AbstractStudy RegionThe study region is the Magdalena River basin in Colombia. The basin was divided into three distinct regions (Andean, Caribbean, and Pacific) and analyzed across different elevations.Study FocusThe study proposes a Spatiotemporal Non-Linear Dynamics Assessment (SNLDA) framework to compare ERA5-Land reanalysis data with in-situ rain gauge observations. It specifically examines the constraints imposed by nonlinear dynamical processes and their associated space-time complexities on the representation of precipitation, particularly in a tropical region. The SNLDA framework incorporates three main components: (i) standard performance metrics (e.g., correlations, RMSE, and dry spell duration), (ii) rainfall spatiotemporal objects (characterizing precipitation events through attributes such as volumes and start-end centroids), and (iii) non-linear dynamics complexity (reconstructing dynamical behavior from time series and evaluating attractors properties, including the Hurst and Lyapunov exponents). These elements were analyzed both individually and in combination. Daily ERA5-Land information (0.1°x0.1°) and in-situ rain gauge data comprising 558 stations from 1980 to 2020 were used, enriched by an Inverse Distance Weighting (IDW) interpolation (0.1°x0.1°) to facilitate comparison across spatial scales.New Hydrological Insights for the RegionOverall, ERA5-Land overestimates precipitation, producing shorter, more frequent events while poorly representing extreme wet and dry spells.Andean region: ERA5-Land overestimates rainfall, with largest errors at low elevations, driven by unresolved spatiotemporal object volumes displacements and nonlinear processes.Caribbean region: ERA5-Land shows the highest errors in nonlinear dynamics and extremes, despite lower annual bias and RMSE.Pacific region: ERA5-Land strongly overestimates precipitation volumes and RMSE, while nonlinear errors remain low; these biases are mainly driven by spatiotemporal objects displacement.

Sub-seasonal drought forecasting is crucial for early warning in estimating agricultural production and optimizing irrigation management, as forecasting skills are relatively weak during this period. Soil moisture exhibits stronger persistence compared to other climate system quantities, which makes it especially influential in shaping land-atmosphere feedback, thus supplying a unique insight into drought forecasting. Relying on the soil moisture memory, this study investigates the combination of multiple deep-learning modules for sub-seasonal drought indices hindcast in the Huai River basin of China, using long-term ERA5-Land soil moisture records with a noise-assisted data analysis tool. The inter-compared deep-learning models include a hybrid model and a committee machine framework. The results show that the performance of the committee machine framework can be improved with the help of series decomposition and the forecasting skill is not impaired with the lead time increases. Overall, this study highlights the potential of combining deep-learning models with soil moisture memory analysis to improve sub-seasonal drought forecasting.