Modelling organic matter dynamics in headwater streams of south-western British Columbia, Canada

Modelling organic matter dynamics in headwater streams of south-western British Columbia, Canada

Ecological Modelling 183 (2005) 463–476 Modelling organic matter dynamics in headwater streams of south-western British Columbia, Canada O. Magnus Ka...

307KB Sizes 1 Downloads 16 Views

Ecological Modelling 183 (2005) 463–476

Modelling organic matter dynamics in headwater streams of south-western British Columbia, Canada O. Magnus Karlssona,∗ , John S. Richardsonb , Peter M. Kiffneyc a AF-Environmental Research Group, P.O. Box 8133, SE-104 20 Stockholm, Sweden Department of Forest Sciences, 3041-2424 Main Mall, University of British Columbia, Vancouver, BC, Canada V6T 1Z4 National Marine Fisheries Service, Northwest Fisheries Science Center, 2725 Montlake Blvd. East, Seattle, 98112 WA, USA b


Received 20 August 2003; received in revised form 12 August 2004; accepted 31 August 2004

Abstract A mass-balance model was developed to simulate organic matter (OM) dynamics in headwater stream ecosystems of southwestern British Columbia, Canada. Empirical data from two streams were used to structure and test a mass-balance model of the riparian–stream system. The model was driven by data on inputs, outputs, processing rates, discharge and water temperature. Statistical sub-models were derived for different processes (e.g. decomposition rates and periphyton growth). Inputs and outputs of OM were modelled on the basis of a series of assumptions of system properties, such as temperature and hydrological regimes. Major uncertainties identified through Monte-Carlo simulations of model predictions and variables important in controlling OM dynamics in these streams were dissolved OM (DOM) import and export, stream area and litterfall import. DOM was quantitatively the most important source of OM, accounting for 80% of total export of OM, followed by export of fine particulate organic matter (FPOM) at 15%. Different scenarios of logging and temperature regimes on the system were simulated to predict how these factors would affect standing stock of OM in the stream. When inputs of riparian litterfall were simulated to mirror reductions predicted from forest harvesting in the riparian area particulate OM (POM) standing stock was reduced by almost 80%. In comparison, a 3 ◦ C increase in water temperature resulted in only a 20% reduction of POM standing stock due to enhanced mineralisation. © 2004 Elsevier B.V. All rights reserved. Keywords: British Columbia; Detritus budget; Headwater; Modelling organic matter dynamics; Riparian; Streams

1. Introduction ∗

Corresponding author. Tel.: +46 8 657 12 88; fax: +46 8 657 37 57. E-mail addresses: [email protected] (O.M. Karlsson), [email protected] (J.S. Richardson), [email protected] (P.M. Kiffney). 0304-3800/$ – see front matter © 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.ecolmodel.2004.08.022

The linkages between streams and their terrestrial setting are of fundamental importance for the function of stream ecosystems (Minshall et al., 1985; Ward, 1989). Small streams in particular are dependent on the input of allochthonous material as an energy source


O.M. Karlsson et al. / Ecological Modelling 183 (2005) 463–476

(Minshall et al., 1985; Richardson, 1991; Wallace et al., 1999). Headwater streams also influence the ecology and geomorphology of higher order streams, because headwater streams carry water, sediment, organic matter, woody debris and nutrients, which determine the character of downstream habitats (Vannote et al., 1980; Gomi et al., 2002; Wipfli and Gregovich, 2002). In coastal forests of western North America, one of the major land use practices is timber harvest, which has the potential to alter the structure of the riparian–stream ecosystem (e.g. Naiman et al., 2000). In forested headwater streams, organic matter inputs from allochthonous sources are generally higher than in-stream primary production (Bilby and Bisson, 1992; Kiffney et al., 2000). There is a strong relationship between organic matter (OM) standing crop and invertebrate abundance, biomass and production, and also production of predators (Gregory et al., 1987; Richardson, 1991; Wallace et al., 1999). Thus, understanding factors (e.g. forest harvest) that influence organic matter dynamics is critical to managing lotic ecosystems. Removal of riparian vegetation will reduce the allochthonous input of OM to streams (Webster et al., 1990), but may stimulate aquatic primary production due to reduced shading (Gregory et al., 1991; Bilby and Bisson, 1992; Kiffney et al., 2003). Alteration of detritus standing crop and composition may have a strong bottom-up effect on the aquatic community and propagate through detritivores to predators (Hetrick et al., 1998; Wallace et al., 1999). Large Woody Debris (LWD) may also physically affect stream processes, as LWD can be important for detrital storage, substrate stabilisation (Murphy et al., 1986; Chamberlain et al., 1991), and forms important fish habitat (Bechie and Sibley, 1997). Accurately quantifying OM dynamics is a major challenge since the time scale varies for different processes (e.g. inputs versus breakdown) and the amount of OM in streams is affected by different processes (e.g. allochthonous input, primary production and retention) (Gregory et al., 1987; Webster and Meyer, 1997). Moreover, most of the organic matter transported in the water column occurs during major storm events, and it is logistically difficult to capture these large events (Wallace et al., 1995). Ecosystem modelling has been used to provide insights into factors most important for describing the dynamics of materials as they move through a system, identifying important uncertainties, and predict-

ing how future actions may alter these. Uncertainty in model outputs may be generated from incomplete data sources, variability in physical and biological process rates, and temporal and spatial variation (Webster and Meyer, 1997). Mathematical models of OM dynamics have been developed by measuring several components directly and using indirect measurements of others (Fisher and Likens, 1973; Boling et al., 1975; McIntire and Colby, 1978; Webster, 1983; Cazelles et al., 1991; Adams, 1998; Buzby and Perry, 2000; Pers, 2000; Welty et al., 2002). The objective for this research was to structure, calibrate and validate a mass-balance model for organic matter dynamics in headwater streams in southwestern British Columbia (Richardson, 1992; Kiffney et al., 2000). This is part of a larger study examining the effects of riparian zone management around small streams. Our focus was to identify major transport mechanisms of OM, as well as determine the magnitude and effects of uncertainties in model variables. A secondary objective was to simulate how varying inputs of organic matter and temperature could affect the stream detrital subsystem. The approach was a simple non-distributed mass-balance, ecosystem-scale model where the riparian–stream system was treated as one compartment to which different fluxes of OM were simulated. It has been shown (H˚akanson, 1999) that lake and coastal ecosystem models of this kind, with a few simple, accessible variables, can provide more accurate predictions than detailed physical models, which may be easy to build but hard to validate. It is therefore interesting to test this approach on a stream–riparian system in combination with an extensive empirical dataset and experimental knowledge of various processes.

