Frequently Asked Questions

Why Predict Water Levels Along the Texas Coast?

Who Measures Water Levels Along the Texas Coast?

What Exactly is an Artificial Neural Network (ANN)?

How Does an Artificial Neural Network (ANN) Work?

Are Water Levels the Same Thing as Tide?

Why Not Simply look at the NOAA Tides Information?

Why Are The Present Forecasting Methods Inadequate For the Texas Coast?

How Does the Artificial Neural Network (ANN) Produce a Better Water Level Prediction?

What are the Consequences of Inaccurate Predictions?

How are Wind Observations Applied to ANN Forecasting?

How does the Persistence Model Predict Water Levels?

How is the Persistence Model Different than the ANN?

Can ANN Predict Water Levels for Tropical Storms and Hurricanes?

What is a Hydrodynamic Model?

Why Predict Water Levels Along the Texas Coast? The knowledge of future water levels is important to coastal communities for a number of reasons including emergency management, engineering activities, safety, economic factors, and recreation. The prediction of the magnitude of anticipated coastal flooding can assist communities in avoiding dangerous low-lying regions or evacuating such regions in a timely fashion. The shipping industry is particularly impacted by changes in water levels, since the amount of cargo carried into Texas ports depends on the depth of the waterways. For example, tankers potentially must unload a portion of their cargo before entering ship channels. Mariners in general and specifically commercial fishermen and anglers rely on water level forecasts for safe navigation in the shallow near-shore waters.

Who Measures Water Levels Along the Texas Coast? The Texas Coastal Ocean Observation Network (TCOON) has many, typically more than 30, hydro meteorological stations along the coast of Texas. The National Ocean Service operates  seven long-term stations established as part of its National Water Level Observation Network (NWLON). TAMUCC is one of the named partners for the TCOON cooperative project in the Texas Natural Resources Code. Most stations provide additional data such as wind speed and direction, air temperature, and water temperature, and some stations provide water current, salinity, pH, and dissolved oxygen data. The TCOON database was the source of input and verification data for the ANN and persistence water level forecasts. The database provided water level observations, barometric pressure, as well as wind speed and direction for these applications.

What Exactly is an Artificial Neural Network (ANN)? An Artificial Neural Network (ANN), sometimes referred to as just a Neural Network, refers to a type of artificial intelligence that attempts to imitate the way a human brain processes information. A neural network functions through the creation of connections between processing elements which function as the equivalent of neurons. These connections are weighted such that a particular input stimulus will produce the desired output. Neural Networks are typically used when the relationship between the input and output is known to depend on several factors, but the interaction of those factors is not well known.

How Does an Artificial Neural Network Work? There are two phases in neural information processing:
1.) Learning or “training” phase and
2.) Retrieval phase
In the training phase, a data set is used to determine the weight parameters that define the neural model. The weights are initially chosen at random and then the network is presented with a set of inputs that have known output. The output of the ANN is compared to this known output and the weights are adjusted to bring the Neural Network output closer to the actual output. This process is repeated until some criteria signify that the ANN has “learned” to recognize the pattern hidden within the data set (or there is no more data available to train the ANN). In other words, an ANN “learns” from previous examples (for example: as humans learn to recognize types of birds from examples of birds) and exhibits some capability for generalization beyond the training data.
Once trained, the ANN can then be used to process real data sets. The same types of input data are given to the ANN in the same order as during the training phase and the ANN outputs a value or values that closely mimics the learned relationship. The ANN is particularly effective for predicting events when the networks have a large database of prior examples to draw on. There are no methods for training Neural Networks that can create information that is not contained in the training data.

 Are Water Levels the Same Thing as Tides? No. Tides are defined as the alternating rise and fall of sea level with respect to land, as influenced by the gravitational attraction of the moon and sun. In certain locations where water elevation is predominately dictated by astronomical factors water level and tide are often used interchangeably. In other regions like south Texas where the average tidal range is minimal (< 1 ft) other natural forces such as wind, shoreline orientation, topography, and bathymetry dominate over the astronomical forces. In these cases, the observed water level is not the same as the predicted tide influenced water level. The combination of the tide and other influences can be referred to as the total water level fluctuation.

