Minnesotans For Sustainability©
Sustainable Society: A society that balances the environment, other life forms, and human interactions over an indefinite time period.
A Plan for a New Science
Predictability of Variations in
Report to the USGCRP from the Water Cycle Study
From both scientific and practical points of view, predicting the components of the Earth's water cycle is central to improving our understanding of climate variability and change. As discussed in Chapter 2, water cycle variability and process interactions (in the world's oceans, global atmosphere, land surface biosphere, surface and groundwater zones, and through fluxes among these reservoirs) vary over a wide range of spatial and temporal scales (Figure 3.1).
Some of these processes manifest themselves at long time scales, as in the deep ocean and in groundwater, while others, as in atmospheric water vapor and surface moisture, are dominated by relatively fast fluctuations. As noted in Chapter 2, land surface processes act to link these slow and fast processes. A scientific challenge is to understand the scales at which interactions and feedbacks of these processes occur, isolating the slower modes of variability from the faster, for enhanced predictive ability at a wide range of scales.
Because the water cycle is intimately related to living organisms, transport of chemicals, and ecosystem dynamics, the problem of modeling these complex and nonlinear interactions necessarily involves meteorology, hydrology, climatology, geomorphology, chemistry, ecology, biology, forestry, water resources management, and socioeconomics.
Water cycle variations can be considered as occurring over one of three time scales. Different processes limit predictability in these distinct regimes. At the shortest time scale of days to several weeks, relatively fast fluctuations associated with day-to-day weather are dominant. Box 3.1 provides an example of how more advanced weather forecasts on this time scale could improve mitigation of infrastructure damage from large-scale floods. Water cycle variability at a given location depends on both the propagation of disturbances from elsewhere (e.g., through an atmospheric front or flood crest) and the development of disturbances due to instabilities (e.g., thunderstorms). The predictability of such fluctuations is largely determined by the state of the system just prior some disturbance.
To what degree the future state of a dynamical system can be estimated, based on complete knowledge of the initial state, is a measure of the system's predictability. Predictability is an inherent attribute of a system, not to be confused with our current ability to make forecasts. Accurate forecasts depend on how fully the initial state can be accurately characterized and the fidelity of the forecasting model (Figure 3.2). There is enough evidence to suggest that atmospheric circulation is a chaotic system and is therefore depends sensitively on the initial conditions; that is, small errors in specifying initial conditions will amplify as the prediction evolves, as illustrated in Figure 3.3.
Relative contributions of observations and models in resolving the structure of an atmospheric or hydrologic phenomenon as a function of the forecast lead time. The upper dashed curve represents the (unknown) theoretical limit to predictability. The three lower curves depict hypothetical contributions of observations and models and the sum of the two to the total predictability of the system. At the initial time (t = 0), observations provide most of the information on the phenomenon, aided by models that may contribute some information.
In this case, models may provide a first-guess field from a previous forecast or assimilate limited observations to provide a consistent analysis of the variables of interest. In this early part of the forecast period, the value of observations exceeds that of models in providing information; however, the contribution of observations at time t to prediction decays rapidly. With lead time, the information provided by the models may actually increase, because dynamic adjustments in the model bring components of the system into balance.
For example, numerical model predictions of precipitation in the 12 to 24 hour range are often superior to the forecasts in the 0 to 12 hour range because it takes time for realistic vertical motion and moisture fields to develop in the models. (Diagram adapted from Anthes, 1984).
Example of error growth in a chaotic dynamical system, illustrated by a series of weather maps showing the distribution of geopotential height (analogous to pressure) at 500 hPa (a mid-tropospheric level in the atmosphere). The panels on the left represent the initial state (top) and subsequent forecasts at three and six days' lead time made using the National Weather Service Medium Range Forecast model. The panels on the right are identical to the panels on the left except that a small perturbation was added to the initial state (top right) before the forecast model was run.
The two initial states are nearly indistinguishable in this presentation. After three days, the forecasts have diverged somewhat, but the general weather pattern they predict over the continental United States is quite similar. By day six, the forecasts have diverged so much that quite different weather forecasts would be issued based on the model guidance from one or the other of these two runs. The two forecasts illustrate both the growth rate and the amplitude to which small errors can grow in a chaotic fluid such as the Earth's atmosphere.
Climate predictions several seasons in advance of the present can help narrow the uncertainty in hydrologic predictions for better water resources management, natural hazard mitigation, and related decision making and policy guidance. There is evidence that the low-frequency predictable components of water cycle variability (e.g., seasonal variations) that are superposed on higher frequency chaotic fluctuations (e.g., daily variations) can be isolated and predicted (Shukla, 1998).
