Using Regression Tree Analysis to Improve Predictions of Low-Flow Nitrate and Chloride in Willamette River Basin Watersheds

The use of regression tree analysis is examined as a tool to evaluate hydrologic and land use factors that affect nitrate and chloride stream concentrations during low-flow conditions. Although this data mining technique has been used to assess a range of ecological parameters, it has not previously been used for stream water quality analysis. Regression tree analysis was conducted on nitrate and chloride data from 71 watersheds in the Willamette River Basin to determine whether this method provides a greater predictive ability compared to standard multiple linear regression, and to elucidate the potential roles of controlling mechanisms. Metrics used in the models included a variety of watershed-scale landscape indices and land use classifications. Regression tree analysis significantly enhanced model accuracy over multiple linear regression, increasing nitrate R ² values from 0.38 to 0.75 and chloride R ² values from 0.64 to 0.85 and as indicated by the ΔAIC value. These improvements are primarily attributed to the ability for regression trees to more effectively handle interactions and manage non-linear functions associated with watershed heterogeneity within the basin. Whereas hydrologic factors governed the conservative chloride tracer in the model, land use dominated control of nitrate concentrations. Watersheds containing higher agricultural activity did not necessarily yield high nitrate concentrations, but agricultural areas combined with either small proportions of forested land or greater urbanization generated nitrate levels far exceeding water quality standards. Although further refinements are recommended, we conclude that regression tree analysis presents water resource managers a promising tool that improves on the predictive ability of standard statistical methods, provides insight into controlling mechanisms, and helps identify catchment characteristics associated with water quality impairment.     

DEVELOPMENT OF WATERSIIED MODELS FOR TWO SIERRA NEVADA BASINS USING A GEOGRAPHIC INFORMATION SYSTEM1

ABSTRACT: Techniques were developed using vector and raster data in a geographic information system (GIS) to define the spatial variability of watershed characteristics in the north-central Sierra Nevada of California and Nevada and to assist in computing model input parameters. The U.S. Geological Survey's Precipitation-Runoff Modeling System, a physically based, distributed-parameter watershed model, simulates runoff for a basin by partitioning a watershed into areas that each have a homogeneous hydrologic response to precipitation or snowmelt. These land units, known as hydrologic-response units (HRU's), are characterized according to physical properties, such as altitude, slope, aspect, land cover, soils, and geology, and climate patterns. Digital data were used to develop a GIS data base and HRIJ classification for the American River and Carson River basins. The following criteria are used in delineating HRU's: (1) Data layers are hydrologically significant and have a resolution appropriate to the watershed's natural spatial variability, (2) the technique for delineating HRU's accommodates different classification criteria and is reproducible, and (3) HRU's are not limited by hydrographic-subbasin boundaries. HRU's so defined are spatially noncontiguous. The result is an objective, efficient methodology for characterizing a watershed and for delineating HRU's. Also, digital data can be analyzed and transformed to assist in defining parameters and in calibrating the model.     

ESTIMATING ATRAZINE LEACHING IN THE MIDWEST1

Data from seven Management Systems Evaluation Areas (MSEA) were used to test the sensitivity of a leaching model, Pesticide Root Zone Model-2, to a variety of hydrologic settings in the Midwest. Atrazine leaching was simulated because it was prevalent in the MSEA studies and is frequently detected in the region's groundwater. Short-term simulations used site specific soil and chemical parameters. Generalized simulations used data avail. able from regional soil databases and standardized variables. Accurate short-term simulations were precluded by lack of antecedent atrazine concentrations in the soil profile and water, suggesting that simulations using data for less than five years underestimate atrazine leaching. The seven sites were ranked in order of atrazine detection frequency (concentration > 0.2 μg L^-1) in soil water at 2 m depth in simulations. The rank order of the sites based on long-term simulations were similar to the ranks of sites based on atrazine detection frequency from groundwater monitoring. Simulations with Map Unit Use File (MUUF) soils data were more highly correlated with ranks of observed atrazine detection frequencies than were short-term simulations using site-specific soil data. Simulations using the MIJUIF data for soil parameters were sufficiently similarity to observed atrazine detection to allow the credible use of regional soils data for simulating leaching with PRZM-2 in a variety of Midwest soil and hydrologic conditions. This is encouraging for regional modeling efforts because soil parameters are among the most critical for operating PRZM-2 and many other leaching models.     

