Temporal Transfer Research Articles

From Geospatial to Temporal Separation: A Review on Carbon Accounting Endogenizing Fixed Capital

A scientific and systematic approach that provides accurate carbon accounting is a fundamental prerequisite for responding to the climate crisis. The carbon accounting system has evolved from production-based accounting measuring territorial emissions to consumption-based accounting, gradually covering spatial and temporal separation. Its scientificity is thus enhanced. The temporal separation (between past production and current consumption) from fixed capital has become of recent interest, but a panoramic overview still rarely exists because of longstanding neglect. Combining quantitative analysis with a literature review, this study systematically analyzed the historical development and current state of carbon accounting, especially on studies endogenizing fixed capital, to locate hotspots and predict future trends. The current studies with fixed capital endogenization mainly use a flow matrix and augmented method to analyze the total and structure and the difference with traditional results, which are still in the initial rise stage; however, the process analysis and influence factor identification are insufficient. Therefore, this study proposes future avenues for analyzing the entire process of spatial and temporal carbon transfer based on a metabolic perspective and considering the initial input factors to identify the key drivers, which will provide new ideas and perspectives for policy formulation.

Artificial Intelligence Algorithms for Rapeseed Fields Mapping Using Sentinel-1 Time Series: Temporal Transfer Scenario and Ground Sampling Constraints

Artificial Intelligence Algorithms for Rapeseed Fields Mapping Using Sentinel-1 Time Series: Temporal Transfer Scenario and Ground Sampling Constraints

Temporal-Domain Adaptation for Satellite Image Time-Series Land-Cover Mapping With Adversarial Learning and Spatially Aware Self-Training

Nowadays, Satellite Image Time Series (SITS) are commonly employed to derive land cover maps (LCM) to support decision makers in a variety of land management applications. In the most general workflow, the production of LCM strongly relies on available ground truth data to train supervised machine learning models. Unfortunately, this data is not always available due to time-consuming and costly field campaigns. In this scenario, the possibility to transfer a model learnt on a particular year ( <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">source domain</i> ) to a successive period of time ( <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">target domain</i> ), over the same study area, can save time and money. Such a kind of model transfer is challenging due to different acquisition conditions affecting each time period thus, resulting in possible distribution shifts between <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">source</i> and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">target</i> domains. In the general field of machine learning, Unsupervised Domain Adaptation (UDA) approaches are well suited to cope with the learning of models under distribution shifts between <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">source</i> and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">target</i> domains. While widely explored in the general computer vision field, they are still under investigated for SITS-based land cover mapping, especially for the temporal transfer scenario. With the aim to cope with this scenario in the context of SITS-based land cover mapping, here we propose Spatially Aligned Domain-Adversarial Neural Network, a framework that combines both adversarial learning and self-training to transfer a classification model from a time period (year) to a successive one on a specific study area. Experimental assessment on a study area located in Burkina Faso characterized by challenging operational constraints demonstrates the significance of our proposal. The obtained results have shown that our proposal outperforms all the UDA competing methods by 7 to 12 points of F1-score across three different transfer tasks.

Effect of Contralateral Acoustic Stimulation on Temporal Processing Abilities in Individuals with Normal Hearing.

The aim of the article was to compare the conditions of silent and contralateral noise on the temporal processing parameters. A total of 40 participants (20 males and 20 females) were enrolled in the study with a mean age of 21.7years, participants with normal hearing thresholds and no history of middle ear pathology were enrolled for the study. The temporal processing tests such as duration discrimination test, gap detection test, and temporal modulation transfer function tests were carried out in these 40 individuals in the two conditions of silent and contralateral noise using MATLAB. Statistical analysis was carried out using the SPSS version 25.0 were descriptive and inferential statistics were carried out. Data was normally distributed on the Shapiro-Wilk's test of normality due to which a paired t test was carried out to establish the nature of significance between the silent and contralateral noise condition. Results reveal the presence of significant difference (p < 0.01) between the groups for all the parameters of temporal resolution with contralateral noise condition performing better than silent condition for the parameters. There is a positive effect of the efferent auditory pathway on the temporal resolution parameters thus implying that speech perception in noise is improved in the presence of background noise for normal hearing individuals due to this effect.

