Post-processing Model Research Articles

Probabilistic fire-danger forecasting: A framework for week-two forecasts using statistical post-processing techniques and the Global ECMWF Fire Forecast System (GEFF)

AbstractWildfire guidance two weeks ahead is needed for strategic planning of fire mitigation and suppression. However, fire forecasts driven by meteorological forecasts from numerical weather prediction models inherently suffer from systematic biases. This study uses several statistical-postprocessing methods to correct these biases and increase the skill of ensemble fire forecasts over the contiguous United States 8–14 days ahead. We train and validate the post-processing models on 20 years of European Centre for Medium-range Weather Forecast (ECMWF) reforecasts and ERA5 reanalysis data for 11 meteorological variables related to fire, such as surface temperature, wind speed, relative humidity, cloud cover, and precipitation. The calibrated variables are then input to the Global ECMWF Fire Forecast (GEFF) system to produce probabilistic forecasts of daily fire-indicators which characterize the relationships between fuels, weather, and topography. Skill scores show that the post-processed forecasts overall have greater positive skill at Days 8–14 relative to raw and climatological forecasts. It is shown that the post-processed forecasts are more reliable at predicting above- and below-normal probabilities of various fire indicators than the raw forecasts and that the greatest skill for Days 8–14 is achieved by aggregating forecast days together.

Intravoxel Incoherent Motion Magnetic Resonance Imaging in Skeletal Muscle: Review and Future Directions.

Throughout the body, muscle structure and function can be interrogated using a variety of noninvasive magnetic resonance imaging (MRI) methods. Recently, intravoxel incoherent motion (IVIM) MRI has gained momentum as a method to evaluate components of blood flow and tissue diffusion simultaneously. Much of the prior research has focused on highly vascularized organs, including the brain, kidney, and liver. Unique aspects of skeletal muscle, including the relatively low perfusion at rest and large dynamic range of perfusion between resting and maximal hyperemic states, may influence the acquisition, postprocessing, and interpretation of IVIM data. Here, we introduce several of those unique features of skeletal muscle; review existing studies of IVIM in skeletal muscle at rest, in response to exercise, and in disease states; and consider possible confounds that should be addressed for muscle-specific evaluations. Most studies used segmented nonlinear least squares fitting with a b-value threshold of 200 sec/mm2 to obtain IVIM parameters of perfusion fraction (f), pseudo-diffusion coefficient (D*), and diffusion coefficient (D). In healthy individuals, across all muscles, the average ± standard deviation of D was 1.46 ± 0.30 × 10-3 mm2 /sec, D* was 29.7 ± 38.1 × 10-3 mm2 /sec, and f was 11.1 ± 6.7%. Comparisons of reported IVIM parameters in muscles of the back, thigh, and leg of healthy individuals showed no significant difference between anatomic locations. Throughout the body, exercise elicited a positive change of all IVIM parameters. Future directions including advanced postprocessing models and potential sequence modifications are discussed. LEVEL OF EVIDENCE: 2 TECHNICAL EFFICACY: Stage 2.

Going with the Trend: Forecasting Seasonal Climate Conditions under Climate Change

AbstractFor managing climate variability and adapting to climate change, seasonal forecasts are widely produced to inform decision-making. However, seasonal forecasts from global climate models are found to poorly reproduce temperature trends in observations. Furthermore, this problem is not addressed by existing forecast postprocessing methods that are needed to remedy biases and uncertainties in model forecasts. The inability of the forecasts to reproduce the trends severely undermines user confidence in the forecasts. In our previous work, we proposed a new statistical postprocessing model that counteracted departures in trends of model forecasts from observations. Here, we further extend this trend-aware forecast postprocessing methodology to carefully treat the trend uncertainty associated with the sampling variability due to limited data records. This new methodology is validated on forecasting seasonal averages of daily maximum and minimum temperatures for Australia based on the SEAS5 climate model of the European Centre for Medium-Range Weather Forecasts. The resulting postprocessed forecasts are shown to have proper trends embedded, leading to greater accuracy in regions with significant trends. The application of this new forecast postprocessing is expected to boost user confidence in seasonal climate forecasts.

