Fully distributed rainfall-runoff modeling using spatial-temporal graph neural network

Recent studies using latest deep learning algorithms such as LSTM (Long Short-Term Memory) have shown great promise in time-series modeling. There are many studies focusing on the watershed-scale rainfall-runoff modeling or streamflow forecasting, often considering a single watershed with limited generalization capabilities. To improve the model performance, several studies explored an integrated approach by decomposing a large watershed into multiple sub-watersheds with semi-distributed structure. In this study, we propose an innovative physics-informed fully-distributed rainfall-runoff model, NRM-Graph (Neural Runoff Model-Graph), using Graph Neural Networks (GNN) to make full use of spatial information including the flow direction and geographic data. Specifically, we applied a time-series model on each grid cell for its runoff production. The output of each grid cell is then aggregated by a GNN as the final runoff at the watershed outlet. The case study shows that our GNN based model successfully represents the spatial information in predictions. NRM-Graph network has shown less over-fitting and a significant improvement on the model performance compared to the baselines with spatial information. Our research further confirms the importance of spatially distributed hydrological information in rainfall-runoff modeling using deep learning, and we encourage researchers to incorporate more domain knowledge in modeling.

Toroidal topology of population activity in grid cells

The medial entorhinal cortex is part of a neural system for mapping the position of an individual within a physical environment1. Grid cells, a key component of this system, fire in a characteristic hexagonal pattern of locations2, and are organized in modules3 that collectively form a population code for the animal’s allocentric position1. The invariance of the correlation structure of this population code across environments4,5 and behavioural states6,7, independent of specific sensory inputs, has pointed to intrinsic, recurrently connected continuous attractor networks (CANs) as a possible substrate of the grid pattern1,8–11. However, whether grid cell networks show continuous attractor dynamics, and how they interface with inputs from the environment, has remained unclear owing to the small samples of cells obtained so far. Here, using simultaneous recordings from many hundreds of grid cells and subsequent topological data analysis, we show that the joint activity of grid cells from an individual module resides on a toroidal manifold, as expected in a two-dimensional CAN. Positions on the torus correspond to positions of the moving animal in the environment. Individual cells are preferentially active at singular positions on the torus. Their positions are maintained between environments and from wakefulness to sleep, as predicted by CAN models for grid cells but not by alternative feedforward models12. This demonstration of network dynamics on a toroidal manifold provides a population-level visualization of CAN dynamics in grid cells.

Habitat use of common dolphin (Delphinus sp.) in the Mochima National Park, Sucre, Venezuela.

In Venezuela, common dolphin (Delphinus sp.) is considered the cetacean with the highest incidence. Studies in the region indicate a possible isolated coastal population so called Venezuelan stock settled mainly in the northeast of the country. . The objective of this study is to describe the habitat use of common dolphin in the Mochima National Park (MNP), a protected area with a high and growing anthropic pressure. Seventy surveys were carried out, with predefined survey route, from September 2009 to August 2010. Each group sighted was monitored while possible to a maximum of 30min.. During this time we registered location (Latitude-longitude), behaviour, group size and composition every 5min. Additionally, environmental variables were assessed from the sight location in a nautical chart. The study area was divided into a grid (cell: 500 x 500m) and the Coefficient of Area Use (CAU) was calculated for each cell. The proportion of the total observation time where the common dolphin displayed behaviours into the areas being used was estimated. A logistic regression model was applied to identify the variables that better explained usage pattern. In 55h of observation, 111 groups were recorded. The common dolphin used the habitat differentially, showing preferences for shallow areas near to the coast. Areas of greatest intensity of use were Tigrillo inlet and the northeast of the Caracas Islands. The probability of presence of dolphins decreased with depth and distance to the coast. Common dolphin invested more time in feeding and socializing activities. Behaviours were significantly dependent of season, group size, composition, depth and distance to the coast. Finally, these data on habitat use and behaviour allow the identification of priority habitats. Throughout the year, the MNP provided areas for refuge, feeding and resting. , It is therefore imperative to promote management and conservation policies that prevent the negative impacts of the increasing tourism and fishing activities we observed in this Park.