A Spatial–Temporal Depth-Wise Residual Network for Crop Sub-Pixel Mapping from MODIS Images

To address the problem caused by mixed pixels in MODIS images for high-resolution crop mapping, this paper presents a novel spatial–temporal deep learning-based approach for sub-pixel mapping (SPM) of different crop types within mixed pixels from MODIS images. High-resolution cropland data layer (CDL) data were used as ground references. The contributions of this paper are summarized as follows. First, we designed a novel spatial–temporal depth-wise residual network (ST-DRes) model that can simultaneously address both spatial and temporal data in MODIS images in efficient and effective manners for improving SPM accuracy. Second, we systematically compared different ST-DRes architecture variations with fine-tuned parameters for identifying and utilizing the best neural network architecture and hyperparameters. We also compared the proposed method with several classical SPM methods and state-of-the-art (SOTA) deep learning approaches. Third, we evaluated feature importance by comparing model performances with inputs of different satellite-derived metrics and different combinations of reflectance bands in MODIS. Last, we conducted spatial and temporal transfer experiments to evaluate model generalization abilities across different regions and years. Our experiments show that the ST-DRes outperforms the other classical SPM methods and SOTA backbone-based methods, particularly in fragmented categories, with the mean intersection over union (mIoU) of 0.8639 and overall accuracy (OA) of 0.8894 in Sherman County. Experiments in the datasets of transfer areas and transfer years also demonstrate better spatial–temporal generalization capabilities of the proposed method.

Coordinated Planning of Power Grids in Information Parks considering the Time-Space Transfer Characteristics of Data Center Loads

With the development of green energy, the penetration rate of new energy in the power grid is rising rapidly. Incorporating a large amount of new energy into the power grid will affect the power supply balance of the power grid, and even in severe cases will affect the safe and stable operation of the power grid. The information park has various small and medium data centers. The adjustable load of the data center can be transferred in time and space through the optical fiber and other communication networks, which can enhance the controllability of the power grid to a certain extent and promote the consumption of new energy. This paper proposes a collaborative planning method for information parks that considers the temporal and spatial transfer characteristics of data center loads, builds a load transfer model for data centers and builds a park power grid expansion model with the goal of minimizing the total cost of construction and operation. In this paper, an example of a park is used to verify the calculation example, and it is compared with other schemes to verify the feasibility of the method in this paper.

Building and exploiting spatial–temporal knowledge graph for next POI recommendation

Next Point-of-Interest (POI) recommendation has shown great value for both users and businesses in the field of location-based services. Many spatial–temporal inferring methods have been developed to perform this task, but the data sparsity of POI trajectories greatly hinders the recommendation performance. Knowledge graphs (KGs) have been demonstrated as an effective way to tackle data sparsity in the general recommendation field by leveraging the valuable information of entities and relations. Yet, few studies have explored applying KGs for the next POI recommendation task because of the following challenges: (1) how to represent the dynamic mobility behaviors of users with the static entities and relations in KGs; and (2) how to utilize the different types of entities and relations in KGs to capture long- and short-term preferences of users. In this work, we investigate building a spatial–temporal KG (STKG) from check-in sequences of users to promote the next POI recommendation, without introducing any external attributes of users and POIs. In STKG, we design a novel spatial–temporal transfer relation to intuitively capture users’ transition patterns between neighboring POIs. Then, based on the STKG, we propose an innovative model, named STKGRec, for the next POI recommendation, which explicitly models long- and short-term preferences of users in an end-to-end manner. In particular, STKGRec learns both the spatial–temporal correlation of consecutive and nonconsecutive visits in the current check-in sequence to comprehensively capture the short-term preferences of users. Extensive experiments on four real-world datasets demonstrate the superiority of STKGRec against the state-of-the-art baseline methods. The code of our proposed model is available athttps://github.com/WeiChen3690/STKGRec.

Diversity of temporal response patterns in midbrain auditory neurons of frogs Batrachyla and its relevance for male vocal responses.

We investigated response selectivities of single auditory neurons in the torus semicircularis of male frogs Batrachyla leptopus (72 neurons) and B. taeniata (57 neurons) to synthetic stimuli of different temporal structures. Series of stimuli in which note and pulse rate, note and pulse structure and call duration varied systematically were presented. Neuronal responses quantified in terms of proportions of units displaying diverse temporal transfer functions are related in different modes with patterns of evoked vocal responses studied previously in these frogs. Correspondences and mismatches occurred between the auditory and vocal domains. The analysis of this evidence together with corresponding information from previous neuronal and behavioral studies in the third species of this genus, B. antartandica, indicates that different modes of preferences for acoustic communication signals can coexist within this anuran group.