AutoDep: automatic depression detection using facial expressions based on linear binary pattern descriptor.

The psychological health of a person plays an important role in their daily life activities. The paper addresses depression issues with the machine learning model using facial expressions of the patient. Some research has already been done on visual based on depression detection methods, but those are illumination variant. The paper uses feature extraction using LBP (Local Binary Pattern) descriptor, which is illumination invariant. The Viola-Jones algorithm is used for face detection and SVM (support vector machine) is considered for classification along with the LBP descriptor to make a complete model for depression level detection. The proposed method captures frontal face from the videos of subjects and their facial features are extracted from each frame. Subsequently, the facial features are analyzed to detect depression levels with the post-processing model. The performance of the proposed system is evaluated using machine learning algorithms in MATLAB. For the real-time system design, it is necessary to test it on the hardware platform. The LBP descriptor has been implemented on FPGA using Xilinx VIVADO 16.4. The results of the proposed method show satisfactory performance and accuracy for depression detection comparison with similar previous work.

Post-processing numerical weather prediction ensembles for probabilistic solar irradiance forecasting

In order to enable the transition towards renewable energy sources, probabilistic energy forecasting is of critical importance for incorporating volatile power sources such as solar energy into the electrical grid. Solar energy forecasting methods often aim to provide probabilistic predictions of solar irradiance. In particular, many hybrid approaches combine physical information from numerical weather prediction models with statistical methods. Even though the physical models can provide useful information at intra-day and day-ahead forecast horizons, ensemble weather forecasts from multiple model runs are often not calibrated and show systematic biases. We propose a post-processing model for ensemble weather predictions of solar irradiance at temporal resolutions between 30 min and 6 h. The proposed models provide probabilistic forecasts in the form of a censored logistic probability distribution for lead times up to 5 days and are evaluated in two case studies covering distinct physical models, geographical regions, temporal resolutions, and types of solar irradiance. We find that post-processing consistently and significantly improves the forecast performance of the ensemble predictions for lead times up to at least 48 h and is well able to correct the systematic lack of calibration.

Operational Wave Forecast Selection in the Atlantic Ocean Using Random Forests

The existence of multiple wave forecasts leads to the question of which one should be used in practical ocean engineering applications. Ensemble forecasts have emerged as an important complement to deterministic forecasts, with better performances at mid-to-long ranges; however, they add another option to the variety of wave predictions that are available nowadays. This study developed random forest (RF) postprocessing models to identify the best wave forecast between two National Centers for Environmental Protection (NCEP) products (deterministic and ensemble). The supervised learning classifier was trained using National Data Buoy Center (NDBC) buoy data and the RF model accuracies were analyzed as a function of the forecast time. A careful feature selection was performed by evaluating the impact of the wind and wave variables (inputs) on the RF accuracy. The results showed that the RF models were able to select the best forecast only in the very short range using input information regarding the significant wave height, wave direction and period, and ensemble spread. At forecast day 5 and beyond, the RF models could not determine the best wave forecast with high accuracy; the feature space presented no clear pattern to allow for successful classification. The challenges and limitations of such RF predictions for longer forecast ranges are discussed in order to support future studies in this area.

Image quality of ultralow-dose chest CT using deep learning techniques: potential superiority of vendor-agnostic post-processing over vendor-specific techniques.

