Accumulation-based Advection Field for Rainfall Nowcasting

Journal of Hydrology, 2006

A multiple linear regression model complemented by a correction procedure (REG) is applied to nowcasting of 1-h precipitation for warm seasons. The model aims at predicting the mean area precipitation in the 9 km • 9 km squares for the territory of the Czech Republic. Precipitation amounts are calculated from the radar reflectivity measured by two C band radars, which are operated by the Czech Hydrometeorological Institute. The model predictors are obtained from: (i) radar-derived precipitation in the squares; (ii) radar-derived precipitation advected by wind fields at the 700 hPa level produced by ALADIN/LACE NWP model forecast; (iii) variables derived from forecasts of the ALADIN/LACE NWP model. Accuracy of the REG forecast is evaluated and compared with the forecast obtained by the advection of precipitation fields by the NWP wind at the 700 hPa level (ADV). The ADV forecast serves as one of the predictors (ii). Results show that REG yields apparently better forecasts than ADV in terms of RMSE, correlation coefficients and categorical measures. Also local precipitation maxima have better positions with respect to the real ones. However, if ADV did not indicate any precipitation, REG was not capable to forecast a convective precipitation development. This fact deteriorates the REG accuracy during afternoon hours. We may say that the REG forecast is more accurate in night and morning hours since the situation mentioned occurs quite scarcely during those time periods.

International Journal of Remote Sensing and Earth Sciences (IJReSES), 2021

Nowcasting, or the short-term forecasting of precipitation, is urgently needed to support the mitigation circle in hydrometeorological disasters. Pangkalan Bun weather radar is single-polarization radar with a 200 km maximum range and which runs 10 elevation angles in 10 minutes with a 250 meters spatial resolution. There is no terrain blocking around the covered area. The Short-Term Ensemble Prediction System (STEPS) is one of many algorithms that is used to generate precipitation nowcasting, and is already in operational use. STEPS has the advantage of producing ensemble nowcasts, by which nowcast uncertainties can be statistically quantified. This research aims to apply STEPS to generate stochastic nowcasting in Pangkalan Bun weather radar and to analyze its advantages and weaknesses. Accuracy is measured by counting the possibility of detection and false alarms under the 5 dBZ threshold and plotting them in a relative operating characteristic (ROC) curve. The observed frequency ...

Atmosphere, 2019

Radar rainfall nowcasts are subject to many sources of uncertainty and these uncertainties change with the characteristics of a storm. The predictive skill of a radar rainfall nowcasting model can be difficult to understand as sometimes it appears to be perfect but at other times it is highly inaccurate. This hinders the decision making required for the early warning of natural hazards caused by rainfall. In this study we define radar spatial and temporal rainfall variability and relate them to the predictive skill of a nowcasting model. The short-term ensemble prediction system model is configured to predict 731 events with lead times of one, two, and three hours. The nowcasting skill is expressed in terms of six well-known indicators. The results show that the quality of radar rainfall nowcasts increases with the rainfall autocorrelation and decreases with the rainfall variability coefficient. The uncertainty of radar rainfall nowcasts also shows a positive connection with rainfal...

Water Resources Research, 1991

Covariance analysis was used to determine the reduction in rainfall forecast and estimation variances offered by radar reflectivity data. Covariance analysis of a particular nonadvective linear physically based model indicated that the utility of the radar reflectivity data of various elevation angles is limited in mean areal rainfall predictions, even when a very small density of rain gauges exists over the region of interest and good quality radar data are used. This applies to both raw reflectivity and radar rainfall data converted through a Z-R relationship. The ratio of mean areal rainfall prediction variances, defined as variance with radar data divided by variance without radar data, was found to be greater than 0.8 in most cases. On the other hand, the radar data reduced the estimated variance of the vertically integrated liquid water content considerably, even when high-density rain gauge data were present. !. INTRODUCTION Forecasting mean areal rainfall accurately and reliably on a basin scale and in real time has been one of the pressing needs of hydrology. Recently, it has been addressed in review papers of Georgakakos and Hudlow [1984] and Georgakakos and Kavvas [1987]. They discuss the suitability of spatially lumped quantitative rainfall prediction models developed by Georgakakos and Bras [1984a, b] and Georgakakos [1984, 1986c] for use in real-time flood forecasting. The models were developed in an effort to improve shortterm precipitation predictions on the scale of small-and medium-sized hydrologic basins (100-1000 km2). They have already been coupled to hydrologic models [Georgakakos, 1986a, b] to form integrated hydrometeorological forecast systems for the real-time prediction of floods and flash floods [Georgakakos, 1987]. The models, as originally formulated, use point data of readily available variables as input, such as surface air temperature, dew point temperature, and pressure. Real-time observations of rain gauge rainfall are used in model state updating. It seems reasonable to expect improvements in model predictions if additional information on the relevant processes could be included. In particular, further improvements are expected if real-time observations of radar reflectivity were used. The rainfall prediction models are based on the principle of the conservation of liquid water mass and utilize adiabatic and pseudo-adiabatic air parcel ascent processes for the determination of the condensation and deposition source term. Computations of cloud precipitation rate are based on parameterizations of cloud microphysics. Evaporation of cloud drops in the subcloud layer of unsaturated air is also modeled. Those models predict hourly precipitation rates, given input in the form of surface air temperature, pressure, and dew point temperature. The models have been formulated in state space form, and state estimators have been designed to process mean areal rainfall observations in real time for (1) state updating and (2) determination of forecast uncertainty. The models, when tested with field data,

