Grantee Research Project Results
2008 Progress Report: Hydrologic Forecasting for Characterization of Non-linear Response of Freshwater Wetlands to Climatic and Land Use Change in the Susquehanna River Basin
EPA Grant Number: R833013Title: Hydrologic Forecasting for Characterization of Non-linear Response of Freshwater Wetlands to Climatic and Land Use Change in the Susquehanna River Basin
Investigators: Wardrop, Denice Heller , Ready, Richard C , Easterling, William Ewart , Brooks, Robert P. , Shortle, James S. , Duffy, Christopher , Dressler, Kevin , Najjar, Raymond
Current Investigators: Wardrop, Denice Heller , Easterling, William Ewart , Brooks, Robert P. , Shortle, James S. , Dressler, Kevin , Duffy, Christopher , Najjar, Raymond , Ready, Richard C
Institution: Pennsylvania State University
EPA Project Officer: Packard, Benjamin H
Project Period: April 20, 2007 through April 19, 2011
Project Period Covered by this Report: April 20, 2008 through April 19,2009
Project Amount: $899,656
RFA: Nonlinear Responses to Global Change in Linked Aquatic and Terrestrial Ecosystems and Effects of Multiple Factors on Terrestrial Ecosystems: A Joint Research Solicitation- EPA, DOE (2005) RFA Text | Recipients Lists
Research Category: Climate Change , Aquatic Ecosystems
Progress Summary:
Our project team is organized into four working groups dealing with: (a) climate scenarios, (b) land cover scenarios, (c) hydrologic modeling, and (d) ecological responses. Progress in each of these areas is described below:
1a. Climate scenarios
The goal of the climate model analysis portion of this project is to make estimates of the likely changes in the climate of the Susquehanna River Basin (SRB) by the middle of the 21st century.
General Circulation Model (GCM) output from 21 models, and 2 observational data sets, were placed on a 1º grid within the SRB. GCM output was linearly interpolated to this grid, whereas observational data were averaged within each 1º grid box.
Ten metrics were computed for model evaluation (Table 1). The first four are based on monthly means between 1961 and 1997 and the rest are based on daily means between 1979 and 1997. The last four metrics are considered extreme precipitation indices (Frich et al. 2002), which may be of particular relevance for wetlands. The multi-model average of these metrics was also computed, after first averaging multiple realizations for a given model.
Table 1. Metrics for the evaluation of climate models.
(1) Annual cycle of mean temperature
|
(2) Annual cycle of mean precipitation
|
(3) Annual cycle of interannual temperature variability (standard deviation)
|
(4) Annual cycle of interannual precipitation variability (standard deviation)
|
(5) Mean annual cycle of intramonthly temperature variability (std. dev.)
|
(6) Mean annual cycle of intramonthly precipitation variability (std. dev.)
|
(7) Mean annual cycle of the maximum number of consecutive dry days within a month
|
(8) Mean annual cycle of the maximum 5-day precipitation total within a month
|
(9) Mean annual cycle of precipitation intensity (total monthly precipitation divided by the number of wet days*) *A wet day is considered to be a day in which precipitation exceeds 1 mm.
|
(10) Mean annual cycle of the number of days with precipitation exceeding 10 mm
|
Models were compared with each other using a target diagram (Jolliff et al. 2008), which breaks the root mean square error into two components: the overall bias and the pattern error. In the target diagram, the bias is plotted on the y-axis and the pattern error is plotted on the x-axis. Both quantities are normalized by the standard deviation of the observed data set, which allows different data types to be plotted and compared on the same diagram. The pattern error is always positive, but is multiplied by -1 on a target diagram if the model standard deviation is greater than the standard deviation of the observations.
We also computed an overall performance index for each model based on multiple metrics, following the approach of Reichler and Kim (2008). First, an overall squared error is computed for each variable (v, one of the 10 metrics) and for each model (m). The resulting value is then normalized by its average over all of the models. Finally, the overall performance index is computed by averaging over all of the metrics. Three performance indices were used: one is based on the first six metrics (non-extreme event metrics), the second is based on the extreme event metrics only, and the final index included all 10 metrics. 95% confidence intervals were constructed about the indices using bootstrapping methods. Model skill does not vary dramatically among the 1º grid cells chosen for evaluation.
