Using a Convolution Neural Network to Improve Ensemble Tropical Cyclone Track Forecasts across the Atlantic Basin

Since 1970, tropical cyclone (TC) track forecasts have improved steadily in the Atlantic basin. This improvement has been linked primarily to advances in numerical weather prediction (NWP) models. Concurrently, with few exceptions, the development and operational use of statistical track prediction schemes have experienced a relative decline. Statistical schemes provided the most accurate TC track forecasts until approximately the late 1980s. In this note, it is shown that increased reliance on the global NWP models does not always guarantee the best forecast. Here, Hurricane Ivan is used from the 2004 Atlantic TC season as a classical example, and reminder, of how strong climatological signals still can add substantial value to TC track forecasts, in the form of improved accuracy and increased timeliness at minimal computational cost. In an 8-day period in early September 2004, Hurricane Ivan was repeatedly, and incorrectly, forecast by 12 operational NWP models to move with a significant northward (poleward) component. It was found that the mean 24-h trajectory forecasts of a consensus of five commonly used NWP track prediction aids had a statistically significant right-of-track bias. Furthermore, the official track forecasts, which relied heavily on erroneous numerical guidance over this period, were also found to have significant poleward trajectory errors. At the same time, a climatology-based prediction technique, drawn entirely from the historical record of motion characteristics of TCs in geographical locations similar to Ivan, correctly and consistently indicated a more westward motion component, had a small directional spread, and was supported by a large number of archived cases. This climatological signal was in conflict with the deterministic NWP model output, and it is suggested that the large errors in the official track forecast for TC Ivan could have been reduced considerably by taking into greater account such a strong climatological signal. The potential impact of such an error reduction is a saving of lives and billions of dollars in both actual damage and unnecessary evacuations costs, for just this one hurricane. We also suggest that this simple strategy of examining the strength of the climatological signal be considered for all TCs to identify cases where the NWP and official forecasts differ significantly from strong, persistent climatological signals.

An accurate track forecast for tropical cyclones (TCs) is critical for disaster prevention and mitigation. Although various models, including statistical, numerical, and artificial intelligence (AI) based, have demonstrated good skill in TC track forecasting, they have not ever been (and will not ever be) perfect, resulting in large uncertainties among them. To reduce the uncertainties, the Spatiotemporal Integrated Forecast Network (ST-IFNet), an integrated model, is developed using deep learning techniques and the track forecasts from seven operational agencies. ST-IFNet utilizes agency-time dual-branch attentions to dynamically adjust the weights of each agency at each time step and concurrently mines the physical trends of the changes in TC movement by the incorporation of environmental data. Experimental results show that ST-IFNet outperforms the operational forecast from any of the agencies as well as a simple average of them, especially in the medium-term (36–48 h) and long-term (60–120 h) TC track forecasts, achieving a track position error (TPE) as low as 62.89, 78.93, 123.96, 139.91, and 177.69 km for 24, 48, 72, 96, and 120 h, respectively. Further analysis reveals that large-scale potential height fields (≥1000 km) and mesoscale flow fields (200–1000 km) obtained from scale separation provide the most valuable guidance for ST-IFNet. This work builds a bridge between numerical forecasts and artificial intelligence techniques and provides a useful and efficient tool for operational TC track forecasts in the future. Significance Statement This study develops Spatiotemporal Integrated Forecast Network (ST-IFNet), an integrated forecast model, based on deep learning techniques to forecast tropical cyclone (TC) tracks. It effectively utilizes multiagency forecasts of TC tracks as well as the corresponding environmental data to achieve higher accuracy of track forecast with minimal computational costs, significantly enhancing the TC track forecast in the South China Sea. In particular, the large-scale potential height fields (≥1000 km) and flow fields that span 200–1000 km are found to provide the most valuable guidance for ST-IFNet.

A New Verification Measure for Tropical Cyclone Track Forecasts and its Experimental Application

In the application of Tropical Cyclone Track (TCT) forecast in South China Sea (SCS), pure linear neural network (PLNN) is used as the expert in the committee machine model, and it partly determines the model output. Data normalization is one of the most important factors, which affect the performance of the individual expert net. This paper aims to find how much data normalization affects the convergence process and forecast accuracy of the independent samples, and whether data normalization could bring reasonably accuracy and diversity that the committee machine model requires.

