Meta-model Weather Forecasting: Integrating Public and Private Weather Forecasts

Abstract:

Accurate weather forecasting is essential for the success of many weather-sensitive sectors, such as water resources, energy, and agriculture. In addition to public weather forecasts provided by government agencies, several private weather forecast companies provide such a service at a global scale. Here, we evaluate real-time daily weather forecasts of near-surface air temperature (also known as reference temperature or “tref”) and precipitation across various locations around the globe. We compare forecasts from 6 private weather companies and 3 public-sector agencies for a continuous period of 3 months (due to access limitations to private weather forecasts) at lead times of 1 to 4 days. Additionally, we acquire a multitude of gauge station observations across 16 countries in different parts of the world to implement the analyses. We find that private weather forecasts are considerably more accurate than the public forecasts. Nonetheless, there is a lot of variance in the skill and uncertainty of private companies (spatially, temporally, and for variables of interest). Therefore, we develop a meta-model ensemble approach to integrate all the forecasts and provide location-specific weather forecasts. Our results indicate that despite data limitations, our meta-model approach is capable of improving the current forecast skills by 15% (on average) for temperature. We also assess the accuracy of weather forecasts for predicting extreme weather (near-freezing air temperatures) which can substantially impact the agriculture sector. Notably, we illustrate that efficient post-processing considerably improves regional weather forecast skill at a farm-level and beyond.