The Depositional Web on the Floodplain of the Fly River, Papua New Guinea Geoff Day,1 William E

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 113, F01S02, doi:10.1029/2006JF000622, 2008 Click Here for Full Article

The depositional web on the floodplain of the Fly River, Papua New Guinea Geoff Day,1 William E. Dietrich,2 Joel C. Rowland,2 and Andrew Marshall3 Received 1 July 2006; revised 28 June 2007; accepted 31 July 2007; published 20 March 2008.

[1] Floodplain deposition on lowland meandering rivers is usually interpreted as either lateral accretion during channel migration or overbank deposition. Previous studies on the Fly River in Papua New Guinea suggest, however, that floodplain channels (consisting of tie channel and tributary channels) play an important role in conveying sediment out across the floodplain. Here we report the results of an intensive field study conducted from 1990 to 1998 that documents the discharge of main stem water from the Fly River onto its floodplain and maps the spatial pattern of sediment deposition on the floodplain (using as a tracer elevated particulate copper introduced into the system by upstream mining). An extensive network of water level recorders demonstrates significant hydraulic heads from the main stem out the floodplain channels. For the monitoring period 1995–1998, net water discharge into the floodplain channels was about 20% of the flow. Another 20% is estimated to spill overbank from the main stem in wet years. Annual floodplain coring from 1990 to 1994 obtained over 800 samples across the 3500 km2 Middle Fly floodplain for use in documenting temporal and spatial patterns of sediment deposition. Early samples record the rapid spread of sediment up to 10 km away from the main stem via floodplain channels. Later, more intensive coring samples documented a well-defined exponential decline in sediment deposition from the nearest channel (which differed little between floodplain and main stem channels). Deposition, averaging about 6–9 mm/a, occurred in a 1 km corridor either side of these channels and effectively ceased beyond that distance. About 40% of the total sediment load was deposited on the floodplain, with half of that being conveyed by the over 900 km of floodplain channels (equal to about 0.09% sediment deposition/km of main stem channel length). Levee topographies along the main stem and floodplain channels are similar but cannot be explained by the observed exponential functions. Channel margin shear flow during extended periods of flooding may give rise to the localized levee deposition. Our study demonstrates that tie and tributary floodplain channels can inject large volumes of sediment-laden main stem waters great distances across the floodplain where they spill overbank, forming a narrow band of deposition, thereby creating a depositional web. Citation: Day, G., W. E. Dietrich, J. C. Rowland, and A. Marshall (2008), The depositional web on the floodplain of the Fly River, Papua New Guinea, J. Geophys. Res., 113, F01S02, doi:10.1029/2006JF000622.

1. Introduction ment discharge to oceans (and addressing biogeochemical cycling issues), and practical matters of river management. [2] A simple question that can be asked about any reach An important distinction must be made between sediment of river bordered by floodplain is: what controls the load diverted to the floodplain that contributes to net proportion of a river’s sediment load that is deposited on deposition or aggradation versus sediment that is deposited its floodplain? Answers to this question are crucial to on the floodplain, but is eventually re-entrained, typically understanding floodplain evolution, interpreting the strati- by lateral bank erosion [e.g., Dunne et al., 1998]. graphic record that floodplains preserve, estimating sedi- [3] The two primary mechanisms of floodplain deposition are lateral shifting (and bar accretion), and overbank depo- 1Rio Tinto, London, UK. sition during flood events [e.g., Wolman and Leopold, 2Department of Earth and Planetary Science, University of California, 1957]. Generally, lateral shifting leaves coarse bed material Berkeley, California, USA. (sand or coarser), while overbank deposits tend to be 3Andrew Marshall and Associates, Sydney, New South Wales, dominated by mud (except for local splay and levee Australia. deposits rich in sand) [Bridge, 2003]. A balance of processes may arise such that no net accumulation occurs [e.g., Wolman Copyright 2008 by the American Geophysical Union. 0148-0227/08/2006JF000622$09.00 and Leopold, 1957]. Bank erosion during lateral shifting

F01S02 1of19 F01S02 DAY ET AL.: DEPOSITONAL WEB F01S02

Figure 1. A 1990 satellite image of upper Middle Fly. Flow is from top to bottom, average channel width is about 320 m, and horizontal distance from top to bottom of the scene is about 11 km (north is at top of image). River entering from east is the Binge (see Figure 3). Note the tie channels connecting the main stem to oxbows and blocked valley lakes. may return coarse material to the channel, balancing bar Amazon above Manaus divert and capture on their beds as accretion on the opposite bank, and undermine overbank much as 37% of the total annual load. These channels shift sediments, thus offsetting overbank deposition elsewhere. and rework the floodplain surface, are active through a On aggrading river systems, however, channel bed level may range of discharge, and are not associated with avulsion rise with floodplain deposition [e.g., Bridge, 2003], elevating processes. the near-channel deposition zone relative to more distal [5] Dietrich et al. [1999] describe two distinct types of regions of the floodplain, and thereby creating the potential channels that disperse sediment great distances across the for significant flood currents from the channel across flood- floodplain of the Fly River in Papua New Guinea. Tie plain. This may lead to avulsions and the formation of channels, first described by Blake and Ollier [1971], are transient multiple channels that split and rejoin, forming an small (typically 0.1 the width of the main stem) channels anastomosing pattern [e.g., Makaske, 2001, Bridge, 2003; which form when sediment laden flow from the main stem Slingerland and Smith, 2004]. Slingerland and Smith [2004] enter the still water of oxbow and blocked valley lakes [e.g., argue that deposition associated with avulsion dominates Rowland et al., 2005; Rowland and Dietrich, 2006]. The floodplain formation of aggrading alluvial rivers. These second channel type is the low-gradient tributary channel deposits will tend to be mostly overbank mud [Makaske, from adjacent uplands. Flow reversals, i.e. flow both to the 2001]. The question of what controls the rate of deposition of floodplain from the main stem and from the floodplain to a river’s load on its floodplain raises questions about the the main stem channel, are common and apparently essen- controls on rates of channel bed aggradation as well as on tial to channel maintenance of tie and tributary channels. rates of lateral and vertical accretion. It also points to the role Sediment-laden main stem waters have been observed to of multiple channels in spreading sediment across the travel over 40 km up tributary channels and spread out floodplain. overbank onto the Fly floodplain. Figure 1 shows a satellite [4] Anastomosing and anabranching (in the sense of image of the Fly floodplain where two tributaries connect to Nanson and Knighton [1996]) rivers create multiple path- the main stem, and tie channels join oxbow lakes. Dietrich ways down which sediment is carried and dispersed across et al. [1999] proposed that such channels play an important floodplains, leading to distinct patterns of deposition and role in delivering sediment to the floodplain. potentially large rates of sediment extraction [e.g., Bridge, [6] Multichannel pathways across floodplains, especially 2003; Slingerland and Smith, 2004]. There are other types low-gradient aggrading ones, appear to be common. De- of channels across floodplains that also contribute to sedi- tailed stratigraphic studies of deltaic plains channels reveal ment dispersion and extraction. Crevasse channels are most the complex interweaving of channel and overbank deposits commonly mentioned and are ephemeral; that is, they and a growing number of field studies have documented operate primarily during flood events [e.g., Bridge, 2003]. stratigraphy associated with avulsive evolution of alluvial Dunne et al. [1998] propose that floodplain channels on the floodplains [Slingerland and Smith, 2004].

