 177

0.055 <0.0001 <0.0001 0.285

<0.0001 0.999 0.349 <0.0001

to the concurrent dataset to test for equality of sample medians among the three plots and between quarters at each plot (Kruskal and Wallis, 1952). The non-parametric ANOVA on ranks between plots confirmed that the throughfall rate for events with medium and high PT were significantly different among the three plots (p < 0.0001) (Table 5). However, rainfall for events with medium and high PT was not statistically different at the 0.05 significance level. Conversely, the one-way ANOVA for rainfall at low PT was significantly different among the three plots while the throughfall rate was only slightly above the 0.05 significance level (p = 0.055). These results suggest that other differences between the plots, such as vegetation, are the source of differences in throughfall rate. Non-parametric ANOVA on ranks within each plot indicated that significant seasonal differences in throughfall were evident at stations 1 (p = 0.044) and 4 (p = 0.012) but not at station 5 (p = 0.46) (Table 6). The highest throughfall rates at all three plots occurred during the first quarter (January–March), while the lowest throughfall rates occurred during the third quarter (July–September). Seasonal differences in rainfall were also evident at station 1 and may help to explain the observed difference in throughfall rates. Consistent with long-term patterns of seasonal rainfall variability, rainfall event magnitudes were greater during the first and fourth quarters of each year at each plot. At station 4, the lack of significant seasonal differences in rainfall suggests that seasonal changes in vegetation may be a cause of throughfall variation. Seasonal phenological changes in C. arabica, the dominant vegetation at station 4, and defoliation resulting from wind damage may have significantly altered canopy interception during the monitoring period. The seasonal variability in potential evaporation rates is likely an additional contributing factor to observed differences in throughfall rate (Fig. 11). 5. Discussion Little information on throughfall in similar Hawaiian forests is available for comparison purposes. Throughfall rates ranging Table 6 Summary of non-parametric ANOVA testing for seasonal changes in throughfall and rainfall. Pr > 2

Station

Magnitude (mm)

N Q1

Q2

Q3

Q4

TF (%)

PT

1

≤5 5–20 >20 All

20 9 5 34

64 11 2 77

29 4 3 36

19 4 7 30

0.088 0.437 0.484 0.044

0.389 0.112 0.743 0.0036

4

≤5 5–20 >20 All

18 12 4 34

52 13 2 67

28 4 3 35

20 3 8 31

0.061 0.896 0.356 0.012

0.727 0.412 0.428 0.142

5

≤5 5–20 >20 All

18 13 6 37

42 19 3 64

27 8 3 38

25 3 10 38

0.553 0.31 0.811 0.46

0.835 0.94 0.662 0.13

Fig. 11. Potential evapotranspiration computed using the Penman method at stations 1 and 6 from September 2006 to August 2008.

from 70 to 91% were reported in a forest comprised of entirely different non-native species in the southern Wai‘anae mountains (Gaskill, 2004). A value of 84% was reported in a cloud forest comprised of C. trigynum and M. polymorpha on the island of Hawai‘i (DeLay, 2005). Thus, the throughfall rates reported in this study are well below what has been reported in a small sampling of other Hawaiian forests. Anecdotal evidence in the form of abundant snags suggests that portions of upper Makaha valley, particularly around station 5, were formerly dominated by native M. polymorpha. However, the lack of information from similar environments in Hawaii with native canopies makes it difficult to conclude whether the observed throughfall in this study were below what may have occurred when the area was vegetated with native tree species. Giambelluca et al. (2009) recently reported much higher annual rates of ET (3.4 mm d−1 ) than previous estimates of forest ET in Hawai‘i. They highlighted the importance of wet-canopy evaporation in controlling variations in ET and contributing to high annual ET in Hawai‘i. In this study, mean potential ET (PET) ranged from 1.9 mm d−1 at station 6 in December 2007 to 6.2 mm d−1 at station 1 in July 2007; mean annual PET was 3.0 mm d−1 at station 6 and 4.5 mm d−1 at station 1 (Fig. 11). The combination of high PET and ¯ low-intensity rainfall observed in upper Makaha valley are likely resulting in high rates of actual ET. Thus, the relatively low rates of throughfall observed during this study are likely, in part, the result of high evaporation rates. Throughfall at station 5, where the canopy is comprised entirely of P. cattleianum, was substantially lower (44%) than the other plots (59–60%). Over the course of the concurrent 563-day monitoring period, nearly the same throughfall was recorded at station 5 as at station 1 despite 38% more rainfall (+638 mm) at station 5. The smaller proportion of rainfall recorded as throughfall at station 5, particularly at medium and high magnitude events, suggests that canopy differences may be a contributing factor to throughfall variability. However, the lower throughfall at station 5 should not be interpreted as an indication of greater interception loss under stands of P. cattleianum since the proportion of rainfall diverted to stemflow is not yet known. Indeed, recent measurements collected on the island of Hawai‘i suggest that interception losses may actually be less under stands of P. cattleianum than stands of native M. polymorpha (T. Giambelluca, personal communication). Observations over the course of the study suggest that stemflow within mature stands of P. cattleianum in the upper valley could be a large but yet unknown component of the canopy water balance. The characteristically smooth bark and relatively high

