Figure 1. Patterns of phytoplankton concentrations around the island of Tasmania. Image acquired by the CZCS on 27 November 1981.
Figure 2. Closeup of eddy outlined in white. Pixel resolution is 1 km.
Figs. 1 &;2 courtesy of NASA SeaWiFS
Thermal imaging systems can observe the surface temperature of the water, revealing thermal mixing phenomena such as snowmelt running into the sea and gulf stream eddies. Visible-light phenomena, such as an algae bloom or sediment runoff from shore, are revealed to the EOSs as changes in the albedo of the imaged area at selected frequency bands (Figures 1 and 2). Some events, such as anomalous high reflectance patches near Bermuda, occur over a large area in a short period of time just after major storms [5]. These events pass before research vessels can be safely dispatched to the area of interest. Finally, radar altimeter systems coupled with accurate orbital navigation data can reveal detailed information about wave heights and the presence of mesoscale topography such as seamounts or the mid-ocean ridge, but cannot give us information about the details of bathymetric structures.
All of the sensor systems described depend upon use of the electromagnetic spectrum. Because sea water is effectively opaque[6], ocean imaging by EOS systems is generally restricted to the top few meters (or tens of meters), leaving the three-dimensional nature of the ocean unknown (Fig. 3a). This lack of information is further compounded by the low resolution of remote sensing data, which is determined by the imaging platform, its location, dynamics and the spectral range of the sensors. With the exception of military-grade spy satellites, typical visual light observation image resolutions vary from 1 to 120 meters, according to the altitude of the system, the operational bandwidth of the sensor, its velocity and its data capture rate (Table 1).
Unlike moored sensors or fixed ground stations, AUVs can be redirected to locate, track, and sample the water column or feature of interest anywhere that is needed. However, their typically low operating speeds and limited battery capacity means that efficient data collection strategies are very important. This motivates research into the problems of adaptive sampling and search strategies, which has been investigated by Bellingham and Willcox[7] and Burien et al[8], among others.
On other occasions the AUV may be sent out to investigate some sort of dynamic phenomena whose characteristics may not be well known or predictable in advance. Such situations might be a thermal front or a chemical spill. In these cases the AUV is assigned to locate and map a dynamic feature in order to characterize its microscale structure. For example: "Which is correct, Fig. 3b, Fig. 3c or neither?" This dynamic phenomena may also be poorly characterized in advance, requiring that one or more of the feature's characteristics be determined while the vehicle is in transit or in situ. Examples of this might be determining what change in temperature constitutes a front or what change in turbidity constitutes the edge of a spill.
To direct the vehicles intelligently, we need some idea of what the greater structure of the phenomena of interest looks like, where the phenomena is actually manifesting itself and what are its characteristic features. EOSs are designed to produce this kind of information.
| System | Typical Image Resolution |
| visible light | |
| Landsat (Thematic Mapper) | 60 Meters |
| Landsat (Multispectral Scanner) | 120 Meters |
| SPOT | 1-2 Meters |
| IR/Thermal | |
| NOAA IR System | 900 Meters |
| NASA/AVHRR1 | 1-14 km |
| Hybrid | |
| NASA ER2/AVIRIS Airborne System | 11 Meters |
EOSs and AUVs have complementary capabilities. In the case of ground-truthing, the AUV measurements can be compared to the remote observation data, allowing us to verify or improve instrument calibration and measure atmospheric effects.
In the case of directed sensing or rapid-response systems the remote sensing data can be used to determine when an event of interest is occurring, where it is taking place and what the characteristic features are. The AUV is then dispatched to the appropriate area, armed with the latest information, for a closer look. Upon arriving, the AUV then proceeds to measure the phenomena in detail, confirming and/or calibrating the remote sensor measurements as well as augmenting those measurements with additional data not otherwise obtainable. In our simple example, we might find that while our model predicted Fig. 3c, the vehicle actually discovers a structure resembling 3d.
