H33C-0473 INVITED 1340h
From process heterogeneity at the hillslope scale to model predictability at the catchment scale
Process representation in catchment model structure continues to be a vexing issue in hydrology and is core issue in the PUB movement. Complex hydrological descriptions at the hillslope scale have been difficult to incorporate within a catchment modeling framework due to the disparity between the scale of measurements and the scale of model sub-units. As a result, parameters represented in many conceptual models are often not physically-based or related to physical properties, and therefore cannot be established prior to a model calibration. This talk examines some new strategies for moving from calibration reliant model schemes to new approaches based on better field-based process understanding of water flowpath, source and age. The talk is given from the perspective of an experimentalist, trying to whittle down complexity at the hillslope scale to achieve model simplicity at the catchment scale.
http://www.cof.orst.edu/cof/fe/watershd/
H33C-0474 1340h
Spatially Explicit Observations Elucidate Simple Scalars of Forest Canopy Transpiration Along Moisture Gradients in Semi-Arid and Humid Climates
The ability to scale from point measurements to watersheds is critical for making predictions in hydrology. Assumptions are often made that averaging point measurements and scaling them up using a cookie-cutter or paint-by-numbers approach will capture relevant spatial gradients. Two issues are whether such assumptions are valid and how these assumptions hold across environmental conditions. We made spatially explicit measurements and modeling at two field sites representing semi-arid and humid climates. For our semi-arid site we identified a topography-soil moisture gradient in an alpine watershed near Laramie, Wyoming. For our humid climate we chose a wetland-to-upland gradient in the Chequamegon National Forest near Park Falls, Wisconsin. At both sites we used cyclic sampling designs to efficiently quantify spatial trends using geostatistics. Spatial data was collected for sap flux using Granier type sensors. In addition, we collected spatial soil moisture, vapor pressure deficit, and leaf area index with the same level of spatial detail, at both study sites. At the Wyoming site our dominant species was Lodgepole pine, as it spans most of the topography-soil gradients from the edges of riparian areas to upslope positions. At the Wisconsin site we selected aspen as our focus species, as it is a dominant species in terms of transpiration in the region and it grows over a wide variation in topographic positions from wetland to upland. We found that the semivariagrams of soil moisture at both sites showed ranges of about 110 meters on low soil moisture days and 80 meters on high soil moisture days. Neither site showed differences in sap flux per unit xylem along soil moisture gradients. However, once we scaled the sap flux measurements to the whole tree using basal area, we found that the semi-arid site had much higher fluxes near the stream compared to upslope, but at the humid site the uplands had much higher scaled sap fluxes than the wetlands. It appears that both sites allow simple scalars to spatially predict transpiration, but we hypothesize that the scaling behavior differs because the semi-arid site is water-limited because of too little water, and the humid site is water-limited because of too much water. To test this hypothesis we used a forest canopy model to interpret the opposing responses, and a stochastic parameter restriction and selection scheme to assess predictive uncertainty associated with the model-supported interpretations.
H33C-0475 1340h
Geology Broadly Predicts Summer Streamflow in Volcanic Terrains: Lessons From the Oregon Cascades
The western slope of the Cascade volcanic arc is comprised of two distinct geologic provinces, both with similar climate and vegetation, making it an excellent place for examining geologic controls on hydrologic variability. Analysis of streamflow from USGS gauges has shown that summer streamflow characteristics are related to the percent of basin area underlain by High Cascade (0-7 Ma) basalt flows and related rocks. Watersheds with High Cascade geology exhibit higher unit streamflows and slower summer recessions than those with Western Cascade (7-40 Ma) geology. Since July 2003, gauging of thirteen 1st to 4th order High Cascade streams in the McKenzie River basin has shown that some High Cascade streams are runoff-dominated, like Western Cascade streams, while others are fed by large volume cold springs, with relatively steady flow. During the summer, these spring-fed streams provide over 80% of the flow to the McKenzie River. In winter months, runoff-dominated streams respond rapidly to rain and rain-on-snow events and become the major water source to the McKenzie River. Spring-fed streams also respond to precipitation events, but show muted and delayed hydrograph peaks. Summer flow behavior and response to individual events varies between springs, even between those that are located less than 1 km from each other. Oxygen isotope analysis suggests that closely spaced springs may have recharge areas differing by over 150 m in average altitude. These springs emanate from lava flow toes or contacts; thus, paleotopography, including buried channel networks, and lava flow characteristics, such as primary and secondary porosity, are also likely to be important determinants of event and seasonal streamflow response. Therefore, geologic differences are useful for predicting streamflow in large basins and over seasonal to interannual timescales, but at the headwater catchment spatial scale, or event time scale, groundwater-fed streams exhibit variability that cannot be predicted by regional geology alone. Prediction at these scales will require more detailed knowledge of local geologic and meteorologic variability.
