Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 1
Report of a Workshop on Predictability & Limits-to-Prediction in Hydrologic Systems 1 Background and Goals NRC COMMITTEE ON HYDROLOGIC SCIENCE 1999 REPORT Predictions of hydrologic phenomena such as floods, seasonal precipitation deficit, aquifer response, subsurface contaminant dispersion, land use and global change impacts, etc. are practical ways of dealing with hazards. Predictions are often the foundations of hazards mitigation strategies. Predictions begin with a characterization of the current state of the system, i.e. initialization. Then the states of the system are predicted into the future according to our current understanding of the dynamic behaviors of the system. As a result observing systems and conceptual understanding are the engine behind operational prediction systems. In this way hydrologic science and operational hydrology work together to reduce the hazards faced by the public. Better characterizations of the system and more effective observing networks are needed to improve predictions. Predictability and limits-to-prediction are themes in hydrologic science that are at the core of both the research community and the operational field. Predictability and limits-to-prediction in the hydrologic sciences emerged as a priority topic for COHS during the development of its first report on the hydrologic science content of the USGCRP plan. The NRC (1999) study titled Hydrologic Science Priorities for the U.S. Global Change Research Program: An Initial Assessment was published in September 1999 and it identified
OCR for page 2
Report of a Workshop on Predictability & Limits-to-Prediction in Hydrologic Systems predictability and variability of regional and global water cycles as one of its two priority themes (together with coupling of hydrologic systems and ecosystems through chemical cycles). In NRC (1999) a series of specific science questions are posed in order to direct hydrology predictability research in a progressive direction. Addressing these science questions and bringing them to closure mark milestones in this research path. NRC (1999) proposes that research in this area should be driven towards three main goals. The first goal is the identification of predictable patterns on all pertinent spatial and temporal scales in the water cycle. Because the water cycle is composed of components with varying memory (e.g., mean residence time in atmosphere is about nine days whereas in the active groundwater system the memory may be decades), there is a natural dampening and potential for predictability based on persistence. The prevalence of autoregressive and Markovian statistical models in hydrology are testament to the recognition that the dampening of signals in various component of the water cycle may be effectively harnessed to make useful predictions. It is now recognized that there may be even more opportunities for prediction if the climate system as a whole is considered. The oceans are the long memory components of the climate system and their influence on interannual to decadal climate variability is becoming more clearly understood. This understanding may be used to make long-range predictions of regional water cycle processes. The identification of predictable patterns and linking them to large memory processes is now emerging as a promising strategy for understanding predictability of hydrologic systems and is supporting the development of useful prediction tools. It is also recognized that memory in the regional climate systems may be due to factors other than the inertia of heat and moisture reservoirs. Establishment of positive feedback mechanisms may also have the same effects as memory by prolonging an event or excursion. Box 1.1 lists a number of specific science questions from NRC (1999) associated with this first goal of predictability research in the hydrologic sciences. The second goal identified by NRC (1999) is understanding the sources of uncertainty and the propagation of uncertainty in hydrologic systems. Innovative monitoring and multi-source data fusion techniques are required to quantify variability and its
OCR for page 3
Report of a Workshop on Predictability & Limits-to-Prediction in Hydrologic Systems Box 1.1 Science questions on predictability and limits-to-prediction from the 1999 NRC report: Hydrologic Science Priorities for the US Global Change Research Program: An Initial Assessment Distinguish the Predictable and the Unpredictable Patterns of Variability What type and location of measurements will most enhance predictability? To what extent is regional-scale hydrology predictable? Across which regions and seasons can predictability of regional water cycling be enhanced by robust coupled land-atmosphere modeling? What special physical and statistical features (e.g., process pathways, influences across scales) can be used to link large-scale climate and regional-scale hydrology in the case of extreme events and how are these features different for the case of floods and the case of persistent droughts? Identify Sources of Variability and their Propagation in Hydrologic Systems What combination of remote and in situ observations and paleohydrologic records are required to identify shifts in regional and local hydrologic properties due to both natural and human-induced factors? Are there spatial patterns in the variability in the hydrologic record that may serve as reliable predictors of the impacts of global change? Understanding the Scaling and Linkages of System Components At what scales and for which processes should the spatial structure of surface heterogeneity be incorporated into the upscaling strategy for hydrologic models? What physical constraints arising due to coupling of water and energy cycles with other systems may be used to bound the estimates of local and regional hydrologic cycles?
