Crowdsourced wind observations can be an invaluable data resource for meteorological studies

02 September 2021

Can citizen scientists help KNMI to improve forecasts and warnings for extreme wind events? The answer is yes according to the new paper by Jieyu Chen, Dr. Kirien Whan, and Dr. Kate Saunders, published in the Quarterly Journal of the Royal Meteorological Society. The research shows that by checking the quality of the data and adjusting for biases there is value in crowdsourced data and that it can be used to complement official observations.

The growing popularity of citizen weather stations enables real-time weather observations at many more locations to be possible. However, cheap devices, non-standard equipment settings and poor placement prevent the practical use of such crowdsourced data. A new study published in the Quarterly Journal of the Royal Meteorological Institute, showed that the quality of wind speed observations from citizen weather stations from the WOW-NL network could be assured by implementing quality control procedures. The method comprises two parts, one is to detect suspect observations, and the other is to adjust the biases.

Compared to precipitation, wind doesn’t vary as much in time and in space, and there is a plausible range for wind speed depending on the location. Suspicious wind speed observations are detected using multiple quality control checks (Figure 1). The researchers first check the range of the crowdsourced observations to remove implausible values such as negative wind speed. Next, they check the temporal consistency so that observations that change too slowly or too quickly with time are identified as suspect. This removes observations that keep constant during an extensive period. Finally, they detect spatially inconsistent wind speed observations using a set of neighboring stations to estimate a reasonable interval of spatially consistent values. 

Official weather stations measure wind at 10 m in an open field but citizen weather stations are typically placed much lower and are often surrounded by obstructions. This can lead to an underestimation of the wind speed. These systematic biases should be adjusted (Figure 1) before the data can be used any further. The researchers used a bias-correction method that avoids making assumptions about the wind speed distribution. This makes the method the best fit for the bias adjustment job and is easily tailored for individual stations. The underestimation of citizen weather stations also results in low wind speeds being erroneously recorded as zeros. These zeros are adjusted by taking interpolated neighboring observations.

A flowchart showing the four steps in the quality control and bias-adjustment procedure
Figure 1: Diagram of the overall quality control and bias adjustment system for wind speed observations collected at citizen weather stations.
Figure 2: Scatter plot of simultaneous observations between an example WOW station (serial number: 956296001) and a nearby KNMI station (Cabauw), showing both the raw WOW data (red circles) and final WOW data (blue squares).

The observations data at citizen weather stations in the study is provided from Weather Observation Website (WOW, The quality of wind speed observations at citizen weather stations is measured by comparing with simultaneous official observations. The research shows that that the wind speed observations between citizen weather stations and official stations are comparable (Figure 2), after implementing the quality control and bias adjustment procedures.  The final WOW data matches much more closely to KNMI, as is shown in Figure 2 for an example station (956296001). A detailed investigation of several statistical indicators, presented in the paper.


The work clearly demonstrates that the quality of crowdsourced wind speed data has been greatly improved, and that the data has the potential to complement an official observing network.