2of19 F01S02 DAY ET AL.: DEPOSITONAL WEB F01S02

[7] Here we make a distinction between avulsive and [10] Mining in the headwaters of the Fly (on the Ok Tedi anabranching floodplain systems, which generate primarily River) led to a large influx of mining waste to the river unidirectional flow down diverse channels that divide and beginning in 1985. This caused the sediment load to reconnect, and the tie and tributary system observed on the increase by about 4.5 times on the Middle Fly, leading to Fly River in which sediment-laden main stem flows inject progressive aggradation of the channel bed, increased flood- sediment great distances through nonmigrating floodplain- ing, and accelerated delivery of sediment to the floodplain. crossing channels. These channels typically experience It led to the delivery of copper-rich sediment across the bidirectional flow depending on the timing of main stem floodplain far above natural or background metal concen- hydrographs and rain in the lowlands. This injection process trations (in this initially pristine catchment), which enabled creates a distinct depositional web. sampling to trace the spread and rate of sediment accretion. [8] In this paper we build upon earlier work by Dietrich We will focus on floodplain sedimentation rates up to 1994 et al. [1999] to quantify for the first time the depositional and floodplain hydrology to 1998. At these times, aggrada- web created by tie and tributary channels across a flood- tion due to mine loading was concentrated near the Ok Tedi plain. We do so along the Middle Fly River, which has an junction and no discernable changes in process had occurred extensive floodplain channel system, and because of up- downstream. Since then, sweeping changes have occurred stream mining is discharging sediment with a distinct metal downstream. Though the basic processes described below rich signature, which allows us to document spatial and still pertain, the relative rates have shifted. Our description temporal patterns of deposition. Our field data show that of floodplain conditions only applies up to 1998. these networks of channels, with a total active length of [11] The focus of this paper is the Middle Fly Reach about 900 km, may convey as much as 20% of the total (Figure 3) bordered by the upstream junction with the Ok discharge of the Fly onto the floodplain, and, correspond- Tedi (D’Albertis Junction) and the downstream junction ingly, convey and deposit about 20% of the total sediment with the Strickland (Everill Junction). Through this reach load onto the plain as well. Overbank discharge of sediment the longitudinal profile along the river flattens from about along the 450 km main stem causes another 20% of the 6 105 to about 1 105 approaching Obo (Figure 4), sediment load carried by the river to be deposited on the and, associated with this decline, the median bed grain size floodplain. Although deposition is limited to a narrow band (premining) decreased from 0.3 to approximately 0.1 mm. bordering channels, the extensive floodplain channel net- In response to Holocene sea level rise, river aggradation work causes rapid delivery of sediment to remote parts of elevated the Fly River, blocking lowland tributaries and the floodplain. Overbank deposition equaled about 0.09% forming some 30 blocked valley lakes. The central scroll bar of sediment load deposited on the floodplain per km of main complex rises above surrounding back swamp areas, and stem channel per year. On the basis of a comparison with contains about 47 oxbow lakes (Figure 3). All of the the Strickland River [Aalto et al., 2008; Swanson et al., blocked valley lakes, and 29 of the oxbow lakes have 2008], we suggest that the tie and tributary driven deposi- distinct floodplain channels (tie channels) connecting the tional web process is associated with low gradient, wet, lakes to the main stems (see Rowland et al. [2005] and wide floodplains with relatively low sediment load. Rowland and Dietrich [2006] for more details about tie channels). Eight additional oxbows receive overbank flow 2. Study Site directly from the main stem at high flow. The total off-river water body area of 420 km2 is about 12% of the total [9] The Fly River system (Figure 2) consists of steep floodplain area (3500 km2). bouldery rivers that cut through the rapidly uplifting South- [12] The channel morphology and floodplain character- ern Fold Mountains of Papua New Guinea (reaching 4 km istics vary downstream. The first one third of the better in height) then progressively join downstream into two drained floodplain is covered in rain forest (pre-1997) primary main stem channels (the Fly and the Strickland which gradually shifts downstream to predominantly rivers). These two primary channels drop their gravel by swamp grass in the lower one third (we follow local and 900 and 730 kilometers, respectively, upstream from the previous usage in calling this area ‘‘swamp grass’’ even delta outlet and then transform into downstream fining, though the forested parts (which makes the term ‘‘swamp’’ meandering, sand-bedded rivers bounded by 4 to 14 km appropriate) are limited to levees and scroll bars). Corre- wide floodplains crossing intensively dissected lowlands spondingly, the channel narrows from about 350 m to 250 m which are underlain by an extensive carbonate platform before reaching Obo, point bars disappear by Manda, and [Pigram et al., 1989]. The rivers join at a bank elevation of the rate of lateral channel migration declines from about about 6 m, and the main stem heads downstream into a 3.5 m/a to 1 m by 200 km downstream. Where the channel progressively more tidally dominated flow as it enters into enters the swamp grass reach, though oxbows are common the Fly delta. Where the rivers join, the drainage of the Fly and well-defined scroll bars emerge, the channel has ceased 2 3 is 18,400 km and the average annual discharge is 2244 m /s, migrating laterally for at least 50 years [Dietrich et al., 2 while the Strickland area is 36,000 km and carries an 1999]. This cessation of migration is inferred to be due to 3 average annual discharge of 3110 m /s [Dietrich et al., reduced slope forced by either backwater effects of the 1999]. The natural annual load of the Middle Fly is Strickland aggradation or sea level rise, or by local effects estimated to be about 10 million tonnes, while that of the of tectonics [Dietrich et al., 1999]. Strickland is about 70 to 80 million tonnes. Only about 30% [13] Starting near the Binge River and running parallel to 2 of the 75, 000 km area of the entire Fly catchment is in the the Middle Fly along the eastern edge of its floodplain is the steep, uplands source area. Agu River, which lies lower than the Fly and drains it

3of19 F01S02 DAY ET AL.: DEPOSITONAL WEB F01S02

Figure 2. Fly River catchment showing general geology, major rivers, and catchment boundary (dashed lines). The numbers in parentheses next to site names are bank elevations in meters relative to Australian Height Datum (modified from Dietrich et al. [1999]). #1999 John Wiley & Sons Limited. Reproduced with permission. through five channels (Figure 3). The Agu River cuts Often the floodplain was inundated with dark, organic rich through bedrock and returns discharge to the Fly just north water, and even during flood events when the flow was well of Manda and the start of the swamp grass reach. While the above bank, the main stem Fly would follow the main stem floodplain channels connecting the Fly to Agu consistently channel path, hemmed in by the surrounding sediment-free drain toward the Agu, we have observed sediment laden floodplain waters (Figure 6). As Figure 5 suggests, by the flows injected over 40 km upstream in the Agu from its time flood pulses reach the lower Middle Fly, they are junction near Manda. These flows followed small channels largely dissipated. Instead, the primary cause of large off the Agu and onto the floodplain as well. We estimate discharge variation in the lower Middle Fly is drought (as that there are over 900 km of tie and floodplain tributary in 1997) driven by El Nino cycles. channels (a length approximately twice the length of the main stem Middle Fly) which transmit Fly River sediment 3. Hydrologic Monitoring and Rates of laden water to the floodplain. Discharge to the Floodplain [14] Rainfall in the headwaters of the Fly reaches 10 m/a and drops progressively downstream to about 2 m/a near [15] Sediment laden Fly River water entered the flood- Obo [Dietrich et al., 1999]. Peak flows entering the Middle plain by three paths: tie channels and tributaries (when flow Fly become strongly damped downstream (Figure 5), and is pushed up the tributaries), levee breaches, and overbank there were periods of over 18 months of continuous flood- flows. In order to document flow via the floodplain chan- ing in the lower Middle Fly during our monitoring period. nels onto and from the floodplain, water level recorders

4of19 F01S02 DAY ET AL.: DEPOSITONAL WEB F01S02

Figure 3. Depositional environments of the Fly River and lower Strickland River (modified from Dietrich et al. [1999]). #1999 John Wiley & Sons Limited. Reproduced with permission. using pressure gauges were installed along 13 tie and to maintain. The recorders were surveyed to a common tributary channels to record stage and slope. Additional datum through differential GPS survey with a vertical recorders were installed along the main stem and across the accuracy of 5 to 10 mm. The period of measurement floodplain, so that 65 instruments were operational starting spanned an El Nino drought that broke in 1998. Figure 7 December 1995. The recorders were set for 15 minute shows station locations for a portion of the floodplain. sampling intervals and collected data with variable success [16] The Manning relation for velocity coupled with until May 1998. Initially, Tuber Series 1 (M Squared monitored water surface slope and measured cross sections Technology) water level recorders were used, but short were used to calculate flux of water to the floodplain battery life caused us to also use Dataflow Systems Pty (following a procedure similar to that of Dunne et al. Ltd. equipment. Battery failure, instrument fouling, bank [1998]). Longitudinal profiles and cross-sectional surveys collapse and other processes made this system very difficult were conducted on 12 of the 13 tie and tributary channels