In this study, the general impact of high-resolution moving nesting domains on tropical cyclone (TC) intensity and track forecasts was verified, for a total of 107 forecast cases of 33 TCs, using the Weather Research and Forecasting (WRF) model. The experiment, with a coarse resolution of 12 km, could not significantly capture the intensification process, especially for maximum intensities (>60 m s−1). The intense TCs were better predicted by experiments using a moving nesting domain with a horizontal resolution of 4 km. The forecast errors for maximum wind speed and minimum sea-level pressure decreased in the experiment with higher resolution; the forecast of lifetime maximum intensity was improved. For the track forecast, the experiment with a coarser resolution tended to simulate TC tracks deviating rightward to the TC motions in the best-track data; this erroneous deflection was reduced in the experiment with a higher resolution. In particular, the track forecast in the experiment with a higher resolution improved more frequently for intense TCs that were generally distributed at relatively lower latitudes among the test cases. The sensitivity of the track forecast to the model resolution was relatively significant for lower-latitude TCs. On the other hand, the track forecasts of TCs moving to the mid-latitudes, which were primarily influenced by large-scale features, were not sensitive to the resolution.

High-altitude, long-endurance unmanned aircraft systems (HALE UAS) are capable of extended flights for atmospheric sampling. A case study was conducted to evaluate the potential impact of dropwindsonde observations from HALE UAS on tropical cyclone track prediction; tropical cyclone intensity was not addressed. This study employs a global observing system simulation experiment (OSSE) developed at the National Oceanic and Atmospheric Administration/Earth System Research Laboratory (NOAA/ESRL) that is based on the NOAA/National Centers for Environmental Prediction gridpoint statistical interpolation (GSI) data assimilation system and Global Forecast System (GFS) model. Different strategies for dropwindsonde deployment and UAS flight paths were compared. The introduction of UAS-deployed dropwindsondes was found to consistently improve the track forecast skill during the early forecast up to 96 h, with the caveat that the experiments omitted both vortex relocation and dropwindsondes from manned flights in the tropical cyclone region. The more effective UAS dropwindsonde deployment patterns sampled both the environment and the body of the tropical cyclone.

Understanding the learning mechanism of convolutional neural networks in spectral analysis

The orthogonal conditional nonlinear optimal perturbations (O-CNOPs) approach for measuring initial uncertainties is applied to the Weather Research and Forecasting (WRF) Model to provide skillful forecasts of tropical cyclone (TC) tracks. The hindcasts for 10 TCs selected from 2005 to 2020 show that the ensembles generated by the O-CNOPs have a greater probability of capturing the true TC tracks, and the corresponding ensemble forecasts significantly outperform the forecasts made by the singular vectors, bred vectors, and random perturbations in terms of both deterministic and probabilistic skills. In particular, for two unusual TCs, Megi (2010) and Tembin (2012), the ensembles generated by the O-CNOPs successfully reproduce the sharp northward-turning track in the former and the counterclockwise loop track in the latter, while the ensembles generated by the other methods fail to do so. Moreover, additional attempts are performed on the real-time forecasts of TCs In-Fa (2021) and Hinnamnor (2022), and it is shown that O-CNOPs are very useful for improving the accuracy of real-time TC track forecasts. Therefore, O-CNOPs, together with the WRF Model, could provide a new platform for the ensemble forecasting of TC tracks with much higher skill.

Among the many deep learning methods, the convolutional neural network (CNN) model has an excellent performance in image recognition. Research on identifying and classifying image datasets using CNN is ongoing. Animal species recognition and classification with CNN is expected to be helpful for various applications. However, sophisticated feature recognition is essential to classify quasi-species with similar features, such as the quasi-species of parrots that have a high color similarity. The purpose of this study is to develop a vision-based mobile application to classify endangered parrot species using an advanced CNN model based on transfer learning (some parrots have quite similar colors and shapes). We acquired the images in two ways: collecting them directly from the Seoul Grand Park Zoo and crawling them using the Google search. Subsequently, we have built advanced CNN models with transfer learning and trained them using the data. Next, we converted one of the fully trained models into a file for execution on mobile devices and created the Android package files. The accuracy was measured for each of the eight CNN models. The overall accuracy for the camera of the mobile device was 94.125%. For certain species, the accuracy of recognition was 100%, with the required time of only 455 ms. Our approach helps to recognize the species in real time using the camera of the mobile device. Applications will be helpful for the prevention of smuggling of endangered species in the customs clearance area.