Geophysical Research Letters, 2005

[1] Short term precipitation forecasts based on Lagrangian advection of radar echoes are robust and have more skill than numerical weather prediction models over time scales of several hours. This is because the models do not generally capture well the initial ...

Automated Forecasting System have been shown to be highly effective in providing a wide range of integrated capabilities for time series analysis and forecasting, econometrics and systems modeling and financial analysis and reporting. We have presented a platform for the practical development and design of a Rainfall Prediction System for the Kano City Metropolitan Area. The system automatically makes future forecasts of rainfall using the rainfall forecasting model derived from the inputted rainfall time series data. The system when fully implemented will ensure accurate analysis and forecasting processes that takes place over time in Kano. The language of implementation is Visual Basic. NET

International Journal of Scientific Research in Computer Science, Engineering and Information Technology, 2024

Weather is influenced by several factors such as temperature, pressure, air movement, moisture/water vapor, and the Earth's rotation. Forecasting weather with high geographic precision is both challenging and computationally demanding. This study explores a nowcasting method for meteorological radar images. Building on the concept of an unsupervised representation problem in deep learning, next-frame prediction is an emerging area in computer vision where future images are predicted based on previous ones. This has numerous applications, including in robotics and autonomous driving. The paper reviews the latest next-frame prediction networks, categorizing them into two main approaches: machine learning and deep learning. It compares these strategies by discussing their respective advantages and disadvantages. Lastly, the paper outlines the most promising directions for future research.

Water Resources Research, 1993

A model based on the two-dimensional stochastic advection-diffusion equation is developed to forecast properties of individual rain cells in urban areas such as speed and spatial rainfall intensity. Two different modeling approaches are employed, and examples of the results are given. The first approach involves a Gaussian distribution as an analytic solution to the advection-diffusion equation, whereas the second one entails a double Fourier series expansion of the rainfall intensity field. Both modeling approaches are used to predict the rainfall intensity field over a small 12-gage urban catchment in southern Sweden. The model parameters are continuously updated by extended Kalman filtering. The Fourier series approach is shown to be the most flexible for practical applications and to give the most accurate forecasts. This model approach gives acceptable forecasts for a lead time of 1-5 min. It gives consistently smaller prediction errors compared to both the Gaussian solution and simple extrapolation calculations. The effect of system noise level on the forecast accuracy and model performance is discussed. The model can be used not only to predict in real time the spatial rainfall, but also to parameterize the variability pattern of small-scale spatial rainfall into a set of physically based parameters, thus separating the effects of advective velocity, turbulent diffusion, and development/decay. 1. INTRODUCTION Real-time prediction of the space-time rainfall field in urban areas is motivated for on-line decision making to optimize the operation of urban hydrological systems, thus minimizing pollutant discharge from drainage systems, avoiding flooding, and making maximum use of the available storage volume in sewers. Prediction of the space-time rainfall field and its dynamic behavior is required as input to runoff simulation models. The requirements of the temporal and spatial resolution of the rainfall input are especially high for urban catchments [e.g., Berndtsson and Niemczynowicz, 1988]. Small catchment areas, dense building structure, and thus a high degree of impermeable areas with a resulting rapid runoff, imply that the smallest spatial units of the rainfall field (individual cells) are of primary importance. Most of the existing models for real-time control of hydrological systems use a stochastic description of the rainfall field [Georgakakos and Hudlow, 1984; Georgakakos and Bras, 1984; Georgakakos, 1986]. Models for short-term rainfall forecasting or nowcasting have in most cases used radar data as input [Browning and Collier, 1989; Bellon and Austin, 1984; Einfalt and Denoeux, 1987]. However, most urban catchments are still not equipped with radar; instead urban rainfall is usually observed by rain gages. Thus in order to extend the use of measurements from rain gages 1On leave from Department of Water Resources Engineering, there is a need to develop methods for rainfall forecasting with respect to the specific requirements of urban hydrological management. If radar data also are at hand, the radar can provide information on a larger scale embedding the rain gage system (e.g., by specifying boundary conditions for the forecasting area). Real-time prediction of urban-scale rainfall relies on two fundamentals: (1) an understanding of the small-scale spacetime characteristics of the rainfall field, and (2) an effective technique to determine the rainfall intensities in time and space. The former has been elaborated on in a related paper (R. Berndtsson et al., Some Eulerian and Lagrangian statistical properties of rainfall at small space-time scales, submitted to Journal of Hydrology, 1992; hereinafter Berndtsson et al., submitted manuscript, 1992), and the latter will be dealt with in this paper. There are several features that a useful forecasting model must exhibit due to the special characteristics of small-scale rainfall variability and the requirements of the urban hydrological management system. First, the model should be capable of reconstructing the irregular shape of the smallscale rainfall field from a set of physically meaningful mathematical parameters. Second, the model should be flexible enough to allow for a dynamic behavior of the rainfall field in both time and space. Third, the model should be able to predict the rainfall field trend in time. Fourth, the model should permit a high degree of automation allowing for use in an interactive urban management system [e.g., Browning and Collier, 1989]. Kumar and Foufoula-Georgiou [1990] described whole contour methods as one of the most promising future techniques for extrapolative short-time forecasting (see also Kalman, R. E., A new approach to linear filtering and prediction problems, J. Basic Eng., 82D, 35-45, 1960. Kalman, R. E., and R. S. Bucy, New results in linear filtering and prediction theory, J. Basic Eng., 83D, 95-108, !961. On the temporal and spatial characteristics of short-term urban-scale rainfall and its real-time prediction, (in Japanese with English abstract), Proc. Hydraul. Eng. Jpn. Soc. Civ. Eng., 35, 63--68, 1991. Kumar, P., and E. Foufoula-Georgiou, Fourier domain shape analysis methods' A brief review and an illustrative application to rainfall area evolution, Water Resour. Res., 26, 2219-2227, 1990. Marshall, R. J., The estimation and distribution of storm movement . and storm structure, using a correlation analysis technique and raingauge data, J. Hydrol., 48, 19--39, 1980. Niemczynowicz, J., An investigation of the areal and dynamic properties of short-term rainfall and its influence on runoff generating processes, Rep. 1005, 2!5 pp., Dep. of Water Resour. Eng.,

