Grantee Research Project Results
Final Report: Regional Analysis of Variation in Adirondack Lake Ecosystems: Landscape Scale Determinations of Dissolved Organic Carbon
EPA Grant Number: R826762Title: Regional Analysis of Variation in Adirondack Lake Ecosystems: Landscape Scale Determinations of Dissolved Organic Carbon
Investigators: Pace, Michael L. , Canham, Charles D.
Institution: Cary Institute of Ecosystem Studies
EPA Project Officer: Packard, Benjamin H
Project Period: January 1, 1999 through December 31, 2001 (Extended to December 31, 2002)
Project Amount: $453,775
RFA: Regional Scale Analysis and Assessment (1998) RFA Text | Recipients Lists
Research Category: Aquatic Ecosystems , Ecological Indicators/Assessment/Restoration
Objective:
The objectives of the research project were to: (1) develop a regional scale modeling approach to predict variations in lake ecosystem properties as a function of landscape characteristics; (2) analyze landscape and within-lake processes that determine variations in lake dissolved organic carbon (DOC); and (3) analyze patterns in lake DOC related to landscape and lake processes. In this report, we briefly summarize the sequence of activities during the project and primary results.
Summary/Accomplishments (Outputs/Outcomes):
The research project was conducted in three phases. During the initial phase, the effort focused on obtaining and utilizing geographic information system (GIS) databases to provide the land coverage data needed for model analyses. The major challenge was delineating watershed boundaries for a large number of watersheds using objective, repeatable criteria. In the middle phase of the work, we developed the mass balance models and programs for likelihood analyses. We also conducted a planned field survey of 100 lakes to reassess chemical conditions in a subset of the lakes used for this study. In the final phase of the work, we conducted model analyses and various hypotheses tests to develop the final results summarized in this report. These results will be published in Ecological Applications (Canham, et al., in press).
Terrestrial ecosystems contribute significant amounts of DOC to aquatic ecosystems. Much of the DOC in lakes derives from terrestrial sources, and this material has a significant effect on light penetration, particularly ultraviolet light, with implications for photosynthesis, mixing rates, and oxygen concentrations. DOC also influences the bioavailability of phosphorus and metals, and is implicated in the transformation of pollutants, such as mercury in lakes. Temperate lakes can vary dramatically in the concentration of DOC, as a result of variation in the spatial configuration and composition of vegetation within the watershed, hydrology, and within-lake processes. We developed and parameterized a mass balance model of lake DOC concentrations using data from lakes and associated watersheds in the Adirondack Park, NY. The watersheds encompassed a wide range of vegetation, soils, hydrology, and lake volume. Our analyses made use of 10-m resolution spatially distributed data sets on the 12 major upland and wetland cover types within each of the watersheds and data on lake DOC concentrations from a synoptic survey of Adirondack lake chemistry conducted from 1984-1987. The analyses estimated watershed loading to each lake as a function of the cover type of each 10 x 10 m grid cell and the flow path distance from individual cells to the lake. The estimated export rates for the three main forest cover types were 38-47 kg/ha/year and the estimated export from the four main wetland cover types was much higher (188-227 kg/ha/year). Loading from both the main wetland and upland cover types showed very little decline with distance from the lakeshore. Although wetlands had much higher rates of DOC export per unit area, they represented, on average, only 12 percent of watershed area. As a result, upland forests were, on average, the source of approximately 70 percent of DOC loading. There was evidence of significant interannual variation in DOC loading, correlated with interannual variation in precipitation. Estimated net in situ DOC production within the lakes was low (< 1 kg/ha/year) relative to loading. The estimated average in-lake decay coefficient (k) was 0.92 year-1. This parameter indicates that a mass roughly equivalent to the entire pool of DOC in a lake turns over via various chemical and biological mechanisms in 1 year. Examination of alternate models revealed a significant decline in k with lake depth, meaning that in larger, deeper lakes, DOC decays more slowly.
Many of the lakes have large watersheds relative to lake volume and have correspondingly high flushing rates (median = 4.34 year-1 for 355 headwater lakes). As a result, losses resulting from lake discharge generally had a larger effect on lake DOC concentrations than the in-lake decay coefficient. One of the strengths of this approach is that the method can be applied to large numbers of watersheds at a regional scale. This allows estimation of the cumulative impacts of alteration in the spatial distribution and types of land cover within a watershed. Availability of spatially distributed data sets of the type required for this approach is increasing rapidly, and resource agencies rely on GIS to manage those data sets. Our approach can be readily incorporated within a GIS framework and allows examination of different scenarios, such as how losses of wetlands, alterations in forest management, or increases in conserved areas might affect aspects of water quality.
