Context Coastal saline regions of the Ganges Delta often contain excess soil water after the monsoon season, delaying Rabi crop establishment. Subsequent unseasonal heavy rainfall further exacerbates waterlogging, reducing crop productivity. Although surface drainage can alleviate waterlogging, the optimum surface drainage depth for sunflower remains unclear. Objective This study aimed to identify the optimal surface drainage depth for sunflower cultivation and assess long–term (LT, 1975–2024) yield variability under different seasonal rainfall conditions. Methods Field trials in Dacope, Khulna, tested three surface drainage depths (no drain, 15 cm, and 30 cm, the depth of the open channel) from 2022 to 2024. Data on crop, soil water and soil salinity were used to calibrate the APSIM crop simulation model for local conditions. Model performance was evaluated using R2, Root Mean Square Error (RMSE), normalised RMSE, and Student’s t–test. Subsequent scenario analyses then compared LT yields across three dry season rainfall ranges: low (0–100 mm), medium (100–300 mm), and high (300–637 mm) based on 50 years of records. Results Model calibration and parameter adjustment helped address previously observed shortcomings of the APSIM-Sunflower module and enhanced its simulation accuracy for the variables of interest. High R2 values (0.79–0.98) and acceptable RMSE values between observed and simulated data (7.4 & 6.4 day, 106 & 153 kg/ha, 543 & 549 kg/ha, 0.026 & 0.036 cm3/cm3, and 46 & 119 mg/kg) were achieved for both calibration and validation across phenology, yield, biomass, soil water, and chloride content (CL), respectively. From LT yield prediction, a 30 cm drainage depth was optimal for all seasonal rainfall categories, resulting in yields 1.1–1.8 times higher than those with 15 cm or no drainage. The LT yields showed a significant (p < 0.001) positive correlation with seasonal rainfall and solute potential, and a significant (p < 0.001) negative correlation with CL. All results were within the range of experimental error, confirming the model's reliability. Conclusion The APSIM model demonstrated strong performance in simulating daily phenology, yield, biomass, soil water, salinity, and LT yield for surface drainage depths in coastal soils, and 30 cm was shown to be the best depth. Implications With further improvement by adding a surface drainage module, APSIM could serve as a valuable decision–support tool for researchers and policymakers for managing LT risks in dry season cropping and increasing smallholders’ farm income through cropping systems intensification in the saline coastal zone of the Ganges Delta and similar environments.
Accurate land cover classification from satellite imagery is crucial in environmental monitoring and sustainable resource management. However, it remains challenging due to the complexity of natural landscapes, the visual similarity between classes, and the significant class imbalance in the available datasets. To address these issues, we propose a dual encoder architecture that independently extracts modality-specific features from optical and Synthetic Aperture Radar (SAR) imagery, which are then fused using a cross-modality attention-fusion module named Cross-modality Land cover segmentation with Attention and Imbalance-aware Reasoning-Enhanced Explanations (CLAIRE). This fusion mechanism highlights complementary spatial and textural features, enabling the network to better capture detailed and diverse land cover patterns. We incorporate a hybrid loss function that utilizes Weighted Focal Loss and Tversky Loss named RIFT (Rare-Instance Focal-Tversky) to address class imbalance and improve segmentation performance across underrepresented categories. Our model achieves competitive performance across multiple benchmarks: a mean Intersection over Union (mIoU) of 56.02% and Overall Accuracy (OA) of 84.56% on the WHU-OPT-SAR dataset; strong generalization with a mIoU of 59.89% and OA of 73.92% on the OpenEarthMap-SAR dataset; and remarkable robustness under cloud-obstructed conditions, achieving an mIoU of 86.86% and OA of 94.58% on the PIE-RGB-SAR dataset. Additionally, we introduce a metric-driven reasoning module generated by a Small Language Model (Phi-3), which generates sample-specific justifications for model predictions, thereby enhancing transparency and interpretability.