To compare the image quality between the vendor-agnostic and vendor-specific algorithms on ultralow-dose chest CT. Vendor-agnostic deep learning post-processing model (DLM), vendor-specific deep learning image reconstruction (DLIR, high level), and adaptive statistical iterative reconstruction (ASiR, 70%) algorithms were employed. One hundred consecutive ultralow-dose noncontrast CT scans (CTDIvol; mean, 0.33 ± 0.056 mGy) were reconstructed with five algorithms: DLM-stnd (standard kernel), DLM-shrp (sharp kernel), DLIR, ASiR-stnd, and ASiR-shrp. Three thoracic radiologists blinded to the reconstruction algorithms reviewed five sets of 100 images and assessed subjective noise, spatial resolution, distortion artifact, and overall image quality. They selected the most preferred algorithm among five image sets for each case. Image noise and signal-to-noise ratio were measured. Edge-rise-distance was measured at a pulmonary vessel, i.e., the distance between two points where attenuation was 10% and 90% of maximal intravascular intensity. The skewness of attenuation was calculated in homogeneous areas. DLM-stnd, followed by DLIR, showed the best subjective noise on both lung and mediastinal windows, while DLIR yielded the least measured noise (ps < .0001). Compared to DLM-stnd, DLIR showed inferior subjective spatial resolution on lung window and higher edge-rise-distance (ps < .0001). Additionally, DLIR showed the most frequent distortion artifacts and deviated skewness (ps < .0001). DLM-stnd scored the best overall image quality, followed by DLM-shrp and DLIR (mean score 3.89 ± 0.19, 3.68 ± 0.24, and 3.53 ± 0.33; ps < .001). Two among three readers preferred DLM-stnd on both windows. Although DLIR provided the best quantitative noise profile, DLM-stnd showed the best overall image quality with fewer artifacts and was preferred by two among three readers. • A vendor-agnostic deep learning post-processing algorithm applied to ultralow-dose chest CT exhibited the best image quality compared to vendor-specific deep learning algorithm and ASiR techniques. • Two out of three readers preferred a vendor-agnostic deep learning post-processing algorithm in comparison to vendor-specific deep learning algorithm and ASiR techniques. • A vendor-specific deep learning reconstruction algorithm yielded the least image noise, but showed significantly more frequent specific distortion artifacts and increased skewness of attenuation compared to a vendor-agnostic algorithm.

Sensitivity evaluation of the different physical parameterizations schemes in regional climate model RegCM4.5 for simulation of air temperature and precipitation over North and West of Iran

In this research the dynamic downscaling method by Regional Climate Model (RegCM4.5) was used to assess the performance and sensitivity of seasonal simulated North and West of Iran (NI&WI) climate factors to different convection schemes, and transforms the large-scale simulated climate variables into land surface states over the North of Iran (NI) and West of Iran (WI). A 30-year (1986–2015) numerical integration simulation of climate over NI&WI was conducted using the regional climate model RegCM4.5 nested in one-way ERA-Interim reanalysis data. The Grell, Kuo and MIT-Emanuel cumulus convection with Holtslag and University of Washington (UW) planetary boundary layer (PBL) parameterization schemes were applied in the running of RegCM4.5 to test their capability in simulating precipitation and temperature in winter-spring (January–April) over NI and WI. The results demonstrated that the RegCM4.5 model has a good potential for simulating the variables and trend of surface temperature over the NI and WI region. Magnitude of the model bias for land surface temperature over different regions of Iran varies by convection parameterization schemes. In most cases, the root mean square error between post-processed simulated seasonal average temperature and observation value was less than 1 °C, but there is a systematic “cold bias”. In general, with respect to land surface temperature simulations, a better performance is obtained when using post-processing model’s data with Holtslag PBL-Grell and Holtslag PBL-Kuo configuration schemes, compared to the other simulations, over the NI&WI region. Also, the UW PBL convection schemes show a relatively excellent spatial correlations and normalized standard deviations closer to 1 for thirty-year seasonal land surface temperature anomalies over the entire NI&WI region. However, the simulation accuracy of model for precipitation is not as optimal as for temperature. The dominant feature in model simulations is a dry bias with the largest average value (∼1.04 mm/day) over NI region, while the lowest mean bias precipitation (∼−0.47 mm/day), mainly located in WI region. In the comparison of six configuration convection schemes, the Emanuel scheme has been proven to be the most accurate for simulating winter-spring seasonal mean precipitation over NI&WI region. The accuracy of the scheme also showed great difference in simulated station interpolation of precipitation, which urges the improvement for the simulation capability of spatial distribution of precipitation. In general, for seasonal variation of precipitation, the Emanuel convection with two (Holtslag, UW) PBL configuration schemes outperforms with a good correlation score between 0.7−0.8 and normalized standard deviations closer to 1.