2. Material and methods 2.1. Study site The University of British Columbia’s Malcolm Knapp Research Forest (MKRF) is located in the Coast Range Mountains approximately 60 km northeast of Vancouver (N49◦ 16 W122◦ 34 ). MKRF lies in the coastal western Hemlock biogeoclimatic zone. The predominant tree species are coniferous and include western hemlock (Tsuga heterophylla), western red cedar (Thuja plicata) and Douglas-fir (Pseudotsuga

O.M. Karlsson et al. / Ecological Modelling 183 (2005) 463–476

menziesii). Small amounts of red alder (Alnus rubra), vine maple (Acer circinatum) and salmonberry (Rubus spectabilis) occur in riparian areas. The climate is maritime, warm to temperate with an annual precipitation of 2200–2700 mm of which 70% falls between October and March (Feller and Kimmins, 1979). However, for the period during which data were collected for this study, the 1996–1997 hydrological year was the wettest on record at the research forest since 1945, whereas 1997–1998 was drier and warmer than average due to a strong El Ni˜no event (Kiffney et al., 2002). The following year 1998–1999, was a wet and cool year, and designated as a La Ni˜na event. Warm and dry El Ni˜no years, a climate phenomena of the Pacific region are often followed by cool and wet La Ni˜na events. Snow falls occasionally at this elevation (140–450 m). The average daily mean temperature is 17◦ for the warmest month and 0◦ for the coldest month (Feller and Kimmins, 1979). The soils are of glacial origin and are primarily shallow, coarsetextured humoferric podzols underlain by acid igneous bedrock (Feller, 1977). Due to the mild climate the soils are seldom frozen (Feller, 1977). The two streams used in this study were East Creek (EC) and South Creek (SCK), and both have a southerly aspect. The average gradients of the streams are 14% for EC and 26% for SCK (Kiffney et al., 2000). In 1931, a forest fire burned most of the area. Since then the watershed study area at EC has been left undisturbed while parts of the catchment area of SCK were logged in 1998–1999 leaving 30 m buffer strips along the stream channel. The two streams have confined channels, although some short reaches are widened into more of a floodplain. Channel wetted width gradually increases from a few centimetres at the stream head to about 2 m wide at the downstream measurement sites. Much of the two channels were characterised by step-pool morphology, with substrate dominated by gravel, cobbles and boulders. There was also a strong temporal variability in the actual stream area. During summer droughts the creeks can be reduced to small, isolated pools (Kiffney et al., 2000). East Creek had a few ephemeral tributaries, which contribute various amounts of water during the year. Much LWD (>10 cm diameter) is deposited in and over the stream channel. A survey of EC found an average LWD content of 60 kg/m2 (Young et al., 1999), which is very high compared to other studies in this region (Caza, 1993).


2.2. Data availability We used OM data collected from February 1997 to April 1999. The data from EC were used to calibrate the model and run different scenarios. The data from SCK were used as an independent dataset for validation of the model. Data on stream area calculations from Kiffney et al. (2000) were used, augmented with a brief field survey in March 2000. Stream lengths and catchment areas were measured using planimetry from a UBC Research Forest map (scale 1:5000). Both streams have V-notch weirs where gauging heights were recorded continuously. The gauging heights gave discharge through a rating curve. The discharge in the streams varied between 0 and 129 L/s. Average discharge during the sampling period was 33 L/s for EC and 13 L/s for SCK. Water temperature was recorded every 3 h at each reach using Onset® temperature loggers. Samples of OM concentrations, including dissolved OM (DOM), in stream water were available from Kiffney et al. (2000). Sampling was performed monthly May–September and twice a month October–April. Kiffney et al. (2000) also provided data on primary production in the stream using ceramic tiles that were placed in the stream and cleaned monthly. For details on sampling procedures and laboratory analysis of all OM, see Kiffney et al. (2000). Input rates of litter from Spring and Mayfly Creeks, two other streams at MKRF, were used (Richardson, 1992). Temperature-dependent decomposition rates of leaf litter from the research forest were also available (Richardson, 1992). There were no measurements available on input of blow-in. There were also no data on input, fragmentation and leaching rates of LWD (see below). To estimate initial amounts of LWD in streams, data from EC downstream of the actual study area were used (Young et al., 1999). Coefficients of variation (CV) were used to characterise uncertainty of variables used in the model. However, for most of the variables in the dataset it was not possible to calculate a CV. Instead, literature values or estimations were used (Table 1). 2.3. The model Our model treated the stream from the sampling site and everything upstream as one compartment. Imports


O.M. Karlsson et al. / Ecological Modelling 183 (2005) 463–476

Table 1 Estimated CV values for some variables in the stream ecosystem that were used to quantify total model uncertainty through Monte-Carlo simulation Variable Stream area

(m2 )




Estimation from brief field survey WMO (1980) Richardson (1992) Estimation from brief field survey Caza (1993)

Q (L/s) Litterfall (g/m2 /year) Import of DOM (g/week)

0.1 0.1 0.5

Fragmentation rate of LWD (1/year) Decomposition rate of litter (1/day) CPOM concentration (mg/L) FPOM concentration (mg/L) DOM concentration (mg/L)

