Spatiotemporal Sequence Research Articles

The Statistical Estimation Averaging Method to Express the Effective Electromagnetic Parameters over a Planar Information Meta-Surface

The electromagnetic scattering of a meta-surface, formed by a honeycomb substrate or periodical arranged meta-materials, has great meanings to communication technologies. In a conventional treatment to investigate the properties of these interfaces, either a variational approach or series expansions had been applied, instead of estimating the effective permittivity or permeability by the mean value of its spatial discretized statistics due to spatiotemporal fusion. Following this, this study has re-examined the problem by the Statistic Estimation Averaging method (SEAm), through the transferred conditional structural Probability Density Functions (PDFs) to realize the structural determinations by homogenization. The parameters estimated by SEAm, which exploited the concept of a homogenized medium to express properties of a structural complex medium, has been verified of validity and accuracy by comparing with the measured results of a honeycomb structure. The method can be extended to estimate the parameters of an equivalent surface, such as randomly scattering from information metamaterials. As a new wireless communication relay technology, considering that information metamaterials can modulate the electromagnetic characteristics of communication links and wireless channels simultaneously by means of spatiotemporal sequence coding, the study also gives a preliminary proposition on state estimation method of information meta-surface, which would interpret the modulation effect of wireless channels caused by its inhomogeneity of antenna wavefront by statistical estimation average information entropy.

Prediction of Oil Saturation during Water and Gas Injection Using Controllable Convolutional Long Short-Term Memory

Oil saturation is a kind of spatiotemporal sequence that changes dynamically with time, and it is affected not only by the reservoir properties, but also by the injection–production parameters. When predicting oil saturation during water and gas injection, the influence of time, space and injection–production parameters should be considered. Aiming at this issue, a prediction method based on a controllable convolutional long short-term memory network (Ctrl-CLSTM) is proposed in this paper. The Ctrl-CLSTM is an unsupervised learning model whose input is the previous spatiotemporal sequence together with the controllable factors of corresponding moments, and the output is the sequence to be predicted. In this way, future oil saturation can be generated from the historical context. Concretely, the convolution operation is embedded into each unit to describe the interaction between temporal features and spatial structures of oil saturation, thus the Ctrl-CLSTM realizes the unified modeling of the spatiotemporal features of oil saturation. In addition, a novel control gate structure is introduced in each Ctrl-CLSTM unit to take the injection–production parameters as controllable influencing factors and establish the nonlinear relationship between oil saturation and injection–production parameters according to the coordinates of each well location. Therefore, different oil saturation prediction results can be obtained by changing the injection–production parameters. Finally, experiments on real oilfields show that the Ctrl-CLSTM comprehensively considers the influence of artificial controllable factors such as injection–production parameters, accomplishes accurate prediction of oil saturation with a structure similarity of more than 98% and is more time efficient than reservoir numerical simulation.

Hidden spatiotemporal sequence in transition to shear band in amorphous solids

Localization of plastic flow into a narrow shear band is a fundamental and ubiquitous nonequilibrium phenomenon in amorphous solids. Because of the intrinsic entangling of three types of elementary local atomic motion---shear, dilatation, and rotation---the precise physical process of shear band emergence is still an enigma. Here, to unveil this mystery, we formulate a theoretical protocol covering both affine and nonaffine components of deformation, to decode these three highly entangled local atomic-scale events. In contrast to the broad concept of the shear transformation zone, the plastic behavior can be demonstrated comprehensively as the operative manipulation of more exact shear-dominated zones, dilatation-dominated zones, and rotation-dominated zones. Their spatiotemporal evolution exhibits a transition from synchronous motion to separate distribution at the onset of the shear band. The hidden mechanism is then revealed with the help of extreme value theory and percolation analysis. Numerical evidence from extreme value theory indicates that dilatation is the dominant mode at the embryonic stage of the initial plastic units, as evidenced through the larger degree of dilatation localization compared with shear and rotation. The percolation analysis points towards the critical power-law scaling nature at the transition from stochastic activation to percolation of plastic regions. Then the comprehensive picture underlying shear banding emergence is uncovered. Firstly, dilatation triggers initial shear and rotation in soft regions, leading to the embryos of the initial flow units, which are followed by the secondary activation of rotation in neighboring hard material, thus causing an alternating distribution of rotation and shear-dilatation regions. Such rotation activation contributes to further perturbation in these regions and ultimately leads to percolation transition and shear band formation. Our findings also reinforce that the discussion of plastic behavior in disordered materials must take into account both affine and nonaffine component deformation.