Implications of Conceptual Channel Representation on SWAT Streamflow and Sediment Modeling

Hydrologic modeling outputs are influenced by how a watershed system is represented. Channel routing is a typical example of the mathematical conceptualization of watershed landscape and processes in hydrologic modeling. We investigated the sensitivity of accuracy, equifinality, and uncertainty of Soil and Water Assessment Tool (SWAT) modeling to channel dimensions to demonstrate how a conceptual representation of a watershed system affects streamflow and sediment modeling. Results showed the amount of uncertainty and equifinality strongly responded to channel dimensions. On the other hand, the model performance did not significantly vary with the changes in the channel representation due to the degree of freedom allowed by the conceptual nature of hydrologic modeling in the parameter calibration. Such findings demonstrated good modeling performance statistics do not necessarily mean small output uncertainty, and partial improvements in the watershed representation may neither increase modeling accuracy nor reduce uncertainty. We also showed the equifinality and uncertainty of hydrologic modeling are case‐dependent rather than specific to models or regions, suggesting great caution should be used when attempting to transfer uncertainty analysis results to other modeling studies, especially for ungauged watersheds. Editor's note: This paper is part of the featured series on SWAT Applications for Emerging Hydrologic and Water Quality Challenges. See the February 2017 issue for the introduction and background to the series.     

Streamflow Hydrology Estimate Using Machine Learning (SHEM)

Continuity and accuracy of near real‐time streamflow gauge (streamgage) data are critical for flood forecasting, assessing imminent risk, and implementing flood mitigation activities. Without these data, decision makers and first responders are limited in their ability to effectively allocate resources, implement evacuations to save lives, and reduce property losses. The Streamflow Hydrology Estimate using Machine Learning (SHEM) is a new predictive model for providing accurate and timely proxy streamflow data for inoperative streamgages. SHEM relies on machine learning (“training”) to process and interpret large volumes (“big data”) of historic complex hydrologic information. Continually updated with real‐time streamflow data, the model constructs a virtual dataset index of correlations and groups (clusters) of relationship correlations between selected streamgages in a watershed and under differing flow conditions. Using these datasets, SHEM interpolates estimated discharge and time data for any indexed streamgage that stops transmitting data. These estimates are continuously tested, scored, and revised using multiple regression analysis processes and methodologies. The SHEM model was tested in Idaho and Washington in four diverse watersheds, and the model's estimates were then compared to the actual recorded data for the same time period. Results from all watersheds revealed a high correlation, validating both the degree of accuracy and reliability of the model.     

Determining regional water quality patterns and their ecological relationships

A multivariate statistical method for analyzing spatial patterns of water quality in Georgia and Kansas was tested using data in the US Environmental Protection Agency's STORET data system. Water quality data for Georgia and Kansas were organized by watersheds. We evaluated three questions: (a) can distinctive regional water quality patterns be detected and predicted using only a few water quality variables, (b) are regional water quality patterns correlated with terrestrial biotic regions, and (c) are regional water quality patterns correlated with fish distributions? Using existing data, this method can distinguish regions with water quality very different from the average conditions (as in Georgia), but it does not discriminate well between regions that do not have diverse water quality conditions (as in Kansas). Data that are spatially and temporally adequate for representing large regions and for multivariate statistical analysis are available for only a few common water quality parameters. Regional climate, lithology, and biotic regimes all have the potential to affect water quality, and terrestrial biotic regions and fish distributions do compare with regional water quality patterns, especially in a state like Georgia, where watershed characteristics are diverse. Thus, identifiable relationships between watershed characteristics and water quality should allow the development of an integrated landaquatic classification system that would be a valuable tool for resource management. Because geographical distributions of species may be limited by Zoogeographic and environmental factors, the recognition of patterns in fish distributions that correlate with regional water quality patterns could influence management strategies and aid regional assessments.     