0.5 0.2 0.5 0.4 0.2

Richardson (1992); Richardson et al. (2004) Kiffney et al. (2000) Kiffney et al. (2000) Kiffney et al. (2000)

Both empirical and literature values have been used.

and exports of OM, as gram ash-free dry mass (AFDM) per unit time were used in the model (Fig. 1). The imports (with variable codes in brackets) were: • litterfall (L) (g/week) from the canopy above the stream – empirically measured (Richardson, 1992); • blow-in (B) (g/week) wind or water transported litter from the riparian zone – estimated as a function of litterfall (Richardson, 1992); • DOM import (DOMimp ) (g/week) DOM entering the stream through subsurface flow – empirically measured but also a calibration “wheel”;

• primary production (PP) (g/week) the periphyton growth inside the stream channel – empirically measured (Kiffney et al., 2000) transformed into a regression as a function of stream flow; • POM or DOM derived from LWD (LWDd ) – literature values (Caza, 1993). The exports were: • Fluvial export of OM (OMexp ) (g/week) – OM exported with the flowing water – empirically measured (Kiffney et al., 2000); • Mineralisation of OM (MIN × Y) (g/week) – OM decomposed by organisms, loss of OM through respiration – empirically measured transformed into a temperature-dependent function (Richardson, 1992). The basic mass-balance equation for this model was: dY = L + B + DOMimp + PP + LWDd − OMexp dt − (MIN × Y ) (1) where Y is the amount of OM in the stream (g) and dY is the change in amount of OM (g) during the time interval dt (week). This is a simplified description of the system taking into account all relevant primary flows of OM into and from the defined area while internal processes have been omitted (e.g. gross sedimentation and resuspension of POM as well as physical phenomena as ve-

Fig. 1. The basic structure of the OM dynamics model. Arrows indicate direction of fluxes, boxes represent amounts, grey circles represent processes and transparent circles represent variables or parameters.

O.M. Karlsson et al. / Ecological Modelling 183 (2005) 463–476

locity and stream area variation effects occurring along the channel, generally dealt with by using traditional convection-diffusion partial differential equations (e.g. Dingman, 1994) that can be solved numerically e.g. by the Crank–Nicolson method). This is motivated mainly because of the fact that, since there were no empirical data available, either on spatial variability of primary flows of OM along the channel, or on internal processes in the stream. Introducing mathematical expressions of processes that cannot be verified would increase total model uncertainty but not necessarily the predictive power of the model. The focus of this paper was to rank primary flows of OM into and from the stream and not to describe what happened along the channel. To solve Eq. (1), a first order ordinary differential equation (ODE), the simulation computer program ithink® was used. We solved Eq. (1) numerically using the Euler method. The time interval was set to 0.01, which means that the computer divides each timestep (dt) into 100 units and for every unit approximates the derivatives with the difference through that time interval. The timestep for simulations was set to 1 week primarily because that was the best time interval to correlate OM export with Q. With the available data that meant that the model could be run for 116 weeks.


The default value of blow-in was set to 25% of litterfall (e.g. Richardson, 1992). The import of DOM was modelled as a concentration of DOM in the subsurface water entering the stream multiplied by Q. There was also a DOM reduction factor incorporated in the model, which allowed us to take into account if a part of the catchment area had been clear-cut. Meyer and Tate (1983) found that annual DOM export from a clear-cut watershed was reduced by 28% in a study performed in North Carolina, United States. That value was set as the default value for the model. Estimates of primary production, measured as ash-free dry mass on ceramic tiles, were regressed against Q (Table 2), which was necessary since there were not data available on PP for every week. The value from this regression was multiplied by the estimated stream area. The calculation of fragmented LWD import was based on two assumptions: (1) the input of LWD was 200 g/m2 /year (Caza, 1993) which was measured in a coastal forest of Washington, United States; (2) the system was in steady state in terms of LWD. The most common method to model decomposition in aquatic systems is the negative exponential model

Table 2 Equations and significance of statistical sub-models used in the dynamic model for describing different processes and transport mechanisms Regression equation


East Creek (EC) CPOM concentration (mg/L) = 0.00193(log10 (7-day Q)) (L/s) FPOM concentration (mg/L) = 0.286 + 0.315(log10 (7-day Q)) (L/s) Arcsin (DOM concentration/10) (mg/L) = 0.248 + 0.0557(T) (◦ C) Periphyton growth (␮g/m2 /day) = 1.70–0.825(log10 (7-day Q)) (L/s)

0.16 0.18 0.61 0.55

28 37 20 19

<0.05 <0.01 <0.0001 <0.001

South Creek (SCK) CPOM concentration (mg/L) = 0.00130 + 0.00277(log10 (7-day Q)) FPOM concentration (mg/L) = 0.374 + 0.0151(log10 (7-day Q)) (L/s) DOM concentration (mg/L) = 3.56 + 0.227(T) (◦ C) Periphyton growth (␮g/m2 /day) = 1.30–0.490(log10 (7-day Q)) (L/s)

0.43 0.27 0.22 0.68

25 32 18 14

<0.001 <0.01 <0.05 <0.001

Temperature correlation between EC and SCK T EC (◦ C) = 1.44 + 0.779(T SC) (◦ C)



<1 × 10−14

Decomposition rate of leaf litter Decomposition rate (1/day) = 0.0032e0.148T (◦ C)




FPOM concentration during rainstorm event FPOMemp (mg/L) = −0165 + 2.96(FPOMmod ) (mg/L)




Fluvial export correlated to modelled amount of POM Export (g/week) = 873 + 0.0392(amount of POMmod ) (g)



r2 : degree of explanation, n: number of empirical data, p: significance level.



