Grantee Research Project Results
Final Report: A Watershed Classification System for Improved Monitoring and Restoration: Landscape Indicators of Watershed Impairment
EPA Grant Number: R831369Title: A Watershed Classification System for Improved Monitoring and Restoration: Landscape Indicators of Watershed Impairment
Investigators: Prince, Stephen D. , Weller, Donald E. , Jordan, Thomas E. , Goetz, Scott J.
Institution: University of Maryland - College Park , Woods Hole Research Center , Smithsonian Environmental Research Center
Current Institution: University of Maryland - College Park , Smithsonian Environmental Research Center , Woods Hole Research Center
EPA Project Officer: Packard, Benjamin H
Project Period: February 1, 2004 through January 31, 2007
Project Amount: $896,497
RFA: Development of Watershed Classification Systems for Diagnosis of Biological Impairment in Watersheds and Their Receiving Water Bodies (2003) RFA Text | Recipients Lists
Research Category: Watersheds , Water
Objective:
To develop a watershed classification scheme based on recent, much improved, comprehensive watershed data sets to diagnose aquatic ecosystem impairment and to target resource management. To use hydrologic metrics, nutrient budgets incorporating point and non-point source/sinks, and landscape function metrics to provide indicators of aquatic ecosystem condition (hydrology, plant, fish, macroinvertebrates, water quality) in reference watersheds. To identify the watershed variables most relevant to prediction of impairment of the receiving water bodies by developing a set of empirical classification models for multiple scales. To develop classifications for mid-Atlantic training watersheds, test them in the mid-Atlantic, apply the entire methodology in southern New England (MA, RI, CT), and to generalize the methods for future national application.
Summary/Accomplishments (Outputs/Outcomes):
Enhanced land cover and riparian buffer mapping
This project component focused on how the expanding urbanization in the Chesapeake Bay watershed is likely to impact stream water quality and whether land cover is a reliable indicator. To address this we mapped land cover variables across the study area (impervious and tree cover) and used these in conjunction with other land cover products (e.g. from the National Landcover Database and National Agricultural Statistical Service) for the work described in the following sections. A summary of this work was published as part of an invited journal paper (Jantz and Goetz 2007, see publication list).
Following the advancements made in riparian buffer mapping described in the earlier progress reports, we prepared an invited overview paper for a special issue of JAWRA on riparian buffers (Goetz 2006). The various sources of satellite image data and map information were reviewed, examples of their application to riparian buffer mapping and stream health assessment provided, and future prospects for improved buffer monitoring discussed. The paper focused not only on mapping vegetation cover in riparian zones, but also on monitor the changes taking place, targeting restoration activities, and assessing the success of previous management actions (see publication list).
Additional work on this topic has been reported in a special issue of Remote Sensing of Environment entitled Monitoring Freshwater and Estuarine Systems, edited by one of the investigators (S. Goetz), which contained 15 peer-reviewed papers focused on this topic (including three from personnel supported by this project - see publication list). An attachment includes the cover and table of contents of this special issue.
Our advances in enhancing land cover and riparian buffer mapping contributed fundamental results for a long-term study (The State of Chesapeake Forests) released in conjunction with the establishment of an historic Forest Conservation Directive signed by members of the Chesapeake Executive Council (comprised of the Governors of Maryland, Virginia, and Pennsylvania, the Mayor of Washington DC, and the EPA Administrator for all federal agencies). The Directive sets agreements to conserve private forest lands in the Chesapeake Bay watershed that are most vulnerable to development and most important to maintaining water quality, a key topic in the current research.
Nutrient budgets
We developed county-level agricultural nutrient budgets and evaluated the utility of variables from those budgets to predict watershed nutrient losses and to help classify streams by likely nutrient status. Using procedures previously used for nitrogen (Jordan and Weller 1996), county data (fertilizer application, land use, crop types, animal numbers, plant and animal harvests, human population) were integrated with general information on nitrogen and phosphorus fluxes (e.g., atmospheric deposition, animal feeding efficiencies, N fixation rates) to estimate the net anthropogenic phosphorus input (NAPI) for each county in the Chesapeake Bay watershed.
