ACCUMULATED RAINFALL AND POST-RAINFALL RECOVERY FOR NOWCASTING UNSUITABLE BATHING CONDITIONS

Código

V-END_II-EBHE0192

Autores

Tainah Frota Carvalho, IRAN EDUARDO LIMA NETO, Newton Celio Becker de Moura, Francisca Dalila Menezes Vasconcelos, Matheus Castelo

Tema

18 - Aspectos qualitativos de água em situações de desastre

Resumo

Urban beaches face sanitary risks after rainfall, when runoff and inadequate sanitation transport fecal contamination to recreational waters. In Fortaleza, Brazil, weekly regulatory bulletins may not capture rapid changes in bathing water conditions. This study develops a nowcasting framework to estimate the risk of unsuitable bathing conditions using routine water-quality and rainfall data. Its contribution is to integrate antecedent rainfall, post-rainfall recovery, and spatial monitoring information into a predictive layer for Fortaleza?s Climate Risk Observatory. The dataset included 33 monitoring points along 34 km of coastline, organized longitudinally by monitoring week. SEMACE bathing-water classifications were coded as a binary response and combined with weighted FUNCEME rainfall estimates for 24 h, 48 h, 72 h, and 7-day antecedent windows. Critical rainfall events and recovery periods were assessed, and an XGBoost classifier was evaluated with a temporal split and F2 threshold prioritizing detection of unsuitable conditions. Among 15,706 valid observations, the association between rainfall and unsuitable conditions increased with longer windows, reaching Pearson?s r = 0.23 and Spearman?s ? = 0.28 for seven-day accumulation. Suitability declined from 67.1% under the lowest 7-day rainfall conditions to 34.0% under the highest. After critical events of at least 10 mm in seven days, unsuitable conditions decreased from 61.9% on the event day to 29.3% after 4?7 days. The final model achieved recall of 0.951, ROC-AUC of 0.857, and 26 false negatives. The framework supports preventive health alerts and future territorial-vulnerability analyses under intense rainfall for local risk management.