DEVELOPMENT OF EARTH SATELLITE TECHNOLOGY FOR THE TELEMETRY OF HYDROLOGIC DATA1

ABSTRACT The use of satellite telemetry is playing a major role in the collection of hydrologic data. Advancing technology and availability of government satellites have permitted many agencies to take advantage of new procedures for acquiring data from automated remote data collection stations. Experiments with Earth satellite technology started in the 1960's and 1970's, with the polar-orbiting National Aeronautics and Space Administration Nimbus and Landsat satellites. Subsequent advancements took place through the development phase to operational systems using the Geostationary Operational Environmental Satellite (GOES) of the National Oceanic and Atmospheric Administration. This satellite system supports more than 2,500 active telemetry sites, of which approximately 1,200 are Geological Survey stream-gaging stations for the collection of hydrologic data. A satellite data collection system is made up of three primary components; a small battery-operated radio, and Earth-orbiting satellite, and an Earth receive and data processing station. The data relay satellites' vast aerial view of the Earth's surface gives satellite telemetry a large advantage over ground-based systems for the collection of real-time hydrologic data for flood warning, reservoir management, irrigation water control, hydropower generation, and the operation of hydrologic stations.     

METHODOLOGIES FOR CALIBRATION AND PREDICTIVE ANALYSIS OF A WATERSHED MODEL1

ABSTRACT: The use of a fitted parameter watershed model to address water quantity and quality management issues requires that it be calibrated under a wide range of hydrologic conditions. However, rarely does model calibration result in a unique parameter set. Parameter nonuniqueness can lead to predictive nonuniqueness. The extent of model predictive uncertainty should be investigated if management decisions are to be based on model projections. Using models built for four neighboring watersheds in the Neuse River Basin of North Carolina, the application of the automated parameter optimization software PEST in conjunction with the Hydrologic Simulation Program Fortran (HSPF) is demonstrated. Parameter nonuniqueness is illustrated, and a method is presented for calculating many different sets of parameters, all of which acceptably calibrate a watershed model. A regularization methodology is discussed in which models for similar watersheds can be calibrated simultaneously. Using this method, parameter differences between watershed models can be minimized while maintaining fit between model outputs and field observations. In recognition of the fact that parameter nonuniqueness and predictive uncertainty are inherent to the modeling process, PEST's nonlinear predictive analysis functionality is then used to explore the extent of model predictive uncertainty.     

NWSRFS CALIBRATION PARAMETER SELECTION AND GEOLOGIC REASONING: PACIFIC NORTHWEST CASES1

ABSTRACT: The National Weather Service River Forecast System (NWSRFS) is the new hydrologic prediction model for the National Weather Service (NWS) and provides guidance to meteorologists who issue NWS Flood Warnings to the public. The primary submodel within NWSRFS is the Sacramento Soil Moisture Accounting (SAC-SMA) model, which predicts surface runoff as a function of meteorological, geological, and soil data calibrated over a watershed. The research presented here focuses on a different approach to NWSRFS calibrations: greater utilization of geologic and soil data, in order to give the model better predictive capability. Geologic understanding can create better insights for the initial estimation and subsequent adjustment of SAC-SMA parameters. Fifteen calibrated Pacific Northwest drainages reveal a variety of hydrogeologic responses. For example, results for the Mount Rainier drainages show the complex interaction between active glaciers, impermeable volcanic surfaces, and glacial sedimentary valleys. Unweathered volcanic terrains show flashy peak flows, fast flow recessions, and low baseflow. Sedimentary terrains display subdued peak flows, slow flow recessions, and higher baseflow. Operational implementation of these calibrations at the NWS's Northwest River Forecast Center has yielded more accurate predictive results since 1995. NWS hydrologic forecasters nationwide could benefit from using drainage basin geologic characteristics in understanding and improving model calibrations and real time forecasts.     