The value of seasonal predictions was seen in the successful forecast of the El Niño event of 1997-1998. The combination of improvements in the global observing system (primarily the deployment of moored buoys in the tropical Pacific Ocean), in the understanding of coupled ocean-atmosphere dynamics responsible for ENSO, and in global climate models together led several research groups around the world to predict tropical Pacific sea surface temperature. The predictions were used in a number of regions in several sectors for the first time to provide socioeconomic benefits (Box 3.2).
Successful modeling of hydrologic processes and the global water cycle on all three time scales require detailed representations of the physical processes controlling water and energy fluxes. Such modeling also requires higher spatial resolution by one to two orders of magnitude than that currently used in dynamical weather forecasting models and climate change simulation models. The predictability of global water cycle components is typically quantified in the context of models that embody both theoretical understanding of the relevant processes and representations of all available observations. Thus, formulating an effective modeling strategy that encompasses the diverse temporal and spatial scales is a high priority.
Global and regional climate models capture the fluid dynamic and physical processes that cause large-scale (100 to 10,000 km) fluctuations in the atmosphere and oceans. They are typically expressed as numerical approximations to the continuous equations of motion that describe atmospheric and oceanic flows. Such models were first devised to predict weather, and only later applied to simulate the general circulation of atmosphere and oceans that is driven by radiative forcing at the top of the atmosphere and by the Earth's rotation. Water is a critically important component of this general circulation. Its vapor phase is the most active absorber/emitter of radiation, thus warming the Earth's surface by the so-called greenhouse effect.
In liquid and solid phases, water forms clouds in the atmosphere and snow and ice at the Earth's surface that are the principal determinants of the planetary albedo. Finally, water vapor that is transported by atmospheric winds and liquid and solid water flows over the surface influence the distribution of vegetation on the land surface, and through weathering, the slow changes in the nature and structure of the land surface. General circulation models and regional climate models approximate each of the water cycle components.
In general, the smaller the scale of a phenomenon's natural variability, the cruder the representation of the phenomenon in the model. Climate models solve equations governing the fluid dynamics and thermodynamics of the atmosphere and oceans by stepping the numerical approximations to those equations forward in small increments. Thus, climate model output can, in principle, include all the state variables and fluxes relevant to the water cycle with fairly high temporal resolution (steps of tens of minutes).
Hydrologic models are generally formulated on the catchment scale. This spatial scale is appropriate for water resources management and natural hazard mitigation (Figure 3.4). However, uncertainties remain in describing key processes at this scale. Much of the hydrologic research in the past three decades has focused on water movement within hydrologic reservoirs, with minimal attention to interactions between these reservoirs, such as freshwater flux between the atmosphere and oceans, groundwater recharge and evapotranspiration between land and atmosphere, and stream-groundwater interactions near the land surface.
As a resource to evaluate and manage, water must be considered along a continuum encompassing all its reservoirs. A major gap in our knowledge of the water cycle is quantification and modeling of fluxes between reservoirs. Some of these fluxes (e.g., evaporation and transpiration) provide critical information and feedback to the climate models. This knowledge gap results in large uncertainties in predictions.
Decision making in managing water resources and ecosystems is a complex, multifaceted process involving many stakeholders. Such decision making involves the economics of supply and demand, regulatory actions of government authorities at various levels, and litigation of disputes among stakeholders. This process is iterative and dynamic. Multiple contractual, legislative, and judicial actions change the priorities and rights of interested individuals and groups as well as the water resources and ecosystems themselves.
The historical record and predictions of water cycle variations can be used in this process to provide quantitative information about the physical system. This information typically is provided in a probabilistic format, so the various stakeholders need to be able to use such information effectively. Finally, the decisions that are made result -- both intentionally and unintentionally -- in changes to the water cycle, both directly and indirectly.
The limitations of historical records for policy analyses are highlighted by the scientific and political debates about the adequacy of methods and data in analyzing extreme floods -- including the 1927 and 1993 Mississippi River floods, the 1986 and 1997 American River floods, the 1982-87 Great Salt Lake flood, and the 1992-present Devils Lake flood. Each such event appears to be dramatic in the context of the local historical record. However, the connections of such basin-scale events to the larger climate picture are now emerging. Both the Devils Lake and Great Salt Lake variations have been tied to persistent, hemispheric scale, atmospheric and oceanic circulation features.
The longer (1847-present) Great Salt Lake record allows for a considerably better characterization of the probabilities of different climate regimes and their ability to forecast the 1980s extreme, than the 1910-present record of Devils Lake does to forecast this lake's levels. For both lakes, anecdotal evidence of 19th century lake levels and prior sediment records provided qualitative inputs into flood control decisions. Similar climate connections were established for the American and Mississippi River floods. Significant advances have also been made in constructing paleo-flood bounds for a number of rivers. The time is now ripe to move from such demonstrations to a more comprehensive synthesis of joint space-time-flood-climate state properties, across continental North America.
