Prediction of reservoir quality using well logs and seismic attributes analysis with an artificial neural network: A case study from Farrud Reservoir, Al-Ghani Field, Libya

Accepted Manuscript Prediction of reservoir quality using well logs and seismic attributes analysis with artificial neural network: A case study from Farrud reservoir, Al-Ghani field, Libya

Abdulaziz M. Abdulaziz, Hameeda A. Mahdi, Mohamed H. Sayyouh PII: DOI: Reference:

S0926-9851(18)30075-2 doi:10.1016/j.jappgeo.2018.09.013 APPGEO 3600

To appear in:

Journal of Applied Geophysics

Received date: Revised date: Accepted date:

28 January 2018 29 June 2018 12 September 2018

Please cite this article as: Abdulaziz M. Abdulaziz, Hameeda A. Mahdi, Mohamed H. Sayyouh , Prediction of reservoir quality using well logs and seismic attributes analysis with artificial neural network: A case study from Farrud reservoir, Al-Ghani field, Libya. Appgeo (2018), doi:10.1016/j.jappgeo.2018.09.013

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT Prediction of Reservoir Quality Using Well Logs and Seismic Attributes Analysis with Artificial Neural Network: A Case Study from Farrud Reservoir, Al-Ghani Field, Libya Abdulaziz M. Abdulaziz†*, Hameeda A. Mahdi†† and Mohamed H. Sayyouh† †

T

Mining, Petroleum, and Metallurgical Engineering department, Faculty of Engineering, Cairo University, Giza 12316, Egypt †† Petroleum Engineering Department, Faculty of Engineering, Benghazi University- Benghazi 9476, Libya * Corresponding Author; [email protected]

IP

Abstract

CR

This paper presents an innovative technique that aims at predicting the quality of a petroleum reservoir using Artificial Neural Networks (ANN) analysis to seismic, well logs and SCAL data.

US

A supervised Probabilistic Neural Network (PNN) has been deployed to predict several reservoir properties, once at a time, through training and validation of the PNN to determine the seismic

AN

attributes that best fit a measured property in well logs. The validated PPN models accurately converted the available 3D seismic data into shale volume, porosity, permeability and water

M

saturation cubes. Alternatively, these predicted reservoir properties have been used by an unsupervised ANN (K-means Clustering Algorithm) to determine the reservoir quality

ED

throughout the area covered by seismic data. This technique is applied to Al-Ghani Field and four grades of reservoir quality have been classified as very good, good, bad, and very bad. The

PT

highest grade is characterized by good porosity and permeability and significantly low water saturation. Based on variability of reservoir properties, the reservoir geobody is characterized

CE

into the predefined four grades, and their spatial distribution is displayed. Such information is highly valuable for optimum reservoir management and well placement. This not only

AC

maximizes reservoir profitability through production schemes, but also minimizes uncertainties in drilling, production, injection, and modeling processes. In addition, adopting the proposed methodology would entail cost reduction in well logging programs and improve the sweeping efficiency of water flooding operations.

Key Words: Reservoir quality, Seismic attributes, ANN analysis, Al-Ghani field, Farrud reservoir

Introduction 1

ACCEPTED MANUSCRIPT The transformation of geophysical data into physical reservoir properties such as elastic or petrophysical parameters represents an inverse problem. The inverse theory adopts mathematical applications with a final goal to deduce reservoir properties with lithofacies characteristics including porosity, lithology, and fluid information from geophysical data (Krief et al., 1990). Well logs and core data represent the only source for real reservoir sampling or measurements taken at a direct contact with the reservoir. Therefore, integration of well logs and core data to

T

other data such as seismic, well test, and production is essential for formation evaluation

IP

(Ahmadi et al., 2014; Abdulaziz et al., 2017), reservoir characterization (Ibrahim et al., 2017;

CR

Ahmadi, 2015) and other engineering applications (Ahmadi et al., 2015; Habib et al., 2016). Core samples provide reservoir description on microscopic (pore size) level, whereas well logs

US

and wireline tests develop megascopic reservoir properties which vary based on log/test resolution (Abdulaziz et al., 2017). Despite the high resolution of well log and core data in

AN

vertical domain, it lacks these capabilities in the spatial domain among wells. Therefore, applications of artificial intelligence (Gholami et al., 2014; Amin and Sadeghnejad, 2017),

M

geostatistical approaches (Burgos et al., 2008) and/or integration with seismic data (Sarhan et al., 2017; Saadu and Nwankwo, 2017; Nguyen et al., 2018) became popular practices to predict the

ED

properties in spatial domain. Of seismic data, seismic attributes provide a versatile resource for integration with log data to enhance reservoir description in three dimensions. Seismic attributes

PT

involve the basic properties of wave signal derived either from direct measurements on seismic data (internal attributes) or by logical or experienced passed reasoning (Taner et al., 1994) such

CE

as seismic inversion (external attributes). Seismic inversion is the process that transforms seismic reflection data into acoustic impedance that helps predicting important reservoir properties for

AC

both rock and saturating fluids (Hampson et al. 2005). Estimating reservoir properties from seismic inverted data are sometimes very important compared to seismic models because most seismic models do not account for the effect of poroelasticity. Seismic inversion entails several categories including pre-stack (fluid indicator) or post-stack (rock indicator) based on seismic data type and deterministic, stochastic, or geo-statistical based on the analysis technique. These techniques typically implement real reservoir measurements such as SCAL and well logs (Russell, 2014; Hampson and Russell, 2006; Hampson et al., 2005; Moosavi and Mokhtari 2016). Seismic inversion proved valuable in reservoir characterization and monitoring (Farfour et al., 2015), mapping permeability barriers and high permeable paths (Ramirez et al., 2013), dynamic 2

ACCEPTED MANUSCRIPT reservoir characterization and EOR (Wang et al., 2017), and sweep mechanisms (Xu et al.1997; Staples et al., 2006; Bousaka and Donovan 2000). Internal seismic attributes have been widely applied in exploration of hydrocarbon resources (Xu et al. 1997; Wang, 2001; Nanda, 2016), facies variation (Farfour et al., 2015; Partyka et al. 1999), and in identifying the predominant reservoir drive mechanism (Xu et al. 1997, Staples et

T

al 2006, Bousaka and Donovan 2000, He,et al.,1997). These attributes involve spectral

