Restorebot: Developing autonomous robotic platforms to support restoration of degraded grazing land ecosystems

Background/Question/Methods

Grazing lands cover approximately 45% of the terrestrial land surface and are highly vulnerable to degradation processes. Restoration of degraded grazing lands is highly prone to failure and is met with limited success, especially with using more traditional ecological restoration approaches. Plant recruitment in degraded grazing lands can be limited by a broad variety of factors, but two of the most critical bottlenecks to successfully restoring native plant communities are available seeds and favorable microsites that are suitable for plant recruitment. Increasing seed availability through seeding has long been a “business as usual” approach for restoring degraded drylands but is often met with low success rates due to a suite of environmental and physical barriers to plant establishment. Following this, we posit that with better microsite selection and more precise seed applications, plant establishment may be greatly improved. To achieve more precise seed application on a large scale, we are developing a fully autonomous robotic platform to determine the most suitable microsites for planting and to continuously monitor vegetation, thereby improving our understanding of the factors which contribute to success in these restoration efforts over time.

Results/Conclusions

In this study we were able to accurately identify and map favorable microsites for plant germination and recruitment such as litter, soil cracks, and micro-depressions using computer vision methods. We were able to achieve this by using convolutional neural networks to analyze images taken by the robot in the field. In addition to identifying conditions suitable for seedling establishment, the robot monitors plant maturation by uniquely identifying individual plants in a dynamic map. This iterative mapping allows us to understand how ongoing restoration efforts affect plant mortality and growth over long periods of time. We then use these data to correlate environmental factors (e.g. soil erosion, microsite type, presence of existing vegetation) to the success of restoration projections. We were able to monitor plant cover and seasonal terrain changes by means of this dynamic map which is updated by state of the art techniques in statistical data association and state estimation. As global efforts to reverse land degradation are strengthened by the recent launch of the UN Decade of Restoration (2021-2030), the development of autonomous robotics platforms adds to the suite of cutting edge technological approaches to repair damaged ecosystems through “precision” restoration strategies.