Optimized placement of sensor networks by machine learning for microclimate evaluation

Microclimate mapping and monitoring are of fundamental importance to manage natural resources and optimize agricultural procedures. Precision agriculture is based on the management of spatial-temporal microclimatic variation in fields monitored by IoT systems. Networks of microclimate sensors provide point- based measurements that can be used as input data for physical and artificial intelligence (AI) models to study variations of microclimatic conditions over several spatial and temporal scales. We propose and experimentally validate a computational framework based on AI algorithms to optimize and validate the placement of sensors networks according to local temperature variations within a study area located in the Lombardian foothills, Italy. The strategy involves a clustering procedure to extract spatial locations with a similar thermal behavior. An experimental validation has been then performed by deploying sensors in the optimized clusters to record temperature data. These data have been processed by a Nhits neural network trained to predict future temperature scenarios, to verify that predictions made inside each cluster are consistent and representative of a real temperature pattern. Our results indicate that the clustering optimization framework successfully identifies real temperature patterns within the study area.