IP

decomposition, tuning thickness, instantaneous amplitude (Reflection Strength), instantaneous

CR

frequency, and instantaneous phase. In addition, geometric Attributes (Dip and Azimuth, Curvature, Coherence) are used for subsurface structural characterization, and identification of lateral variations in stratigraphic features (Nanda, 2016). Instantaneous frequency provides an

US

estimate of wave frequency with time which is very useful in depicting stratal details of thin reservoirs (Partyka et al. 1999). Alternatively, instantaneous frequency and amplitude related to

AN

tuning are capable of revealing the presence of hydrocarbon through high frequency loss by adsorption. Consequently, low frequency regions define the reservoir target for drilling

M

(Robertson and Nogami 1984). However, other causes other than hydrocarbon such as reservoir lithology and facies change, particularly thin formations of deltaic environment may induce high

ED

frequency loss (Ebrom 2004). Spectral decomposition attributes involve isolating the seismic frequency content into domains of interest that helps the evaluation of stratigraphic details in thin

PT

layers (Castagna and Sun, 2006). Thus, spectral decomposition exposes lateral discontinuity and identifies lateral facies change that is useful for selecting optimal drilling trajectory (Partyka et

CE

al.1999, Rijks and Jauffred, 1991). Dip and azimuth attributes present a quantitative estimate to the shape of geologic bodies. Sharp discontinuity in geometric attributes typically indicates the

AC

presence of faults or fractures important to hydrocarbon production. For example, the fracture density and trends were mapped using curvature attribute analyses in Argentine basins (Sigismondi and Soldo 2003). Recently, ANN has been extensively applied in property prediction and/or problem solving in petroleum engineering studies (e.g. Al-Kaabi, 1993; Zhangxin, 2007; Habib et al., 2017). This includes drilling engineering (Ahmadi, 2016), reservoir fluids characterization (Ahmadi et al., 2015a; Ahmadi et al., 2015b; Ahmadi and Ebadi, 2014 ), Enhanced Oil Recovery (Ahmadi, 2015a; Ahmadi et al., 2015c), heavy oil recovery (Ahmadi et al., 2014a) and reservoir 3

ACCEPTED MANUSCRIPT engineering (Ahmadi, 2014; Ahmadi et al., 2014b; Ahmadi and Mahmoudi, 2016). ANN adopts nonlinear parallel processing approaches using numerous algorithms (Anifowose et al., 2016). To select the optimum algorithm, neural network must be trained with real and representative data (Anifowoseet al., 2013). Generally, applications of neural network fall within two categories; the first category involves classification problems, and second category, as being adopted in this work, involves property prediction (Hampson et al. 2005). Prediction with ANN

T

has many advantages including; high prediction power, better resolution, enables cross

IP

validation, and allows using various attributes combined with amplitude and time domains

CR

(Russell, 2014; Habib et al., 2017). Neural networks are utilized through two main techniques; supervised and unsupervised techniques. The supervised ANN technique is used in solving

US

problems with input and known output data. This technique allows estimating a pertinent relationship between input and output data (Habib et al., 2017). Therefore, supervised ANN

AN

would provide reasonable interpretation such as the relationship between input and output variables which is already specified during the teaching process. One of supervised ANN

M

limitations is that the numerical values of input data should be set in order to get accurate output appropriate to ANN training,. Numerous methods of supervised ANN have been developed

ED

including Multi-layer Feed Forward (MLFN), Radial Basic Function (RBF) and the Probabilistic Neural Networks (PNN) (Russell and Hampson1991; Hampson et al., 2001; Russell, 2014;

PT

Hampson and Russell, 2006; Hampson et al. 2005). Alternatively, unsupervised neural network utilizes a solving algorithm that investigates data patterns of a series of input data without the

CE

need for any specific outputs. A main advantage of this method is that it doesn’t necessarily require to know the output at the beginning of any trial (Russell, 2014; Habib et al., 2017).

AC

Al-Ghani Field is discovered in 1972 within the western part of Sirt Basin with estimated oil reserve of 0.5 billion barrels, and is currently producing 10000 barrel per day (AL-Harouge, 1980). Despite the long production history, the reservoir quality of Farrud pay zones is only known at well location leaving large areas among wells unknown. Such gaps in reservoir description represent a major challenge to future development plans. In the present study, a novel technique which adopts seismic, well logs, and core data with ANN is developed to map out important reservoir parameters and identify the overall quality throughout the reservoir geobody. Geological Setting and Data Set 4

ACCEPTED MANUSCRIPT Farrud reservoir, also known as Beda-C, covers 7200 acres and represents the main reservoir in Al-Ghani field. It is typically made of dolomitic to calcareous facies that trap sweet crude (Figure 1-upper). Farrud sediments formulate the largest member within Beda Formation that includes Facha and Gir members as well (Figure 2). It represents a regressive carbonate depositional cycle with an average thickness of 76 ft deposited in a shallow to fairly open marine environment. Although Farrud is conventionally called a Formation as stated in the operating

T

company reports, but it represents a stratigraphic member of the Beda Formation (Figure 2).

IP

Lithostratigraphically, Farrud member is sandwiched between Hagfa Shale and the Dahra

CR

Formation of predominant shale lithofacies. Thus during Farrud deposition, marine conditions had changed radically from deep to shallow marine, and the transport of land derived clay

US

sediments into the basin had entirely stopped. While Farrud reservoir had initial poor porosity, dolomitization changed this initial setting dramatically. Dolomitization was so quick and

AN

complete that left only a little of the original structure and texture of the sediments. Structurally, the important features in Ghani reservoir involve the Northwest to Southeast trending faults

M

which lead to a sudden increase in pay thickness at the center of the field (AL-Harouge, 1980). The present study involves log data of 19 wells, core data from three wells and 3D poststack

ED

seismic data (Figure 1-lower). The log data involves density, sonic, gamma ray, resistivity, and neutron porosity logs. The available wells are separated into 13 wells for initial ANN model

PT

training and the remaining 6 wells are utilized for verification purposes (Table 1). The seismic cube involves 300 in-lines and 194 crosslines that extend between 600 ms and 1,200 ms covering

Methodology

CE

the reservoir interval that dominantly fall between 960 ms, and 1,150 ms.