Using Conditional Generative Adversarial 3-D Convolutional Neural Network for Precise Radar Extrapolation

Radar echo extrapolation is a basic but essential task in meteorological services. It could provide radar echo prediction results with high spatiotemporal resolution in a computationally efficient way, and effectively enhance the operational system's forecasting capability for meteorological hazards. Traditional methods perform extrapolation by estimating echo motions between contiguous radar data. This strategy is difficult to characterize complex nonlinear meteorological processes effectively, and it is difficult to benefit from large historical data. Recently, machine learning (ML) models have been used for radar echo extrapolation. These methods have effectively improved extrapolation quality in a data-driven way and from the statistical perspective. Although the ML-based methods show excellent performance, they usually produce blurry extrapolations. This leads to underestimating radar echo intensity and making echo lack small-scale details. Moreover, it makes models difficult to predict severe convective hazards. To solve this problem, a two-stage extrapolation model based on 3-D convolutional neural network and conditional generative adversarial network is proposed. These two models form the “pre-extrapolation” and “postprocessing” paradigm. The pre-extrapolation model is trained in the traditional way and performs rough extrapolation. The postprocessing model uses the pre-extrapolation result as input and is trained with the adversarial strategy. It could correct the echo intensity and increase the echo's details. In the experiment, our model could provide more precise radar echo extrapolations than other methods, especially for intense echoes and convective systems, in the data of North China from 2015 to 2016.

Model-Agnostic Post-Processing Based on Recursive Feedback for Medical Image Segmentation

In medical image segmentation, post-processing can effectively improve the performance of a segmentation model. Existing post-processing methods generally require additional training of a post-processing model using training data or designing a post-processing procedure based on a high level of domain knowledge. Their application is limited in many real-world situations due to the lack of prerequisites. In this study, we present a post-processing method that can be applied to any existing segmentation model without requiring the use of training data or domain knowledge. Given a segmentation model of any type, the proposed method improves its prediction based on the recursive feedback mechanism. For a query image, we first obtain its prediction mask by using the segmentation model. Based on the prediction mask, we modify the original image by selectively blurring the area in which the target object is expected to be absent. Subsequently, the modified image is fed again into the model to acquire a refined prediction mask. We repeat this process to obtain multiple prediction masks, which are then combined to yield the final prediction. We verified the effectiveness of the proposed method through experiments using real-world medical datasets.

A SAR Target Image Simulation Method With DNN Embedded to Calculate Electromagnetic Reflection

Electromagnetic (EM) scattering calculation is a very important part of most synthetic aperture radar (SAR) target image simulation methods. It affects the intensity of the radar echo signal to a great extent, thus affecting the quality of the final simulation image. EM reflection models are usually approximate formulas derived under certain assumptions. The errors between these models and the actual situation can cause significant differences between simulated images and real images. To solve this problem, we propose a novel modified SAR target image simulation framework, in which the deep neural network (DNN) is embedded to calculate the EM reflection, so that the DNN can directly learn and fit the EM reflection models from real SAR images. First, the intensity calculation of radar signal in a single reflection is separated from the cumulative calculation of multiple radar reflection signals intensity in each pixel. Thus, the approximate calculation formulas of EM reflection can be replaced with the DNN models. Next, the DNN model is trained with the backpropagation algorithm to learn the actual EM reflection model from real SAR images. Finally, the fitted EM reflection models and an image post-processing model are applied to simulate images of the target under different imaging angles. In the simulation framework, the functions of the neural network models are limited to calculating the reflection coefficient and adding sidelobe and speckle noise. The imaging model is still the original simulation method based on ray tracing, which ensures the correctness and generalization of the simulation method. Experiments show that the proposed simulation method can significantly improve the quality of the simulation image. When the image is normalized to [0, 1], the minimum mean square error between the simulated SAR images and the real images of the Sandia laboratory implementation of cylinders target can reach 0.003. The visualization results of the DNN models show that the fitted reflection coefficient calculation curve and the convolution kernel used for image post-processing are consistent with the laws in the theoretical model. In addition, when the proposed method is used to simulate complex targets, the similarity of simulation images can also be significantly improved.