Droughts are also natural hazards with large economic impacts, particularly on agriculture. In the semiarid southwest, water availability in times of drought is an important driver of water resources management. Drought characterization and forecasting across North America is limited by the historical data. Droughts are often multiyear events that make the short historical data base even more limited in its value for risk assessment. Examples are the 1930s and 1950s multiyear droughts that devastated the northern and southern Great Plains, respectively. Considerable progress has been made in recent years in using proxy indices of past drought, especially those derived from tree rings, to improve estimates of the return time of such extreme events.
Such work, however, has focused on climatic drought indices, and hence is not directly relevant to river system management, which would require hydrologic drought indices. Yet, with the proper choice of paleoclimatic records, it is clearly possible to develop useful estimates of important hydrologic. Streamflow reconstruction for the Upper Colorado River flow from tree rings has had an impact on renegotiation of the Colorado River Compact. As in the case of floods, there is an opportunity to develop and test low flow and climate reconstructions for drought using paleoclimatic and historical data sources, reconstructions that could significantly aid in water resources management. At a minimum, future water availability could be assessed on the basis of projected future demands calculated from population growth and the historical record of past drought conditions.
Traditionally, surface water and groundwater have been managed separately. However, interactions between these two reservoirs are forcing the integrated management of them. Increased groundwater production (pumping) generally results in reduced baseflow to streams and may ultimately result in capture of streamflow, which would have very harmful effects on ecosystems. Surface water and groundwater should be managed as a single resource to ensure sustainable development.
Most of the above-described physical models and processes in place to manage water resources and ecosystems have been developed to answer particular questions about the environment or about the human systems that use and control elements of the environment. The models differ among each other significantly in design and function, and typically, the output from the physical models is used as information input to decision-making models. This view that users of scientific information exist outside the process of developing that information provokes difficulties in the effective use and exchange of information. By considering physical and socioeconomic models together in a single systems modeling framework, it is possible to break the linear pathway from models to users as illustrated in Figure 3.6.
The systems modeling framework works effectively on daily, seasonal to interannual, and decadal to centennial time scales. For example, seasonal and longer lead predictions of tropical sea surface temperatures and their effects on climate variability in other parts of the globe have been provided to the public. However, different sectors have made use of this information with varying degrees of effectiveness. By incorporating the needs of users into the ways that predictions are made, the information can be made significantly more useful, even for regimes for which predictability is quite modest (e.g., Barnston et al., 1999).
Systems modeling framework for prediction and transfer of information to end users. Users of scientific information and environmental predictions should not be seen as outside the process. A systems modeling framework can explicitly provide a valuable feedback loop by stating societal needs in terms already identified by users through national and regional assessments, and by including social science perspectives on the value of information. In this way, the value of information that can be extracted from predictions of water cycle variations can be better evaluated.
In the systems modeling framework, all elements of the system, including the methods for observing and monitoring the natural environment, the models used to make predictions, and the risks and values associated with the predictive information can all be evaluated within a common framework. Notice that this systems modeling framework is generic and applies to all three time scales (daily, seasonal-to-interannual, and decadal-to-centennial). Typically, users require probabilistic predictions for risk-based decision making. To achieve these predictions, the model must run in an ensemble mode where initial conditions or model physics are perturbed to produce a suite of predictions and characterize their likelihood (and conversely, their level of uncertainty).
In this scheme, prognostic state variables of the physical processes (such as precipitation, air temperature, and surface and groundwater storage), are passed from one scale to another to estimate the basin response, for example, the flood hydrograph needed for water management decisions. Global, climate, and hydrologic models must also include uncertainty propagation from the climate models to the hydrologic models and the additional uncertainty generated from the hydrologic models (owing to uncertainty in model parameterizations and input data) to provide valuable information to the end users (decision makers, resource users, interest groups, economic models, policy scenario analysts, etc.).
Predictions from the climate-hydrologic models are more useful when accompanied by uncertainty estimates, provide better information for decision making and management of risk. Information produced by models (as shown in Figure 3.7) and transferred to users (as shown in Figure 3.5) should not only project "mean states," but also identify measures of uncertainty for the projected states. Uncertainty estimates are crucial inasmuch as vulnerability to climate change varies regionally and by sector, and depends to a large degree on the lead times at which decisions must be made and the time scales over which decisions have an effect.
The quantities that numerical weather and climate models correctly predict are not necessarily the quantities needed by decision makers for managing water resources or mitigating natural hazards. Such information gaps indicate a need to develop a framework in which model predictions can be related to decision-making requirements for the complex context of multiple parties, uses and values.