Dynamic prediction of global monthly burned area with hybrid deep neural networks.

Wildfires not only severely damage the natural environment and global ecological balance but also cause substantial losses to global forest resources and human lives and property. Unprecedented fire events such as Australia's bushfires have alerted us to the fact that wildfire prediction is a critical scientific problem for fire management. Therefore, robust, long-lead models and dynamic predictions of wildfire are valuable for global fire prevention. However, despite decades of effort, the dynamic, effective, and accurate prediction of wildfire remains problematic. There is great uncertainty in predicting the future based on historical and existing spatiotemporal sequence data, but with advances in deep learning algorithms, solutions to prediction problems are being developed. Here, we present a dynamic prediction model of global burned area of wildfire employing a deep neural network (DNN) approach that produces effective wildfire forecasts based on historical time series predictors and satellite-based burned area products. A hybrid DNN that combines long short-term memory and a two-dimensional convolutional neural network (CNN2D-LSTM) was proposed, and CNN2D-LSTM model candidates with four different architectures were designed and compared to construct the optimal architecture for fire prediction. The proposed model was also shown to outperform convolutional neural networks (CNNs) and the fully connected long short-term memory (FcLSTM) approach using the refined index of agreement and evaluation metrics. We produced monthly global burned area spatiotemporal prediction maps and adequately reflected the seasonal peak in fire activity and highly fire-prone areas. Our combined CNN2D-LSTM approach can effectively predict the global burned area of wildfires 1 month in advance and can be generalized to provide seasonal estimates of global fire risk.

Urban Traffic Flow Prediction with Deep Neural Network

It is critical to realize accurate collecting, visualization, rule mining, and prediction analysis of the traffic flow operating state in order for the intelligent transportation system to achieve exact management and control of traffic flow. Traffic flow prediction is primarily concerned with traffic data on roadways, which has both temporal and spatial correlations. Aiming at the spatiotemporal characteristics, this paper studies two aspects and designs a traffic flow prediction model with a deep neural network. First, this work proposes a traffic flow spatial feature learning algorithm with the combination of graph convolutional neural network and attention mechanism. Distinct weights are assigned to the degree of mutual impact between different nodes, and node adaptive learning is implemented at the same time, which modifies the standard parameter sharing mode, allowing for improved expressive ability and spatial feature extraction. Secondly, a learning algorithm for temporal characteristics of traffic flow based on the temporal convolutional network is proposed, which ensures that the dimensions of input and output data are consistent through causal convolution. The dilated convolution can flexibly control the receptive field by setting the sampling interval and can also extract temporal features well for long-length spatiotemporal sequence data. Finally, a spatiotemporal graph attention-based traffic flow prediction approach is constructed. To learn features, learn parameters for multiple modes and improve the model effect, this model employs a combination of graph convolutional neural networks and an attention mechanism. It uses a temporal convolutional network to expand the receptive field, better capture temporal features, and finally add residual connections to prevent problems such as overfitting caused by too deep network layers.

Clustering-Assisted Regional Spatio-Temporal Sequence Pattern Mining in Crime Database

With the recent advances in IoT and other smart devices, an explosive amount of data, both spatially and temporally significant are generated. Discovering interesting or useful patterns from these spatiotemporal data is referred to as spatiotemporal data mining. These patterns could be unordered, totally ordered or partially ordered based on the temporal ordering. This work focusses on the totally ordered patterns or sequential patterns from spatiotemporal event database. Spatiotemporal event sequence miner finds sequence of events that overlaps spatially and temporally. Traditional approaches discover patterns that are frequent in the entire dataset. In this work a clustering-assisted approach to find regionally or locally frequent spatiotemporal pattern is proposed. The proposed Clustering assisted Regional Spatiotemporal Event Sequence (CReST) mining approach overcomes the bias caused by uneven distribution of spatiotemporal events while mining patterns. Chicago crime dataset is used for evaluating the proposed approach with traditional sequence mining algorithm.