Modeling subgrid lake energy balance in ORCHIDEE terrestrial scheme using the FLake lake model

The freshwater 1-D FLake lake model was coupled to the ORCHIDEE land surface model to simulate lake energy balance at the global scale. A multi-tile approach has been chosen to allow the modelling of various types of lakes within the ORCHIDEE grid cell. The different categories have been defined according to lake depth which is the most influential parameter of FLake, but other properties could be considered in the future. Several depth parameterization strategies have been compared, differing by the way to aggregate the depth of the subgrid lakes, i.e., arithmetical, geometrical, harmonical mean and median. Five atmospheric reanalysis datasets available at 0.5° or 0.25° resolution, have been used to force the model and assess model systematic errors. Simulations have been performed, evaluated and intercompared against observations of lake water temperatures provided by the GloboLakes database over about 1000 lakes and ice phenology derived from the Global Lake and River Ice Phenology database. The results highlighted the large impact of the atmospheric forcing on the lake energy budget simulations and the improvements brought by the highest resolution products (ERA5 and E2OFD). The median of the Root Square Mean Errors (RMSE) calculated at global scale range between 3.2 K and 2.7 K among the forcings, CRUJRA and ERA5 leading respectively to the best and worst results. Depth parameterization strategy appeared to be less influent, with RMSE differences less than 0.1 K for the four aggregation scenarios tested. The simulation of ice phenology presented systematic errors whatever the forcing used and the depth parameterization. Freezing onset was shown to be the less sensitive to forcing and depth parameterization with median of the errors ranging between 10 and 14 days. Larger errors were observed on the simulation of the end of the freezing period significantly influenced by the atmospheric forcing used. Such errors already highlighted in previous works, could be the result of deficiencies in the modeling of snow/ice parameterization processes. Various pathways are drawn to improve the model results, including the use of remote sensing data to better constrain the lake radiative parameters (albedo and extinction coefficient) as well as the lake depth thanks to the recent and forthcoming high resolution satellite missions.

Effects of harvester positioning errors on merchantable timber volume predicted and estimated from airborne laser scanner data in mature Norway spruce forests

Newly developed positioning systems in cut-to-length harvesters enable georeferencing of individual trees with submeter accuracy. Together with detailed tree measurements recorded during processing of the tree, georeferenced harvester data are emerging as a valuable tool for forest inventory. Previous studies have shown that harvester data can be linked to airborne laser scanner (ALS) data to estimate a range of forest attributes. However, there is little empirical evidence of the benefits of improved positioning accuracy of harvester data. The two objectives of this study were to (1) assess the accuracy of timber volume estimation using harvester data and ALS data acquired with different scanners over multiple years and (2) assess how harvester positioning errors affect merchantable timber volume predicted and estimated from ALS data. We used harvester data from 33 commercial logging operations, comprising 93 731 harvested stems georeferenced with sub-meter accuracy, as plot-level training data in an enhanced area-based inventory approach. By randomly altering the tree positions in Monte Carlo simulations, we assessed how prediction and estimation errors were influenced by different combinations of simulated positioning errors and grid cell sizes. We simulated positioning errors of 1, 2, …, 15 m and used grid cells of 100, 200, 300 and 400 m. Values of root mean square errors obtained for cell-level predictions of timber volume differed significantly for the different grid cell sizes. The use of larger grid cells resulted in a greater accuracy of timber volume predictions, which were also less affected by positioning errors. Accuracies of timber volume estimates at logging operation level decreased significantly with increasing levels of positioning error. The results highlight the benefit of accurate positioning of harvester data in forest inventory applications. Further, the results indicate that when estimating timber volume from ALS data and inaccurately positioned harvester data, larger grid cells are beneficial.2

Matrix Refinement in Mass Transport Across Fracture-Matrix Interface: Application to Improved Oil Recovery in Fractured Reservoirs

In this paper, we will show that it is highly beneficial to model dual-porosity reservoirs using matrix refinement (similar to the multiple interacting continua, MINC, of Preuss, 1985) for water displacing oil. Two practical situations are considered. The first is the effect of matrix refinement on the unsteady-state pressure solution, and the second situation is modeling water-oil, Buckley-Leverett (BL) displacement in waterflooding a fracture-dominated flow domain. The usefulness of matrix refinement will be illustrated using a three-node refinement of individual matrix blocks. Furthermore, this model was modified to account for matrix block size variability within each grid cell (in other words, statistical distribution of matrix size within each grid cell) using a discrete matrix-block-size distribution function. The paper will include two mathematical models, one unsteady-state pressure solution of the pressure diffusivity equation for use in rate transient analysis, and a second model, the Buckley-Leverett model to track saturation changes both in the reservoir fractures and within individual matrix blocks. To illustrate the effect of matrix heterogeneity on modeling results, we used three matrix bock sizes within each computation grid and one level of grid refinement for the individual matrix blocks. A critical issue in dual-porosity modeling is that much of the fluid interactions occur at the fracture-matrix interface. Therefore, refining the matrix block helps capture a more accurate transport of the fluid in-and-out of the matrix blocks. Our numerical results indicate that the none-refined matrix models provide only a poor approximation to saturation distribution within individual matrices. In other words, the saturation distribution is numerically dispersed; that is, no matrix refinement causes unwarranted large numerical dispersion in saturation distribution. Furthermore, matrix block size-distribution is more representative of fractured reservoirs.