ScoreNet: A Neural Network-Based Post-Processing Model for Identifying Epileptic Seizure Onset and Offset in EEGs.

We design an algorithm to automatically detect epileptic seizure onsets and offsets from scalp electroencephalograms (EEGs). The proposed scheme consists of two sequential steps: detecting seizure episodes from long EEG recordings, and determining seizure onsets and offsets of the detected episodes. We introduce a neural network-based model called ScoreNet to carry out the second step by better predicting the seizure probability of pre-detected seizure epochs to determine seizure onsets and offsets. A cost function called log-dice loss with a similar meaning to the F1 score is proposed to handle the natural data imbalance inherent in EEG signals signifying seizure events. ScoreNet is then verified on the CHB-MIT Scalp EEG database in combination with several classifiers including random forest, convolutional neural network (CNN), and logistic regression. As a result, ScoreNet improves seizure detection performance over lone epoch-based seizure classification methods; F1 scores increase significantly from 16-37% to 53-70%, and false positive rates per hour decrease from 0.53-5.24 to 0.05-0.61. This method provides clinically acceptable latencies of detecting seizure onset and offset of less than 10 seconds. In addition, an effective latency index is proposed as a metric for detection latency whose scoring considers undetected events to provide better insight into onset and offset detection than conventional time-based metrics.

Automated Segmentation and Severity Analysis of Subdural Hematoma for Patients with Traumatic Brain Injuries.

Detection and severity assessment of subdural hematoma is a major step in the evaluation of traumatic brain injuries. This is a retrospective study of 110 computed tomography (CT) scans from patients admitted to the Michigan Medicine Neurological Intensive Care Unit or Emergency Department. A machine learning pipeline was developed to segment and assess the severity of subdural hematoma. First, the probability of each point belonging to the hematoma region was determined using a combination of hand-crafted and deep features. This probability provided the initial state of the segmentation. Next, a 3D post-processing model was applied to evolve the initial state and delineate the hematoma. The recall, precision, and Dice similarity coefficient of the proposed segmentation method were 78.61%, 76.12%, and 75.35%, respectively, for the entire population. The Dice similarity coefficient was 79.97% for clinically significant hematomas, which compared favorably to an inter-rater Dice similarity coefficient. In volume-based severity analysis, the proposed model yielded an F1, recall, and specificity of 98.22%, 98.81%, and 92.31%, respectively, in detecting moderate and severe subdural hematomas based on hematoma volume. These results show that the combination of classical image processing and deep learning can outperform deep learning only methods to achieve greater average performance and robustness. Such a system can aid critical care physicians in reducing time to intervention and thereby improve long-term patient outcomes.

Automatic bridge crack identification from concrete surface using ResNeXt with postprocessing

In this study, the bridge concrete structure is taken as the research object, and the real image is used for crack identification. In structural engineering, surface cracks are the main indexes of durability and service performance of structures. Artificial visual inspection is often considered ineffective in terms of cost, safety, evaluation accuracy, and reliability. In this article, a simple, high-classification framework based on ResNeXt with postprocessing (ResNeXt+PP) model is provided to effectively identify concrete cracks. During the training phase of the method, image binarization approach is used to extract the candidate crack regions. It is difficult to automatic identify cracks from images containing actual cracks and noises, especially, shadows, stains, masses, and holesoften occur in concrete surfaces. Thereafter, classification models are constructed based on ResNeXt+PP module. Based on the new concrete surface images including cracks and noncracks, the obtained methods for crack identification are compared quantitatively and qualitatively. Besides, the five complete raw images are used to study the robustness and practicability of the method. The binary transformation process based on a binarization method of adaptive crack width is adopted to identify crack pixels in subimages. Results show that the trained improved ResNeXt+PP can automatically detect cracks and noncracks in the raw image. The obtained results that the method is superior to multiple methods and the application prospect of autonomous concrete structure driver for bridge detection robot are presented.

