Researcher in reliable spatiotemporal artificial intelligence and Associate Lecturer at the University of Alicante

I study how artificial intelligence systems can produce dependable forecasts when sensor networks, observational data and deployment conditions are imperfect. My research connects spatiotemporal machine learning, graph neural networks, environmental forecasting, sensor data quality, distribution shift and robust model evaluation.

I am a member of the Network Data Analysis and Visualisation research group (ANVIDA) in the Department of Computer Science and Artificial Intelligence at the University of Alicante.

I publish scientific work primarily as Marc Semper.

Website · Google Scholar · ORCID · Scopus · LinkedIn

• Reliable and robust spatiotemporal artificial intelligence
• Graph neural networks for distributed sensor systems
• Environmental and meteorological forecasting
• Distribution shift and model transferability
• Model selection under observational uncertainty
• Quality control for environmental station networks
• Missing data, sensor failures and spatial misalignment
• Extreme precipitation and high-impact weather
• Satellite, reanalysis and surface-observation comparison
• Reproducible computational research

• Decision-Support Protocol for Spatiotemporal Forecasting
  Experimental code for evaluating whether the ranking of trained forecasting models remains stable when the evaluation reference is perturbed by lag, missingness, spatial shifts or support loss.

• PhyK-TAS
  Experimental decision-support pipeline for assessing transferability risk in spatiotemporal precipitation forecasters by combining physical regime descriptors, distribution-shift diagnostics and degradation-inference models.

• Controlled coarsening of hourly precipitation extremes
  Reproducibility package for separating temporal-support loss, phase alignment, reporting cadence and rainfall-regime effects in hourly precipitation products.

• GPM IMERG correctability benchmark
  Minimal reproducibility package for testing the out-of-sample statistical correctability of sub-daily GPM IMERG precipitation estimates against dense surface observations.

• Operational limits of GPM IMERG
  Code and derived results for evaluating representativeness, attenuation and pixel-scale displacement in sub-daily satellite precipitation extremes.

• Auditable quality control for environmental station networks
  Reproducibility package for translating quality-control evidence into auditable actions, including deletion, quarantine, flagging and maintenance escalation.

• Global forecasting of aerosol optical depth
  Code associated with graph-based global aerosol forecasting using atmospheric composition and meteorological data.

• Inter-city air-quality forecasting
  Spatiotemporal graph-learning experiments for forecasting air quality across a distributed monitoring network in Spain.

Some projects use licensed or restricted observational archives. In those cases, the repositories provide code, schemas, derived results and reproduction instructions without redistributing the original raw data.

1. A Practical Decision-Support System for Robust Post-Training Model Selection in Spatiotemporal Forecasting
   Marc Semper, Manuel Curado, Jose F. Vicent and Leandro Tortosa.
   Knowledge-Based Systems, article 116673.
   DOI: 10.1016/j.knosys.2026.116673

2. Multi-Dataset Training for Improved Accuracy in Spatio-Temporal Problems: An Explainable Analysis
   Javier García-Sigüenza, Alberto Real-Fernández, Faraón Llorens-Largo, Rafael Molina-Carmona and Marc Semper.
   Mathematics, 14(5), article 908.
   DOI: 10.3390/math14050908

3. Global Forecasting of Aerosol Optical Depth through a Deep Learning Spatiotemporal Modeling
   Marc Semper, Manuel Curado and Jose F. Vicent.
   International Journal of Environmental Science and Technology, 23(1), article 69.
   DOI: 10.1007/s13762-025-06905-4

4. Spatio-Temporal Graph Neural Network for Inter-City Air Quality Forecasting
   Jose F. Vicent, Manuel Curado and Marc Semper.
   International Journal of Environmental Science and Technology, 23(1), article 63.
   DOI: 10.1007/s13762-025-06850-2

5. Noise Pollution Prediction in a Densely Populated City Using a Spatio-Temporal Deep Learning Approach
   Marc Semper, Manuel Curado, Jose Luis Oliver and Jose F. Vicent.
   Applied Sciences, 15(10), article 5576.
   DOI: 10.3390/app15105576

6. Global Forecasting of Carbon Concentration through a Deep Learning Spatiotemporal Modeling
   Marc Semper, Manuel Curado and Jose F. Vicent.
   Journal of Environmental Management, 371, article 122922.
   DOI: 10.1016/j.jenvman.2024.122922

For the complete and updated publication record, see marcsemperlloret.com/publications.

Modelado espacio-temporal con redes neuronales para la predicción de fenómenos ambientales
Spatiotemporal modelling with neural networks for forecasting environmental phenomena
University of Alicante, 2025.

The thesis investigates neural architectures for modelling environmental phenomena across interconnected spatial locations and multiple temporal scales.

Thesis overview

My work is guided by several methodological principles:

1. Strict validation — temporal and spatial leakage must be explicitly prevented.
2. Realistic evaluation — models should be tested under missing data, sensor failures, measurement shifts and changing deployment conditions.
3. Decision robustness — improved average accuracy is insufficient when the selected model changes under plausible observational uncertainty.
4. Strong baselines — complex models should be compared against transparent and competitive alternatives.
5. Uncertainty awareness — inconclusive or non-identifiable outcomes are valid scientific results.
6. Reproducibility — data provenance, preprocessing decisions, configurations and evaluation protocols should be auditable.
7. Operational relevance — statistical improvements should be interpreted according to their practical consequences.

• Artificial intelligence and machine learning
• Python and scientific computing
• Graph-based learning
• Computer networks and network automation
• Cloud computing
• Data engineering and reproducible experimentation
• Applied projects using meteorological and environmental sensor data

I am interested in collaborations involving reliable artificial intelligence, graph neural networks, environmental observation, meteorology, climatology, high-impact weather, Earth observation, sensor networks, distribution shift and reproducible scientific software.

Research enquiries can be sent to marc.semper@ua.es.

• Personal website
• University of Alicante research profile
• University of Alicante CVNet profile
• Google Scholar
• ORCID
• Scopus Author ID 59404623900
• ResearchGate
• LinkedIn
• Lens

My publications may appear under the following name variants:

• Marc Semper
• Marc Semper Lloret
• M. Semper

ORCID: 0009-0002-5552-1420