Multimodal Spatio-Temporal Prediction with Stochastic Adversarial Networks

Spatio-temporal (ST) data is a collection of multiple time series data with different spatial locations and is inherently stochastic and unpredictable. An accurate prediction over such data is an important building block for several urban applications, such as taxi demand prediction, traffic flow prediction, and so on. Existing deep learning based approaches assume that outcome is deterministic and there is only one plausible future; therefore, cannot capture the multimodal nature of future contents and dynamics. In addition, existing approaches learn spatial and temporal data separately as they assume weak correlation between them. To handle these issues, in this article, we propose a stochastic spatio-temporal generative model (named D-GAN) which adopts Generative Adversarial Networks (GANs)-based structure for more accurate ST prediction in multiple time steps. D-GAN consists of two components: (1) spatio-temporal correlation network which models spatio-temporal joint distribution of pixels and supports a stochastic sampling of latent variables for multiple plausible futures; (2) a stochastic adversarial network to jointly learn generation and variational inference of data through implicit distribution modeling. D-GAN also supports fusion of external factors through explicit objective to improve the model learning. Extensive experiments performed on two real-world datasets show that D-GAN achieves significant improvements and outperforms baseline models.

Temporal Prediction of Paralytic Shellfish Toxins in the Mussel Mytilus galloprovincialis Using a LSTM Neural Network Model from Environmental Data

Paralytic shellfish toxins (PSTs) are produced mainly by Alexandrium catenella (formerly A. tamarense). Since 2000, the National Institute of Fisheries Science (NIFS) has been providing information on PST outbreaks in Korean coastal waters at one- or two-week intervals. However, a daily forecast is essential for immediate responses to PST outbreaks. This study aimed to predict the outbreak timing of PSTs in the mussel Mytilus galloprovincialis in Jinhae Bay and along the Geoje coast in the southern coast of the Korea Peninsula. We used a long-short-term memory (LSTM) neural network model for temporal prediction of PST outbreaks from environmental data, such as water temperature (WT), tidal height, and salinity, measured at the Geojedo, Gadeokdo, and Masan tidal stations from 2006 to 2020. We found that PST outbreaks is gradually accelerated during the three years from 2018 to 2020. Because the in-situ environmental measurements had many missing data throughout the time span, we applied LSTM for gap-filling of the environmental measurements. We trained and tested the LSTM models with different combinations of environmental factors and the ground truth timing data of PST outbreaks for 5479 days as input and output. The LSTM model trained from only WT had the highest accuracy (0.9) and lowest false-alarm rate. The LSTM-based temporal prediction model may be useful as a monitoring system of PSP outbreaks in the coastal waters of southern Korean.

Training-Associated Superior Visuomotor Integration Performance in Elite Badminton Players after Adjusting for Cardiovascular Fitness

Badminton is recognized as the fastest racket sport in the world based on the speed of the birdie which can travel up to 426 km per hour. On the badminton court, players are not only required to track the moving badminton birdie (visual tracking and information integration) but also must anticipate the exact timing to hit it back (temporal estimation). However, the association of training experience related to visuomotor integration or temporal prediction ability remains unclear. In this study, we tested this hypothesis by examining the association between training experience and visuomotor performances after adjusting for age, education, and cardiovascular fitness levels. Twenty-eight professional badminton players were asked to perform a compensatory tracking task and a time/movement estimation task for measuring visuomotor integration and temporal prediction, respectively. Correlation analysis revealed a strong association between training experience and performance on visuomotor integration, indicating badminton training may be promoted to develop visuomotor integration ability. Furthermore, the regression model suggests training experience explains 32% of visuomotor integration performances. These behavioral findings suggest badminton training may facilitate the perceptual–cognitive performance related to visuomotor integration. Our findings highlight the potential training in visuomotor integration may apply to eye–hand coordination performance in badminton sport.

Empirical Rainfall Thresholds for Landslide Occurrence in Serra do Mar, Angra dos Reis, Brazil

In the tropical environment such as Brazil, the frequency of rainfall-induced landslides is particularly high because of the rugged terrain, heavy rainfall, increasing urbanization, and the orographic effect of mountain ranges. Since such landslides repeatedly interfere with human activities and infrastructures, improved knowledge related to spatial and temporal prediction of the phenomenon is of interest for risk management. This study is an analysis of empirical rainfall thresholds, which aims to establish local and regional scale correlations between rainfall and the triggering of landslides in Angra dos Reis in the State of Rio de Janeiro. A statistical analysis combining quantile regression and binary logistic regression was performed on 1640 and 526 landslides triggered by daily rainfall over a 6-year period in the municipality and the urban center of Angra dos Reis, in order to establish probabilistic rainfall duration thresholds and assess the role of antecedent rainfall. The results show that the frequency of landslides is highly correlated with rainfall events, and surprisingly the thresholds in dry season are lower than those in wet season. The aspect of the slopes also seems to play an important role as demonstrated by the different thresholds between the southern and northern regions. Finally, the results presented in this study provide new insight into the spatial and temporal dynamics of landslides and rainfall conditions leading to their activation in this tropical and mountainous environment.