Distribution and Occurrence Frequency of dB/dt Spikes During Magnetic Storms 1980–2020

AbstractThe physical magnetospheric cause for geomagnetically induced currents (GICs) are rapid time‐varying magnetic fields (dB/dt), which occur mainly during magnetic substorms and storms. When, where and why exactly such rapid dB/dt may occur is insufficiently understood. We investigated all storms since 1980 and analyzed the negative and positive dB/dt spikes (>|500| nT/min) in the north and east component using a worldwide coverage (SuperMAG). Our analysis confirmed the existence of two dB/dt spikes “hotspots” located in the pre‐midnight and in the morning magnetic local time sector, independently of the geographic location of the stations. The associated physical phenomena are probably substorm current wedge onsets and westward traveling surges (WTS) in the evening sector, and wave‐ or vortex‐like current flows in the morning sector known as Omega bands. We observed a spatiotemporal evolution of the negative northern dB/dt spikes. The spikes initially occur in the pre‐midnight sector, and then develop in time toward the morning sector. This spatiotemporal sequence is correlated with bursts in the AE index, and can be repeated several times throughout a storm. Finally, we investigated the peak value of Dst and AE during the storm period in comparison with the dB/dt spike occurrence frequency, we did not find any correlation. This result implies that a moderate storm with many spikes can be as (or more) dangerous for ground‐based infrastructures than a major storm with fewer dB/dt spikes. Our findings regarding the physical causes and characteristics of dB/dt spikes may help to improve the GIC forecast for the affected regions.

Rock mass fracture maps prediction based on spatiotemporal image sequence modeling

AbstractDiscontinuities in rock mass are the characteristic challenge of rock tunnel engineering projects, which have a vital impact on rock mass exposures' mechanical and hydrological characteristics. There is a growing demand for predicting the fracture maps during tunnel excavation to ensure a smooth tunnel excavation process. The computer vision measurement of fractures in the tunnel surface is a current hot spot, but the traditional statistical analysis methods for fractures are still mainstream. This paper uses a novel perspective of the time‐space sequence to explain the continuously exposed rock mass during tunneling. A spatial‐aware recurrent neural network is proposed, which takes the historical fracture maps as the input to predict the unexcavated part. The experimental results suggest that the proposed model produces reliable performance and is superior to the other two state‐of‐the‐art deep learning models. Moreover, the test on the site rock tunnel data suggested promising results for fracture map predictions.

An Attention Encoder-Decoder Dual Graph Convolutional Network with Time Series Correlation for Multi-Step Traffic Flow Prediction

Accurate traffic prediction is a powerful factor of intelligent transportation systems to make assisted decisions. However, existing methods are deficient in modeling long series spatio-temporal characteristics. Due to the complex and nonlinear nature of traffic flow time series, traditional methods of prediction tasks tend to ignore the heterogeneity and long series dependencies of spatio-temporal data. In this paper, we propose an attentional encoder-decoder dual graph convolution model with time-series correlation (AED-DGCN-TSC) for solving the spatio-temporal sequence prediction problem in the traffic domain. First, the time-series correlation module calculates the sequence similarity by fast Fourier transform and inverse fast Fourier transform, while obtaining multiple possible lengths as possible solutions for the sequence period length. Then, K possible periods fetches are selected and the corresponding sequences are weighted and aggregated to the target sequence. Then, the gated dual graph convolution recurrent unit uses the graph convolution operation, which combines the ideas of node embedding, and dual graph, as an operation inside the gated recurrent structure to capture the spatio-temporal heterogeneity relationship of long sequences. The gated decomposition recurrent module decomposes the time series into the period and trend terms, which are modelled by convolutional gated recurrent unit (ConvGRU) and then fused with features, respectively, and output after graph convolution. Finally, multi-step prediction of future traffic flow is performed in the form of encoder-decoder. Experimental evaluations are conducted on two real traffic datasets, and the results demonstrate the effectiveness of the proposed model.

Electrocorticographic evidence of a common neurocognitive sequence for mentalizing about the self and others

Neuroimaging studies of mentalizing (i.e., theory of mind) consistently implicate the default mode network (DMN). Nevertheless, the social cognitive functions of individual DMN regions remain unclear, perhaps due to limited spatiotemporal resolution in neuroimaging. Here we use electrocorticography (ECoG) to directly record neuronal population activity while 16 human participants judge the psychological traits of themselves and others. Self- and other-mentalizing recruit near-identical cortical sites in a common spatiotemporal sequence. Activations begin in the visual cortex, followed by temporoparietal DMN regions, then finally in medial prefrontal regions. Moreover, regions with later activations exhibit stronger functional specificity for mentalizing, stronger associations with behavioral responses, and stronger self/other differentiation. Specifically, other-mentalizing evokes slower and longer activations than self-mentalizing across successive DMN regions, implying lengthier processing at higher levels of representation. Our results suggest a common neurocognitive pathway for self- and other-mentalizing that follows a complex spatiotemporal gradient of functional specialization across DMN and beyond.