Hydrologic Landscape Classification to Estimate Bristol Bay,Alaska Watershed Hydrology

Hydrologic landscapes (HLs) have proven to be a useful tool for broad scale assessment and classification of landscapes across the United States as they help organize larger geographical areas into areas of similar hydrologic characteristics. We developed a HL classification for the Bristol Bay watershed of southwest Alaska that incorporates indices of annual climate and seasonality, terrain, geology, and the influences of large lakes and glaciers. A HL classification is particularly useful in this large watershed because of its hydrologic and landscape variability, important salmon fishery, variety of environmental and potential anthropogenic stressors, and lack of widespread hydrologic data. Following creation of Bristol Bay basin‐wide HL classes, we compared the HL distributions within watersheds grouped by two calculated runoff parameters derived from available long‐term streamflow records and found HL distributions within these groups provided predictive insight on hydrologic behavior. Using these developed runoff groups, we estimated expected hydrologic behavior in watersheds across the larger Bristol Bay watershed that lacked gauged streamflow records. The HL approach provides a scientific basis for estimating the first‐order hydrologic behavior of watersheds and landscapes that lack detailed hydrologic information.     

A Quantitative Method for Delineating Regions: An Example for the Western Corn Belt Plains Ecoregion of the USA

/ A method was developed to systematically delineate boundaries forecological classification of regions. The process entailed the use ofsmall-scale digital data to quantify spatial concordance among environmentalattribute data sets. The data sets were grouped into spatially related themesusing cluster analysis and multidimensional scaling. Selected data sets werethen used either individually or collectively to divide the study area intosubregions that exhibited different environmental attributes. The method wasapplied to a previously defined ecological unit, the western Corn Belt of thecentral United States. The results showed that the portion of the study areawith intensive corn and soybean production was identifiable using each of thethree input data sets selected for partitioning (soil associations; AVHRRremote-sensing imagery; and a combined data set of landform, forest, andsoils data). The classification of other portions of the study area washighly dependent on the type and scale of the input data. The systematicmethodology used here offers advantages over other methods for identifyingecological regions in that the results from the systematic approach can bereproduced, the boundaries between ecological units can be revised based onnew or more accurate data, important ecological processes are explicitlychosen to delineate boundaries, and transition zones between regions can bequantified.KEY WORDS: Ecoregions; Spatial analysis; Corn Belt; Iowa; GIS;Regionalization     

INTERDISCIPLINARY MODELING IN THE ANALYSIS OF THE SALINITY PROBLEMS OF THE SAFFORD VALLEY1

ABSTRACT: A groundwater quality change of +0.13 millimhos electrical conductivity was documented between 1940 and 1 972 in the Safford Valley. The change is attributable to four principal mechanisms: pumping-encouraged saline artesian aquifer leakage, natural recharge of the water table aquifer by saline waters, leaching of agricultural waters into the aquifer and the lateral flow of groundwater through saline lacustrine beds. A hydrologic study of the area has shown the first of these mechanisms to be predominant. Salinity modeling has shown three regions of salinity change, and salinity increase projections for each are determined. An economic analysis and an economic model are then combined with the physical model, yielding information as to when certain economic conditions are reached with respect to the salinity increase. This combined model shows that, based on projected salinity trends, cotton, the principal agricultural crop of the valley, will remain economical to cultivate for a significant time beyond the model's limit of prediction. Alfalfa, on the other hand, should go out of production in large areas of the valley by 1990, and not be under economical cultivation by 2040. A sociologic model, based on the cluster analysis of questionnaire data, shows an awareness of the salinity problems of the area but little concern over them. Interdisciplinary model based salinity control regulations are made.     