Exploring complementarities between modelling approaches that enable upscaling from plant community functioning to ecosystem services as a way to support agroecological transition

Promoting plant diversity through crop mixtures is a mainstay of the agroecological transition. Modelling this transition requires considering both plant-plant interactions and plants’ interactions with abiotic and biotic environments. Modelling crop mixtures enables designing ways to use plant diversity to provide ecosystem services, as long as they include crop management as input. A single modelling approach is not sufficient, however, and complementarities between models may be critical to consider the multiple processes and system components involved at different and relevant spatial and temporal scales. In this article, we present different modelling solutions implemented in a variety of examples to upscale models from local interactions to ecosystem services. We highlight that modelling solutions (i.e. coupling, metamodelling, inverse or hybrid modelling) are built according to modelling objectives (e.g. understand the relative contributions of primary ecological processes to crop mixtures, quantify impacts of the environment and agricultural practices, assess the resulting ecosystem services) rather than to the scales of integration. Many outcomes of multispecies agroecosystems remain to be explored, both experimentally and through the heuristic use of modelling. Combining models to address plant diversity and predict ecosystem services at different scales remains rare but is critical to support the spatial and temporal prediction of the many systems that could be designed.

Temporal Monitoring and Predicting of the Abundance of Malaria Vectors Using Time Series Analysis of Remote Sensing Data through Google Earth Engine

Background: In many studies in the field of malaria, environmental factors have been acquired in single-time, multi-time or a short time series using remote sensing and meteorological data. Selecting the best periods of the year to monitor the habitats of Anopheles larvae can be effective in better and faster control of malaria outbreak.Methods: In this article, high-risk times for three regions in Iran, including Qaleh-Ganj, Sarbaz and Bashagard counties with history of malaria prevalence had been estimated. For this purpose, a series of environmental factors affecting the growth and survival of Anopheles had been used over a seven-year period through the GEE. Environmental factors used in this study include NDVI and LST extracted from Landsat-8 satellite images, daily precipitation data from PERSIANN-CDR, soil moisture data from NASA-USDA Enhanced SMAP, ET data from MODIS sensor, and vegetation health indices included TCI and VCI extracted from MODIS sensors. All these parameters were extracted on a monthly average for seven years and, their results were fused at the decision level using majority voting method to estimate high-risk time in a year.Results: The results of this study indicated that there were two high-risk times for all three study areas in a year to increase the abundance of Anopheles mosquitoes. The first peak occurred from late winter to late spring and the second peak from late summer to mid-autumn. If there is a malaria patient in the area, after the end of the Anopheles larvae growth period, the disease will spread throughout the region. Further evaluation of the results against the entomological data available in previous studies showed that the high-risk times predicted in this study were consistent with the increase in the abundance of Anopheles mosquitoes in the study areas. Conclusions: The proposed method is very useful for temporal prediction of the increase of the abundance of Anopheles mosquitoes and also the use of optimal data with the aim of monitoring the exact location of Anopheles habitats. This study extracted high-risk time based on the analysis of the time series of remote sensing data.

Spatial-temporal habitat suitability for lemuru fish (Sardinella lemuru) using the Second-generation Global Imager (SGLI) and Maximum Entropy model in the Bali Strait, Indonesia

Lemuru fish (Sardinella lemuru), the most dominant fishery resource, has economic values for the fisherman fishing activities in the Bali Strait (between Jawa and Bali islands), Indonesia. Spatial and temporal prediction for the fishing location is essential information for effective fisheries management. The high spatial resolution of sea surface temperature (SST) and Chlorophyll-a (Chl-a) by the second-generation global imager (SGLI) on the global change observation mission (GCOM-C) satellite was employed for the input of the Maximum Entropy Model (MaxEnt) to predict the potential fishing area of lemuru fish in 2020. This study analyzed SST and Chl-a using the SGLI data and shows the variability of SST and Chl-a for lemuru fish-catching data. The MaxEnt model performance to predict the habitat suitability for lemuru fish in the Bali Strait has been shown in this study. As a result, the maximum average Chl-a estimated in August 2020 was around 1.62 mg m−3 and maximum SST in March 2020 around 28.12°C. The correlation between SST and Chl-a with total lemuru fish-catching were -0.209 and 0.375 for SST and Chl-a, respectively. The prediction of lemuru fishing areas using the MaxEnt model showed excellent model evaluations with a correlation value higher than 0.80.