5of19 F01S02 DAY ET AL.: DEPOSITONAL WEB F01S02

Figure 5. Middle Fly River hydrographs recorded at Figure 4. Long profile of the Middle Fly River showing Kuambit (located just downstream of the confluence with bank elevations (right and left) and bed elevations (mean the Ok Tedi) and Obo (just upstream of the confluence with and minimum) based on differential GPS surveys. AHD Strickland River). The hydrographs show the strong refers to Australian Height Datum. Values on the plot show damping of peak flows as the river flows from forested progressive decrease in river slope from the D’Albertis reaches into the lower-gradient swamp reaches. The Junction (confluence of Fly River and the Ok Tedi) to hydrological effects of El Nino–induced droughts on river Everill Junction (confluence of Middle Fly and Strickland discharge are visible in the late 1997 measurements rivers). Thick vertical line marks the junction between the presented in the figure. At this time, Obo water levels Ok Tedi and Fly rivers. dropped to where tidal influences are significant. monitored. Along the Middle Fly River, tie channels typi- tributaries was about 15 to 24% of the main stemflow. cally exhibit maximum widths on the order of 25 m and The relative contribution varied greatly among the channels. depths of 5 m but vary from 15 to 39 m in width and from Figure 8 shows two cases, one near the upstream end of the 4.5 to 8.5 m in depth. The two main tributaries monitored, floodplain where, during the period of measurement, more the Binge and Agu, were 74 and 100 m wide and 8.9 and flow exited from the floodplain to the main stem, and the 8.4 m deep, respectively. A stage cross-sectional area other on one of the floodplain channels connected to the relationship was calculated for each channel (when flows overtopped tie channel banks, an effective cross-sectional area was calculated from the channel width and flow depth over the tie channel). Surveyed main stem stage range relative to bed height of the tie channels indicates that for 60% of the range in river stage water may flow through tie channels onto the floodplain. This agrees with visual impressions. Outflow via tie channels was commonly seen, even at relatively low stages, and at high stages, rain on the floodplain could induce outflow to the river. Measured water surface slopes ranged from 6 104 to +8 104 (+ is from floodplain to river channel), hence the steepest slopes exceed main stem slopes by a factor of 10. During episodes of equipment failure, tie channel water surface slopes were sometimes estimated by projecting the main stem water surface slope to the month of the tie channel in order to determine the relative difference in water surface elevations between the main stem and the floodplain. Manning’s roughness term, n, was estimated to be between 0.03 and 0.04. A time series of discharge was then calculated for each monitored channel. For unmoni- Figure 6. Photograph depicting sharp boundary between tored channels, discharge was estimated by assigning dis- sediment-laden river water and sediment-free, organic-rich charge rates from the most similar nearby monitored tie waters of the inundated floodplain of the Middle Fly. channel. Channels in the forested, transitional and swamp Despite river stages above bankfull, distribution of grass reaches were treated separately. sediment-laden water on the floodplain is limited to a [17] For the period of monitoring, calculated total dis- narrow band focused over the levees. The width of the charge outflow to the floodplain via tie channels and Middle Fly downstream of the confluence is approximately 250 m.

6of19 F01S02 DAY ET AL.: DEPOSITONAL WEB F01S02

Figure 7. Map of a portion of the Middle Fly River (see inset) showing the network of water level recorders established to monitor flow to and from the main channel and the floodplain. Dark areas are blocked valley lakes, and the last two letters in the station location names indicate whether the station monitored tie channel (TC) or floodplain (FP) water levels.

Agu River, the flow was nearly always toward the low-lying whereas in 1996 the estimated value was as high as 28% of tributary. the main stem discharge while in 1998 it was as low at [18] Levee breaches are transient features associated with 2.5%. Overbank flow was calculated to be highest in the slumping, which are erased by subsequent sediment infill- forested reach where greater differences between flood ing. During a survey in 1997, 27 levee breaches were height and floodplain water levels occurred. counted in a 40 km length, and one breach dimension was [20] Field observations through the drought and subse- surveyed with a cross-sectional area of 16 m2. Assuming quent discharge rise showed a distinct stage-dependent this density of levee breach along the entire Middle Fly, we pattern of river water injection onto the floodplain. For an estimate the number of such breaches to be 190. Discharge initially dry floodplain (most common in the forested was estimated using the Manning relation, which was reach), initial stage rise would drive water out through the applied assuming that the water surface slope through the tie channels and tributaries, filling the oxbows and blocked breach was the same as that measured at the nearest tie valley lakes. Continued stage rise caused water to spill out channel slope. The small size of the breaches and the of the blocked valley lakes into surrounding channel net- relatively rare frequency of occurrence lead to a small (less works. With further stage rise, water spilled out through the than 1%) contribution to discharge to the floodplain. levee breaches and eventually over the levee tops. In a two [19] Overbank flow was calculated to occur when water week period following the onset of rain in December 1997, surface elevation exceeded bankfull height. As part of a the river rose rapidly in the upstream forested reach and regular monitoring program conducted by Ok Tedi Mining forced water out the tie and tributary channels, but by the Limited, 94 cross sections have been repeatedly surveyed time the flood wave had reached the lower swamp reach, the along the length of the Middle Fly (and tied into a common blocked valley lakes and floodplain behind the levees were datum). Water level records along the river were projected already flooded because of tie channel flows and direct rain between successive cross sections, and, for periods when on the floodplain. As a result of floodplain inundation in the flow height exceed bank height, the nearest tie channel lower reaches, there was little surface gradient outward water surface slope was used to calculate the slope toward across the plain when the main channel stage exceeded the floodplain. Channel bank length, overbank depth and bankfull conditions. Mertes [1997] describes this process of the Manning’s equation for velocity were used to estimate floodplain inundation due to direct precipitation in several discharge. Attempts to measure the flood flow velocities in other rivers. the field failed, so estimates of Manning’s n were crudely [21] Although the discharge values are approximate, both set to 0.07 to 0.1. In the drought year of 1997, these quantitative assessment and visual impressions during stage calculations correctly assess essentially no overbank flow, rise suggest that there is significant Fly River water dis-

7of19 F01S02 DAY ET AL.: DEPOSITONAL WEB F01S02

Figure 8. Plots showing magnitude and duration of flow to (positive discharge) and away from (negative discharge) the Middle Fly River. The top plot (T0102TC) connects to an off-river water body (ORWB), and inflows and outflows are nearly balanced. The other location (T0701TC) records flow through a channel connected to the Agu River. At this location, flow is predominantly away from the Fly to the Agu. charge into its adjacent floodplain. One consequence of sampling followed a protocol (applied particulate level or discharge is a great dampening of the flood waves down- APL) established with government agreement [OTML, stream (Figure 5). Our data suggest that at least 20% and 1993a, 1993b] but, when it was discovered that the sedi- perhaps as much as 40% of the discharge flows out onto the ment was rapidly crossing the floodplain, a more intensive floodplain. Some of this water is stored in lakes and sampling effort was initiated. The sampling was done evaporates, but much returns at lower stages. The returning annually from 1990 until 1994. waters are largely sediment free because of deposition and [23] The APL sampling program, started in 1990, con- trapping of sediment in lakes and on the floodplain. These sisted of 100 floodplain samples and 100 lake samples sediment-poor waters can appear as distinct inputs of black (oxbows and blocked valley lakes, referred to as off-river water that mix with turbid waters of the main channel at water bodies or ORWB). The floodplain was divided into confluences with tie and tributary channels. The sediment 10 equal area reaches (zones) from the junction with the Ok that does return to the Fly River commonly results from Tedi River to the junction with the Strickland River. Within localized scour of tie channel beds because of high hydrau- each reach the floodplain was divided into five approxi- lic gradients and velocities generated during return flow mately equal area strips that ran parallel to the valley axis. periods [Dietrich et al., 1999]. This created 50 sample cells (5 cells per ten reaches) (Figure 9a). Two sediment cores were collected from 4. Sediment Coring and Rates of Floodplain floodplain sites and two from ORWB sites in each of the Sedimentation sample cells (a total of 100 samples each of floodplain and ORWB sites). All sampling was accomplished with a float 4.1. Field and Laboratory Methods helicopter and location was established using GPS (at that [22] A campaign to monitor the spread of mine contam- time accurate to approximately 100 m). In 1990–1992, inated sediment across the Middle Fly floodplain began in shallow cores were collected and the top 5 to 10 mm sample 1990, 5 years after the onset of dumping into the river. A was analyzed for particulate copper concentration (pCu). hydrologic network and suspended sediment monitoring This shallow coring approach was taken under the incorrect program had been established from the mine site to Obo assumption that deposition rates would be relatively slow [Ok Tedi Mining, Limited (OTML), 1993a]. The initial