Effect of Classification Procedure on the Performance of Numerically Defined Ecological Regions

Ecological regionalizations define geographic regions exhibiting relative homogeneity in ecological (i.e., environmental and biotic) characteristics. Multivariate clustering methods have been used to define ecological regions based on subjectively chosen environmental variables. We developed and tested three procedures for defining ecological regions based on spatial modeling of a multivariate target pattern that is represented by compositional dissimilarities between locations (e.g., taxonomic dissimilarities). The procedures use a “training dataset” representing the target pattern and models this as a function of environmental variables. The model is then extrapolated to the entire domain of interest. Environmental data for our analysis were drawn from a 400 m grid covering all of Switzerland and consisted of 12 variables describing climate, topography and lithology. Our target patterns comprised land cover composition of each grid cell that was derived from interpretation of aerial photographs. For Regionalization 1 we used conventional cluster analysis of the environmental variables to define 60 hierarchically organized levels comprising from 5 to 300 regions. Regionalization 1 provided a base-case for comparison with the model-based regionalizations. Regionalization 2, 3 and 4 also comprised 60 hierarchically organized levels and were derived by modeling land cover composition for 4000 randomly selected “training” cells. Regionalization 2 was based on cluster analysis of environmental variables that were transformed based on a Generalized Dissimilarity Model (GDM). Regionalization 3 and 4 were defined by clustering the training cells based on their land cover composition followed by predictive modeling of the distribution of the land cover clusters using Classification and Regression Tree (CART) and Random Forest (RF) models. Independent test data (i.e. not used to train the models) were used to test the discrimination of land cover composition at all hierarchical levels of the regionalizations using the classification strength (CS) statistic. CS for all the model-based regionalizations was significantly higher than for Regionalization 1. Regionalization 3 and 4 performed significantly better than Regionalization 2 at finer hierarchical levels (many regions) and Regionalization 4 performed significantly better than Regionalization 3 for coarse levels of detail (few regions). Compositional modeling can significantly increase the performance of numerically defined ecological regionalizations. CART and RF-based models appear to produce stronger regionalizations because discriminating variables are able to change at each hierarchic level.     

EVALUATION OF THE APPLICABILITY OF THE SWAT MODEL FOR COASTAL WATERSHEDS IN SOUTHEASTERN LOUISIANA1

ABSTRACT: The Soil and Water Assessment Tool (SWAT) has been used for hydrologic analyses at various watershed scales. However, little is known about the model's performance in coastal watersheds. In this study SWAT was evaluated for its applicability in three Louisiana coastal watersheds: the Amite, Tickfaw, and Tangipahoa River watersheds. The model was calibrated with daily discharge from 1976 to 1977 and validated from 1979 to 1999 for the Amite and Tangipahoa and with daily discharge from 1979 to 1989 for the Tickfaw. Deviation of mean discharge and the Nash‐Sutcliffe model efficiency were used to evaluate model behavior. The study found that Manning's roughness coefficient for the main channel, SCS curve number, and soil evaporation compensation factor were the most sensitive parameters for these coastal watersheds. The Manning's roughness coefficient showed the greatest effect on the response time of surface runoff, suggesting the critical role of channel routing in hydrologic modeling for lowland watersheds. The SWAT model demonstrated an excellent performance, with Nash‐Sutcliffe efficiencies of 0.935, 0.940, and 0.960 for calibrations of the Amite, Tickfaw, and Tangipahoa watersheds, respectively, and of 0.851, 0.811, and 0.867 for validations. The modeling results demonstrate that SWAT is capable of simulating hydrologic processes for medium scale to large scale coastal lowland watersheds in Louisiana.     

MANAGEMENT AND ANALYSIS OF WATER-USE DATA USING A GEOGRAPHIC INFORMATION SYSTEM1

ASSTRACT: As part of its mission, the U.S. Geological Survey conducts water-resources research. Site-specific and aggregate water-use data are used in the Survey's National Water-Use Information Program and in various hydrologic investigations. Both types of activities have specific requirements in terms of water-use data access, analysis, and display. In Kansas, the Survey obtains water-use information from several sources. Trpically, this information is in a format that is not readily usable by the Survey. Geographic information system (GIS) technology is being used to restructure the available water-use data into a format that allows users to readily access and summarize site-specific water-use data by source (i.e., surface or ground water), type of use, and user-defined area.     