Nesting climate and hydrologic models. A coupled ocean-atmosphere model is used to compute global sea surface temperatures (SSTs), which are transferred as input, along with the land-surface state, to a global atmospheric model. The global model output drives a large-scale atmospheric circulation model, whose output serves as an input to a regional atmospheric dynamics model. This regional model computes surface temperature, precipitation, and other radiation components which, when spatially disaggregated, and upon specifying the land state again at basin resolution, provide the input for a basin-scale coupled hydrology-ecosystem model. The hydrologic model predictions along with risk statistics can provide input to a complex decision making process.
Goal 1: Demonstrate the degree of predictability of variations in the water cycle on three time scales -- daily, seasonal to interannual, and decadal to centennial -- regimes for which the limits of predictability differ.
Why? It is necessary to concentrate efforts on predictable components of the water cycle, because progress in this area will be the most cost-effective and fast. At the same time, less predictable or inherently unpredictable processes must be estimated and their limits of predictability assessed. Because predictability at daily, seasonal to interannual, and decadal to century time scales is limited by different processes, the predictability of water cycle components must be assessed for each time scale.
How? This goal will be accomplished by identifying predictable components of the water cycle at daily, seasonal to interannual, and decadal to centennial time scales, and all spatial scales; by assessing the limits of predictability for less predictable components; and by quantifying prediction uncertainties through a program of monitoring, process studies, and model development.
California's San Joaquin Valley exemplifies the tensions that now exist between irrigated agriculture and a growing social interest in ecosystem preservation and restoration. In the summer of 1999, after 10 years of litigation by environmental groups, the Bureau of Reclamation and the valley's agricultural community agreed to release 35,000 acre-feet of water from Friant Dam. The release was part of an experiment to test the feasibility of restoring a frequently dry part of the San Joaquin River between the reservoir and Mendota Pool.
Many observers predicted that a large fraction of the released water would quickly be lost to infiltration as it passed over the normally dry section of the river bed. However, most of it flowed through that section directly to Mendota Pool, where it became available for use by irrigators in the lower San Joaquin Valley. The success of the $2.5 million experiment was a surprise. Was it simply a fortuitous result of high groundwater levels, resulting from a number of years with normal to above-normal precipitation, or can such success be sustained under California's frequent excursions into drought conditions? Without a better understanding of water fluxes between the valley's surface and groundwater systems, we cannot answer that question well.
Evaporation. Evaporation from land surfaces serves as the primary feedback mechanism that links land cover to the water vapor and energy budgets within large-scale climate models. To understand basin-scale evaporation and transpiration requires vegetation observations at basin-scale (using remote sensing), models of vegetation that cover growth and plant succession, and (using theories of hydraulic limitations within plants, xylem, cavitation etc.) photosynthesis and biochemical assimilation of nutrients, and soil moisture limitations. Remote-sensing data should be complemented with point estimates of evapotranspiration measured with eddy correlation systems or Bowen ratio systems. Recent advances in these ground-based measuring systems have greatly increased their reliability and reduced their costs.
The ability of catchment-scale models to predict spatial and temporal variability of recharge depends on accurate partitioning of residual water (after runoff is calculated) between evapotranspiration and recharge. As water resources become more limited, particularly in the semiarid southwest, there may be increased emphasis on reducing evapotranspiration and enhancing recharge by replacing deep-rooted nonnative vegetation, such as creosote and mesquite, with shallow-rooted grasses.
Spatial Inhomogeneity. One important difficulty for the science is the pronounced spatial inhomogeneity of most variables involved -- precipitation, soil moisture, topography, land surface and other characteristics -- and the need to derive upscaling relationships for modeling purposes. Since the interactions of these processes are nonlinear, subgrid-scale variability in one variable can significantly affect the grid-average estimate of another variable. Thus, fluxes between water reservoirs must be studied at a range of space-time scales, and the relative roles of these processes and scales in improving water cycle predictions must be assessed.
The hypothetical transfer of information model is shown in Figures 3.6 and 3.7, in which forcing functions and state variable information cascade from climate models to socioeconomic models involving multiple decision makers using intermediate basin-scale hydrologic models. For this transfer to be successful, disaggregation strategies must be developed for downscaling large-scale climate model predictions to provide input to smaller scale hydrologic models. Precipitation, temperature, and radiation are generally provided at 100 km2 scale from climate models. These quantities must then be disaggregrated to provide forcing functions for hydrologic models as illustrated in Figure 3.7.
Uncertainties in these parameters necessarily propagate to the hydrologic models. When hydrologic models are used to forecast droughts, floods, and other hazards, such predictions are then uncertain because of the uncertainties in forcings generated by the climate models, with additional uncertainties inherent in the hydrologic models. Output from hydrologic models will provide probabilistic input to decision makers whose interactions determine the allocation of water use across various sectors (Figure 3.8). This information can then be used to assess the vulnerability of various sectors to climate variability. Information from this process may be valuable in improving the coordination of decision-making entities. For example, public agencies and other parties engaged in local watershed management initiatives could use this information to develop alternative management strategies to address climate variability.