One way to summarize the results of the modeling study is to consider the example of two specific lakes. Clear Lake has a watershed dominated by forest with an estimated annual export of 6,100 kg C per year and a small additional contribution from wetlands. These inputs result in an area l loading rate (per unit of lake surface area) of 15 g C m-2 year-1. Hence, Clear Lake lives up to its name by having a very low DOC concentration with clear water. The carbon input to Clear Lake is still significant but probably is a less important feature of the overall carbon budget than in many other lakes. Muir Lake, in contrast, has a high average area l loading rate (100 g C m-2 year-1) because of extensive wetland areas within the watershed. Wetlands dominate inputs to Muir Lake, but a third of the annual load still comes from uplands. Therefore, relative wetland coverage among watersheds is important in explaining differences among lakes in DOC, but forests are most often the dominant source of loadings because this cover type dominates the landscape.
Technical Aspects of the Project
The main technical aspects of the project involved used GIS databases and lake chemistry data measured by others. These data sets are described in detail below under Computer Modeling Methods. Quality issues related to the databases we used are carefully documented in the sited publications that present the original methods.
We also conducted a lake survey primarily to revisit a subsample of the hundreds of lakes sampled by the Adirondack Lake Survey Corporation (ALSC) in the 1980s (Kretser, et al., 1989). We measured DOC and watercolor using well-documented methods (Pace and Cole, 2002). The color measurement was based on measuring light absorbance at a specific wavelength (440 nm) and was not directly comparable to qualitative methods used during the ALSC survey. DOC values from the two surveys were comparable. Direct comparison of the two data sets revealed similar means and standard deviations (Institute of Ecosystem Studies [IES] data mean ± 1 SD: 5.78 ± 2.56, ALSC data: 5.84 ± 3.55, n = 98). It is not possible to draw strong conclusions about significant changes from these data because of the limited number of temporal samples for each lake. This is best done from temporal monitoring as has been carried out for selected lakes in the Adirondacks (Driscoll, et al., 2003). The comparison of the ALSC and IES data, however, indicates that lakes represent a similar population in terms of the mean and variance observed in DOC, and so we conclude that the model is generally applicable over the broad time period. The comparison also reveals, however, that there is variability within lakes in DOC over time that affects our attempts to model variation in DOC among lakes. Thus, the model explains about one-half of the overall variation among lakes and some of the residual, unexplained variation is related to using single sample values from the ALSC survey. A better estimate of the long-term mean value of DOC and other water quality variables from surveys would likely improve model results.
Computer Modeling Methods
Model Description. Our analyses used principles of mass-balance, in which variation in DOC concentration can be understood as a balance between total inputs to the lake, primarily from the surrounding watershed, and net losses, as a result of in-lake processes and output in lake discharge. In the formal terms of a difference equation:
DOCt+1 = DOC2 + Inputst→t+1 - Degradationt→t+1 - Discharget→t+1
where DOC is measured as a concentration (g C/m3), and inputs and losses are scaled to a predefined time interval (e.g., 1 year). Inputs to the lake are assumed to be independent of in-lake DOC concentration, whereas losses are assumed proportional to in-lake DOC concentration. This results in a predicted steady-state when DOC concentration reaches a level where losses balance inputs. The analyses predict average, midsummer concentrations within individual lakes based on steady state assumptions. Studies have shown significant year-to-year variability in midsummer averages, often with significant regional synchrony (e.g., Pace and Cole, 2002). This potential variability is incorporated in the model through additional terms that account for the effects of interannual differences in climate and hydrology on nutrient loading and lake discharge (see Model Formulation, Input Data Sets, Data Requirements).
Performance Criteria. We evaluated the model by assessing bias, goodness of fit, and root mean square error of prediction. For the analyses, the data were divided into headwater lakes that had no lake above them higher in the watershed, and a total lake data set that included both headwater lakes and those lakes that were higher in the watershed (see discussion of lake types below). The analyses produced unbiased fits to the data (i.e., slope of regression of observed versus predicted ~ 1.0), explaining 55 percent of the variation in DOC for the 355 headwater lakes and 48 percent of the variation in the larger sample of 428 lakes. Root mean squared error was 2.49 for the headwater lakes model and 2.75 for the 428 lakes.