This critical review examines the definition, impacts, underlying causes, and potential solutions for waste mountains in developing countries. The analysis highlights three major knowledge gaps: inconsistent terminology; the absence of reliable national inventories documenting the number, location, size, and composition of waste mountains; and limited information on sustainable waste-to-energy solutions. The findings reveal that waste mountains generate interconnected environmental, public health, social, and economic consequences, while the full economic costs of continued dumping remain largely unquantified. Rather than being primarily a technological problem, their persistence stems from systemic challenges, including weak governance, fragmented policies, insufficient financing, inadequate waste management infrastructure, and poor data availability. For existing waste mountains, waste mining combined with resource recovery and thermochemical waste-to-energy technologies offers a promising remediation pathway. However, the suitability of individual technologies depends on local conditions, including waste characteristics, moisture content, contamination levels, infrastructure capacity, financial feasibility, and environmental safeguards. Among available options, incineration remains the most widely established technology, whereas pyrolysis presents opportunities for resource recovery and circular economy outcomes through the production of valuable co-products. Nevertheless, its application to heterogeneous legacy municipal waste remains insufficiently proven at scale. The review concludes that effective management of waste mountains will require the development of comprehensive national inventories, risk-based prioritisation of remediation efforts, robust valuation of the costs of inaction, inclusive governance arrangements, and comparative techno-economic and lifecycle assessments to guide the selection of appropriate strategies.
Background Parental beliefs about infant sleep shape caregiving practices, infant development, and parental wellbeing. The Views oN Infant Sleep (VNIS) questionnaire was designed to assess these beliefs across three domains: Closeness, Independence, and Night-Waking. This study aimed to validate the questionnaire for the first time in an Australian context, in a sample of 111 pregnant women. Methods Women completed an anonymous online questionnaire containing the 12-item VNIS. Confirmatory factor analysis was conducted in MPlus, using a weighted least squares estimator to evaluate the proposed three-factor structure. Results Findings provided partial support for the three-factor model. Incremental and residual fit indices were strong (comparative fit index = 0.97; Tucker-Lewis index = 0.96; weighted root mean square residual = 0.76) and the chi-square to degrees-of-freedom ratio indicated acceptable fit (1.75). Global fit was borderline (root mean square error of approximation = 0.08, 90% confidence interval = 0.05–0.11). All items loaded significantly onto their intended factors, however item performance varied. Independence items showed consistently strong loadings, while Closeness items ranged from moderate to strong with one weaker indicator. Night-Waking items were more variable, including one item below acceptable thresholds. Inter-factor correlations were strong (r = 0.58–0.71), indicating substantial coherence and limited distinctiveness between components. Conclusion The VNIS questionnaire captures multidimensional parental beliefs about infant sleep but requires refinement to strengthen reliability and discriminant validity. Further cross-cultural and longitudinal research is needed to support its broader applicability in diverse clinical and research contexts including to fathers and non-birthing partners.
Machine unlearning is essential for complying with privacy regulations such as GDPR and CCPA, which require the removal of user data upon request. This paper proposes an efficient approximate machine unlearning framework for deep learning models that removes specific data points or classes without full retraining, enabling real-time compliance for large-scale web applications. Unlike traditional methods such as naive data deletion or computationally prohibitive retraining, our approach balances precision and efficiency through a novel Hessian-guided gradient unlearning technique. This method combines gradient-based updates for speed with Hessian refinements for accuracy, achieving practical data removal while preserving model utility. A post-processing step further reduces residual data traces via masking and inpainting, enhancing compliance for sensitive web-based systems. Evaluations on CIFAR-10, ImageNet-Subset, and web-specific datasets (Twitter100k, Amazon Products) demonstrate that our method (1) closely approximates the performance of retrained models in downstream tasks, (2) reduces unlearning time by 43% compared to Hessian-only baselines, and (3) maintains 89.8% retained accuracy while lowering forgotten-class accuracy to 48.4%. These results position our solution as a practical alternative for web platforms that prioritize both regulatory compliance and operational efficiency. We further validate scalability through an additional experiment on ResNet-50, confirming that the proposed framework generalizes to deeper architectures. we further validate the method through a SOTA comparison against recent methods (SalUn, SCRUB), distribution-level analysis via Maximum Mean Discrepancy (MMD), and a systematic hyperparameter sensitivity study. While our approach achieves significant reductions in the influence of forgotten data and closely approximates retraining in practice, it does not provide certified guarantees of exact data removal.