Long-term atmospheric emissions for the Coal Oil Point natural marine hydrocarbon seep field, offshore California

In this study, we present a novel approach for assessing nearshore seepage atmospheric emissions through modeling of air quality station data, specifically a Gaussian plume inversion model. A total of 3 decades of air quality station meteorology and total hydrocarbon concentration, THC, data were analyzed to study emissions from the Coal Oil Point marine seep field offshore California. THC in the seep field directions was significantly elevated and Gaussian with respect to wind direction, θ. An inversion model of the seep field, θ-resolved anomaly, THC′(θ)-derived atmospheric emissions is given. The model inversion is for the far field, which was satisfied by gridding the sonar seepage and treating each grid cell as a separate Gaussian plume. This assumption was validated by offshore in situ data that showed major seep area plumes were Gaussian. Plume total carbon, TC (TC = THC + carbon dioxide, CO2, + carbon monoxide), 18 % was CO2 and 82 % was THC; 85 % of THC was CH4. These compositions were similar to the seabed composition, demonstrating efficient vertical plume transport of dissolved seep gases. Air samples also measured atmospheric alkane plume composition. The inversion model used observed winds and derived the 3-decade-average (1990–2021) field-wide atmospheric emissions of 83 400 ± 12 000 m3 THC d−1 (27 Gg THC yr−1 based on 19.6 g mol−1 for THC). Based on a 50 : 50 air-to-seawater partitioning, this implies seabed emissions of 167 000 m3 THC d−1. Based on atmospheric plume composition, C1–C6 alkane emissions were 19, 1.3, 2.5, 2.2, 1.1, and 0.15 Gg yr−1, respectively. The spatially averaged CH4 emissions over the ∼ 6.3 km2 of 25 × 25 m2 bins with sonar values above noise were 5.7 µM m−2 s−1. The approach can be extended to derive emissions from other dispersed sources such as landfills, industrial sites, or terrestrial seepage if source locations are constrained spatially.

Method for Fast Map Construction Based on GPS Data and Compressed Grid Algorithm

Electronic maps play an important role in the field of urban traffic management, but the interface functions provided by map service agencies are limited, and commercial maps are generally expensive. Furthermore, the map generation algorithms based on the Global Positioning System (GPS) data can be very complex and take up a lot of storage space, which limits their application to specific practical problems, such as the real-time update of area maps, temporary road control, emergency route planning, and other scenarios. In order to solve this problem, an intuitive, extensible, and flexible method of constructing urban road maps is proposed. Using the Othello-coordinated method, the representation of the unit grid cell was redesigned. Through this method, the disadvantages of the raster map’s large storage space and computing resource requirements are compensated for during processing, improving the topological expression ability of the raster map and the speed with which the construction of the map is realized. The application potential of the proposed method is demonstrated by the evaluation of public transport service and road network resilience. In our experiments, the optimization efficiency of storage space was up to 99.914%, and the calculation accuracy of bus coverage was about 99.86%.

Effect of reward on electrophysiological signatures of grid cell population activity in human spatial navigation

The regular equilateral triangular periodic firing pattern of grid cells in the entorhinal cortex is considered a regular metric for the spatial world, and the grid-like representation correlates with hexadirectional modulation of theta (4–8 Hz) power in the entorhinal cortex relative to the moving direction. However, researchers have not clearly determined whether grid cells provide only simple spatial measures in human behavior-related navigation strategies or include other factors such as goal rewards to encode information in multiple patterns. By analysing the hexadirectional modulation of EEG signals in the theta band in the entorhinal cortex of patients with epilepsy performing spatial target navigation tasks, we found that this modulation presents a grid pattern that carries target-related reward information. This grid-like representation is influenced by explicit goals and is related to the local characteristics of the environment. This study provides evidence that human grid cell population activity is influenced by reward information at the level of neural oscillations.

A structured scaffold underlies activity in the hippocampus

Place cells are believed to organize memory across space and time, inspiring the idea of the cognitive map. Yet unlike the structured activity in the associated grid and head-direction cells, they remain an enigma: their responses have been difficult to predict and are complex enough to be statistically well-described by a random process. Here we report one step toward the ultimate goal of understanding place cells well enough to predict their fields. Within a theoretical framework in which place fields are derived as a conjunction of external cues with internal grid cell inputs, we predict that even apparently random place cell responses should reflect the structure of their grid inputs and that this structure can be unmasked if probed in sufficiently large neural populations and large environments. To test the theory, we design experiments in long, locally featureless spaces to demonstrate that structured scaffolds undergird place cell responses. Our findings, together with other theoretical and experimental results, suggest that place cells build memories of external inputs by attaching them to a largely prespecified grid scaffold.