Theory Behind the Model. Our analyses required a large sample of extensive and detailed watershed spatial data. Where such data are available, our spatially explicit inverse modeling approach allows estimation of the key terms that govern regional-scale variation in lake DOC concentrations. The method has a number of advantages in comparison with multivariate analyses that are not spatially explicit and not based on mass-balance principles. For example, our approach partitions loading among specific source areas within the watershed as a function of cover type and distance to the lake. Because the model is based on mass-balance principles, the estimates of loading are in units (i.e., kg C ha-1 year-1) that can be directly related to carbon fluxes within both the watershed and the lake. The method also should be generally applicable to analyses of watershed loading and in-lake processing of other nutrients and elements such as P, N, and S, which are parameters of concern to both lake eutrophication and acidification in the Adirondack Park and elsewhere. The approach also offers an advantage over simulation modeling methods in that the parameters are estimated from the data and not based on averages from the literature, guesses, or direct measurement (often impractical for many parameters). The disadvantage of the approach is that it is less mechanistic then many detailed, hydrologic watershed models. These types of models (hydrologically detailed) offer a better representation of key processes, but it is often difficult to parameterize such models for a large number of watersheds.
Mathematics, Formulas, and Calculation Methods
Watershed Delineation. We delineated the watershed for each lake using GIS software (ArcView 3.1) combined with our own scripts. Ten-meter resolution digital elevation model (DEM) data were downloaded from the Cornell University Geospatial Data Information Repository. These data were imported into ArcView and merged into one grid data layer. An ArcView script (Spatial.DEMFill) was used to remove sinks from the grid layer. The ALSC field manual was used to identify lakes for which DOC was measured. These were extracted from the photointerpreted GIS wetlands data layer and converted to grid format. The contributing area above each lake was calculated using the ArcView command "Watershed" on the sink free DEM data. The resulting watersheds were verified using the Adirondack Park Agency (APA) delineation from U.S. Geological Survey (USGS) topographic maps.
Stream Networks. Part of the watershed delineation procedure requires the calculation of a flow direction map. These data were used to calculate a flow accumulation map. This was, in turn, used to create the stream network by applying a threshold to identify cells with high accumulated flow. Results were compared to USGS topographic maps to give a reasonable approximation to the mapping of perennial streams. This method alone did not generate stream networks that corresponded to USGS maps in both steep and flat areas. Different thresholds could be selected that optimized for one at the expense of the other, but not both. We developed another procedure that weighted upstream cells according to the landscape type. Unsaturated areas were given a weight of 1, whereas saturated areas were given a weight of 50. This, combined with a stream threshold of 5,500, resulted in an acceptable approximation of the USGS mapped streams. The stream vector coverage was converted to a grid layer with a width of 10 m (the minimum resolution of our grid data layers).
Flow Path Distances. Flow path lengths were calculated from each point (i.e., 0.01 ha grid cell) in each watershed to the drainage lake using ArcView's "FlowLength" command. "FlowLength" calculates the flow path length using the flow direction map from each point to the outlet at the lake edge.
Compiled Watershed Datasets. For the 428 watersheds in our final data set, we classified each 10 x 10 m grid cell into either a nonsource area (lakes, streams, and roads, see data sets below) or 1 of the 12 wetland or upland cover types, based on the GIS data layers. For each cell, we used the 10-m resolution DEM to calculate flow path distance (as above) to the lakeshore. Data from the ALSC surveys provided the midsummer lake DOC concentrations, lake volume, and lake flushing rate (based on watershed runoff calculations) (Kretser, et al., 1989). To increase the speed of the iterative process used to estimate model parameters (see below) for each cover type in each watershed, we calculated the average flow path distance to the lake for all cells of that cover type in each of 20 distance classes for the headwater watersheds or 26 distance classes for the analyses of all 428 lakes (which included larger watersheds). The sizes of the distance classes were chosen to provide more precise discrimination of flow path distances near the lake (starting at 10-m intervals) and increased in size with greater distance from the lake. Thus, rather than integrate across all grid cells in each watershed, we summed across the 20 or 26 distance classes, using the mean flow path distance for grid cells in that class.
Parameter Estimation Through Inverse Modeling and Maximum Likelihood Methods. Our analyses are a form of inverse modeling conceptually similar to a spatial regression in which lake DOC concentration is the dependent variable and the independent parameters are: (1) lake volume and surface area; (2) lake flushing rate; (3) the cover type and distance from lake for each of the grid cells in the immediate watershed; and (4) the year in which the lake was sampled (as a categorical variable). The basic model requires 3*n + 5 parameters, where n is the number of cover types, for a total of 41 parameters given 12 cover types. The parameters are analogous to regression coefficients. We solve for the parameter estimates that maximize the likelihood of the observed lake DOC concentrations, using simulated annealing (Goffe, et al., 1994), an iterative, global optimization procedure. Residuals were assumed to be normally distributed. Although a large number of parameters are estimated in this procedure, it is important to note that the number of watersheds and individual grid cells on which the estimation is based also is very large.
