Water use and availability are critical determining factors for economic and social conditions in the Pacific Northwest. Steady population growth in the Northwest implies future shortages and competition for water resources that will require a sophisticated water management system. The NWS may be called upon to play an increasingly important role in forecasting water availability on time scales from a few hours to a season. The proper management of water resources requires a great deal of information, most of which is currently based on data from rain gauges distributed unevenly within the area. These gauges are of many different types and are maintained and operated by several different organizations. For a variety of reasons, the quality and reliability of the data vary greatly. In the case of river and avalanche forecasting, the collection and processing of data is extremely time-critical. Accordingly, some form of real-time quality control is required.
The purpose of this project was to lay the groundwork and prove the functionality of a real time QC algorithm which would address the following issues:
1) Computer resources are a constraint on methodologies used, both in processing time and data transmission
2) The process must occur in real time
3) The algorithm must be able to function if data from stations normally used in the procedure are missing. It should also be able to incorporate extra data, as they become available
4) Screening methods must be able to detect the difference between extreme events and erroneous events with a great deal of dependability. It is mostly extreme events that we are interested in; erroneously flagging a correct extreme event as an error could be disastrous
5) The screening method must be able to incorporate data from different sensor types and time frames, and with differing historical record periods
6) The algorithm should work in a modular fashion with the network of data transmission and software packages already in use
The method developed under this project uses statistical probabilities, operates in real time, and offers no correction of suspect values. It is essentially an expert system which uses climatic data as the knowledge base, and a system of normalizing the data in the working memory to make an estimate in the inference engine about the quality of the data, based on climatically-derived probabilities of exceedence. The data are labeled with various quality indicators (i.e., “verified”, “questionable”, “rejected”) for later scrutiny by a forecaster.
The prototype work conducted under this project concentrated on rainfall data. However, the same scheme may be applied to any normally distributed data field.
The first part of the research was a test of Oregon State University's one-dimensional planetary boundary layer model as a "what-if" forecasting tool in the Portland NWSFO and as a possible candidate for improving the early-fall soil moisture estimates for use as initial conditions by the NWS River Forecast Center. The model simulates the physical properties of the atmosphere, soil, and vegetated surfaces from given large-scale conditions and has the potential to provide physically-based guidance to the forecaster on the detailed effects of interactions between the atmosphere, soil, and vegetated surface. The model has been modified to allow the user to change the input for sensitivity experiments in numerical weather forecasting. Continuation of operational testing of the model has been suspended, due to the transfer of the NWS employee involved in the project.
During the second part of the research, a comprehensive data survey of weather stations operating within or near Oregon was conducted. Various data sources were identified, and the focus of the project centered on the development of improved methods for determining the validity of real-time weather data, particularly precipitation information. A method developed by NWS Northwest River Forecast Center was used to transform historical precipitation data to obtain probabilities for extreme 1-day precipitation events. Data from the NWS cooperative network were used to develop the coefficients for the transform, and data from other networks have been, or will be collected, to augment the cooperative station data set. The derived coefficients will form the basis for a screening scheme to be used for identifying potential outliers in real-time precipitation data, along with an isohyetal analysis for the state.