Heidke Skill Score Research Articles

Precipitation Retrieval over the Tibetan Plateau from the Geostationary Orbit—Part 1: Precipitation Area Delineation with Elektro-L2 and Insat-3D

The lack of long term and well distributed precipitation observations on the Tibetan Plateau (TiP) with its complex terrain raises the need for other sources of precipitation data for this area. Satellite-based precipitation retrievals can fill those data gaps. Before precipitation rates can be retrieved from satellite imagery, the precipitating area needs to be classified properly. Here, we present a feasibility study of a precipitation area delineation scheme for the TiP based on multispectral data with data fusion from the geostationary orbit (GEO, Insat-3D and Elektro-L2) and a machine learning approach (Random Forest, RF). The GEO data are used as predictors for the RF model, extensively validated by independent GPM (Global Precipitation Measurement Mission) IMERG (Integrated Multi-satellitE Retrievals for GPM) gauge calibrated microwave (MW) best-quality precipitation estimates. To improve the RF model performance, we tested different optimization schemes. Here, we find that (1) using more precipitating pixels and reducing the amount of non-precipitating pixels during training greatly improved the classification results. The accuracy of the precipitation area delineation also benefits from (2) changing the temporal resolution into smaller segments. We particularly compared our results to the Infrared (IR) only precipitation product from GPM IMERG and found a markedly improved performance of the new multispectral product (Heidke Skill Score (HSS) of 0.19 (IR only) compared to 0.57 (new multispectral product)). Other studies with a precipitation area delineation obtained a probability of detection (POD) of 0.61, whereas our POD is comparable, with 0.56 on average. The new multispectral product performs best (worse) for precipitation rates above the 90th percentile (below the 10th percentile). Our results point to a clear strategy to improve the IMERG product in the absence of MW radiances.

MLC-LSTM: Exploiting the Spatiotemporal Correlation between Multi-Level Weather Radar Echoes for Echo Sequence Extrapolation.

Weather radar echo is the data detected by the weather radar sensor and reflects the intensity of meteorological targets. Using the technique of radar echo extrapolation, which is the prediction of future echoes based on historical echo observations, the approaching short-term weather conditions can be forecasted, and warnings can be raised with regard to disastrous weather. Recently, deep learning based extrapolation methods have been proposed and show significant application potential. However, there are two limitations of existing extrapolation methods which should be considered. First, few methods have investigated the impact of the evolutionary process of weather systems on extrapolation accuracy. Second, current deep learning methods usually encounter the problem of blurry echo prediction as extrapolation goes deeper. In this paper, we aim to address the two problems by proposing a Multi-Level Correlation Long Short-Term Memory (MLC-LSTM) and integrate the adversarial training into our approach. The MLC-LSTM can exploit the spatiotemporal correlation between multi-level radar echoes and model their evolution, while the adversarial training can help the model extrapolate realistic and sharp echoes. To train and test our model, we build a real-life multi-level weather radar echoes dataset based on raw CINRAD/SA radar observations provided by the National Meteorological Information Center, China. Extrapolation experiments show that our model can accurately forecast the motion and evolution of an echo while keeping the predicted echo looking realistic and fine-grained. For quantitative evaluation on probability of detection (POD), false alarm rate (FAR), critical success index (CSI), and Heidke skill score (HSS) metrics, our model can reach average scores of 0.6538 POD, 0.2818 FAR, 0.5348 CSI, and 0.6298 HSS, respectively when extrapolating 15 echoes into the future, which outperforms the current state-of-the-art extrapolation methods. Both the qualitative and quantitative experimental results demonstrate the effectiveness of our model, suggesting that it can be effectively applied to operational weather forecasting practice.

