SWaBME: A PARSIMONIOUS LARGE-SCALE MODEL TO SIMULATE WATER BALANCE COMPONENTS OF TYPICAL LAND COVER TYPES IN BRAZIL

Accurately identifying the interactions between large-scale land cover and regional climate in the water balance components is crucial for our understanding of how the transformation of native vegetation into agricultural areas impacts the water cycle. Yet the available regional models to access water balance components are often too complex and typically highly dependent on a large number of inputs and parameters. This inadvertently leads to relatively high uncertainty in the model components and their interactions, undermining their use for identifying controlling factor and mechanisms associated with key hydrological processes. In this work, we address the need for a parsimonious model by introducing the Soil Water Balance Modelling Environment (SWaBME). SWaBME is a novel parsimonious hydrological model used to assess the water balance partitioning of typical land cover types in Brazil, a country that is constantly affected by high rates of deforestation and agricultural expansion. The SWaBME model uses a Penman-Monteith formulation to estimate, separately and explicitly, the evapotranspiration (ET) in the three main components (bare soil evaporation, transpiration, and evaporation from canopy interception), which allow it to distinguish the effects of climate and land cover on the ET. The SWaBME model requires only five parameters to be prescribed a priori, and also contains a set of parameters which are directly provided by the recent development of global georeferenced data products. SWaBME is calibrated by following an alternative approach which evaluates hundreds of thousands of randomly generated parameter sets against observed monthly evapotranspiration and soil moisture data (when available) that are ultimately tested at a pre-defined set of soft rules to ensure model consistency. The model calibration were done individually at 10 flux sites in Brazil,  but we also investigate whether such preferred parameter combinations produce plausible model performances at the country main land-cover and land-use classes: forests, cerrado/woodlands, pasture/grasslands, and soybean and sugarcane crops. From all the parameters combinations, the model was able to satisfactorily retain about 70 to 90% of the sets for forests and cropland biomes, but appears to constrain much more strongly for pasture/grasslands and cerrado biomes with respectively 30% and 1% of the set retained. Most of the introduced soft rules have low to moderate constraining power, and we found that differences in the calibrated parameters for each biome are more pronounced only when the prior information from literature review was used to constrain specific parameters ranges. The performance with the selected parameters showed Root Mean Squared Error of about 20 to 36 mm/month [RR1] at forest and cropland biomes, 23 to 26 mm/month at the cerrado/woodland and 30 to 36 mm/month at pasture/grasslands; ranking slight better when compared to the more complex (in terms of structure and number of parameters) NOAH/GLDAS model with a RMSE ranging from 30 to 60 mm/month. Overall, SWaBME is a parsimonious model aimed at large-scale application of water balance assessment focusing on assessing the impacts of climate and land-use/land-cover changes primarily in Brazil. However, the structure and approach used here can be widely transferred to other regions of the world.