PearTree

Inspiration

Urban trees in cities like Barcelona are under increasing stress from heatwaves, drought, and hidden structural decay that is not visible until failure occurs. We wanted to build a system that turns trees into continuously monitored living infrastructure, making environmental stress visible and understandable in real time.

What it does

PearTree is a real-time urban tree health monitoring system that uses low-cost sensors to measure structural vibration, temperature, and humidity. It computes a live stress index for each tree and translates it into an intuitive “mood” that the public can understand. Users can also interact with the tree through a chatbot interface that responds based on its real-time health state.

How we built it

We used an Arduino UNO Q with Modulino sensors for vibration (accelerometer) and environmental data (temperature and humidity). On the software side, we built a Python pipeline that performs signal filtering, feature extraction (RMS, FFT-based features), and unsupervised anomaly detection using a baseline model. The system fuses environmental and structural stress into a single health score, which is streamed to a web UI in real time. A chatbot layer connects a generative AI API to the system, conditioning responses on the current tree state.

Challenges we ran into

We struggled with correctly interpreting vibration data due to noise, gravity effects, and inconsistent sampling timing. Another major challenge was stabilizing the anomaly detection so that it reflected meaningful structural changes instead of sensor jitter. Integrating real-time sensor streaming with AI responses without latency issues was also non-trivial.

Accomplishments that we're proud of

We successfully built a full end-to-end system that combines embedded sensing, real-time signal processing, anomaly detection, and a public-facing interaction layer. We are especially proud that the system works continuously in real time and translates raw physical signals into an intuitive and engaging “tree health” experience.

What we learned

We learned how sensitive real-world sensor systems are to noise and timing assumptions, and how important proper filtering and windowing are for stable inference. We also gained experience in building unsupervised anomaly detection pipelines and learned how to meaningfully combine physical signals with environmental data into a single interpretable metric.

What’s next for PearTree

We plan to implement peer-to-peer communication between nearby trees, allowing them to share local environmental and stress information to build a more collaborative, city-wide understanding of urban health. Additionally, we aim to improve the chatbot with more grounded environmental reasoning and integrate city infrastructure data (traffic, weather, pollution) to better contextualize tree stress at a broader scale.

Built With

• arduino
• python