The GCMs vary dramatically in their ability to simulate the mean annual cycle in temperature, with maximum errors for any given month being as large as 5º C. However, the multi-model average agrees very well with the observations, with the overall bias being only -0.2º C. The annual cycle in the multi-model mean is slightly too large, with July 0.6º C too warm and January 0.9º C too cold. Similar biases (too cool and too seasonal) were found by Najjar et al. (2008) in their analysis of models from IPCC’s Third Assessment Report (TAR), but the errors were much larger. Thus, GCMs have improved markedly in their ability to simulate the mean annual cycle of temperature in the SRB.
Errors are greater for the simulations of the mean annual cycle of precipitation than they are for that of temperature. Precipitation errors occasionally reach 100% for some models in some months. The multi-model mean is too wet by 14 mm mon-1, or 18%. Models tend to have too much variability in the mean annual cycle, and there is a tendency for models to have the largest precipitation in the spring, in contrast to the observations, where summer is the wettest season.
Results for interannual and intramonthly variability in temperature and precipitation are similar to those for the mean annual cycles in these variables: any individual model can have large errors, but the multi-model mean is much more accurate. Overall, the models have slightly too much interannual variability in temperature and not enough in precipitation.
One might expect numerical models with grid resolution of several hundreds of km to do a poor job of capturing extreme precipitation. However, our results suggest otherwise, with the multi-model average in fairly good agreement with the observations. The main error is a slight tendency for models to be not extreme enough, except for extreme precipitation days in the winter. In addition, the multi-model mean is significantly (at the 95% confidence level) superior to any individual model when comparing performance indices computed for all 10 metrics, as well as performance indices for the first six metrics.
Given the superiority of the multi-model mean in simulating the past, we have used it for projections into the 21st century. The projections show an average warming of 2.8º C, which is spread out fairly evenly throughout the year. Precipitation increases mainly during the spring and winter, with an average increase of about 10% between November and May. Interannual variability of temperature is projected to increase in the summer and decrease in the winter, with changes of about 10% in each season; the latter may be related to decreases in snow cover. Interannual variability in precipitation is projected to increase rather uniformly throughout the year by about 10%. Submonthly variability in temperature is projected to increase by roughly 10% in all seasons except the summer. Submonthly precipitation variability is projected to increase by a few percent from late fall to early spring.
Extreme precipitation indices show modest changes by mid century, except perhaps the number of days above 10 mm, which increases by about 10% in the spring. This is a surprising finding given that other studies have found widespread increases in precipitation extremes in GCMs for many regions of the world.
1b. Land Use Change Modeling and Projection
The following data were assembled for the SRB and other parts of Pennsylvania:
- Land cover in 1992 (NLCD)
- Land cover in 2001 (NLCD)
- Land cover change between 1992 and 2001 (USGS)
- Elevation and Slope (USGS)
- Protected Lands (GAP Analysis Project)
- Location of major roads
- County and MCD boundaries (Census)
- Population and Demographics in 1990 and 2000 (Census)
For each minor civil division (MCD) in the study area, the following statistics were calculated (land cover measures exclude protected lands):
- Amount of open space in 1992 (forests, agricultural land, and wetlands)
- Amount of open space in 2001
- Percent loss of open space between 1992 and 2001
- Population density in 1990
- Population density in 2000
- Change in population density 1990 to 2000
- Distance to nearest interstate highway
- Demographics of population (income, % over 65 years old)
Using this dataset, an open space loss model was estimated that explains how much open space was lost in each MCD during the period 1992 to 2001. This model found that MCDs with more open space initially lost a smaller proportion of that open space, MCDs with higher population growth lost more open space, and that MCDs with higher initial population lost more open space.
To predict future open space loss, MCD-level population projections were estimated from county-level data for the period 2010 to 2050. A model for allocating population change was estimated based on observed population change between 1990 and 2000. The model regressed MCD-level population change on county-level population change, initial population density in the MCD, and quantity of open space available for development. This model was used to project population change for each MCD in the study area. These predictions were adjusted proportionally to assure that the total population change across all MCDs in a county matched the predicted change for the county. Using the MCD-level population predictions combined with the MCD-level land use change model, land use change (open space loss) was projected for each MCD in the study area.
