Philippe Tissot

Neural Network Forecasting of Storm Surges along the Gulf of Mexico

Accurate water level forecasts are of vital importance along the Gulf of Mexico as its waterways play a critical economie role for a number of industries including shipping, oil and gas, tourism, and fisheries. While astronomical forcing (tides) is well tabulated, water level changes along the Gulf Coast are frequently dominated by meteorological factors. Their impact is often larger than the tidal range itself and unaccounted for in present forecasts. We have taken advantage of the increasing availability of real time data for the Texas Gulf Coast and have developed neural network models to forecast future water levels. The selected inputs to the model include water levels, wind stress, barometric pressures as well as tidal forecasts and wind forecasts. A very simple neural network structure is found to be optimal for this problem. The performance of the model is computed for forecasting times between l and 48 hours and compared with the tide tables. The model is alternatively trained and tested using three-month data sets from the 1997, 1998 and 1999 records of the Pleasure Pier Station located on Galveston Island near Houston, Texas. Models including wind forecasts outperform other models and are considerably more accurate than the tide tables for the forecasting time range tested, demonstrating the viability of neural network based models for the forecasting of water levels along the Gulf Coast.

Spatiotemporal Predictive Models for Juvenile Southern Flounder in Texas Estuaries

Abstract Southern Flounder Paralichthys lethostigma supports a multimillion dollar commercial and recreational fishery in the Gulf of Mexico. Despite its economic importance, the Southern Flounder population has been declining for decades. To improve the management of this fishery, both population trends and changes in environmental conditions need to be considered. Using two different statistical modeling techniques, boosted regression tree (BRT) and artificial neural network (ANN), a 29-year fisheries-independent record of juvenile Southern Flounder abundance in Texas was examined to illustrate how environmental factors influence the temporal and spatial distribution of juvenile Southern Flounder. Boosted regression trees show the presence of juvenile Southern Flounder is closely associated with relatively low temperatures, low salinity levels, and high dissolved oxygen concentrations. Both ANN and BRT models resulted in high predictive performance with slight spatial differences in predicted distributi...

Using an artificial neural network to improve predictions of water levels where tide charts fail

Tide tables are the method of choice for water level predictions in most coastal regions. In the United States, the National Ocean Service (NOS) uses harmonic analysis and time series of previous water levels to compute tide tables. This method is adequate for most locations along the US coast. However, for many locations along the coast of the Gulf of Mexico, tide tables do not meet NOS criteria. Wind forcing has been recognized as the main variable not included in harmonic analysis. The performance of the tide charts is particularly poor in shallow embayments along the coast of Texas. Recent research at Texas A&M University-Corpus Christi has shown that Artificial Neural Network (ANN) models including input variables such as previous water levels, tidal forecasts, wind speed, wind direction, wind forecasts and barometric pressure can greatly improve water level predictions at several coastal locations including open coast and deep embayment stations. In this paper, the ANN modeling technique was applied for the first time to a shallow embayment, the station of Rockport located near Corpus Christi, Texas. The ANN performance was compared to the NOS tide charts and the persistence model for the years 1997 to 2001. This site was ideal because it is located in a shallow embayment along the Texas coast and there is an 11-year historical record of water levels and meteorological data in the Texas Coastal Ocean Observation Network (TCOON) database. The performance of the ANN model was measured using NOS criteria such as Central Frequency (CF), Maximum Duration of Positive Outliers (MDPO), and Maximum Duration of Negative Outliers (MDNO). The ANN model compared favorably to existing models using these criteria and is the best predictor of future water levels tested.

Developing a goodness criteria for tide predictions based on fuzzy preference ranking

The paper deals with the developing of the tool to measure quality of predictions of water levels in estuaries and shallow waters of the Gulf of Mexico, when tide charts cannot provide reliable predictions. In future this goodness criteria of predictions will be applied to different regions.