AC

To predict the 3D distribution of inherent reservoir properties from the available data, a PNN is trained using the pertinent well logs derived-reservoir property and the corresponding seismic attributes at well locations. Such a method is accurate with high resolution as it involves both detailed log response tied with the comprehensive seismic response throughout the geologic media. The detailed workflow of this analysis is summarized in Figure 3, and some details of the key steps are briefly described. First, log data set is reviewed for consistency and removal of spurious signals, while formation tops, check shot, and well survey are checked for consistency and quality assurance. Seismic data is primarily preconditioned to improve the seismic data quality, and reduce noise using OpendTect version 6.0 software (dGB Earth Sciences). Seismic 5

ACCEPTED MANUSCRIPT and log data analysis involve well-to-seismic tie, picking horizons, and wavelet extraction which is, in the present case, a zero phase wavelet. Time-depth conversion of seismic data is completed using the check shot of well RRR01, while the depth-time conversion is automatically completed for wells using the sonic-check shot relationship to enable data integration and facilitate well tie in time domain. Wavelet extraction is accomplished using zero phase statistical approach that showed higher correlation compared to other techniques, and converges faster than minimum

T

phase wavelet (Hampson et al., 2001). Horizon picking is completed to guide interpretation of

IP

reservoir properties within the important zones of Farrud member. Log data analysis includes

CR

calculations of shale volume (Stieber, 1970), porosity (Bertozzi et al., 1981) and water saturation (Archie, 1942) using the relationships presented in Table 2. In water saturation calculations, the

US

Archie parameters, a and m, are determined in each well using Hingle (Hingle, 1959) and Pickett (Pickett, 1966) plots and “n’ was set as 2. In addition, the impedance from log data at layers

AN

interfaces (Table 2) is also calculated to build the initial guess inversion model. Inversion analysis uses correlation ratio, error ratio, and matching ratio to check the performance

M

of inversion method that best fit the available post stack seismic data. Inversion is accomplished using 3D seismic data and 11 wells. Figure 4 demonstrates the correlation between the inverted

ED

trace (red trace to the left) and the original acoustic impedance log, blue trace calculated from sonic and density logs. Among the available inversion types, colored inversion reported high

PT

correlation ratio and low error ratio. In addition, it showed the best inversion results in both elastic and acoustic impedance and, therefore, is adopted for executing inversion process. In the

CE

present, study both internal and external attributes are exploited by PNN, and over a hundred attributes have been tested in extracting impeded properties within seismic records. The PNN

AC

structure adopts a probabilistic algorithm in simple network structure with a single hidden layer and the number of neurons is adjusted to achieve the best predictions. The input layer comprises a number of neurons adjusted to fit the number of seismic attributes involved in the analysis while the output layer involves a single neuron for target property. The trend of multi attribute transform is cascaded and the result is smoothen with a smoother length of 50 samples. PNN implements 25 sigmas that range between 0.1 and 3.0, and 20 conjugate gradient iterations. The success of attributes analysis varies dramatically, and some types proved to be diagnostic. Therefore, the results of attributes analysis are strictly verified using well data. Accordingly,

6

ACCEPTED MANUSCRIPT seismic attribute analysis is applied to predict porosity, water saturation and absolute permeability as key reservoir properties. For predicting each reservoir property, three consecutive steps are accomplished. This includes integration of inversion results with original seismic data using PNN for attributes analysis, verifying the neural networks results, and converting seismic data into petrophysical property of

T

interest. PNN facilitates defining the attributes in seismic traces located nearby a well that best fit

IP

the property calculated in the corresponding well log (training process) and subsequently

CR

deriving PNN model based on log-seismic correlation for a number of attributes. This involves assigning a weight for each attribute based on this correlation, and subsequently predicting the reservoir property from available seismic attributes at well location. This process is repeated

US

using different numbers of attributes (Try and Error technique), until the best fit between the predicted and actual values is achieved in training wells. Then, the PNN model is applied to the

AN

entire seismic data set. The resulting reservoir property is subsequently validated using independent wells (verification wells) with actual reservoir property at the corresponding

M

location where seismic attributes are used. In such analysis, cross correlation, on well bases, is used as an indication for PNN performance in property prediction. Generally, the PNN model is

ED

accepted where cross correlation exceeds 0.70 in all verification wells, otherwise the number of attributes is changed. Table 3 presents the number of wells used in PNN training and verification

PT

for each studied reservoir property. The available seismic data cube is converted into four property cubes using the verified PNN for each corresponding property. Finally, the four

CE

property cubes are analyzed with unsupervised ANN using class and match techniques (K-means Clustering Algorithm) to define four grades of reservoir quality (very good, good, bad and very

AC

bad). In this technique, 500 vectors (Sw, phi, Vsh, and K) representing the different reservoir grades are plotted. And based on a predefined n-classes, four in this case, the data are characterized into n-clusters. Then, the statistics for each cluster are calculated, including the mean value, mode and standard deviation to formulate data base for the match technique. In match technique, K-means Clustering algorithm assigns all data points in Farrud reservoir to the suitable class based on similarity to the mean value.

Results and Discussion

7

ACCEPTED MANUSCRIPT Petrophysical analysis for all available wells are completed and a simple statistics of the results is presented in Table 4. Table 5 presents the results of well tie correlation with the reported depths for Farrud Formation and the application(s) of each well in the present study. Inversion of available seismic data, 600 - 1200 ms, imaging the reservoir interval is completed using the colored inversion. Despite the significant correlation between P-impedance and both porosity and water saturation in Farrud reservoir, inversion data is preferred to be integrated in attribute

T

analysis instead of using the P-impedance as a proxy for porosity and water saturation. Four

IP

reservoir properties including porosity, volume of shale, water saturation and absolute

CR

permeability are analyzed using seismic attributes and well data for property prediction throughout the reservoir. Table 6 lists the seismic attributes used for each property prediction,

US

while the results of error analysis applied to both training and verification wells are presented in Figure 5. In addition, for each property the error analysis is calculated for individual attributes,

AN

and Table 7 shows an example for the error analysis calculated for each attribute involved in permeability prediction. Detailed discussion to the results of converting seismic data to each

M

petrophysical property using PNN is presented independently.

