Integration of web service technologies and smart sensors to create a unified decision support platform for forest ecosystem management

Abstract

The study addresses the growing risks to forest ecosystems caused by climate change, anthropogenic pressure and extreme events, requiring a transition from reactive monitoring to proactive management. The goal is to develop and test a unified decision support platform based on the integration of smart sensors and web services for data analysis in quasi-real time. Novelty – a service-oriented architecture that levels the heterogeneity of hardware and software and provides standardized access to key indicators of the state of the forest. The methods included the design of a multi-level architecture and the deployment of a field prototype. Physical layer – a network of nodes on ESP32 with solar power and SHT31 sensors, capacitive soil moisture, MQ-7 and all-weather microphones connected via LoRaWAN and NB-IoT. Raspberry Pi 4 border gateways performed aggregation, filtering, and buffering. Web layer – microservices on FastAPI in Docker with REST/HTTPS and JSON. Storage – InfluxDB; analytics – LSTM for fire risk and CNN for acoustics; interface – React with GIS. Polygon 10 hectares, 50 nodes, 3 gateways, 3 months; measurements of PDR, delays, energy efficiency, API load and metrics Precision, Recall, F1. The results showed the reliability of LoRaWAN on small packets, but the increase in latency with size, while NB-IoT provided PDR >99.6% and subsecond delays by 128 bytes; LoRaWAN power consumption is 40–60% lower, which justifies a hybrid strategy. The API maintained low latencies at 50-500 RPS: /data/meteo 45.18-112.73 ms; /data/acoustic 68.91-185.44 ms. The models achieved high quality: LSTM for elevated fire risk with Precision 91,47%, Recall 88,23%, F1 = 0,898; CNN with F1 = 0,953 for chainsaw detection and F1 = 0,993 for normal conditions. The discussion revealed the resource intensity of Edge AI: aggregation consumed ~4,87% CPU and 180,3 mW, filtering 11,23% CPU, while local inference raised CPU load to 65,71% and added ~1,55 W, making complex models on gateways impractical. The microservice architecture ensures fault tolerance, horizontal scalability, and faster functional expansion. The platform shifts monitoring toward predictive analytics and rapid response. Future directions include the integration of remote sensing data, MLOps practices, enhanced cybersecurity, and organizational resilience.