Rapid urban population growth in Bandung City has led to the development of slums due to inadequate housing facilities and urban planning. However, it remains unclear how these slums are distributed and evolve spatially and temporally. Therefore, it is necessary to map their distribution and trends effectively. This study aimed to classify slum areas in Bandung City using a machine learning-based local knowledge approach; this classification exercise contributes towards Sustainable Development Goal 11 related to sustainable cities and communities. The methods included settlement and commercial/industrial classification from 2021 SPOT-6 satellite data by the Random Forest classifier. A knowledge-based classifier was used to derive slum and non-slum settlements from the settlement and commercial/industrial classification, as well as railway, river, and road buffering. Our findings indicate that these methods achieved an overall accuracy of 82%. The producer’s accuracy for slum areas was 70%, while the associated user’s accuracy was 92%. Meanwhile, the Kappa coefficient was 0.63. These findings suggest that local knowledge could be a potent option in the machine learning algorithm.

This paper presents a comparative analysis of four machine learning algorithms-Random Forest (RF), Gradient Boosting (GTB), Support Vector Machine (SVM), and Classification and Regression Trees (CART)-for mapping drought-driven forest and land fires in Kalimantan. We utilized multi-source Geo-Artificial Intelligence data, including satellite remote sensing products and spatial data, incorporating physical, environmental, climatological, biological, and socio-economic (anthropogenic) factors as inputs. The EI Niño year of 2019 was selected as the study period for data analysis. The Global Annual Burned Area Maps (GABAM) product was refined using nighttime active fire data, which represent actual fire occurrences and served as our reference map. We applied the Graph Regularization fusion method to all the classified images to enhance mapping accuracy. Our results indicate that RF and GTB significantly outperformed SVM and CART in terms of precision, recall, Kappa, and overall accuracy, with both RF and GTB achieving an overall accuracy of 97%. The fused image yielded results similar to those of the RF and GTB images. These findings offer valuable insights for developing a national early warning system for drought-related forest and land fires, using Geo-AI data to support forest management and the goal of achieving net-zero emissions.

The disposal of waste materials such as plastic, foam, cloth, paper, rubber, and wood in river areas significantly impacts local environments, affecting regions several kilometers away and even reaching the open sea. This study utilized data from the Sentinel-1 C-SAR spanning 2017–2024 to detect water bodies. Additionally, an adaptive decision tree algorithm method with the Otsu threshold was applied to the Sentinel-2 optical data of March 19, 2024, to identify aquatic vegetation and floating waste (non-vegetative objects) within the water body. The study focused on the inlet of the Saguling Reservoir, where the Citarum River flows near the Bridge of Babakan Sapan (BBS) in West Java Province. This location is known for frequent waste accumulation. The methodology was then employed to monitor accumulated waste along the Citarum River on June 17, 22, and July 12, 2024. The model demonstrated an overall accuracy of 86% and a Kappa coefficient of 79% in distinguishing waste, water, and vegetation, with most errors occurring in the vegetation and waste classes. Additionally, the model's effectiveness in detecting waste in narrow water bodies with densely populated buildings on both sides, as observed in a narrow river along slum areas in Cirebon City, was comparatively lower.

Riverine plastic waste is difficult to detect and map accurately because debris ranges from small items to large floating clusters, and tropical rivers present challenging conditions, such as murky water, floating vegetation, and variable lighting. This study develops and benchmarks a deep learning method for pixel-level, multi-scale mapping of riverine waste from five-band UAV multispectral imagery. We deployed a drone equipped with a five-band multispectral sensor over the Brantas River in Surabaya, East Java, Indonesia, and introduce the Multi-Scale Hierarchical Attention Ensemble (MHAE), which combines three backbone networks across three image resolutions through learned scale- and backbone-attention weighting. MHAE was evaluated against 12 CNN-, transformer-, state-space-, and traditional-machine-learning-based baselines (including Random Forest, U-Net, UNet++, and DeepLabV3+) on the accompanying BrantasRiverWaste-UAV dataset (882 image tiles from a single-site, single-season orthomosaic covering approximately 0.54 km² of river surface, labelled as water, land, organic waste, or inorganic waste), with pairwise comparisons assessed using Wilcoxon signed-rank tests with Bonferroni correction. MHAE achieved the highest pixel-level waste detection rate among the 12 evaluated models (86.76%), with a mean intersection-over-union of 78.33% (third-highest, behind UNet++ and U-Net). This work provides an initial, single-site benchmark and reference architecture for near-real-time riverine waste monitoring, and introduces the BrantasRiverWaste-UAV as, to our knowledge, one of the first five-band multispectral UAV datasets for tropical riverine waste mapping. Balanced sampling and multi-scale attention fusion can substantially improve waste-pixel detection under severe class imbalance; because the benchmark derives from a single site and season, future work should extend evaluation across additional seasons and river systems before the approach is generalised operationally.