<1 × 10−5


O.M. Karlsson et al. / Ecological Modelling 183 (2005) 463–476

(Webster and Benfield, 1986). That model can be written into the following form: A = A0 e−kt


where A is the amount of decomposable OM (g) at time t, A0 the initial amount of decomposable OM (g), k the rate coefficient (1/day) and t is the elapsed time (day); which is the solution of: ∂D = kD ∂t


which is stated in Eq. (1) (MIN × Y). Richardson (1992) found that k-values for red alder (A. rubra) leaf packs were related to water temperature in an exponential fashion (Table 2). The decomposition rates of conifer needles are ∼80% of what has been measured for alder leaves in the MKRF (Richardson et al., 2004). Sampling in EC and SCK during fall 1999 showed that 98% of litterfall was of coniferous origin. That value was used as a partition coefficient for detritus composition in the model. The export of CPOM (particle size > 1.0 mm), FPOM (particle size 1.0 ␮m–1.0 mm) and DOM (particle size < 1.0 ␮m) was based on their concentration multiplied by Q. Regressions were determined for OM concentrations and different variables (Kiffney et al., 2000). CPOM and FPOM were best related to Q, whereas DOM was best predicted by water temperature (T) (Table 2). The regressions were rather weak but the results are in agreement with the general view on sediment transport (e.g. Lid´en et al., 2001). Particulate matter especially coarse matter concentration is dependent on runoff, creating erosion, while the concentration of dissolved substances, which do not depend on erosion, are independent of Q. The reason for DOM being best related to T could be that biological activity transferring POM into DOM is dependent on T or that during periods of high T the streams are fed mainly with groundwater holding high concentrations of DOM. The rather weak regressions made it difficult to further investigate and explain the relation of different OM particle sizes with other variables. Preliminary results from a special storm sampling programme with high temporal resolution, not included in the basic dataset for this study, suggested that FPOM export on average was three times higher during storm events than during base flow. This behaviour was not handled by the original regressions between POM and

Q based on monthly sampling. Therefore, a routine was added to that model so that when gauging height increased more than 5 cm, which was considered a storm event, POM export was multiplied by three. This is also in agreement with general observations of sediment transport in running waters (e.g. Lid´en et al., 2001) and for detritus transport in headwaters in particular (Wallace et al., 1991; Adams, 1998). This simplification did not take into account the fact that the source pool of erodable OM is smaller shortly after a storm event and therefore the next storm will generate a smaller washout. Since sample frequency for OM and epilithic biomass was monthly or bimonthly, it was necessary to establish regressions with one or several dynamic parameters (i.e. Q or T) to run the model with an accurate timestep. The first step was to log10 transform Q and use arcsine transformation for DOM to satisfy assumptions of normality. The different transformations and variants where included in a stepwise multiple regression performed in Statistica® where the critical value of the F-distribution to enter was set to 4.0, which is approximately the same as not allowing a variable to enter unless p < 0.05 (H˚akanson and Peters, 1995). The model was calibrated under the assumption that the system was in steady state. That assumption is unlikely to be true for such a short time period over which the model was run (Webster and Meyer, 1997), but it was necessary to make the model work. The calibration “wheels” were DOM concentration in groundwater and the fragmentation rate of LWD, since no reliable data were available for those variables. With the DOM concentration in groundwater it was possible to adjust the input of DOM while fragmentation rate of LWD made it possible to adjust input of POM. Calibration of the model was performed as an iterative process where the total input and output of OM for the entire time period were balanced. Sensitivity analysis was performed by adjusting one variable or rate, while all other parameters were kept constant to see how that one parameter affected the model’s prediction of OM standing stock in the stream. Sensitivity analyses were conducted with two different methods: first, all model variables were altered by the same factor; second, the model variables were given a characteristic CV (Table 1) which was used to create a normal distribution of the variable. We ran the model

O.M. Karlsson et al. / Ecological Modelling 183 (2005) 463–476

300 times with the distributed values, which created a distribution in the prediction of the standing stock. Uncertainty analysis was done using Monte-Carlo simulation. Five hundred runs were created where all variables differed simultaneously according to their CVs so that their collective uncertainty could affect predictions of standing stock of OM. This was repeated when the model variables were omitted from the analysis one at the time.

3. Results 3.1. Building and calibrating the model Sensitivity analysis indicated that the prediction of OM standing stock was most sensitive to variation in DOM import and export. Stream area and the litterfall import also had a significant impact on the uncertainty in the prediction of OM standing stock. On the other hand, changes in CPOM and FPOM export had little effect on prediction of standing stock. The same results were obtained when the variables were given an identical CV. Uncertainty analysis showed the same pattern as the sensitivity analysis. Uncertainties in DOM were most important in explaining total model uncertainty. When uncertainty in DOM import was eliminated, the total model uncertainty decreased CV from 1.4 to 1.0.


Table 3 A ranking of the different inputs of OM to EC Source of OM

Percent of total input

DOM Litterfall Fragmentation Blow-in Primary production

67 20 8 5 <0.5

Variability in the transport of OM was high due to the variable discharge conditions, since fluvial export was concentration of OM × Q (Fig. 2). Dissolved organic matter from groundwater flow was the most important source of OM followed by litterfall, whereas primary production was very small (Table 3). Estimates for DOM and fragmentation of wood came from the calibration of the model for EC. The DOM concentration in groundwater was calibrated to a value of 4.1 mg/L in order to reach steady state. When three samples were taken from the stream head of three small seeps to EC in March 2000 average DOM concentration was 3.7 mg/L. Dissolved organic matter was also the predominant fraction of OM export, whereas CPOM was very small (Table 4). The overriding importance of DOM in determining OM budgets is why the model was most sensitive to changes in DOM import and export. Since DOM was the dominating fraction and overwhelmed most of the model, which made it difficult to simulate varying inputs of OM and varying tempera-

Fig. 2. Fluvial export of OM (g/week) in EC from February 1997 to April 1999. The export shows a high variability mainly due to the variable discharge conditions.