ED

Prediction of Porosity

In porosity analysis, 12 wells are used for PNN training to predict total porosity using seismic

PT

attributes from original seismic data and inversion results. The porosity used for PNN training is calculated from Neutron and Density logs using standard relationships (Dobrin.et al, 1988 and

CE

Archie, 1942) (Table 2). Teaching of PNN using five seismic attributes is considered acceptable, as the correlation coefficient approaches 0.90. These attributes are filter 35/40-45/50, amplitude

AC

weighted phase, instantaneous phase, filter 25/30-35/40, and the dominant frequency (Table 6). This PNN model is applied to the seismic data to develop total porosity cube shown in Figure 6. The error analysis of the porosity prediction indicated that most wells had an average error of 3.5 % with maximum error (5%) reported at RRR21 well, and the minimum error (2.0%) in RRR31 well (Figure 5). Alternatively, the prediction errors typically fall close to 7% except the porosity predictions in wells RRR07, RRR09, and RRR21, where error approaches 10% (Figure 5). Validation of the porosity results is accomplished using two wells (RRR08 and RRR33) shown in Figure 7. A good correlation can be seen in both wells even if the exact rhythm is not precisely followed due to resolution issues, as logs typically resolve 1.0 m features while seismic data, in 8

ACCEPTED MANUSCRIPT the best cases, resolve 10 m objects. Generally the predicted values are 3 to 4 porosity unites underestimated compared to the corresponding values calculated for density and porosity logs. This could be due to the fact that the presented values of predicted porosity represent the average of all traces located nearby the well. The analysis of the resulting porosity cube showed that the wells drilled in the area targeted the high porosity zones within Farrud reservoir (Figure 6). In addition, a close view at the predicted porosity cube nearby wells showed that predicted porosity

T

captured correctly the porosity response measured in wells (e.g. RRR08 and RRR33 wells). This

US

Predicting Shale Volume

CR

identifying the pay zone in Farrud reservoir (Figure 7).

IP

can be depicted where the higher porosity values correlate well with the high resistivity zones

Testing the performance of manual analysis of single and multiple attributes in predicating the

AN

shale volume at eight well locations showed low capabilities as indicated by the low correlation coefficients (~0.60). Therefore, Neural Network is applied to help selecting the proper attributes to improve the prediction of shale volume. Hence, shale volume logs of eight wells and the

M

corresponding seismic attributes are used to train PNN at well locations. The optimum

ED

correlation (0.83) between predicted and actual shale volumes is obtained after PNN selection of five attributes (Table 6). Accepting this PNN model as a proxy for converting the seismic data

PT

into shale volume, the converted cube shown in Figure 8 is obtained. The detailed investigation of the converted shale volume helps defining the distribution of reservoir rock in the study area

CE

using a suitable Vsh cutoff, 15% for example. Generally, well placement was successful to target low shale content places (Green color), but higher shale content (> 80%) may indicate a good seal for potential reservoir units. The error analysis of the input data is prepared to show the

AC

contribution of each well to the error in Vsh prediction. The errors calculated at most wells were analogous (~1%), except PRR07 well where the deviation had increased to report about 4.5% (Figure 5). In addition, two wells (RRR59 and RRR 30) are used for results verification and a very good correlation is obtained in both wells, 0.75 and 0.65 respectively (Figure 9). This is particularly true within the Farrud pay zones, but the mismatch significantly increases outside the pay intervals (e.g. RRR 59, Figure 9). Water Saturation Prediction

9

ACCEPTED MANUSCRIPT In water saturation prediction of Farrud units, 15 wells are used with 12 wells randomly selected for PNN training, while 3 wells are reserved for results verification. In PNN teaching, several attributes are tested and results showed the best six seismic attributes (Table 6) that provide acceptable prediction. The PNN model is considered acceptable as the correlation coefficient between predicted water saturation and log-derived water saturation approaches 0.98. Then, the PNN model is applied to convert the available seismic data to the water saturation cube

T

presented in Figure 10. Error analysis is applied to the 12 wells used in PNN teaching, and the

IP

results indicated that most wells attained 2.5 to 4.0 % errors in Sw prediction (Figure 5). In this

CR

error analysis, the maximum error (5.25%) is reported at RRR21 well while the minimum error (2.0%) is calculated at RRR31 well (Figure 5). In addition, a comparison of the predicted water

US

saturation versus interpreted water saturation from log data is prepared at two wells, RRR08, and RRR71, for results verification (Figure 11). In RRR42, RRR08 and RRR71 wells the water

AN

saturation pattern of log data is very well captured in the converted traces, particularly within Farrud zones. This is indicated with correlation coefficient of 0.79 and 0.80 calculated at RRR08

M

and RRR71wells, respectively (Figure 11).

Figure 12 presents the investigations applied to a section within the converted water saturation

ED

cube using RRR71 well to verify the obtained results. Low water saturation zones (red color) are recognized at four intervals indicating potential hydrocarbon accumulation in the proximity of

PT

Farrud sediments (790 ms), which are significantly correlated with the high resistivity log records (Figure 12). This indicates a feasible occurrence of three other pay zones that are

CE

respectively encountered at 735 ms, 755ms, and 765 ms (Figure 12). The upper most zone (735 ms) represents the Gir Formation with confirmed oil production at this depth in Al Ghani Field

AC

as documented in the operating company reports. The second (755 ms) and third (765 ms) predicted pay zones correspond, respectively, to Facha and Mabruk lithostratigraphic units. Despite the lack of log data for these two Formation, the internal company reports indicated production from these zones also. Such a result indicates a successful prediction of water saturation from the present analysis and successful tracking of oil bearing intervals across the sedimentary section. A similar result, but with different pattern, is recognized at lower depth around RRR8 well where the pay zones are not continuous. In other locations, the converted water saturation section shows a successful depiction of the low water saturation zones confirmed by high resistivity log response in many other wells penetrating Farrud reservoir. To 10

ACCEPTED MANUSCRIPT further verify the Sw results, the converted sections of water saturation and porosity are compared to the impedance data for the area around RRR08 and RRR59 wells. The low water saturation zones (red tone in water saturation section) are correlated well with intermediate porosity (8-12%), and both consistently showed intermediate P-impedance values. Furthermore, the time slice (Figure 13) shows numerous wells with low water cut placed within the interpreted low water saturation zones. For other wells with increasing water cut, a sidetrack to a nearby low

T

water saturation spots would probably improve the oil production of these wells and

IP

consequently increase the overall recovery.