Dissecting Performances of PERSIANN-CDR Precipitation Product over Huai River Basin, China

Satellite-based precipitation products, especially those with high temporal and spatial resolution, constitute a potential alternative to sparse rain gauge networks for multidisciplinary research and applications. In this study, the validation of the 30-year Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks-Climate Data Record (PERSIANN-CDR) daily precipitation dataset was conducted over the Huai River Basin (HRB) of China. Based on daily precipitation data from 182 rain gauges, several continuous and categorical validation statistics combined with bias and error decomposition techniques were employed to quantitatively dissect the PERSIANN-CDR performance on daily, monthly, and annual scales. With and without consideration of non-rainfall data, this product reproduces adequate climatologic precipitation characteristics in the HRB, such as intra-annual cycles and spatial distributions. Bias analyses show that PERSIANN-CDR overestimates daily, monthly, and annual precipitation with a regional mean percent total bias of 11%. This is related closely to the larger positive false bias on the daily scale, while the negative non-false bias comes from a large underestimation of high percentile data despite overestimating lower percentile data. The systematic sub-component (error from high precipitation), which is independent of timescale, mainly leads to the PERSIANN-CDR total Mean-Square-Error (TMSE). Moreover, the daily TMSE is attributed to non-false error. The correlation coefficient (R) and Kling–Gupta Efficiency (KGE) respectively suggest that this product can well capture the temporal variability of precipitation and has a moderate-to-high overall performance skill in reproducing precipitation. The corresponding capabilities increase from the daily to annual scale, but decrease with the specified precipitation thresholds. Overall, the PERSIANN-CDR product has good (poor) performance in detecting daily low (high) rainfall events on the basis of Probability of Detection, and it has a False Alarm Ratio of above 50% for each precipitation threshold. The Equitable Threat Score and Heidke Skill Score both suggest that PERSIANN-CDR has a certain ability to detect precipitation between the second and eighth percentiles. According to the Hanssen–Kuipers Discriminant, this product can generally discriminate rainfall events between two thresholds. The Frequency Bias Index indicates an overestimation (underestimation) of precipitation totals in thresholds below (above) the seventh percentile. Also, continuous and categorical statistics for each month show evident intra-annual fluctuations. In brief, the comprehensive dissection of PERSIANN-CDR performance reported herein facilitates a valuable reference for decision-makers seeking to mitigate the adverse impacts of water deficit in the HRB and algorithm improvements in this product.

A proposed hybrid rainfall simulation model: bootstrap aggregated classification tree–artificial neural network (BACT-ANN) for the Langat River Basin, Malaysia

AbstractClimate change is a global issue posing threats to the human population and water systems. As Malaysia experiences a tropical climate with intense rainfall occurring throughout the year, accurate rainfall simulations are particularly important to provide information for climate change assessment and hydrological modelling. An artificial intelligence-based hybrid model, the bootstrap aggregated classification tree–artificial neural network (BACT-ANN) model, was proposed for simulating rainfall occurrences and amounts over the Langat River Basin, Malaysia. The performance of this proposed BACT-ANN model was evaluated and compared with the stochastic non-homogeneous hidden Markov model (NHMM). The observed daily rainfall series for the years 1975–2012 at four rainfall stations have been selected. It was found that the BACT-ANN model performed better however, with slight underproductions of the wet spell lengths. The BACT-ANN model scored better for the probability of detection (POD), false alarm rate (FAR) and the Heidke skill score (HSS). The NHMM model tended to overpredict the rainfall occurrence while being less capable with the statistical measures such as distribution, equality, variance and statistical correlations of rainfall amount. Overall, the BACT-ANN model was considered the more effective tool for the purpose of simulating the rainfall characteristics in Langat River Basin.

Validation of Inner Magnetosphere Particle Transport and Acceleration Model (IMPTAM) With Long‐Term GOES MAGED Measurements of keV Electron Fluxes at Geostationary Orbit