8of19 F01S02 DAY ET AL.: DEPOSITONAL WEB F01S02

established at approximately 10 km intervals down the floodplain and samples were collected across the floodplain, where possible, at 50 m, 100 m, 250 m, 500 m and then every 500 m to the floodplain boundary (Figure 9). Such sampling could not be conducted on the portion of the floodplain that lies in Indonesia. Up to 30 cores were taken in individual transects. Additional cores were also taken in the ORWB. Within each oxbow 3 cores (in the middle and 10 m from each bank) were collected from 3 transects across each oxbow (both ends and the center of the oxbow), i.e., 9 cores per oxbow. The APL and oxbow samples are those points lying off the transects shown in Figure 9b. The 17 transect sites were surveyed in with kinematic GPS by Andrew Marshall and Associates, Sydney, Australia in the same program that surveyed the water level recorders. [24] Cores taken in 1993 and 1994 (both transects and APL sites) were deeper, reaching 1 m in some cases. These longer cores were collected when it was realized that the depth of burial greatly exceeded initial expectations. Sub- samples of these cores could then be used to obtain an inventory of elevated pCu with depth. The coring device was either a specifically made hand corer or a commercial gravity corer, and both consisted of 50 mm diameter stainless steel tube with interchangeable polycarbonate core liners. In 1996, longer cores were obtained in seven ORWB using a pneumatically driven vibrocorer, with sample re- covery up to 5 m. [25] Subsamples collected from cores were transferred to plastic bags and frozen at the end of each day of collection. Analyses followed OTML APL protocol [OTML, 1993b]. Once in the laboratory, samples were thawed, wet sieved through 100 micron sieves (to focus on copper rich mine- derived sediment which was less than 100 microns). The finer sediment was dried at 104°C, digested in hot agua regia (HNO3 and HCl, at ratio of 1 to 2) and the pCu analyzed by flame atomic absorption spectroscopy. Elimi- nation of the >100 micron sediment had little effect on the analysis as there was little sediment in this size fraction. Sample analysis of suspended sediment collected in 1992 showed that pCu was relatively constant (about 500 mg/g) between 70 and 15 microns and then rose to about 840 mg/g for 10 to 5 micron sizes and 1840 mg/g for less than 5 micron size sediment. [26] In 1995, surface samples from the transects were collected for grain size analysis. Air dried samples were placed in plastic bags and crushed by hand to remove large aggregates and a subsample suspended in water and placed in an ultrasonic bath. These samples were then analyzed Figure 9. Floodplain sample locations for the initial APL using a Coulter LS130 laser diffraction analyzer with a size program and subsequent transect method: (a) detail of upper range sensitivity of 0.4 to 900 microns. If organic matter Middle Fly at the junction with the Ok Tedi River showing was visibly present, the sample was first combusted at the APL system of zones and sample strips (delineated by 550°C for 8 to 12 h. If coarse tails in the grain size dashed lines) and (b) map of entire Middle Fly River distribution were observed in uncombusted samples indi- showing APL points and transect sample locations. Labels cating the presence of organic matter, these samples were (T1, T2, etc.) indicate transect identification numbers. then also combusted and reanalyzed. Sample sites not along transects record the location of the [27] Suspended sediment samples were collected with original 200 APL sites and the additional oxbows analyzed. Niskin samplers at four locations in 1994 and at 7 locations in 1997. During the 1997 sampling, multiple vertical and there would be no vertical mixing. Starting in 1992, profiles were conducted at each sampling cross section coring was conducted at evenly spaced points along trans- and velocity measurements were made in conjunction with ects (in addition to the APL samples) across the floodplain. the Niskin sampling. The sampling effort spanned river By 1993 and continuing into 1994, 17 transects were stages at and just below bankfull, capturing both rising and

9of19 F01S02 DAY ET AL.: DEPOSITONAL WEB F01S02

Background values, based on deepest core samples preceding mining, averaged about 31 mg/g in mineral rich samples and 15 mg/g in organic rich samples. Hence commonly measured values between 65 and 250 mg/g (about 1/3 of all samples analyzed, for example in 1993) have pCu too low to be due to temporal changes in contaminated suspended sediment. It should be added that samples taken in 1994 showed near-surface values in this range, despite the well- documented high pCu in the source suspended load. [29] The concern about postdeposition remobilization of copper led to a field study of pore waters at several locations across the floodplain in 1993. Pore waters were found to be highly reducing below the ground surface with minor dissolved copper in the water. This favors formation of insoluble copper sulfides and limited mobilization. Mobili- zation has become a central issue on the aggraded, seasonally drained levees since this study [OTML, 2006]. For the period of this study, we can conclude that postdepositional movement of copper was minor. [30] Because of the sudden increase and dominance of sediment load on the Fly from mine-derived sediment and the high particulate copper content of that load, we argue that core samples with elevated pCu values that lie well below the incoming contaminated values record the effects of mixing between the contaminated and premine sediments on the floodplain. Field evidence of such mixing was plentiful. During seasonal drying of the floodplain, the surface would crack deeply, allowing contaminated surface material to tumble deep into the premine sediment deposit below. Vigorous vegetation growth (relative to the modest Figure 10. Schematic diagram of a sediment core with rates of sedimentation) would tend to move elevated pCu mixing between incoming Cu-rich (black) and premine, Cu- downward. Given the large disparity between background poor (white) floodplain sediments. The diagram on the left and mine-derived sediment, relatively minor vertical mix- shows graphically the effect of mixing on measured pCu ing can greatly increase the signal at depth. This suggests (areas of gray). The diagram on the right provides that, to calculate sedimentation rate, we need to get an definitions of the components of the mass balance inventory of elevated pCu with depth and calculate the calculations used to determine floodplain deposition rates equivalent thickness of deposit at the concentration of the on the basis of the Cu inventory of the core as described in incoming sediment. Middelkoop [2002] introduces a coupled the text. deposition-mixing model to account for the redistribution effects of postdeposition mixing on metal concentration and sedimentation rates along the lower River Rhine. Our falling stages. Suspended sediment samples were decanted shorter period of metal contamination allows us to take a to concentrate the sediment and run untreated in the laser simpler approach, as follows. analyzer to determine grain size. [31] Figure 10 shows a schematic representation of a core subsample of length L (for example, the top 10 mm of a [28] Between 1990 and 1994 a total of 423 floodplain s and 420 ORWB sites were cored, leading to approximately 1990 core) with an average pCu concentration, es, and an 4000 samples analyzed for pCu (note the APL 1994 were average bulk density of rs through the sample. The thick- not analyzed correctly and not used here). In those cores in ness of deposited copper-rich (unmixed) sediment in the which pCu was mapped through the vertical, concentrations subsample (Lc; Figure 10) can then be determined on the generally decreased with depth. This could arise for three basis of the mass balance: reasons: (1) increasing concentration with time arriving at the core site, (2) postdeposition mobilization of copper, and Ls es rs A ¼ Lc ec rc A þ ðÞLs Lc eb rb A ð1Þ (3) postdeposition vertical mixing by physical and biotic disturbance. Suspended sediment measurements indicate where Ls is the length of the subsample analyzed; es and rs that from 1990 to 1994 pCu increased from 900 to 1300 mg/g represent the average pCu and bulk density of the core near the Ok Tedi junction and from 700 to1200 mg/g at subsample throughout Ls respectively; A is the cross- Obo. Earlier samples were less systematically collected but sectional area of the core; Lc is the length or thickness of indicate a pCu closer to 800 mg/g near the Ok Tedi junction deposited copper-rich sediment; eb and rb are the back- in the late 1980s. These values are all greatly in excess of ground pCu and bulk density of premine floodplain background values and indicate an increase in time and a sediment; and ec and rc are the average pCu and bulk decrease downstream in the pCu delivered to the floodplain. density of deposited copper-rich sediment.

10 of 19 F01S02 DAY ET AL.: DEPOSITONAL WEB F01S02

Table 1. Particulate Copper Concentration in the Suspended pCu as discussed below) in the upstream reaches, and a Sediment Load of the Middle Fly River dilution by natural sediment delivered from the tributaries 1990 1991 1992 1993 1994 and bank erosion along the Middle Fly. Table 1 lists the pCu Nukumba 900 900 1000 1100 1300 used for ec for each sampling year. These data were Obo 700 700 800 1000 1200 obtained from OTML reports [e.g., OTML, 1993a, 1993b] aUnits are mg/g. and resampled to coincide with the years between the coring program rather than the calendar year averages reported by OTML. The mean suspended sediment pCu data from [32] The bulk density of the core, r , varies according to Nukumba were used for ec for all cores within the forest s reach and the data for Obo were used for cores from the the relative contributions of rc and rb by transition and swamp reaches. [35] Given the spatial and temporal dependency of ec rc Lc þ rb ðÞLs Lc i rs ¼ ð2Þ shown in Table 1, a consistent reach and time-dependent Ls scheme was adopted for assigning which value of eci to use Rearrangement of equation (2) into (1) allows calculation of for each depth interval (zi). We assigned 1994 values to the the copper-rich sediment thickness (L )via surface samples in the cores and then progressively earlier c incoming values for incremental samples at depth. For most samples from 1993 and 1994, analyses were performed at Ls rb ðÞes eb Lc ¼ ð3Þ depth intervals below the surface of 0–5 mm, 20–25 mm, ðrcðÞec es rbðÞeb es 60–65 mm, (and if cores went deeper) at 100–105 mm, 205–210 mm or the deepest part of the core. In such a The calculated value of Lc, the date the core was collected, sequence we assigned 1994 to the first layer, 1993 to the and the time the mine had been in operation were then used second, 1992 to the third, and so forth with the lowest value to determine the rate of sediment deposition over that having an eci set equal to that of 1990. If the incoming interval. This procedure is incomplete, however, if the concentrations preceding 1990 were lower, then our depo- contamination is spread down the core with varying sition rates are an underestimate. concentrations. [36] In about 20 of the samples, the pCu concentration of [33] The above method can be modified to calculate the the surface sample was greater than the average incoming thickness of copper-rich sediment in a core with elevated value of ec (Table 1). In these 20 cases the value of Lc was pCu over several depth horizons. Each horizon subsample 1 i set equal to Lsi in equation 4. Possible explanations for esi > of length Ls has an average pCu of es . Assuming that each i i ec may include a greater percentage of fine grained sedi- value of e represents the mean pCu averaged over half the i si ment in the sample (pCu increases with decreasing grain core length between each of the immediate upper and lower size, as discussed above), or variation (above the annual ec horizon subsamples (zi), we used equation (4) to estimate mean) of the pCu in the main channel as a result of specific the total thickness of copper-rich sediment (LCu) throughout flow conditions during a flood event, or short-term operat- the entire core by summing the individual L values for ci ing fluctuations at the mine. The increase in pCu on finer each zi to lowest sampled horizon, n, sediments affects the deposition analysis if the percentage of X the finest grain sizes; that is, the fraction <5 mmis n zi rbðÞesi eb LCu ¼ ð4Þ preferentially deposited across the floodplain. As shown i¼1 ðrcðÞeci esi rbðÞeb esi below, beyond the channel levee deposited sediment grain size varies relatively little across the floodplain. Further-