H33C-0476 1340h
Multiscale Trend Analysis of River Basin Dynamics
Storm runoff hydrographs are the signature of a basin's dynamics to a given precipitation forcing. As such, their study across basins and over a range of scales offers the opportunity to detect and quantify nonlinearities and scaling laws in river basin dynamics. Manual extraction of hydrograph characteristics (e.g., time to peak, relaxation time and peak magnitude) however, is tedious and has prevented extensive regional analyses of such observations for the purpose of scaling and regionalization. In this paper, (1) we propose a new methodology, called multiscale trend analysis (MTA) for the automatic and reliable extraction of hydrograph characteristics from hourly or finer scale streamflow series, and (2) we report the results of a regional multiscaling analysis of hydrologic response characteristics from 31 stations for drainage areas ranging from 10 to $10^4$ km$^2$ over the Kansas/Oklahoma region. Our results suggest the presence of statistical multiscaling in hydrologic response characteristics and a change of scaling regimes at a scale of approximately 700 km$^2$. We relate this scale to the scale at which channel morphometry properties, fluvial regimes, and statistical properties of floods also change and highlight the interconnection of physical processes and statistical laws in river basin dynamics.
H33C-0477 1340h
Improving Temporal and Spatial Variability of Hydrological and Energy Parameters in Global Analysis through Precipitation Assimilation
Understanding climate variability over a wide range of space-time scales requires a comprehensive description of the earth system. Global analyses produced by a fixed assimilation system (i.e., re-analyses) - as their quality continues to improve - have the potential of providing a vital tool for meeting this challenge. At the present time, the usefulness of re-analyses is limited by uncertainties in such basic fields as clouds, precipitation, and evaporation - especially in the tropics, where observations are relatively sparse. Yet, for many hydrological and climate applications it is essential that analyses can accurately reproduce the observed rainfall intensity and variability. Analyses of the tropics have long been shown to be sensitive to the treatment of cloud/precipitation processes, which remains a major source of uncertainty in current generation of atmospheric models. NASA Goddard Space Flight Center has been exploring the use of satellite-based microwave rainfall measurements in improving global analyses and has recently produced a 1o x 1o "TRMM re-analysis" that assimilates 6-hourly TMI and SSM/I surface rain rates over tropical oceans using the GEOS-3 global data assimilation system. The goal is to provide a multi-year global analysis that is dynamically consistent with available tropical precipitation observations. A distinct feature of the GEOS-3/TRMM re-analysis is that its precipitation analysis is not derived from a short-term forecast (as for most operational systems) but given by a time-continuous model integration directly constrained by precipitation observations within a 6-h analysis window, while the wind, temperature, and pressure fields adjust to the improved precipitation and associated latent heating structures within the same analysis window. In this talk, we show that precipitation assimilation leads to significant improvements in the intensity and variability of hydrological and climate parameters in the GEOS-3/TRMM re-analysis and compare results against other operational and reanalysis products.
