Understanding land cover dynamics is one of the most important challenges in the study of global change. Many of these changes take place at very fine scales (less than 1 km cell size), and require the analysis of high resolution satellite images for accurate measurement. Databases for land cover dynamics are essential for global carbon models, biogeochemical cycling, hydrological modeling and ecosystem response modeling. These needs have been recognized in a number of international programs such as the International Geosphere Biosphere Program (IGBP), the World Climate Research Program and the International Satellite Land Surface Climatology Project.
Some aspects of global change modeling depend on relatively coarse resolution data sets, as in the case of General Circulation Models, where resolutions as coarse as 250 km cell size are typical. Even in these situations, reliable parameterizations of sub-grid scale variability require detailed knowledge of what is happening at very fine scales. The need for the analysis of fine resolution data sets arises from the low spatial autocorrelation of land surface properties. Even in areas with near natural land cover much spatial variability can exist at the finest scales, largely because of the influence of local terrain variability on vegetation response for a given set of climate characteristics. Human-induced changes are the source of many of the finest scale alterations, and are often the most significant drivers of global changes. Currently we have very poor data on where these changes are occurring and what are the resultant land cover properties of the altered land.
As an example of the need to consider such local changes, the Landsat Pathfinder project is generating global wall-to-wall detailed databases of all the Earth's tropical moist forest, to reduce major uncertainties in the global carbon budget. Without the use of fine resolution Landsat data, it has been demonstrated conclusively that data on deforestation cannot be derived with sufficient accuracy. This task is being carried out for only roughly 10% of the Earth's surface and is largely conducted through visual analysis of the data. Currently we do not have the algorithms and computational resources available that might allow us to match human interpreters in terms of accuracy.
- At the science level we are developing new models for fundamental science problems such as atmospheric correction, determining ground reflectivity from imagery, mixture modeling, image feature extraction, image pixel and region classification, and spatial data structures.
- At the algorithms level we are designing and developing portable and scalable parallel implementations of our algorithms and data structures, and testing them - first in terms of their scientific merit, and then in terms of their parallel scalability and absolute performance, on large image and map data sets.
- At the high performance computing level we are developing new tools and techniques for supporting applications such as image processing and spatial data handling on parallel machines. Our research here focuses on object oriented parallel programming and on new models and techniques for parallel I/O of large image and map data structures.
Links to other CRPC Research
The Maryland Land Cover Dynamics Grand Challenge project leverages a variety of collaborative projects being carried out with other CRPC institutions. Rice, Syracuse, Argonne, CalTech and Maryland are collaborating in the development of Scalable I/O compiler methods and runtime support. The I/O optimization library developed by Maryland in the context of the Land Cover Dyanamics Grand Challenge project is being as one of the starting points in the development of compiler runtime support for scalable I/O.The CHAOS runtime support library is used by both CRPC Fortran D compilers to provide irregular problem runtime support. As part of the Grand Challenge project, Maryland has produced a preliminary C++ version of the CHAOS runtime support library. CHAOS++ improves on CHAOS in being able to handle distributed data structures that have global pointers. CHAOS++ will be used as compiler runtime support in future CRPC Fortran D projects as well as to support future CRPC projects in High Performance C/C++. CHAOS++ will also be used as one of the starting points for work in the Parallel Runtime Consortium, lead by Geoffrey Fox.