Statistical Analyses. We compared alternate models with different numbers of parameters using likelihood ratio tests (Hilborn and Mangel, 1997). This tested the significance in improvement (if any) in likelihood of a model because of the incorporation of additional parameters. Under principles of parsimony, we accepted a simpler model (i.e., with fewer parameters) if it did not have a significantly lower likelihood. For alternate models with the same number of parameters, no significance tests were necessary; parsimony dictated choosing the model with the highest likelihood. We calculated asymptotic 95 percent support limits (analogous to traditional confidence intervals) for each of the parameters by holding all other parameters at their maximum likelihood value, and then systematically increasing or decreasing the parameter of interest until the likelihood of the resulting model was significantly worse (at a 5 percent alpha level) than the maximum likelihood model.
Review of Theory and Mathematical Algorithms
The equations used in this study were peer reviewed for publication (Canham, et al., in press). In addition, the basic outline of the model approach was given in the original proposal that also was peer reviewed. Strengths and weaknesses of the approach are summarized above (under Theory Behind the Model).
Parameter Uncertainty
Because the parameters are estimated from the data with the approach used, parameter uncertainty is explicit. Table 1 presents each parameter and the estimated uncertainty. For the major land cover areas (forests, major wetland types), 95 percent uncertainty of export coefficients was approximately +/- 20 percent (see Table 1). For minor cover types, uncertainty was greater, but these areas were much less important as overall contributors to export. Estimates of distance decay factors had large uncertainty, but most cover types also had little distance decay (meaning the contribution of DOC to lakes from near and far areas was similar). Hence, in practice, the cover types for which distance decay was important were those with coefficients greater than 100. Only a few cover types were in this category, and they represent relatively small portions of the overall landscape.
Model Formulation, Input Data Sets, and Data Requirements
The approach depends heavily on spatial data and other data inputs. This section describes the model formulation in the context of the data sets used and the implementation of this data in the model.
Inputs. There are three major allochthonous inputs of DOC to lakes: (1) atmospheric deposition; (2) streams that carry DOC exported from upstream lakes and their associated watersheds; and (3) inflowing stream water and groundwater from wetlands and upland areas within the immediate watershed. In addition, there is in situ production of DOC within lakes. For the purposes of our model, we assumed that both in situ DOC production and atmospheric deposition of DOC directly to the lake are linearly proportional to lake surface area (SA, in m2), so we combined these two sources into a single, net lake surface area input (SAI, in g C/m2).
We considered the watershed of a given lake as a grid of source areas of fixed size (10 x 10 m), in which each source area is classified as a discrete cover type based on vegetation, drainage, and land use. Inputs arise from grid cells and move along flow paths that conceptually include both overland and groundwater flow, until they reach surface water (either the lake shore or streams feeding into the lakes). The model does not discriminate between overland versus groundwater flow, but instead lumps them as "ground" flow, as distinct from "stream" flow inputs to the lake. In the simplest model, total annual input (grams) of DOC to the lake is specified by:
(2)
ULE is the export (in grams) from j = 1.M upstream lakes, and is the average proportion of upstream lake export that is not lost through processing within a stream before it reaches the downstream lake. For the sake of simplicity, is assumed to be independent of stream length. Exportc is the export (in grams) of the i'th grid cell (0.01 ha) of type c within the immediate watershed. The fraction of the export that reaches the lake (i.e., loading) is specified by an exponential loss as a function of the flow path distance (Di) from the grid cell to the lake. The loss function is flexible enough to accommodate a wide range of shapes, according to the estimated parameters and . Loss of DOC along a flow path is assumed to occur because of several processes, including: (1) decomposition; (2) sedimentation and mineral complexing in soils and sediments along the flow path; and (3) loss to deep groundwater.
Equation 2 is a simple additive model of nonpoint inputs, in which each unit area of the watershed is a potential source, and the amount of DOC from each source area that reaches the lake is a declining function of the distance of the source area from the lake. In this simple model, loss along a flow path that originated from an upslope source area does not depend on the nature of the cover type through which DOC moves.