NAPI is an index of phosphorus pollution potential estimated by quantifying all phosphorus inputs and outputs for each county. Inputs include fertilizer applications and non-food phosphorus uses, while trade of food and feed can be an input or an output. We used land cover data to apportion the NAPI for each county to four land use types (row crops, developed land, pasture land, and enclosed animal facilities) and to a fifth category representing losses in transfer and storage. Our phosphorus budgets include many refinements over our previous calculations for nitrogen (Jordan et al. 1996), so we also incorporated those refinements into improved county nitrogen budgets to produce revised estimates of net anthropogenic nitrogen input (NANI) for each county.
These NAPI calculations and implications for the Chesapeake Bay were published in the journal Biogeochemistry (Russell et al. 2008). The average of 1987, 1992, 1997, and 2002 NAPI for individual counties ranged from 0.02 to 78.46 kg P ha-1year-1. The overall area-weighted average NAPI for 266 counties in the region was 4.52 kg P ha-1year-1, indicating positive net phosphorus input that can accumulate in the landscape or can pollute the water. Large positive NAPI values were associated with agricultural and developed land cover. County area-weighted NAPI increased from 4.43 to 4.94 kg P ha-1year-1 between 1987 and 1997 but decreased slightly to 4.86 kg P ha-1year-1 by 2002. Human population density, livestock unit density, and percent row cropland combined to explain 83% of the variability in NAPI among counties. Around 10% of total NAPI entering the Chesapeake Bay watershed is discharged into Chesapeake Bay. The developed land component of NAPI had a strong direct correlation with measured phosphorus discharges from major rivers draining to the Bay (R2 = 0.81), however, the correlation with the simple percentage of developed land was equally strong. Our results help identify the sources of P in the landscape and evaluate the utility of NAPI as a predictor of water quality.
For Maryland only, we also enhanced our budget calculations to explore the role of urban fertilizer application to county NAPI and NANI (Russell et al. in prep. b). In more urbanized counties, fertilizer applications to lawns and golf courses dominated net anthropogenic N and P inputs, but cropland fertilizer became more important above about 9% cropland in a county. Annual urban fertilizer input in MD ranged up to 46.8 kg N ha-1 and 1.87 kg P-1. Simple regression models successfully predicted urban fertilizer application from land cover or human population data (R2 between 0.83 and 0.99), and these relationships proved useful in extending the Maryland analysis to the rest of the Chesapeake drainage.
Landscape indicators of stream N and P loads
Stream buffer efficacy
New geographical and statistical analyses were developed to better estimate stream buffer effects in whole catchments. This geographical analysis accounts for the length of buffer traversed by every surface flow path connecting a nutrient source area to a stream. This approach correctly scales up from transects (flow paths), considered in field studies, to whole catchments. The statistical analysis relates stream nutrients to land cover proportions (cropland, developed land, and grassland) and to the prevalence of riparian buffers between croplands and streams (as determined by the flow-path analysis).
Two papers on geographic metrics for quantifying riparian buffer potential in landscapes have been published. The first paper (Baker et al. 2006) developed new landscape buffer potential estimates that account for hydrologic connectivity of nutrient source areas with streams. The new metrics are more functionally based and more independent of whole-watershed land cover traditional analyses that estimate buffer potential using land cover proportions within a fixed distance of streams. The metrics revealed important differences in land cover arrangement among different physiographic provinces. The second paper (Baker et al. 2007) applied the metrics to examine the effects of the resolution of stream maps on estimates of riparian buffer potential. Overall, increasing stream map resolution led to reduced estimates of nutrient retention potential in riparian buffers. In some watersheds, switching from a coarse resolution to a fine-resolution stream map completely changed our perception of a stream network from well buffered to largely unbuffered. Because previous, broad-scale analyses of riparian buffers used coarse-resolution stream maps, those studies may have overestimated landscape-level buffer prevalence and effectiveness. Stream map resolution affects inferences about whether retention occurs in streams or riparian zones.
