Rainfall estimation using moving cars as rain gauges (RainCars)
Geodesy, Photogrammetry, Remote Sensing, Geoinformatics, Cartography
Final Report Abstract
The project investigated the use of cars with windscreen wipers or optical rain sensors as a potential new data source for rainfall estimation. In the laboratory, a clear dependency between visibility through the car´s windscreen and rainfall intensity could be derived. Both, the wiper frequency and the readings of two different types of optical sensors have shown a significant correlation with rainfall measured by standard devices (termed W-R relationship). No sensor type showed a strong saturation behavior in the stationary case: rain intensities up to ~80 mm/h (wiper frequency) and 100mm/h (optical sensors), measured by a tipping bucket device, could be detected. The results for moving optical sensors show that with increasing speed, the rain amount detected by the sensor increases. The car´s windscreen angle is the determining factor in this relationship. In contrast to the theoretical considerations in the literature, advocating a linear relationship between speed and rain amount, a saturation behavior could be observed for this relationship, possibly due to the sensor readings´ upper bound, which might be reached when high rain intensities occur at high car speeds. Due to the large number of influencing factors such as road spray, shading of rainfall by buildings or trees or wind direction and speed, the W-R relationship determined in the laboratory cannot be expected to hold in reality under all circumstances. Establishing a W-R relationship in the field turned out to be more difficult due to the lack of a valid ‘ground truth’ at the car positions and the high space-time variability of rainfall, especially strong rainfall. Further, most of the field data was collected at rain intensities significantly lower compared to those simulated in the laboratory. Therefore, both relationships are not directly comparable to each other. In the field, the influence of car speed on the sensor readings could be determined with a high degree of certainty. It turned out that the wiper frequency of cars with discrete wiper classes (operated manually) is strongly coupled with the car speed and seems to be only marginally influenced by the rain amount. Therefore, such cars are currently not considered suitable for rainfall estimation. The optical sensor readings (and the wiper frequency of wipers controlled by an optical sensor) showed a linear relationship to the car speed. However, for strong rainfall intensities at high car speeds, a saturation behavior similar to that of the laboratory can be expected. The optical sensor readings, corrected by the car speed influence, show a linear dependency to rainfall estimated from radar. However, the errors are rather large and no reliable confidence bounds could be derived. For doing this, a better ground truth is required, e.g. in the form of a denser network of rain gauges in conjunction with significantly more data of a large-scale experiment, especially at high rain intensities relevant for the envisioned use cases. Using the uncertainties derived from laboratory experiments in a computer simulation, it was observed that cars can be beneficial for areal rainfall estimation and discharge simulation. Testing different uncertainties (for taking unforeseen uncertainties in practice, e.g. tree coverage, into account), it was observed that when errors lay below a certain threshold and the car density is above a certain level using cars as the direct means of rainfall measurement can be useful. Considering large uncertainties for cars resulted in deterioration of results. In this case, cars might be useful when regarded as additional information in interpolation techniques. Both the temporal resolution of data and number of observations were observed to be important concerning the quality of the interpolation techniques assessed by cross-validation. Considering weather radar as an important source of rainfall data, it was observed that original radar as additional information in interpolation techniques may not be useful for fine temporal resolution, whereas using smoothed radar data resulted in better results than when only rain gauges were considered. Furthermore, it was observed that correcting radar data using the proposed quantile mapping correction method results in a significant improvement of the quality of rainfall estimation. It was also observed that implementing CM is more convenient than other investigated techniques. Radar is provided in 5-min resolution. For 1-min resolution, interpolation methods using field motion information derived from radar have shown to be promising extensions of spatial, symmetric spatio-temporal or radar-rain gauge merging techniques. It could be shown that the motion estimation might not necessarily require data on a regular grid, e.g. weather radar images, but is possible in a decentralized way by the irregularly distributed nodes of a GSN. In addition to the number of observations and the temporal resolution of the data, the location of the stations was observed to be important. A method for relocating the stations (optimizing) was proposed and a significant improvement was observed when comparing the optimized network with the original DWD network by means of cross validation. The improvement for fine temporal resolutions was observed to be more significant, illustrating the high spatial variation of the rainfall data in fine temporal resolutions. Future research necessary include field experiments on a larger scale together with the investigation of current technology available in the cars (‘Floating Car Data’), as well as the integration of all available information (rain gauges, radar, and cars).
Publications
-
(2013): Rainfall estimation using moving cars as rain gauges - laboratory experiments. Hydrol. Earth Syst. Sci., 17(11): 4701-4712
Rabiei, E., Haberlandt, U., Sester, M., Fitzner, D.
-
(2013): Rainfall Estimation with a Geosensor Network of Cars - Theoretical Considerations and First Results - PFG Photogrammetrie, Fernerkundung, Geoinformation, vol. 2013, no. 2, p. 93-103, 2013
Fitzner, D., Sester, M., Haberlandt, U., Rabiei, E.
-
(2014): Geostatistical merging of rain gauge and radar data for high temporal resolutions and various station density scenarios - Journal of Hydrology, Volume 508, Pages 88–101
Berndt, C., Rabiei, E., Haberlandt, U.
-
(2015). Estimation of precipitation fields from 1-minute rain gauge time series–comparison of spatial and spatio-temporal interpolation methods. International Journal of Geographical Information Science, 29(9), 1668- 1693
Fitzner, D., and Sester, M.
-
(2015): Applying bias correction for merging rain gauge and radar data. Journal of Hydrology, 522(0): 544-557
Rabiei, E. and Haberlandt, U.
-
(2016). Decentralized Gradient-based Field Motion Estimation with a Wireless Sensor Network.In Proceedings of the 5th International Conference on Sensor Networks, Rome, Italy
Fitzner D. and Sester M.
-
(2016): Areal rainfall estimation using moving cars – computer experiments including hydrological modeling - Hydrol. Earth Syst. Sci. Discuss., 2016: 1-38
Rabiei, E., Haberlandt, U., Sester, M., Fitzner, D., Wallner, M.
