This study compares the performance of three different neural network models to estimate daily streamflow in a watershed under a natural flow regime. Based on existing and public tools, different types of NN models were developed, namely, multi-layer perceptron, long short-term memory, and convolutional neural network. Precipitation was either considered an input variable on its own or combined with air temperature as another input variable. Different periods of accumulation, average, and/or delay were considered. The models’ structures were optimized and automatically showed that CNN performed best, reaching, for example, a Nash–Sutcliffe efficiency of 0.86 and a root mean square error of 4.2 m³ s⁻¹. This solution considers a 1D convolutional layer and a dense layer as the input and output layers, respectively. Between those layers, two 1D convolutional layers are considered. As input variables, the best performance was reached when the accumulated precipitation values were 1 to 5, and 10 days and delayed by 1 to 7 days.

The southern Iberian Peninsula is characterized by evergreen oak woodlands (locally known as montado), which constitute an important savanna-type agro-silvo-pastoral ecosystem. This ecosystem is facing a progressive decline for several reasons, with the foremost being overgrazing. Better management tools are necessary to accurately quantify the systems’ carrying capacity and the sustainable stocking rates that prevent land degradation. The purpose of this study was to determine whether the MOHID-Land model could adequately simulate soil water dynamics and pasture growth in the montado ecosystem. The study area was located in the Alentejo region of southern Portugal. The model successfully simulated soil water contents and aboveground biomass during the 2010–2011 and 2011–2012 growing seasons, producing acceptable errors of the estimates (0.015 ≤ RMSE ≤ 0.026 cm³ cm⁻³; 279 ≤ RMSE ≤ 1286.5 kg ha⁻¹), and relatively high modeling efficiencies (0.481 ≤ EF ≤ 0.882). The model was further used to simulate the same variables for a longer period (1979/2009 seasons), to account for the effect of climate variability on model estimates. Water balance and dry biomass estimates were found to be significantly different between rainfed and irrigated pastures, as well as between the ten driest and ten wettest seasons, with the model responding well to climate variability. The results showed the potential of using the MOHID-Land model for improving pasture management in the montado ecosystem.

Hydrological modeling at the catchment scale requires the upscaling of many input parameters for better characterizing landscape heterogeneity, including soil, land use and climate variability. In this sense, remote sensing is often considered as a practical solution. This study aimed to access the impact of assimilation of leaf area index (LAI) data derived from Landsat 8 imagery on MOHID-Land’s simulations of the soil water balance and maize state variables (LAI, canopy height, aboveground dry biomass and yield). Data assimilation impacts on final model results were first assessed by comparing distinct modeling approaches to measured data. Then, the uncertainty related to assimilated LAI values was quantified on final model results using a Monte Carlo method. While LAI assimilation improved MOHID-Land’s estimates of the soil water balance and simulations of crop state variables during early stages, it was never sufficient to overcome the absence of a local calibrated crop dataset. Final model estimates further showed great uncertainty for LAI assimilated values during earlier crop stages, decreasing then with season reaching its end. Thus, while model simulations can be improved using LAI data assimilation, additional data sources should be considered for complementing crop parameterization.

Hydrological models are increasingly used for studying watershed behavior and its response to past and future events. The main objective of this study was to conduct a sensitivity analysis of the MOHID-Land model and identify the most relevant parameters/processes influencing river flow generation. MOHID-Land is a complex, physically based, three-dimensional model used for catchment-scale applications. A reference simulation was implemented in the Ulla River watershed, northwestern Spain. The sensitivity analysis focused on sixteen parameters/processes influencing water dynamics at that scale. River flow generation was influenced by the resolution of the simulation grid, soil water infiltration, and crop evapotranspiration. Baseflow was affected by soil hydraulic properties, the depth of the soil profile, and the dimensions of the river cross-sections. Peak flows were mostly constrained by Manning’s coefficient in the river network, as well as the dimensions of the river cross-sections. The MOHID-Land model was then used to simulate daily streamflow during a 10-year period (2008−2017). Model simulations were compared against measured data at four hydrometric stations characterizing the natural flow regime of the Ulla River, resulting in coefficients of determination (R²) from 0.56 to 0.85; ratios of the standard deviation of the root mean square error to observation (RSR) between 0.4 and 0.67, and Nash and Sutcliffe model efficiency (NSE) values ranging from 0.55 to 0.84. The MOHID-Land model thus has the capacity to reproduce watershed behavior at a daily scale with reliable accuracy, constituting a powerful tool to improve water governance at the watershed scale.

The main goal of this study was to estimate inflows to the Maranhão reservoir, southern Portugal, using two distinct modeling approaches: a one-dimensional convolutional neural network (1D-CNN) model and a physically based model. The 1D-CNN was previously trained, validated, and tested in a sub-basin of the study area where observed streamflow values were available. The trained model was here subject to an improvement and applied to the entire watershed by replacing the forcing variables (accumulated and delayed precipitation) to make them correspond to the values of the entire watershed. The same way, the physically based MOHID-Land model was calibrated and validated for the same sub-basin, and the calibrated parameters were then applied to the entire watershed. Inflow values estimated by both models were validated considering a mass balance at the reservoir. The 1D-CNN model demonstrated a better performance in simulating daily values, peak flows, and the wet period. The MOHID-Land model showed a better performance in estimating streamflow values during dry periods and for a monthly analysis. Hence, results show the adequateness of both modeling solutions for integrating a decision support system aimed at supporting decision-makers in the management of water availability in an area subjected to increasing scarcity.