O.M. Karlsson et al. / Ecological Modelling 183 (2005) 463–476

Table 4 A ranking of the different outputs of OM from EC Export of OM

Percent of total export

DOM FPOM Mineralisation CPOM

83 13 4 <0.5

tures due to riparian zone management, the model was also run for POM, omitting the fluxes of DOM. The target variable for the model was then the POM standing stock, defined as the total amount of POM in the stream divided by the stream area (g/m2 ). The model was calibrated so that over time total imports and exports of POM would be equal, which led to a seasonally varying amount of POM in the stream (Fig. 3). The initial amount of POM standing stock at EC was set at 32 g/m2 , which have been empirically measured in Spring and Mayfly Creek, second-order streams of MKRF slightly larger than EC (Richardson, 1992). The model predicted an average standing stock of 117 g/m2 in EC but there was high variability in POM standing stock (95% CI = 0 and 235 g/m2 ). The predicted higher average POM standing stock in EC compared to the streams studied in Richardson (1992) likely reflects a higher retention capacity in the relatively smaller EC. This result may also be due to the discrepancy between the amount of POM in the stream with the flow of OM

through the system (Fig. 2). The average fluvial export of OM was 1 × 105 g/week while the amount of POM in the stream averaged 3 × 105 g, which means an average retention time of only 3 weeks and that over the year total OM export are 15–20 times larger than average standing stock. This means that standing stock of POM is small compared to what is flushed through the system and POM in the stream could easily be washed out of the system during a large storm and high stream flows. Peaks in POM standing stock occurred in early autumn (i.e. September and October) when inputs of litterfall were high and discharge generally low. 3.2. Validation It was not possible to do a formal validation of model performance since there was not any available empirical data on POM standing stock. However, running the model for SCK with model variables locked, showed same pattern in the seasonal dynamics of POM standing stock in SCK as in EC (Fig. 4). The average value on POM standing stock in SCK was 146 g/m2 (CI 0–292) compared to 117 g/m2 in EC. The results from the validation indicate that the basic structure of the model seems to be appropriate, although an iterative calibration using data from several streams could probably better tune the model.

Fig. 3. Prediction of the POM standing stock (g/m2 ) in EC from February 1997 to April 1999. The peaks occur in early autumn when litterfall input reaches maximum values.

O.M. Karlsson et al. / Ecological Modelling 183 (2005) 463–476


Fig. 4. Validation of the model using data from SCK to predict the POM standing stock (g/m2 ) from February 1997 to April 1999. The POM standing stock shows the same pattern as in EC (Fig. 3) with peaks in early autumn.

3.3. Scenarios We used regression analysis to make long-term predictions of OM dynamics. This was done by running the POM model for the available period of data and then relating the data on fluvial export with what the model predicted as POM standing stock for that period. Logging of riparian trees would potentially reduce input of OM to the stream (Fig. 5). The model shows that there would still be measurable amounts of POM in the stream even when litterfall was reduced to zero. This result is due to the large amount of wood in the stream that continues to fragment (e.g. Ward and Aumen, 1986). The average decrease in POM standing stock was 77% when litterfall was completely excluded (Table 5). Large woody debris deposited in the stream would continue to generate OM available for transport for a long time. If the input of wood ceased, there would Table 5 The decrease in POM standing stock (%) when litterfall was gradually simulated to be reduced from a normal input downwards to no input Reduction of litterfall (%)

Average POM standing stock in % of undisturbed value

0 25 50 75 100

100 81 62 42 23

still be enough material in the stream that would fragment for at least 50 years keeping the input of fragmented wood near the same level. Hence, LWD has the potential to partly compensate for the losses in OM input that would occur after harvesting. Another possible impact of timber harvest may be increased water temperatures due to reduced shading. This could stimulate decomposition and primary production. However, in the model increased T only affected the decomposition rate since primary production was so small. We increased average T by 1 ◦ C increments, up to a maximum of 3 ◦ C, while all other variables were kept at constant, to see how POM standing stock may be affected by increased T. The effect of increased T was relatively small (Table 6) compared to the effects of reduced litterfall (Table 5).

Table 6 The decrease in POM standing stock (%) due to a simulated increase in average water temperature (T) up to 3 ◦ C Increase in average T (◦ C)

Average POM standing stock in % of undisturbed value

1 2 3

92 85 77

The permanent average reduction in POM standing stock is due to an enhanced decomposition. The possible stimulation of periphyton growth and thereby increased POM standing stock was so small that it was not noticeable.


O.M. Karlsson et al. / Ecological Modelling 183 (2005) 463–476

Fig. 5. A simulation where litterfall has been successively reduced. The graph shows how this affects the POM standing stock (g/m2 ). The conditions are undisturbed system; litterfall reduction of 25%; litterfall reduction of 50%; litterfall reduction of 75%; litterfall reduction of 100%. The amount of POM standing stock left when litterfall is totally reduced is due to wood fragmentation.

Fig. 6. Prediction of POM standing stock (g/m2 ) in a 50-year perspective. Litterfall is simulated to cease after a logging operation and then recovers exponentially until canopy closure after 30 years. T is simulated to increase by 2 ◦ C directly after harvesting and then successively decrease due to shading as the canopy is established. The conditions are: undisturbed system; increased T; reduced litterfall and finally both reduced litterfall and increased T.

We also modelled successional changes in POM standing stock that may occur after logging (Fig. 6). Canopy closure generally takes place 30 years after a clear-cut and recovers exponentially (Cindy Prescott, UBC, personal communication). Increased T had little impact on the standing stock, whereas reduction in litterfall had a dramatic impact.