AbstractSurface charging by keV (kiloelectron Volt) electrons can pose a serious risk for satellites. There is a need for physical models with the correct and validated dynamical behavior. The 18.5‐month (2013–2015) output from the continuous operation online in real time as a nowcast of the Inner Magnetosphere Particle Transport and Acceleration Model (IMPTAM) is compared to the GOES 13 MAGnetospheric Electron Detector (MAGED) data for 40, 75, and 150 keV energies. The observed and modeled electron fluxes were organized by Magnetic Local Time (MLT) and IMPTAM driving parameters; the observed Interplanetary Magnetic Field (IMF) BZ, BY, and |B|; the solar wind speed VSW; the dynamic pressure PSW; and Kp and SYM‐H indices. The peaks for modeled fluxes are shifted toward midnight, but the ratio between the observed and modeled fluxes at around 06 MLT is close to 1. All the statistical patterns exhibit very similar features with the largest differences of about 1 order of magnitude at 18–24 MLT. Based on binary event analysis, 20–78% of threshold crossings are reproduced, but Heidke skill scores are low. The modeled fluxes are off by a factor of 2 in terms of the median symmetric accuracy. The direction of the error varies with energy: overprediction by 50% for 40 keV, overprediction by 2 for 75 keV, and underprediction by 18% for 150 keV. The revealed discrepancies are due to the boundary conditions developed for ions but used for electrons, absence of substorm effects, representations of electric and magnetic fields which can result in not enough adiabatic acceleration, and simple models for electron lifetimes.

Prediction Model for Solar Energetic Proton Events: Analysis and Verification

AbstractSolar energetic protons (SEPs) can cause radiation damage to satellites. The Space Environment Prediction Center of China defines the start of a SEP event as the time when three consecutive instances of the 5‐min‐averaged integral flux of >10 MeV protons equal or exceed the 10 proton flux unit threshold. In this study, we analyzed the 5‐min‐averaged soft X‐ray flux and the differential and integral proton flux obtained by the Geostationary Operational Environmental Satellites for the period from January 1990 to September 2017 and developed SEP prediction models and their products. The statistical models take one or two data products and the given thresholds as the predictor and predict whether a SEP event will occur or not in the next 24 hr. The quality of the forecast models was measured by comparing the model results against certain verification metrics. Our model taking the products of 5‐min‐averaged integral flux of >10 MeV protons and long wavelength of soft X‐ray flux as predictors can provide a probability of detection of 0.80 (152/190), a false alarm ratio of 0.26 (53/205), and an average warning time of 2.6 hr for the correctly predicted events. In addition, the model can provide a critical success index of 0.63, a Gilbert skill score of 0.62, and a Heidke skill score of 0.76.

Evaluation of the Global and Regional Assimilation and Prediction System for Predicting Sea Fog over the South China Sea

In the South China Sea, sea fog brings severe disasters every year, but forecasters have yet to implement an effective sea-fog forecast. To address this issue, we test a liquid-water-content-only (LWC-only) operational sea-fog prediction method based on a regional mesoscale numerical model with a horizontal resolution of about 3 km, the Global and Regional Assimilation and Prediction System (GRAPES), hereafter GRAPES-3km. GRAPES-3km models the LWC over the sea, from which we infer the visibility that is then used to identify fog. We test the GRAPES-3km here against measurements in 2016 and 2017 from coastal-station observations, as well as from buoy data, data from the Integrated Observation Platform for Marine Meteorology, and retrieved fog and cloud patterns from Himawari-8 satellite data. For two cases that we examine in detail, the forecast region of sea fog overlaps well with the multi-observational data within 72 h. Considering forecasting for 0–24 h, GRAPES-3km has a 2-year-average equitable threat score (ETS) of 0.20 and a Heidke skill score (HSS) of 0.335, which is about 5.6% (ETS) and 6.4% (HSS) better than our previous method (GRAPES-MOS). Moreover, the stations near the particularly foggy region around the Leizhou Peninsula have relatively high forecast scores compared to other sea areas. Overall, the results show that GRAPES-3km can roughly predict the formation, evolution, and dissipation of sea fog on the southern China coast.