Boyer, et al. (1996) have shown that overland flow from areas with saturated soils is a proximate source of significant DOC loading from uplands. A "topographic index" (Beven and Wood, 1983), based on the slope of a grid cell and the upslope contributing area, is frequently used to identify areas prone to saturated soil conditions. We calculated the topographic index (TI) for each grid cell in the study region to explore whether the index would improve our predictions of watershed-scale inputs. We tried several model variants incorporating TI in our mass-balance model, but none of the variants improved the fit between data and model. Some of the information contained in TI is already incorporated in our model in a different form, because areas with high TI values are generally occupied by wetlands, and our wetland data layers (described below) allow us to take that into account.
Interannual Variability in DOC Loading. The lakes in our data set (described below) were sampled in midsummer in 1 of 4 years (1984-1987). In each year, the sampled lakes were widely distributed across the Adirondack region studied and were well stratified across watershed characteristics, such as lake size and flushing rate. Nonetheless, lakes sampled in 1986 had a significantly higher DOC concentration than lakes sampled in the other 3 years. In a separate study, Pace and Cole (2002) examined temporal variation of DOC in a set of Michigan lakes and found a high degree of synchrony. Years with high midsummer DOC concentrations were associated with higher-than-normal runoff in spring and early summer. On this basis, we incorporated a term in our model to allow for interannual variation in total DOC loading from within the watershed. The year 1984 was set as a benchmark, and the analyses estimated the variation in total within-watershed loading for the 3 other years (1985-1987) needed to account for the observed interannual variation in lake DOC concentration.
Losses. Losses of DOC from a lake are conceptually separated into: (1) lake discharge, and (2) within-lake losses. Loss via lake discharge is estimated from flushing rates based on data on runoff from within the immediate watershed, lake morphometry, and discharge from upstream lakes. Degradation of DOC in aquatic systems is actually an amalgamation of processes that include direct photodecay, microbial degradation, and flocculation/sedimentation (Wetzel, 2001; Molot and Dillon, 1997). Following previous studies (Engstrom, 1987; Dillon and Molot, 1997), we combined these processes into a single decay constant:
Degradation = K*volume* DOC
Thus, the overall model at steady state has the following formulation:
(4)
We also considered alternative formulations of within-lake losses that were related to three factors: (1) lake depth (Rasmussen, et al., 1989; Dillon and Molot, 1997); (2) proportion of watershed DOC loading from wetlands (Engstrom, 1987); and (3) lake acid neutralizing capacity (Reche, et al., 1999). All three factors have been shown to influence rates of degradation of DOC in lakes as a result of very different mechanisms.
Lake and Watershed Data. Data for this study came from several sources. Between 1984 and 1987, the ALSC sampled 1,469 lakes within the boundaries of the Adirondack Park (Kretser, et al., 1989). The wetlands and forests for the major river drainage systems in the Park are being mapped and classified by the APA (Roy, et al., 1997; Primack, et al., 2000). As a companion to the wetlands mapping program, APA also has assembled an extensive set of GIS-referenced data layers on the physical and biological characteristics of the watersheds in those drainages (Roy, et al., 1997; Primack, et al., 2000).
During this research project, watershed data were available for four major river drainages in the Park: the Oswegatchie River, the Black River, the Sacandaga River, and the Upper Hudson River. Within these drainages, 610 lakes were sampled for DOC by the ALSC. Each lake was sampled twice (spring and summer or summer and fall) for a spectrum of physical, chemical, and biological variables, including DOC, with roughly equal numbers of lakes sampled in each year. We analyzed two categories of watersheds. First, we considered only headwater lakes; those that had no upstream ponded waters more than 1 ha in size. This allowed for an initial examination of model results without the complication of inputs from upstream lakes. A second analysis included all lakes (n = 610). A total of 182 of the 610 lakes could not be used in our analyses for a variety of reasons. Twenty-two lakes were dropped because they were downstream from very large reservoirs that would constitute a large, unmeasured input of DOC. Twenty-five "lakes" were dropped because they were actually emergent marshes rather than open water, or had a mean depth less than 1 m. One hundred and twenty-five lakes had ponds greater than 1 ha in size upstream, for which there were no ALSC DOC data available to estimate downstream exports. We were unable to produce acceptable watershed delineations for 10 of the lakes that were in areas of very low relief. In most of these cases, the 10-m resolution DEM produced watershed boundaries that split apart wetlands bordering the lake into adjacent watersheds. Of the 428 remaining lakes, 355 were headwater lakes.