Quantitative information on water resources and associated uncertainties is critical for planning and developing a policy framework to respond to and mitigate extreme hydrologic events. One of the challenges to successfully implementing a climate-hydrologic-economic model framework is clear coordination of the scientific requirements of the water management decision process and the predictive ability of the hydrologic models.
Figure 3-8 Schematic of how a Probabilistic Model Forecast
can be used for Risk-based Decision-making
Depending on a tolerance level that is user-specified and application-dependent and the sensitivity of decisions to that tolerance level and prediction uncertainty, valuable feedback might result for model verification (i.e., no need to demand from the model more than the user can make use of), model improvements (when the model predictions are not reliable enough to support a decision), and observational requirements (collection or assimilation of more observations to improve the reliability of predictions).
Efforts to predict water cycle components are particularly useful for forecasting extremes (e.g., droughts and floods) that may be sensitive to large-scale climate variability and local atmosphere-land-groundwater coupling. For this reason, an initiative must be directed toward improving the capabilities of physical (numerical) climate, hydrology, and water quality models. However, improvements in these physical models must be guided by end users' requirements, so the modeling framework must include a knowledge transfer element.
Determine optimal modeling strategies to handle multiple sources of uncertainty and multiple scales of natural variability. Uncertainty in predicting water cycle variations is introduced from the inherently chaotic character of the atmosphere and limitations in the process of making predictions, including undersampling, short record length, measurement error, analysis error, parameter value identification, and systematic errors in numerical models. These uncertainties arise at different spatial scales and different lead times for a given prediction.
A fundamental question in seasonal and longer lead prediction of water fluxes is the choice of a modeling strategy to minimize propagation of uncertainty among components of a predictive model. For example, predictability of regional hydrologic systems is limited by predictability of large-scale climate fluctuations when large-scale climate forces regional hydrology. If there is two-way land-atmosphere coupling or climate modulation by local hydrologic processes, there is potentially greater predictability that can be exploited through coupled modeling.
Given the relevant scales for predicting floods and droughts -- at global, continental, regional, and catchment scales -- two methodologies may be employed to model the relevant processes: nested models and high-resolution global models. The nested model approach most often used predicts each phenomenon with the model relevant to that scale and uses the output of larger scale models to drive smaller scale models. For example, the large-scale atmospheric seasonal mean atmospheric circulation that is in balance with the globally distributed SST anomalies is most appropriately modeled with a global atmospheric general circulation model. A regional model in which seasonal predictability is high can be nested in the global model to help predict regional details of the distribution of precipitation and temperature anomalies consistent with the large-scale moisture and energy fluxes in the global model.
Nested regional climate models help bridge the spatial scales of atmospheric, land-surface, and subsurface processes. Much weather and climate simulation research has recently been done with nested regional models; and several operational weather forecasting products are generated using regional models. Further nesting is needed of surface and subsurface hydrology models to simulate the runoff and streamflow from the precipitation predicted by regional models for many catchments where basin response is sensitive to small-scale inhomogeneities.
While the suitability of the nested model approach has been demonstrated, such models have well-known limitations; for example, they are not formulated to obey integral constraints associated with mass and energy conservation laws, which can limit their accuracy for climate simulation. In principle, such limitations could be removed by using a high-resolution global model strategy. The initiative proposed here is a systematic test of the two approaches for optimal prediction of seasonal water fluxes.
Integrated system models typically link atmospheric, land-surface, stream, and groundwater component models into a single modeling framework. These models are inherently nonlinear, because they include feedbacks among components that are themselves nonlinear. Further, these component systems exhibit variability at many time and space scales. The challenge is to link global coupled ocean-atmosphere models to integrated system models such that resulting predictions and uncertainty estimates that provide a useful basis for decision making in water resources management: such a result requires high spatial and temporal resolution, placing huge demands on computational resources and observational data to initialize and verify the models.
A systematic approach to model design and development
is therefore needed to help determine, based on basic research outcomes, the
scales at which predictive information should be exchanged within a nested
modeling approach, or whether a global integrated system model is required.
This research in modeling strategy will be heavily computational and will require enhancements to the nation's computing capabilities. In particular, a virtual distributed modeling software engineering infrastructure must be developed to facilitate model comparison. Also needed will be the interchange of modeling components and data sets among models, and adequate computing facilities distributed to participating modeling centers.
Additionally, new possibilities for developing distributed hydrologic models (Figure 3.9) are provided by the expanded analytic capabilities of geographical information systems, along with the availability on the web of information on input parameters, including elevation, vegetation type, soil type, land use land cover, river reaches, and hydrologic unit boundaries, at relatively fine spatial and temporal scales. Improvements in visualization enhance our ability to understand hydrologic systems and allow researchers to transfer model results more readily to the user community.
