Future Growing Seasons: Bias Correction with SVR and QDM for Indonesian Temperature Projection under RCP 2.6 and RCP 8.5

Human activities have significantly contributed to greenhouse gas (GHG) emissions, escalating global temperatures and exacerbating extreme weather events, which pose serious agricultural threats by disrupting crop growth. Climate researchers employ Representative Concentration Pathways (RCPs) to forecast future GHG scenarios. Downscaling techniques have improved predictions of Growing Season Lengths (GSL) predictions and mean temperatures (Tmean), both of which are crucial for agricultural planning. This study evaluated Support Vector Regression (SVR) and Quantile Delta Mapping (QDM) for projecting Tmean and its impact on GSL under RCP 2.6 and RCP 8.5. Bias correction was applied to historical Tmean data using both methods, based on ERA5 data. SVR showed a lower Root Mean Square Error (RMSE) (0.6 vs 1.1) and a slightly higher correlation (0.6 vs. 0.5) than QDM. However, QDM was chosen for Tmean projection due to its superior data homogeneity and better alignment of standard deviation with observed values. Projections indicated a significant Tmean increase after 2026 under both RCPs, with Tmean under RCP 8.5 expected to exceed 30°C between 2050 and 2100, necessitating heat-resistant crop varieties. Greater increases in GSL under RCP 2.6 underscored the need for effective mitigation strategies. This study emphasizes adaptive farming practices and recommends integrating quantile-based and machine learning methods into future climate projections to enhance agricultural resilience.