OCR for page 4
Report of a Workshop on Predictability & Limits-to-Prediction in Hydrologic Systems changes with scale. The focus of the specific science questions associated with this goal (see Box 1.1) is on the quantitative characterization of the impacts of perturbations in hydrologic systems. These perturbations may be due to human-induced factors such as land use change or they may be due to natural variability in climate forcing. The third and final major goal posed in NRC (1999) for predictability research in the unique context of hydrologic sciences is to understand how variability in hydrologic processes change with spatial scale. Heterogeneity in landscape properties (e.g. geology, ecology, terrain) are ubiquitous in hydrology. How processes such as water, energy, and biogeochemical fluxes are dependent on the variabilities in these properties needs to be addressed to advance understanding. Box 1.1 also lists a number of specific science questions for this goal. The task of developing the full science and implementation plan for the USGCRP element went to the sixteen-member Water Cycle Study Group, organized by the federal agencies and chaired by Dr. George Hornberger (University of Virginia). THE USGCRP WATER CYCLE INITIATIVE In June 2001 the USGCRP Water Cycle Study Group published a comprehensive report titled: A Plan for a New Science Initiative on the Global Water Cycle (Hornberger et al. 2001). The report poses several priority science questions as well as several specific goals for predictability research that are listed in Box 1.2. In the USGCRP report, a series of research needs are articulated that can be mapped to activities that will have to be undertaken by the appropriate federal agencies. A sampling of these needs indicates the wide scope of the efforts. A new program is needed in the science and mathematics of water cycle predictability to guide applications of atmospheric and hydrologic theories over a broad range of space and time scales. Climate predictions on seasonal and longer time scales must be made within a probabilistic framework that takes into account the uncertainty of initial and boundary conditions, as well as the inherent characteristics of the distribution of possible states that
OCR for page 5
Report of a Workshop on Predictability & Limits-to-Prediction in Hydrologic Systems Box 1.2 Key elements of A Plan for a new Science Initiative on the Global Water Cycle (Hornberger et al., 2001) Priority Science Questions What are the underlying causes of variation in the water cycle on both global and regional scales, and to what extent is this variation induced by human activity? To what extent are variations in the global and regional water cycle predictable? How will variability and changes in the cycling of water through terrestrial and freshwater ecosystems be linked to variability and changes in cycling of carbon, nitrogen and other nutrients at regional and global scales? Specific Goals in Predictability and Limits-to-Prediction in Hydrologic Systems Demonstrate the degree of predictability of variations in the water cycle on a range of time scales (daily to centennial). This goal is to be reached through a number of program elements that include: nested modeling to deal with scales interactions, probabilistic modeling to deal with uncertainty in modeling, and scaling models for interpreting observations. Improve predictions of water resources by quantifying fluxes between key hydrologic reservoirs using observations, process understanding, and numerical modeling. The program elements required to reach this second goal include observations using surface networks: precipitation, basin-scale recharge that links the surface and subsurface reservoirs, stream-aquifer interaction which also is related to these two reservoirs, and evaporation that links the surface and the atmospheric reservoirs. Establish a systems modeling framework for making predictions and estimates of uncertainty that are useful for water-resources management, natural hazards mitigation, and policy guidance. The program elements for this goal include transfer of information from physical to socioeconomic models.