Forest fire risk indices and zoning of hazardous areas in Sorocaba, São Paulo state, Brazil

This study compares the performance of three fire risk indices for accuracy in predicting fires in semi-deciduous forest fragments, creates a fire risk map by integrating historical fire occurrences in a probabilistic density surface using the Kernel density estimator (KDE) in the municipality of Sorocaba, Sao Paulo state, Brazil. The logarithmic Telicyn index, Monte Alegre formula (MAF) and enhanced Monte Alegre formula (MAF+) were employed using data for the period 1 January 2005 to 31 December 2016. Meteorological data and numbers of fire occurrences were obtained from the National Institute of Meteorology (INMET) and the Institute for Space Research (INPE), respectively. Two performance measures were calculated: Heidke skill score (SS) and success rate (SR). The MAF+ index was the most accurate, with values of SS and SR of 0.611% and 62.8%, respectively. The fire risk map revealed two most susceptible areas with high (63 km2) and very high (47 km2) risk of fires in the municipality. Identification of the best risk index and the generation of fire risk maps can contribute to better planning and cost reduction in preventing and fighting forest fires.

Establishing a daily rainfall occurrence simulation model for the Langat River catchment, Malaysia

For the study of water resources of a catchment, an immediate task would be to establish a good model for predicting the probable daily rainfall occurrence and rainfall amount. This study presents the simulation of daily rainfall occurrence using the generalized linear model (GLM), the non-homogeneous hidden Markov model (NHMM) and the bootstrap aggregated classification tree (BACT) model. The major challenge of NHMM is the determination of optimum number of hidden states, which can be achieved using the Bayesian information criterion score. While the determination of number of grown tree is another challenge for BACT model, this critical task can be achieved with the help of out-of-bag classification error. Both the NHMM and BACT model outperformed the GLM to capture the rainfall persistence and spell lengths distribution. Through the validation phase, the BACT model exhibited better performance with the higher indices of probability of detection, critical success index, Heidke skill score and Peirce skill score, than other models. The prediction ability of the NHMM is equivalent to an unskilled random forecast with the skill scores nearly equal to zero. At the end, the BACT model was recommended as the appropriate daily rainfall occurrence model for this study.

A Multi-Categorical Probabilistic Approach for Short-Term Bike Sharing Usage Prediction

Most existing bike usage prediction studies aim at building models to fit continuous bike usage data rather than categorical data, which may result in an over-fitting problem and therefore reduce the potential of the model to capture more generalized trends in bike usage predictions. This study explores a multi-categorical probabilistic approach for sharing bike demand prediction. In order to overcome the weakness of using single point measurements to describe bike usage conditions, we prepare three alternatives to capture the range, local variation, and trend of bike usage over a short-time period. The suitable indicator variables are determined based on the Principal Component Analysis (PCA) results. The Gaussian Mixture Models (GMM) is adopted to cluster homogeneous bike usage states. Then, a Markov chain model is developed based on the identified states to forecast the categorical changes of bike usage. Finally, to examine the effectiveness of the proposed approach, the persistence model is employed as a benchmark and two measures-Percent Correct (PC) and Heidke Skill Score (HSS), are introduced to quantify categorical data prediction performance. The results show that the proposed approach is able to offer high accuracy, skill, reliability, and discrimination at suitable prediction intervals.

Application of the Deep Convolutional Neural Network to the Forecast of Solar Flare Occurrence Using Full-disk Solar Magnetograms

Abstract In this study, we present the application of the Convolutional Neural Network (CNN) to the forecast of solar flare occurrence. For this, we consider three CNN models (two pretrained models, AlexNet and GoogLeNet, and one newly proposed model). Our inputs are SOHO/Michelson Doppler Imager (from 1996 May to 2010 December) and SDO/Helioseismic and Magnetic Imager (from 2011 January to 2017 June) full-disk magnetograms at 00:00 UT. Model outputs are the “Yes or No” of daily flare occurrence (C, M, and X classes) and they are compared with GOES observations. We train the models using the input data and observations from 1996 to 2008, covering the entire solar cycle 23, and test them using the data sets from 2009 to 2017, covering solar cycle 24. Then we compare the results of the CNN models with those of three previous flare forecast models in view of statistical scores. The major results from this study are as follows. First, we successfully apply CNN to the full-disk solar magnetograms without any preprocessing or feature extraction. Second, the results of our CNN models are slightly better in Heidke skill score and true skill statistics, and considerably better in false alarm ratio (FAR) and critical success index than the previous solar flare forecasting models. Third, our proposed model has better values of all statistical scores except for FAR, than the other two pretrained models. Our results indicate a sufficient possibility that deep learning methods can improve the capability of the solar flare forecast as well as similar types of forecast problems.