Examination of the Temporal Stability of Daily and Monthly Adjustment Factors of Bicycle Traffic

Monthly adjustment factors (MFs) and daily adjustment factors (DFs) are typically calculated using complete annual data from permanent count stations. These factors are applied at other locations where short-term counts are collected using temporary counters. In this research, the long-term temporal stability of MFs and DFs for bicycle traffic was investigated. Annual daily count data collected from permanent count stations in Vancouver, Canada between 2010 and 2019 were used in the study. A longitudinal analysis of MFs and DFs in the 10 years was undertaken. The coefficient of variation (CoV) of the MFs of the 10 years was found to be less than 12% for the months between April and September, indicating the potential temporal transferability of these factors from one year to a future year. The other 4 months of the year showed a higher CoV with December being the highest (i.e., 25%). As for the DFs, it was found that, depending on the day of week, weather, and month, substantial variation could exist in the calculated factors. Nevertheless, factors developed for dry weekdays in the summer months were shown to exhibit the lowest variability over the 10 years and as such these factors were deemed transferable from one year to another. The results presented in this paper could be beneficial for transportation planners and traffic engineers managing traffic monitoring programs of cycling traffic for which the potential for transferring historical adjustment factors to future years was demonstrated.

Temporal transferability of species abundance models to study the changes of breeding bird species based on land cover changes

AimSpecies distribution models have become important tools for studying changes in biodiversity. Most studies use these models to evaluate the impact of global changes on biodiversity. For that purpose, scenarios are used that are based on changes in land use and/or land cover, or on climatic changes. However, the temporal transferability of such models depends heavily on modeling methods, environmental predictors and the ecological traits of species. Here, we evaluate the power of modeling tools to predict changes in bird species abundances based on observed changes in land cover. LocationWallonia, Belgium MethodsTo assess this temporal transferability, this research makes use of two biological and two environmental datasets, both sampled with a 10-year interval. This allows us to compare the predictions of models for another period than the period used to fit the models, with actual values for species abundance. We also analyzed the impact of ecological traits on the temporal transferability. Generalized additive models were fitted for 75 breeding birds. While a lot of studies use occurrence data, we used abundance data for fitting models. Abundance data contains more information and should allow us to better capture abundance changes in bird populations. ResultsFor the majority of species studied, the results show a low temporal transferability. With a few exceptions, e.g., species with softwood habitats, predicted changes do not correspond to observed changes. For certain bird species, e.g., those on arable lands, we observed an increase in predicted abundances in the future, while these species actually decreased. Few ecological traits seem to significantly impact the models’ transferability. Main conclusionsOur findings show that it is difficult to predict abundance changes of bird species based only on land cover changes. It is necessary to add other predictors into the models, e.g., predictors of habitat quality or spatial configuration.

Listening Effort in Prelingual Cochlear Implant Recipients: Effects of Spectral and Temporal Auditory Processing and Contralateral Acoustic Hearing.

Considering the impact of listening effort (LE) on auditory perception, attention, and memory, it is a significant aspect in the daily hearing experiences of cochlear implant (CI) recipients. Reduced spectral and temporal information on an acoustic signal can make listening more difficult; as a result, it is important to understand the relationship between LE and spectral and temporal auditory processing capacities in CI receivers. This study used spectral ripple discrimination and temporal modulation transfer function to evaluate 20 prelingually deafened and early implanted CI recipients. The speech perception in noise test (primary) and the digit recall task (DRT-secondary) were used to assess LE using the dual-task paradigm. To assess the effects of acoustic hearing, contralateral acoustic hearing thresholds between 125 Hz and 8 kHz with a hearing aid were also acquired. To examine the relationship between the research variables, correlation coefficients were generated. Furthermore, the Mann-Whitney U test was used to compare unilateral and bimodal users. There was a statistically significant correlation between LE and spectral ripple discrimination (r = 0.56; p = 0.011), 125 Hz (r = 0.51; p = 0.020), 250 Hz (r = 0.48; p = 0.030), 500 Hz (r = 0.45; p = 0.045), 1,000 Hz (r = 0.51; p = 0.023), 2000 Hz (r = 0.48; p = 0.031), and 4,000 Hz (r = 0.48; p = 0.031), whereas no statistically significant correlations were observed between temporal modulation transfer function in four frequencies and LE. There was no statistically significant difference between unilateral and bimodal CI recipients ( p > 0.05). As a result of the improved signal-to-noise ratio in the auditory environment, CI users with better spectral resolutions and acoustic hearing have a reduced LE. On the other hand, temporal auditory processing, as measured by temporal modulation detection, does not contribute to the LE.