CR

Permeability Prediction

The permeability analysis in the present study considers the intrinsic permeability, a rock

US

property measured in milli-Darcy. According to Hubbert (1956), Shepherd (1989) and Fetter (2001) intrinsic permeability (Ki) of a basin is dependent only to the mean pore size of the

AN

sediments, essential petrofabrics property. Since seismic wave propagation and interaction at the interface are controlled by the petrofabrics of the geologic media, the change in pore size would

M

be depicted within seismic records. Spatial and vertical distribution of absolute reservoir permeability is predicted from seismic data using PNN-based model built with the aid of the core

ED

data measured permeability at well locations (Figure 14). This involves merging core data of two wells with seismic data using PNN to define the best 26 attributes (Table 6) as an optimum

PT

function for predicting the absolute permeability in Farrud reservoir. The higher number of versatile attributes that involve amplitude, phase, spectral decomposition and frequency together

CE

with their derivatives indicates the complex data relationship to predict permeability. The amplitude and phase attributes inherit their characteristic at layer interface while spectral

AC

decomposition (filter attributes) and frequency are strongly affected by the propagation media. The PNN model is considered acceptable as the correlation between predicted permeability and core permeability of RRR01 and RRR37 wells approaches 0.97. Error analysis is applied to the data points of these two wells used in PNN teaching, and the results reported 10.5 % permeability prediction errors in RRR37 well, while 1.5 % error is calculated at RRR01 well (Figure 5). This PNN model is applied to convert the available seismic data to a permeability cube. The results showed that the permeability in Farrud reservoir falls between 12 mD and 141 mD, which closely agree with the values documented in the internal reports of the operating company. To further investigate the validity of results, the core permeability RRR10 well is 11

ACCEPTED MANUSCRIPT compared to the predicted permeability at well location (Figure 15) .The comparison indicated a high correlation, and generally the dominant pattern of measured permeability is depicted with a correlation coefficient of approximately 0.70. Prediction of Reservoir Quality To develop a reservoir quality for Farrud reservoir, unsupervised ANN with K-means Clustering

T

Algorithm is applied to categorize the Farrud sediments into four categories (good, medium,

IP

poor, and very poor). Table 8 presents the 500 vectors used in building the classification data base and the contribution of each class with average match over 90%. As shown in Table 8,

CR

permeability and calculated Sw reported a relative importance of 100 and 90 respectively, and appear to be the key parameters that distinctly determine the category to which a particular

US

portion within the reservoir is assigned. Alternatively, the volume of shale reported a moderate relative importance of 73, but porosity carried relatively insignificant influence on reservoir

AN

quality classification with 39 relative importance. Using a match technique, the data points throughout Farrud pay zone are assigned a reservoir quality using the criteria presented in Table

M

8, and the final reservoir quality volume is presented in Figure 16. In general, the majority of drilled wells tap the high quality reservoir (yellow color) with few wells drilled in the medium

ED

quality (light green) area. The high and medium quality regions in Figures 16 are outlined with a polygon to define the reservoir extension in Farrud sediments. This polygon defining Farrud

PT

reservoir geobody closely matches that display in Figure 1-upper. Generally, the good reservoir quality regions present 26.4% of the total classified vectors and

CE

indicate the best match of the characterization parameters, where low water saturation (30%) and volume of shale with moderate porosity and permeability are encountered. Together with the

AC

good reservoir quality, the medium quality constitute the dominant categories that formulate Farrud hydrocarbon pool and the outlined geobody presented in Figure 16. Alternatively, the very poor class encompasses 5% of the studied reservoir volume where Sw fall close to 80%, while the other parameters fall below the predefined thresholds outlined in Table 8. The majority of the studied area (46%) fall within the poor reservoir quality with significantly high Sw (76%) , while the other reservoir properties report values matchable to those assigned to the good category. Finally, the medium reservoir category represents 22% of the studied reservoir volume and encompasses all regions with median Sw of 70% with comparable characteristics to the other categories (Table 8). 12

ACCEPTED MANUSCRIPT Conclusions This work presents an innovative technique that aims at predicting the quality of a petroleum reservoir using Neural Networks analysis of seismic, well logs and core data. A supervised Probabilistic Neural Network has been deployed to predict several reservoir properties, once at a time, through training the PNN to determine the seismic attributes that accurately represent the

T

measured properties in well logs, such as shale volume, porosity, permeability and water

IP

saturation. These predicted reservoir properties have been used by unsupervised ANN with Kmeans Clustering Algorithm to determine the reservoir quality throughout the area covered by

CR

seismic data. Four grades of reservoir quality; good, medium poor and very poor, have been determined, and their spatial distribution is displayed. The good reservoir category is

US

characterized by low water saturation and shale volume with moderate porosity and permeability. These results are valuable for optimum reservoir management and well placement

AN

that not only maximize reservoir profitability through the development of field production

M

schemes but also minimize uncertainty in drilling, production, injection and modeling processes.

ED

Recommendation

Future work is expected to target testing other algorithms of artificial intelligence to determine

PT

the algorithm(s) that best fit integration of seismic and well log data sets. In addition, the applications of supervised algorithms should be extended to classify the reservoir quality, and

CE

uncertainty analysis should be implemented to evaluate the reservoir classification performance.

AC

Acknowledgement

The authors appreciate very much the kind approval of the National Petroleum Corporation in Libya for providing the data set used in this study and approving to publish the results of this work. The proof reading by Prof. Dr. Abdel Zaher Abu Zaid, Faculty of Engineering-Cairo University is highly appreciated.

References

13

ACCEPTED MANUSCRIPT Abdulaziz A.M., Dahab A.A. 1, Najmuddin M.Y., 2017. Well Log Petrophysics of the Cretaceous Pay Zones in Zubair Field, Basrah, South Iraq. Open Journal of Geology, 7 (10): 1552-1568. https://doi.org/10.4236/ojg.2017.710104 Ahmadi M.A., 2014. Prediction Breakthrough Time of Water Coning in the Fractured Reservoirs by Implementing Low Parameter Support Vector Machine Approach, Fuel 117 (2014) 579–589

T

Ahmadi M.A., 2015. Connectionist Approach Estimates Gas-Oil Relative Permeability in Petroleum Reservoirs: Application to Reservoir Simulation. Fuel, 140: 429-439 (2015), https://doi.org/10.1016/j.fuel.2014.09.058