A Versatile Method for Ice Particle Habit Classification Using Airborne Imaging Probe Data

AbstractA versatile method to automatically classify ice particle habit from various airborne optical array probes is presented. The classification is achieved using a multinomial logistic regression model. For each airborne probe, the model determines the particle habit (among six classes) based on a large set of geometrical and textural descriptors extracted from the two‐dimensional image of a particle. The technique is applied and evaluated using three probes with significantly different specifications: the high volume precipitation spectrometer, the two‐dimensional stereo probe, and the cloud particle imager. Performance and robustness of the method are assessed using standard machine learning tools on the basis of thousands of images manually labeled for each of the considered probes. The three classifiers show good performance characterized by overall accuracies and Heidke skill scores above 90%. Depending on the application and user preferences, the classification scheme can be easily adapted. For a more precise output, intraclass subclassification can be achieved in a nested fashion, illustrated here with columnar crystals and aggregates. A comparative study of the classification output obtained with the three probes is presented for two aircraft flight periods selected when the three probes were operating together. Results are globally consistent in term of proportions of habit identified (once blurry and partial images have been automatically discarded). A perfect agreement is not expected as the three considered probes are sensitive to different particle size range.

Model Evaluation Guidelines for Geomagnetic Index Predictions

AbstractGeomagnetic indices are convenient quantities that distill the complicated physics of some region or aspect of near‐Earth space into a single parameter. Most of the best‐known indices are calculated from ground‐based magnetometer data sets, such as Dst, SYM‐H, Kp, AE, AL, and PC. Many models have been created that predict the values of these indices, often using solar wind measurements upstream from Earth as the input variables to the calculation. This document reviews the current state of models that predict geomagnetic indices and the methods used to assess their ability to reproduce the target index time series. These existing methods are synthesized into a baseline collection of metrics for benchmarking a new or updated geomagnetic index prediction model. These methods fall into two categories: (1) fit performance metrics such as root‐mean‐square error and mean absolute error that are applied to a time series comparison of model output and observations and (2) event detection performance metrics such as Heidke Skill Score and probability of detection that are derived from a contingency table that compares model and observation values exceeding (or not) a threshold value. A few examples of codes being used with this set of metrics are presented, and other aspects of metrics assessment best practices, limitations, and uncertainties are discussed, including several caveats to consider when using geomagnetic indices.

Subseasonal Prediction of Wintertime East Asian Temperature Based on Atmospheric Teleconnections

AbstractA composite-based statistical model utilizing Northern Hemisphere teleconnection patterns is developed to predict East Asian wintertime surface air temperature for lead times out to 6 weeks. The level of prediction is determined by using the Heidke skill score. The prediction skill of the statistical model is compared with that of hindcast simulations by a climate model, Global Seasonal Forecast System, version 5. When employed individually, three teleconnections (i.e., the east Atlantic/western Russian, Scandinavian, and polar/Eurasian teleconnection patterns) are found to provide skillful predictions for lead times beyond 4–5 weeks. When information from the teleconnections and the long-term linear trend are combined, the statistical model outperforms the climate model for lead times beyond 3 weeks, especially during those times when the teleconnections are in their active phases.

