The Application of Radial Basis Network Model, GIS, and Spectral Reflectance Band Recognition for Runoff Calculation

The Application of Radial Basis Network Model, GIS, and Spectral Reflectance Band Recognition for Runoff Calculation

Hadeel Qays HashimKhamis Naba Sayl 

Dams & Water Resources Department, Engineering College, Anbar University, Ramadi 31001, Iraq

Corresponding Author Email: 
hadeel_ded@uoanbar.edu.iq
Page: 
441-447
|
DOI: 
https://doi.org/10.18280/ijdne.150318
Received: 
13 March 2020
|
Accepted: 
10 June 2020
|
Published: 
30 June 2020
| Citation

OPEN ACCESS

Abstract: 

Runoff estimation in a watershed is very important for efficient management of scarce water resources. Soil information is essential information for runoff estimation. Data collecting and determination of soil textural classification for large territory using the traditional method, i.e. laboratory testing is time-consuming and costly. Therefore, this study suggested a model based on the combination of Radial Basis Neural Network (RBNN) model, Geographic Information System (GIS), Remote Sensing (RS) and field data to create a digital soil map. This model was studied as a case study in western Iraq, and it was tested using performance parameters. The findings of this model were further confirmed using the hydrological soil group developed by the United States Geological Survey (USGS). The adopted model has been successful in predicting the spatial distribution of clay soil, followed by both silt and sand. It was also noted that the Root Mean Square Error (RMSE) for clay, silt and sand is 4.2 percent, 9.5 percent and 11.0 percent respectively, while the highest value was for the coefficient of clay soil correlation (0.749). Furthermore, there are only four samples out of 25 that have minor variations in the estimated and measured soil texture category defined by USGS. The methodology adopted in this study is therefore well practical for soil classification. Additionally, a broad scale will produce high-quality runoff measurement.

Keywords: 

artificial neural network, hydrological soil group, Remote Sensing (RS), Geographic Information System (GIS), runoff depth, Soil Conservation Service-Curve Number (SCS-CN)

1. Introduction

For both arid and semi-arid regions one of the most significant natural resources is freshwater. Nevertheless, according to the paper [1], only 1 percent of the 2.5 percent extensive freshwater is obtainable for human use. One of the promising alternatives for freshwater is water that is stored and used especially in the arid and semi-arid area from the surface runoff. The west desert of Iraq is categorized as an arid region and it suffers from water scarcity [2]. Surface runoff has the potential to be available and safe to human needs. Precise measurement of surface runoff and its volume plays a key role in the management of the watershed in arid and semi-arid regions. Nonetheless, large scale runoff estimation was a difficult task. Therefore, more work is needed.

Recently, uncalibrated distribution runoff-rainfall models were obtained when runoff data are unachievable to simulate catchment rainfall-runoff response in semi-arid regions [3-5]. These models all have drawbacks. They need several input factors that reflect different catchment characteristics, and they also do not completely produce spatially distributed predictions of runoff values. This is because data collection for large catchment area is complicated and time-consuming. However, optimized runoff-rainfall model was successfully applied to achieve practical runoff simulation where such data is available in these regions [6]. The construction and implementation of a large-scale optimized rainfall-runoff model is difficult due to the lack of detailed field data. The rate of runoff to rainfall relies on the texture of the soil which is known to be the key parameter influencing the catchment runoff. Soil texture is therefore an important factor in terms of the stability and the surface runoff capacity [1].

The Soil Conservation Service (SCS) method is the most widely used in hydrology to predict runoff from rainfall from a satisfied rainfall event [1]. GIS is capable of handling various spatial data needed for modeling and can be used as a tool in distributed hydrological modeling. GIS spatially distributed modeling approach applying the curve number (CN) framework has been used in several studies [7-14]. This modeling process depends on which CN is given. CN is a derivative of land use/cover and texture of soils. Many hydrological models use CN as input to estimate storm runoff, such as Environmental Policy Integrated Climate (EPIC) [15], Soil and Water Assessment Tool (SWAT) [16] and Agricultural Non-Point Source Pollution (AGNPS) [17].

Unadventurously, the determination of the soil texture has been checked in the laboratory, and this method is expensive and time-consuming. RS is one of the most available approaches and the best solutions could provide the ability to extend obtainable soil survey data sets. Indeed, remote sensing is a very significant technique focused on different ranges of the electromagnetic spectrum to predict soil characteristics on various spatial scales [18]. The chemical and physical properties of materials define their spectral reflectance and emittance spectra, which can be used to identify them. Spectral reflectance refers the ratio of radiant energy reflected to the incident energy on a body [19].

Many studies have found that some textual classification of soil can be easily determined based on specific absorption characteristics on a local and laboratory scale [20-23]. The connection between laboratory analysis and image data of French Satellite for observation of Earth (SPOT), Landsat TM, and airborne spectroscopy was shown by Proctor et al. to regulate differences in soil texture class [24]. Five curves formed by Stoner and Baumgardner describe the relationship between spectral reflection and classification of soil that is based on texture and grain size [25]. The image dimension with band 2 and band 8 of Advance Spaceborne Thermal Emission and Reflection (ASTER) was used to estimate the soil texture groups [22]. ASTER`s short wave with band 5 and band 6 can detect clay soil, sandy soil and dark clay soil [23, 26].

The reflectance of the soil is very complex. Consequently, estimating the soil properties based on the physical model is a difficult task [27]. Therefore, we need a method that is able to reveal the complex relationships between reflectance and soil properties.

The current research provides an intelligible method for predicting the surface runoff using integrated GIS and RS data based on soil textural classification. The main aim of this study is to generate a digital soil map based on the spectral reflectance bands with laboratory testing of soil data using the Artificial Neural Network (ANN) model known as RBNN hybrid with GIS model. This map is a base of the hydrological model, which uses GIS to implement the CN method. The current study has proposed a methodology that is of paramount importance to obtain data which are almost minimal and hard to collect. The suggested method has been tested to investigate a field of research in Iraq, particularly in the west desert.