Ride hailing services, such as Uber, Lyft, and Grab, have become a major transportation mode in the last decade. The number of current passenger requests is one of the important factors for such services routing and pricing algorithms. Therefore, predicting future passenger request for ride hailing services can boost the efficiency of the service for both drivers and riders by pre-planning the allocation of vehicles and avoiding traffic congestions. Demand forecasting for ride hailing services relies upon the spatial and temporal correlations of its features. The existing literatures mostly divide the target area into rectangular grids (based on the longitude and latitude), consider only adjacent grids for spatial correlation, and calculate demand for each grid independently. An individual grid can contain different regions with high and low demand or have a major part of it outside the land area, which obscures the granularity and precision of estimations and predictions. This paper attempts to mitigate the limitations of grid-based methods by estimating and predicting ride hailing service demand between geographic regions as pickup and destination zones. For predicting demand, a convolutional neural network is integrated with a recurrent autoencoder network to best capture the spatial–temporal correlations of features, including time of the day, month, year, weekend, holiday, pickup zone, destination, and demand. In our experiments, we forecast the demand for each pickup–destination pair for the next day at a certain hour by observing the demands over the past 2 weeks during the same hour in the New York City hire vehicle data set. Using the same model (CNN-biLSTM-AE) to predict demand for geographical regions, it achieved an [Formula: see text] of 0.984, while predicting demand for cells in the grid achieved an [Formula: see text] 0.545. While using the geographical regions instead of grids for partitioning the space, we compared our deep learning model with LSTM, CNN, CNN-LSTM, and LSTM-AE models and observed an improvement in [Formula: see text] from 0.632 to 0.767 and an improvement in RMSE from 20.53 to 16.33 against CNN.

El Niño-Southern Oscillation (ENSO) is a climate phenomenon caused due to irregular periodic oscillation in easterly winds and sea surface temperature (SST) over the tropical Pacific Ocean. ENSO is one of the main drivers of Earth’s inter-annual climate variability, which causes climate anomalies in the form of tropical cyclones, severe storms, heavy rainfalls and droughts. Due to the impact of ENSO on global climate, forecasting ENSO is of great importance. However, forecast accuracy of ENSO for a lead time of one year is low. ENSO events are forecasted through Oceanic Niño Index (ONI), which is the three-month running mean of SST anomalies over the Niño 3.4 region (5∘N-5∘S, 120∘W-170∘W). Features, such as SST, sea level pressure, zonal wind speed and meridional wind speed that contribute in determining ONI are mapped on spatial or geographical grids, where each spatial or geographical grid represents the values of one feature at a snapshot. Juxtaposing the spatial grids of all features creates a layered map at a snapshot. The layered spatial feature map is constructed at different snapshots, and they all are fed to the CLSTM to forecast ONI at lead times of 1, 3, 6, 9 and 12 months. This study employs backward stepwise feature selection based on generalization accuracy to find the most effective features. SST showed to be the best feature for forecasting ONI. Experiments showed that the CLSTM outperforms Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM) and Standard Neural Network (SNN) in terms of coefficient of determination ([Formula: see text]), Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE). More specifically, an improvement in [Formula: see text] values by 27.6%, 20.9%, 25% and 15.2% over CNN is observed for lead times of 3, 6, 9 and 12 months, respectively.

Accurate traffic flow prediction is required in traffic management and optimal route selection. Traffic data patterns are influenced by factors such as road characteristics, time of the day, day of the week, weather conditions, and special events. The accuracy of traffic flow prediction depends on the spatial and temporal extent and diversity of the available data, along with the prediction algorithm. This paper presents a long short-term memory (LSTM) network for traffic prediction which is boosted by optimizing the spatial and temporal extent of the features that are used as input. A grid search method is employed to choose the spatial and temporal window sizes that result in the highest generalization accuracy. The optimal temporal window size is the number of time steps for input features and the spatial window size refers to the size of the geographical neighborhood, where features are used as input. Our experiments with traffic prediction in the State of California showed that optimizing the spatial and temporal window sizes in an LSTM network could improve the RMSE and R2by up to 40.02% and 25.21%.