Multi-Variable Classification Approach for the Detection of Lightning Activity Using a Low-Cost and Portable X Band Radar

This work proposes a multi-parameter method for the detection of cloud-to-ground stroke rate (SRCG) associated to convective cells, based on the measurements of a low-cost single-polarization X-band weather radar. To train and test our procedure, we built up a multi-year dataset, collecting 1575 radar reflectivity volumes that were acquired in the pilot study area of Naples metropolitan environment matched with the LIghtning NETwork (LINET) strokes and meteorological in-situ data. Three radar-based variables are extracted simultaneously for each rain cell and properly merged together, using “ad hoc” classification methods, to produce an estimation of the expected lightning activity for each rain cell. These variables, proxies of mixed-phase particles and ice amount into a convective cell, are combined into a single label to cluster the SRCG into two categories: SRCG = 0 (no production of strokes) or SRCG > 0 (stroke production), respectively. Overall, the main results are comparable with those that were obtained from more advanced radar systems, showing a Critical Success Index of 0.53, an Equitable Threat Score of 0.34, a Frequency Bias Index of 1.00, a Heidke Skill Score of 0.42, a Hanssen-Kuiper Skill Score of 0.42, and an area under the curve of probability of detection as a function of false alarm rate (usually referred as ROC curve) equal to 0.78. The developed technique, although with some limitations, outperforms those based on the use of single stroke proxy parameters.

Fast Cloud Segmentation Using Convolutional Neural Networks

Information about clouds is important for observing and predicting weather and climate as well as for generating and distributing solar power. Most existing approaches extract cloud information from satellite data by classifying individual pixels instead of using closely integrated spatial information, ignoring the fact that clouds are highly dynamic, spatially continuous entities. This paper proposes a novel cloud classification method based on deep learning. Relying on a Convolutional Neural Network (CNN) architecture for image segmentation, the presented Cloud Segmentation CNN (CS-CNN), classifies all pixels of a scene simultaneously rather than individually. We show that CS-CNN can successfully process multispectral satellite data to classify continuous phenomena such as highly dynamic clouds. The proposed approach produces excellent results on Meteosat Second Generation (MSG) satellite data in terms of quality, robustness, and runtime compared to other machine learning methods such as random forests. In particular, comparing CS-CNN with the CLAAS-2 cloud mask derived from MSG data shows high accuracy (0.94) and Heidke Skill Score (0.90) values. In contrast to a random forest, CS-CNN produces robust results and is insensitive to challenges created by coast lines and bright (sand) surface areas. Using GPU acceleration, CS-CNN requires only 25 ms of computation time for classification of images of Europe with 508 × 508 pixels.

Real‐Time SWMF at CCMC: Assessing the Dst Output From Continuous Operational Simulations

AbstractThe ground‐based magnetometer index of Dst is a commonly used measure of near‐Earth current systems, in particular the storm time inner magnetospheric current systems. The ability of a large‐scale, physics‐based model to reproduce, or even predict, this index is therefore a tangible measure of the overall validity of the code for space weather research and space weather operational usage. Experimental real‐time simulations of the Space Weather Modeling Framework (SWMF) are conducted at the Community Coordinated Modeling Center (CCMC). Presently, two configurations of the SWMF are running in real time at CCMC, both focusing on the geospace modules, using the Block Adaptive Tree Solar wind‐type Roe Upwind Solver magnetohydrodynamic model, the Ridley Ionosphere Model, and with and without the Rice Convection Model. While both have been running for several years, nearly continuous results are available since April 2015. A 27‐month interval through July 2017 is used for a quantitative assessment of Dst from the model output compared against the Kyoto real‐time Dst. Quantitative measures are presented to assess the goodness of fit including contingency tables and a receiver operating characteristic curve. It is shown that the SWMF run with the inner magnetosphere model is much better at reproducing storm time values, with a correlation coefficient of 0.69, a prediction efficiency of 0.41, and Heidke skill score of 0.57 (for a −50‐nT threshold). A comparison of real‐time runs with and without the inner magnetospheric drift physics model reveals that nearly all of the storm time Dst signature is from current systems related to kinetic processes on closed magnetic field lines.

