RESEARCH ARTICLE Sand fraction prediction from seismic attributes using optimized support vector regression in an oil reservoir Mohammad Sadegh Amiri Bakhtiar 1 & Ghasem Zargar 1 & Mohammad Ali Riahi 2 & Hamid Reza Ansari 3 Received: 17 April 2019 /Accepted: 15 January 2020 # Springer-Verlag GmbH Germany, part of Springer Nature 2020 Abstract In this study, a new strategy based on integrating geostatistical seismic inversion and optimized support vector regression (OSVR) will be utilized to transform multi seismic attributes to sand fraction log. In first step, owing to compatibility relation between acoustic impedance (AI) and sand fraction, a high resolution value of this important attribute was extracted through a geostatistical seismic inversion (GSI). In second step, in addition to AI, several physical attributes are obtained from seismic data and then all of extracted attributes (AI and other seismic attributes) evaluated by step-wise regression for selecting best attributes that have highest effect on predicting sand fraction. In final step, selected attributes have been fed in the bat inspired optimized support vector regression as input and the sand fraction log is estimated. For the assessment of proposed strategy, the values of predicted sand fraction are compared with their real corresponding values in a blind well. It will be evident from the results that the proposed strategy is qualified for modeling the sand fraction as a function of seismic attributes. Keywords Sand fraction . Seismic attribute . Geostatistical seismic inversion . Support vector regression . Bat algorithm optimization Introduction In the clastic reservoirs, prediction of sand fraction as a stra- tegic parameter is necessary to indicate reservoir net to gross. Since the sand fraction measurement from core data is expen- sive and also is not available in pre-drilling phase, this param- eter can be computed from indirect methods such as geophys- ical logs or seismic attributes. Horizontal distribution makes impossible the prediction sand fraction away from the wells when the well log data is used alone. In such cases, seismic data is a guidance at all traces of the area of interest (Bosch et al. 2010). Seismic attributes are geometric, kinematic, dynamic or statistical transform extracted from seismic data that can be used to obtain enhanced information that might be absent in a traditional seismic sections leading to a better interpretation (Chen and Sidney 1997; Taner et al. 1994). There are many seismic attributes introduced by researchers over past 50 years (Taner et al. 1979; Lavergne 1975; Dalley et al. 2007; Sonneland et al. 1989; Conticini 1984; Vossler 1989). Acoustic impedance (AI) is an important attribute to pre- dict the lithology or other rock properties. One of the major contribution from seismic attributes is inversion of seismic reflectivity into AI. The seismic data are band limited; hence deterministic inversion methodology cannot recover the abso- lute values of the AI from seismic trace directly (Ansari et al. 2014). Furthermore, inversion of the seismic traces can only recover the relative changes in the impedance. Geostatistical seismic inversion (GSI) which introduced firstly by Haas and Dubrule (1994) is a geostatistical method to improve the estimation results of the deterministic inver- sion. Seismic and well logs data are each represented as a probability density function (PDF) and geostatistical descrip- tion based on variograms provides. Geostatistical inversion recovers low frequency component of seismic data, estimates the uncertainty, and it can be computes AI at fine scale. Communicated by: H. Babaie * Mohammad Sadegh Amiri Bakhtiar amiribakhtiar.m.s@gmail.com 1 Abadan Faculty of Petroleum University of Technology, Petroleum University of Technology, Abadan, Iran 2 Institute of Geophysics, University of Tehran, Tehran, Iran 3 Geoscience Department of Kish Petroleum Engineering Company, Tehran, Iran Earth Science Informatics https://doi.org/10.1007/s12145-020-00443-y