Kühnlein, M.; Appelhans, T.; Thies, B. & Nauss, T. (2015): <b>Precipitation estimates from MSG SEVIRI daytime, night-time and twilight data with random forests</b>. <i>Journal of Applied Meteorology and Climatology</i> <b>53</b>, 2457-2480.
Resource Description
Title:
Precipitation estimates from MSG SEVIRI daytime, night-time and twilight data with random forests
FOR816dw ID:
246
Publication Date:
2015-11-10
License and Usage Rights:
PAK 823-825 data user agreement. (www.lcrs.de/dataagreementp3.do)
A new rainfall retrieval technique for determining rainfall rates in a continuous manner (day, twilight, and night) resulting in a 24-h estimation applicable to midlatitudes is presented. The approach is based on satellite-derived information on cloud-top height, cloud-top temperature, cloud phase, and cloud water path retrieved from Meteosat Second Generation (MSG) Spinning Enhanced Visible and Infrared Imager (SEVIRI) data and uses the random forests (RF) machine-learning algorithm. The technique is realized in three steps: (i) precipitating cloud areas are identified, (ii) the areas are separated into convective and advective-stratiform precipitating areas, and (iii) rainfall rates are assigned separately to the convective and advective-stratiform precipitating areas. Validation studies were carried out for each individual step as well as for the overall procedure using collocated ground-based radar data. Regarding each individual step, the models for rain area and convective precipitation detection produce good results. Both retrieval steps show a general tendency toward elevated prediction skill during summer months and daytime. The RF models for rainfall-rate assignment exhibit similar performance patterns, yet it is noteworthy how well the model is able to predict rainfall rates during nighttime and twilight. The performance of the overall procedure shows a very promising potential to estimate rainfall rates at high temporal and spatial resolutions in an automated manner. The near-real-time continuous applicability of the technique with acceptable prediction performances at 3–8-hourly intervals is particularly remarkable. This provides a very promising basis for future investigations into precipitation estimation based on machine-learning approaches and MSG SEVIRI data.