4. Discussion Our modelling efforts showed that the most crucial variable for total model uncertainty was DOM because it is the major component of OM in the stream system. Hence, a sampling programme investigating the spatial and temporal variability of DOM concentration

O.M. Karlsson et al. / Ecological Modelling 183 (2005) 463–476

in the groundwater and surface water would likely improve the predictive power of the model. Little is known about the function and importance of DOM in stream ecosystems (Webster and Meyer, 1997), but it may provide the energy to support a rich and productive food web as has been observed in ocean and lake ecosystems (Hessen and Tranvik, 1998; McArthur and Richardson, 2002). The calibrated value of DOM concentration in groundwater is within the range of what has been measured for most headwater stream ecosystems (Wallis et al., 1981; Webster and Meyer, 1997; except see Fiebig, 1995). Kiffney et al. (2000) showed that over 80% of the total export of OM from headwater streams was as DOM. The model also showed that wood was an important source of OM in these small, forested streams. In fact, when the amount of wood is compared with the amount of all other OM in the stream, wood is a factor 1000 higher than other OM. However, there were no data for the input, standing stock, fragmentation and mineralisation of wood, except literature values (Harmon et al., 1986; Caza, 1993; Young et al., 1999). The literature review on breakdown and input rates of wood by Harmon et al. (1986) shows a tremendous variation among different study sites. This indicates that any attempt to calculate budgets for OM dynamics in the stream will face the difficulty of quantifying the import of OM through fragmentation of wood. The breakdown of wood into finer particles occurs by erosion over time by the scouring of suspended load (Ward and Aumen, 1986), and by the activities of benthic invertebrates. It is known that many lotic invertebrates ingest wood (Anderson et al., 1978), likely the biofilm associated with the wood surface. Wallace et al. (1999) showed that the removal of wood from a small stream contributed to the reduction of benthic productivity, although it could not be shown whether this was through changes in physical habitat or its contribution to OM dynamics. The quantitative contribution of wood to the trophic basis of production in streams has not been evaluated, but it may be substantial given the magnitude of the biomass of wood in most streams. Regressions describing the relationship between FPOM and CPOM concentration and Q were weak (r2 = 0.26 on average). These statistical sub-models may be improved as the sampling programme continues. Wallace et al. (1991) observed a strong relationship between FPOM export and annual discharge (r2 = 0.91)


over a 5-year period in three headwater streams of the Appalachian Mountains. Wallace et al. (1995) also found that approximately 70% of CPOM export occurred during major rainstorm events. Thus, CPOM export in this study may be underestimated since CPOM was not specifically sampled during storm events. However, in the streams used in this study average discharge was 20–30 times higher and the leaves were smaller than in the Appalachian study. This indicates that CPOM export may occur even during baseflow conditions and that the retention time of leaf litter may be relatively short in the streams used in this study. Also fragmentation of wood may explain the flow of CPOM during baseflow. The model also showed how factors affecting OM operate at different temporal scales. The system’s response to a decrease in litterfall and blow-in was fast (weeks) while the effect of a lack of input of wood is hardly noticeable in a 50-year perspective. Naiman (1983) found that a large fraction of FPOM in small streams was apparently derived from wood fragments, consistent with our model results. This prediction was based on the assumption that wood would continue to fragment by the same rate as time goes by. Since the amount of LWD is high, the fragmentation remains almost constant over time. This indicates that it is difficult to compare results from different riparian management studies unless the role of wood has been investigated. Since the lifespan and distribution of trees are highly variable among sites, the response to different riparian treatments likely varies among streams depending on the initial amount of woody debris. Like all ecosystem models, the model built here has a certain range, i.e. a domain where the model is applicable. This range depends on how many streams the model has been validated against as well as the range in the data that formed the statistical sub-models. The stream-specific variables such as Q, T, stream area and the fluvial concentrations of OM all have to be recorded in any stream where the model is to be applied. To summarise, the model is applicable for small, heavily shaded forest streams (Qaverage < 33 L/s, wetted width < 2.4 m, watershed area < 44 ha), where the percentage of wetlands in the watershed is <5, the slope is moderate and the vegetation is dominated by conifers. However, the basic structure of the model could be used for any riparian–stream ecosystem but would require a new calibration and validation with independent data.


O.M. Karlsson et al. / Ecological Modelling 183 (2005) 463–476

There are differences in models of OM dynamics related to target variables, modelling scales (hourly to annual predictions), modelling structures (from empirical/regression models to partial differential equations) and driving variables (accessed from site-specific studies or from general physics theory). We have not found any model of OM dynamics originating from the same domain as ours. It is therefore not a simple matter to make meaningful model comparisons, and this was not the focus of this paper. The advantages of our empirical and rather simple modelling approach, compared to a more detailed physical and distributed model, depend on what purpose the model is used for. Introducing new processes or compartments may increase the predictive power to a degree, but the number of unknown parameters will certainly also increase, as well as the total model uncertainty (H˚akanson and Peters, 1995). It is partly a question of philosophy as to which level of complexity one chooses. Omitting or over-simplifying important processes necessarily creates a need for compensation somewhere else in the model. Doing so, the model may become more or less powerful as a predictive tool and always less useful in mechanistic and explanatory contexts. In our case, the availability and structure of empirical data from the site in combination with the well-known variability of ecosystem processes in headwaters and the time scale of interest (weeks–years) supported the box model approach in favour of a physical and distributed approach. Keeping the number of model variables low also simplified the procedure for calibration and validation and the interpretation of simulation results. This paper compiled information of OM dynamics of the studied system based on an extensive sampling programme and experiments complemented with necessary flows of OM to satisfy the principle of continuity by putting together processes that earlier have been studied separately. This gives new information on which flows of OM that actually determines the character of the system. It also gives the possibility to simulate and predict how future actions may affect the system. The scientific value of this study does not lie in the model as a tool for exact quantitative predictions of POM standing stock. The sensitivity analysis clearly showed that the uncertainties in driving variables generated wide confidence intervals around the

predictions of standing stock. The model should instead be used for identifying important flows and transport mechanisms in the extremely complex web of physical, chemical and biological interactions, which determine the character of the headwater stream ecosystem. There are two main results regarding the OM dynamics models for these streams. First, DOM is quantitatively the most important variable for driving the model and potentially the stream system in these small, heavily shaded streams. Second, wood is an important source of OM to the system due to its contribution of a large amount of the benthic biomass of OM in these streams. This raises questions for future research to investigate DOM and wood as food resources in headwater stream ecosystems. Another conclusion from the modelling of different scenarios is that logging and thereby removal of litterfall input will have a dramatic impact on POM standing stock. On the other hand, the effects of increased water temperature, which may be another effect of logging or from climate warming, have much smaller impact. In order to minimise the negative impacts of logging operations, the retention of buffer zones alongside stream channels has been encouraged, and clearly can have a large influence on the maintenance of OM budgets in these small streams.