OCR for page 6
Report of a Workshop on Predictability & Limits-to-Prediction in Hydrologic Systems may ensue from the given initial state. Research is required to place current ad hoc methods of producing ensemble model predictions on a firmer theoretical basis. A systematic approach to model design and development is needed that will permit determining the scales at which predictive information should be exchanged within a nested modeling approach. This research will be heavily computational, requiring enhancements to available national computing capabilities. The development of coupled land-atmosphere models should be accelerated through the better use of data assimilation techniques. One existing vehicle for this development is the new U.S. multiagency initiative known as Land Data Assimilation System, or LDAS. LDAS should be supported and expanded to include data representing snowpack and high-latitude glaciers. Studies should examine whether two-way land-atmosphere coupling or climate modulation by local hydrologic processes results in predictability that can be exploited through coupled modeling. For seasonal and longer lead prediction of water fluxes, a modeling strategy must be developed to minimize the propagation of uncertainty among the components of such predictive models. Field campaigns and intensive observation programs to better understand interactions among land, ocean, and atmosphere are needed to isolate the effects of fast and slow processes in the hydrological cycle. Enhanced field campaigns should take place over multiple years to observe large-scale surface conditions, surface fluxes, and atmospheric variables. These large-scale observations would be supplemented with simultaneous observations of the slower components of the land system, such as groundwater levels. A continuing effort to use observations to close water budgets is critical. New data sets geared specifically for budget studies are needed. Because analysis budgets are the main link between models and observations, they should be rigorously tested against all observations, especially those hydrometeorological observations developed to cover broad space and time scales. New continental and global hydrometeorological data sets will be required to support these activities. These data sets include gridded (or equivalent) observations of streamflow over continental domains, and gridded high-resolution precipitation data. Expanded budget studies covering snow accumulation, melt, runoff, and
OCR for page 7
Report of a Workshop on Predictability & Limits-to-Prediction in Hydrologic Systems evaporation of snow in continental regions should also be undertaken to understand how snow contributes to the water cycle. Satellite observations are needed for hydrological variables not yet remotely sensed, and for which technology development may be required. These include surface soil moisture at high resolution (~10 km for hydrometeorology applications, ~40 km for hydroclimatology applications), surface freeze/thaw condition, diurnal cycle of precipitation, river and water bodies altimetry, and snow cover and water equivalent. An initiative should be designed to integrate users needs into the development of the research agenda and to ensure that research results are provided in a form useful for users. Estimates should be developed of the natural variability of surface hydrological processes that can be incorporated into water resource systems design and management, with reduced dependence on historical observations. Ensemble forecast products for operating water resource systems should be produced, with a primary focus on reservoir systems (or, in some cases, free-flowing rivers), but with implications for groundwater in systems that conjunctively use surface water and groundwater. Model testing facilities should be established at existing weather and climate prediction centers (like National Centers for Environmental Prediction (NCEP)), which would be charged with facilitating model evaluation and the transfer of methods from the general research to the operational modeling community and vice versa. THE COHS WORKSHOP These two reports (NRC 1999 and Hornberger et al. 2001) summarized above provide context for material presented at the COHS workshop in Boulder, Colorado. The workshop additionally served to engage the larger community outside of these two study groups. The primary goal was to start along the path defined by the reports by making contributions in three main areas: 1) provide concise definitions for predictability and limits to predictability, 2) identify the key technical challenges for research advances in predictability science, and 3) make recommendations to federal
OCR for page 8
Report of a Workshop on Predictability & Limits-to-Prediction in Hydrologic Systems agencies in implementing a research program in this topic. These three contributions are each contained in the sections of this report. In preparing for the workshop it became necessary to choose a particular example application from the diversity of topics in hydrologic science to demonstrate the current status and future prospects for predictability science in hydrologic systems. Predictability and understanding the limits-to-prediction are core elements of research in many specializations in hydrology. Subsurface flow and transport, surface water hydrology and chemistry, ecohydrol-ogy, snow hydrology, hydrometeorology and hydroclimatology as well as other topics in hydrologic science all involve predictions and predictability in one form or another. COHS decided that it is preferable to choose one context and delve deeply into it in order to identify some of the structural issues associated with predictability research. Such structural issues include the criteria for posing science questions and for defining metrics of progress. In the workshop the context of hydrometeorology a hydroclimatology was used to introduce some of the issues associated with predictability research. Nonetheless the findings are applicable to most other contexts in hydrologic science. In preparing for the workshop it also became necessary to recognize that the evolution of predictability science and operational prediction systems are intimately linked. During the workshop it was recognized that the two are complementary and that together they form a synergistic approach to advancing understanding in hydrologic science that is valuable to applications. Predictability science and the quest to define the limits-of-prediction may be directed towards the goal of gaining basic understanding of hydrologic systems. Alternatively predictability science may be directed towards the goal of improving operational predictions in the context of an application. There are circumstances possible where the two goals are aligned together and they cannot be separated. Nonetheless the two goals for predictability research are distinguished here because they often require different driving science questions, data, and metrics of progress. Progress in predictability research directed towards improving operational prediction is measured in terms of increased prediction accuracy or forecast skill. Such improvements may be achieved by empirical means that are useful in the application con-