Using Machine Learning Methods to Forecast if Solar Flares Will Be Associated with CMEs and SEPs

Abstract Among the eruptive activity phenomena observed on the Sun, those that threaten human technology the most are flares with associated coronal mass ejections (CMEs) and solar energetic particles (SEPs). Flares with associated CMEs and SEPs are produced by magnetohydrodynamical processes in magnetically active regions (ARs) on the Sun. However, these ARs do not only produce flares with associated CMEs and SEPs, they also lead to flares and CMEs, which are not associated with any other event. In an attempt to distinguish flares with associated CMEs and SEPs from flares and CMEs, which are unassociated with any other event, we investigate the performances of two machine learning algorithms. To achieve this objective, we employ support vector machines (SVMs) and multilayer perceptrons (MLPs) using data from the Space Weather Database of Notification, Knowledge, Information of NASA Space Weather Center, the Geostationary Operational Environmental Satellite, and the Space-Weather Heliospheric and Magnetic Imager Active Region Patches. We show that True Skill Statistics (TSS) and Heidke Skill Scores (HSS) calculated for SVMs are slightly better than those from the MLPs. We also show that the forecasting time frame of 96 hr provides the best results in predicting if a flare will be associated with CMEs and SEPs (TSS = 0.92 ± 0.09 and HSS = 0.92 ± 0.08). Additionally, we obtain the maximum TSS and HSS values of 0.91 ± 0.06 for predicting that a flare will not be associated with CMEs and SEPs for the 36 hr forecast window, while the 108 hr forecast window gives the maximum TSS and HSS values for predicting that CMEs will not be accompanying any events (TSS = HSS = 0.98 ± 0.02).

An Event-Based Verification Scheme for the Real-Time Flare Detection System at Kanzelh\xf6he Observatory

In the framework of the Space Situational Awareness program of the European Space Agency (ESA/SSA), an automatic flare detection system was developed at Kanzelhöhe Observatory (KSO). The system has been in operation since mid-2013. The event detection algorithm was upgraded in September 2017. All data back to 2014 was reprocessed using the new algorithm. In order to evaluate both algorithms, we apply verification measures that are commonly used for forecast validation. In order to overcome the problem of rare events, which biases the verification measures, we introduce a new event-based method. We divide the timeline of the Hupalpha observations into positive events (flaring period) and negative events (quiet period), independent of the length of each event. In total, 329 positive and negative events were detected between 2014 and 2016. The hit rate for the new algorithm reached 96% (just five events were missed) and a false-alarm ratio of 17%. This is a significant improvement of the algorithm, as the original system had a hit rate of 85% and a false-alarm ratio of 33%. The true skill score and the Heidke skill score both reach values of 0.8 for the new algorithm; originally, they were at 0.5. The mean flare positions are accurate within {pm},1 heliographic degree for both algorithms, and the peak times improve from a mean difference of 1.7pm 2.9~mbox{minutes} to 1.3pm 2.3~mbox{minutes}. The flare start times that had been systematically late by about 3 minutes as determined by the original algorithm, now match the visual inspection within -0.47pm 4.10~mbox{minutes}.

Application of Hidden Markov Model for avalanche danger simulations for road sectors in North-West Himalaya

Hidden Markov Model (HMM) has been developed for avalanche warning on 10 different road sectors in Pir-Panjal and Great Himalayan mountain ranges of North-West Himalaya. The model uses a data set of nine snow and meteorological variables—average air temperature, snow temperature index, snow drift index, snowfall in 24 h, snowfall in 48 h, snow water equivalent, snowfall intensity, standing snow and snowpack settlement collected during past 20 winters (1992–2012). The HMM is composed of four observations derived from the model input variables and four state variables. The state variables of the model are four levels of avalanche danger (No, Low, Medium and High). Single HMM has been developed to provide avalanche warning for both direct and delayed/wet avalanches with a lead time of two days. The HMM has been validated with (Case-1) and without (Case-2) incorporating delayed/wet avalanches using data collected during four winters (2012–2016) and compared with official Avalanche Warning Bulletin issued by Snow and Avalanche Study Establishment during these winters. The model has been validated through computation of accuracy measures such as percent correct (PC), bias, false alarm rate, probability of detection and Heidke Skill Score. The PC of the HMM for different stations for Case-1 varies from 80.1 to 98.6% for day-1 and 81.2 to 98.3% for day-2 and that for Case-2 from 82.2 to 98.6% for day-1 and 83.3 to 98.3% for day-2.