Why Not Simply Look at the NOAA Tides Information? National Oceanic and Atmospheric Administration tide forecasts are predicted based on the periodicity of astronomical factors (the respective motion of the moon and the sun). NOAA has stated that “presently published predictions do not meet working standards” when assessing the performance of current tidal predictions, tides being closely related to water level predictions, for regular weather conditions in Aransas Pass and Corpus Christi Bay NOAA 1991, NOAA 1994. Water level forecasting is complicated by meteorological influences (in particular wind forcing) and is often further complicated by the unique shoreline orientation and shallow water depth associated with the bays and lagoon system that exists along the Texas Gulf Coast. For example, strong winds have been observed to drive water out of the shallow lower Laguna Madre resulting in a lower than predicted water level or in contrast pile water up along the shore resulting in an increased observed water level. Differences in predicted verses observed water levels could be on the order of 1 to 2 feet along Texas shorelines. Such dramatic differences are typically observed under the action of two opposing wind regimes; a) winds directed out of the southeast – dominating greater than 50% of the time or b) winds directed out of the north – strong pulses of energy associated with frontal passage.

Why are the Present Forecasting Methods Inadequate for the Texas Coast? Water levels are typically predicted using tide tables computed based on the gravitational influence of celestial bodies (Harmonic Analysis). While this method of prediction is adequate for most coastlines, along the Gulf of Mexico meteorological effects can significantly influence the observed water level and make the tide tables ineffective. Although astronomical forcing is very predictable meteorological influences are not always periodic and are difficult to predict for periods longer than 1 or 2 days. The influence of meteorological effects, however, is frequently stronger than the influence from the celestial bodies. For example, the Corpus Christi, Texas, airport is ranked by the National Weather Service as the third windiest in the U.S. based on a multiannual average wind speed of 23.5 kph (Smith 1978).
Until recently it was not possible to make accurate water level predictions for the Texas coast. This is because the present models do not include both meteorological and astronomical effects. For example, a fourteen-day comparison between measured water levels and tidal forecasts for the Port Aransas station is shown below. The tide station is located in the Corpus Christi ship channel near Port Aransas, Texas. The difference between tidal forecasts and actual water level can be larger than 1 foot (~31 cm) for several consecutive days. A predicted difference of 1 foot is particularly significant because of the small tidal range (< 1 ft) observed along the Texas Coast.

How Does the Artificial Neural Network (ANN) Produce a Better Water Level Prediction? ANN takes meteorological parameters into account
Non linear modeling capability (Allows modeling of complex systems consisting of reciprocal relationships or feedback loops)
Ability to learn dynamically
Robustness to noisy data

What are the Consequences of Inaccurate Predictions? the inability of present models to accurately predict water level fluctuations can result in severe consequences, such as ship groundings. In an effort to improve information required for safe navigation NOAA established the Physical Oceanographic Real-Time System (PORTS). Although the PORTS system is beneficial to navigators in the Galveston area it is not available at other Texas ports, thus the ANN water level forecasts will fill this gap for navigators. In addition to navigation hazards present forecasting inaccuracies impair the management of roadways and low-lying coastal regions during presently unanticipated increases in water levels.

How are Wind Observations Applied to ANN Forecasting? ANN models were tested with exact wind and tidal forecasts as input sets. In addition, the ANN models were trained to compute their outputs using these exact forecast inputs. The National Weather Service (NWS) provided wind forecasts derived from the comparison of output from several models. The MesoETA model was selected because it provided the range of forecast needed for the ANN models with the proper geographic resolution. This model is called Meso-Eta model from NCEP [17]. Although wind predictions of 6, 12, 18, 24, 30, 36, 42, 48, 54 and 60 hours were provided, the longest ANN prediction time is 48 hours and just the past winds influence the water levels so the 54 and 60 hour forecasts were not applied to the ANN models. The use of Meso-Eta forecasts will also allow for other modeling possibilities. In this work and in the previous works [3], only the wind forecast at the station of interest was used. As the Meso-Eta model gives forecasts for Texas and the Gulf of Mexico, future models will be able to use a wider geographic range of forecasts. Including wind forecasts at several locations and in particular in the Gulf of Mexico itself should lead to models that have a more direct link between the forcing functions and the changes in water levels.

How Does the Persistence Model Predict Water Levels? This models assumes:
Water level anomalies build progressively. This is particularly true for bays and estuaries.
Water level anomaly at the time of forecasts will persist throughout the forecasting period.
In other words, take the difference between the observed water level and harmonic prediction during a past time interval and apply it to forecast over a future time interval.

How is the Persistence Model Different than the ANN? The persistence model provides a forecast with more moderate differences between observed and predicted for extreme high and low water levels.
It does not utilize wind observations as a model parameter so is not as optimized as ANN during seasons when wind is responsible for set-up and set-down in bays and estuaries.
Works well when water level anomaly changes slowly over time.

Can ANN Predict Water Levels for Tropical Storms and Hurricanes? Preliminary results indicate that ANN may be able to predict water levels at least over short temporal scales for Tropical storms.

What is a Hydrodynamic Model? Hydrodynamic models are an efficient, comprehensive approach to representing coastal water dynamics. These numerical computational models can be used to simulate currents, water levels, sediment transport and salinity.