Acknowledgements We thank Forest Renewal BC for research funding. Jennifer Bull and Natalie Lissimore are thanked for their assistance in collecting field data and Michael Feller for his willingness to share hydrological data.

References Adams, J.C., 1998. The role of leaf litter and small wood in the retention of fine particles during storms in an Appalachian headwater stream. M.Sc. Thesis. Virginia Polytechnic Institute and State University. Anderson, N.H., Sedell, J.R., Roberts, L.M., Triska, F.J., 1978. The role of aquatic invertebrates in processing of wood debris in coniferous forest streams. Am. Midland Nat. 100, 64–82. Bechie, T.J., Sibley, T.H., 1997. Relationships between channel characteristics, woody debris, and fish habitat in northwestern Washington streams. Trans. Am. Fish. Soc. 126, 217– 229.

O.M. Karlsson et al. / Ecological Modelling 183 (2005) 463–476 Bilby, R.E., Bisson, P.A., 1992. Allochthonous versus autochthonous organic matter contributions to the trophic support of fish populations in clear-cut and old-growth forested streams. Can. J. Fish. Aquat. Sci. 49, 540–551. Boling, R.H., Goodman, J.R., Van Sickle, E.D., Zimmer, J.A., Cummins, K.W., Petersen, R.C., Reice, S.R., 1975. Toward a model of detritus processing in a woodland stream. Ecology 56, 141–151. Buzby, K.M., Perry, S.A., 2000. Modeling the potential effects of climate change on leaf pack processing in central Appalachian streams. Can. J. Fish. Aquat. Sci. 57, 1773–1783. Caza, C.L., 1993. Woody Debris in the Forests of British Columbia. Land Management Report. Ministry of Forests, Victoria. Cazelles, B., Fontivielle, D., Chau, N.P., 1991. Self-purification in a lotic ecosystem: a model of dissolved organic carbon and benthic microorganisms dynamics. Ecol. Model. 58, 98–117. Chamberlain, T.W., Harr, R.D., Everest, F.H., 1991. Timber harvesting silviculture and watershed processes. In: Meehan, W.R. (Ed.), Influences of Forest and Rangeland Management on Salmonid Fishes and Their Habitat, vol. 19. American Fisheries Society Special Publication, Bethesda, Maryland, pp. 181–205. Dingman, S.L., 1994. Physical Hydrology. Prentice-Hall, New York, p. 575. Feller, M.C., 1977. Nutrient movement through western Hemlock–western Redcedar ecosystems in southwestern British Columbia. Ecology 58, 1269–1283. Feller, M.C., Kimmins, J.P., 1979. Chemical characteristics of small streams near Haney in southwestern British Columbia. Water Resour. Res. 15, 247–257. Fiebig, D.M., 1995. Groundwater discharge and its contribution of dissolved organic carbon to an upland stream. Archiv f¨ur Hydrobiologie 134, 129–155. Fisher, S.G., Likens, G.E., 1973. Energy flow in Bear Brook New Hampshire: an integrative approach to stream ecosystem metabolism. Ecol. Monogr. 43, 421–439. Gomi, T., Sidle, R.C., Richardson, J.S., 2002. Headwater and channel network understanding processes and downstream linkages of headwater systems. BioScience 52, 905–916. Gregory, S.V., Lamberti, G.A., Erman, D.C., Koski, K.V., Murphy, M.L., Sedell, J.R., 1987. Influence of forest practices on aquatic production. In: Salo, E.O., Cundy, T.W. (Eds.), Streamside Management: Forestry and Fishery Interactions, vol. 57. Univ. Washington, Inst. For. Resour. Contr., pp. 233–255. Gregory, S.V., Swanson, F.J., Mckee, W.A., Cummins, K.W., 1991. An ecosystem perspective of riparian zones. BioScience 41, 540–551. Harmon, M.E., Franklin, J.F., Swanson, F.J., Sollins, P., Gregory, S.V., Lattin, J.D., Anderson, N.H., Cline, S.P., Aumen, N.G., Sedell, J.R., Lienkamper, G.W., Cromack Jr., K., Cummins, K.W., 1986. Ecology of coarse woody debris in temperate ecosystems. Adv. Ecol. Res. 15, 133–302. Hessen, D.O., Tranvik, L.J., 1998. Aquatic Humic Substances Ecology and Biogeochemistry. Ecological studies, vol. 133. SpringerVerlag, Berlin Heidelberg, 346 pp. Hetrick, N.J., Brusven, M.A., Bjornn, T.C., Keith, R.M., Meehan, W.R., 1998. Effects of canopy removal on invertebrates and diet of juvenile coho salmon in a small stream in southeast Alaska. Trans. Am. Fish. Soc. 127, 876–888.