Wetland Data. The APA identified and mapped all wetlands within the Oswagatchie, Black, Sacandaga, and Upper Hudson drainages (Roy, et al., 1997; Primack, et al., 2000). Wetlands were delineated from the 1:40,000 scale USGS National Aerial Photography Program that used color infrared imagery taken in the mid-1990s, and the 1:58,000 scale USGS National High Altitude Photography Program that used color infrared imagery taken in the mid 1980s, as described in Roy, et al., 1997; Primack, et al., 2000. The classification was based on National Wetlands Inventory techniques (Cowardin and Golet, 1995) and identified the dominant and subordinate strata in each wetland, as well as modifiers for hydrology and disturbance (by beavers, etc.). We lumped the wetlands into seven major groups: (1) emergent marshes, typically dominated by cattails and sedges; (2) deciduous shrub swamps, dominated by speckled alder (Alnus incana ssp. rugosa) and willows (Salix spp.); (3) broadleaved evergreen shrub swamps, primarily bogs dominated by a variety of ericaceous shrubs; (4) needle-leaved evergreen shrub swamps, typically bogs dominated by stunted black spruce (Picea mariana); (5) deciduous forest swamps, typically dominated by red maple (Acer rubrum); (6) conifer forest swamps, dominated by red spruce (Picea rubens), black spruce, or balsam fir (Abies balsamea); and (7) “dead tree” swamps, in which most of the canopy trees were dead, usually as a result of beaver activity (Roy, et al., 1996; Primack, et al., 2000). To keep parameters in the model to a manageable number, we did not further divide these groups based on the estimated frequency and duration of flooding.
Forest Data. The APA also mapped and classified upland forests using LANDSAT 5 Thematic Mapper imagery (Roy, et al., 1997; Primack, et al., 2000). The classification delineated forests into four major cover types: deciduous forests, coniferous forests, mixed deciduous/coniferous forests, and mixed deciduous/open forests, and two nonforest cover types: "deciduous/open" vegetation with a mix of herbaceous and young woody vegetation, and "open vegetation" for areas dominated by nonwoody vegetation. The much coarser resolution of forest cover types was, in part, dictated by the nature of the remote sensing analyses. Previous studies, however, suggest that this is an appropriate level of resolution for characterizing the effects of variation in forest composition on inputs of DOC to lakes (e.g., D'Arcy and Carignan, 1997). We combined the two “mixed” forest types into a single type, giving five upland vegetation types: deciduous forest, mixed forest, conifer forest, deciduous/open vegetation, and open vegetation that included most residential and developed areas.
Roads. Many of the watersheds are in road-less wilderness areas. For watersheds that contained roads, we used a road data layer compiled by the APA and assigned a width to each road category: 10 m for local and town roads, 20 m for secondary state highways, and 30 m for primary state highways.
Hardware and Software Requirements
The analyses were done with software written using Delphi (Borland International) for a personal computer (PC) running Windows 95 or higher (Microsoft Corp.). At least a 200 mHz central processing unit (CPU) is required. A fast computer (1 GHz CPU) is very helpful, as analyses may take 24 hours or more to run on a PC.
Documentation
The model is summarized in a forthcoming peer-reviewed journal publication. A reprint of the paper will be available from the principal investigators (PIs) as soon as the paper is published. Model code is available from the PIs upon request.
One of the strong impacts of terrestrial environments on aquatic ecosystems is the transfer of dissolved organic matter from land to water. This material arises from the degradation of plants in the soil and litter. Some of the material eventually dissolves in solution and flows to streams and lakes. Water rich in terrestrially derived DOC has a brown color and is readily visible in "stained" lakes. DOC has significant effects on light penetration in lakes, especially on harmful ultraviolet light. DOC also is implicated in the transformation of mercury into a toxic form that concentrates as it moves through lake food chains. Lakes can vary dramatically in the concentration of DOC from clear water low in DOC to tea-colored and even black water, high in DOC. This variation is the result of vegetation cover within the watershed, hydrology, the presence or absence of wetlands, and within-lake processes. We developed and parameterized a model to predict the variation in lake DOC concentrations using data from 428 lakes and associated watersheds in the Adirondack Park, NY. The watersheds encompassed a wide range of vegetation, soils, hydrology, and lake types. Our analyses made use of spatially detailed GIS data. The resulting model explained more than one-half of the observed variation in lake DOC. Wetlands contribute the most DOC to lakes per unit area, but upland forests also are important sources because forests dominate overall land coverage. The approach used in this study is valuable because it can be applied to large numbers of watersheds over a broad region. This allows estimation of the cumulative impacts of alteration in the spatial distribution and types of land cover within a watershed. Availability of spatial data of the type required for our approach is increasing rapidly, and resource agencies rely on GIS to manage these data. Our approach can be readily incorporated within a GIS framework and allows examination of scenarios of the impacts on water quality variables, such as DOC of losses of wetlands, alterations in forest management, or increases in conserved areas.
