Potential Sites for Runoff Water Harvesting using Geographic Information System and Remote Sensing Data

Abstract

One of the most important challenges in the field of engineering hydrology and water resources management is predicting and quantifying surface runoff. The western desert of Iraq is an arid region that has suffered from a lack of metrological stations and the lack of available data for predicting and calculating surface runoff. The present study aims to model the spatially distributed infiltration and surface runoff using Artificial Neural Networks (ANN) integrated with GIS, RS, and field measurements of infiltration for both dry and wet seasons. For the dry season, field infiltration measurements for 75 soil samples in Al-Ratga catchment area are achieved. The Multilayer Perceptron (MLP-ANN) model is developed to predict the infiltration rate based on the spectral reflectance data. A good agreement was found between actual and estimated infiltration rate values ( R2 = 0.8443). As for the wet season, the infiltration measurements were conducted at 105 points. The Radial Basis Function (RBF-ANN) model is developed to predict the basic infiltration rate. The results refer to a good agreement between estimated and measured infiltration (R2 =0.768). The water balance equation is used to model the surface runoff based on the infiltration data. This model has a good agreement with the Soil Conservation Services-Curve Number (SCS-CN) model. The RBF-ANN model was developed to predict the soil type (sand%, silt%, and clay%), based on spectral reflectance data. The good performance of this model shows that the maximum absolute error was 9.3%, 14.6%, and 4.4% for sand, silt, clay respectively, and the sandy soil has the highest correlation coefficient (0.825). On other hand, infiltration data collected were analyzed to check the ability of the common infiltration models to accurately estimate the infiltration rate. The results showed that all models provided acceptable values for Root Mean Square Error (RMSE) as 1.45, 2.01, 1.88 cm.hr-1 for Horton‟s, Kostikov‟s, and Philip‟s models respectively. The highest model efficiency (ME) was 99% for all models. This indicates that infiltration can be well-described by Horton‟s model than other models in the study area. The results revealed that the total annual (for period 2018-2019) runoff for Wadi Al-Ratica basin by the proposed method is 56235000 m3 , while by VI the SCS-CN method is 49260420 m3 . The combination of artificial neural networks with RS data can develop a numerical model for runoff estimation in an arid region, such as the western desert of Iraq.