H˚akanson, L., 1999. Water Pollution – Methods and Criteria to Rank, Model and Remediate Chemical Threats to Aquatic Ecosystems. Backhuys Publishers, Amsterdam, 238 pp. H˚akanson, L., Peters, R.H., 1995. Predictive Limnology: Methods for Predictive Modelling. SPB Academic Publishing, Amsterdam, 464 pp. Kiffney, P.M., Richardson, J.S., Feller, M.C., 2000. Fluvial and epilithic organic matter dynamics in headwater streams of southwestern British Columbia. Can. Arch. Hydrobiol. 149, 109–129. Kiffney, P.M., Richardson, J.S., Bull, J.P., 2003. Responses of periphyton and insects to experimental manipulation of riparian buffer width along forest streams. J. Appl. Ecol. 40, 1060–1076. Kiffney, P.M., Bull, J.P., Feller, M.C., 2002. Climatic and hydrologic variability in coastal watershed of southwestern British Columbia. J. Am. Water Resour. Assoc. 38, 1432–1451. Lid´en, R., Harlin, J., Karlsson, M., Rahmberg, M., 2001. Hydrological modelling of fine sediments in the Odzi River, Zimbabwe. Water SA 27 (3), 303–314. McArthur, M.D., Richardson, J.S., 2002. Microbial utilization of dissolved organic carbon leached from riparian litterfall. Can. J. Fish. Aquat. Sci. 59, 1668–1676. McIntire, C.D., Colby, J.A., 1978. A hierarchical model of lotic ecosystems. Ecol. Monogr. 48, 167–190. Meyer, J.M., Tate, C.M., 1983. The effects of watershed disturbance on dissolved organic carbon dynamics of a stream. Ecology 64, 33–44. Minshall, G.W., Cummins, K.W., Petersen, R.C., Cushing, C.E., Burns, D.A., Sedell, J.R., Vannote, R.L., 1985. Developments in stream ecosystem theory. Can. J. Fish. Aquat. Sci. 42, 1045–1055. Murphy, M.L., Heifetz, J., Johnson, S.W., Koski, K.V., Thedinga, J.F., 1986. Effects of clear-cut logging with and without buffer strips on juvenile salmonids in Alaskan Streams. Can. J. Fish. Aquat. Sci. 43, 1521–1533. Naiman, R.J., 1983. The influence of stream size on the food quality of seston. Can. J. Zool. 61, 1995–2010. Naiman, R.J., Bilby, R.E., Bisson, P.A., 2000. Riparian ecology and management in the Pacific coastal rain forest. BioScience 50, 996–1011. Pers, B.C., 2000. Modelling organic matter dynamics in aquatic systems. Ph.D. Thesis. Department of Water and Environmental studies, Link¨oping University. ISBN 91-7219-825-7, ISSN 0282-9800. Richardson, J.S., 1991. Seasonal food limitation of detritivores in a montane stream: an experimental test. Ecology 72, 873–887. Richardson, J.S., 1992. Coarse particulate detritus dynamics in small, montane streams of southwestern British Columbia. Can. J. Fish. Aquat. Sci. 49, 337–346. Richardson, J.S., Shaughnessy, C.R., Harrison, P.G., 2004. Litter breakdown and invertebrate association with three types of leaves in a temperate rainforest stream. Archiv f¨ur Hydrobiologie 159, 309–325. Vannote, R.L., Minshall, G.W., Cummins, K.W., Sedell, J.R., Cushing, C.E., 1980. The river continuum concept. Can. J. Fish. Aquat. Sci. 37, 130–137. Wallace, J.B., Cuffney, T.F., Webster, J.R., Lugthart, G.J., Chung, K., Goldowitz, B.S., 1991. Export of fine organic particles from


O.M. Karlsson et al. / Ecological Modelling 183 (2005) 463–476

headwater streams: effects of season, extreme discharge and invertebrate manipulation. Limnol. Oceanogr. 36, 670–682. Wallace, J.B., Whiles, M.R., Eggert, S., Cuffney, T.F., Lugthart, G.J., Chung, K., 1995. Long-term dynamics of coarse particulate organic matter in three Appalachian Mountain streams. J. N. Am. Benthol. Soc. 14, 217–232. Wallace, J.B., Eggert, S.L., Meyer, J.L., Webster, J.R., 1999. Effects of resource limitation on a detrital-based ecosystem. Ecol. Monogr. 69, 409–442. Wallis, P.M., Hynes, H.B.N., Telang, S.A., 1981. The importance of ground water in the transportation of allochthonous dissolved organic matter to the streams draining a small mountain basin. Hydrobiologia 79, 77–90. Ward, G.M., Aumen, N.G., 1986. Woody debris as a source of fine particulate organic matter in coniferous forest stream ecosystems. Can. J. Fish. Aquat. Sci. 43, 1635–1642. Ward, J.V., 1989. The four-dimensional nature of lotic ecosystems. J. N. Am. Benthol. Soc. 8, 2–8. Webster, J.R., 1983. The role of benthic macroinvertebrates in detritus dynamics of streams: a computer simulation. Ecol. Monogr. 53, 383–404.

Webster, J.R., Meyer, J.L., 1997. Stream organic matter budgets. J. N. Am. Benthol. Soc. 16, 3–161. Webster, J.R., Golladay, S.W., Benfield, E.F., D’Angelo, D.J., Peters, G.T., 1990. Effects of forest disturbance on particulate organic matter budgets of small streams. J. N. Am. Benthol. Soc. 9, 120–140. Webster, J.R., Benfield, E.F., 1986. Vascular plant breakdown in freshwater ecosystems. Annu. Rev. Ecol. Syst. 17, 567–594. Welty, J.J., Beechie, T., Sullivan, K., Hyink, D.M., Bilby, R.E., Andrus, C., Pess, G., 2002. Riparian aquatic interaction simulator (RAIS): a model of riparian forest dynamics for the generation of large woody debris and shade. Forest Ecol. Manag. 162, 299–318. Wipfli, M.S., Gregovich, D.P., 2002. Invertebrates and detritus export from fishless headwater streams in southeast Alaska: implications for downstream salmonid production. Freshwater Biol. 47, 957–970. WMO, 1980. Manual on Stream Gauging. Computation of Discharge, vol. 2. World Meteorological Organisation, 258 pp. Young, K.A., Hinch, S.G., Northcote, T.G., 1999. Status of resident coastal cutthroat trout and their habitat twenty-five years after riparian logging. N. Am. J. Fish. Manag. 19, 901–911.