Leveraging Environmental Monitoring UAS in Rainforests

This project is funded in part by the National Science Foundation through grant #NSF-NRI: 1925368.

Project Description:

We will develop and assess techniques, tools, and systems that dramatically improve the potential for UASs to track water transportation through forest canopies and below ground for monitoring in difficult to access environments. This proposal builds on four years of successful deployments of a water sampling system and two years of successful development and emplacement of sensors in target soils, while extending capabilities to sample from trees and address key NRI 2.0 themes. The proposed work will enable Unmanned Aerial Systems (UASs) to collect samples in an hour, which would have previously been inaccessible or taken a human team multiple days while also improving the ability of the scientific team to evaluate sample sites through targeted communications, the collection of samples (soil, water, and leaf) in challenging environments, and facilitate better scientific decision-making through improved resolution of data and expansion of areas that can be monitored. This project will be assessed on both the capabilities of the developed system, but also on its ability to impact the monitoring of increasingly endangered ecosystems and advance fundamental science in forest canopies.

This proposal presents a vision aimed at advancing heterogeneous multi-UAS technologies, practices, and understanding to increase the reach of human sensing in challenging, hard-to-access environments while increasing scientific understanding of forest canopy health. The vision addresses key goals in co-robotic system development with regards to the available attention of the humans involved, site selection for complementary sampling, and improvements in robot design and decision making for sample collection. These goals will be developed in local, well-understood environments before being refined in yearly tests in the harsh, cluttered forest contexts, all while contributing to progress in fundamental co-robotic challenges.

  • Goal 1: Targeted communications for expert feedback. Although it can be assumed that a scientist will be onsite, there is the potential that not every UAS will be individually monitored by a scientist. During sample collection, the scientist(s) will likely be focused on the success of all vehicles rather than monitoring one. In this case, the vehicle will need to attract attention (to either itself or the interface) only when necessary to gain important feedback from the expert user.
  • Goal 2: Automated perceptive-assistive site selection. In order for these systems to facilitate sample collections in complementary locations at the same time, site selection will need to be collaboratively completed in an efficient manner. This will require both an understanding of the science (to monitor locations with appropriate characteristics) and the ability to perceive the environment to identify optimal sites based on vegetation type, proximity to water, and amount of sunlight.
  • Goal 3: Actuating and Assessing Sample Collection Success. While the systems for sensor emplacement and water sampling have been developed and extensively tested in well-defined areas and temperatures, deployment in new environments and collection of new types of samples (e.g., leaf) will require additional research on both the mechanisms and algorithms. We also need sampling algorithms that can succeed over a range of conditions and that can determine if the sample is properly collected (e.g., that the leaf samples were taken from the sunny top of the canopy and not further into the shaded canopy). To do so, we will focus on using reinforcement learning, but segmenting the problem to close the reality gap between simulations and lab experiments and between lab experiments and field trials.
  • Cross-cutting Goal 4: Enable safer, less destructive, and more complete data collection to improve scientific decision-making. Using the methods developed here, scientists will be able to obtain more complete information while limiting disruption to the surrounding ecosystem as well as their risk of falls and bites. Spatially explicit data on plant hydraulics, canopy leaf temperature and wetness, streamflow, water chemistry, and soil moisture will lead to richer, more accurate representations of the forest water cycle at hillslope to watershed scales. In turn, these will help constrain Earth system models of tropical rainforests, potentially leading to better weather and climate predictions.


  • PI: Brittany Duncan
  • Co-PI: Carrick Detweiler
  • GRA: Siya Kunde
  • GRA: Alisha Bevins
  • GRA: Paul Fletcher
  • UGRAD: Evan Palmer


  • Joshua Peschel (Iowa State University)
  • Gretchen Miller (Texas A&M University)