Close Window

San Jose State University (Scott Archer, COMET graduate student fellow): "An intercomparison of QPF predictions by three mesoscale models"

Final Report

(Excerpted from Report)


The National Weather Service Forecast Office (NWSFO) in Monterey, California utilizes model data from the National Center for Environmental Prediction in producing quantitative precipitation forecasts (QPF). The Eta suite is utilized most frequently in making QPF's at the NWSFO in Monterey because of its relatively high resolution and decent microphysical package as compared to other operational models. This study addressed the question of whether increased horizontal resolution improves model QPF. Model precipitation biases were calculated for the winter 1996-1997 and the January 1997 flood.

Model precipitation bias show that increased horizontal resolution improved model QPF, with the 29-km mesoscale Eta showing less of a bias than the 48-km Eta. Both models under predicted precipitation at all thresholds considered. Precipitation bias increased with increased threshold, suggesting that models are not capable of accurately predicting heavier precipitation. Increased horizontal resolution improved the spatial distribution of precipitation over the complex topography of California.


This study was conducted in an attempt to answer the question of "how does improved model resolution/topography affect California QPF's". It is clear that at the eight locations studied for the winter QPF bias, that the higher resolution 29-km Eta model had improved QPF'S. In this case, the 29-km Eta under predicted precipitation less often that the 48 km Eta at all thresholds (0.1-3.0"). Where terrain resolution is poor for both models, such as the central California coast, both models performed poorly, with the 29-km and 48 km showing nearly the same bias.

The effects of highly orographic precipitation events are clearly seen for the January 1997 flood event in northern and central California. In this case, all models (Eta interpolated to 20, 40, 80 km, the 12 km MM5, and the Rhea orographic model) under predicted rainfall in most mountainous locations that favor orographic enhancement. For the Eta suite this was likely due to the poor resolution of topography, since the precipitation fell over complex terrain above 1800 m. The MM5 model performed relatively well, except for the placement and timing of precipitation.

It is important to note that in areas of complex terrain, vertical velocities can exceed 1 m/sec, while vertical velocities in the models rarely exceed 0.25 m/sec due to resolution restrictions. Zhao (1997) states that the microphysical packages available are improving rainfall predictions. Given they can resolve topography and vertical motions forced by topography, even more improvement is expected. Most West Coast heavy precipitation events are non-convective; therefore convection contributes little to the overall precipitation totals.

Although model initialization was not considered in the study, it is clear that much of the QPF error seen in models is due to the poor placement and timing of fronts. The January 1997 case shows how the southerly displacement of the front from the MM5 model leads to over prediction of rainfall over the southern Sierra Nevada, and the under prediction of rainfall over the Northern and Central Sierra Nevada. With improved initialization, models should be able to forecast the placement, intensity, and timing of fronts over California, thus improving QPF'S.

From this study, it can be assumed that increased model resolution so that dx,dy £ 10 km is needed to capture the very heavy rainfall totals observed during flood events in California. With California's complex topography, including coastal mountains/valleys, the Sierra Nevada, and the Shasta/Siskiyous, increased model resolution is needed to properly model the location and amount of precipitation. In addition, the current suite of models is not capable of modeling the intense and narrow cold fronts that are operationally observed at the NWSFO Monterey, moving southeast over California (although the 12-km MM5 has shown hope). These cold fronts are capable of producing very heavy rainfall in a short period of time, resulting in flooding.

One would expect that with an increase in model resolution (i.e. decreased horizontal grid spacing and improved topography) that model grid scale/stable QPF would improve. This study compared the QPF for the 29-km mesoscale Eta and the 48-km Eta model, which are the same model except for improved resolution and the running time of 03/15Z vs. 00/12Z. Bias statistics determined for the winter of 1996-1997 (November through January) at eight locations around northern and central California. The bias statistics are capable of showing whether a model over or under predicted precipitation at a single location. Model data is interpolated from the nearest grid point to the latitude/longitude of a station. In addition to the bias for a three-month period, statistics were determined for the flood of early January 1997.

Based on data collected from this study, the 29-km mesoscale Eta clearly produces an improved QPF over the 48-km Eta model during the three-month winter period. Bias scores for seven of the eight locations clearly show that the 48-km Eta under predicted rainfall more often and to a larger degree than did the 29-km Eta. The degree of error was reduced over the northern coastal areas of California and the Santa Lucia range of Big Usr. The largest errors were seen in the orographically favored locations of the northern and central Sierra, where typical rainfall amounts are 2-3 times greater than valley locations.

Recent studies at NCEP (Gartner et al 1996) show that both the 48 km Eta and 29-km Mesoscale Eta have a tendency to under-predict precipitation. Model bias in this study was found to be similar to those of Gartner et al.

The main purpose of this study was to determine if improved model resolution could improve model QPF'S. From the simple bias statistics determined over much of Northern and Central California, it was determined that improved model resolution leads to improved model QPF. Therefore, forecasters should utilize the higher resolution Eta products such as QPF products from the 29-km Mesoscale Eta, rather than those of the 48-km Eta.