In this paper, a Kalman framework is proposed for dynamic energy-saving in wireless sensor networks used to monitor urban noise pollution. The energy-saving framework implements a dynamic power management strategy (DPMS) with the Kalman algorithm that varies the sensor node’s sleep period according to the measured noise levels. An Internet-of-Things (IoT) edge-based self-sustaining Long Range (LoRa) network is developed and used for ubiquitous monitoring and analysis of urban noise pollution. The network consists of a star topology with six LoRa battery-free wireless sensing nodes deployed in Mérida city downtown, each node powered by a green facade structure consisting of an array of Plant-Microbial Fuel Cells (P-MFC). The sensor node’s prototype was implemented following Mexican regulations, transmitting the data packets with the open frequency band of 915 MHz and monitoring the LoRa network from a web page. Experimental results proof a sustainable operation with a green facade P–MFC array power generation of 112.1 mW with an open circuit voltage of 2.7 V and a short-circuit current of 180 mA. The sensor node’s average power consumption is 11.2 mW, therefore sufficient energy is generated for continuous monitoring. The efficient Kalman-DPMS is also tested with the urban noise measurement estimation and adjusting the sleep period only if the urban measurement state estimation is above the threshold normativity. The system’s low-power consumption allows to perform 70 continuous LoRa transmissions even when the energy harvester source is absent, and the power capacity of the green facade restores a supercapacitor full charge after only 4 minutes of a LoRa transmission. A 23.6% of energy was saved with Kalman-DPMS in comparison with a continuous measurement system of 10 min uniform-sleep period.

Since 2014, the Mexican Caribbean has faced an ecological and socio-economic crisis due to massive coastal landings of pelagic sargassum. This study presents a comprehensive methodology for sargassum detection using Machine Learning (ML) models applied to imagery from a coastal video monitoring station (EVMC). Three machine learning techniques were implemented to classify sargassum on sand and water, Support Vector Machines (SVM), Random Forest (RF), and a Multi-Layer Perceptron (MLP) Artificial Neural Network. The performance of these ground-based models was then compared against sargassum detections from Sentinel-2 satellite data using the Floating Algae Index (FAI) and the Normalized Difference Vegetation Index (NDVI). The results demonstrated high efficacy, with the MLP model proving most effective for detecting sargassum on sand with a F1-score > 0.86, whilst the Random Forest model performing best in water with a F1-score > 0.75. A significant positive correlation was found between the video-based detections and indices derived from satellite data, with NDVI showing a consistently stronger correlation in both environments. This study validates coastal video monitoring as a reliable tool for local sargassum quantification and suggests that fusing this high-resolution ground data with wide-area satellite imagery offers a promising path toward a more accurate and comprehensive sargassum monitoring system.