Figure - available from: Geophysical Research Letters
This content is subject to copyright. Terms and conditions apply.
Two‐stage deep learning system for Tropical Rainfall Measuring Mission (TRMM) Precipitation Radar (PR) rainfall estimation: (a) overall system diagram; Block 1 shows the conceptual diagram of the MLP‐1 model designed for ground radar using rain gauge as target labels; Block 2 illustrates the geometry and alignment between ground‐based and spaceborne radar measurements; Block 3 sketches the MLP‐2 model for PR using ground radar rainfall estimates (from MLP‐1) as target labels; (b) MLP model optimization for a predefined hyperparameter; (c) details of the MLP‐1 model; (d) details of the MLP‐2 model.
Source publication
Plain Language Summary
The Tropical Rainfall Measuring Mission (TRMM) Precipitation Radar (PR) was the first spaceborne active sensor for observing precipitation over the tropics and subtropics. During its 17 years (1997–2014) in orbit and beyond, PR has been an important tool to characterize tropical precipitation microphysics and quantify rainfal...
Similar publications
We compare more than three years (between 2014 and 2018) of precipitation estimates over Germany from the Dual-frequency Precipitation Radar (DPR) operating on the core satellite of the Global Precipitation Mission (GPM) with the radar-derived precipitation product RADOLAN RY provided by the German national meteorological service (DWD, Deutscher We...
Global precipitation demonstrates a substantial role in the hydrological cycle and offers tremendous implications in hydro-meteorological studies. Advanced remote sensing instrumentations, such as the NASA Global Precipitation Measurement mission (GPM) Dual-Frequency Precipitation Radar (DPR) can estimate precipitation and cloud properties, and has...
Rain cells are the most elementary unit of precipitation system in nature. In this study, fundamental geometric and physical characteristics of rain cells over tropical land and ocean areas are investigated by using 15-yr measurements of the Tropical Rainfall Measuring Mission (TRMM) Precipitation Radar (PR). The rain cells are identified with a mi...
As part of the Second Tibetan Plateau Scientific Expedition and Research (STEP) field campaign, raindrop size distribution (DSD) measurements were conducted with a laser optical particle size velocity disdrometer in Motuo on the Tibetan Plateau (TP). The DSD characteristics of five different rain rate classes and convective and stratiform precipita...
Rain cells are the most common units in the natural precipitation system. Enhancing the understanding of these rain cell characteristics can significantly improve the cognition of the precipitation system. Previous studies have mostly analyzed rain cells from a single radar data. In this study, we merged the precipitation parameters measured by the...
Citations
... In recent years, deep learning models based on neural networks have provided new research directions for QPE due to their powerful nonlinear representation capabilities [20][21][22][23]. Compared with the traditional Z-R relationship, the deep learning model can learn the complex mapping relationship between radar data and precipitation intensity from extensive historical data, significantly improving the accuracy of QPE. ...
Quantitative precipitation estimation (QPE) by radar observation data is a crucial aspect of meteorological forecasting operations. Accurate QPE plays a significant role in mitigating the impact of severe convective weather. Traditional QPE methods mainly employ an exponential Z–R relationship to map the radar reflectivity to precipitation intensity on a point-to-point basis. However, this isolated point-to-point transformation lacks an effective representation of convective systems. Deep learning-based methods can learn the evolution patterns of convective systems from rich historical data. However, current models often rely on 2 km-height CAPPI images, which struggle to capture the complex vertical motions within convective systems. To address this, we propose a novel QPE model: combining the classic extrapolation model ConvLSTM with Unet for an encoder-decoder module assembly. Meanwhile, we utilize three-dimensional radar echo images as inputs and introduce the convolutional block attention module (CBAM) to guide the model to focus on individual cells most likely to trigger intense precipitation, which is symmetrically built on both channel and spatial attention modules. We also employ asymmetry in training using weighted mean squared error to make the model concentrate more on heavy precipitation events which are prone to severe disasters. We conduct experiments using radar data from North China and Eastern China. For precipitation above 1 mm, the proposed model achieves 0.6769 and 0.7910 for CSI and HSS, respectively. The results indicate that compared to other methods, our model significantly enhances precipitation prediction accuracy, with a more pronounced improvement in forecasting accuracy for heavy precipitation events.