A. Mair, A. Fares / Agricultural and Forest Meteorology 150 (2010) 1453–1466

NPVAD of P. cattleianum may be structural features that enhance the amount of intercepted rainfall converted to stemflow. While P. cattleianum appears to generate low throughfall rates, the impact on the amount of rainfall reaching surface and subsurface soils is not yet understood. The relatively low PAI and low throughfall are a further indication of differences in canopy function at station 5. Thus, the presence of P. cattleianum in substantial numbers within the canopy may be a controlling factor for the generation of release throughfall and stemflow at the stand level.

6. Conclusions Gross rainfall and throughfall were measured at three loca¯ tions in upper Makaha valley over 18- and 23-month periods from September 2006 to August 2008. During a concurrent monitoring period, mean throughfall was 60% at station 1 under a canopy dominated by S. terebinthifolius, 59% at station 4 under a canopy dominated by C. arabica, and 44% at station 5 under a monotypic stand of P. cattleianum. Annualized throughfall rates were computed as 62, 60, and 45% at stations 1, 4, and 5, respectively. The free throughfall coefficient p ranged from a mean of 0.16 at station 5 to 0.31 at station 1, while the mean canopy storage capacity S varied from 2.5 mm at station 1 to 4.0 mm at station 4. The low throughfall percentages across the valley are likely due, in part, to a combination of frequent low-intensity/low magnitude rainfall and high evaporation. However, additional investigative work is necessary to quantify the amount of ET occurring in the upper valley. The exceptionally low throughfall at station 5, recorded under a mature stand of P. cattleianum with low PAI, may be the result of structural features (smooth bark, high NPVAD) that increase the amount of intercepted rainfall diverted to trunk storage and stemflow, and decrease the amount of release throughfall. Further investigation is also needed to quantify the amount of stemflow occurring at all three locations. The spread of one non-native tree species (e.g., P. cattleianum) has the ability to dramatically alter the canopy water balance with potentially significant consequences for groundwater recharge and streamflow generation. As suggested by Giambelluca et al. (2009), estimates of groundwater recharge for critical recharge areas in Hawai‘i may need to be revised to account for both high rates of wet-canopy evaporation and the spread of non-native species that have the potential to create large monotypic stands in high rainfall, critical recharge areas. This study has provided estimates of throughfall rates for non-native forests that could be used in such calculations, although further work is necessary to evaluate the potential additional contribution of stemflow. Additional research is also needed to better understand wet-canopy evaporation in the valley.

Acknowledgements The project was supported by two grants from the U.S. Department of Agriculture: (1) Cooperative State Research, Education and Extension Service grant number 2004-34135-15058, (2) McIntire–Stennis formula grant number 2006-34135-17690. The authors wish to thank Tom Giambelluca and Alan Ziegler for their assistance with designing and fabricating the stationary throughfall gauges. Special thanks to Nghia Dai Tran, Domingos Maria, Amjad Ahmad and Mohammad Safeeq for all their assistance with the field installation and monitoring. In addition, the authors extend thanks to the Honolulu Board of Water Supply and members of Mohala I ka Wai for their assistance and cooperation. Finally, the authors wish to thank two anonymous reviewers for their constructive comments.