We applied the geographic analyses in statistical models to explain nitrate concentrations in the streams draining 321 catchments in the Coastal Plain (CP), Piedmont (PD), and Appalachian Mountain (AM) physiographic provinces of the Chesapeake Bay drainage (Baker et al. submitted). Estimates of the loading of nitrate from cropland, the potential of buffers to remove nitrate, the prevalence of buffers within each watershed, and the nitrate removed by buffers in each watershed all differed among physiographic provinces. The water leaving PD croplands contained more than three times more nitrate than CP croplands, and AM croplands were intermediate. The percentage of inflowing nitrate that can be removed by a riparian buffer was highest in CP (81%), intermediate in the PD (41%), and lowest in AM (24%). In the CP, drainage from 53% of the cropland passes through a riparian buffer, but only 32% of PD croplands and 13% of AM croplands are buffered. We combined the percentages of unbuffered cropland with the nitrate removal potentials to estimate the maximum benefits of buffer restoration. The PD province provides the greatest opportunities to reduce nitrate concentrations through buffer restoration. The highest percentage reductions in stream nitrate relative to current concentrations (54%) could be achieved in the CP. However, because average current nitrate concentration is higher in the PD (3.9 mg/l) than the CP (1.0 mg/l), in absolute quantities, buffer restoration in the PD could reduce stream nitrate loads more (by 1.0 mg/l) than in the CP (0.6 mg/l) even though the percentage reduction in the PD is lower (29%). The catchment-scale effects of riparian buffers differ among and within physiographic provinces, and the aggregate removal by riparian buffers across watersheds is not as high as the potential suggested by studies of field-to-stream transects. These differences should be considered in developing expectations for the water quality benefits of riparian buffer restoration and in estimating the likely nitrate concentration in unmonitored streams.
Land cover and county nutrient budgets as predictors
The county nutrient budgets were used to provide predictors of water quality in study watersheds (Russell et al. in prep a). Models were built relating land cover and net anthropogenic nutrient inputs to measured nutrient discharges from 78 streams in all physiographic provinces of the Chesapeake basin. Land cover percentages in each study watershed were calculated using the RESAC land cover map (Goetz et al. 2004). We apportioned previously calculated county phosphorus and nitrogen budgets (see above) to watersheds based on land cover proportions within the counties and the watersheds. County net anthropogenic nitrogen inputs (NANI) averaged 63.35 kg N ha yr-1 between 1987 and 2002. When county NANIs were apportioned to specific land uses, row crop NANI averaged 47.74 kg N ha-1 yr-1, pasture NANI 1.93 kg N ha-1 yr-1, and developed NANI 6.80 kg N ha-1 yr-1. County NANI components were then compared to their respective land cover percentages. Depending on physiographic province and nutrient form, univariate models relating N or P discharge to land cover or budget components explained between 0 and 74% of total phosphorus flux and between 44 and 96% of total nitrogen flux in streams.
Landscape indicators of stream biological condition
Statistical models were built to predict stream biological condition from landscape characteristics, and then applied the models to classify likely conditions in unstudied streams (Maloney et al. in revision a). We used data from the Maryland Biological Stream Survey (MBSS) to evaluate the utility of several relatively new classification algorithms. We compared five modeling methods (classification trees, conditional inference trees, random forests [RF], conditional random forests [cRF], and ordinal linear regression [OLR]) to predict categorical stream benthic macroinvertebrate integrity scores (Very Poor, Poor, Fair, and Good) taken from the Maryland Biological Stream Survey. Predictor variables included land use and land cover (e.g., impervious surface, row crop agriculture, and population density) and landscape measures (annual precipitation and watershed area). There were 1561 sites; 80% (1248) were used as a training data set to build models and 20% (313) were used as an independent evaluation data set. The RF and cRF models most accurately predicted observed integrity scores in the evaluation data set, but we selected the cRF as the best model due to known weaknesses in the RF model (e.g., biased variable selection). Impervious surface percentage was the most important variable in the cRF model, and there was a rapid increase in the probability a site was Very Poor or Poor as impervious cover increased up to 20%.
