Soft Data in Hydrologic Modeling: Prediction of Ecologically Relevant Flows with Alternate Land Use/Land Cover Data

Henrique Haas,Haw Yen,Furkan Dosdogru,Latif Kalin

doi:10.3390/w13212947

Abstract

Watershed-scale hydrological models have become important tools to understand, assess, and predict the impacts of natural and anthropogenic-driven activities on water resources. However, model predictions are associated with uncertainties stemming from sources such as model input data. As an important input to most watershed models, land use/cover (LULC) data can affect hydrological predictions and influence the interpretation of modeling results. In addition, it has been shown that the use of soft data will further ensure the quality of modeling results to be closer to watershed behavior. In this study, the ecologically relevant flows (ERFs) are the primary soft data to be considered as a part of the modeling processes. This study aims to evaluate the impacts of LULC input data on the hydrological responses of the rapidly urbanizing Upper Cahaba River watershed (UCRW) located in Alabama, USA. Two sources of LULC data, i.e., National Land Cover Database (NLCD) and Digitized Landsat 5 Thematic Mapper (TM) images, were used as input in the Soil and Water Assessment Tool (SWAT) model for the years 1992 and 2011 using meteorological data from 1988 to 2013. The model was calibrated at the watershed outlet against daily streamflow from 1988 to 1993 using the 1992 LULC data and validated for the 2008–2013 period using the 2011 LULC datasets. The results show that the models achieved similar performances with both LULC datasets during the calibration and validation periods according to commonly used statistical rating metrics such as Nash Sutcliffe efficiency coefficient (NSE), coefficient of determination (R2), and model percent bias (PBIAS). However, LULC input information had substantial impacts on simulated ERFs such as mean monthly streamflow, maximum and minimum flows of different durations, and low flow regimes. This study demonstrates that watershed models based on different sources of LULC and applied under different LULC temporal conditions can achieve equally good performances in predicting streamflow. However, substantial differences might exist in predicted hydrological regimes and ERF metrics depending on the sources of LULC data and the LULC year considered. Our results reveal that LULC data can significantly impact the simulated flow regimes of the UCRW with underlaying influences on the predicted biotic and abiotic structures of aquatic and riparian habitats.

Highlights

This study demonstrates that watershed models based on different sources of LULC and applied under different LULC temporal conditions can achieve good performances in predicting streamflow
What interests us is to investigate an unexplored scientific question: How do hydrological models that were set up based on different sources of LULC data respond in terms of ecologically relevant flows (ERFs) conditions? we aim to address the following related questions: (i) If LULC data from different sources are used as model input data, is it possible to achieve good model performances in streamflow prediction through automated model calibration? (ii) If yes, would those models predict similar streamflow under future LULC? (iii) What is the impact of different LULC data sources on ERF metrics? To answer these questions, we employed the Soil and Water Assessment Tool (SWAT) model in an urbanizing watershed with high biodiversity, the Upper Cahaba River Watershed in Alabama, USA
The results presented here indicate that different sources and years of LULC input data can yield similar performances in watershed models based on statistical rating metrics such as Nash Sutcliffe efficiency coefficient (NSE), R2, and percent bias (PBIAS), yet marked differences were found in analyzing other simulated flow metrics

Summary

Introduction

Hydrologic models have been widely used to assess the interplays between land use/cover (LULC) changes and hydrological processes and play a pivotal role in regulatory, planning, research, and decision-making efforts [1,2]. The simulated hydrologic fluxes and processes contain uncertainties from various sources since hydrologic models are essentially a simplified representation of natural systems [3]. Model predictive uncertainties can be attributed to forcing data (e.g., time-series of precipitation), input parameters (e.g., parameterization of soil physical properties), and an oversimplified model structure and representation of hydrologic processes (e.g., streamflow) [4–6]. Studies such as [4,7,8] examined the parameter uncertainty stemming from different

Objectives

Methods

Results

Discussion

Conclusion