Decision-making about the sustainability of water resources requires supporting tools to manage the resources effectively. Hydrological models are useful tools to capture spatiotemporal variations in hydrological fluxes in many basins. A model consists of various parameters that define the characteristics of the basin. Hydrologic models have become progressive tools for studying the effects of anthropogenic activities and environments on hydrology and ecology [1
]. Good modeling should be characterized by a high degree of confidence in the simulated outputs and minimization of uncertainty in model results. These uncertainties may be due to data input integrity, measured output data, non-optimal parameters, and model bias [2
The conventional knowledge is that the best simulation is associated with less uncertainty in its output. The best model is the one that gives results close to reality with the use of the least parameters and model complexity, i.e., a parsimonious model [3
]. Hydrologists are faced with improving the quality of model outputs by using various procedures to reduce these uncertainties [4
]. However, errors due to model bias are difficult to address and are usually mentioned by the modeler when reports are made. To address uncertainties relating to non-optimum parameter values, model calibration is employed. This procedure is targeted at obtaining optimum parameter values for the model setup in the specific basin, such that the error due to non-optimal parameters is insignificant compared with errors due to data integrity [6
Various strategies have been adopted to calibrate hydrological models. Scientists have introduced series of mathematical functions that guide parameter optimization during the calibration of the model [7
]. Hydrologists are also engaged in exploring model setups and data inputs to obtain the best model performances. Various resolutions of Digital Elevation Models (DEM), LULC source and resolutions, soil map resolution, and data length have been studied to obtain their effects on model performance. These responses vary among models due to model types and structures [10
]. Physically distributed hydrological models are capable of simulating multi-layered hydrological processes. Rather than utilizing extensive hydrological and meteorological data during calibration, physically distributed hydrologic models require the evaluation of many parameters that describes the physical characteristics of the catchment [11
]. These models provide output upon which decisions about effective water resources management are based especially with spatiotemporal changes in LULC [12
]. Therefore, the accuracy of these outputs is of high importance to hydrologist, hence the need for efficient calibration procedure to design simulations representing the interested basin.
The effect of input data resolution on SWAT model performance was studied by Bouslihim et al. [15
], which focused on simulating hydrologic fluxes using various spatial resolution of soil data. Likewise, the temporal resolution effect was observed by Ficchì et al. [16
] using different precipitation time-steps to simulate hydrologic events. These studies showed that quality and resolution of data affect the estimation of hydrological fluxes. According to Li et al. [17
] in their study on the effect of calibration length on lumped hydrological model performance, longer calibration periods do not necessarily result in better model performance and optimum parameter values can be attained with fewer calibrations. This was enhanced in recent research by Ilampooranan et al. [18
], which further clarified the need for additional calibration data sources to improve the robustness and predictive ability of distributed models. The rise in technology, data accessibility and technical capabilities are gradually improving the complexity of hydrologic models. This raises the question of how model complexity affects model performance and was elucidated by Orth et al. [19
], which concluded that complexity does not necessarily cause increased model performance.
Vegetation can have a significant effect on hydrological fluxes due to variations in the physical characteristics of the land surface, soil, and vegetation; such as the roughness, albedo, infiltration capacity, root depth, architectural resistance, leaf area index (LAI), and stomatal conductance [20
]. Setyorini et al. [22
] utilized SWAT model calibrated with the LULC map obtained from supervised classification of Landsat images to evaluate the impact of LULC changes and climatic variables on hydrological parameters. The combined effect of both variables decreased surface runoff, groundwater, lateral flow and stream flow while evapotranspiration increased. Boongaling et al. [23
] further studied the impact of LULC changes using the SWAT model configured with land cover data generated from Satellite Pour l’Observation de la Terre (SPOT) 5 imagery with a resolution of 10 m and revealed that vegetated sub-basins tend to loosen the soils and permit improved rate of infiltration leading to increase base flow and decreased overland flow. The sensitivity of vegetation parameters to hydrological processes was also assessed by Das et al. [24
] using Land Use Land Cover (LULC) changes in the eastern India basin between 1995 and 2005 to set up Variable Infiltration Capacity (VIC) model, which revealed the LAI as the most sensitive parameter that influences hydrological fluxes. Yearly variability in LAI is essential in model calibration and monthly water balance as observed by Tessema et al. [25
] in the simulation of runoff in Goulburn–Broken catchment of Australia using VIC model.
Furthermore, Chen et al. [26
] simulated the effects of irrigation waters on surface water and energy balance fluxes using VIC model and irrigation scheme for Heihe River basin in China, which revealed a greater increase in evapotranspiration as irrigation activities persists. Additionally, the impacts of future climate changes on hydrological characteristics resulted in the decline in streamflow as studied by Hashim et al. [27
], using HBV to simulate fluxes in Harver River catchment, although these impacts were later found by Bisht et al. [28
] to be flow type and time-step dependent, as observed using integrated MIKE 11 NAM-HD. In a different study by Jasrotia et al. [29
] to predict streamflow under changing climate conditions using VIC, precipitation projections in Jhelum catchment was observed to have a huge influence on runoff. The application of cell to cell routing and modified parametrization was revealed to increase model performance in a conceptual model as studied by Paul et al. [30
] using Satellite-based Hydrologic Model (SHM). LULC changes through deforestation, urbanization, expansion of croplands led to reduction in the extent of canopy cover for interception and transpiration, which affects hydrologic model performance. Jin et al. [1
] reiterated this inference in their study on the impact of multiple LULC datasets and resolutions on hydrologic performance and suggested using a high-resolution LULC dataset for modeling when only one year LULC dataset is available.
