The ED2 model calculates the water, carbon, and energy dynamics of the land surface. One of ED2’s distinguishing features is its ability to describe, in a physically consistent manner, the coupled water, carbon, and energy dynamics of heterogeneous landscapes. This ability to incorporate the land-use change and resulting heterogeneous patterns of land cover, and its impacts on the hydrological processes, ecology, and energy balance means that the model is ideally suited for investigating the combined impacts of rainfall trends and land-use change within the upper Paraná River basin.
The heterogeneous landscape of the region was represented by the mixtures of primary forest, secondary forest, and agricultural area. Four plant functional types (PFTs) were used: (1) early successional trees (fast growing, low density, water demanding), (2) mid-successional trees, (3) late successional trees (slow growing, high density, shade tolerant), and (4) C4 type grass. While a grid-cell shared a homogeneous meteorological forcing, evaporation fluxes were computed from the sub-divided, multi-layer canopy structure by size and age within a grid-cell. The stomatal conductance that regulates the plant’s transpiration was calculated by the water vapor deficit functions (Leuning 1995). A multi-layer soil model (Walko et al. 2000) was used to simulate surface runoff, soil moisture, and drainage. With the land-use disturbance rate updated annually, the model simulated the hydrologic changes associated with sub-grid heterogeneity of vegetation dynamics in the model time steps that span from seconds to 5 min (e.g., energy and water cycles) up to 15 min (e.g., photosynthesis and soil respiration). For further description of the ED2 model, please see Medvigy et al. (2009), and also Knox et al. (2015) and Swann et al. (2015) that used the ED2 model’s couple version to a regional climate model to investigate the past, current, and future land-use impact on hydroclimate of the surrounding regions in South America.
Study domain and experimental design
Model simulations were conducted for a region encompassing the five major sub-basins of the upper Paraná River: the Grande, Paranapanema, Tiete, Verde, and Paranaiba River basins (a total drainage area of approximately 787,000 km2) that feed the river discharge measured at the Itaipu Dam (Fig. 1). Note that the regional boundary of the basin of this study does not include the nearby Pantanal, the large-scale wetland area that consists of marsh.
Following the approach of Hurtt et al. (2006) and Albani et al. (2006), two representative ecosystem states, corresponding to the region’s land cover in 1970 and contemporary land cover patterns in 2008 (hereafter 1970LC and 2008LC, respectively), were developed by forcing the model with the historical land-use transition dataset of Hurtt et al. (2006). This dataset specifies the historical patterns of the land-use transitions between three land-use states: agricultural land, primary vegetation, and secondary vegetation. In 1970LC, the Paraná region consisted of 17% of agricultural land, 4% of primary forest, and 79% of secondary forest, while in 2008LC, the region was comprised of 75% of agricultural land, less than 1% of primary forest, and 24% of secondary forest (Fig. 2).
Five simulations (Table 1) were conducted at a spatial resolution 0.5° × 0.5° (approximately 50 km × 50 km) for 280 grid-cells spanning the upper Paraná River basin that contributes to the discharge at the Itaipu dam (green dot in Fig. 1). The pink dots in Fig. 1 correspond to the centroids of the grid-cells. The atmospheric CO2 concentration was kept constant at 378 ppm over the simulation period.
In the initial simulation (LCC-fullClim), the model was forced with both the observed 1970–2008 time series of meteorological forcing specified from the dataset by Sheffield et al. (2006) and the abovementioned time series of land-use transitions that occurred during this period. The contributions of changes in climate forcing and land-use change to the pattern and magnitude of discharge at Itaipu were estimated by conducting two additional pairs of simulations. In the first pair (1970LC-fullClim and 2008LC-fullClim), the model was forced with the observed time series of meteorological forcing for the period of 1970 to 2008 specified from the dataset by Sheffield et al. (2006) but with the land cover states corresponding to either the land-use pattern of the year 1970 (1970LC) or present-day land-use patterns (2008LC). In the second pair of simulations (LCC-70sClim and LCC-00sClim), the model was forced with the Hurtt et al. (2006) time series of land-use transitions that occurred during the period 1970–2007, with either climate specified for the 1970s (LCC-70sClim) or for the 2000s (LCC-00sClim). The 2008 climatology was applied for the 10th year of each decade. In LCC-70sClim and LCC-00sClim simulations, land conversion rates were applied to introduce disturbances originating in human activity specified from the Global Land Use dataset, for 1970–1999 from Hurtt et al. (2006). We applied the rate of year 1999 onward until 2008 because of little change in total agricultural area for 2000–2008 within the basin. The LCC-70sClim was forced by the climatological condition of the 1970s, and the LCC-00sClim by the climatology of the 2000s. All simulations generated 39-year multi-decadal monthly time series on the predicted changes in the water cycle and its major components (i.e., patterns of water discharge, storage, and evapotranspiration) via the coupled water, carbon, and energy dynamics across the basin.
Datasets and model evaluation metrics
The spatially integrated monthly discharge fluxes predicted by this scenario were then evaluated against the estimated natural flow at the Itaipu dam gauge station (25.43 S and 54.59 W). The natural flow was reconstructed from the daily discharge measurements at the Itaipu Dam, by taking into account the temporal gains and losses arising from reservoir operation and water withdrawals at the upstream of the gauge station (data available from the ANA (Brazilian National Water Agency) website (http://www2.ana.gov.br/). In addition, the model’s predictions of the temporal dynamics of total water storage (TWS) within the upper Paraná River basin were evaluated against the GRACE satellite measurements of this quantity (Rodell et al. 2004; Güntner 2008; Syed et al. 2008; Han et al. 2009). Because GRACE measures only relative differences in TWS, the respective mean values of TWS from model simulations were removed, and the anomalies were compared against the change in GRACE-estimated TWS over the period of 2002–2008. For large river basins such as the Amazon, the speed of the river flow must be factored into TWS storage calculations because the amount of laterally transported water within the basin is significant (Han et al. 2010). However, the relatively small size of the upper Paraná River basin and the monthly time scale of the analysis meant that river routing effects can be neglected, and thus the model’s predictions of TWS can be simply calculated based on the differences in the aggregate fluxes of water (i.e., precipitation minus evapotranspiration and runoff) within the river basin.
We applied the Mann-Kendall test with a confidence level of 95% for the trend analyses of the discharge at Itaipu Dam and the precipitation over the upper Paraná River basin. The goodness-of-fit of the ED2 model to GRACE was assessed with the Nash-Sutcliff Efficiency (NSE) value (Nash and Sutcliffe 1970), a popular measure of the goodness-of-fit between the model predictions and the measurements. A summary of the validation methodology is shown in Fig. 3.