Parameter Estimate | Lower Support Interval | Upper Support Interval | ||
---|---|---|---|---|
Export (kg/ha/year | ||||
Dediduous Forest | 41.76 | 32.47 | 47.19 | |
Mixed Forest | 51.09 | 43.94 | 56.71 | |
Conifer Forest | 37.07 | 26.69 | 48.93 | |
Deciduous/Open | 23.80 | 0.00 | 127.79 | |
Open Vegitation | 932.12 | 419.45 | 1267.68 | |
Emergent Marsh | 140.49 | 25.29 | 240.24 | |
DeciduousForest Swamp | 1510.91 | 0.00 | 2000.00 | |
Conifer Forest Swamp | 146.73 | 150.22 | 215.61 | |
Dead Tree Swamp | 1350.10 | 634.55 | 1917.15 | |
Deciduous Shrub Swamp | 192.03 | 145.94 | 240.03 | |
Boadleaved Evergreen Shrub Swamp | 223.51 | 165.40 | 277.15 | |
Needle-leaved Evergreen Shrub Swamp | 213.91 | 162.57 | 256.70 | |
Distance Decay (/m)(*10^5) | ||||
Deciduous Forest | 0.2333 | 0 | 27.0107 | |
Mixed Forest | 7.8971 | 0 | 25.1919 | |
Conifer Forest | 1.0532 | 0 | 49.0359 | |
Deciduous/Open | 146.8400 | 0 | 1.0000 | |
Open Vegetation | 7130.6770 | 4064.4859 | 1.0000 | |
Emergent Marsh | 26.2005 | 0 | 1.0000 | |
Deciduous Forest Swamp | 6907.7929 | 2901.2730 | 1.0000 | |
Conifer Forest | 0.5727 | 0 | 8.4823 | |
Dead Tree Swamp | 9734.7103 | 5451.4378 | 1.0000 | |
Deciduous Shrub Swamp | 16.9852 | 0 | 162.2090 | |
Broadleaved Evergreen Shrub Swamp | 36.9025 | 0 | 410.7247 | |
Needle-leaved Evergreen Shrub Swamp | 2.1739 | 0 | 149.41.4 | |
In-Lake Decay (/year) | 0.921 | 0.755 | 1.123 | |
In-Lake Production (kg/ha/year) | 0.677 | 0.000 | 18.164 | |
1985 Loadin G (as % of 1984) | 101.906 | 93.754 | 111.078 | |
1986 Loadin G (as % of 1984) | 127.363 | 117.174 | 136.278 | |
1987 Loadin G (as % of 1984) | 111.485 | 98.107 | 118.174 | |
% of Upstream Lake Export Received | 64.178 | 50.859 | 75.966 |
References:
Beven K, Wood EF. Catchment geomorphology and the dynamics of runoff contributing areas. Journal of Hydrology 1983;65:139-158.
Boyer EW, Hornberger GM, Bencala KE, McKnight DM. Response characteristics of DOC flushing in an alpine catchment. Hydrological Processes 1997;11:1635-1647.
Cowardin LM, Golet FC. U.S. Fish and Wildlife Service 1979 wetland classification-a review. Vegetatio 1995;118:139-152.
Curran RP. Biological resources and diversity of the Adirondack Park. In: Proceedings of the Adirondack Park in the Twenty First Century, Technical Report 17, Volume 1, Albany, NY, 1990, pp. 414-461.
D'Arcy P, Carignan R. Influence of catchment topography on water chemistry in southeastern Quebec shield lakes. Canadian Journal of Fisheries and Aquatic Sciences 1997;54:2215-2227.
Dillon PJ, Molot LA. Dissolved organic and inorganic carbon mass balance in central Ontario lakes. Biogeochemistry 1997;36:29-42.
Driscoll CT, Driscoll KM, Roy KM, Mitchell MJ. Chemical response of lakes in the Adirondack region of New York to declines in acidic deposition. Environmental Science and Technology 2003;37:2036-2042.
Engstrom DR. Influence of vegetation and hydrology on the humus budgets of Labrador lakes. Canadian Journal of Fisheries and Aquatic Sciences 1987;44:1306-1314.