If esn > eb (i.e., the core had not sampled all the way more, the deposited sediment grain size distribution closely through the copper-rich sediment), a value of zi for this mirrors that in the suspended load of the Fly River and tie particular horizon subsample was rounded up generally to channels. While there are undoubtedly areas across the the next 50 or 100 mm on the basis of a review of the pCu floodplain where preferential deposition of fine-grained through the core profile. The average background bulk high pCu sediment does occur, it appears that it is not a 3 density, rb, was either 1.1 gm/cm for mineral rich deposits widespread occurrence and therefore the impact of the 3 or 0.5 gm/cm for organic-rich deposits. Similarly, the higher pCu fine-grained sediments was regarded to be background pCu (eb) varied between these two sediment minor. types. On the basis of deep and distant samples, for mineral [37] Equation (4) was used to calculate sediment deposi- rich samples eb is approximately 31 mg/g and for organic tion since the start of mining from the pCu profile data from rich samples it is 15 mg/g. 1993 and 1994 at all coring sites. The blocked valley lake [34] As mentioned above, monitoring results reported by deposits were included in the floodplain data set because OTML reveal that the pCu of Fly River suspended sediment they tended to dry up seasonally, acting more like floodplain (ec) was not constant, varying both temporally (increasing surfaces both in depositional patterns and mechanics (wet- concentration between years) and spatially (decreasing ting, drying and mixing). Oxbow lakes remain partially downstream). The increase between years was likely due inundated throughout the year and were treated separately to increased ore and waste rock throughput from the mine from the floodplain data. Application of equation (4) to the and the gradual saturation of background sediment in the oxbow samples was problematic because the majority bed of the river with copper-rich sediment. The decrease (71%) of oxbow cores sampled did not penetrate the entire along the length of the Fly possibly reflects a combination depth of copper-rich sediment. For those oxbow cores the of deposition of fine-grained sediment (generally higher in value of zn was rounded up to the next 50 mm increment.

11 of 19 F01S02 DAY ET AL.: DEPOSITONAL WEB F01S02

majority of oxbow samples yield a minimum estimate of deposition thickness. [38] For the majority of oxbows sampled, cores were collected from between 1 and 9 locations. In order to determine the total amount of sediment deposited in each oxbow the mean value of Lcu from all cores in each oxbow was calculated and applied across the area of each particular oxbow. Not all oxbows were cored, such as those on the Indonesian side of the border. An estimate of the total deposition across all oxbows was calculated by using the total amount of deposition calculated in those oxbows that were cored and multiplying that value by the ratio of total oxbow area to cored oxbow area. The total area of cored oxbows was 24.9 km2 and the total of all oxbows was 34.3 km2, which required the total deposition volume from the cored oxbows to be multiplied by 1.4 to estimate the total deposition in all oxbows. The computed inflow from tie channels discussed above was also used in combination with estimated suspended sediment concentrations to estimate oxbow deposition rates for comparison. Figure 11. Histograms of pCu in shallow floodplain cores between 1990 and 1993. The data show a progressive 4.2. Sedimentation Patterns increase in floodplain pCu with time. In 1990, only 3 years [39] The floodplain sampling reveals significant gradients after the first significant release of mine waste, half the in the spatial and temporal patterns of deposition. Figure 11 floodplain samples exhibited elevated pCu. The rapid presents histograms of the 200 APL pCu values from 1990 dispersal of copper across the floodplain illustrates the to 1993 (shallow cores only). Even though significant effectiveness of the depositional web to widely distribute release of mine waste didn’t begin until 1987, by 1990 sediment tens of kilometers from the main channel. nearly one half of all the floodplain samples and two thirds This procedure was repeated for all cores not sampling the of the ORWB samples had elevated pCu above background values of 15 to 31 mg/g. Figure 12 shows the difference entire depth of copper-rich sediment and the value of Lcu calculated using equation (4). Hence the calculations for the between 1990 and 1993 for surface samples of all cores collected in the lower Middle Fly. These data reveal a

Figure 12. Distribution and magnitude of pCu in floodplain surface samples along the Middle Fly River in 1990 and 1993. Comparison of the two maps highlights both the downstream progression of pCu distribution along the Middle Fly and the increase in pCu across the entire floodplain system. The junction of the Middle Fly and Strickland rivers is located in the bottom right of each image.

12 of 19 F01S02 DAY ET AL.: DEPOSITONAL WEB F01S02

old), Rowland [2007] estimates that oxbow lakes captured approximately 1.5% of the premine sediment load of the river. [41] Figure 13 shows sediment deposition (derived from equation (4)) as a function distance from the nearest channel for the sample year 1993. Although the transects were designed to sample at intervals from the channel bank of the Fly, we found that often our cores would be closer to a floodplain channel, and, correspondingly the sedimentation would be higher. Therefore in Figure 13 we use the shortest distance to any channel as a measure of travel distance across the floodplain. Three striking features stand out in Figure 13. First, there is a well-defined exponential decline in sedimentation with distance from the channel bank. Second, the sedimentation abruptly declines at 1 km from the channel and remains roughly constant farther out. Third, the variance in deposition rate declines sharply with distance from channel bank. Figure 13. Cumulative deposition of 7 years of mine- [42] The exponential function for the first 1 km from the derived sediment from all channel types as of 1993. The channel bank shown in Figure 13 takes the form, d = aebx , inset shows the exponential decrease in deposition rate for in which d is the total sediment deposition (in mm) since the first 1 km in the forested reach. start of elevated mine-derived sediments entered the system up to the time of sampling, a is the deposition at the bank (x =0),b has units of km1 and x is the distance from the downstream advance of the contaminants, a large increase channel bank (km). Regression analysis for this first 1 km in pCu with time, but at values below suspended sediment was done separately for the forested, transitional and swamp source concentrations, and a rapid lateral transport across grass reaches and for deposition from the main stem versus the entire floodplain (over 10 km in just 4 years). The latter the floodplain channels and ORWB (these lakes, connected observation was most surprising and careful inspection of by tie channels to the main stem also became sources of the sample sites shows that this rapid spread was associated sediment when the lakes filled with water; Table 2). The with dispersion via floodplain channels (tie and tributary insert in Figure 13 shows the regression for the forested channels). reach for 1993 for the first 1 km. Regressions for 1993 [40] The total sediment flux to oxbow lakes is poorly showed significant differences in deposition between the constrained because the shallow cores consistently failed to three floodplain reaches. Across channel types only in the reach uncontaminated sediment after the first few years. swamp reach did the exponential function differ significantly Using the 420 shallow cores collected as part of the APL between tie channels and the main stem. In 1994, no program and estimating rates in unmonitored lakes, we significant differences were found between floodplain estimate the minimum deposition to be about 8.5 Mt channel types (perhaps due to reduced data density as a (megatons) up to 1994. As a separate analysis, we used consequence of flaws in the 200 APL laboratory analyses). the discharge calculated into the oxbow lakes and estimated No correlation was found between deposition rate and the suspended sediment concentration entering the lakes to relative elevation on the floodplain, unlike that proposed arrive at a value of about 19 Mt. These deposition values by Howard [1992] and Middelkoop and Asselmann [1998]. represent 2.4 to 4.2% of the total sediment load that entered [43] Integration of the exponential function out to 1 km the Middle Fly River between 1985 and 1994. In an for each distinct channel type and floodplain reach and independent analysis examining the volume of sediments division by the depositional length (1 km) gives the average deposited in two dated oxbow lakes (350 and 900 years deposition amount for the active portion of the floodplain.