Spatiotemporal graph convolutional network for automated detection and analysis of social behaviours among pre-weaning piglets

In the pig industry, social behaviors of preweaning piglets are critical indicators of their livability, growth, health, and welfare status, for which there is an urgent need for using precision livestock farming tools. In this study, a novel method based on graph convolutional networks (GCNs) was developed to characterize preweaning piglet social behaviors such as snout–snout as well as snout-body social nosing and snout–snout as well as snout-body aggressive/playing behavior. Using an integrated CNN-based network, the proposed method first detected and tracked individual piglets. After that, a self-adaptive spatial affinity kernel function was used to detect suspected social behaviors and spatiotemporal graphs with high-quality node features (coordinates, node-estimation confidence, distance from each node to the centroid, and node motion) were built for pairwise piglets for further analysis. The spatiotemporal graph sequences were fed into a self-adaptive GCN combined with an attention mechanism to classify the suspected social behaviours. Our method performed well in detecting piglet social behaviours with a recall of 0.9405, a precision of 0.9669, and an F 1 score of 0.9535. In an 8-h video episode, the time budges of snout–snout as well as snout-body social nosing and snout–snout as well as snout-body aggressive/playing behaviour were 33.82%, 38.34%, 13.74%, and 14.10%, respectively. The findings show that detecting body part-associated social behaviours in piglets using a GCN is feasible, yielding practical computer vision technologies for improved piglet behaviour monitoring and management. • Used a graph convolutional network to recognize animal behaviors. • Constructed spatiotemporal and self-adaptive graphs with high-quality node features. • Detected social behaviors among preweaning piglets using computer vision. • Achieved high accuracy in piglet social behavior detection.

Automatic Quality Assessment of Speech-Driven Synthesized Gestures

The automatic synthesis of realistic gestures has the ability to change the fields of animation, avatars, and communication agents. Although speech-driven synthetic gesture generation methods have been proposed and optimized, the evaluation system of synthetic gestures is still lacking. The current evaluation method still needs manual participation, but it is inefficient in the industry of synthetic gestures and has the interference of human factors. So we need a model that can construct an automatic and objective quantitative quality assessment of the synthesized gesture video. We noticed that recurrent neural networks (RNN) have advantages in modeling advanced spatiotemporal feature sequences, which are very suitable for use in the processing of synthetic gesture video data. Therefore, to build an automatic quality assessment system, we propose in our work a model based on Bi-LSTM and make a little adjustment on the attention mechanism in it. Also, the evaluation method is proposed and experiments are designed to prove that the improved model of the algorithm can complete the quantitative evaluation of synthetic gestures. At the same time, in terms of performance, the model has an improvement of about 20% compared to before the algorithm adjustment.

GP-ETAS: semiparametric Bayesian inference for the spatio-temporal epidemic type aftershock sequence model

The spatio-temporal epidemic type aftershock sequence (ETAS) model is widely used to describe the self-exciting nature of earthquake occurrences. While traditional inference methods provide only point estimates of the model parameters, we aim at a fully Bayesian treatment of model inference, allowing naturally to incorporate prior knowledge and uncertainty quantification of the resulting estimates. Therefore, we introduce a highly flexible, non-parametric representation for the spatially varying ETAS background intensity through a Gaussian process (GP) prior. Combined with classical triggering functions this results in a new model formulation, namely the GP-ETAS model. We enable tractable and efficient Gibbs sampling by deriving an augmented form of the GP-ETAS inference problem. This novel sampling approach allows us to assess the posterior model variables conditioned on observed earthquake catalogues, i.e., the spatial background intensity and the parameters of the triggering function. Empirical results on two synthetic data sets indicate that GP-ETAS outperforms standard models and thus demonstrate the predictive power for observed earthquake catalogues including uncertainty quantification for the estimated parameters. Finally, a case study for the l’Aquila region, Italy, with the devastating event on 6 April 2009, is presented.