Goffe WL, Ferrier GD, Rogers J. Global optimization of statistical functions with simulated annealing. Journal of Econometrics 1994;60:65-99.
Hilborn R, Mangel M. The Ecological Detective: Confronting Models With Data. Princeton, NJ: Princeton University Press, 1997, 330 pp.
Kretser WA, Gallagher J, Nicholette J. Adirondack Lakes study 1984-87: an evaluation of fish communities and water chemistry. Presented at the Adirondack Lakes Survey Corporation, Ray Brook, NY, 1989.
Molot LA, Dillon PJ. Photolytic regulation of dissolved organic carbon in northern lakes. Global Biogeochemical Cycles 1997;11:357-365.
Pace ML, Cole JJ. Synchronous variation of dissolved organic carbon in lakes. Limnology and Oceanography 2002;47:333-342.
Primack AGB, Spada DM, Curran RP, Roy KM, Barge JW, Grisi BF, Bogucki DJ, Allen EB, Kretser WA, Cheeseman CC. Watershed scale protection for Adirondack wetlands: implementing a procedure to assess cumulative effects and predict cumulative impacts from development activities to wetlands and watersheds in the Oswegatchie, Black, and Greater Upper Hudson River watersheds of the Adirondack Park. Part I. Resource mapping and data collection. Part II. Resource data analysis, cumulative effects assessment, and determination of cumulative impacts. Presented at the New York State Adirondack Park Agency Meeting, Ray Brook, NY, 2000.
Rasmussen JB, Godbout L, Schallenberg M. The humic content of lake water and its relationship to watershed and lake morphometry. Limnology and Oceanography 1989;34:1336-1343.
Reche I, Pace ML, Cole JJ. Relationship of trophic and chemical conditions to photobleaching of dissolved organic matter in lake ecosystems. Biogeochemistry 1999;44:259-280.
Roy KM, Allen EB, Barge JW, Ross JA, Curran RP, Bogucki DJ, Franz DA, Kretser WA, Frank MM, Spada DM, Banta JS. Influences on wetlands and lakes in the Adirondack Park of New York State: a catalog of existing and new GIS datalayers for the 400,000 hectare Oswegatchie/Black River watershed. Presented at the New York State Adirondack Park Agency, Ray Brook, NY, 1997.
Wetzel RG. Limnology: Lake and River Ecosystems. San Diego, CA: Academic Press, 2001.
Journal Articles on this Report : 1 Displayed | Download in RIS Format
Other project views: | All 16 publications | 1 publications in selected types | All 1 journal articles |
---|
Type | Citation | ||
---|---|---|---|
|
Canham CD, Pace ML, Papaik MJ, Primack AGB, Roy KM, Maranger RJ, Curran RP, Spada DM. A spatially explicit watershed-scale analysis of dissolved organic carbon in Adirondack lakes. Ecological Applications 2004;14(3):839-854. |
R826762 (2001) R826762 (Final) |
Exit Exit |
Supplemental Keywords:
ecological effects, aquatic, terrestrial, aquatic ecosystems, land water interactions, limnology, dissolved organic carbon, DOC, likelihood analysis, landscape, ecosystem protection/environmental exposure risk, geographic area, ecology, ecosystem, state, water, watersheds, Adirondack lake, geographic information system, GIS, New York, NY, carbon allocation, ecological variation, ecosystem assessment, integrated ecological assessment, land use, model ecosystem effects, modeling, regional scale impacts, remote sensing imagery, temporal scale, water management options, water quality., RFA, Scientific Discipline, Water, Geographic Area, Ecosystem Protection/Environmental Exposure & Risk, Water & Watershed, Hydrology, State, Regional/Scaling, Ecology and Ecosystems, Ecological Risk Assessment, Watersheds, aquatic, carbon allocation, landscape scale determinations, model ecosystem effects, regional analysis, ecosystem assessment, modeling, temporal scale, ecological variation, dissolved organic carbon, regional scale impacts, aquatic ecosystems, water quality, GIS, water management options, Adirondack Lake, integrated ecological assessment, remote sensing imagery, wetland, land useRelevant Websites:
Progress and Final Reports:
Original AbstractThe perspectives, information and conclusions conveyed in research project abstracts, progress reports, final reports, journal abstracts and journal publications convey the viewpoints of the principal investigator and may not represent the views and policies of ORD and EPA. Conclusions drawn by the principal investigators have not been reviewed by the Agency.