1c. Hydrologic modeling
The Little Juniata/ Spruce Creek subwatershed served as a pilot for this study, and was the focus of initial hydrologic modeling, as well as field data collection, including monitoring wells and ecological sampling. Some results of the hydrologic modeling are described below.
The predicted spatial distribution of recharge in the Little Juniata/Spruce Creek show high values along the higher order stream segments and lower values in the upland lower order portions.
Figure 1. Predicted (a) spatial distribution of recharge and (b) gaining and losing streams, in the Little Juniata/ Spruce Creek subwatersheds.
We also predict the presence of gaining and losing streams. The streams in this example are predominantly gaining, however, at the junctions of differing stream orders, the reach can change seasonally. Together, these estimates aid in determining the possible sources and sinks to wetlands and how they evolve.
The model did very well in simulating well observations throughout the basin over a range of recharge and groundwater head values. The calibrated Little Juniata/Spruce Creek model shows good performance in simulating both the magnitude and timing of runoff in the basin. Particularly there was good agreement in the recession limb dynamics of lower flows in the hydrograph, and improvement could be made on some of the larger flow events.
Transpiration and interception loss closely resembled the vegetation distribution pattern. Evaporation from ground and overland flow has a spatial pattern that bears a resemblance to topography. At higher elevations, the evaporative losses from land appear to be lowest while the highest values are found at lower elevation. Evaporative flux components along an elevation transect indicate that evapotranspiration has an inverse relationship to average ground water depth. Shallow water table conditions in the topographic valleys result in higher evaporative losses since the capillary fringe supplies water to the unsaturated soil above the water table. The relationship is accentuated in regions of large. Therefore, topography and depth to groundwater add to the complex spatial pattern of evaporative losses which are primarily influenced by the spatial distribution of precipitation, heterogeneity of land cover types and geology.
We ran a limited time period simulation of climate change for the Little Juniata/Spruce Creek basin. In this test we worked with the calibrated hydrologic model for the time period of November 1983 – October 1985. We then applied the monthly normal differences in temperature and precipitation between the 1961-1997 and 2032-2068 periods for the average of all the IPCC fourth assessment GCM outputs (multi-model mean). The monthly differences were applied to the observed, daily time series of temperature and precipitation in the basin to force the hydrologic model. Temperature changes averaged +2.8°C, and ranged from +2.54°C in June to +3.06°C in December. Precipitation change averaged +6.07%, and ranged from -6.19 in October to +14.3% in March.
Reach level results indicate that the geographic position of wetlands relative to the right or left bank of stream channel may be important seasonally as the signatures of the left and right bank can diverge. Baseflow is highly variable among headwater, first order streams of interest. Fluctuations can be either or low or high frequency and low or high magnitude. Exploring this relationship of stream/aquifer dynamics at the reach level appears important in assessing wetland condition/occurrence.
Recharge and streamflow increased, in general, throughout the simulation even though evapotranspiration increased. Streamflow, for example showed an increase of nearly 9% (total flow volume over the simulation period) at the Spruce Creek gauge location. It was very clear that even though we applied a geographically homogeneous climate change to the model, there was a geographically heterogeneous response. This heterogeneous response will be explored in the next round of analysis associated with the full application of the climate change scenario described in the subsequent activity section.
1d. Ecological responses
We are focusing our intensive ecological assessment on a series of “super sites” within one of the physiographic provinces (the Ridge and Valley) in the study area. The sites represent a range of land cover change and reach types: wetland-dominated, mixed wetland and floodplain, and floodplain-dominated reaches. Super sites were chosen within the Upper Juniata River and Shavers Creek watersheds, which were also the focus of hydrologic modeling.
To control for natural variation, we applied the following constraints for site selection:
· Unconstrained reach identified through GIS classification and topographical maps;
· Less than 10% carbonate bedrock within the drainage area;
· Little or no influence from the Allegheny Front;
· Comparable stream size within each reach type with drainage areas < 5 km² for wetland sites, 10 – 40 km² for mixed sites, and > 75 km² for floodplain sites;
· Within or immediately upstream of downstream of an NWI wetland (or found to contain wetland areas during site visits).