1465

References Calder, I.R., 1998. Water use by forests, limits and controls. Tree Physiology 18, 625–631. Cavalier, J., Jaramilla, M., Solis, D., de Leon, D., 1997. Water balance and nutrient inputs in bulk precipitation in tropical montane cloud forest in Panama. Journal of Hydrology 193, 83–96. DeLay, J., 2005. Canopy water balance of an elfin cloud forest at Alakahi, Hawai’i. Master’s Thesis, University of Hawaii at Manoa. Denslow, J., 2003. Weeds in paradise: thoughts on the invasibility of tropical islands. Annals of the Missouri Botanical Garden 90, 119–127. Dykes, A.P., 1997. Rainfall interception from a lowland tropical rainforest in Brunei. Journal of Hydrology 200, 260–279. Gash, J.H.C., 1979. An analytical model of rainfall interception by forests. Quarterly Journal, Royal Meteorological Society 105, 43–55. Gash, J.H.C., Morton, A.J., 1978. An application of the Rutter model to the estimation of the interception loss from Thetford Forest. Journal of Hydrology 38, 49–58. Gaskill, T.G.R., 2004. Hydrology of forest ecosystems in the Honouliuli preserve: implications for groundwater recharge and watershed restoration. Ph.D. Thesis, University of Hawaii at Manoa, Honolulu. Giambelluca, T.W., Asner, G.P., Martin, R.E., Nullet, M.A., Huang, M., DeLay, J.K., Mudd, R.G., Takahashi, M., 2007. Impacts of alien tree invasion on evapotranspiration in tropical montane cloud forest in Hawaii. Eos Transactions, American Geophysical Union 88 (52) (Fall Meeting Supplement, Abstract H21I-02). Giambelluca, T.W., Martin, R.E., Asner, G.P., Huang, M., Mudd, R.G., Nullet, M.A., DeLay, J.K., Foote, D.E., 2009. Evapotranspiration and energy balance of native wet montane cloud forest in Hawaii. Agricultural and Forest Meteorology 149, 230–243. Giambelluca, T.W., Nullet, M.A., Schroeder, T.A., 1986. Rainfall atlas of Hawai’i: Report R76. Department of Land and Natural Resources, Honolulu, p. 267. Gingerich, S.G., Oki, D.S., 2000. Ground water in Hawai‘i. U.S. Geological Survey Fact Sheet 126-00. Harman, M.R., 2006. Mapping invasive species in Makaha valley, Oahu using fine resolution satellite imagery. Bachelor’s Thesis, University of Hawaii at Manoa. Hommon, R.J., 1970. Report 9, final report on the upper valley survey. In: Green, R.C. (Ed.), Makaha Valley Historical Project, Interim Report No. 2. Department of Anthropology, Bernice Pauahi Bishop Museum, Honolulu. Hormann, G., Branding, A., Clemen, T., Herbst, M., Hinrichs, A., Thamm, F., 1996. Calculation and simulation of wind controlled canopy interception of a beech forest in northern Germany. Agricultural and Forest Meteorology 79, 131–148. Kagawa, A., Sack, L., Duarte, K., James, S., 2009. Hawaiian native forest conserves water relative to timber plantation: species and stand traits influence water use. Ecological Applications 19 (6), 1429–1443. Kruskal, W.H., Wallis, W.A., 1952. Use of ranks in one-criterion variance analysis. Journal of the American Statistical Association 47 (260), 583–621. Lanza, L., Leroy, M., van der Muelen, J., Ondras, M., 2005. The WMO Laboratory Intercomparison of Rainfall Intensity Gauges. World Meteorological Organization. Lau, L.S., Mink, J.F., 2006. Hydrology of the Hawaiian Islands. University of Hawai‘i Press, Honolulu, 274 pp. Le Maitre, D.C., Wilgen, B.W.v., Gelderbloom, C.M., Bailey, C., Chapman, R.A., Nel, J.A., 2002. Invasive alien trees and water resources in South Africa: case studies of the cost and benefits of management. Forest Ecology and Management 160, 143–159. Levia, D.F., Frost, E.E., 2003. A review and evaluation of stemflow literature in the hydrologic and biogeochemical cycles of forested and agricultural ecosystems. Journal of Hydrology 274, 1–29. Levia, D.F., Frost, E.E., 2006. Variability of throughfall volume and solute inputs in wooded ecosystems. Progress in Physical Geography 30 (5), 605–632. LI-COR Biosciences, 2007. LAI-2000 File Viewer v. 1.10. LI-COR Biosciences, Lincoln, NE. Link, T.E., Unsworth, M., Marks, D., 2004. The dynamics of rainfall interception by a seasonal temperate rainforest. Agricultural and Forest Meteorology 124, 171–191. Mair, A., 2009. Effects of rainfall variability and groundwater pumping on streamflow in upper Makaha valley. Ph.D. Thesis, University of Hawaii at Manoa. Mair, A., Fares, A., 2010. Assessing rainfall data homogeneity and estimating missing ¯ records in Makaha valley, O‘ahu, Hawai‘i. Journal of Hydrologic Engineering 15 (1), 61–66. Mair, A., Fares, A., El-Kadi, A., 2007. Effects of rainfall and groundwater pumping on ¯ streamflow in Makaha, O‘ahu, Hawai‘i. Journal of the American Water Resources Association 43 (1), 148–159. McJannet, D., Wallace, J., Reddell, P., 2007. Precipitation in Australian tropical rainforests. I. Measurement of stemflow, throughfall and cloud interception. Hydrological Processes 21, 1692–1702. Oki, D.S., 2004. Trends in streamflow characteristics at long-term gaging stations, U.S. Geological Survey Scientific Investigations, Hawai‘i, Report 2004-5080. Penman, H.L., 1948. Natural evaporation from open water, bare soil and grass. Royal Society of London Series A, 120–146. Scatena, F.N., 1990. Watershed scale rainfall interception on two forested watersheds in the Luquillo Mountains of Puerto Rico. Journal of Hydrology 113, 89–102. Schellekens, L., Bruijnzeel, L.A., Scatena, F.N., Bink, N.J., Holwerda, F., 2000. Evaporation from a tropical rain forest, Luquillo Experimental Forest, eastern Puerto Rico. Water Resources Research 36, 2183–2196. Shapiro, S.S., Wilk, M.B., 1965. An analysis of variance test for normality (complete samples). Biometrika 52 (3–4), 591–611.