Selective transfer of maternal antibodies in preterm and fullterm children

Preterm newborns are more likely to suffer from infectious diseases at birth compared to children delivered at term. Whether this is due to compromised cellular, humoral, or organ-specific development remains unclear. To begin to define whether maternal–fetal antibody transfer profiles differ across preterm (PT) and fullterm (FT) infants, the overall quantity and functional quality of an array of 24 vaccine-, endemic pathogen-, and common antigen-specific antibodies were assessed across a cohort of 11 PT and 12 term-delivered maternal:infant pairs from birth through week 12. While total IgG levels to influenza, pneumo, measles, rubella, EBV, and RSV were higher in FT newborns, selective Fc-receptor binding antibodies was noted in PT newborns. In fact, near equivalent antibody-effector functions were observed across PT and FT infants, despite significant quantitative differences in transferred antibody levels. Moreover, temporal transfer analysis revealed the selective early transfer of FcRn, FcγR2, and FcγR3 binding antibodies, pointing to differential placental sieving mechanisms across gestation. These data point to selectivity in placental transfer at distinct gestational ages, to ensure that children are endowed with the most robust humoral immunity even if born preterm.

Assessing multitemporal calibration for species distribution models

Species distribution models (SDMs), the most prominent tool in modern biogeography, rely on the assumptions that (i) species distribution is in equilibrium with the environment and (ii) that climatic niche has been conserved throughout recent geological time. These issues affect the spatial and temporal transferability of SDMs, limiting their reliability for applications such as when studying effects of past climate change on species distribution and extinctions. The integration of paleontological and neontological data for a multitemporal calibration and validation of SDMs has been suggested for improving SDMs flexibility. Here, we provide an empirical test for a multitemporal calibration, employing virtual species (i.e., with perfectly-known distributions) and comparing them directly with monotemporal SDMs (i.e., SDM calibrated in a single time layer). We used 1kyr-interval scenarios throughout the last 22 kyr BP for two ecologically different species in South America (a “hot and wet” species and a “cold and dry” species). Models with multitemporal calibration performed similarly to models with monotemporal calibration, regardless of species, sample sizes, and time frame. However, multitemporal calibration performed better when dealing with non-analogous climates among time layers. By improving the temporal SDMs transferability, multitemporal calibration opens new avenues for integrating fossil and recent occurrence data, which may substantially benefit biogeography and paleoecology.

The predictive performance of process‐explicit range change models remains largely untested

Ecological models used to forecast range change (range change models; RCM) have recently diversified to account for a greater number of ecological and observational processes in pursuit of more accurate and realistic predictions. Theory suggests that process‐explicit RCMs should generate more robust forecasts, particularly under novel environmental conditions. RCMs accounting for processes are generally more complex and data hungry, and so, require extra effort to build. Thus, it is necessary to understand when the effort of building a more realistic model is likely to generate more reliable forecasts. Here, we review the literature to explore whether process‐explicit models have been tested through benchmarking their temporal predictive performance (i.e. their predictive performance when transferred in time) and model transferability (i.e. their ability to keep their predictive performance when transferred to generate predictions into a different time) against simpler models, and highlight the gaps between the rapid development of process‐explicit RCMs and the testing of their potential improvements. We found that, out of five ecological processes (dispersal, demography, physiology, evolution, species interactions) and two observational processes (sampling bias, imperfect detection) that may influence reliability of forecasts, only the effects of dispersal, demography and imperfect detection have been benchmarked using temporally‐independent datasets. Only nine out of twenty‐nine process‐explicit model types have been tested to assess whether accounting for processes improves temporal predictive performance. We found no benchmarks assessing model transferability. We discuss potential reasons for the lack of empirical validation of process‐explicit models. Considering these findings, we propose an expanded research agenda to properly test the performance of process‐explicit RCMs, and highlight some opportunities to fill the gaps by suggesting models to be benchmarked using existing historical datasets.

