Search for: All records

Forest disturbances, such as an eastern spruce budworm ( Choristoneura fumiferana ) outbreak, impact the strength and persistence of forest carbon sinks. Salvage harvests are a typical management response to widespread tree mortality, but the decision to salvage mortality has large implications for the fate of carbon stocks (including forest carbon and harvested wood products) in the near and long terms. In this study, we created decision-support models for salvage harvesting based on carbon after an eastern spruce budworm outbreak. We used lasso regression to determine which stand characteristics (e.g., basal area) are the best predictors of carbon 40 years after an outbreak in both salvage and no salvage scenarios. We modeled carbon at year 40 for different treatment scenarios and discount rates. Treatment scenarios represent residual stand conditions that may be present when an outbreak occurs. Economic discount rates were applied to 40-year carbon values to account for near and long-term carbon storage aspects. We found that the volume and size of eastern spruce budworm host species are significant predictors of salvage preference based on carbon. We found overall that salvaging less volume is recommended to avoid major swings in carbon budgets and that discounting carbon values to apply weight to near or long-term sequestration greatly affects whether salvaging is preferred. Lasso models are constructed for the northeastern US, however, similar concepts may be applied beyond our study area and potentially for other insect outbreaks similar to spruce budworm, such as mountain pine beetle ( Dendroctonus ponderosae ) or hemlock woolly adelgid ( Adelges tsugae ). From a policy standpoint widespread salvaging could create a large carbon emissions deficit with the risk of not being fully replenished within a desired timeframe. Since salvaging is often financially driven, especially for private landowners, carbon market payments or incentives for not salvaging is a consideration for future policy. 

The difficulty in specifying rewards for many real world problems has led to an increased focus on learning rewards from human feedback, such as demonstrations. However, there are often many different reward functions that explain the human feedback, leaving agents with uncertainty over what the true reward function is. While most policy optimization approaches handle this uncertainty by optimizing for expected performance, many applications demand risk-averse behavior. We derive a novel policy gradient-style robust optimization approach, PG-BROIL, that optimizes a soft-robust objective that balances expected performance and risk. To the best of our knowledge, PG-BROIL is the first policy optimization algorithm robust to a distribution of reward hypotheses which can scale to continuous MDPs. Results suggest that PG-BROIL can produce a family of behaviors ranging from risk-neutral to risk-averse and outperforms state-of-the-art imitation learning algorithms when learning from ambiguous demonstrations by hedging against uncertainty, rather than seeking to uniquely identify the demonstrator’s reward function. 

This paper evaluates the ability of two different data-driven models to detect and localize simulated structural damage in an in-service bridge for long-term structural health monitoring (SHM). Strain gauge data collected over 4 years is used to characterize the undamaged state of the bridge. The Powder Mill Bridge in Barre, Massachusetts, U.S., which has been instrumented with strain gauges since its opening in 2009, is used as a case study, and the strain gauges used in this study are located at 26 different stations throughout the bridge superstructure. A linear regression (LR) model and an artificial neural network (ANN) model are evaluated based on the following criteria: (a) the ability to accurately predict the strain at each location in the undamaged state of the bridge; (b) the ability to detect simulated structural damage to the bridge superstructure; and (c) the ability to localize simulated structural damage. Both the LR and the ANN models were able to predict the strain at the 26 stations with an average error of less than 5%, indicating that both methodologies were effective in characterizing the undamaged state of the bridge. A calibrated finite element model was then used to simulate damage to the Powder Mill Bridge for three damage scenarios: fascia girder corrosion, girder fracture, and deck delamination. The LR model proved to be just as effective as the ANN model at detecting and localizing damage. A recommended protocol is thus presented for integrating data-driven models into bridge asset management systems. 

What policy should be employed in a Markov decision process with uncertain parameters? Robust optimization answer to this question is to use rectangular uncertainty sets, which independently reflect available knowledge about each state, and then obtains a decision policy that maximizes expected reward for the worst-case decision process parameters from these uncertainty sets. While this rectangularity is convenient computationally and leads to tractable solutions, it often produces policies that are too conservative in practice, and does not facilitate knowledge transfer between portions of the state space or across related decision processes. In this work, we propose non-rectangular uncertainty sets that bound marginal moments of state-action features defined over entire trajectories through a decision process. This enables generalization to different portions of the state space while retaining appropriate uncertainty of the decision process. We develop algorithms for solving the resulting robust decision problems, which reduce to finding an optimal policy for a mixture of decision processes, and demonstrate the benefits of our approach experimentally.