<p>The accuracy of tropical cyclone (TC) track forecast has been improved year by years, on the other hand, the forecast of tropical cyclone intensity still has a difficulty of improvement. Recently the relationship between lightning activity and tropical cyclone intensity has been investigated. Lightning tends to increase during the rapid intensification of the TC. Therefore, monitoring the lightning activity becomes important for a TC intensity forecast. Lightning observation network are deployed over the western north Pacific by five very long frequency events trigger measurements called V-POTEKA at Palau, Guam, Manila, Okinawa Japan and Serpong Indonesia under the ULAT (Understanding Lightning and Thunderstorm) of SATREPS (Science and Technology Research Partnership for Sustainable Development) in the Philippines.</p><p>Tropical storm (TS) Bavi and Maysak were generated over the Philippine Sea on August 2020. We found that lightning activity drastically increased in the TC during the TC genesis stage when the TC reached TS criteria. Numbers of lightning reached the maximum during the life cycle of the TCs. Lightning was concentrated in the convective clouds about 100 to 200 km size. They are located around 400 to 500 km and 100 to 200 km from the TC center respectively and lasted about few hours. We called this phenomenon as “lightning burst”. On the other hand, when TS Haishen and Dolphin were generated over the Philippine Sea in September 2020, they did not observe lightning burst. About half of the TCs observed lightning burst in the Philippine Sea in 2020. We will investigate further what kind of structure occurred during the lightning burst and what kind of mechanism responsible for the lightning burst.</p>

Distinguishing pericarpium citri reticulatae of different origins using terahertz time-domain spectroscopy combined with convolutional neural networks

Accurately identifying epileptogenic zone (EZ) using high-frequency oscillations (HFOs) is a challenge that must be mastered to transfer HFOs into clinical use. We analyzed the ability of a convolutional neural network (CNN) model to distinguish EZ and non-EZ HFOs. Nineteen medically intractable epilepsy patients with good surgical outcomes 2 years after surgery were studied. Five-minute interictal intracranial electroencephalogram epochs of slow-wave sleep were selected randomly. Then 5 s segments of ripples (80–200 Hz) and fast ripples (FRs, 200–500 Hz) were detected automatically. The EZs and non-EZs were identified using the surgery resection range. We innovatively converted all epochs into four types of images using two scales: original waveforms, filtered waveforms, wavelet spectrum images, and smoothed pseudo Wigner–Ville distribution (SPWVD) spectrum images. Two scales were fixed and fitted scales. We then used a CNN model to classify the HFOs into EZ and non-EZ categories. As a result, 7,000 epochs of ripples and 2,000 epochs of FRs were randomly selected from the EZ and non-EZ data for analysis. Our CNN model can distinguish EZ and non-EZ HFOs successfully. Except for original ripple waveforms, the results from CNN models that are trained using fixed-scale images are significantly better than those from models trained using fitted-scale images (p < 0.05). Of the four fixed-scale transformations, the CNN based on the adjusted SPWVD (ASPWVD) produced the best accuracies (80.89 ± 1.43% and 77.85 ± 1.61% for ripples and FRs, respectively, p < 0.05). The CNN using ASPWVD transformation images is an effective deep learning method that can be used to classify EZ and non-EZ HFOs.

This study leveraged Convolutional Neural Network (CNN) models to estimate canopy height in Southeast Australian forests before and after the 2019–2020 bushfire event, using inputs from Sentinel-1, Sentinel-2, spectral indices, and terrain features and GEDI canopy height (rh99) as target variables. Our research consisted of three primary objectives: (1) an evaluation of GEDI rh99 canopy height in relation to an nDSM derived from airborne LiDAR, (2) the development of two CNN models for pre- and post-fire scenarios, and (3) utilizing the trained CNN models to generate pre- and post-fire canopy height maps and canopy height change maps across the continuous landscape. GEDI rh99 accuracy analysis revealed R 2 = 0.85, RMSE = 7.25 m and nRMSE = 0.23 for pre-fire GEDI and R 2 = 0.78, RMSE = 10.1 m and nRMSE = 0.3 for post-fire GEDI. The pre-fire CNN model achieved R 2 = 0.75, RMSE = 7.29 m and nRMSE = 0.22, and post-fire CNN model achieved R 2 = 0.62, RMSE = 11.67 m, and nRMSE = 0.34 when validated against nDSM. Bias analysis was conducted, revealing that GEDI rh99 underestimated nDSM across all canopy height ranges, and slope has no significant impact on GEDI accuracy. Notably, post-fire GEDI rh99 exhibited significant underestimation in high to extreme fire severity. The CNN models’ prediction displayed overestimation in short forests < 20 m and increasing underestimation in taller forests > 20 m and underestimation in high and extreme fire severity. Subsequent investigation indicated that canopy layer thinning in high fire severity conditions resulted in weak GEDI waveform signals and consequent bias. Additionally, the GEDI rh99 bias was propagated to the CNN models, together with insensitivity of spectral band values to canopy height in dense and burnt forests potentially contributed to models’ bias. Finally, we demonstrated CNN models’ ability to monitor forest recovery by generating canopy height maps spanning East Gippsland from 2019 to 2023.

