Evaluating the US National Digital Forecast Database for Use as Virtual Weather Stations to Drive the WSU-Decision Aid System

摘要

We evaluated whether the gridded forecasts available from the US National Digital Forecast Database (NDFD); of the US National Oceanic and Atmospheric Administration, could be used as virtual weather stations to run models for insect pests of deciduous fruit, in Washington State. Observed maximum and minimum temperatures, from 97 weather stations maintained by Washington State University's AgWeather Net and the paired NDFD locations, showed that error rates using the raw NDFD data were unacceptably high, with only 40% of the stations having an error rate <3 days. Error rates of the models, over the course of a season, also generally increased for five of the seven models evaluated, which showed the errors were cumulative. We also explored the calibration of the NDFD using a linear regression of the daily heat units observed and predicted for each station. We found that the regressions were stable for most locations over time and greatly increased the accuracy of the NDFD data, so that across all models, only 20% of the stations had error rates >3 days. Using the calibration also removed seasonal trends in all the models. Our data show that while the NDFD data can be used as virtual weather stations with a suitable calibration, error rates at somestations are still unacceptably high for pest management purposes.