... Tao et al. (2017) proposed a DL model to extract features from bispectral satellite infrared (IR) and water vapor (WV) channels for detecting rain areas. Chen et al. (2019) proposed a two-stage hybrid neural network to estimate precipitation using ground-based radar and satellite observations. Wang et al. (2021) proposed a transfer learning based method, which uses data-riched Continental US (CONUS) IR data set from the Geostationary Operational Environment Satellite for pre-training of the model, and then transferred to China through re-training with multi-band IR signals from Chinese Fangyuan (FY) satellite. ...
Quantitative precipitation forecasting in numerical weather prediction (NWP) models is contingent upon physicals parameterization schemes. However, uncertainties abound due to limited knowledge of the precipitating processes, leading to degraded forecasting skills. In light of this, our study explores the application of a Swin‐Transformer based deep learning (DL) model as a supplementary tool for enhancing the mapping trajectory between the NWP fundamental variables and the most downstream variable precipitation. Constrained by the observational satellite precipitation product from NOAA CPC Morphing Technique (CMORPH), the DL model serves as the post‐processing tool that can better resolve the precipitation patterns compared to solely based on NWP estimation. Compared to the baseline Weather Research and Forecasting simulation, the DL post‐processing effectively extracts features over meteorological variables, leading to improved precipitation skill scores of 21.7%, 60.5%, and 45.5% for light rain, moderate rain, and heavy rain, respectively, on an hourly basis. We also evaluate two case studies under different driven synoptic conditions and show promising results in estimating heavy precipitation during strong convective precipitation events. Overall, the proposed DL model can provide a vital reference for capturing precipitation‐triggering mechanisms and enhancing precipitation forecasting skills. Additionally, we discuss the sensitivities of the fundamental meteorological variables used in this study, training strategies, and performance limitations.
... However, the non-linear character of rainfall renders NWP models susceptible to initial meteorological conditions, necessitating complex atmospheric models for improved predictions (Coiffier, 2011;Li et al., 2021;Shi et al., 2015). Alternatively, radar observations coupled with mathematical models have offered viable solutions (Cheung and Yeung, 2012;Chen et al., 2019;Chen et al., 2020;Woo and Wong, 2017;Bárdossy and Pegram, 2017). Beyond radar-observation-based approaches, the utilization of ML models for rainfall prediction has gained prominence in various hydrological applications, including flood and drought forecasting, flow prediction, reservoir routing, and urban drainage modeling (Piadeh et al., 2023;Palmitessa et al., 2022;Tang, 2023;Lim and Wang, 2022;Wang et al., 2022;Chen et al., 2023). ...
Arid regions like the United Arab Emirates (UAE) face a dire challenge of scarce water resources and unpredictable climate patterns. This study investigates the efficacy of advanced Machine Learning (ML) techniques in enhancing rainfall prediction within hyper-arid environments. Leveraging an extensive 30-year dataset from 1991 to 2020, this study harnessed the power of XGBoost, LSTM, Random Forest (RF), Gradient Boost (GB), Support Vector Machine (SVM), Multilayer Perceptron (MLP), Linear Regression (LR), and ensemble methods to significantly enhance the prediction accuracy of monthly rainfall over UAE. In the initial univariate analysis, focused solely on rainfall as the predictor, the ML models displayed encouraging performance during the training phase, achieving an impressive correlation coefficient (CC) of 0.88 for both XGBoost and the ensemble models. However, their predictive efficacy witnessed a decline during the testing phase, where the maximum CC reached 0.45. In contrast, traditional models like Linear Regression and SVM, yielded subpar results in both training and testing, exhibiting correlation values lower than 0.3. To address these limitations, a multivariate analysis is conducted by incorporating additional meteorological parameters, including wind speed, temperature, humidity, and evapotranspiration. This augmentation proved highly beneficial as it substantially enhanced the models' predictive capacities during the testing period. The XGB achieves a CC of 0.76, LSTM improves from 0.21 to 0.71, and stacked models exhibit promising behavior jumping from an average of 0.44 to 0.82 during the testing periods. Additionally, we performed a sensitivity analysis utilizing LASSO regression, which revealed that wind speed and minimum temperature emerged as the most influential parameters for monthly rainfall prediction in the arid context. These two meteorological factors exerted a substantial impact on the accuracy of our predictive models, underscoring their significance in understanding and forecasting rainfall patterns in hyper-arid regions, such as the United Arab Emirates. The identification of these key drivers further strengthens the foundation for effective water resource management and climate adaptation strategies in such challenging environments. This study provides valuable insights for water resource planning, agriculture, and climate resilience strategies in hyper-arid regions. Further research can build upon these results to enhance rainfall prediction models and support sustainable development in arid regions.