CR

IP

Ahmadi M. A., 2015a. Developing a Robust Surrogate Model of Chemical Flooding Based on the Artificial Neural Network for Enhanced Oil Recovery Implications, Mathematical Problems in Engineering, Volume 2015 (2015), Article ID 706897, 1-9. http://dx.doi.org/10.1155/2015/706897

US

Ahmadi M.A., 2016. Toward reliable model for prediction Drilling Fluid Density at wellbore conditions: A LSSVM model, Neurocomputing, Volume 211, 26 October 2016, Pages 143-149. http://dx.doi.org/10.1016/j.neucom.2016.01.106i

AN

Ahmadi M.A. and Ebadi M., 2014. Evolving Smart Approach for Determination Dew Point Pressure through Condensate Gas Reservoirs, Fuel (2014)117 Part B, pp. 1074-1084

ED

M

Ahmadi M. A. and Mahmoudi B., 2016. Development of robust model to estimate gas–oil interfacial tension using least square support vector machine: Experimental and modeling study, The Journal of Supercritical Fluids, 107 (2016) 122–128. DOI: 10.1016/j.supflu.2015.08.012

CE

PT

Ahmadi, M.A., Ahmadi, M.R., Hosseini, M. and Ebadi, M., 2014. Connectionist model predicts the porosity and permeability of petroleum reservoirs by means of petro-physical logs: Application of artificial intelligence. Journal of Petroleum Science and Engineering. 123. 183200. 10.1016/j.petrol.2014.08.026.

AC

Ahmadi, M, Bahadori, A and Shadizadeh, SR 2015b, A Rigorous Model to Predict the Amount of Dissolved Calcium Carbonate Concentration through Oil Field Brines: Side Effect of Pressure and Temperature, Fuel 139 (2015) 154–159 Ahmadi M. A., Ebadi M., Samadi A. and Siuki M.Z., 2014b. Phase Equilibrium Modeling of Clathrate Hydrates of Carbon Dioxide + 1,4-Dioxine Using Intelligent Approaches, Journal of Dispersion Science and Technology, Volume: 36, Issue: 02, pages 236 – 244. https://doi.org/10.1080/01932691.2014.904792 Ahmadi M. A., Masoumi M. and Askarinezhad R., 2014a. Evolving Connectionist Model to Monitor Efficiency of the In-situ Combustion Process: Application to Heavy Oil Recovery. Journal of Energy Technology, 2 (9-10): 811–818 (2014) https://doi.org/10.1002/ente.201402043

14

ACCEPTED MANUSCRIPT Ahmadi M.A., Masoumi M., Askarinezhad R., 2015c. Evolving Smart Model to Predict Combustion Front Velocity throughout In-Situ Combustion Process Employment, Energy Technol. 2015, 3, 128 – 135. https://doi.org/10.1002/ente.201402104 Ahmadi M.A., Soleimani R., Lee M., Kashiwao T., Bahadori A., 2015. Determination of oil well production performance using artificial neural network (ANN) linked to the particle swarm optimization (PSO) tool, Petroleum, (2015) 1(2): 118-132.118e13210.1016/j.petlm.2015.06.004

IP

T

Ahmadi M. A., Zahedzadeh M. Shadizadeh S.R. Abbassi R., 2015a. Connectionist Model for Predicting Minimum Gas Miscibility Pressure: Application to Gas Injection Process, Fuel 148 (2015) 202–211

CR

AL-Harouge, 1980, “field information internal report”, AL-Harouge Oil Company, Tripoli, Libya.

US

Al-Kaabi, A.U. 1993. Using Artificial Neural Networks to Identify the Well Test Interpretation Model. Society of Petroleum Engineers, 1993. SPE-20332-PA.

AN

Amin S. M. and Sadeghnejad P.S., 2017. Fractal parameters and well-logs investigation using automated well-to-well correlation. Computers & Geosciences, 103: 59-69. https://doi.org/10.1016/j.cageo.2017.03.004

M

Anifowose F., Abdulazeez A., Al-Shuhail A.A., Schmitt D.P. (2013). “Improved Permeability Prediction from Seismic and Log Data using Artificial Intelligence Techniques”, SPE Paper 164465 Presented at the 18th Middle East Oil & Gas Show and Conference (MEOS) held in Manama, Bahrain on 10 – 13 March, 2013.

PT

ED

Anifowose F., Adeniye S. and Abdulazeez A., 2016, "Integrating Seismic and Log Data for Improved Petroleum Reservoir Properties Estimation Using Non-Linear Feature-Selection Based Hybrid Computational Intelligence Models”, Journal of Petroleum Science and Engineering, Volume 145, pp. 230 - 237, September 2016.

CE

Archie, G.E. 1942. “The Electrical Resistivity Log as an Aid in Determining Some Reservoir Characteristics”. Trans.

AC

Bertozzi, W., Ellis, D.V., and Wahl, J.S., 1981, "The Physical Foundations of Formation Lithology Logging with Gamma Rays", Geophysics 46 (10), pp-1439-1455 Burgos P., Madejón E., Pérez-de-Mora A., Cabrera F., 2008. Horizontal and vertical variability of soil properties in a trace element contaminated area. International Journal of Applied Earth Observation and Geoinformation, 10(1): 11-25. https://doi.org/10.1016/j.jag.2007.04.001 Bousaka J., O’Donovan A. (2000), “Exposing the 4D seismic time-lapse signal imbedded the Foinaven active reservoir management project”, OTC paper 12097. Castagna J.P., Sun S. (2006), “Comparison of spectral decomposition methods”, First Break 24:75–79. Dobrin, Milton B., and Carl H. Savit. 1988. “Introduction to Geophysical Prospecting”. Fourth. New York: McGraw-Hill Companies.

15

ACCEPTED MANUSCRIPT Ebrom D. (2004), “The low-frequency gas shadow on seismic sections”, Lead Edge 23:772. Farfour M., Yoon W.J., Kim J., 2015. Seismic attributes and acoustic impedance inversion in interpretation of complex hydrocarbon reservoirs. Journal of Applied Geophysics, 114: 68-80. https://doi.org/10.1016/j.jappgeo.2015.01.008 Fetter C.W., 2001. Applied Hydrogeology 4th edition. Prentice Hall, Upper Saddle River, NJ 07458, ISBN 0-13-088239-9, pp.598.