An Improved Transition Probability Matrix for Crime Distribution Prediction.

The occurrence of crime has always been the main problem affecting urban public security and social security environment. Therefore, the prevention and control of crime is the focus of public security work. The traditional police strategy has poor timeliness and cannot respond and adjust in real time with the occurrence of criminal activities, and its deterrence and control of criminal activities are limited. To address the problem of low accuracy of existing crime distribution prediction models, an improved transition probability matrix for crime distribution prediction is proposed in this paper. Based on a large number of trajectory data of criminals, this paper quantitatively describes the temporal and spatial characteristics of crowd movement in different areas of the city by using the temporal and spatial transfer probability. Then, combining Markov chain and Bayes' theorem, the probability model of spatio-temporal transfer of criminal groups in regions is constructed. Finally, the model predicts the number of crimes in urban grid areas.

Evaluation of phase-locking to parameterized speech envelopes.

Humans rely on the temporal processing ability of the auditory system to perceive speech during everyday communication. The temporal envelope of speech is essential for speech perception, particularly envelope modulations below 20 Hz. In the literature, the neural representation of this speech envelope is usually investigated by recording neural phase-locked responses to speech stimuli. However, these phase-locked responses are not only associated with envelope modulation processing, but also with processing of linguistic information at a higher-order level when speech is comprehended. It is thus difficult to disentangle the responses into components from the acoustic envelope itself and the linguistic structures in speech (such as words, phrases and sentences). Another way to investigate neural modulation processing is to use sinusoidal amplitude-modulated stimuli at different modulation frequencies to obtain the temporal modulation transfer function. However, these transfer functions are considerably variable across modulation frequencies and individual listeners. To tackle the issues of both speech and sinusoidal amplitude-modulated stimuli, the recently introduced Temporal Speech Envelope Tracking (TEMPEST) framework proposed the use of stimuli with a distribution of envelope modulations. The framework aims to assess the brain's capability to process temporal envelopes in different frequency bands using stimuli with speech-like envelope modulations. In this study, we provide a proof-of-concept of the framework using stimuli with modulation frequency bands around the syllable and phoneme rate in natural speech. We evaluated whether the evoked phase-locked neural activity correlates with the speech-weighted modulation transfer function measured using sinusoidal amplitude-modulated stimuli in normal-hearing listeners. Since many studies on modulation processing employ different metrics and comparing their results is difficult, we included different power- and phase-based metrics and investigate how these metrics relate to each other. Results reveal a strong correspondence across listeners between the neural activity evoked by the speech-like stimuli and the activity evoked by the sinusoidal amplitude-modulated stimuli. Furthermore, strong correspondence was also apparent between each metric, facilitating comparisons between studies using different metrics. These findings indicate the potential of the TEMPEST framework to efficiently assess the neural capability to process temporal envelope modulations within a frequency band that is important for speech perception.

ALOS-2 L-band SAR backscatter data improves the estimation and temporal transferability of wildfire effects on soil properties under different post-fire vegetation responses

Remote sensing techniques are of particular interest for monitoring wildfire effects on soil properties, which may be highly context-dependent in large and heterogeneous burned landscapes. Despite the physical sense of synthetic aperture radar (SAR) backscatter data for characterizing soil spatial variability in burned areas, this approach remains completely unexplored. This study aimed to evaluate the performance of SAR backscatter data in C-band (Sentinel-1) and L-band (ALOS-2) for monitoring fire effects on soil organic carbon and nutrients (total nitrogen and available phosphorous) at short term in a heterogeneous Mediterranean landscape mosaic made of shrublands and forests that was affected by a large wildfire. The ability of SAR backscatter coefficients and several band transformations of both sensors for retrieving soil properties measured in the field in immediate post-fire situation (one month after fire) was tested through a model averaging approach. The temporal transferability of SAR-based models from one month to one year after wildfire was also evaluated, which allowed to assess short-term changes in soil properties at large scale as a function of pre-fire plant community type. The retrieval of soil properties in immediate post-fire conditions featured a higher overall fit and predictive capacity from ALOS-2 L-band SAR backscatter data than from Sentinel-1 C-band SAR data, with the absence of noticeable under and overestimation effects. The transferability of the ALOS-2 based model to one year after wildfire exhibited similar performance to that of the model calibration scenario (immediate post-fire conditions). Soil organic carbon and available phosphorous content was significantly higher one year after wildfire than immediately after the fire disturbance. Conversely, the short-term change in soil total nitrogen was ecosystem-dependent. Our results support the applicability of L-band SAR backscatter data for monitoring short-term variability of fire effects on soil properties, reducing data gathering costs within large and heterogeneous burned landscapes.