... Additionally, it is also important to note that the detection of underground pipelines shares many characteristics with the measurement of the relative position between adjacent wells. Techniques for detecting underground pipelines include electromagnetic sensing, mining radar, direct current, sound wave, and seismic wave methods, among others (Chen et al. 2019;Ge et al. 2021;El-Qady et al. 2014). Therefore, in recent years, several new methods have been proposed to measure the distance based on underground pipe detection. ...
In order to realize the efficient development of unconventional oil and gas, the measurement accuracy of wellbore spacing in the drilling of parallel horizontal wells is more and more required. Although the Rotating Magnet Ranging System or Magnetic Guidance Tool is used to achieve a good ranging effect in the drilling of dual horizontal wells, the position measurement of the magnetic sub leads to a large ranging error. A new ranging algorithm for the Two Sensor Packages-Rotating Magnet Ranging System is presented in this paper. The algorithm takes the magnetic signal generated by the rotation of the magnetic sub at a fixed position, the tilt measurement data of the two wells, the length of the magnetic sub, and the distance between the two fluxgate sensors as input parameters to avoid measuring the position of the magnetic sub and to reduce the influence of the degree of non-parallelism and the length of the magnetic sub. The simulation and experiment demonstrate that the inclination and azimuth angles of the two wells have a significant impact on the magnetic ranging results when the ranging well sections are not parallel and that the distance between the bottom of the drill bit and the center of the magnetic sub cannot be ignored. Moreover, the accuracy of the relative distance calculated by this new algorithm can reach 97%, and the error of direction calculation is less than 3°. Applying this algorithm in the field can successfully aid in controlling the spacing of cluster horizontal wells more accurately.
... Tao et al. [45]employed PERSIANN-SDAE in conjunction with Stacked Denoising Auto-Encoders (SDAE) to predict large-scale precipitation from GEOS imagery. In 2019, Hen and colleagues developed a two-step deep neural network (DNN) approach for predicting rainfall using data from ground radar reflectivity, TRMM precipitation radar, and gauge stations [46]. Their DNN method demonstrated excellent performance in both regional and global rainfall mapping. ...
Because of the accessibility of massive data from remote sensing data and developments in ML, machine learning (ML) techniques have been extensively applied in environmental remote sensing research. Modern machine learning (ML) frameworks like deep learning (DL) have significantly outperformed older models in terms of performance. This study focuses on the software that uses a traditional neural network (NN) as well as Deep Learning (DL) approaches in environmental remote sensing, which also covers land cover mapping, retrieval of environmental parameters, data fusion, image compression, and information reconstruction and prediction. It is also explained how DL may be used to monitor other aspects of the ecosystem, including the environment, water management, ground and air surface temperatures, transpiration, ultraviolet (UV) rays, and sea color all factors to consider. Following that, the essay explores the challenges and prospective uses of DL in environmental remote sensing.
... As informações de precipitação foram geradas na plataforma do Google Earth Engine-GEE de análise geoespacial baseada em nuvem (https://cade.earthengine.google.com), onde permite visualizar e analisar imagens como do TRMM (Tropical Rainfall Measuring Mission),considerado uma ferramenta promissora nas pesquisa deste âmbito(CHEN et al., 2019;BATTAGLIA et al., 2020). Foi utilizado para extrair informações da precipitação estimada de 20 anos (1998-2019) totalizando de 1.369 mm a 1.421 mm de média anual no recorte da área de estudo.A precipitação média anual foi calculada e dividida pelo número de meses chuvosos da região, de acordo com IBGE (1993). ...