Table 2. Regression Results of Fits of Exponential Function (d = a*exp(bx)) for All Samples Collected in 1993 and 1994 Distinguished by Floodplain Type and Closest Channela a,mm b, 1/km R2 Forest Trans. Swamp Forest Trans. Swamp Forest Trans. Swamp 1993 Data All channels 243 165 139 3.79 3.69 3.23 0.96 0.97 0.85 Fly R. only 243 169 128 3.78 3.66 2.92 0.95 0.96 0.81 TC/ORWB only 245 159 179 3.82 3.60 3.77 0.94 0.98 0.89

1994 Data All channels 288 169 122 3.64 3.18 2.45 0.92 0.90 0.75 Fly R. only 307 187 113 3.62 3.20 2.04 0.91 0.87 0.70 TC/ORWB only 204 119 187 3.36 2.70 3.59 0.94 0.93 0.93 aFloodplain types are forest, transition, and swamp. Closest channel is main stem Fly or floodplain channel, i.e., tie channels and tributaries (TC) and off river water bodies (ORWB). The distance from the channel, x, is in km, and a is deposition in mm since start of mine loading.

13 of 19 F01S02 DAY ET AL.: DEPOSITONAL WEB F01S02

Table 3. Sediment Budget of the Fly Rivera Wall Suspended Mine-Derived Year Tailing, t 106 Waste Rock, t 106 Erosion, t 106 Sediment Load,b t 106 Sediment Load,c t 106 1985 2.38 3.80 0 17 7 1986 3.10 9.50 0 22 12 1987 9.72 11.80 0 32 22 1988 14.93 23.40 0 51 41 1989 23.60 29.70 0 53 43 1990 27.42 27.80 1 62 52 1991 27.01 27.80 19 49 39 1992 26.74 21.96 24 51 41 1993 28.62 14.03 16 56 46 1994 29.71 27.47 22 62 52 Total 193 197 82 455 355 aTailings, waste rock, and wall erosion (erosion due to waste dumping) define the input to the Ok Tedi River. bShown is the total suspended sediment load entering the Middle Fly at D’Albertis Junction from the Ok Tedi and Upper Fly. cShown is the total mine derived sediment input to the Middle Fly (assuming an annual background load to the Middle Fly of 5 106 t each from the Ok Tedi and Upper Fly). Data are supplied by Ok Tedi Mining Limited.

By 1994, the deposition ranged from 77 mm in the forested system and the load entering the Middle Fly. Comparison reach to 48 mm in the swamp reach. Dividing by the period with the GIS derived sedimentation rates indicates that of of significant mine waste addition to the river (7 and 8 years the 455 Mt in the sediment load from 1985 to 1994, 164 Mt for 1993 and 1994, respectively), the average deposition or 36% of the total load entering the Middle Fly was rate across the 1 km active zone of the floodplain ranged deposited onto the floodplain as overbank deposits. Ac- from 9.6 mm/a in the forested reach to 6.0 mm/a in the counting for loss into the ORWB and recognizing that the swamp reach. mine derived tracers probably did not show up until 1987 [44] Using GIS coverage of the Middle Fly and the (thus lowering the total sediment to 416 Mt) would raise this exponential relations listed in Tables 2 and 3 we calculated value to at least 40%. The difference between the sediment the total overbank deposition onto the floodplain. The GIS load onto the floodplain between 1994 and 1993, 31 Mt, is model used 20 m by 20 m grid cells within the floodplain the net deposition for the year 1994. The suspended boundary, and excluded channels, oxbows and elevated sediment load entering the Middle Fly estimated from areas deemed above the common floodplain inundation monitoring data was 62 Mt in 1994, which means the level. The locally appropriate exponential function was floodplain has a trap efficiency of 50%. This higher value applied along the main stem and active floodplain channels may reflect the effects of aggradation, which by 1994 had throughout the Middle Fly floodplain. Deposition beyond caused the bed at Kuambit to rise about 2 to 3 m, leading to 1 km was set equal to the observed 1993 amount (0.5 mm) more frequent and longer-duration flooding. and 1994 amount (0.7 mm) as appropriate. Where channel curvature causes 1 km strips to overlap, only one rate was 5. Discussion permitted and deposition was eliminated if the 1 km strip crossed into an adjacent channel. The cumulative mass [47] The approximately 40% loss of sediment to the deposited by 1993 is calculated to be 110 million m3 or floodplain by overbank deposition matches the indepen- 133 Mt (for a surface bulk density of 1.2 t m3) and in 1994 dently estimated discharges of water to the floodplain. it is 137 million m3 or 164 Mt. We checked these calcu- There are large uncertainties in the sediment flux and lations by computing the mean deposition and multiplying discharge determination but it seems safe to say the number by the estimated channel length. This manual method gives is at least 20% and could well exceed 40%. An earlier about 25% higher amounts because it does not consider the reported loss rate (3%) by Dietrich et al. [1999] was based effects of channel curvature on deposition rate, nor does it on the early APL results before it was realized that the cores exclude elevated areas or oxbows. were too shallow, hence this rate is clearly wrong. The 40% [45] Figure 14 shows the resulting predicted pattern of deposition rate is equivalent to 0.09% of sediment load deposition from the GIS model (1993 data) for a portion of deposited overbank on the floodplain per km of main stem the Middle Fly at the upper end of the swamp reach where channel length, or 0.03% of the load deposited per km total the Agu drains into the Fly. Sample points are shown as channel length (including tie and tributary pathways). As well. Most of the floodplain sedimentation occurs along a summarized by Swanson et al. [2008], the Strickland River relatively narrow corridor on either side of the main stem loses 0.05% of its sediment load as overbank deposition per and floodplain channels with the rest of the floodplain km, the Brahmaputra-Jamuna River deposits 0.07% per km, receiving very little sediment. The Agu River, which runs the Amazon below Obidos deposits 0.065% per km and parallel to the Fly and drains it in 5 places, is a major above Obidos about 0.1% per km, and the Mississippi pathway for sediment dispersion across the floodplain. This deposits about 0.02% per km. Data reported by Middelkoop demonstrates that tie and tributary channels act as distrib- and Asselman [1998] indicate that the river Waal, a distrib- utary systems forming a distinct depositional web. utary of the lower Rhine floodplain, captured 19% of the [46] Table 3 summarizes data derived from OTML reports sediment load on its floodplain in a December 1993 flood, that define the introduction of sediment by mining to the Fly or about 0.16% sediment load per km. Hence the Fly deposition rate per channel is relatively high on the basis

14 of 19 F01S02 DAY ET AL.: DEPOSITONAL WEB F01S02

Figure 14. Predicted pattern of deposition from the GIS model of the exponential functions (1993) for a portion of the Middle Fly at the upper end of the swamp reach where the Agu (along the eastern margin of the floodplain) drains into the Fly. The magnitude of deposition decreases from dark red to pink with distance from the channel; white areas are regions of the floodplain outside the influence of the depositional web. Sample locations are shown as dark dots. The main channel, tributary, and tie channels are shown in bright yellow. The margins of ORWB are outlined in gray. The channel network is shown at low water level to reveal the extensive network that extends into the blocked valley lakes (as shown in Figure 3). of the main channel length, but when the entire network of long reach downstream of D’Albertis junction to the head of deposition is accounted for, the rate is relatively low. the swamp grass reach, the premine average bank erosion [48] Aalto et al. [2008] and Swanson et al. [2008] rate is about 1 m/a (0.0045 channel widths per year). This emphasize that the Strickland River sweeps back into its erosion rate would cause about one half the natural load overbank deposits through bend migration, causing much of (about 10 Mt/a) to be exchanged with the banks and would the deposits to be returned to the channel relatively rapidly. return some of the overbank deposits. In the lower Middle Dunne et al. [1998] estimate that overbank deposition on Fly no migration was occurring, hence overbank deposits the Amazon was 1.7 times the load that arrived at the were not being returned to the channel. The extensive downstream gauging station of their study reach, but that a transport of sediment away from the main stem via tie similar amount was swept back into the channel during and tributary channels on the Fly causes much of the lateral migration. Unlike the Strickland and Amazon, how- overbank deposits to go into net storage (Figure 14). ever, the Middle Fly has very modest lateral migration rates. [49] We can obtain another estimate of floodplain depo- Dietrich et al. [1999] estimate that for the active 250 km sition rate from our observations by calculating what the

