The 2023–2024 Amazon drought, driven by the concurrence of an El Niño and positive sea surface temperature anomalies in the tropical North Atlantic, resulted in historically low streamflow and widespread social and ecosystem impacts. To evaluate terrestrial subseasonal-to-seasonal (S2S) predictability during this event and assess its cascading impacts on regional wildfire potential, we develop an S2S Amazon Land Data Assimilation System (Amazon-LDAS). By integrating dynamical meteorological forecasts into a high-resolution land surface and routing model, we generated 5 km daily hydrometeorological forecasts at up to a 6-month lead to evaluate the propagation of predictability from atmospheric forcings to land surface states. Results indicate that dynamical forecasts initialized at up to 3-month leads captured the drought's spatial footprint, although 1-month lead forecasts provided overall best detection. Root zone soil moisture (RZSM) forecasts demonstrated significantly high probabilistic skill with a 74% improvement over persistence forecasts, highlighting the improvement in skill due to hydrologic memory in the basin. Streamflow forecasts also showed high skill throughout the drought event at all forecast initializations. However, integrated autocorrelation analysis indicated that this statistical skill in streamflow is partially influenced by internal hydrologic memory biases. This means that the modelling framework exhibited routing sluggishness, retaining dry anomalies within the channel network longer than the retrospective baseline, which inflated streamflow forecast metrics. Additionally, utilizing S2S RZSM forecasts as a proxy for fire risk doubled the probability of detection (POD) of fires in the arc of deforestation compared to concurrent meteorological metrics. Hydrologic memory in land surface and routing allows improvement of terrestrial skill in hydrometeorological forecasting at long lead times (up to 3 months ahead) and improve the prediction of extreme event impacts on the Amazon ecosystem when accounting for intra-regional memory biases. Building on these findings, we provide an operational hydrometeorological outlook for October 2026 in the context of currently developing El Niño to demonstrate the system's capacity to support early warning.