This presentation is accessible only to the Illinois State University community.

  • Off-Campus ISU Users: To download this item, click the "Off-Campus Download" button below. You will be prompted to log in with your ISU ULID and password.

Publication Date

4-5-2019

Document Type

Poster

Degree Type

Graduate

Department

Geography, Geology and the Environment

Mentor

Wondwosen Seyoum

Mentor Department

Geography, Geology and the Environment

Abstract

Historically and recently, many people have suffered from severe droughts and/or flooding due to climate changes. In the future, 10-40% of streamflow may be decreased or increased according to accelerated climate changes. Consequently, governments around the world will face a serious challenge regarding water resources management strategy. Since the Rainfall-Runoff (R-R) process is one of the fundamental factors of the hydrological cycle, it is necessary to establish and monitor the rainfall and streamflow responses in various watersheds to attain effective water resource management policy. However, the R-R process is not a function of a single variable but a function of multiple variables such as precipitation, evapotranspiration, infiltration, and topography, which means the streamflow process is complex and is highly nonlinear. Furthermore, most of the watersheds around the world are ungauged which means filed measured data could be scarce in many cases. These constraints could increase the uncertainties of streamflow predictions from conventional approaches such as process-based models and simple empirical models. Machine learning techniques combined with remote sensing data can be an effective tool to overcome the difficulties. The main objective of this study is to characterize a watershed and evaluate the effectiveness of a machine learning-based hydrologic model in simulating the water cycle using remote sensing data that can potentially aid the conventional watershed analyses. By integrating spatial land surface and climate data that describe a watershed as an input dataset in a machine learning model (MLM), and in-situ streamflow discharge data for an output learning dataset, a relationship between watershed characteristics and streamflow is established. The results are validated using in-situ data and compared with results from previous conventional modeling studies. The overall performance of monthly streamflow prediction shows the land surface integrated MLM could be effectively used for streamflow prediction.

Off-Campus Download

Share

COinS