Acta Scientiarum Polonorum Formatio Circumiectus

Forecasts from nowcasting models are increasingly becoming a crucial input to the rainfall-runoff models. A basic approach to the nowcast generation is based on extrapolation (advection) of current precipitation field. The main limitation of such nowcasting is the rapid decrease in accuracy with forecasting lead time, due to dynamical evolution of precipitation, especially when convection appears, therefore recent studies are focused on taking into account also the evolution of precipitation. According to subject literature, the conceptual cell lifecycle models are not sufficient to significantly increase forecast accuracy, thus at present new approaches based on autoregressive models are investigated. This paper presents the SNAR (Spectral Nowcasting with Autoregression) nowcasting model developed at IMGW-PIB. The aim of the present research is to improve the nowcasting reliability, and to extend the lead time. The model proposes two innovative solutions: (i) decomposition of precipitation field to layers associated with their spatial scale, (ii) forecasting based on autoregressive model. The paper gives an overview of algorithms used in the SNAR model and provides preliminary results.

Journal of Engineering Science and Technology Review, 2022

Precipitation nowcasting is important for a multitude of applications like the planning and preparation of flights in the aviation industry and also for daily routine management. Existing methods are complex, computationally extensive. Also, inherent model configurations involved in assimilation and simulation of current atmospheric state limit their application in forecasting precipitation for next two to four hours. So, in the proposed paper, the development of a data driven algorithm for precipitation nowcasting using the Automatic Weather Station (AWS) data obtained from Vikram Sarabhai Space Centre, Thiruvananthapuram has been carried out. Three different artificial intelligence (AI) models have been developed based on ensemble learning techniques because of low variance and robustness i.e., bagging and boosting, and the potential of these ensembles viz. Random Forest, Adaboost, and Xgboost have been investigated. The accuracy of Random Forest and Xgboost are comparable and slightly better than Adaboost in predicting the occurrence of rainfall events. The model can capture the instances with good accuracy of 80 % for the Random Forest and Xgboost and 65 % for the Adaboost. Sensitivity of the in-situ observations has been carried out. The results indicate that for the prediction of rainfall, time series of pressure and temperature are found to be more influential than other predictors in all the State-of-the-art, machine learning techniques. The study is important from the perspective to apply as an approach for rainfall nowcasting applications and also can be applied for the other heterogenous and dynamic systems.