Os estudos ambientais em bacia hidrográfica são fundamentais para o entendimento dos processos de uso e ocupação devido à urgência de equilíbrio entre as práticas econômicas e as condições ambientais que acabam por alterar drasticamente as paisagens naturais. O objetivo do presente estudo foi analisar a fragilidade ambiental da Bacia Hidrográfica do Ribeirão Jatobá- BHRJ localizada no médio curso do rio Aquidauana, sendo de grande contribuição para o Pantanal no estado de Mato Grosso do Sul. A metodologia adotada fundamentou-se na proposta de Ross (1994) e de Bacani et al. (2015), onde se considerou a análise integrada do ambiente a partir das variáveis, solo, declividade, precipitação, uso e cobertura da terra, e adição das Áreas Preservação Permanentes - APPs. Os resultados demonstraram que a BHRJ possui potencial natural de baixa fragilidade ocupando 56,9% da área da bacia, porém o tipo de uso predominante refere-se à classe de pastagem (46% da área), essa influência resulta em fragilidade ambiental emergente média, ocupando 74% da área de estudo. Para validar o modelo de fragilidade ambiental desenvolvido foram mapeados locais de erosão e conflitos de uso da terra que foram correlacionados com as classes de fragilidade ambiental e de uso e cobertura da terra. Constatou-se que as áreas mais frágeis próximas às APPs possuem maior número de locais (pontos) com erosão indicando uma correlação com os altos graus de fragilidade, onde o tipo de uso e cobertura da terra influencia na fragilidade juntamente com as condições naturais.
... CNN architectures can also include different normalization and/or dropout operators embedded among the different layers, which can further improve the network's performance. CNNs have been extensively researched, with models applied in many contexts of remote sensing, and in particular EO image classification [93][94][95][96]. ...
We propose AiTLAS—an open-source, state-of-the-art toolbox for exploratory and predictive analysis of satellite imagery. It implements a range of deep-learning architectures and models tailored for the EO tasks illustrated in this case. The versatility and applicability of the toolbox are showcased in a variety of EO tasks, including image scene classification, semantic image segmentation, object detection, and crop type prediction. These use cases demonstrate the potential of the toolbox to support the complete data analysis pipeline starting from data preparation and understanding, through learning novel models or fine-tuning existing ones, using models for making predictions on unseen images, and up to analysis and understanding of the predictions and the predictive performance yielded by the models. AiTLAS brings the AI and EO communities together by facilitating the use of EO data in the AI community and accelerating the uptake of (advanced) machine-learning methods and approaches by EO experts. It achieves this by providing: (1) user-friendly, accessible, and interoperable resources for data analysis through easily configurable and readily usable pipelines; (2) standardized, verifiable, and reusable data handling, wrangling, and pre-processing approaches for constructing AI-ready data; (3) modular and configurable modeling approaches and (pre-trained) models; and (4) standardized and reproducible benchmark protocols including data and models.
... Due to the success in dealing with images, DL is applied in the identification of weather patterns 33,34 , the category of precipitation types 35 , and the classification of the wave breaking in the ocean 36 . Based on satellite and radar images, DL is further used in estimations of rainfall and tropical cyclone intensity [37][38][39] . DL is also employed as a useful tool for forecasting air temperature, rainfall, and air quality [40][41][42][43] . ...
The quasi-periodic signals in the earth system could be the predictability source for sub-seasonal to seasonal (S2S) climate prediction because of the connections among the lead-lag time of those signals. The Madden–Julian Oscillation (MJO) is a typical quasi-periodic signal, which is the dominant S2S variability in the tropics. Besides, significantly periodic features in terms of both intensity and location are identified in 10–40 days for the concurrent variation of the subtropical and polar jet streams over Asia in this study. So far, those signals contribute less and are not fully applied to the S2S prediction. The deep learning (DL) approach, especially the long-short term memory (LSTM) networks, has the ability to take advantage of the information at the previous time to improve the prediction after then. This study presents the application of the DL in the postprocessing of S2S prediction using quasi-periodic signals predicted by the operational model to improve the prediction of minimum 2-m air temperature over Asia. With the help of deep learning, it finds the best weights for the ensemble predictions, and the quasi-periodic signals in the atmosphere can further benefit the S2S operational prediction.