The calibrated cRF model was applied to predict stream biological conditions in all 7,908 small, non-tidal stream reaches within the Maryland portion of the Chesapeake basin (Maloney et al. in revision a). The model predicted that 33.8% of these reaches are Fair, 29.9% Good, 22.7% Poor, and 13.6% Very Poor. This work indicates that model predictions for unsurveyed streams can help target field studies to identify high-quality streams deserving of conservation efforts and impaired streams needing restoration.
The same classification methods (classification trees, conditional inference trees, random forests [RF], conditional random forests [cRF], and ordinal linear regression [OLR]) were applied to benthic IBI scores from small streams throughout the Chesapeake Bay basin (Maloney et al. in revision b). We used a data set that combined stream data from three stream surveys (MAIA, MAHA, and MBSS). The manuscript presenting this analysis was not accepted for publication because reviewers objected to combining the data from the different stream surveys. We are now collaborating with the Chesapeake Bay Program to repeat this analysis with a new Bay-wide data set that they are currently assembling. Their methods include extensive consultation with experts from the different stream surveys that are being combined, and extensive comparisons of key metrics to reference sites within each survey to ensure consistency throughout the combined data set. When their effort is complete this fall (2009), we will repeat our analysis. As with our MBSS analysis (Maloney et al in revision a), we will identify the best classification model for relating stream condition to landscape characteristics, and then apply that model to predict the most likely biological condition for every first to third order stream reach within the Chesapeake drainage (Maloney et al. in revision b).
The MBSS data were also used to develop an improved method for detecting, quantifying, and diagnosing ecological community thresholds in response to anthropogenic disturbance (Baker and King in revision). Threshold Indicator Taxa Analysis (TITAN) combines change-point analysis and indicator species analysis. Change-point analysis is a non-parametric technique that orders and partitions observations along an environmental gradient in a manner identical to parametric single-level multivariate or univariate regression tree analysis. As optimal partitioning can be sensitive to sample distribution along the environmental gradient, a bootstrap resampling procedure assesses uncertainty associated with the observed change-point value. TITAN replaces conventional univariate (e.g., species richness or an index of integrity [IBI]) or multivariate (Euclidean or Bray-Curtis distance) community responses with taxon-specific, indicator value (IndVal) scores from indicator species analysis. IndVal scores are the product of cross-group relative abundance (within-species) and within- group occurrence frequency, so they effectively eliminate the influence of group size on comparisons among groups.
We applied the TITAN method to the MBSS data set to explore the responses of stream invertebrates to impervious surfaces (King et al. in revision). Analysis of more than 1800 streams and their watersheds revealed sharp declines in 30% of the stream invertebrate taxa in response to less than 2% difference in watershed impervious cover. This result was robust to bootstrap resampling and consistent across physiographic provinces, although province-specific responses to impervious cover were evident. More importantly, comparisons of TITAN with other forms of threshold analysis demonstrate that TITAN is unique in its ability to detect these losses and to establish taxon-specific evidence of a threshold effect. In addition to its diagnostic and interpretive advantages, TITAN can be employed in a tree-based framework of recursive partitioning or variable importance assessment for watershed classification, characterization, and assessment.