Methodology for Applying GIS to Evaluate Hydrodynamic Model Performance in Predicting Coastal Inundation

ABSTRACT Reid, S.K.; Tissot, P.E., and Williams, D.D., 2014. Methodology for applying GIS to evaluate hydrodynamic model performance in predicting coastal inundation. Accurate inundation predictions are critical to habitat conservation, littoral boundary definition, and coastal planning. Validating a hydrodynamic model against field observations is essential for evaluating model performance. The output of a typical hydrodynamic model provides georeferenced predictions that can be used to delineate a wet/dry boundary. This paper presents two geospatial methods that complement in-situ point-based validation. The methods utilize the Coastal Modeling System (CMS) hydrodynamic model and ArcMap software. In general, the techniques are applicable to other calibrated hydrodynamic models and may be applied independently or in conjunction. Each method was validated with an extensive data set available for tidal flats located along Packery Channel, Texas. Method 1 compared model predictions to the observed conditions of high-resolution aerial and satellite imagery, which were classified using one of two geospatial processes. Method 2 applied the model output to delineate the maximum flood extent and then compared the predicted extent to topographic surveys, which define the true time-dependent flood line. Method 1 was applied to 11 test cases for which reliable high-resolution aerial and satellite imagery was available, while method 2 was based on five topographic surveys. The hydrodynamic model was well calibrated, with average absolute errors ranging from 0.026 to 0.125 m/s for current predictions. Analyses quantified agreement between model predictions and classified imagery ranging from 69% to 91%. This methodology expands present capabilities to assess and improve hydrodynamic model predictions, particularly for inundation delineation in shallow coastal environments and coastal settings strongly influenced by nontidal factors such as wind.

PERFORMANCE AND COMPARISON OF WATER LEVEL FORECASTING MODELS FOR THE TEXAS PORTS AND WATERWAYS

The ports and waterways of the Texas Gulf Coast are of vital importance to the shipping industry as well as the overall US economy. Safe navigation, particularly underkeel clearance, within these shallow, confined waterways requires accurate water level forecasts. While tide tables are tabulated for a number of locations along the Texas Gulf coast, they do not meet National Ocean Service (NOS) standards due to meteorological forcing. This paper presents and compares alternative models to improve real-time water level forecasts, including a new model based on Artificial Neural Networks (ANN). All models include real-time measurements collected by the Texas Coastal Ocean Observation Network (TCOON) and the forecasts are published on the World Wide Web. The new ANN model is shown to improve considerably upon the tide tables and the other models tested and to meet NOS criteria for many locations for up to 48-hour forecasts. Model performances are compared for Corpus Christi Bay and Galveston Bay and present model limitations and future improvements are discussed.

Local and Remote Forcing of Subtidal Water Level and Setup Fluctuations in Coastal and Estuarine Environments

The relative importance of remote and local forcing on the subtidal response in Galveston Bay was studied using water level and wind data observed during the winter and spring months from 1997 to 2000. The study confirmed the importance of remote forcing through Eckman transport for the water level response and local forcing for the surface slope response. These two forcing mechanisms act independently since the estuary axis was oriented roughly orthogonal to the coastline. A neural network model was introduced which used the meteorological data to predict the water level anomaly which, when added to the harmonic tides, provided good estimates of the total water level at the bay entrance (remote forcing). 1. I N T R O D U C T I O N The need for reliable water level forecasting is increasing with the tread toward deepdraft vessels, particularly for shallow water ports along the Gulf of Mexico (NOAA, 1999). Nine of the twelve largest U.S. ports are located along the Gulf of Mexico, and ports served by the Mobile Bay Entrance and Galveston Bay Entrance account for 46% of the total U.S. tonnage (NOAA, 1999). Although the astronomical tides in the Northern Gulf of Mexico are easily predicted by conventional harmonic analysis, it is difficult to accurately predict the total water level fluctuations because of frequent meteorological events~ such as the passage of strong cold fronts. Our ilmbility to accurately predict water level anomalies (difference between the observed water level and the tide prediction) can have severe consequences. In Galveston Bay there were over 1,240 ship groundings between 1986 and 1991, with a significant number of incidents involving petrochemicals. 1Die. of Coastal and Ocean Engrg., Dept. of Civil Engrg., Texas A&M Univ., College Station, TX 77843-3136 USA; dtc~tamu.edu :Die. of Nearshore Research, Conrad Blucher Institute for Surveying and Science, Texas A&M Univ.-Corpus Christi, Corpus Christi, TX 78412 USA; PTissot~envcc00.cbi.tamucc.edu