Figure 3-9 Example of
Available Data that can be used
to help Develop Distributed Hydrologic Models
Many kinds of information are readily available from the web for distributed hydrologic modeling, such as vegetation type (U.S. Forest Service), soil type (USDA), and land use/land cover (USGS).
Determine the parameter, accuracy, and sampling requirements through observing system simulation experiments at all spatial scales. A powerful method for determining gaps in our measurements to best improve our understanding of predictability is the so-called observing system simulation (OSS) methodology. In an OSS experiment, a model that faithfully represents some natural process is used to generate a long time series of states that serve as a proxy for nature. The model can then be "sampled" in the same way as a prospective observing system would sample nature, to determine whether the sampling strategy adequately captures the relevant variability modes in the natural process. The OSS methodology provides important information for predictability studies because it involves the use of a predictive model, a set of measurements that are either already operational or are prospective, and a data assimilation system that optimally combines observations and model output.
Systematically combining observational data and models will require a systematic expansion of four-dimensional data assimilation research and observing-system simulation experiments. At present, assimilation of atmospheric observations has been advanced to some degree, but assimilation of land surface observations has only recently begun. Assimilation of observations in hydrologic (surface and groundwater) models has yet to be attempted in a systematic way. To adequately assess the impact of new observing systems on the understanding of predictability and on the ability to predict at seasonal and longer lead times, it will be necessary to develop an OSS methodology to effectively combine hydrology models and observations. This approach can then be used to estimate the scales appropriate to couple or nest component models. New observing systems may also affect the ability of predictive models to make more accurate forecasts.
In addition to OSS experiments to help shape future observing systems, companion observing-system experiments (OS) are needed to assess prediction impacts, sampling and deployment strategy, and improved use of already existing observing systems. Both OSS and OS strategies have been formally defined by the North American Observing System Program (NAOS), established in 1995 to develop scientifically based procedures, centered on 4DDA, to determine optimal atmospheric observing-system strategies, including impacts on prediction model accuracy.
For these observing systems to be possible, methods will be required for optimally merging data from heterogeneous observing systems, such as weather radar, satellite and surface and upper air stations.
Make fundamental advances in mathematical theories of predictability, based on principles in probability, stochastic processes, nonlinear dynamics, and numerical methods. Mathematical theories are continually developed in other physical sciences, but they have not been systematically applied often to characterize water cycle components. A new program in the science and mathematics of water cycle predictability is needed to guide applications of atmospheric and hydrologic theories over a range of space and time scales. This program would forge collaboration and links among mathematicians, statisticians, hydrologists, and climate modelers to advance predictability in hydrology. Such a program can capitalize on the proposed long-term monitoring network (described below), as well as planned remote-sensing measurements and climate model outputs for wealth of measurement and model results on hydrologic processes.
Until recently, climate and hydrologic models were used for deterministic predictions. In other words, initial and boundary conditions have been applied to obtain a single realization of the weather or climate system, and to predict a single value. As is well known, however, deterministic predictions of, for example, the atmosphere's instantaneous state cannot be made with lead times longer than a few days because of chaotic fluctuations (Figure 3.3). Climate predictions on seasonal and longer time scales must be made within a probabilistic framework. Such predictions are currently made by using an ensemble of model predictions, each with slightly different initial or boundary conditions that are nearly indistinguishable from the actual environmental state given the available measurements. Research is required to place this ad hoc methodology on a firmer theoretical basis.
Promote knowledge transfer among scientists and end users. Earth system models may be able to predict global water cycle variations with some skill. For these predictions to be useful to water resources managers or other decision makes, however, the predictions must take the right form. Coarse-resolution predictions, for example, may need to be converted to the local watershed scale using empirical disaggregation techniques or the additional application of a higher resolution nested model. Uncertainties in the model predictions must also be communicated to the end users in an understandable and quantitatively useful way. The key to achieving this knowledge transfer, of course, is communication and coordination among the scientists and society's end users early in developing prediction systems. Modelers should design their end-to-end systems with the users in mind. Users, in turn, must be sufficiently versed in the limitations of the prediction systems and in the proper interpretations of model-generated data.
Toward this end, a prediction system framework must be developed to identify and quantify the characteristics (parameter, spatial and temporal resolution, lead time, and accuracy) of what can be observed, what can be predicted, and what can be controlled or managed. Users of water cycle prediction information must help define what aspects of water resources and ecosystems should be managed, and help establish the requirements for predictive information. The modelers must use this information in designing and realizing predictability experiments and operational prediction systems. The information exchange among them all must be iterated with new developments in observing systems, new predictive models, and new water resources and ecosystems management experience. The effective transfer of knowledge gained through this process must be enabled and strengthened.
Large-scale and basin-scale experiments are needed to improve models' representations of the relevant processes, estimate model parameters, validate model simulations and predictions, and evaluate fluxes among hydrologic reservoirs (as through evapotranspiration, recharge, and surface water -- groundwater interactions in watersheds with different types of land cover or major anthropogenic effects. Large-scale models must have greatly improved representations of cloud dynamics; microphysics of precipitation; interaction of cloud liquid water, water vapor, and radiation; transport of water vapor in the planetary boundary layer; and aggregation/disaggregation of surface hydrologic processes over complex landscapes. In land surface models, a number of processes are poorly represented. Model parameter estimation and model validation require the integrating of observations, process-resolving models, and large-scale (parameterized) models through data assimilation and adjoint techniques.
Biospheric responses are inadequately represented for both short- and long-term processes. For example, how vegetation in different biomes around the world will respond to increasing concentrations of atmospheric CO2 is poorly known; only a limited number of plant species have been studied for such applications. Further, the geographic distribution of different biomes is strongly controlled by climate, especially precipitation and temperature regimes. The manner in which boundaries between major biomes may change is unknown and certainly has not been quantitatively represented in models. The potential for such long-term change is clear in comparing representative climate diagrams for temperate grasslands, temperate forests, and boreal forests.
The southern boundaries of boreal forests are either temperate grasslands or temperate forests; and warmer conditions, which shorten the winter period and promote forest fires, could allow northward expansion of temperate grasslands and forests. These transitions in biome distribution may limit the predictability of biospheric responses to atmospheric processes over longer time scales, and should be considered as observations are integrated with process-resolving models and large-scale parameterized models.
The experiments for evaluating fluxes should help develop techniques for quantifying fluxes between reservoirs at the catchment scale. Such fluxes include evapotranspiration and recharge. The experiments should incorporate both remote sensing and in situ measurements. A major effort should be made to downscale remote-sensing data and to upscale point measurements of fluxes to a common scale where process comparison can be made. Water balance modeling may be used to provide initial estimates of evapotranspiration, recharge, and surface water -- groundwater interactions.
Remote-sensing information on vegetation dynamics can be used as input to these models. Environmental tracers can provide integrated measures of recharge over large spatial and temporal scales and may also provide valuable insights into flow mechanisms. Existing numerical models that incorporate surface water -- groundwater interactions are extremely limited. New models must be developed based on quantitative understanding of interactions between these reservoirs, as determined from detailed field experiments.
It is critical to develop a monitoring network to determine initial states for model predictions, verify model forecasts, and validate model components. Advances in measurement techniques now allow land surface properties (e.g., topography, vegetation type and state, and snow depth and moisture content) to be determined at significantly finer scales than was possible only a few years ago. A high priority is exploiting these advances to verify and validate predictive models, and to determine the scales at which measurements are required to simulate hydrologic variability realistically and accurately. Beyond developing new monitoring sites, the proposed monitoring program should take full advantage of existing sites, such as the NSF Long-Term Ecological Research sites; USDA experimental watersheds; USGS Water, Energy and Biogeochemical Budget research watersheds; and the AmeriFlux network.
They should be augmented with an intensive array of soil moisture sensors, precipitation and interception gauges, a network of groundwater monitoring wells, and frequent remote-sensing measurements, particularly using NASA's new canopy lidar. The addition of soil moisture instruments to existing eddy-covariance instruments will help quantify soil moisture content controls for transpiration modeling and for feedback to climate models. Long-term experimental sites for characterizing water balance and flow pathways provide the data essential for developing models and methods to scale hydrologic variables, characterize basin-scale variability, and understand the limits of predictability. The monitoring program covers several components.
Precipitation. Assessing precipitation over the entire globe at sufficiently high resolution will capture its diurnal variability and spatial inhomogeneities to improve understanding of water cycle exchanges and predictions at all scales. The Global Precipitation Mission, according to currently planned specifications, promises 3-hourly, global, 4-kilometer precipitation coverage, and could be the cornerstone of the needed effort. Accurate snow cover mapping by integrating satellite and ground-based networks is also a high priority.
Ocean Fluxes. A global ocean surface flux monitoring program is needed to obtain, for the first time, an estimate of the freshwater flux from atmosphere to ocean (precipitation) and from the oceans to the atmosphere (evaporation). These aspects of the global water cycle remain a major source of uncertainty in attempting to characterize the fluxes of water between reservoirs and to predict fluctuations in these fluxes at seasonal time scales. Changes in the Earth's two major ice sheets represent the largest source of water for interannual- to millennial-scale changes in ocean circulation. Accurate measurement programs are needed to estimate interannual variability and long-term changes in these ices sheet masses.