For example, a satellite observes an interesting phenomenon from orbit and transmits the imaging data to a ground station (perhaps the plankton bloom in Figure 1). On the ground an operator notes the phenomenon, its location when observed and its characteristics. Information about the characteristics of the phenomenon (relative phytoplankton concentration, estimated thermal ranges, etc.) and a region where the phenomenon is estimated to be are then transmitted to a waiting AUV [9].
The AUV then proceeds to the survey area and begins to search for the phenomenon. The vehicle maps it out in detail, taking measurements and water samples as desired. Upon completion the vehicle returns to base or surfaces for pickup and data download.
Because earth observation systems are large, expensive to use and difficult to get time on, we have chosen to perform our first tests in simulation. However, we wish to use as much real data as possible in our experiments to preserve realism. Also, because we would like to test multiple strategies both in simulation and in the field, we would like to use a dataset which allows us to move easily from simulation to field tests. For these reasons we have chosen to represent our feature of interest with a low resolution bathymetric map. A bathymetric feature is repeatable and, since it is based on a real bathymetric dataset, strategies which prove successful in simulation can be verified in the field.
There are several issues being investigated by this study. Among them are:
€What (and how much) a priori information does the AUV need to perform its function?
€How to characterize the feature of interest?
€How will the AUV extract the feature from the incoming sensory data?
€How to determine where to go next based on the information already obtained?
In this example, the vehicle has been assigned a search zone (the dark-line box) and sent in to investigate and map the feature, using two different strategies. The first mission (Fig. 6) employs a simple "lawnmower" search strategy. In this simple search, the vehicle merely traverses the search zone until the area is searched or until the mission timer expires. The second mission uses a reactive approach (Fig. 7). Here the AUV uses the a priori data obtained from the EOS to make an estimate of where to start the search and what some characteristics of the feature are. The vehicle then proceeds to locate and map the feature in detail, turning back to re-acquire the feature every time it passes outside of the feature.
The key difference between the two missions is the amount of useful data gathered in a given amount of time and/or using a specific amount of energy. The mission of Fig. 6 ran for 100 minutes, while Fig. 7 was only half that at 50 minutes. The path of the search in Fig. 7 also shows that the bulk of the active mission time was spent gathering useful data in the feature. In other words, given the restricted energy supply of an AUV, the vehicle gathered more useful data per watt expended in Fig. 7 than in Fig. 6. Clearly the a priori information provided by the EOS proved to be of value to the AUV.
In the future, it is hoped that enlarged fleets of such vehicles will allow a truly three-dimensional "snapshot" of a portion of the ocean to be taken for the first time: an EOS observer overhead taking in the surface characteristics of a feature while multiple AUVs (and perhaps surface vessels as well) gather complementary data simultaneously with it.
The EOS Homepage
The SEAWiFS Homepage
Gulf of Maine Sea Surface Temperature Data
MIT Sea Grant Home Page
Florida Atlantic University
WHOI Deep Submergence Group
2. Costello, D. K., Carder, K. L. And Smith, S. M., "Unmanned Underwater Vehicles as Platforms for Optical Oceanography in Coastal Waters," AGU Proceedings, 1996
3. Szekielda, K. H., Satellite Monitoring of the Earth, Wiley Interscience, 1988, pp.125-128.
4. Bellingham, J. G., et al, "A Second Generation Survey AUV," AUV 95 Proceedings, IEEE 1995, pp. 148-155.
5. Acker, J. G., personal communication.
6. Jackson, J.D., Classical Electrodynamics, 2nd ed., John Wiley &;Sons, 1975, pp. 291.
7. Bellingham, J. G., Willcox, J. S, "Optimizing AUV Oceanographic Surveys," AUV 96 Proceedings, IEEE 1996, pp. 391-398.
8. Burian, E., Yoerger, D., Bradley, A., Singh, H., "Gradient Search with Autonomous Underwater Vehicles Using Scalar Measurements," AUV 96 Proceedings, IEEE 1996, pp. 86-98.
9. Curtin, T. B., Bellingham, J. B., Catipovic, J., Webb, D., "Autonomous Oceanographic Sampling Networks," MITSG 94-16J, MIT Sea Grant, 1994.