Landscape indicators of stream hydrological responses
In addition to nutrient loads and biological indicators, metrics and models for classifying the hydrological condition of streams (Baker et al in prep) have also been studied. We developed a program in the R statistical programming language to analyze hydrologic time series and calculate a suite of published flow metrics that have been used as indicators of effects of watershed disturbances on stream flow. These include measures of flow magnitude, variability, baseflow, and floods. Study watersheds included the SERC automated flow sampling stations (n=74 across 3 physiographic regions) as well as 15 USGS gauges from throughout the Coastal Plain, Piedmont, and Appalachian Highland portions of the Chesapeake Basin. Watersheds were selected to represent a broad range of anthropogenic disturbance gradients, but focused primarily on urban and suburban development. Watersheds were characterized by land cover, measures of in-line impoundments, as well as a suite of terrain- based descriptors linked to hydrologic response and water residence time in the literature. All land cover patterns were analyzed as simple proportions as well as weighted proportions according to their spatial distribution relative to streams and sampling stations.
Regional climate provided the strongest correlation with flow regimes. Therefore, the analysis was repeated using the precipitation time series to filter precipitation signals from the hydrologic record and thereby isolate the effects of watershed properties on hydrologic response. Filtering removed different precipitation signals from the hydrologic record that would otherwise confound analysis of watersheds in distinct portions of the Chesapeake Basin. Resulting stream flow classes exhibited a strong spatial structure. Impoundments had a substantial impact on flow regimes, but this impact varied with dam management. Watershed imperviousness was the best anthropogenic predictor of hydrologic response across provinces leading to increased variability and the magnitude of the responses to extreme events, but the strongest overall effects resulted from terrain descriptors and elevation derivatives, especially once regional climate signals were removed from metric values. As with the water quality and biological models (above), we are applying the hydrological response models to classify likely hydrological conditions in all first-third order stream reaches of the Chesapeake basin (Baker et al in prep).
Testing indicators of impairment in New England
The maps and stream macroinvertebrate data sets were also used to study the power of much simpler land cover variables than those discussed above with stream biota. Initially the methods were applied to watersheds of the mid-Atlantic region (particularly the Maryland Biological Stream Survey, MBSS). The paper reporting this work was included in the special issue described above (TOC in appendix). As with related work for much smaller watersheds in Montgomery county Maryland (Snyder and Goetz 2005, see publication list), impervious and tree cover alone were found to have significant predictive ability for stream biota, and this varied with physiographic province, as was shown in the analyses of nutrient budgets (above). Landscape configuration (see below) was important in many, but not all cases for improving the predictive quality of statistical models estimating stream biota metrics.
The approach was extended to model stream biota using simplified land cover information in watersheds in southern New England. We adapted the statistical models of the relationship of stream health indicators (specifically the richness and abundance of stream macroinvertebrates, and integrated indices of stream biological integrity) into simple procedures that can be conducted in a Geographic Information System (GIS).The result of that analysis was also prepared for publication, which is submitted and still pending publication at this time.
We also applied the models to the Upper Delaware Scenic and Recreational River (UPDE). As part of this effort we developed what we refer to as a Standard Operating Procedure (SOP) document that leads users through the necessary steps to predict stream biota using the models together with widely available land cover data sets (e.g. from the NLCD). That SOP is attached. This work advances the estimation of stream health characteristics in areas where MBSS measurements do not exist but can be estimated using comparable land cover information, and informs guidelines for management and restoration.
