29 June 2018 in Water Resources Research
Studying the hysteretic relationships embedded in high‐frequency suspended‐sediment concentration and river discharge data over 600+ storm events provides insight into the drivers and sources of riverine sediment during storm events. However, the literature to date remains limited to a simple visual classification system (linear, clockwise, counter‐clockwise, and figure‐eight patterns) or the collapse of hysteresis patterns to an index. This study leverages 3 years of suspended‐sediment and discharge data to show proof‐of‐concept for automating the classification and assessment of event sediment dynamics using machine learning. Across all catchment sites, 600+ storm events were captured and classified into 14 hysteresis patterns. Event classification was automated using a restricted Boltzmann machine (RBM), a type of artificial neural network, trained on 2‐D images of the suspended‐sediment discharge (hysteresis) plots. Expansion of the hysteresis patterns to 14 classes allowed for new insight into drivers of the sediment‐discharge event dynamics including spatial scale, antecedent conditions, hydrology, and rainfall. The probabilistic RBM correctly classified hysteresis patterns (to the exact class or next most similar class) 70% of the time. With increased availability of high‐frequency sensor data, this approach can be used to inform watershed management efforts to identify sediment sources and reduce fine sediment export.
11 April 2022 in IEEE Sensors Journal
Imputation of missing sensor-collected data is often an important step prior to machine learning and statistical data analysis. One particular data imputation challenge is filling large data gaps when the only related data comes from the same sensor station. In this paper, we propose a framework to improve the popular multivariate imputation by chained equations (MICE) method for dealing with missing data. One key strategy we use to improve model accuracy is to reshape the original sensor data to leverage the correlation between the missing data and the observed data. We demonstrate our framework using data from continuous water quality monitoring stations in Vermont. Because of possible irregularly spaced peaks throughout the time series, the reshaped data is split into extreme and normal values and two MICE models are built. We also recommend that sensor-collected data should be transformed to meet the machine learning model assumptions. According to our experimental results, these strategies can improve MICE data imputation model accuracy at least 23% for large data gaps based on $\text {R}^{{2}}$ values and are promising to be applied for other data imputation algorithms.
1 February 2014 in Natural Hazards Review
Tropical Storm Irene and spring flooding in 2011 exposed the vulnerabilities of mobile home parks in Vermont when 154 mobile homes in parks were destroyed. The question of mobile home parks’ relationship to floodplains was a pressing concern to state officials as displaced households sought to resettle. Little analysis had been done about the extent to which Vermont’s mobile home parks were exposed to flooding and the data to do this analysis had not been assessed or assembled. A spatial overlay analysis largely using multiple sources of existing data revealed that nearly 32% of all mobile home parks in the state have some of their land in floodplains, and more than 20% of all mobile home parks have at least one house in the floodplain. Statewide, nearly 12% of mobile homes in parks are in floodplains. A key element in this assessment was an existing geographic information system (GIS) data set showing the location and E911 addresses of residences, an outcome of Vermont’s decision to geolocate all dwellings in the state and make that data set publically available. Similar data are available in most states but have not been made public. The research demonstrates the benefits to policy makers and emergency planners of creating and making available accurate GIS databases of residences. The utility of this rapid assessment in planning for mobile home park communities is discussed.
1 January 2019 in Geomatics, Natural Hazards and Risk
Excessive streambank erosion is a significant source of fine sediments and associated nutrients in many river systems as well as poses risk to infrastructure. Geomorphic change detection using high-resolution topographic data is a useful method for monitoring the extent of bank erosion along river corridors. Recent advances in an unmanned aircraft system (UAS) and structure from motion (SfM) photogrammetry techniques allow acquisition of high-resolution topographic data, which are the methods used in this study. To evaluate the effectiveness of UAS-based photogrammetry for monitoring bank erosion, a fixed-wing UAS was deployed to survey 20 km of river corridors in central Vermont, in the northeastern United States multiple times over a two-year period. Digital elevation models (DEMs) and DEMs of difference allowed quantification of volumetric changes along selected portions of the survey area where notable erosion occurred. Results showed that UAS was capable of collecting high-quality topographic data at fine resolutions even along vegetated river corridors provided that the surveys were conducted in early spring, after snowmelt but prior to summer vegetation growth. Longer term estimates of streambank movements using the UAS showed good comparison to previously collected airborne lidar surveys and allowed reliable quantification of significant geomorphic changes along rivers.
4 January 2019 in Journal of Geophysical Research: Biogeosciences
Movement of sediment, and associated phosphorus, from stream banks to freshwater lakes is predicted to increase with greater frequency of extreme precipitation events. This higher phosphorus load may accelerate harmful algal blooms in affected water bodies, such as Lake Champlain in Vermont, New York, and Québec. In the Mad River, a subwatershed in central Vermont's Lake Champlain Basin, extreme flooding from Tropical Storm Irene in 2011 caused extensive erosion. We measured stream channel change along the main stem between 2008 and 2011 by digitizing available prestorm and poststorm aerial imagery. Soils were sampled post Irene at six active stream erosion sites, using an experimental design to measure differences in soil texture and phosphorus both with depth (90 cm) and distance from the stream. In addition to total phosphorus (TP), we determined bioavailable (soil test) phosphorus (STP) and the degree of phosphorus saturation (DPS). The six sites represented a 0.87‐km length of stream bank that contributed an estimated 17.6 × 10 3 Mg of sediment and 15.8 Mg of TP, roughly the same as average annual watershed export estimates. At four sites, the STP and DPS were low and suggested little potential for short‐term phosphorus release. At two agricultural sites where the lateral extent of erosion was high, imagery showed a clear loss of well‐established riparian buffer. Present‐day near‐stream soils were elevated in STP and DPS. An increase in these extreme events will clearly increase sediment loads. There will also be increasing concentration of sediment phosphorus if stream banks continue to erode into actively managed agricultural fields.
25 July 2017 in River Research and Applications