Application of these constraints yielded a limited number of possible sites for intensive monitoring. We randomly selected sites from the potential site list, which consisted of reaches within the Shavers Creek and Bald Eagle Creek watersheds. The first potential sites for which owner permission for long-term monitoring was granted were selected until the matrix was filled (Table 2). Regarding land cover change, rather than assume a simple gradient of change in percent forest (which would imply an assumed linear response to this change), we defined each column within the floodplain/mixed/wetland category as possessing different scenarios of land cover, forested buffer, agricultural and urban usage, and apparent degree of wetland/stream/floodplain connectivity.
Table 2. Characteristics of sites selected for ecological assessment.
After establishing a baseline and series of perpendicular transects throughout the reach (crossing the stream, floodplain, and wetland), we mapped all aquatic habitat and randomly selected areas for macroinvertebrate sampling. Sample sizes were based on reach size and habitat complexity and represented at least 50% of the available sampling plots (approximately 10 – 20 samples per site). Prior to sampling, we photographed each plot, visually estimated substrate composition and measured average water depth, as well as the standard water chemistry suite. In addition, we collected average water depths at approximately weekly intervals to estimate inundation patterns (i.e., ephemeral, seasonal, or permanent) and establish relationships (if any) between a particular habitat type, precipitation data, and the stream hydrographs (from in-stream well recordings).
Super sites have been or will be instrumented with the following automatic water-level monitoring wells: in-stream wells to serve as stream gages; in-wetland wells to calibrate the hydrologic models and inform the ecological sampling by determining amount and direction of flow (e.g., from one hillside into the wetland, from the stream into the wetland, etc.). One to two in-stream wells were installed at all sites, and four to six in-wetland wells will be installed in the wetland and mixed sites.
Future Activities:
Planned activity for Year 3:Climate Scenarios: Final section of climate scenarios for use in hydrologic modeling will be made. In addition to the multi-model average scenario, which agreed best with observations, results of one or two specific GCMs will be selected as an alternate scenario.
Land Cover Scenarios: A model will be estimated to allocate land use change within an MCD. First, land use change between 1992 and 2001 will be measured for each 100m2 grid cell within each MCD. Then a model will be estimated that explains how much open space was lost in each cell, based on characteristics of the cell and on the total land use change in the MCD. Next, using the grid cell level land use change model, the MCD-level land use change projections will be allocated among the grid cells in each MCD. This will allow construction of maps showing the location and quantity of land use change within each MCD at decadal intervals.
Hydrologic Modeling: Model set-up will be completed for the remaining four study basins and then on the alternate basins as time permits. We will continue to examine domain decomposition configurations to determine what technique is best in satisfying the needs of lower resolution estimates over the larger HUC 11 domain and higher resolution in the smaller wetland domains within the HUC 11 basins. After the study basin set-up and domain decomposition is complete, calibration of the model in the remaining basins will begin, to be followed by applying the climate and land cover scenarios developed by these working groups to yield a series of predictive hydrologic scenarios.
Ecological Analysis:
Planned activities for the remainder of 2009 include final well installations, completion of the rapid assessment protocol, collection of elevation measurements along one or more transects, collection of plant community data, additional macroinvertebrate collections for the fall season and processing of macroinvertebrate sample data. Once data collection is complete, these data will be used to characterize the relationships between hydrologic and landcover parameters and ecosystem characteristics and services in wetlands of various types,
Integrative Analyses:
The predictive hydrologic scenarios will be used to forecast changes in wetland ecosystems and their services in the selected study basins. A probability surface will be developed to identify where non-linearities and/or thresholds in wetland response occur.
Journal Articles:
No journal articles submitted with this report: View all 5 publications for this projectSupplemental Keywords:
RFA, Scientific Discipline, Air, Ecosystem Protection/Environmental Exposure & Risk, Hydrology, climate change, Air Pollution Effects, Monitoring/Modeling, Regional/Scaling, Environmental Monitoring, Atmospheric Sciences, Ecological Risk Assessment, Atmosphere, coastal ecosystem, aquatic species vulnerability, biodiversity, environmental measurement, ecosystem assessment, meteorology, global change, climate, anthropogenic, climate models, UV radiation, greenhouse gases, environmental stress, coastal ecosystems, water quality, ecological models, climate model, Global Climate Change, land use, regional anthropogenic stresses, atmospheric chemistry, stressor response modelRelevant Websites:
www.wetlands.psu.eduProgress 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.