Search

Ecological communities are dynamic. These dynamics are influenced by many sources of variation, making it difficult to understand or predict future change. Biotic interactions, and other sources of multi-species dependence, are major contributors. But ecological prediction overwhelmingly focuses on models that treat individual species in isolation. Here, we model the relative importance of nonlinear environmental responses and multi-species temporal dependencies for a community of semi-arid rodents. We use a hierarchical, Dynamic Generalized Additive Model (DGAM) to analyze monthly capture time series for nine rodents across a 25-year period. A vector autoregression to model unobserved trends allowed us to ask targeted questions about population dynamics. We find that multi-species dependencies are important for capturing unmeasured drivers of community change. Variation in captures for some species are expected to have delayed, often nonlinear effects on captures for others. These complexities are useful for inference but also for prediction. Models that captured multi-species dependence gave better near-term forecasts of community change than models that ignored it. We also quantify nonlinear effects of temperature change and positive effects of vegetation greenness on captures for nearly all species. Models that describe biological complexity, both through nonlinear covariate functions and multi-species dependence, are useful to ask targeted questions about population dynamics and drivers of change.

Advances in artificial intelligence for computer vision hold great promise for increasing the scales at which ecological systems can be studied. The distribution and behavior of individuals is central to ecology, and computer vision using deep neural networks can learn to detect individual objects in imagery. However, developing supervised models for ecological monitoring is challenging because it requires large amounts of human-labeled training data, requires advanced technical expertise and computational infrastructure, and is prone to overfitting. This limits application across space and time. One solution is developing generalized models that can be applied across species and ecosystems. Using over 250,000 annotations from 13 projects from around the world, we develop a general bird detection model that achieves over 65% recall and 50% precision on novel aerial data without any local training despite differences in species, habitat, and imaging methodology. Fine-tuning this model with only 1000 local annotations increases these values to an average of 84% recall and 69% precision by building on the general features learned from other data sources. Retraining from the general model improves local predictions even when moderately large annotation sets are available and makes model training faster and more stable. Our results demonstrate that general models for detecting broad classes of organisms using airborne imagery are achievable. These models can reduce the effort, expertise, and computational resources necessary for automating the detection of individual organisms across large scales, helping to transform the scale of data collection in ecology and the questions that can be addressed.

Advances in artificial intelligence for computer vision hold great promise for increasing the scales at which ecological systems can be studied. The distribution and behavior of individuals is central to ecology, and computer vision using deep neural networks can learn to detect individual objects in imagery. However, developing supervised models for ecological monitoring is challenging because it needs large amounts of human-labeled training data, requires advanced technical expertise and computational infrastructure, and is prone to overfitting. This limits application across space and time. One solution is developing generalized models that can be applied across species and ecosystems. Using over 250,000 annotations from 13 projects from around the world, we develop a general bird detection model that achieves over 65% recall and 50% precision on novel aerial data without any local training despite differences in species, habitat, and imaging methodology. Fine-tuning this model with only 1000 local annotations increase these values to an average of 84% recall and 69% precision by building on the general features learned from other data sources. Retraining from the general model improves local predictions even when moderately large annotation sets are available and makes model training faster and more stable. Our results demonstrate that general models for detecting broad classes of organisms using airborne imagery are achievable. These models can reduce the effort, expertise, and computational resources necessary for automating the detection of individual organisms across large scales, helping to transform the scale of data collection in ecology and the questions that can be addressed.

Remote sensing of forested landscapes can transform the speed, scale, and cost of forest research. The delineation of individual trees in remote sensing images is an essential task in forest analysis. Here we introduce a new Python package, DeepForest, that detects individual trees in high resolution RGB imagery using deep learning.While deep learning has proven highly effective in a range of computer vision tasks, it requires large amounts of training data that are typically difficult to obtain in ecological studies. DeepForest overcomes this limitation by including a model pre-trained on over 30 million algorithmically generated crowns from 22 forests and fine-tuned using 10,000 hand-labeled crowns from 6 forests.The package supports the application of this general model to new data, fine tuning the model to new datasets with user labeled crowns, training new models, and evaluating model predictions. This simplifies the process of using and retraining deep learning models for a range of forests, sensors, and spatial resolutions.We illustrate the workflow of DeepForest using data from the National Ecological Observatory Network, a tropical forest in French Guiana, and street trees from Portland, Oregon.

Recent security threats against cyber-physical critical power grid infrastructures have further distinguished the differences and complex interdependencies between optimal plant control and infrastructural safety topics. In this paper, we reflect upon few real-world scenarios and threats to understand how those two topics meet. We then propose a practical architectural solutions to address the corresponding concerns. As a first concrete step, we focus on networked industrial control systems in smart grid where several sensing-processing-actuation embedded nodes receive information, make control decisions, and carry out optimal actions. Traditionally, global safety maintenance, e.g., transient stability, is embedded into control and taken into account by the decision making modules. With recent cyber security-induced safety incidents, we believe that the safety-handling modules should also be considered as a part of global trusted computing base (attack surface) for security purposes. Generally, maximizing the system's overall security requires the designers to minimize its trusted computing base. Consequently, we argue that the traditional combined safety-control system architecture is not anymore the optimal design paradigm to follow given existing threats. Instead, we propose PLCLOUD, a new cloud-based safety-preserving architecture that places a minimal trusted safety verifier layer between the physical world and the cyber-based supervisory control and data acquisition (SCADA) infrastructure, specifically programmable logic controllers (PLCs). PLCLOUD's main objective is to take care of infrastructural safety and separate it from optimal plant control that SCADA is responsible for.