H33C-0478 1340h
Assimilating Terrestrial Hydrologic Fluxes Into Land Surface Models Using Remote Sensing Data Products
The state of the land surface plays a critical role in the land-atmosphere interactions, through the dynamic evolution of moisture and energy fluxes at the land surface. The current generation of Land Surface Models (LSMs) estimating these fluxes, however sophisticatedly parameterized they might be, are still constrained by the underlying approximate model physics. With the advent of a variety of land surface remote sensing (LSRS) data products, better estimations of the dynamic state can be obtained by integrating (assimilating) these LSRS products into the predictive models. Numerous techniques at various levels of sophistication have been developed for assimilating remotely sensed near surface soil moisture into LSMs, based on the knowledge about the errors in the model predictions and observations, to update the soil moisture profile and associated fluxes. Though a lot of research is being done to obtain remotely sensed near surface soil moisture as a reliable data product, the current coverage and spatial and temporal scales of the same have limited applications in land surface modeling. However, several other LSRS products are available as reliable global coverage data at desired spatial and temporal scales, which can be incorporated either directly or indirectly into LSMs to obtain reliable estimates of moisture and energy fluxes. In the present work, we use LSRS products which have been derived from radiation measurements by MODIS (MODerate-resolution Imaging Spectrometer) instrument flying on Terra and Aqua satellite platforms. The LSRS products are: Land Surface Temperature (LST), Surface Albedo, Vegetation Indices (NDVI/EVI), Fractional Vegetation Cover and Leaf Area Index (LAI). We use these data products and other boundary layer variables, as surrogate data to provide an index to the energy fluxes, using the Surface Energy Balance System (SEBS) framework, which is then used to update the soil moisture profile and associated fluxes at relevant scales. The SEBS framework [Su et al., 2001; Li et al., 2003] will be used in conjunction with LSM, so as to take advantage of the more accurate parameterization from SEBS and the underlying physics representation in LSM, thus developing a blended system.
H33C-0479 1340h
Rainfall Variability Studies Based on a Long Term Radar-Rainfall Data Set
A radar rainfall climatology has been developed based on approximately 10 years of NEXRAD level III data for several radars in the Carolina region of the southeast U.S. We are using this radar rainfall climatology to investigate the variability and trends of precipitation patterns based on the high resoution radar rainfall data. The data set is provided at hourly 4x4 km2 resolution and will be integrated to daily, monthly, seasonal, and annual temporal scales. Several issues have arisen during the development of the radar rainfall climatology related to inherent biases in the radar rainfall estimates, and several steps have been taken to reduce or eliminate these biases. Image processing techniques have been applied to try and elimate bright band and anomalous propagation. The radar range extent has been reduced to eliminate the effects of range on the rainfall estimates, and radar calibration bias adjustment has been applied to smooth radar to radar bias effects. These improved rainfall estimates will be used in studies that look at the variability of rainfall in space-time domains and they will be used to determine trends of rainfall from seasonal and interannual. Further studies will be carried out to extract extreme events through data mining techniques. Extreme events are important when investigating the variability of rainfall as they help to more accurately define the overall distribution of rainfall events, and this has implications in climate prediction and weather and flood forecasting for example.
H33C-0480 1340h
Distributed snow modeling suitable for use with operational data for the American River watershed.
The mountainous terrain of the American River watershed (~4300 km2) at the Western slope of the Northern Sierra Nevada is subject to significant variability in the atmospheric forcing that controls the snow accumulation and ablations processes (i.e., precipitation, surface temperature, and radiation). For a hydrologic model that attempts to predict both short- and long-term streamflow discharges, a plausible description of the seasonal and intermittent winter snow pack accumulation and ablation is crucial. At present the NWS-CNRFC operational snow model is implemented in a semi distributed manner (modeling unit of about 100-1000 km2) and therefore lump distinct spatial variability of snow processes. In this study we attempt to account for the precipitation, temperature, and radiation spatial variability by constructing a distributed snow accumulation and melting model suitable for use with commonly available sparse data. An adaptation of the NWS-Snow17 energy and mass balance that is used operationally at the NWS River Forecast Centers is implemented at 1 km2 grid cells with distributed input and model parameters. The input to the model (i.e., precipitation and surface temperature) is interpolated from observed point data. The surface temperature was interpolated over the basin based on adiabatic lapse rates using topographic information whereas the precipitation was interpolated based on maps of climatic mean annual rainfall distribution acquired from PRISM. The model parameters that control the melting rate due to radiation were interpolated based on aspect. The study was conducted for the entire American basin for the snow seasons of 1999-2000. Validation of the Snow Water Equivalent (SWE) prediction is done by comparing to observation from 12 snow Sensors. The Snow Cover Area (SCA) prediction was evaluated by comparing to remotely sensed 500m daily snow cover derived from MODIS. The results that the distribution of snow over the area is well captured and the quantity compared to the snow gauges are well estimated in the high elevation.