A Hybrid Multivariate Deep Learning Network for Multistep Ahead Sea Level Anomaly Forecasting

Abstract The accumulated remote sensing data of altimeters and scatterometers have provided new opportunities for ocean state forecasting and have improved our knowledge of ocean–atmosphere exchanges. Studies on multivariate, multistep, spatiotemporal sequence forecasts of sea level anomalies (SLA) for different modalities, however, remain problematic. In this paper, we present a novel hybrid and multivariate deep neural network, named HMnet3, which can be used for SLA forecasting in the South China Sea (SCS). First, a spatiotemporal sequence forecasting network is trained by an improved convolutional long short-term memory (ConvLSTM) network using a channelwise attention mechanism and multivariate data from 1993 to 2015. Then a time series forecasting network is trained by an improved long short-term memory (LSTM) network, which is realized by ensemble empirical mode decomposition (EEMD). Finally, the two networks are combined by a successive correction method to produce SLA forecasts for lead times of up to 15 days, with a special focus on the open sea and coastal regions of the SCS. During the testing period of 2016–18, the performance of HMnet3 with sea surface temperature anomaly (SSTA), wind speed anomaly (SPDA), and SLA data is much better than those of state-of-the-art dynamic and statistical (ConvLSTM, persistence, and climatology) forecast models. Stricter testbeds for trial simulation experiments with real-time datasets are investigated, where the eddy classification metrics of HMnet3 are favorable for all properties, especially for those of small-scale eddies.

A multi-view bidirectional spatiotemporal graph network for urban traffic flow imputation

Accurate estimation of missing traffic data is one of the essential components in intelligent transportation systems (ITS). The non-Euclidean data structure and complex missing traffic flow patterns make it challenging to capture nonlinear spatiotemporal correlations of missing traffic flow, which are critical for the imputation of missing traffic data. In this study, we propose a novel multi-view bidirectional spatiotemporal graph network called Multi-BiSTGN to impute urban traffic data with complex missing patterns. First, three spatiotemporal graph sequences are constructed to comprehensively describe traffic conditions from different temporal correlation views, i.e. temporal closeness view, daily periodicity view, and weekly periodicity view. Then, three bidirectional spatiotemporal graph networks are fused by a parametric-matrix-based method to obtain the final imputation results. To train the Multi-BiSTGN model, a novel loss function that considers the interactions between three temporal correlation views is designed to optimize the parameters of the Multi-BiSTGN model. The proposed model was validated on real-world traffic datasets collected in Wuhan, China. Experimental results showed that Multi-BiSTGN outperformed ten existing baselines under different missing types (random missing, block missing, and mixed missing) and missing rates.

Geomorphological and sedimentary records of Late Quaternary activities of Qiaojia-Jinyang segment of Lianfeng fault zone, Southwest China

Lianfeng fault zone (LFZ) in Southwest China has great significance for understanding the seismogenic environment, but its nature and latest active age is still poor constrained up to now. The Qiaojia-Jinyang segment (QJS) of LFZ intersects with Jinsha River; so the well-developed river terraces provide a potential spatio-temporal constrains for recent faulting of QJS. Based on investigation of the terrace deposits along river, this paper makes a detailed logging and dating of the faulting and liquefaction of QJS. Combined previous data, the spatio-temporal sequence of the Late-Quaternary river terraces in the area was redetermined. It is considered that the river terraces T1 (lowest) and T2 at QJS (~ 10–20 m and 60–70 m, respectively, above the local river level) are roughly developed in the Middle Holocene and the Late Pleistocene, indicating that the valley along QJS was strongly undercut since the Late Pleistocene. Based on the analysis of the morphological characteristics, spatial distribution, material composition and intersecting relationship between the sand veins and the layers, the strong ground motions is preliminarily detected, which indicates the strong seismic activity of the during the Quaternary. Combined with the faulting characteristics and the dating data of strata within the profiles of terrace deposits, it is considered that the LFZ is active at least at QJS, and the latest active time is not earlier than the Early-Middle Holocene. These findings provide a clear geological evidences for the seismicity assessment at LFZ, and help to the understanding of regional tectonic environment and the prevention of earthquake disasters.

