Search

Tropical Storm Irene swept through Vermont with 100-200 mm (4-8 inches) of rain, causing intensive flooding in the Southern and Central regions. Over 300 bridges suffered damage or failed as a result of heavy rains. Resources from the Vermont Agency of Transportation, Vermont Agency of Natural Resources, and Vermont Regional Planning Commission were joined to create a comprehensive bridge database. The database combines the stream characteristics and Irene damage along with the bridge attributes into a common resource that can be used for analysis. The examination of affected bridges showed scour as the dominant mode of damage from Irene. Bridges damaged tended to have floodplain contraction, and spans incapable of passing flood flows. Flow constriction and floodplain blockage were key indications of poor performance.

Stream channel and watershed management requires an interdisciplinary approach by multiple stakeholders with often disparate goals and objectives. Even single management goals ( e.g. mitigation of property loss due to stream bank erosion and flooding, restoring stream channels and aquatic species habitat) require solutions that span multiple spatial and temporal scales. To deal with the complexity associated with goals such as reducing fluvial erosion hazards, sediment and nutrient loading and threats to aquatic habitat associated with geomorphic instability, the Vermont Agency of Natural Resources' River Management Program has been developing and testing 1) protocols for conducting field-based and remote sensing data and 2) a geographical information system (GIS)-based tool since 1999. These protocols and tools facilitate an understanding of channel instability and assist in developing strategies at appropriate scales to restore channel equilibrium. A hierarchical system of data-driven artificial neural networks (ANNs) has been developed to enhance these existing GIS-based watershed management tools for predicting channel geomorphic condition and inherent sensitivity at the reach scale. We research and develop these ANNs as an alternative model that can incorporate large amounts of data for use in the operational management of watersheds. The ANNs have been trained and tested on remotely sensed and in-field data along with corresponding expert opinions of geomorphic condition and sensitivity. The focus is performing sensitivity analyses and addressing inadequacies associated with assimilation activities that are closely tied to large amounts of hydrological-relevant remote sensing and field data.

We apply a modified counterpropagation artificial neural network (ANN) that uses multivariate data to several parameter estimation problems: (1) estimation of small scale Berea sandstone geophysical properties, (2) estimation of apparent conductivity at a leaking landfill using electromagnetic data and (3) estimation of hydraulic conductivity field at a landfill in New York State using pumping test and well log data. The counterpropagation algorithm has been enhanced in this research to allow for spatial interpolation that is comparable to traditional kriging methods. This enhanced ANN is data-driven, can incorporate large amounts of multiple data types to produce parameter estimates in real-time and does not require the computation of large covariance matrices associated with traditional geostatistical methods (kriging).

Accurate detection of a contaminant source is vital to the management of groundwater restoration projects. We developed a technique for identifying the contaminant source areas that combines site data, geostatistics and artificial neural networks (ANNs). The methodology involves replicating the kriging methods with ANNs to provide estimates of spatially dependent concentration fields, and for quantifying the uncertainty associated with the estimates. Once trained the ANNs approximate the results of ordinary kriging in a more computationally efficient manner. This provides a bi-directional prediction (i.e. forecast the future plume, as well as, the original source), which is impossible for many physics-based models.

The risk of fungal and bacterial crop disease can be predicted using risk models with specific environmental parameters such as temperature, relative humidity, solar radiation, wind speed, and leaf wetness duration (LWD). LWD has long been recognized as key in the management of crop disease. Air temperature and wetness influence the majority of fungal plant diseases. Wetness also impacts insect populations, as well as pollution deposits. Many parameters are well understood, readily defined, and easily measured. Unfortunately, LWD is a complex phenomenon, due to its spatial and temporal variability within a crop canopy. The inconvenience and uncertainty associated with monitoring LWD at the local leaf scale and the complexity of upscaling to the crop level prevent existing disease risk models from being used with reliability. In spite of their imprecision, LW projections are already included in a number of online weather products. One non- parametric statistical approach receiving scant attention for the modeling of LWD is that of artificial neural networks (ANNs). In this work, two previously untried ANNs estimate this key environmental variable at local crop scales, using local and regional weather station data and site-specific sensing data. The first ANN combines two statistical methods to accomplish this spatial mapping (a K-nearest means classifier and a Bayesian classifier), while the recurrent nature of the second ANN provides a means of leveraging the temporal property of the data. The ultimate goal is to embed the ANN into a highly-portable tool, designed to predict leaf wetness duration as an SOC (system on a chip) in conjunction with local weather stations, and as input to real-time decision support systems.

Groundwater remediation management does not end with the installation of the cleanup system and flipping the switch. Water quality samples from groundwater monitoring wells are used to verify that the remediation plan is working as planned, and to ensure that there are no unexpected impacts or risks. An important component of the adaptive long-term monitoring and operation system package aLTMOs TM (Rizzo et al., 2000) is a data-driven Bayesian filter, a data assimilation technique that integrates monitoring data and phenomenological modeling results to provide insightful analyses on plume dynamics in both spatial and temporal dimensions. In this process, a model prediction is calculated using the flow and transport simulation. When new measurements are made, they are used systematically to adjust the model prediction. Irregular spatial and temporal data collection can be incorporated. Concentration estimates and their uncertainties are updated as monitoring data are collected. Based on uncertainty analysis in spatial and temporal dimensions and cleanup goals, an optimal monitoring plan is selected. This paper uses a real superfund site problem to demonstrate the application of the data-driven Bayesian modeling technique in uncertainty analysis and redundant sampling reduction optimization in groundwater long term monitoring design.