H33C-0481 1340h
Modeling Infiltration and Runoff During Soil Seal Formation: The Effect of Areal Heterogeneity of Soil Hydraulic Properties
The expected trends in climate change may affect rainfall-runoff relationships. In the case of small, bare watersheds, rainfall-infiltration-runoff relationships are determined by (i) the formation of a sealing layer at the soil surface due to the raindrops impact, (ii) the spatial variability of the soil hydraulic properties within the watershed. The combined effects of soil surface sealing and areal heterogeneity of the soil hydraulic properties on the hydrological response of small bare catchment are simulated. Seal formation during rainfall was simulated according to the dynamic model of Assouline and Mualem (1997). Areal heterogeneity of the soil was represented by a lognormal distribution of the saturated hydraulic conductivity of the initially undisturbed soil, Ks, and by related distributions of the other soil parameters. The cell-model of Diskin et al. (1984) was applied to compute the runoff hydrograph at the outlet of a hypothetical bare catchment of 0.5 km2. Homogeneous and heterogeneous soils, with two different surface conditions, unsealed (mulched) and dynamic sealing, were considered. In terms of infiltration, accounting for areal soil heterogeneity reduces the ponding time and the rate of infiltration decrease, while it increases the steady infiltration rate after a long exposure to rainfall. When soil surface sealing is considered, the ponding time is only slightly affected but significantly more runoff is produced. However, the effects of soil variability on the infiltration curve are reduced, compared to the unsealed field. Compared to the homogeneous unsealed catchment case, runoff during soil surface sealing was increased by a factor of 10, and by a factor of 20 when the soil surface was already sealed before rainfall begun. Accounting for catchment heterogeneity increased runoff. Soil surface sealing reduced the effect of catchment heterogeneity on runoff hydrograph. This effect was also reduced as rainfall intensity increased.
H33C-0482 1340h
Simulation of Crust Effects on Infiltration and on Runoff Generation at Arid Micro-catchments
Generation of runoff in arid areas was studied by analysis and modeling of rainfall - runoff relations over a cluster of six micro-catchments whose total area is about 0.25 km2. The state of the crust, formed by raindrop impact on the soil surface, was found to substantially affect infiltration and runoff. These effects pertain, almost entirely, to overland processes, because channel flow is minimal in the micro-catchments. The experimental catchments lie at Avdat, Israel, where mean depth of precipitation is about 85 mm/y and mean depth of potential evapotraspiration exceeds 2500 mm/y. Rainfall was recorded by two gauges, situated at the margins of the catchments' cluster, and was considered uniformly distributed over their area. Runoff was separately recorded for each catchment. Infiltration rates, for the different geo-morphological units prevailing in the study area, were measured for incrusted conditions. Runoff generation and magnitude were simulated by use of the physical, event-based, distributed, model called KINEROS. Data analysis applied the index and the UH models. An order of magnitude difference was found between measured final infiltration rates for incrusted conditions and the values of hydraulic conductivity obtained in the calibration of the KINEROS model for the actual crusted conditions. The calibration required values of capillary drive substantially smaller than those commonly used to the actual soil types. In addition, the model was found very sensitive to antecedent soil moisture. Fit of index abstraction rates corroborated to the final infiltration rates obtained by the model. The results indicate that the selected model can be used for predicting runoff over and infiltration through crusted surfaces in arid watersheds with assumed soil characteristics for the crust only, and without taking into consideration the underlying conditions. The linearity of unit hydrographs, found for some cases at the micro-catchments, indicates that the thin flow over the crusted soil is either laminar or reached terminal velocity.
H33C-0483 1340h
Incorporating Sub-grid Spatial Variabilities Into Subsurface Flow Parameterizations For Land Surface Modeling In Ungauged Basins
An adequate representation of subsurface flow processes is very important for land surface modeling at the large scales. At the regional scale, for example, the amount of annual averaged subsurface flow is an important component of the water budget used to manage water resources. However, the understanding of the impacts of subgrid spatial variability on subsurface flow processes at the large spatial scales is very limited. Most land surface models use simple parameterizations to estimate the subsurface flow where groundwater discharge is related to a lumped soil moisture state (i.e., the unsaturated zone and saturated zone are lumped), while some models incorporate the effects of landscape and lateral flow partially into the subsurface flow calculation using the framework of TOPMODEL. In this study, we propose a new wetness index for developing a new subsurface flow parameterization. The new wetness index incorporates the concept of dynamic contributing area, spatial variability of the topography, and spatial variability of recharge. Numerical experiments will be carried out to investigate the impacts of different types of spatial variabilities on subsurface flow simulations at two study sites: (1) the Tarrawara catchment in the southeastern Australia, and (2) the Usadievskiy catchment at Valdai, Russia, both of which have multiple measurements on water fluxes. To enable the proposed parameterization to be applied to ungauged basins, efforts are made to keep the number of parameters in the new subsurface flow parameterization as few as possible.