Incorporating landscape metrics in the USGS SPARROW model
The aim here was to determine whether the effects of catchment and riparian stream buffer-wide urban and non-urban land cover/land use (LC/LU) on total nitrogen (TN) and total phosphorus (TP) discharges, as indicated in other parts of this project, could be captured by a highly generalized model represented by the widely-used USGS hybrid statistical-process model, SPAtially Referenced Regression On Watershed Attributes (SPARROW) The model was calibrated with 1997 watershed-wide, average annual TN and TP discharges to the Chesapeake Bay. Two variables were predicted: 1) yield (mass for a specified time normalized by drainage area) per unit watershed area and 2) loadings (mass for a specified time) delivered to the upper estuary. The 166,534- km2 watershed was divided into 2339 catchments averaging 71 km2. LC/LU was described using sixteen classes applied to both the catchments and to riparian stream buffers alone. Seven distinct landscape metrics were evaluated. In all, 167 (TN) and 168 (TP) LC/LU class metric combinations were tested in each model calibration run. Runs were made with LC/LU in six fixed riparian buffer widths (31, 62, 125, 250, 500, and 1000m) and entire catchments. The significance of the non-point source type (land cover, manure and fertilizer application, and atmospheric deposition) and factors affecting land-to-water delivery (physiographic province and natural or artificial land surfaces) was assessed. Yield and loadings were estimated to be 1.449 x 108 and 5.367 x 106 kg yr-1, respectively. Five of the 167 TN and three of the 168 TP landscape metrics were shown to be significant (p value ≤ 0.05) either for non-point sources or land-to-water delivery variables. The model with a 31 m riparian stream buffer width accounted for the highest variance of mean annual TN (r2 =0.94) and TP (r2 =0.75) yield and TN and TP loadings entering the Chesapeake Bay. Land cover metrics were shown to improve the precision of estimated TN and TP annual loadings and may suggest changes in land management that may be beneficial in control of nutrient discharges to the Chesapeake Bay and similar watersheds elsewhere.
The success of the calibration and analysis of significant causative factors has enabled the effects of simulated land cover/land use (LC/LU) change from 2000 to 2030 on nutrient loadings to the Chesapeake Bay to be estimated. Anticipated watershed-wide LC/LU change was provided by a growth forecast model that provides spatially explicit probabilities of conversion to impervious surface. The total nitrogen (TN) and total phosphorus (TP) loadings estimated to enter the Chesapeake Bay were reduced by 20% and 19%, respectively. In general, as developed land replaced other LC/LUs from 2000 to 2030, TN and TP discharge was predicted to be significantly reduced by losses of non-point non-urban source loadings, yields, and land-to-water delivery. The simulation results suggest future changes in landscape composition and configuration at catchment and riparian stream buffer width scales that could lower TN and TP discharges to the estuary.
Further analysis of the LC/LU changes predicted to occur by 2030 suggested that leapfrog growth that creates smaller, detached patches on crop and pasture will lower catchment total nitrogen (TN) discharge. This was in contrast to gains in TN discharge that were predicted to occur with infill and peripheral growth, predominantly in cover types other than crop and pasture. Furthermore, development in smaller, detached patches would also reduce urban, non-point N discharge in catchments over this period. These results suggest that the strategic placement of leapfrog development may be capable of reducing future TN discharge to the Chesapeake Bay and in similar watersheds elsewhere.
These results are reported in three papers currently under review for Ecological Indicators, Ecological
Engineering, and Science of the Total Environment.
Conclusions:
This research has established a set of landscape and land cover properties that can predict stream health for subcatchments of the Chesapeake Bay and in a simpler application to southern New England and the upper Delaware River. While land use and impervious cover were the best indicators, landcover configuration, as quantified by landscape metrics, was found to be a significant factor in nitrogen and phosphorus discharges, biological indices, and hydrological parameters. Each of these relationships was investigated at different levels of detail, using detailed field measurements of watershed nutrient budgets and water quality, and land cover types as proportions of catchments. Classification schemes were demonstrated that can predict the condition of streams based on their watershed characteristics. These models are at different levels of detail, from complete nutrient budgets, through simple biotic indices and land cover to the generalized USGS SPARROW model. In all cases, potential applications to watershed classification were noted.