... Deep learning approaches have been successfully applied in various remote-sensing scenarios, be it learning models from scratch or via transfer learning (Marmanis et al., 2016;Chen et al., 2019), in a fully supervised or self-supervised setting (Castillo-Navarro et al., 2022;Wang et al., 2022a), exploiting the heterogeneity (Neumann et al., https://doi.org/10.1016/j.isprsjprs.2023.01.014 Received 16 September 2022; Received in revised form 5 December 2022; Accepted 20 January 2023 I. Dimitrovski et al. 2020) and temporal properties (Ienco et al., 2017) of the available data. ...
... Received 16 September 2022; Received in revised form 5 December 2022; Accepted 20 January 2023 I. Dimitrovski et al. 2020) and temporal properties (Ienco et al., 2017) of the available data. As a result, this synergy of accurate DL approaches, on the one hand, and accessible high-resolution aerial/satellite imagery, on the other, has led to important contributions in various domains ranging from agriculture (Chlingaryan et al., 2018;Johnson et al., 2016;Xu et al., 2021a), ecology (Ayhan et al., 2020;Jo et al., 2018), geology (Shirmard et al., 2022) and meteorology (Zhang et al., 2018;Sadeghi et al., 2019;Chen et al., 2019) to urban mapping/planning (Longbotham et al., 2012;Lv et al., 2022;Huang et al., 2018) and archaeology (Somrak et al., 2020). ...
... CNN architectures have been widely researched, with models applied in many contexts of remote sensing, and in particular EO image classification (Chen et al., 2019;Weng et al., 2017;Castelluccio et al., 2015;Papoutsis et al., 2022). This includes AlexNet (Krizhevsky et al., 2012), a pioneering architecture that introduced and successfully demonstrated the utility of the CNN blueprint, mentioned earlier, for computer vision tasks. ...
We present AiTLAS: Benchmark Arena – an open-source benchmark suite for evaluating state-of-the-art deep learning approaches for image classification in Earth Observation (EO). To this end, we present a comprehensive comparative analysis of more than 500 models derived from ten different state-of-the-art architectures and compare them to a variety of multi-class and multi-label classification tasks from 22 datasets with different sizes and properties. In addition to models trained entirely on these datasets, we benchmark models trained in the context of transfer learning, leveraging pre-trained model variants, as it is typically performed in practice. All presented approaches are general and can be easily extended to many other remote sensing image classification tasks not considered in this study. To ensure reproducibility and facilitate better usability and further developments, all of the experimental resources including the trained models, model configurations, and processing details of the datasets (with their corresponding splits used for training and evaluating the models) are publicly available on the repository: https://github.com/biasvariancelabs/aitlas-arena.
... [33], [34]. In this work, we use a multilayer perceptron 267 (MLP) network, which is one of the commonly used network 268 architectures in training NN models [35]. ...
Compared with other types of Langmuir probes, segmented probes are arguably the least often used, to diagnose the state of plasma. With linear dimensions of the order of centimeters, these probes would induce perturbations, which would make them impractical in many laboratory plasma experiments. Their size, and the fact that they consist of several equipotential faces or “segments”, from which individual currents are collected, introduces additional complexities in the construction of inference techniques for their characteristics. In this work, we focus on the use of spherical segmented probes mounted on a satellite and present new techniques to infer plasma (ion) densities, flow velocities, and satellite potentials, from currents collected by two segmented probes biased to two different fixed potentials relative to a spacecraft. This is done by carrying out 3-D kinetic self-consistent particle-in-cell (PIC) simulations to compute the response of a probe to space plasma under different environment conditions of relevance to satellites in low-Earth orbit (LEO) at low and mid-latitudes. Computed currents and the corresponding known plasma and satellite parameters used as input in the simulations are then used to create a solution library with which regression-based inference models are constructed, following standard machine learning techniques. The models trained with a subset of our synthetic dataset are found to yield excellent agreement with data in distinct validation sets. The models constructed with synthetic data are then applied to
in situ
measurements made with segmented Langmuir probes mounted on the Proba-2 satellite, and the inferences are compared with densities reported on the Proba-2 data portal. The advantage of our approach is that it readily produces uncertainty margins that are specifically related to the inference technique used.