IP

T

Gholami R., Moradzadeh A., Maleki S., Amiri S., Hanachi J., 2014. Applications of artificial intelligence methods in prediction of permeability in hydrocarbon reservoirs. Journal of Petroleum Science and Engineering, 122: 643-656. https://doi.org/10.1016/j.petrol.2014.09.007

CR

Habib M.A., Abdulaziz A.M. and Dahab A.A., 2016. Improving history match using artificial neural networks. International Journal of Petroleum Engineering, 2(4): 302–319. https://doi.org/10.1504/IJPE.2016.084121

US

Hampson, D., Russell, B., 2006, “The old and the new in seismic inversion”, Telus Convention Centre, Calgary, December 18, 2006.

AN

Hampson, D., Russell, B., and Bankhead, B., 2005, “Simultaneous inversion of pre-stack seismic data”: Ann. Mtg. Abstracts, Society of Exploration Geophysicists.

M

Hampson, D.P., Schuelke, J.S., and J.A. Quirein, 2001, “Use of multiattribute transforms to predict log properties from seismic data”, Geophysics, Vol. 66, No. 1, pp. 220-236, 2001.

ED

He W., Anderson R.N., Boulanger A,. Yu-Chiung Teng, Xu L., Neal R., Meadow B. (1997), “4D seismic The fourth dimension in reservoir management. Part 4 -Inversion of 4-D seismic changes to find bypassed pay”, World Oil Journal, Identifier: 536483, TX (United States) 199707-01.

PT

Hingle, A.T., 1959. The use of logs in exploration problems. Paper presented at the SEG 29ht. intl. annual meeting, Los Angles, Nov., 1959.

CE

Hubbert M.K., 1956. Darcy law and the field equations of flow of underground fluids. Transactions, American Institute of Mining and Metallurgical Engineers, 204:222-39.

AC

Ibrahim M.M. Abdulaziz A.M. Fattah K.A., 2017. STOIIP Validation for a Heterogeneous Multi Layered Reservoir of a Mature Field Using an Integrated 3D Geo- Cellular Dynamic Model. Egyptian Journal of Petroleum, in press. https://doi.org/10.1016/j.ejpe.2018.01.004 Krief M, Garar J, Stellingwerff J, Ventre J (1990),”A petrophysical interpretation using the velocities of P and S waves (fullwaveform sonic)”. Log Anal 31:355–369. Moosavi N. and Mokhtari M., 2016. Application of S-Transform and Simultaneous Seismic Inversion to Interpret Bright Spots. J Pet Environ Biotechnol 7:311. doi: 10.4172/21577463.1000311 Nanda C., 2016 “Seismic Data Interpretation and Evaluation for Hydrocarbon Exploration and Production A “Practitioner’s Guide”, Cuttack, Odisha, India2016.

16

ACCEPTED MANUSCRIPT Nguyen V.T., Hwang S., Jang S., Hoang N.D., Pham H.G., 2018. Well path design by integrating the analysis of wireline logs and the interpretation of seismic data for a fractured basement reservoir in Cuu Long Basin, Viet Nam. Marine and Petroleum Geology, 93: 315-330. https://doi.org/10.1016/j.marpetgeo.2017.12.011 Partyka G, Gridley JM, Lopez J (1999) “Interpretational applications of spectral decomposition in reservoir characterization “. Lead Edge 18:353–360.

T

Pickett G.R., 1966. A review of current techniques for determination of water saturation from logs. JPT (Nov. 1966) 1425-1433.

CR

IP

Ramirez A., White D., Hao Y., Dyer K., Johnson J., 2013.Estimating reservoir permeabilities using the seismic response to CO2 injection and stochastic inversion, International Journal of Greenhouse Gas Control, 16(1): S146-S159, https://doi.org/10.1016/j.ijggc.2012.11.031 . Rijks E.J.H., Jauffred J.C.E.M. (1991), “Attribute extraction: an important application in any detailed 3-D interpretation study”, Lead Edge 10:11–19.

US

Robertson D.J., Nogami H.H. (1984), “Complex seismic trace analysis of thin beds” Geophysics 49:344–352.

AN

Russell B. 2014, “Seismic Reservoir Characterization and Pre-stack Inversion in Resource Shale Plays”, Calgary, Canada, Posted October 27, 2014.

M

Russell, B. and Hampson, D., 1991, “Comparison of poststack inversion method”: 61st Ann. Internat. Mtg., Soc. Expl. Geophys., Expanded Abstracts, 876-878.

ED

Saadu Y.K. and Nwankwo C.N., 2017. Petrophysical evaluation and volumetric estimation within Central swamp depobelt, Niger Delta, using 3-D seismic and well logs, Egyptian Journal of Petroleum, in print. https://doi.org/10.1016/j.ejpe.2017.08.004

CE

PT

Sarhan M.A., Basal A.M.K., Ibrahim I.M., 2017. Integration of seismic interpretation and well logging analysis of Abu Roash D Member, Gindi Basin, Egypt: Implication for detecting and evaluating fractured carbonate reservoirs, Journal of African Earth Sciences, 135: 1-13. https://doi.org/10.1016/j.jafrearsci.2017.08.010 Shepherd R.G., 1989. Correlations of permeability and grainsize. Ground Water 27, no. 5:63338.

AC

Sigismondi M.E., Soldo J.C. (2003), “Curvature attributes and seismic interpretation: case studies from Argentina basins”, Lead Edge 22(11):1122–1126. Staples R., Stammeijer J., Jones S., Brain J., Smit F., Hatchell P. (2006), “Time-lapse seismic monitoring-expanding applications”, CSPG-CSEG-CWLS Convention. Calgary, Canada 181– 189. Stieber S.J. 1970. “Pulsed neutron capture log evaluation - Louisiana Gulf Coast”. SPE Fall Meeting AIME, Conference paper: 2961-MS. Taner M. T., J. S. Schuelke, R.O. “Doherty, and E. Baysal, 1994: “Seismic attributes revisited”, 64th Annual International Meeting, SEG, Expanded Abstracts, 1104–1106.

17

ACCEPTED MANUSCRIPT Wang H., Ma J., Li L., Jia L., Tan M., Cui S., Zhang Y., Qu Z., 2017. Time-lapse Seismic Analysis for Gao89 Area of CO2-EOR Project in SINOPEC Shengli Oilfield, China. Energy Procedia, 114: 3980-3988, https://doi.org/10.1016/j.egypro.2017.03.1530. Wang Z., 2001, “Y2K tutorial fundamentals of seismic rock physics” Geophysics 66:398–412. Zhangxin, Chen 2007, Reservoir simulation: Mathematical Techniques in oil recovery, CBMSNSF Regional Conference Series in Applied Mathematics, ,the society for industrial and applied mathematics, 3600 Market Street, 6th floor, Philadelphia, PA, 19104-2688, 208 pp.