OCR for page 9
Report of a Workshop on Predictability & Limits-to-Prediction in Hydrologic Systems text but do not necessarily provide added insight for guiding the research into the future generation of prediction systems. Similarly predictability research directed towards basic understanding may strive to understand a fundamental feature or define a new paradigm that promises to become the basis for an improved operational prediction system at a later stage. In this context the goal is less to improve the immediate forecast skill but to develop the foundations for a future system. In fact, in the short term there may be a drop in forecast skill as a new paradigm is introduced but the paradigm shift sets a new trajectory that promises great gains (see Figure 1.1). SUCCESS AND FAILURE IN APPLYING ADVANCES IN PREDICTABILITY SCIENCE Because policy decisions are often forward-looking, they are, in part, based on predictions. In the Workshop Roger Pielke, Jr. discussed the makings of an effective research and application program. Under this model there is a parallel undertaking of research and use of predictive models, linked by the communication between researchers and users of predictions. Success in prediction depends on the effective communication between these groups, leading to predictions which, 1) provide the most skill possible given the available data and modeling capabilities, 2) account for the stated needs of prediction users, and 3) have effectively communicated and understood uncertainties. Because the linkages among research, prediction and use involves a series of interacting processes, characterizing success or failure is difficult. Absolute success in the cycle occurs when a skillful prediction is communicated efficiently, and used effectively in formulating a decision that has a value to society. Users communicate their needs to predictors, and the predictors communicate the predictions and the uncertainties associated with the prediction to the users. The users are then able to make decisions, which are beneficial to society. The 1997-1998 El Nino event in California, illustrates the effective interaction between the users and predictors. During this period the users and the predictors communicated the importance of identifying the timing and
OCR for page 10
Report of a Workshop on Predictability & Limits-to-Prediction in Hydrologic Systems FIGURE 1.1: Conceptual figure demonstrating how advances in predictability science transition to improved operational predictions. The effectiveness of the predictions is measured with a skill score. Prediction systems are based on existing paradigms or scientific understanding. Initially the system has a slow rate of increase in skill. Errors in implementation, uneven completion of auxiliary systems, and gradual training of personnel in the prediction system some of the reasons why the initial increase in prediction skill can be modest or even negative. As the prediction system matures it undergoes a period of rapid improvement in its effectiveness. As the prediction system and its supporting science paradigm mature, the system again experiences slower rates of skill increase with time. In this phase the prediction system has essentially reached it highest potential for characterizing and predicting hydrologic phenomena. potential impacts of the anticipated El Nino. Through research programs, assembled data and established models, researchers
OCR for page 11
Report of a Workshop on Predictability & Limits-to-Prediction in Hydrologic Systems were searchers were able to predict the upcoming El Nino and were also able to determine that it would cause heavier than normal precipitation in California. This prediction was ably communicated to decision makers, who were involved in the process throughout the prediction. Because of the interaction between researchers and users, effective decisions were made and communicated at a local, regional and national scale. As a result, communities were able to prepare for and, in some cases, mitigate the potential for disaster. Although the flooding did cause significant damage, the interaction and communication between the policy makers and the researchers is believed to have limited the damage from this event. It is often more difficult to characterize failure in the system. Effective decisions may be made despite poor prediction and skillful predictions do not always result in decisions with a value to society. This breakdown is often a result of poor communication between predictors and policy makers. Catastrophic flood events exact a high toll in lives and property every year. For example the 1997 Red River Flood provides can be used to illustrate the breakdown in the cycle, where failed communication between predictors and policy makers resulted in a disastrous outcome. During this event, predictions were made for record-river flooding to take place. Forecasters gave a deterministic forecast with a value of 49 feet for the flood crest, which was slightly greater than the previous record crest for the river. This prediction did not effectively convey the associated uncertainties, rather indicated a fixed value for the flood crest, which was only marginally larger than the previous high water mark. The river crested at a height of 54 feet, resulting in extensive damage to the unprepared communities, which had prepared for river crests of no greater than 49 feet. (It should be noted that predicting a new record crest level is a remarkable achievement in operational prediction.) Although many elements of the hydrologic system offer varying degrees of predictability, it is crucial, as evidenced from the above examples of success and failure, that the uncertainties and the limits-to-prediction be established and well communicated to user groups.
Representative terms from entire chapter: