EATHER is fickle, especially in the varied terrains and microclimates of the western United States. California and the other western states thrive or languish with their water supply, as the pendulum swings between drought and deluge. All too often, "average" precipitation is merely an artifact of arithmetic. Complicating the picture is the fact that the area receives its year's supply of water during the winter, and water for the dry summer must come almost entirely from reservoir storage and mountain snowpacks. At the start of each winter, everyone-water district official, fire fighter, ski resort operator, homeowner-wants to know if rainfall and snowfall will be above or below average. Accurate assessments of wintertime precipitation are particularly important for regional water management agencies as they attempt to manage reservoir capacity and balance the water demands of agricultural, industrial, urban, recreational, and environmental interests. In addition, since water supply is a limiting factor for urban and industrial development, regional planners are increasingly concerned about the effect of global climate change on local water resources.

























Numerical simulation using general circulation models (GCMs) is one of the most important tools for understanding global climate and for projecting long-term climate change. Great strides have been made in recent years to couple models of atmospheric, terrestrial, and oceanic processes to provide more complete climate simulations. However, because of their coarse spatial resolution (typically 100 km), it is difficult to apply these GCMs directly to regional forecasts. In California, for example, precipitation is closely related to topographic features (e.g., the Coastal Range, the San Francisco Bay) with spatial scales of less than 100 km, too small to be resolved by a GCM. Increasing the resolution of the GCMs to provide regional simulations is beyond the capabilities of present and envisioned computational resources.

Mesoscale models, nested within GCMs, are being developed to assess regional climate. As part of an effort to investigate regional-scale atmospheric flow, precipitation, and hydrology over various time scales and spatial resolutions, four LLNL researchers-Jinwon Kim, Norman Miller, Donald Ermak, and William Dannevik-have developed the Coupled Atmosphere-Riverflow Simulation (CARS) system. The system consists of three unidirectionally coupled models-MAS, LAS, and TOPMODEL (see Figure 1). CARS can be nested either within large-scale weather forecasts to predict regional weather and river flow or within global climate analysis data to assess regional climate and long-term water resources.


The Mesoscale Atmospheric Simulation (MAS) model was developed jointly by LLNL and the University of California at Davis. (Reference 1) It models atmospheric processes, including those involved in storms, from which it computes local precipitation, wind velocity, and other atmospheric variables. MAS computes rainfall and snowfall separately (using a bulk cloud microphysics scheme, Reference 2), an important capability because mountain snowpacks are major sources of summertime water for the western states.

The Land Analysis System (LAS) is a system of codes taken in part from software developed by the U.S. Geological Survey (reference 3) and combined with numerous other codes and scripts developed at LLNL. It provides land surface characteristics (such as flow directions, topographic slopes, water channels, and hydrological characteristics) for individual watersheds, based on digital elevation data provided by LLNL's Atmospheric Release Advisory Capability group. The areas and locations of the LAS watersheds are nested within the grid points of the MAS model.

TOPMODEL is a hydrology model, developed originally in 1979 at Lancaster University, England, (Reference 4) and enhanced and expanded over the years. LLNL's version of TOPMODEL takes the watershed-averaged precipitation and atmospheric variables from MAS together with the land surface characteristics from LAS to simulate surface and subsurface hydrology and river flow for individual watersheds.

The series of storms that struck Northern California in January 1995 provided an effective test of CARS's ability to make accurate short-term forecasts of precipitation and flooding. Between January 7 and January 11, three strong storms hit California. Several areas experienced extensive flooding as soils became saturated after the second and third storms came ashore. The Russian River basin was among the hardest-hit areas, with an estimated $800 million in flood-related damage.

Large-scale forecast data (80-km resolution) from the National Weather Service were used as input to the CARS system, and MAS simulations (20-km resolution) were run producing precipitation fields for all of California for this time period. MAS's ability to calculate rainfall and snowfall separately was essential for predictions of river flow, since snowfall does not immediately affect river flow.

California's complex terrain can cause considerable differences in the precipitation received by areas only a few miles apart. As a result, accurate estimates of local precipitation are essential for accurate estimates of river flow in mountainous areas. To illustrate this dependence, CARS computations were made for the area-averaged daily rainfall for the entire Russian River basin (approximately 7000 km2) and compared with calculations of the Hopland watershed (a smaller area, about 660 km2) within the Russian River basin, north of the Hopland gauge station. The simulated daily rainfall for the two areas differs by factors of two to three (Figure 2a).

To evaluate CARS's ability to predict river flow and flooding, simulated precipitation values for the Hopland watershed were compared with the observed precipitation values for the first 12 days of January (which were used by the National Weather Service's California-Nevada River Forecast Center to model river flow). CARS successfully simulated the amounts and timing of rainfall over the Hopland watershed, except on January 10, where the model overestimated precipitation by a factor of two (Figure 2b). Upon further examination, this overestimation was found to have resulted from excessive amounts of water vapor flux in the input data for the CARS simulation, clearly demonstrating the dependence of regional predictions on accurate large-scale data.

Figure 2c plots the observed and simulated daily-mean river flow volume of the Russian River at the Hopland gauge station from January 1 through January 12. CARS simulated the river flow rate to within 10% accuracy during the flood stage. The overestimation of modeled river flow for January 11 was due in part to the overpredicted rainfall for January 10, as noted above. For the low flow periods before flooding, simulated river flow exceeded the observed river flow mainly because of uncertainties in the initial water content of the soils, a difficult variable to simulate.

These successful predictions of extreme precipitation and river flow demonstrate the applicability of the CARS system to short-term, local weather forecasting. Such modeling will not replace human weather forecasters; rather, modeling can provide another type of data to assist forecasters. As Jinwon Kim, one of CARS's developers, remarks, "The value of front-line forecasting is that the forecasters have the experience to interpret data from various sources. Our goal is to create a modeling system that can help improve the accuracy of a forecast and the time span for which it is valid."

Improving short-term weather forecasts is but one step toward the long-range goal of understanding and predicting global climate change and its regional impacts. Having successfully simulated the Russian River situation, CARS's developers are moving ahead on several fronts. CARS's hydrology simulation model is being extended to include other major river systems in California, specifically the inflow to Lake Shasta, the Feather River, and the American River. This expansion will make it possible to use CARS for simulating local weather and river flows over northern California's major watersheds.

In collaboration with the National Weather Service, the CARS system is being used for experimental weather prediction for the southwestern United States. Simulations are also being run to test CARS's ability to assess water resources over seasonal, multiyear, and decadal time scales, to model the effect of such global phenomena as El Nino on regional climate, and to determine the effects of pollutants such as carbon dioxide and aerosols on climate change.


References
1.


J. Kim and S.-T. Soong, "Simulation of a Precipitation Event in the Western United States," Proceedings of the 6th Conference on Climate Variations (January 1994), pp. 407-410. (UCRL-JC-114412)
2.


H.-R. Cho et al., "A Model of the Effect of Cumulus Clouds on the Redistribution and Transformation of Pollutants," Journal of Geographical Research 94 (ND10), 12,895-12,910 (1989).
3.


S. K. Jenson and J. O. Domingue, "Extracting Topographic Structure from Digital Elevation Data for Geographic Information System Analysis," Photogrammetric Engineering and Remote Sensing 54 (11), 1593-1600 (1988).
4.


K. J. Beven and M. J. Kirkby, "A Physically Based, Variable Contributing Area Model of Basin Hydrology," Hydrological Science Bulletin 24, 43-69 (1979).



For further information contact Jinwon Kim (510) 422-1848 (kim1@llnl.gov), Norman Miller (510) 423-1283 (norm@llnl.gov), Donald Ermak (510) 423-0146 (ermak1@llnl.gov), or William Dannevik (510) 422-3132 (dannevik1@llnl.gov).



Return to July 1995//Return to S&TR 1995//Return to S&TR Homepage