it_team4_FOREst 7 July 2025July 2–3, 2025 | Salzburg, Austria
The It4Forest team members Prof. Dr. Jan-Peter Mund, Gulam Mohiuddin, Dr. Kevin Beiler and two graduate students from HNEE participated at the AGIT 2025 Conference for Geoinformatics, held from July 2–3, 2025, in Salzburg, Austria. AGIT is a leading platform in for geospatial innovation, applied science, and networking among researchers, industry professionals, and decision-makers. This year’s event brought together experts in geoinformation and spatial analysis to address pressing challenges in urban sustainability, digital transformation, and climate resilience. It4Forest team contributed with four oral presentations, showcasing advanced remote sensing and machine learning approaches for forest and urban environmental analysis.
The first presentation was on maximum land surface temperature in different forest types in the Barnim District, Brandenburg, Germany. The study was conducted by Robin Simon Stephan, Prof. Dr. Jan-Peter Mund, Prof. Dr. Peter Spathelf and Gulam Mohiuddin. The study investigates the variation of maximum land surface temperature (LSTmax) across eight distinct forest types in the Barnim district of Brandenburg, Germany, using a 30-year Landsat time series (1982–2023) processed via Google Earth Engine. By analyzing 32 representative plots featuring dominant tree species such as Pinus sylvestris, Fagus sylvatica, and Quercus species, the research highlights how forest composition influences thermal behavior. Results show that pure pine stands consistently exhibit higher LSTmax values than mixed or deciduous-dominated forests, emphasizing the cooling benefits of broadleaf species due to canopy shading and higher transpiration. The findings support regional forest conversion strategies aimed at enhancing microclimate regulation and resilience to climate change.
The second presentation was on assessing the loss of urban waterbodies using a multi-sensor Sentinel and machine learning approach. The study was conducted by HNEE graduate student Ali Dia, Gulam Mohiuddin and Prof. Dr. Jan-Peter Mund. Using Sentinel-1 and Sentinel-2 satellite data, the researchers used Support Vector Machine (SVM) classification to detect and quantify water body changes between 2016 and 2023. Their results revealed a 28 sq km reduction in urban water area, raising important implications for urban heat mitigation and sustainable planning.
The third presentation was a comparative analysis of machine learning based land use and land cover classification with an attempt to enhance the class based accuracy. The study was conducted by HNE graduate student Kazi Jahidur Rahaman, Gulam Mohiuddin and Prof. Dr. Jan-Peter Mund. Using Sentinel-2 data from Phnom Penh and applying Random Forest, Support Vector Machine, and Artificial Neural Network algorithms, the team demonstrated how performance varies across land cover classes. Their proposed ensemble framework improves both class-wise and overall accuracy by selecting the most accurate algorithm for each specific land cover type. This approach achieved an enhanced F1 Score of 90%, outperforming any individual model. The results offer practical applications in LULC mapping and urban change detection, particularly in complex urban environments.
All presentations received positive feedback for their methodological rigor, practical relevance, and strong visual analysis.