Wind Speed Forecast Based on Post-Processing of Numerical Weather Predictions Using a Gradient Boosting Decision Tree Algorithm

With the large-scale development of wind energy, wind power forecasting plays a key role in power dispatching in the electric power grid, as well as in the operation and maintenance of wind farms. The most important technology for wind power forecasting is forecasting wind speed. The current mainstream methods for wind speed forecasting involve the combination of mesoscale numerical meteorological models with a post-processing system. Our work uses the WRF model to obtain the numerical weather forecast and the gradient boosting decision tree (GBDT) algorithm to improve the near-surface wind speed post-processing results of the numerical weather model. We calculate the feature importance of GBDT in order to find out which feature most affects the post-processing wind speed results. The results show that, after using about 300 features at different height and pressure layers, the GBDT algorithm can output more accurate wind speed forecasts than the original WRF results and other post-processing models like decision tree regression (DTR) and multi-layer perceptron regression (MLPR). Using GBDT, the root mean square error (RMSE) of wind speed can be reduced from 2.7–3.5 m/s in the original WRF result by 1–1.5 m/s, which is better than DTR and MLPR. While the index of agreement (IA) can be improved by 0.10–0.20, correlation coefficient be improved by 0.10–0.18, Nash–Sutcliffe efficiency coefficient (NSE) be improved by −0.06–0.6. It also can be found that the feature which most affects the GBDT results is the near-surface wind speed. Other variables, such as forecast month, forecast time, and temperature, also affect the GBDT results.

AR-GARCH with Exogenous Variables as a Postprocessing Model for Improving Streamflow Forecasts

AbstractThe common strategy for real-time streamflow forecasting involves a precalibrated rainfall-runoff model for streamflow simulation together with a statistical postprocessing model of simulat...

Progressive Bilateral-Context Driven Model for Post-Processing Person Re-Identification

Most existing person re-identification methods compute pairwise similarity by extracting robust visual features and learning the discriminative metric. Owing to visual ambiguities, these content-based methods that determine the pairwise relationship only based on the similarity between them, inevitably produce a suboptimal ranking list. Instead, the pairwise similarity can be estimated more accurately along the geodesic path of the underlying data manifold by exploring the rich contextual information of the sample. In this paper, we propose a lightweight post-processing person re-identification method in which the pairwise measure is determined by the relationship between the sample and the counterpart's context in an unsupervised way. We translate the point-to-point comparison into the bilateral point-to-set comparison. The sample's context is composed of its neighbor samples with two different definition ways: the first order context and the second order context, which are used to compute the pairwise similarity in sequence, resulting in a progressive post-processing model. The experiments on four large-scale person re-identification benchmark datasets indicate that (1) the proposed method can consistently achieve higher accuracies by serving as a post-processing procedure after the content-based person re-identification methods, showing its state-of-the-art results, (2) the proposed lightweight method only needs about 6 milliseconds for optimizing the ranking results of one sample, showing its high-efficiency. Code is available at: https://github.com/123ci/PBCmodel .

Vertical profiles of wind gust statistics from a regional reanalysis using multivariate extreme value theory