BACKGROUND AND AIM: Outdoor ultrafine particles (UFP; 100 nm) and black carbon (BC) vary greatly within cities and may have adverse impacts on human health such as cardiovascular mortality and brain tumour incidence. Traditional regression methods for estimating spatial distributions of ambient air pollution depend on extensive, curated geospatial information system databases. This has resulted in a disparity wherein little is known about local air pollution in the data-sparse settings of low- and middle- income countries. We used a hybrid approach to develop new models to estimate within-city spatial variations in outdoor UFP and BC concentrations across Bucaramanga, Colombia. METHODS: We conducted a large-scale mobile monitoring campaign over twenty days in 2019. Land use regression (LUR) models were developed using land use parameters from curated and open-source databases. Convolutional Neural Network (CNN) models were developed using Google Maps satellite view and street view images. Combined models were developed by combining predictions from the LUR and CNN models. After training in the training and validation sets, predictions were generated in the test set and compared to measured values. RESULTS:The combined UFP model (R2=0.54) outperformed the CNN (R2=0.47) and land use regression (LUR) models (R2=0.47) on their own. Similarly, the combined BC model also outperformed the CNN and LUR BC models (R2 = 0.51 vs 0.43 and 0.45 respectively). Spatial variations in model performance were more stable for the CNN and combined models compared to the LUR models suggesting that the combined approach may be less likely to contribute to differential exposure measurement error in epidemiological studies. CONCLUSIONS:Estimates from these models can then be applied to population-based cohorts in order to evaluate population health risks. Additionally, our findings demonstrated that satellite and street-level images can be combined with a traditional LUR modelling approach to improve predictions of within-city spatial variations in outdoor UFP and BC concentrations. KEYWORDS: Ultrafine particles, Black Carbon, Deep learning, Images, Land use regression

Excessive withdrawal of groundwater for agricultural irrigation can cause seawater intrusion into coastal aquifers. Such a case will in turn results in deterioration of irrigation water quality. Determination of irrigation water quality with traditional methods is a time-consuming and costly process. However, machine learning algorithms can be useful tools for modeling and estimating groundwater quality used for irrigation water purposes. In this study, TDS, PS, SAR, and Cl parameters of groundwater were estimated with models based on EC and pH variables. For this purpose, prediction performances of two different deep learning methods (convolutional neural network (CNN) and deep neural network (DNN)) and two different classical machine learning (Random Forest (RF) and extreme gradient boosting (XGBoost)) methods were compared. In addition, predictive uncertainty of the models was determined by quantile regression (QR) analysis. Performance criteria and results of uncertainty analysis revealed that CNN (in testing phase, NSE = 0.95 for TDS, NSE = 0.96 for PS, NSE = 0.67 for SAR and NSE = 0.93 for CI) and DNN (in testing phase, NSE = 0.91 for TDS, NSE = 0.91 for PS, NSE = 0.57 for SAR and NSE = 0.94 for Cl) models had quite a close performance in estimation of TDS, PS, SAR, and Cl parameters and higher than the other two classical machine learning methods. As a result, the CNN model can be considered the best performing model in estimating all quality parameters due to the highest NSE and lowest RMSE values. In addition, the Taylor diagram showed that the values estimated using the CNN model had the highest correlation with the measured data. It was determined that the model with the lowest uncertainty based on the PICP statistics was DNN, followed by the CNN model. However, the CNN model has predicted outliers more accurately. Present findings proved that deep learning models could offer efficient tools for predicting irrigation water quality parameters.