Ride hailing services, such as Uber, Lyft, and Grab have become a major transportation mode in the last decade.Current ride demand is one of the major factors in such services' pricing algorithm.Therefore, forecasting future travel demand for such services is essential to both drivers and riders.This study constructs a deep learning based model for ride hailing demand forecast aiming to achieve high accuracies in solving similar problems.This study attempts to address a limitation in existing ride hailing demand prediction models, where the area is divided into a rectangular grid and all travel demand forecasts are made between rectangular cells, rather than city neighborhood zones.The proposed model forecasts travel demand between city neighborhood zones.The forecast model integrates convolutional and recurrent neural networks and forecasts the demand for each pickup-destination pair for a particular hour, during the next day, by observing the demand over the past two weeks for that particular hour.Our experiments with a real-world hire vehicle dataset in New York City showed that the proposed model outperforms the CNN and LSTM models up to 18.41 % in RMSE and 22.65% in R 2 values.

El Nino-Southern Oscillation (ENSO) is an irregular periodic oscillation in easterly winds and sea surface temperature (SST) over the tropical Pacific Ocean.El Nino and La Nina are warm and cold phases of ENSO.Oceanic Nino Index (ONI) determines ENSO events by calculating the three-month running mean of SST anomalies over the Nino 3.4 region (5°N-5°S and 120°W-170°W).El Nino refers to ONI greater than +0.5℃ and La Nina refers to ONI less than -0.5℃ for five consecutive months across the east-central equatorial Pacific.ENSO is one of the main drivers of Earth's inter-annual climate variability, which causes climate anomalies in the form of tropical cyclones, severe storms, heavy rainfall, and droughts.ENSO not only impacts global climate and oceanic conditions but also impacts food production, human health, and economy.Therefore, forecasting ENSO is of great importance.The main contribution of this study is proposing a convolutional long-short term memory that can capture spatial and temporal relationships between ENSO and environmental variables, such as SST, sea level pressure, meridional wind, and zonal wind.This study not only reports forecast accuracy but also quantifies the uncertainty associated with the forecast.Experimental results show that the proposed model improves the forecast accuracy by 14.8%, 10.4%, 11.8%, and 22.2% for lead times of 3, 6, 9, and 12 months, respectively.

Mining GPS traces is proposed, as an alternative to surveying and satellite image processing, for constructing road and pedestrian networks, trading out the constructed network's perfectness for substantially reduced cost and time. To construct a road or pedestrian network, the proposed algorithm requires only GPS traces, which can be collected with low or no cost using crowd-sourced repositories such as open street map. Network construction algorithms in the literature, mostly do not provide enough details about different steps of constructing a network. Besides, their evaluations are mostly visual, qualitative, and limited to one dataset. This paper not only describes all the steps involved in constructing a network but also quantitatively evaluates the algorithm (via three metrics: precision, completeness, and topology correctness) using numerous datasets from different sources and discusses its time complexity. The proposed algorithm's distinguishing strengths can be summarized as: 1) fixing the constructed network's topological errors while many existing algorithms overlook the constructed network's topological integrity; 2) imposing no restrictions on the shape of GPS traces while many existing algorithms include such implicit or explicit constraints, e.g. GPS traces cannot include U turns; 3) being fully automatic and involving no subjective judgments or human interventions for filtering GPS points unlike many existing algorithms; and 4) achieving high accuracies in practice. The proposed algorithm includes six hyperparameters, all related to GPS traces' geometrical traits and optimized during evaluation.

A map-matching algorithm outputs a vector per GPS point, projecting the moving object on one of the segments of the transportation network. Although developing more sophisticated map-matching algorithms for vehicle and pedestrian navigation systems have been the focus of research in this field, reusability of the historical information already provided by map-matching algorithms has not been addressed yet. In other words, although researchers have been attempting to improve the accuracy of the aforementioned vector to correctly project GPS points on the transportation network, no research has exploited the spatial-temporal pattern in the arrangement of these projection vectors. This pattern, if properly detected, can be used as a rough surrogate for map-matching algorithms, in addition to other applications that require better positional accuracy for moving objects in smart cities. This paper detects and validates the spatial-temporal pattern in projection vectors produced by map-matching algorithms via machine learning. Projection vectors showed a strong spatial-temporal pattern in Chicago, IL, USA, which was captured best via a local nonlinear regressor, K-nearest neighbors, and helped double the positional accuracy of unseen GPS points. While a global nonlinear regressor, multilayer Perceptron was able to slightly improve the positional accuracy of GPS points, the linear least squares had an exacerbating effect on the positional accuracy.