Hypothermic stunning of green sea turtles in a western Gulf of Mexico foraging habitat

Texas waters provide one of the most important developmental and foraging habitats for juvenile green turtles (Chelonia mydas) in the western Gulf of Mexico, but hypothermic stunning is a significant threat and was the largest cause of green turtle strandings in Texas from 1980 through 2015; of the 8,107 green turtles found stranded, 4,529 (55.9%) were victims of hypothermic stunning. Additionally, during this time, 203 hypothermic stunned green turtles were found incidentally captured due to power plant water intake entrapment. Overall, 63.9% of 4,529 hypothermic stunned turtles were found alive, and 92.0% of those survived rehabilitation and were released. Numbers of green turtles recorded as stranded and as affected by hypothermic stunning increased over time, and were most numerous from 2007 through 2015. Large hypothermic stunning events (with more than 450 turtles documented) occurred during the winters of 2009–2010, 2010–2011, 2013–2014, and 2014–2015. Hypothermic stunning was documented between November and March, but peaked at various times depending on passage of severe weather systems. Hypothermic stunning occurred state-wide, but was most prevalent in South Texas, particularly the Laguna Madre. In the Laguna Madre, hypothermic stunning was associated with an abrupt drop in water temperatures strong northerly winds, and a threshold mean water temperature of 8.0°C predicted large turtle hypothermic stunning events. Knowledge of environmental parameters contributing to hypothermic stunning and the temporal and spatial distribution of turtles affected in the past, can aid with formulation of proactive, targeted search and rescue efforts that can ultimately save the lives of many affected individuals, and aid with recovery efforts for this bi-national stock. Such rescue efforts are required under the U.S. Endangered Species Act and respond to humanitarian concerns of the public.

Estimated Increase in Inundation Probability with Confidence Intervals for Galveston, Texas

This study uses bootstrap methods to estimate confidence intervals for increases in inundation probability at the Pier 21 tide gauge in Galveston, Texas. The local surge is modeled using the generalized extreme value (GEV) distribution. Resamples of the historical record are created, and a GEV model is fitted to each resample. This ensemble of models is then used to estimate future water level exceedance probabilities under two possible sea level rise scenarios, a conservative linear continuation of the past centurys trend, and a scenario based on the upper limit of the sea level range in the IPCC AR4 report, i.e. the A1FI scenario. The distribution of future exceedance probabilities is trimmed to estimate 90% and 95% confidence intervals around the estimated proportional change in annual water level exceedance probabilities by 2100. The study shows that even under the conservative scenario and using the wider 95% intervals, the frequency of surges of 1.1 m (current return period of 16 years) becomes at least 4 times as common by the end of the century.

Thunderstorm Predictions Using Artificial Neural Networks

Artificial neural network (ANN) model classifiers were developed to generate ≤ 15 h predictions of thunderstorms within three 400-km2 domains. The feed-forward, multilayer perceptron and single hidden layer network topology, scaled conjugate gradient learning algorithm, and the sigmoid (linear) transfer function in the hidden (output) layer were used. The optimal number of neurons in the hidden layer was determined iteratively based on training set performance. Three sets of nine ANN models were developed: two sets based on predictors chosen from feature selection (FS) techniques and one set with all 36 predictors. The predictors were based on output from a numerical weather prediction (NWP) model. This study amends an earlier study and involves the increase in available training data by two orders of magnitude. ANN model performance was compared to corresponding performances of operational forecasters and multi-linear regression (MLR) models. Results revealed improvement relative to ANN models from the previous study. Comparative results between the three sets of classifiers, NDFD, and MLR models for this study were mixed—the best performers were a function of prediction hour, domain, and FS technique. Boosting the fraction of total positive target data (lightning strikes) in the training set did not improve generalization.