Subpixel-Based Precipitation Nowcasting with the Pyramid Lucas–Kanade Optical Flow Technique

Li, Ling; He, Zhengwei; Chen, Sheng; Mai, Xiongfa; Zhang, Asi; Hu, Baoqing; Li, Zhi; Tong, Xinhua

doi:10.3390/atmos9070260

Cited by 24 publications

(19 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In the context of high spatiotemporal resolutions, traditional methods based on the Numerical Weather Prediction model are said to be computationally expensive, too sensitive to noises, highly dependent on initial conditions and not able to exploit big data [2]. Meanwhile, extrapolation-based approaches using radar reflectivity or remote sensing data can provide more accurate prediction [3,4]. Recently, data-driven approaches that leverage advances in machine learning/deep learning have been used to analyze radar images and perform precipitation nowcasting with promising results.…”

Section: Introductionmentioning

confidence: 99%

Computer Vision in Precipitation Nowcasting: Applying Image Quality Assessment Metrics for Training Deep Neural Networks

Tran

Song

2019

Atmosphere

View full text Add to dashboard Cite

This paper presents a viewpoint from computer vision to the radar echo extrapolation task in the precipitation nowcasting domain. Inspired by the success of some convolutional recurrent neural network models in this domain, including convolutional LSTM, convolutional GRU and trajectory GRU, we designed a new sequence-to-sequence neural network structure to leverage these models in a realistic data context. In this design, we decreased the numbers of channels in high abstract recurrent layers rather than increasing them. We formulated the task as a problem of encoding five radar images and predicting 10 steps ahead at the pixel level, and found that using only the common mean squared error can misguide the training and mislead the testing. Especially, the image quality of last predictions usually degraded rapidly. As a solution, we employed some visual image quality assessment techniques including Structural Similarity (SSIM) and multi-scale SSIM to train our models. Experimental results show that our structure was more tolerant to increasing uncertainty in the data, and the use of image quality metrics can significantly reduce the blurry image issue. Moreover, we found that using SSIM was very effective and a combination of SSIM with mean squared error and mean absolute error yielded the best prediction quality.

show abstract

Section: Introductionmentioning

confidence: 99%

Computer Vision in Precipitation Nowcasting: Applying Image Quality Assessment Metrics for Training Deep Neural Networks

Tran

Song

2019

Atmosphere

View full text Add to dashboard Cite

show abstract

“…However, these methods showed some inconsistencies in the magnitude of the movement vector [52]. In contrast, the optical flow method using intensity gradient information on several image scales (PyrLK) is a more robust method for an advection model using consecutive images [8,52,53]. The algorithm for this method was introduced by [54] and is based on the Lucas Kanade algorithm [55].…”

Section: Rain Field Advectionmentioning

confidence: 99%

“…The probability of detection (POD), false alarm (FAR), and frequency of bias (FBI) indices were calculated to quantify the coincidences between the projected and recorded images every 5 min. [8,18,57,58].…”

Section: Rain Field Advectionmentioning

confidence: 99%

“…In this regard, rainfall data acquired from different remote sensors, such as weather radar systems, have become more accessible over the last few years [5]. This technology has allowed us to study rainfall with high spatial and temporal resolutions [6], and it can be used in several applications, such as input data in rainfall forecasting studies [7][8][9][10][11], different hydrological models [12,13], early warning systems for floods [14], erosion studies [1,15], research related to atmospheric chemistry [16], and many others. Meanwhile, cost-effective systems derived from ship radar technology are available to establish radar networks with limited funds at hand [17].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Tropical Andes Radar Precipitation Estimates Need High Temporal and Moderate Spatial Resolution

Guallpa

Orellana‐Alvear

Bendix

2019

Water

View full text Add to dashboard Cite

Weather radar networks are an excellent tool for quantitative precipitation estimation (QPE), due to their high resolution in space and time, particularly in remote mountain areas such as the Tropical Andes. Nevertheless, reduction of the temporal and spatial resolution might severely reduce the quality of QPE. Thus, the main objective of this study was to analyze the impact of spatial and temporal resolutions of radar data on the cumulative QPE. For this, data from the world’s highest X-band weather radar (4450 m a.s.l.), located in the Andes of Ecuador (Paute River basin), and from a rain gauge network were used. Different time resolutions (1, 5, 10, 15, 20, 30, and 60 min) and spatial resolutions (0.5, 0.25, and 0.1 km) were evaluated. An optical flow method was validated for 11 rainfall events (with different features) and applied to enhance the temporal resolution of radar data to 1-min intervals. The results show that 1-min temporal resolution images are able to capture rain event features in detail. The radar–rain gauge correlation decreases considerably when the time resolution increases (r from 0.69 to 0.31, time resolution from 1 to 60 min). No significant difference was found in the rain total volume (3%) calculated with the three spatial resolution data. A spatial resolution of 0.5 km on radar imagery is suitable to quantify rainfall in the Andes Mountains. This study improves knowledge on rainfall spatial distribution in the Ecuadorian Andes, and it will be the basis for future hydrometeorological studies.

show abstract

“…concern. The use of sub-pixel methods [50] may however help to increase further the spatial resolution of fused product.…”

mentioning

confidence: 99%

The Convective Rainfall Rate from Cloud Physical Properties Algorithm for Meteosat Second Generation Satellites: Microphysical Basis and Intercomparisons using an Object-Based Method

Tapiador¹,

Marcos²,

Sancho³

2019

Preprint

View full text Add to dashboard Cite

The Convective Rainfall Rate from Cloud Physical Properties (CRPh) for Meteosat Second Generation Satellites is a day-only precipitation algorithm developed at the Spanish Meteorological Agency (AEMET) for EUMETSAT’ Satellite Application Facility in support to Nowcasting and Very Short Range Forecasting (NWC SAF). It is therefore mainly intended to provide input for monitoring and near-real-time forecasts for the next few hours. This paper critically discusses the theoretical basis of the algorithm with special emphasis in the empirical values and assumptions in the microphysics of precipitation and compares the performances of the CRPh with its antecessor, the Convective Rainfall Rate algorithm (CRR), using an object-based method. The analyses show that AEMET’s CRPh is physically consistent and that outperforms the CRR. The applicability of the algorithm for nowcasting and the challenges to evolve the product to an all-day algorithm are also presented.

show abstract

Subpixel-Based Precipitation Nowcasting with the Pyramid Lucas–Kanade Optical Flow Technique

Cited by 24 publications

References 36 publications

Computer Vision in Precipitation Nowcasting: Applying Image Quality Assessment Metrics for Training Deep Neural Networks

Computer Vision in Precipitation Nowcasting: Applying Image Quality Assessment Metrics for Training Deep Neural Networks

Tropical Andes Radar Precipitation Estimates Need High Temporal and Moderate Spatial Resolution

The Convective Rainfall Rate from Cloud Physical Properties Algorithm for Meteosat Second Generation Satellites: Microphysical Basis and Intercomparisons using an Object-Based Method

Contact Info

Product

Resources

About