1466

A. Mair, A. Fares / Agricultural and Forest Meteorology 150 (2010) 1453–1466

Sraj, M., Brilly, M., Mikos, M., 2008. Rainfall interception by two deciduous Mediterranean forests of contrasting stature in Slovenia. Agricultural and Forest Meteorology 148, 121–134. Suzuki, T., 2006. Spectral separability among invasive and native plant species for satellite image analysis. Master’s Thesis, University of Hawaii at Manoa. Townscape, Inc., 2009. Wai‘anae watershed management plan, pre-final draft. Honolulu Board of Water Supply. USDI-USGS, 2009. Earth Data Imagery. U.S. Geological Survey, National Geospatial Program. Vitousek, P.M., Loope, L.L., Stone, C.P., 1987. Introduced species in Hawaii: biological effects and opportunities for ecological research. Trends in Ecology and Evolution 2 (7), 224–227.

Yen, D.E., Kirch, P.V., Rosendahl, P., Riley, T., 1972. Report 4, prehistoric agriculture in the upper valley of Makaha, Oahu. In: Ladd, E.J., Yen, D.E. (Eds.), Makaha Valley Historical Project, Interim Report No. 3. Department of Anthropology, Bernice Pauahi Bishop Museum, Honolulu. Zavaleta, E., 2000. The economic value of controlling an invasive shrub. Ambio 29 (8), 462–467. Ziegler, A.D., Giambelluca, T.W., Nullet, M.A., Sutherland, R.A., Tantasarin, C., Vogler, J.B., Negishi, J.N., 2009. Throughfall in an evergreen-dominated forest stand in northern Thailand: comparison of mobile and stationary methods. Agricultural and Forest Meteorology 149, 373–384.