APPLICATION OF INFILTRATION THEORY FOR THE DETERMINATION OF EXCESS RAINFALL HYETOGRAPH1

ABSTRACT: The purpose of this paper is to show through the use of numerical examples that modern infiltration theory can be used in everyday hydrologic practice. The actual use of four methods of calculation of infiltration rates and of excess rainfall rates is demonstrated for the case when simultaneous data of rainfall and stream flow are available for a watershed. The four methods are: (1) the well known Π-index method, (2) the traditional Horton's infiltration capacity formula, (3) the less traditional Green and Ampt infiltration capacity formula, and (4) a ponding time approach. It is recommended that hydrologists become at least familiar with the numerical procedures involved in the ponding time and postponding infiltration approach. This approach, though not flawless, should be preferred to the other three methods if use of the other three is at all considered.     

EFFECTIVE USE OF LANDSAT DATA IN HYDROLOGIC MODELS1

ABSTRACT: In a cooperative demonstration project, NASA and the U.S. Army Corps of Engineers (Corps) compared conventional and Landsat-derived land-use data for use in hydrologic models, and the resulting discharge frequency curves were analyzed. When a grid-based data-management system was used on a cell-by-cell basis (size about 1.1 acres or 0.45 hectare), Landsat classification accuracy was only 64 percent, but, when the grid cells were aggregated into watersheds, the classification accuracy increased to about 95 percent. When both conventional and Landsat land-use data were input to the HEC-1 model for generating discharge frequency curves, the differences in calculated discharge were judged insignificant for subbasins as small as 1.0mi² (2.59 km²). For basins larger than 10mi² (25.9km²), use of the Landsat approach is more cost-effective than use of conventional methods. Digital Landsat data can also be used effectively by local and regional agencies for hydrologic analysis by incorporating the data into grid-based data-management systems. The transfer of this new technology is well under way through inclusion in some Corps training courses and through use by both county government personnel and private consultants.     

Interpolating SRTM Elevation Data to Higher Resolution to Improve Hydrologic Analysis

The Shuttle Radar Topography Mission (SRTM) digital elevation model (DEM) has been a valuable resource for hydrological analysis, providing elevation data at a consistent resolution on a near‐global scale. However, its resolution (three arc‐second or 90 m) is sometimes too low to obtain the desired level of accuracy and precision for hydrologic analysis. We evaluated the performance of several methods for interpolating SRTM three arc‐second data to a 30‐m resolution grid to better represent topography and derive terrain characteristics of the landscape. STRM data were interpolated to 30‐m DEMs on a common grid using spline, inverse distance weighting (IDW), kriging (KR), natural neighbor methods, and cubic convolution (CC) resampling. Accuracy of the methods was assessed by comparing interpolated and resampled 30‐m grids with the reference data. Slope, aspect, sinks, and stream networks were derived for the 30‐m grids and compared on a cell‐by‐cell basis to evaluate their performance in reproducing the derivatives. The comparisons identify spline and KR as the most accurate interpolation methods, of which spline is preferred because of its relative simplicity. IDW provided the greatest bias in all methods with artifacts evident in slope and aspect maps. The performance of CC projection directly to a 30‐m resolution was comparable to spline interpolation, thus is recommended as the most convenient method for interpolating SRTM to a higher resolution.     

THE MISUSE OF HYDROLOGIC UNIT MAPS FOR EXTRAPOLATION,REPORTING, AND ECOSYSTEM MANAGEMENT1

ABSTRACT: The use of watersheds to conduct research on land/water relationships has expanded recently to include both extrapolation and reporting of water resource information and ecosystem management. More often than not, hydrologic units (HUs) are used for these purposes, with the implication that hydrologic units are synonymous with watersheds. Whereas true topographic watersheds are areas within which apparent surface water drains to a particular point, generally only 45 percent of all hydrologic units, regardless of their hierarchical level, meet this definition. Because the area contributing to the downstream point in many hydrologic units extends far beyond the unit boundaries, use of the hydrologic unit framework to show regional and national patterns of water quality and other environmental resources can result in incorrect and misleading illustrations. In this paper, the implications of this misuse are demonstrated using four adjacent HUs in central Texas. A more effective way of showing regional patterns in environmental resources is by using data from true watersheds representative of different ecological regions containing particular mosaics of geographical characteristics affecting differences in ecosystems and water quality.