Comparison of Different Spatial Regression Methods for Predicting Organic Matter Content in Soil

Dynamic changes in soil organic matter content affects the sustainable supply of soil water and fertilizer and impacts the stability of soil ecological function. Understanding the spatial distribution characteristics of soil organic matter will help deepen our understanding of the differences in soil organic matter content, soil formation law; such understanding would be useful for rational land use planning. Taking terrain data, meteorological data, and remote sensing data as auxiliary variables and the ordinary Kriging (OK) method as a control, this study compares the spatial prediction accuracies and mapping effects of various models (MLR, RK, GWR, GWRK, MGWR, and MGWRK) on soil organic matter. Our results show that the spatial distribution trend of soil organic matter predicted by each model is similar, but the prediction of composite models can reflect more mapping details than that of unitary models. The OK method can provide better support for spatial prediction when the sampling points are dense; however, the local models are superior in dealing with spatial non-stationarity. Notably, the MGWR model is superior to the GWR model, but the MGWRK model is inferior to the GWRK model. As a new method, the prediction accuracy of MGWRK reached 47.72% for the OK and RK methods and 40.08% for the GWRK method. The GWRK method achieved a better prediction accuracy. The influence mechanism of soil organic matter is complex, but the MGWR model more clearly reveals the complex nonlinear relationship between soil organic matter content and factors influencing it. This research can provide reference methods and mapping technical support to improve the spatial prediction accuracy of soil organic matter.

Spatial distribution of above ground carbon density in Harana Forest, Ethiopia

Background The need for understanding spatial distribution of forest aboveground carbon density (ACD) has increased to improve management practices of forest ecosystems. This study examined spatial distribution of the ACD in the Harana Forest. A grid sampling technique was employed and three nested circular plots were established at each point where grids intersected. Forest-related data were collected from 1122 plots while the ACD of each plot was estimated using the established allometric equation. Environmental variables in raster format were downloaded from open sources and resampled into a spatial resolution of 30 m. Descriptive statistics were computed to summarize the ACD. A Random Forest classification model in the R-software package was used to select strong predictors, and to predict the spatial distribution of ACD. Results The mean ACD was estimated at 131.505 ton per ha in this study area. The spatial prediction showed that the high class of the ACD was confined to eastern and southwest parts of the Harana Forest. The Moran’s statistics depicted similar observations showing the higher clustering of ACD in the eastern and southern parts of the study area. The higher ACD clustering was linked with the higher species richness, species diversity, tree density, tree height, clay content, and SOC. Conversely, the lower ACD clustering in the Harana Forest was associated with higher soil cation exchange capacity, silt content, and precipitation. Conclusions The spatial distribution of ACD in this study area was mainly influenced by attributes of the forest stand and edaphic factors in comparison to topographic and climatic factors. Our findings could provide basis for better management and conservation of aboveground carbon storage in the Harana Forest, which may contribute to Ethiopia’s strategy of reducing carbon emission.

TARA-Net: A Fusion Network for Detecting Takeaway Rider Accidents

In the emerging business of food delivery, rider traffic accidents raise financial cost and social traffic burden. Although there has been much effort on traffic accident forecasting using temporal-spatial prediction models, none of the existing work studies the problem of detecting the takeaway rider accidents based on food delivery trajectory data. In this article, we aim to detect whether a takeaway rider meets an accident on a certain time period based on trajectories of food delivery and riders’ contextual information. The food delivery data has a heterogeneous information structure and carries contextual information such as weather and delivery history, and trajectory data are collected as a spatial-temporal sequence. In this article, we propose a TakeAway Rider Accident detection fusion network TARA-Net to jointly model these heterogeneous and spatial-temporal sequence data. We utilize the residual network to extract basic contextual information features and take advantage of a transformer encoder to capture trajectory features. These embedding features are concatenated into a pyramidal feed-forward neural network. We jointly train the above three components to combine the benefits of spatial-temporal trajectory data and sparse basic contextual data for early detecting traffic accidents. Furthermore, although traffic accidents rarely happen in food delivery, we propose a sampling mechanism to alleviate the imbalance of samples when training the model. We evaluate the model on a transportation mode classification dataset Geolife and a real-world Ele.me dataset with over 3 million riders. The experimental results show that the proposed model is superior to the state-of-the-art.