Journal Articles on this Report : 14 Displayed | Download in RIS Format
| Other project views: | All 70 publications | 19 publications in selected types | All 15 journal articles |
|---|
| Type | Citation | ||
|---|---|---|---|
|
|
Baker ME, Weller DE, Jordan TE. Improved methods for quantifying potential nutrient interception by riparian buffers. Landscape Ecology 2006;21(8):1327-1345. |
R831369 (Final) R828684 (Final) |
Exit |
|
|
Baker ME, Weller DE, Jordan TE. Effects of stream map resolution on measures of riparian buffer distribution and nutrient retention potential. Landscape Ecology 2007;22(7):973-992. |
R831369 (2006) R831369 (Final) R828684 (Final) |
Exit |
|
|
Goetz S, Fiske G. Linking the diversity and abundance of stream biota to landscapes in the mid-Atlantic USA. Remote Sensing of Environment 2008;112(11):4075-4085. |
R831369 (Final) R828684 (Final) |
Exit Exit Exit |
|
|
Goetz SJ. Remote sensing of riparian buffers: past progress and future prospects. Journal of the American Water Resources Association 2006;42(1):133-143. |
R831369 (2006) R831369 (Final) R828684 (Final) |
Exit Exit |
|
|
Goetz SJ, Gardiner N, Viers JH. Monitoring freshwater, estuarine and near-shore benthic ecosystems with multi-sensor remote sensing:an introduction to the special issue. Remote Sensing of Environment 2008;112(11):3993-3995. |
R831369 (Final) |
Exit Exit Exit |
|
|
Jantz CA, Goetz SJ. Can smart growth save the Chesapeake Bay? Journal of Green Building 2007;2(3):41-51. |
R831369 (Final) |
Exit Exit |
|
|
Lang MW, Kasischke ES, Prince SD, Pittman KW. Assessment of C-band synthetic aperture radar for mapping and monitoring Coastal Plain forested wetlands in the Mid-Atlantic Region, U.S.A. Remote Sensing of Environment 2008;112(11):4120-4130. |
R831369 (Final) |
Exit Exit Exit |
|
|
Maloney KO, Weller DE, Russell MJ, Hothorn T. Classifying the biological condition of small streams: an example using benthic macroinvertebrates. Journal of the North American Benthological Society 2009;28(4):869-884. |
R831369 (Final) |
Exit |
|
|
Maloney KO, Munguia P, Mitchell RM. Anthropogenic disturbance and landscape patterns affect diversity patterns of aquatic benthic macroinvertebrates. Journal of the North American Benthological Society 2011;30(1):284-295. |
R831369 (Final) |
Exit |
|
|
Nielsen EM, Prince SD, Koeln GT. Wetland change mapping for the U.S. mid-Atlantic region using an outlier detection technique. Remote Sensing of Environment 2008;112(11):4061-4074. |
R831369 (Final) |
Exit Exit Exit |
|
|
Roberts AD, Prince SD, Jantz CA, Goetz SJ. Effects of projected future urban land cover on nitrogen and phosphorus runoff to Chesapeake Bay. Ecological Engineering 2009;35(12):1758-1772. |
R831369 (Final) |
Exit Exit Exit |
|
|
Roberts AD, Prince SD. Effects of urban and non-urban land cover on nitrogen and phosphorus runoff to Chesapeake Bay. Ecological Indicators 2010;10(2):459-474. |
R831369 (Final) |
Exit Exit Exit |
|
|
Russell MJ, Weller DE, Jordan TE, Sigwart KJ, Sullivan KJ. Net anthropogenic phosphorus inputs:spatial and temporal variability in the Chesapeake Bay region. Biogeochemistry 2008;88(3):285-304. |
R831369 (Final) |
Exit Exit |
|
|
Snyder MN, Goetz SJ, Wright RK. Stream health rankings predicted by satellite derived land cover metrics. Journal of the American Water Resources Association 2005;41(3):659-677. |
R831369 (2005) R831369 (Final) |
Exit Exit |
Supplemental Keywords:
RFA, Scientific Discipline, Water, Ecosystem Protection/Environmental Exposure & Risk, Water & Watershed, Hydrology, Monitoring/Modeling, Environmental Monitoring, Ecological Risk Assessment, Ecology and Ecosystems, Watersheds, ecosystem modeling, aquatic ecosystem, watershed classification, continuous monitoring, aquatic ecosystems, water quality, ecosystem restoration, environmental stress, hydrologic modeling, land use, land managementProgress 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.