Fluxes Among Reservoirs. Another requirement is the monitoring of hydrologic components to quantify fluxes among atmospheric, surface, and subsurface reservoirs. Monitoring efforts should cover a spectrum of land cover variability, with times ranging from stable to rapidly changing conditions, using in situ instrumentation and remote sensing. Soil moisture is clearly a parameter of great importance in characterizing the water cycle variations. Monitoring parameters should include surface water fluxes (stream gauges), water content, pressure, and temperature in the unsaturated zone, and water levels and temperatures in the saturated zone.
Geophysical measurements using electromagnetic induction or ground-penetrating radar should be used to interpolate and extrapolate information for locations between monitoring locations at a site. Data on environmental tracers, such as chloride, tritium, tritium/helium, and chlorofluorocarbons, should be measured in the unsaturated or saturated zones to date water for recharge estimation and to evaluate flow mechanisms (piston vs preferential flow). A national network of groundwater monitoring wells to assess both water level and water chemistry is essential for groundwater recharge characterization and for identifying long-term trends resulting from pumping, drought, and land use change.
A current barrier to effectively integrating the output from large-scale climate models in hydrology and water resource management is the lack of active collaboration among scientists in these various disciplines and stakeholders. An intellectual forum and creative funding arrangements are needed to assure opportunities for cross-disciplinary collaboration on important water resources and environmental management issues that are sensitive to climate variability. One essential task of this intellectual forum would be identifying basins that are particularly vulnerable to climate variability and developing climate and hydrologic databases including environmental and socioeconomic variables that would be useful in assessing vulnerability and in making decisions.
To understand how uncertainties propagate through these systems we need to consider both uncertainties in climate variability and uncertainties in human impacts and evaluate the feedback between these elements. Output from such analyses can be used to evaluate policy options and assess their implications for mitigating the potential adverse impacts of climate variability and anthropogenic disturbances. Basins in which local watershed forums have been established may provide a useful focus for such efforts, because these communities may be vulnerable to climate variability and, in any event, especially receptive to the use of improved forecast information.
Practical use of a particular forecast is a more involved problem than simply assessing the quality or accuracy of the forecast. The forecast and its uncertainty must be integrated with site- and case-specific information (e.g., flood stage vs damage curves, shear stress vs aircraft safety curves, etc.) to compute a "risk of failure" that can guide decision making. Methods for translating an uncertain forecast (with forecast here meaning a suite of multidimensional variables) to products useful for decision making are in their infancy and much research is needed to advance them. This area must be developed in parallel with the efforts to improve forecast quality -- even forecasts with large uncertainties can provide useful guidance for water management decisions.
Another current barrier to effective use of hydro-meteorological information in water resources decision making is the difficulty of achieving coordinated, rational water use and management decisions in the presence of divided authority. In most basins, a large number of individuals and public agencies, at different levels of government, exercise independent authority over some aspect of the water resource. The resulting myriad lines of communication, overlapping jurisdictional boundaries, differing legal constraints, and multiple decisions made or influenced by parties with strongly divergent interests can result in suboptimal resource allocation and ineffective overall risk management. Efforts to overcome such problems have gathered momentum over the past decade, for example, with the recent proliferation of local watershed management initiatives. These efforts, which are aimed at finding pragmatic solutions to local problems, are likely to benefit from improved access to information on the implications of water cycle variability for local resource management concerns.
The rationale for introducing ensemble forecast methods to the water resources management community is a strong one. Yet despite their clear advantages, ensemble forecasts of weather and climate water cycle variations have not penetrated the water resources management community. The applications initiative should assist water managers in using ensemble forecast products in their operation of water resource systems. The primary focus should be on reservoir systems (or, in some cases, free-flowing rivers). At the same time, implications for groundwater in systems that conjunctively use surface and groundwater should be recognized. The necessary technology transfer might be accomplished through a cooperative applications program such as the program proposed in Chapter 2 for development of improved design protocols incorporating climate information.
The applications initiative would be funded and managed by one or more of the science agencies. This agency would solicit proposals for demonstration applications of ensemble forecast methods to water management problems. These solicitations would include a mechanism for funding of dedicated personnel who would work in an operational setting, but they would also require evidence of the participation of a credible science-based organization. This arrangement might be accomplished through a fellowship system that assigned personnel on a part-time rotating basis to a government research laboratory or university, and part-time to a government agency or university. Applications should go well beyond the water resources management community, to embrace other regional economic development and resource management decisions where climate and hydrologic information can play an important role.
1717 Pennsylvania Ave, NW, Suite 250, Washington, DC 20006.
See original at < http://www.usgcrp.gov/usgcrp/Library/watercycle/wcsgreport2001/wcsg2001chapter3.htm >.
Please send mail to
firstname.lastname@example.org with questions or comments about this web site. Minnesotans For Sustainability
(MFS) is not
affiliated with any government body, private, or corporate entity.
Copyright © 2002, 2003, 2004
Minnesotans For Sustainability