Nowcasting

In this paper, a novel Radar-based technique to forecast probability of rainfall within the next 4 hours is presented. Time series data of Doppler Weather RADARs located at different points in the Philippines allow us to observe and forecast the chance of rain at a more localized level. Trajectories derived from the trail of past observations are used to generate the forecasted location of rainfall. Based on their projected locations, a percent chance of rain (PCOR) per city is calculated. From this technique, automatically obtained forecasts for rain events are accurate with an average accuracy of 82.68%, and with an average success ratio of 57.98% peaking at 76% at the first hour for forecasted rain with actual rain events.

The World Meteorological Organization (WMO) World Weather Research Programme’s (WWRP) Forecast and Research in the Olympic Sochi Testbed program (FROST-2014) was aimed at the advancement and demonstration of state-of-the-art nowcasting and short-range forecasting systems for winter conditions in mountainous terrain. The project field campaign was held during the 2014 XXII Olympic and XI Paralympic Winter Games and preceding test events in Sochi, Russia. An enhanced network of in situ and remote sensing observations supported weather predictions and their verification. Six nowcasting systems (model based, radar tracking, and combined nowcasting systems), nine deterministic mesoscale numerical weather prediction models (with grid spacings down to 250 m), and six ensemble prediction systems (including two with explicitly simulated deep convection) participated in FROST-2014. The project provided forecast input for the meteorological support of the Sochi Olympic Games. The FROST-2014 archive of winter weather observations and forecasts is a valuable information resource for mesoscale predictability studies as well as for the development and validation of nowcasting and forecasting systems in complex terrain. The resulting innovative technologies, exchange of experience, and professional developments contributed to the success of the Olympics and left a post-Olympic legacy.

This paper describes a new multi-sensor approach for convective rain cell continuous monitoring based on rainfall derived from Passive Microwave (PM) remote sensing from the Low Earth Orbit (LEO) satellite coupled with Infrared (IR) remote sensing Brightness Temperature (TB) from the Geosynchronous (GEO) orbit satellite. The proposed technique, which we call Precipitation Evolving Technique (PET), propagates forward in time and space the last available rain-rate (RR) maps derived from Advanced Microwave Sounding Units (AMSU) and Microwave Humidity Sounder (MHS) observations by using IR TB maps of water vapor (6.2 μm) and thermal-IR (10.8 μm) channels from a Spinning Enhanced Visible and Infrared Imager (SEVIRI) radiometer. PET is based on two different modules, the first for morphing and tracking rain cells and the second for dynamic calibration IR-RR. The Morphing module uses two consecutive IR data to identify the motion vector to be applied to the rain field so as to propagate it in time and space, whilst the Calibration module computes the dynamic relationship between IR and RR in order to take into account genesis, extinction or size variation of rain cells. Finally, a combination of the Morphing and Calibration output provides a rainfall map at IR space and time scale, and the whole procedure is reiterated by using the last RR map output until a new MW-based rainfall is available. The PET results have been analyzed with respect to two different PM-RR retrieval algorithms for seven case studies referring to different rainfall convective events. The qualitative, dichotomous and continuous assessments show an overall ability of this technique to propagate rain field at least for 2–3 h propagation time.

The short-term prediction of precipitation is a difficult spatio-temporal task due to the non-uniform characterization of meteorological structures over time. Currently, neural networks such as convolutional LSTM have shown ability for the spatio-temporal prediction of complex problems.
In this research, we propose an LSTM convolutional neural network (CNN-LSTM) architecture for immediate prediction of various short-term precipitation events using satellite data. The CNN-LSTM
is trained with NASA Global Precipitation Measurement (GPM) precipitation data sets, each at 30-min intervals. The trained neural network model is used to predict the sixteenth precipitation
data of the corresponding fifteen precipitation sequence and up to a time interval of 180 min. The results show that the increase in the number of layers, as well as in the amount of data in the training data set, improves the quality of the forecast.