15 of 19 F01S02 DAY ET AL.: DEPOSITONAL WEB F01S02 natural sedimentation rate would have been and comparing (3–10), and by Tornqvist and Bridge [2002] for the Mis- that with other, longer-term values. By the time of the last sissippi delta (5.9). In contrast, Swanson et al. [2008] sediment survey for this project (1994), river-bed aggrada- compare overbank deposition patterns on the Strickland with tion was still modest and significant changes in flooding those reported for the Beni and Brahmaputra-Jamuna rivers regimes were not evident down much of the Middle Fly. and argue that an exponential poorly represents the rate of Therefore we can reason that the main change in sedimen- decline of deposition across the floodplain from the channel tation was due to higher suspended load concentrations bank. The cause for this departure from the trend found on rather than high flood frequencies (this is distinctly not true the Middle Fly and on rivers reviewed by Bridge [2003] is in more recent years). On the basis of the sediment budget not known. given in Table 3, the total load for 10 years entering the Fly [53] Cross-sectional surveys of both the main stem Mid- was about 455 M tonnes (Mt). Dietrich et al. [1999] dle Fly and tie and tributary floodplain channels reveal that estimate the natural load to be about 10 Mt/a at D’Albertis each have well-defined, relatively small levees that differ junction on the Fly, hence over 10 years the total load would little in dimensions. On the Fly the mean height for the three have been 100 Mt. This indicates an increase of a factor of reach types (forest, transition, and swamp) ranged from 1.2 4.6 in suspended sediment over background, a number that to 1.9 m on the main stem and was about 1.4 m on the is roughly similar to the concentration change from before floodplain channels in 1997. The levee widths varied from to during mining. This suggests that a reasonable estimate 56 to 93 m on the main stem, and 43 to 58 m on the of natural floodplain sedimentation rate would be to divide floodplain channels. These topographic similarities and the the observed rate by 4.6. The 9.6 mm/a and 6 mm/a average absence of another sediment source for the construction of deposition rates under current conditions for the first km in these active depositional features support the observations the forested and the swamp reaches, respectively, would be that the floodplain channels are significant conveyors of 2 mm/a and 1.3 mm/a under natural conditions. sediment from the main channel to the floodplain. [50] As summarized by Dietrich et al. [1999], various [54] Although there is clear evidence of lateral transport previous estimates based on a few dated deposits and out to about 1 km from any channel on the floodplain, the exposed sediments suggest that the average Holocene mechanisms responsible for this are not clear. During field sedimentation rate was about 1 mm/a, with the rate probably work on the floodplain, no detectable currents were docu- declining downstream [see Pickup, 1984; Pickup and Warner, mented. In the forested reach, we did see evidence 1984]. Recent seismic surveys on the Fly (B. Bolton, (deflected vegetation and sand deposits) of overbank flows personal communication), indicate that Holocene sediments in the outside of bends. We did not witness currents across areabout5to15mthick,againpointingtoaverage the swamp grass floodplain but, presumably, under the right accretion rate of about 1 mm/a. These rates are not corrected stage history (high flows with head gradients across the for postdeposition consolidation that would happen with floodplain), overbank currents develop. Three observations deeper burial. are more puzzling. First, we note that the exponential [51] Both the hydrologic monitoring and the sediment function predicts a decline over a distance of a kilometer, deposition mapping indicate high rates of water and sedi- but the levees bordering the channels are less than 100 m ment discharge to the floodplain via tie and tributary wide, and are commonly separated from the rest of the channels. About one half of the outflow to the floodplain floodplain by a shallow trough. Such troughs have also been occurs via these channels. The tie and tributary channels observed along the Mississippi River and attributed to local actively conveying sediment to the floodplain are approx- subsidence due to the loading of the levee deposits [Russell, imately twice the main stem channel length, and roughly 1939]. Other mechanisms associated with local flow pat- half of the overbank deposition occurs along these flood- terns may be responsible. Second, as shown in Figure 15, plain channels. Surprisingly, the exponential functions de- the grain size analysis of floodplain sediments shows only fining deposition away from the channels with distance are the levees to contain more than 5% sand, and yet some sand similar on the floodplain channels and main stem. Given is found out to 1 km. We confirmed the presence of sand in that the ultimate sediment source is the main stem, it would the distal samples by making thin sections of the sediment. seem reasonable to expect that overbank deposition should Beyond the levees there is no systematic fining across the be less on distal tie and tributary borders, but this is not floodplain; if anything, the data suggest a slight coarsening evident in the data. away from the bank. Third, as reported by OTML [2004] in [52] Bridge [2003], summarizing work by others especially the 6 year period between 1997 and 2002 of repeated Mackey and Bridge [1995], Middelkoop and Asselmann channel bank surveys at >130 sections along the Middle [1998], and Tornqvist and Bridge [2002], reports fits of an Fly, the average levee height grew by an additional 1 m exponential function for overbank deposition, d = aeb(x/xm) along the entire Middle Fly River in response to the in which xm is the maximum floodplain distance from the sustained mine-derived sediment load on the river. edge of the channel belt. Hence x in this case does not [55] The localized nature of the levee, which is inconsis- necessarily originate at the channel bank. By coincidence, tent with a description of the off-channel sedimentation as a our exponential fit is numerically the same for the evaluation single exponential function, suggests that the mechanism of of b because the Middle Fly deposition rate falls to nearly levee formation differs from that which distributes sediment zero at 1 km (their xm). The b values reported here of 2 to 3.8 farther out. The common occurrence of the Fly rising up (Table 2) are greater than those reported by Mackey and against floodplain waters (formed from direct rain and Bridge [1995] (0.35–1.4), and on the low end of values cited floodplain channel spillage) means that the lateral hydraulic by Middelkoop and Asselmann [1998] and Tornqvist and head is often low to zero even during high floods. This leads Bridge [2002] for studies on the Rhine-Meuse delta system to clearly observable shear zones between the sediment

16 of 19 F01S02 DAY ET AL.: DEPOSITONAL WEB F01S02

Figure 15. Plots of grain size distributions of surface floodplain deposits (first 10 mm) with distance from channels: (top) data for all channel types presenting a representative size distribution of the suspended sediment load of the Middle Fly River and data separated by whether samples bordered (middle) the Fly River or (bottom) tie channels and ORWBs. Measured grain sizes are plotted in five bin classes from <8 mm to sand size. The greatest percentage of sands is found closest to the channel (0–0.01 km), though sand is present at all distances from the channel. There is surprisingly little change in the size distribution with distance from the channel, and the data suggest that a slight coarsening of the distribution may occur away from the channel. Samples were collected in 1995 and, depending on proximity to channels, are a mixture of background and mine-derived sediment. laden main channel flows that track the channel and the load also points to an important sediment supply role in stationary organic rich flood waters (Figure 6). In this shear setting levee heights. zone, sediment laden waters will tend to rain loose sediment [56] Finally, it should be asked why there is a depositional along a corridor that defines the levee. This is a mechanism web on the Middle Fly River; that is, what circumstances similar to that described by Leighly [1934], modeled exper- favor its existence? We suggest that it results from overall imentally by Sellin [1964] and numerically by James low gradient of the valley, sea level rise, significant rainfall [1985], and incorporated into Adams et al.’s [2004] con- on the floodplain, and modest sediment load. Low gradients ceptual model of levee formation. Levee formation during favor low channel migration rate, large channel head extended periods of flooding with limited to no hydraulic gradients on the tie and tributary channels relative to head out of the channel may explain why there were no downstream slopes, poor drainage of the floodplain, and crevasse splay deposits on the Fly. Dispersion farther across frequent flow reversals in the floodplain channels. Sea level the floodplain may occur as eddies are shed outward from rise has forced a slope change and the modest sediment the channel [Adams et al., 2004], though such a processes load, relative to the large accommodation space created by has not been directly observed along the Middle Fly. The sea level rise, means that the channel is still responding to lack of cross-stream fining has been observed by others the sea level rise [Lauer et al., 2008]. The overall low [e.g., Marriott, 1992; Simm, 1993; Nicholas and Walling, gradient of the Fly valley is, to some degree also set by the 1996]. As these authors have suggested, perhaps the pres- carbonate platform over which the Fly is slowly aggrading ence of high dissolved organic carbon has contributed to and by still poorly defined effects of tectonics. Significant particle aggregation which leads to higher fine sediment floodplain rainfall can lead to extended periods of standing deposition near the bank. The rapid and uniform levee rise water on the floodplain that is not due to overbank flows, between 1997 and 2002 due to a large increase in sediment and can prevent significant overbank flows and cause tie