Semantic Trajectory Frequent Pattern Mining Model

A method for mining frequent patterns of individual user trajectories is proposed based on location semantics. The semantic trajectory is obtained by inverse geocoding and preprocessed to obtain the Top-k candidate frequent location item sets, and then the spatio-temporal sequence intersection and the divide and conquer merge methods are used to convert the frequent iterative calculation of long itemsets into hierarchical sets' regular operations, the superset and subset of frequent sequences are found. This kind of semantic trajectory frequent pattern mining can actively identify and discover potential carpooling needs, and provide higher accuracy for location-based intelligent recommendations such as carpooling and HOV lane travel (High-Occupancy Vehicle Lane). Carpool matching and recommendation based on semantic trajectory in this paper is suitable for single carpooling and relay-ride carpooling. the results of simulation carpooling experiments prove the applicability and efficiency of the method.

Thunderstruck: The ACDC model of flexible sequences and rhythms in recurrent neural circuits.

Adaptive sequential behavior is a hallmark of human cognition. In particular, humans can learn to produce precise spatiotemporal sequences given a certain context. For instance, musicians can not only reproduce learned action sequences in a context-dependent manner, they can also quickly and flexibly reapply them in any desired tempo or rhythm without overwriting previous learning. Existing neural network models fail to account for these properties. We argue that this limitation emerges from the fact that sequence information (i.e., the position of the action) and timing (i.e., the moment of response execution) are typically stored in the same neural network weights. Here, we augment a biologically plausible recurrent neural network of cortical dynamics to include a basal ganglia-thalamic module which uses reinforcement learning to dynamically modulate action. This "associative cluster-dependent chain" (ACDC) model modularly stores sequence and timing information in distinct loci of the network. This feature increases computational power and allows ACDC to display a wide range of temporal properties (e.g., multiple sequences, temporal shifting, rescaling, and compositionality), while still accounting for several behavioral and neurophysiological empirical observations. Finally, we apply this ACDC network to show how it can learn the famous "Thunderstruck" song intro and then flexibly play it in a "bossa nova" rhythm without further training.

Using a 3D convolutional neural network and gated recurrent unit for tropical cyclone track forecasting

The tropical cyclone (TC) track forecast is an essential task in meteorological operations. An accurate forecast should be based on a comprehensive understanding and description of TCs. A TC has a complex three-dimensional structure, and the surrounding atmosphere is the driving force for its development. Traditional forecasting methods performed relatively well for the TCs with stable moving speed and direction. However, the forecast accuracy still leaves some space to improve. In recent years, machine learning methods that can extract features from a large amount of historical data have been used in meteorological services and have shown excellent performance. To better forecast 6, 12, 18, and 24 h TC tracks in the Western North Pacific, a hybrid optimization model, combining the 3D convolutional neural network (3DCNN), gated recurrent unit (GRU), and smoothing algorithm is designed, which is called smoothed 3D-GRU. The 3DCNN is used to explore the potential relationship between environmental variables and TC movements at different pressure levels. The GRU is used to convert the TC track forecasting problem into a spatio-temporal sequence problem. The smoothing algorithm is used as a post-processing method to suppress unreasonable jumps of the model output. The mean spherical distances (MSDs) of the proposed smoothed 3D-GRU model at four prediction times are 27.89, 52.37, 79.16, and 112.05 km, which are lower than the comparative machine learning-based forecasting algorithms. Compared with the numerical prediction methods, the MSDs of the smoothed 3D-GRU model are lower in most situations. In general, the smoothed 3D-GRU model can provide reliable guidance for the TC trajectory prediction.

Deep learning via LSTM models for COVID-19 infection forecasting in India.

The COVID-19 pandemic continues to have major impact to health and medical infrastructure, economy, and agriculture. Prominent computational and mathematical models have been unreliable due to the complexity of the spread of infections. Moreover, lack of data collection and reporting makes modelling attempts difficult and unreliable. Hence, we need to re-look at the situation with reliable data sources and innovative forecasting models. Deep learning models such as recurrent neural networks are well suited for modelling spatiotemporal sequences. In this paper, we apply recurrent neural networks such as long short term memory (LSTM), bidirectional LSTM, and encoder-decoder LSTM models for multi-step (short-term) COVID-19 infection forecasting. We select Indian states with COVID-19 hotpots and capture the first (2020) and second (2021) wave of infections and provide two months ahead forecast. Our model predicts that the likelihood of another wave of infections in October and November 2021 is low; however, the authorities need to be vigilant given emerging variants of the virus. The accuracy of the predictions motivate the application of the method in other countries and regions. Nevertheless, the challenges in modelling remain due to the reliability of data and difficulties in capturing factors such as population density, logistics, and social aspects such as culture and lifestyle.