H33C-0484 1340h
On the effectiveness of assimilation of coarse scale near-surface soil moisture imagery and fine scale temporal ratios of soil moisture into land surface models
Remote sensing retrievals of near-surface soil moisture constitute a major asset for constraining and improving the predictions of the water and energy fluxes at the earth surface in ungauged basins. The expected near-future resolution of near-surface soil moisture imagery derived from passive remote sensing sources is approximately 30 km. While active remote sensing retrievals of near-surface soil moisture may be derived at finer resolutions (~100 m), these have been shown to be prone to error and large uncertainties in the presence of vegetation. However, recent evidence suggests that the temporal ratios of near-surface soil moisture retrievals from active remote sensing sources may retain much of the soil moisture signal with decreased uncertainty. In this study, we first evaluate the impacts of the resolution of passive remotely sensed soil moisture imagery (from 800-m to 25-km) on the effectiveness of assimilation of these fields with regards to improving the prediction of soil moisture states and energy fluxes from the Three Layer Variable Infiltration Capacity (VIC-3L) land surface model. We additionally assess the value of fine-scale remotely sensed temporal ratios of near-surface soil moisture when assimilated in conjunction with coarse scale near-surface soil moisture imagery. We conduct our investigation with the near-surface soil moisture retrievals derived during the Southern Great Plains Hydrology experiment of 1997 (SGP97). Assimilation is performed with an extension of multiscale Kalman filtering (MKF). MKF is ideally suited for this study as it permits for the remotely sensed near-surface soil moisture retrievals and land surface model predictions to be available at distinct spatial scales and allows for optimal characterizations of time varying uncertainties in the land surface model predictions and the observations, respectively.
H33C-0485 1340h
Runoff Variability in Field-scale Catchments and the Implications for Rainffall-Runoff Modeling
In this study long-term rainfall runoff records for two agricultural catchments (ca. 0.5 ha) in the USDA - Agricultural Research Service North Appalachian Experimental Watershed (Coshocton, OH) network were used to address the inter-event and inter-catchment variability of field-scale runoff processes. Through analyses of flood frequency and flow duration, the adjacent fallowed watersheds (WS106 and WS121) were found to be similar in terms of annual flood peaks, but less so in terms of the distribution of their discharge rates. Further investigation was focused on event-scale variations of runoff response and whether these variations can be effectively captured by rainfall-runoff models, which included: a) TR-20 (a lumped model); b) EPA-SWMM (a semi-distributed model); and c) GSSHA (a grid-based, fully distributed model). Each model was used to simulate 41 selected runoff episodes recorded in each of the two catchments, and subsequently calibrated to yield parameter values that maximize the correlation between the simulated and observed runoff peaks. Our results indicate that, despite calibration, the hydrographs derived from all models deviated considerably from actual observations, and on the basis of inter-event fluctuations, which furthermore lacked a conspicuous dependence on the magnitude of runoff peaks. Our findings suggest that, in the absence of information on rainfall distribution and soil moisture, distributed models may not be superior to lumped ones in forecasting runoff responses of field scale catchments; and the correspondence between runoff mechanisms and model representations needs to be better understood and accounted for in order to limit the uncertainties of model predictions.
