Environmental spatiotemporal prediction is crucial for air quality management, where accurate prediction of primary air pollutants is essential for public health and policy-making. This paper introduces the Conditioned Diffusion-based Graph Attention (CDGA) model, a novel Bayesian deep learning framework for spatiotemporal prediction of primary air pollutant ground-level concentrations. CDGA integrates Graph Attention Networks (GAT) and time series model order as conditioning inputs, enabling the model to capture both spatial and temporal dependencies while it provides uncertainty measures for the predictions. The proposed model is validated on two spatiotemporal benchmarks of NO2 and O3 ground-level concentrations, measured by EEA stations in Italy from 2014 to 2022. Experimental results demonstrate that CDGA outperforms state-of-the-art spatiotemporal models in terms of popular error metrics.

The study and modeling of spatio-temporal processes are essential in various fields, including geostatistics and environmental sciences. The presence of temporal and spatial dependencies in the data requires careful consideration to build accurate and reliable models, which imposes several challenges. In light of this, we propose a novel deep learning architecture tailored for spatio-temporal prediction in situations where data is sparse and irregularly sampled across both spatial and temporal dimensions, addressing scenarios of static and mobile sensors, mimicking data collection approaches commonly employed when monitoring large wild areas, such as rain forests. Our approach works by aggregating information from random sets of observed data points, learning a latent neighborhood representation, and mapping this latent vector to produce a real-valued prediction. The aggregation process utilizes the permutation invariant attention mechanism, enabling the model to identify which points from the set provide the most meaningful and unique information. Finally, through multiple predictions based on random sets of points, our architecture is able to provide the estimate of its uncertainty. We validated our proposal by experimentation where it outperformed the baseline considering the sampling biases associated with static and mobile sensors.

This research is a resume of two regarding the applications of machine learning modelling focusing on the power efficiency. The selected researches present two power-efficient models inspired from biological behavior of brain cells. The first one is based on spiking neural networks (SNNs) and its goal is to classify clinical images. SNNs are more energy-efficient compared to traditional artificial neural networks because they use power only when a spike is emitted. In fact, the structure of SNNs encourages sparse network activation. The second one paper is always related to image classification task, but it uses a different model, similar to SNNs, inspired from the field of membrane computing.

Land cover classification is vital in environmental science by supporting urban tracking, deforestation monitoring, and being the future cornerstone for climate solutions. More specifically, the EuroSAT dataset is a critical benchmark for global remote sensing research, advancing land-cover classification. Our study enhances EuroSAT’s utility by introducing a refined ensemble learning methodology to improve classification accuracy and integrate multiple specialized convolutional neural networks (CNN) with a final ensemble neural network. By doing so, this optimizes collective-level predictions, surpassing top global models. Our methodology yielded improvements over top worldwide EuroSAT classification benchmarks, with the model achieving 99.87±0.02% accuracy—a 0.63% gain over these top models. The evaluation framework incorporated multiple performance metrics to assess model reliability: F1-score (99.86±0.02%), precision (99.86±0.02%), recall (99.87±0.02%), Receiver Operating Characteristic Area Under the Curve (ROC AUC) score (99.99±0.00%), and a mean absolute error (MAE) of 0.04% ± 0.01%. Statistical validation confirmed the consistency of these results across multiple test scenarios. These findings highlight the potential of ensemble learning in satellite-based remote sensing applications, as compared to other top models worldwide. Our approach leverages the complementary strengths of multiple specialized models to deliver a more nuanced and accurate land cover classification system. This advancement significantly enhances environmental monitoring and urban planning capabilities while establishing a foundation for future innovations in remote sensing analysis. The results validate not only the effectiveness of ensemble learning in remote sensing but also demonstrate how thoughtful integration of multiple models can push the boundaries of environmental observation techniques.

Performing accurate confidence quantification and assessment in pixel-wise regression tasks, which are downstream applications of AI Foundation Models for Earth Observation (EO), is important for deep neural networks to predict their failures, improve their performance and enhance their capabilities in real-world applications, for their practical deployment. For pixel-wise regression tasks, specifically utilizing remote sensing data from satellite imagery in EO Foundation Models, confidence quantification is a critical challenge. The focus of this research is on developing a Foundation Model using EO satellite data that computes and assigns a confidence metric alongside regression outputs to improve the reliability and interpretability of predictions generated by deep neural networks. To this end, we develop, train and evaluate the proposed Confidence-Aware Regression Estimation (CARE) Foundation Model. Our model CARE computes and assigns confidence to regression results as downstream tasks of a Foundation Model for EO data, and performs a confidence-aware self-corrective learning method for the low-confidence regions. We evaluate the model CARE, and experimental results on multi-spectral data from the Copernicus Sentinel-2 constellation to estimate the building density (i.e. monitoring urban growth), show that the proposed method can be successfully applied to important regression problems in EO. We also show that our model CARE outperforms other methods.

Land cover classification is vital in environmental science by supporting urban tracking, deforestation monitoring, and being the future cornerstone for climate solutions. More specifically, the EuroSAT dataset is a critical benchmark for global remote sensing research, advancing land-cover classification. Our study enhances EuroSAT’s utility by introducing a refined ensemble learning methodology to improve classification accuracy and integrate multiple specialized convolutional neural networks (CNN) with a final ensemble neural network. By doing so, this optimizes collective-level predictions, surpassing top global models. Our methodology yielded improvements over top worldwide EuroSAT classification benchmarks, with the model achieving 99.87±0.02% accuracy—a 0.63% gain over these top models. The evaluation framework incorporated multiple performance metrics to assess model reliability: F1-score (99.86±0.02%), precision (99.86±0.02%), recall (99.87±0.02%), Receiver Operating Characteristic Area Under the Curve (ROC AUC) score (99.99±0.00%), and a mean absolute error (MAE) of 0.04% ± 0.01%. Statistical validation confirmed the consistency of these results across multiple test scenarios. These findings highlight the potential of ensemble learning in satellite-based remote sensing applications, as compared to other top models worldwide. Our approach leverages the complementary strengths of multiple specialized models to deliver a more nuanced and accurate land cover classification system. This advancement significantly enhances environmental monitoring and urban planning capabilities while establishing a foundation for future innovations in remote sensing analysis. The results validate not only the effectiveness of ensemble learning in remote sensing but also demonstrate how thoughtful integration of multiple models can push the boundaries of environmental observation techniques.

In recent years, great progress has been made in the field of forecasting meteorological variables. Recently, deep learning architectures have made a major breakthrough in forecasting the daily average temperature over a ten-day horizon. However, advances in forecasting events related to the maximum temperature over short horizons remain a challenge for the community. A problem that is even more complex consists in making predictions of the maximum daily temperatures in the short, medium, and long term. In this work, we focus on forecasting events related to the maximum daily temperature over medium-term periods (90 days). Therefore, instead of addressing the problem from a meteorological point of view, this article tackles it from a climatological point of view. Due to the complexity of this problem, a common approach is to frame the study as a temporal classification problem with the classes: maximum temperature "above normal", "normal" or "below normal". From a practical point of view, we created a large historical dataset (from 1981 to 2018) collecting information from weather stations located in South America. In addition, we also integrated exogenous information from the Pacific, Atlantic, and Indian Ocean basins. We applied the AutoGluonTS platform to solve the above-mentioned problem. This AutoML tool shows competitive forecasting performance with respect to large operational platforms dedicated to tackling this climatological problem; but with a "relatively" low computational cost in terms of time and resources.