Temporal Stability Analysis of Lighting Conditions in Traffic Accidents

Different lighting conditions can result in accidents of different levels of severity. However, current studies lack the consideration of the heterogeneity and temporal stability of accident data under various lighting conditions. Therefore, three years’ worth of data were used to investigate the critical factors of accident severity. The random parameters logit model was employed to investigate the influence of different lighting conditions on temporal stability and heterogeneity. The critical factors affecting injury severity were also identified. The temporal stability and transferability of the models were investigated by a series of likelihood ratio tests. Based on different lighting conditions (daylight conditions, and night-time conditions with street lighting on), six models were established. Three kinds of accident injury severity levels were classified: property damage only (PDO), severe injury (SI), and fatal injury (FI). The estimation results showed contributing factors of accident severity were significantly different between the two kinds of lighting conditions. Additionally, accidents showed temporal instability. The proposed method can provide a guide for infrastructure construction, operation, and maintenance in traffic-safety management.

Estimation of canopy nitrogen content in winter wheat from Sentinel-2 images for operational agricultural monitoring

Canopy nitrogen content (CNC, kg/ha) provides crucial information for site-specific crop fertilization and the usability of Sentinel-2 (S2) satellite data for CNC monitoring at high fertilization levels in managed agricultural fields is still underexplored. Winter wheat samples were collected in France and Belgium in 2017 (n = 126) and 2018 (n = 18), analysed for CNC and S2-spectra were extracted at the sample locations. A comparison of three established remote sensing methods to retrieve CNC was carried out: (1) look-up-table (LUT) inversion of the canopy reflectance model PROSAIL, (2) Partial Least Square Regression (PLSR) and (3) nitrogen-sensitive vegetation indices (VI). The spatial and temporal model transferability to new data was rigorously assessed. The PROSAIL-LUT approach predicted CNC with a root mean squared error of 33.9 kg/ha on the 2017 dataset and a slightly larger value of 36.8 kg/ha on the 2018 dataset. Contrary, PLSR showed an error of 27.9 kg N/ha (R2 = 0.52) in the calibration dataset (2017) but a substantially larger error of 38.4 kg N/ha on the independent dataset (2018). VIs revealed calibration errors were slightly larger than the PLSR results but showed much higher validation errors for the independent dataset (> 50 kg/ha). The PROSAIL inversion was more stable and robust than the PLSR and VI methods when applied to new data. The obtained CNC maps may support farmers in adapting their fertilization management according to the actual crop nitrogen status.

Improving Spatiotemporal Transferability of Real-Time Crash Likelihood Prediction Models Using Transfer-Learning Approaches

A real-time crash likelihood prediction model is an important component of the proactive traffic safety management system. Over the past decades, numerous models have been proposed for predicting real-time crash likelihood and achieved promising results. However, most studies have ignored the model transferability, especially for deep-learning models. The transferability of a model could be defined as applying the model to new data which obtain from different locations or periods. The purpose of this study is to improve the spatiotemporal transferability of a deep-learning crash likelihood prediction model by using transfer-learning approaches. Trajectory and crash data from five arterials in Florida were used in this study. A two-layer long short-term memory (LSTM) model was developed to predict real-time crash likelihood. Two scenarios were created for investigating spatial and temporal transferability of the developed model. Experimental results suggested that the model could be accurately transferred to new data by using the fine-tuning transfer-learning approach. In the framework of this research, the transferred models achieved higher predictive accuracy than models which were directly developed on the new data. Moreover, the spatial transfer learning outperformed the temporal transfer learning in predictive accuracy. For example, when the percentage of test data was 70%, the spatial transferred model had an average area under the curve (AUC) of 0.74. while the temporal transferred model had an average AUC of 0.65. Results from this study could be applied to transfer pretrained crash likelihood prediction models to new locations when few crashes are available.