H33C-0486 1340h
Complexities Involved in One-Dimensional Infiltration Models for Predicting the Soil Moisture Content for a Hillslope
Most operational hydrologic models use one-dimensional representations of infiltration and soil moisture redistribution. However several field studies have shown lateral flow under unsaturated and near saturation conditions to make a significant contribution to hydrologic response for hillslopes. There is not consensus on a robust technique to predict the occurrence of neither bypass flow nor the volumetric division between matrix flow and bypass flow from hillslopes. This study investigates whether the popular one-dimensional infiltration models based on Richard's equation adequately represent the spatial and temporal patterns of soil moisture content on a hillslope. Two spring rain events of different intensities were chosen to compare the translation of precipitation to changes in soil moisture and streamflow. The soil moisture predicted by numerical approximation of the Richard's equation using Hydrus 1D and by manual computation of the Green-Ampt equations was compared to soil moisture measured by TDR for several locations on north and south facing hill slopes in the Dry Creek watershed, near Boise ID. The streamflow records for the two events differed in lag of time to peak and that the low-constant flux event had one peak and the high-variable flux event had two. Both models replicated the pattern of the measured soil moisture values for the low-steady flux rainfall event, and for the first peak of the high-variable flux event, although the moisture content was overestimated. For the second hydrograph peak of the high-variable flux event the models did not capture the second peak in the measured soil moisture at shallow depths, indicating that shallow lateral flow occurred in response to the higher input flux during that event.
H33C-0487 1340h
Temporal and spatial controls on variability in water repellency for prediction of runoff from burned watersheds
Prediction of runoff in catchments affected by wildfire is an extreme example of the ungauged basin problem. After wildfire, there is an increase in the risk of severe erosion, floods, and debris flows, often resulting from water repellency. The transient nature of wildfire effects on soils makes the gauging information from affected systems anecdotal in nature; so modeling is required to assimilate the scattered observations and provide for prediction of runoff from newly burnt watersheds. While most water repellency research has focused on the strength of water repellency and its duration for time scales of seconds to seasons measured at the point scale, there is a lack of information on spatial variability of water repellency and the temporal variation of the spatial properties at multi-annual time scales. The lack of spatial information on water repellency both results from and perpetuates a lack of modeling using information on water repellency to predict runoff from burned watersheds. The recently developed FERGI model uses information on the fractional water repellent area to predict runoff from burned hillslopes and catchments. Modeling a catchment requires information to describe the spatial and temporal variations in the fractional water repellent area. Spatial controls on initial water repellency include vegetation density, fire severity, and soil texture, which can be mapped with remote sensing. At finer scales, there is high connectivity of water repellent areas on soils with greater than 70% water repellency; so fractional repellency is adequate information for prediction of runoff for the great majority of storms. In addition, observations on the temporal decay of water repellency show a spatial organization to the water repellency caused by erosion where repellent areas contribute water to non-repellent areas, further supporting use of a fractional repellent area. Measurements of burned areas on different parent materials show that the fractional area of water repellency declines to levels less than 70% within 3-years of the fire. The decrease in fractional area is linear in time, falsifying the hypothesis that decay in water repellency is caused solely by random innovations, which would predict an exponential decay.
http://www.fs.fed.us/rm/boise/
H33C-0488 1340h
A Statistical Approach to Estimating the Contribution of Glaciers to Future Sea-level Rise.
Valley glaciers and small ice caps are expected to supply the bulk of the cryosphere's contribution to anthropogenic sea-level rise over the coming century (~ 0.23 m (IPCC, 2001)). The estimation of this contribution is hampered by the lack of quantitative data for the vast majority of glaciers worldwide (only 100 glaciers out of over160,000 present have mass-balance records for longer than 5 years). The issues surrounding the parameterisation of subgrid scale processes, uncertainties in parameter values and the propagation of errors in model prediction of sea-level rise are similar to those experienced in the prediction of discharge from ungauged river basins. Given the similarities between valley glacier systems and their hydrological counterparts, it maybe appropriate to use the techniques developed for hydrological modelling. Therefore, to calculate sea-level rise with an associated error we propose the following four stage procedure. First, a generic valley-glacier system model that allows for variations in width, depth, accumulation and ablation along the glacier is developed. Second, the model is calibrated against the small number of glaciers on which we have sufficient data. Third, a response surface of sea-level contribution as a function of glacier climatology and topography is constructed. Entailed in this stage is a rigorous assessment of the uncertainty propagated through the model due to uncertainties inherent in the input parameters. The final fourth stage is then to sample the response function, in accordance with estimates of the global distribution of glaciers in the climate-topography phase space, in order to estimate sea-level rise with a meaningful estimation of error. The use of methods traditionally employed by the hydrological community to a different study area highlights issues that contribute to an understanding of the limitations in defining variability within a system and quantifying uncertainty.