Abstract. Many applications require wind gust estimates at very different atmospheric height levels. For example, the renewable energy sector is interested in wind and gust predictions at the hub height of a wind power plant. However, numerical weather prediction models typically only derive estimates for wind gusts at the standard measurement height of 10 m above the land surface. Here, we present a statistical post-processing method to derive a conditional distribution for hourly peak wind speed as a function of height. The conditioning variables are taken from the COSMO-REA6 regional reanalysis. The post-processing method was trained using peak wind speed observations at five vertical levels between 10 and 250 m from the Hamburg Weather Mast. The statistical post-processing method is based on a censored generalized extreme value (cGEV) distribution with non-homogeneous parameters. We use a least absolute shrinkage and selection operator to select the most informative variables. Vertical variations of the cGEV parameters are approximated using Legendre polynomials, such that predictions may be derived at any desired vertical height. Further, the Pickands dependence function is used to assess dependencies between gusts at different heights. The most important predictors are the 10 m gust diagnostic, the barotropic and the baroclinic mode of absolute horizontal wind speed, the mean absolute horizontal wind at 700 hPa, the surface pressure tendency, and the lifted index. Proper scores show improvements of up to 60 % with respect to climatology, especially at higher vertical levels. The post-processing model with a Legendre approximation is able to provide reliable predictions of gusts' statistics at non-observed intermediate levels. The strength of dependency between gusts at different levels is non-homogeneous and strongly modulated by the vertical stability of the atmosphere.

Improving NCEP’s global-scale wave ensemble averages using neural networks

The quality of metocean forecasts at longer forecast ranges has a significant impact on maritime safety and offshore operations. A nonlinear ensemble averaging technique is demonstrated using neural networks applied to one year (2017) of Global ocean Wave Ensemble forecast System (GWES) data provided by NCEP. Post-processing algorithms are developed based on multilayer perceptron neural networks (NN) trained with altimeter data to improve the global forecast skill, from nowcast to forecast ranges up to 10 days, including significant wave height (Hs) and wind speed (U10). NNs are applied as an alternative to the typical use of the arithmetic ensemble mean (EM). NN models are constructed using six variables sourced from 21 ensemble members, plus latitude, sin/cos of longitude, sin/cos of time, forecast lead time, and GWES cycle. The NN outputs are the residues of Hs and U10, i.e., the difference from the EM to the observations. One hidden (intermediate) layer is evaluated in terms of the optimum number of neurons (complexity) to map the given problem. The sensitivity test considered 26 different numbers of neurons, 10 seeds for initial conditions, and 3 equally-divided datasets; for a total of 780 NN experiments. Assessments using 2,507,099 paired satellite/GWES fields show that a simple NN model with few neurons is able to reduce the systematic errors for short-range forecasts, while a NN with more neurons is required to minimize the scatter error at longer forecast ranges. The novel method shows that one single NN model with 140 neurons is able to improve the error metrics for the whole globe while covering all forecast ranges analyzed. The bias of the widely used EM of GWES that varies from -10% to 10% for Hs compared to altimeters can be reduced to values within 5%. The RMSE of day-10 forecasts from the NN simulations indicated a gain of two days in predictability when compared to the EM, using a reasonably simple post-processing model with low computational cost.

Impervious Surface Extraction by Linear Spectral Mixture Analysis with Post-Processing Model

Accurate estimations of impervious surface areas are essential for urban planning development. Linear spectral mixture analysis (LSMA) is commonly adopted to extract the impervious surface (IS) fraction in a mixed pixel at the subpixel scale. However, owing to errors in the spectra of pure pixels selected from remote sensing images, incorrect fractions of different land cover types often emerge after unmixing. In this study, two Landsat 8 Operational Land Imager (OLI) images—acquired on 20 September 2019 (Path/Row: 121/44) and 14 November 2019 (Path/Row: 122/44)—of Guangzhou and Shenzhen were unmixed by LSMA using spectral indices in endmember selection. A post-processing model using the Dry Bare-soil Index (DBSI) and Normalized Difference Vegetation Index (NDVI) as thresholds was established to improve the IS fraction of the LSMA result. Comparative analysis reveals that LSMA with the post-processing model achieves better performance for IS fraction extraction (R2 = 0.910 and 0.926 and root mean square error [RMSE] = 10.08% and 10.83% for Guangzhou and Shenzhen, respectively), and the distribution of IS is basically consistent with the IS of the actual areas. The post-processing model solves the problem of overestimation of pervious surface and underestimation of impervious surface.