Maharaja Bir Bikram College

Urban flooding poses a growing threat to rapidly expanding cities in the Global South, with Agartala, the capital of Tripura in Northeast India, experiencing recurrent inundation due to its low-lying topography, intense monsoon rainfall, and deteriorating drainage infrastructure. This paper provides a systematic review of satellite remote sensing and artificial intelligence (AI) models for urban flood mitigation within the context of Agartala Smart City. The study synthesizes findings from peer-reviewed literature, government reports, and remote sensing analyses conducted between 2015 and 2025. Our analysis reveals that approximately 9.2% of Agartala's total geographical area falls within high to very high flood risk zones, with seven out of 35 wards identified as critically vulnerable. Sentinel-1 Synthetic Aperture Radar (SAR) imagery from flood events demonstrates that 374.5 hectares were inundated during a single extreme rainfall event in August 2024. AI-based flood susceptibility models, including Random Forest (RF), Support Vector Machines (SVM), and Convolutional Neural Networks (CNN), achieve classification accuracies of 85-92% for urban flood mapping. Emerging Geo-Foundational Models (GFMs) show promise for transfer learning applications where labeled training data remains scarce. The paper proposes an integrated framework combining real-time SAR monitoring, AI-based predictive modelling, and smart city infrastructure (sensors, command centers, citizen reporting) to enhance flood resilience. Key policy recommendations include updating drainage master plans based on LiDAR-derived Digital Elevation Models (DEMs), deploying IoT-based water level sensors across 50 identified hotspots, and establishing a public-facing flood early warning dashboard. This review contributes to the growing literature on climate-adaptive smart cities in data-sparse environments. Keywords: Urban flood mitigation, Synthetic Aperture Radar (SAR), artificial intelligence, Agartala Smart City, flood susceptibility mapping, resilience planning, geospatial analysis

Wetlands are among the most productive yet vulnerable ecosystems globally, providing essential services including carbon sequestration, water purification, flood regulation, and biodiversity habitat. Rudrasagar Lake, a Ramsar site (No. 1572) in Tripura, India, exemplifies the challenges facing these ecosystems—experiencing rapid degradation due to anthropogenic pressures, agricultural runoff, siltation, and invasive species proliferation. This review paper synthesizes current advances in deep learning (DL) approaches for wetland environmental monitoring using remote sensing data, with special reference to Rudrasagar. We critically evaluate state-of-the-art DL architectures including U‑Net, U‑Net++, DeepLabV3+, and Swamp‑AI for wetland classification, change detection, and water quality parameter estimation. Our analysis reveals that integrating multi‑source satellite data (Sentinel‑1 SAR, Sentinel‑2 optical, and PALSAR‑2 L‑band) with uncertainty‑aware deep learning frameworks achieves superior performance (over 90% overall accuracy) compared to traditional methods. However, significant gaps exist in applying these approaches to small, monsoon‑influenced wetlands like Rudrasagar. We propose three innovative monitoring models: (1) a Hybrid Spectral‑Temporal Deep Learning Framework for wetland health assessment, (2) a Multi‑Task Attention Network for simultaneous water quality and biodiversity monitoring, and (3) a Weakly Supervised Change Detection System for cost‑effective wetland inventory updates. These models integrate physical limnological parameters with satellite‑derived indices to address the unique challenges of tropical Ramsar wetlands. The review concludes with specific recommendations for implementing DL‑based monitoring at Rudrasagar and similar wetland ecosystems in South Asia. Keywords: Deep Learning, Remote Sensing, Wetland Monitoring, Rudrasagar Lake, Ramsar Site, Sentinel‑2, U‑Net, Change Detection, Water Quality, Carbon Sequestration

The rapid proliferation of metaphor-based optimization algorithms has faced criticism from researchers regarding their novelty and contribution. Many of these pseudo-novel metaheuristics suffer from performance inefficiencies, biased verification systems, and excessive similarities in their modelling as they imitate the foraging behaviors of animals and other living beings. This study introduces a population-based and metaphor-free optimization algorithm using the well-known and efficient Lagrange interpolation formula to address these concerns. The proposed algorithm, called Lagrange interpolation-based optimization (LIBO), is designed for numerical optimization and solving real-world optimization problems. The second-order Lagrange interpolation formula is considered to construct the algorithm where the polynomial passes through three given random points, thereby demonstrating a parabolic structure. While updating the algorithm, every trial solution considers three randomly chosen solutions and obtains the minimal point. To strike a suitable balance between exploration and exploitation and avoid local optima trapping, the proposed LIBO algorithm incorporates specific measures in its search mechanism. Its effectiveness is evaluated by IEEE CEC 2017 and IEEE CEC 2019 benchmark functions, three practical engineering problems, and the highly non-linear multiple gravity assist spacecraft trajectory problem. Comparative analyses were conducted, and it was found that in CEC 2017 functions, the proposed LIBO is better on 100

Evaluating the semester examination performance of Science, Technology, Engineering, and Mathematics (STEM) students involves multiple qualitative and quantitative factors such as academic preparation, learning environment, cognitive ability, and institutional support. These factors often contain uncertainty and vagueness due to subjective evaluations. To address this issue, this study proposes a Cosine Similarity Measure (CSM) based Multi-Criteria Decision-Making (MCDM) approach under the Multi-Neutrosophic Set (MNS) environment. The MNS framework represents truth, indeterminacy, and falsity membership values with multiplicities, enabling effective modeling of uncertain evaluation data. A new cosine similarity measure between MNSs is developed and its mathematical properties are examined. Based on this measure, an MCDM algorithm is designed to evaluate and rank the influential factors affecting STEM students’ semester examination performance. A numerical case study demonstrates the applicability of the proposed method. The results show that the model effectively handles uncertainty and provides stable and reliable rankings of the influential academic factors, offering a robust decision-support tool for academic performance evaluation under uncertain environments.

The script debate for Kokborok, the Indigenous language of the Tiprasa people in Tripura, India is more than a linguistic issue - it is a struggle for identity, cultural preservation, and self-determination. While the state enforces the Bengali script, widespread support exists for the Roman script, seen as both linguistically suitable and a symbol of Indigenous resistance. This article examines the historical roots of script imposition, its socio-political implications, and the ongoing movement challenging linguistic hegemony. By analysing the role of political parties, grassroots activism, and educational impact, this study underscores how script choice becomes a battleground for cultural assertion and decolonial resistance in Northeast India.