Observation impacts have been estimated with an ensemble-based approach

It would be very useful if we could estimate the impact of each observation on the weather forecasts. Knowing the impacts of observations, we can optimize the cost-benefit balance of expensive observations and plan future observations wisely.

To estimate observation impacts, we usually perform data-denial experiments. Data-denial experiments are also known as Observing System Experiments (OSE), in which we perform additional experiments with/without our interested observation data. This way, we can estimate how much those observations help improve the weather forecasts, or possibly, degrade the forecasts.

Langland and Baker (2004) developed an adjoint-based approach to estimating observation impacts on forecasts without performing expensive data-denial experiments. Further, Liu and Kalnay (2008) proposed an equivalent approach without using an adjoint model. Instead, they proposed using an ensemble of model states. We call this the ensemble-based approach here.

Masaru Kunii, then visiting researcher in my lab, worked very hard on applying the ensemble-based approach to the real-case observations of Typhoon Sinlaku (2008). This is the first study showing the real observation impacts estimated by the ensemble-based approach. What is shown above is the impacts of atmospheric profile observations from so-called "dropsondes", weather instruments dropped from aircraft. This aircraft is operated by Taiwan, known as "DOTSTAR", which flies when strong typhoons approach Taiwan. The figure shows that those aircraft observations from DOTSTAR flights help improve the forecasts, although 5 out of 18 observations show negative impacts. We found that not using those negative-impact observations actually improved the forecasts. However, we need to be careful when interpreting the results from this method. The method estimates the impacts of observations in the given numerical weather prediction (NWP) system. These negative-impact observations may help improve the forecasts in other NWP systems. Nevertheless, we find a large potential in that we have this efficient method reasonably evaluate the value of observations in a given NWP system. Since our first trial of estimating real observation impacts was successful, we are excited about further studies to draw the full potential of the method.

These results have been published in the following literature. Please refer to the literature for more details.

Kunii, M., T. Miyoshi, and E. Kalnay, 2012: Estimating impact of real observations in regional numerical weather prediction using an ensemble Kalman filter. Mon. Wea. Rev., 140, 1975-1987. doi:10.1175/MWR-D-11-00205.1