17 of 19 F01S02 DAY ET AL.: DEPOSITONAL WEB F01S02 and tributary channels to be primary paths of water for Dietrich, W. E., G. Day, and G. Parker (1999), The Fly River, Papua New Guinea: Inferences about river dynamics, floodplain sedimentation and floodplain exchange. The modest sediment load, as shown fate of sediment, in Varieties in Fluvial Form, edited by A. J. Miller et al., by comparison with the high sediment load on the Strick- pp. 345–376, John Wiley, New York. land River [Swanson et al., 2008], causes low rates of lateral Dunne, T., L. A. K. Mertes, R. H. Meade, J. E. Richey, and B. R. Forsberg (1998), Exchanges of sediment between the flood plain and channel of migration, slow depositional response to sea level rise, and the Amazon River in Brazil, Geol. Soc. Am. Bull., 110(4), 450–467. preservation of tie channels, oxbow lakes and blocked Howard, A. D. (1992), Modeling channel migration and floodplain sedi- valley lakes. mentation in meandering streams, in Lowland Floodplain Rivers, edited by P. A. Carling and G. E. Petts, pp. 165–183, John Wiley, Chichester, U. K. 6. Conclusion James, C. S. (1985), Sediment transfer to overbank sections, J. Hydraul. Res., 23(5), 435–452. [57] Independent data on floodplain hydrology and sed- Lauer, J. W., G. Parker, and W. E. Dietrich (2008), Response of the Strick- imentation rates demonstrate that floodplain channels (tie land and Fly River confluence to postglacial sea level rise, J. Geophys. Res., 113, F01S06, doi:10.1029/2006JF000626. and tributary channels) inject large quantities of water and Leighly, J. (1934), Turbulence and the transportation of rock debris by sediment onto the floodplain of the Middle Fly River. The streams, Geogr. Rev., 24, 453–464. total length of the floodplain channels is twice that of the Mackey, S. D., and J. S. Bridge (1995), Three-dimensional model of alluvial stratigraphy: Theory and application, J. Sediment. Res., Sect. B, 65, main stem, and about half the river discharge and sediment 7–31. load that enters the floodplain is conveyed through them. Makaske, B. (2001), Anastomosing rivers: A review of their classification, Sediment that is carried overbank from these channels origin and sedimentary products, Earth Sci. Rev., 53, 149–196. rapidly deposits, reaching less than 1 km either side of the Marriott, S. (1992), Textural analysis and modeling of a flood deposit— River Severn, UK, Earth Surf. Processes Landforms, 17, 687–697. channels. This leads to the formation of a distinct deposi- Mertes, L. A. K. (1997), Documentation and significance of the perirheic tional web across the floodplain. Because the floodplain zone on inundated floodplains, Water Resour. Res., 33(7), 1749–1762. channels disperse sediment far from the main stem, much of Middelkoop, H. (2002), Reconstructing floodplain sedimentation rates from heavy metal profiles by inverse modeling, Hydrol. Processes, 16, 47–64. the sediment deposited is going into long-term net storage. Middelkoop, H., and N. E. M. Asselman (1998), Spatial variability of Perhaps as much as 40% of the Fly River load is lost to the floodplain sedimentation at the event scale in the Rhine-Meuse delta, floodplain, which is equivalent to 0.09% of sediment load the Netherlands, Earth Surf. Processes Landforms, 23, 561–573. Nanson, G. C., and A. D. Knighton (1996), Anabranching rivers: Their per km of main stem channel length. cause, character, and classification, Earth Surf. Processes Landforms, [58] The overbank deposition along the main stem and 21, 217–239. floodplain channels follows a distinct exponential decline in Nicholas, A. P., and D. E. Walling (1996), The significance of particle rate with distance from the bank. The specific cause of this aggregation in the overbank deposition of suspended sediment on river floodplains, J. Hydrol., 186(1–4), 275–293. decline is not clear. Levees are narrow, often separated by a Ok Tedi Mining Limited (OTML) (1993a), APL compliance and additional shallow trough from the rest of the floodplain, contain environmental monitoring program: 1992 annual report, OTML Rep. ENV coarser sediment and cannot be explained by the general 93004, Tabubil, Papua New Guinea. Ok Tedi Mining Limited (OTML) (1993b), Fly River flood-plain copper exponential deposition function. We suggest that levees monitor, June 1993, OTML Rep. ENV 93-06, Tabubil, Papua New Guinea. emerge from deposition focused by localized shear during Ok Tedi Mining Limited (OTML) (2004), Summary report: Effects of mine periods of extended flooding. The dispersion of sediment life extensions and dredging on the Ok Tedi and Fly River—A sediment transport model study, environment section, September 10, 2004, 68 pp., across the floodplain, its lack of lateral fining, and the Tabubil, Papua New Guinea. persistence of minor sedimentation beyond about 1 km Ok Tedi Mining, Ltd (OTML) (2006), Ok Tedi Mining Limited 2005 an- require further explanation. nual review, 31 pp., Tabubil, Papua New Guinea. Pickup, G. (1984), Geomorphology of tropical rivers: I. Landforms, hydrol- [59] We suggest that the depositional web found on the ogy and sedimentation in the Fly and Lower Purari, Papua-New-Guinea, Fly would be favored by low-gradient channels, carrying Catena Suppl., 5, 1–17. modest sediment load (relative to accommodation space), Pickup, G., and R. F. Warner (1984), Geomorphology of tropical rivers: and experiencing relatively high runoff and wet floodplains. II. Channel adjustment to sediment load and discharge in the Fly and Lower Purari, Papua-New-Guinea, Catena Suppl., 5, s19–41. Pigram, C. J., P. J. Davies, D. A. Feary, and P. A. Symonds (1989), Tectonic [60] Acknowledgments. This work was supported by Ok Tedi Min- controls on carbonate platform evolution in southern Papua New Guinea: ing Limited (OTML). We especially thank OTML environment managers Passive margin to foreland basin, Geology, 17, 199–202. Murray Eagle and Eric Woods. Analysis of the data was also supported in Rowland, J. (2007), Tie channels, Ph.D. dissertation, 176 pp., Univ. of part by the National Center for Earth-surface Dynamics and an NSF Calif., Berkeley. Margins Source to Sink grant. Deanna Sereno and Dino Bellugi performed Rowland, J., and W. E. Dietrich (2006), The evolution of a tie channel, in GIS analysis of floodplain deposition, and Mark Stacey provided valuable River, Coastal and Estuarine Morphodynamics, edited by G. Parker and guidance. The paper benefited from reviews by Gary Parker, Desmond M. H. Garcia, pp. 725–736, Taylor and Francis, Philadelphia, Pa. Walling, Torbjorn Tornqvist, Mary Power, and Kathleen Swanson. Careful Rowland, J. C., K. Lepper, W. E. Dietrich, C. J. Wilson, and R. Sheldon editing by Michael Church greatly improved the manuscript. (2005), Tie channel sedimentation rates, oxbow formation age and channel migration rate from optically stimulated luminescence (OSL) analysis References of floodplain deposits, Earth Surf. Processes Landforms, 30, 1161–1179. Russell, R. J. (1939), Louisiana stream patterns, Bull, Am. Assoc. Pet. Aalto, R., J. W. Lauer, and W. E. Dietrich (2008), Spatial and temporal Geol., 23(8), 1199–1227. dynamics of sediment accumulation and exchange along Strickland River Sellin, R. H. J. (1964), A laboratory investigation into the interaction be- floodplains (Papua New Guinea) over decadal-to-centennial timescales, tween the flow in the channel of a river and that over its flood plain, J. Geophys. Res., 113, F01S04, doi:10.1029/2006JF000627. Houille Blanche, 7, 793–802. Adams, P. N., R. L. Slingerland, and N. D. Smith (2004), Variations in Simm, D. J. (1993), The deposition and storage of suspended sediment natural levee morphology in anastomosed channel flood plain complexes, in contemporary floodplain systems: A case study of the River Culm, Geomorphology, 61(1–2), 127–142. Devon, Ph.D. thesis, Exeter Univ., Exeter, U. K. Blake, D. H., and C. D. Ollier (1971), Alluvial plains of the Fly River, Slingerland, R., and N. D. Smith (2004), River avulsions and their deposits, Papua, Z. Geomorphol. Suppl., 12, 1–17. Annu. Rev. Earth Planet Sci., 32, 257–285. Bridge, J. S. (2003), Rivers and Floodplains: Forms, Processes, and Sedi- mentary Record, 491 pp., Blackwell, Malden, Mass.

18 of 19 F01S02 DAY ET AL.: DEPOSITONAL WEB F01S02

Swanson, K. M., E. Watson, R. Aalto, J. W. Lauer, M. T. Bera, A. Marshall, Wolman,M.G.,andL.B.Leopold(1957),Riverfloodplains:Some M. P. Taylor, S. C. Apte, and W. E. Dietrich (2008), Sediment load and observations on their formation, U.S. Geol. Surv. Prof. Pap., 282-C, floodplain deposition rates: Comparison of the Fly and Strickland rivers, 87–109. Papua New Guinea, J. Geophys. Res., doi:10.1029/2006JF000623, in press. Tornqvist, T. E., and J. S. Bridge (2002), Spatial variation of overbank G. Day, Rio Tinto, 6 St James’s Square, London SW1Y 4LD, UK. aggradation rate and its influence on avulsion frequency, Sedimentology, W. E. Dietrich and J. C. Rowland, Department of Earth and Planetary 49(5), 891–905. Science, University of California, Berkeley, CA 94720, USA. (bill@ eps.berkeley.edu) A. Marshall, Andrew Marshall and Associates, 43 Warrangarree Drive, Woronora Heights, Sydney, NSW 2233, Australia.

19 of 19