Most previous studies have been exploring various strategies to improve model performances by adjusting input data features, model set up and calibration approach. However, little information exists about the combined effect of multiple LULC datasets and LAI on the performance of fully distributed physically based hydrologic models like the mesoscale Hydrologic Model (mHM). Hence, this study aims at (1) exploring the performance of a distributed hydrologic model by calibrating the hydrologic model using a series of land use land cover datasets and (2) analyzing the influence of leaf area index, which is a significant characteristic of forest areas on the performance of the model. This research involves different model configuration cases with varying LULC, potential evapotranspiration (PET) setup, and comparing the performance and hydrograph of each simulated output. Some limitations of the study is that while there exists different land cover classes in nature, the model only identifies three significant land cover classes (Forest, Impervious and Pervious). Hence, error in reclassification of closely related classes is inevitable. Furthermore, only annual LULC data can be defined in the model, despite the use of multiple land cover inputs. Finally, users can manually define LAI classes in look up tables, if LAI is not provided as input.
The need to estimate the impacts of climate and land use change on discharge regime has made hydrologic models popular worldwide, especially with their ability to predict flows at gauges and ungauged catchments. However, these predictions are subjected to uncertainty due to model bias, input data errors, and errors in model parameter values [60
]. The goal of hydrologic modelers is to reduce uncertainties associated with the model outputs upon which decision about hydrologic fluxes are based. There have been numerous studies focusing on different ways to improve model performance by exploring various input data characteristics that affects the predictions in conceptual and physically distributed models [17
]. Models like mHM are spatially distributed since they consist of equations involving one or more space coordinates, which are used for simulation of spatial variation in hydrological variables within a catchment as well as simple outflows and bulk storage volumes. Such models make considerable demands in terms of computational time and data requirements and are costly to develop and operate [6
]. The results presented in this study provide the spatial model’s response to varying characteristics of the input data based on the objective under study. Recall that the scenarios labeled as Case 1 to Case 4 and their varying input characteristics. Case 1 and Case 2 were configured with single LULC data, with LAI influence on the Case 2 only. The same applies to Case 3 and Case 4 but with multiple land cover classes, which makes it possible to observe LAI’s effect on the model performance. LAI is a key biophysical vegetation property for interception that describes biome-specific canopy structure and an essential variable in evaluating the interrelation of the components of the water-soil-plant-atmosphere system, which influences environmental studies [64
]. The vegetation of an area due to its LAI is a significant component of land surface models that influence evapotranspiration simulation and, consequently, affects base flow, recharge, infiltration excess, saturation excess, subsurface storm flow and catchment wetness [25
]. As hypothesized, the cases LAI data performed better than their counterparts during calibration and validation process. The implication is that predictions are more accurate when leaf area index is included in the model for correction on evapotranspiration. Precise evaluation of evapotranspiration is a vital aspect of water balance estimation. About 60% of terrestrial precipitation returns to the atmosphere by plant transpiration (40%), or through direct soil evaporation (20%) [65
]. The correct simulations of LAI seasonal dynamics and stomatal aperture in an eco-hydrological model are prerequisite of good simulation of canopy radiation exchanges and transpiration fluxes which are important components of the accurate water balance estimation [66
Furthermore, the impact of single and multiple land cover data was observed. The result associated better model performance with the cases with multiple land cover data. This corroborates Jin et al., [1
], where the use of multiple land-use datasets improved model performance although with some complexities in simulation complexity due to greater number of LULC patches. LULC maps are essential input of physically distributed models since they spatially delineate the basin’s different morphological characteristics that decide water’s fate on the earth’s surface. A detailed characterization of land cover is essential for urban areas, as coarse spatial description might result in biased estimates by hiding urban heterogeneous evapotranspiration [67
]. Petrucci & Bonhomme [68
] also found a similar result to our study where increasing geographical information clearly improves model performances with the land use classification providing the highest benefit. Finally, considering the combined effect of both morphological characteristics, the case with multiple land cover data and LAI to correct evapotranspiration performed better than the other cases and the case with single land cover data, without LAI correction performed worse. It is noteworthy that the discharge data were used in the model’s calibration, which leaves the possibility of variation in model performance if different calibration method or data was used. The data-intensive nature of physically distributed models often leads to over-parameterization while effectively representing each hydrologic process that constitutes such a model. This opens up another phase of research in hydrologic modeling to solve over-parameterization and identify strategies to improve model performance without overwhelming the model with information.
FDC provides an indepth response of the flow to varying configurations of the model set up. While the total FDC gives a general response of the basin to the stream flow timeseries, focusing on the response of specific flow type sheds essential information for various aspects of water resources management, such as high flows for dam construction and low flows for irrigation mangement [69