AN

US

CR

IP

T

Xu L., Anderson R.N., Boulanger A., Yu-Chiung Teng, He W., Neal R., Meadow B. 1997,“4-D reservoir monitoring, the business driver”. World Oil’s 4-D seismic hand book, pp23–28.

Tables and Figures

M

Tables

ED

Table 1: Well data of AL-Ghani Field used for data analysis.

CE

PT

Table 2 Equation list applied in petrophysical properties calculations for Farrud reservoir. The variables are as follows; Ip is P impedance (Kg/m2.s), Vp is P wave velocity (m/s), ρ is the density (Kg/m3), ϕND is the Neutron-Density porosity (%),ϕ N is the Neutron Porosity (%), ϕ D is the Density porosity (%), Vsh is the volume of shale (%), IGR Shale index, Sw is the water saturation (%), formation water resistivity(Ohm.m), Rt is the true formation resistivity (Ohm.m), a, m and n are Archie parameters, and ϕ is the total porosity.

AC

Table 3: The wells use for PNN training process and results verification in Farrud reservoir. Table 4. A brief statistical analysis for the calculated petrophysical calculations in Farrud reservoir. Table 5. The correlation percentage of well tying to seismic data obtained for all studied wells using a statistical zero phase wavelet. Table 6. Seismic attributes used in reservoir properties prediction with PNN for Farrud reservoir. Table 7. Average training and verification errors calculated in permeability prediction. Table 8. Number of vectors, property relative importance, and center property value used in reservoir quality prediction by K-means Clustering algorithm. 18

ACCEPTED MANUSCRIPT

Figures Figure 1. Location map of Al-Gahni Field (upper) with a base map (lower) showing well sites relative to the seismic survey inlines and crosslines (AL-Harouge, 1980). Figure 2. A composite stratigraphic column for west Sert Basin (AL-Harouge, 1980).

IP

T

Figure 3. A schematic workflow for integrating well logs and seismic data for property prediction using ANN in Farrud reservoir.

CR

Figure 4. The correlation of the inverted trace (red trace) with the original acoustic impedance log (blue trace) calculated at RRR04 well.

US

Figure 5. Training and verification error analysis calculated in all wells for the four properties predicted in this study.

AN

Figure 6. The predicted total porosity cube of Al-Ghani field from seismic data using PNN and five seismic attributes.

M

Figure 7. Validation of the total porosity prediction at Farrud reservoir using RRR08 and RRR33 wells.

ED

Figure 8. The predicted cube for the volume of shale at Al-Ghani field from seismic data using PNN and five seismic attributes.

PT

Figure 9. Validation of the volume of shale prediction at Farrud reservoir using RRR59 and RRR30 wells.

CE

Figure 10. The predicted cube for water saturation at Al-Ghani field from seismic data using PNN and seven seismic attributes.

AC

Figure 11. Validation of water saturation prediction at Farrud reservoir using RRR08 and RRR71 wells. Figure 12. A cross sectional view from the converted water saturation cube at the proximity of RRR71 well. Figure 13. Numerous wells with their water cut presented on a time slice at the converted water saturation cube of Farrud reservoir. Figure 14. The predicted cube for permeability at Al-Ghani field from seismic data using PNN and 25 seismic attributes. Figure 15. Validation of permeability prediction at Farrud reservoir using RRR10 well. Figure 16. Reservoir quality distributed through Al-Ghani field with Farrud reservoir geobody outlined. 19

AC

CE

PT

ED

M

AN

US

CR

IP

T

ACCEPTED MANUSCRIPT

20

ACCEPTED MANUSCRIPT

Y (ft.)

Available data

6,160 6,210 6,130 6,090 6,110 6,060 5,965 6,100 6,040 5985 5995 6,000 6,200 6100 8300 6,205 6,250 6,050 6,214

740,393 739,273 739,335 739,742 738,824 739,282 739,144 739,734 739,070 739,042 739,702 739,611 738,911 740,352 738,065 740,279 741,396 741,470 740,503

3,206,254 3,206,656 3,208,370 3,207,035 3,208,221 3,207,568 3,209,225 3,206,508 3,209,809 3,210,514 3,209,508 3,210,486 3,207,280 3,209,879 3,209,807 3,207,395 3,207,783 3,208,781 3,207,984

Check shot ,sonic, resistivity ,gamma ray, neutron porosity logs Density ,sonic, resistivity ,gamma ray, and neutron porosity logs Density ,sonic, resistivity ,gamma ray, and neutron porosity logs Density ,sonic, resistivity ,gamma ray, and neutron porosity logs Density ,sonic, resistivity ,gamma ray, and neutron porosity logs Density ,sonic, resistivity ,gamma ray, and neutron porosity logs Density ,sonic, resistivity ,gamma ray, and neutron porosity logs Density ,sonic, resistivity ,gamma ray, and neutron porosity logs Density ,sonic, resistivity ,gamma ray, and neutron porosity logs Density ,sonic, resistivity ,gamma ray, and neutron porosity logs Density ,sonic, resistivity ,gamma ray, and neutron porosity logs Density ,sonic, resistivity ,gamma ray, and neutron porosity logs Density ,sonic, resistivity ,gamma ray, and neutron porosity logs Density ,sonic, resistivity ,gamma ray, and neutron porosity logs Density ,sonic, resistivity ,gamma ray, and neutron porosity logs Density ,sonic, resistivity ,gamma ray, and neutron porosity logs Density ,sonic, resistivity ,gamma ray, and neutron porosity logs Density ,sonic, resistivity ,gamma ray, and neutron porosity logs Density ,sonic, resistivity ,gamma ray, and neutron porosity logs

PT

ED

M

AN

US

CR

IP

T

X (ft.)

CE

RRR01 RRR02 RRR04 RRR07 RRR08 RRR09 RRR10 RRR20 RRR21 RRR26 RRR30 RRR31 RRR33 RRR37 RRR40 RRR42 RRR47 RRR59 RRR71

TD(ft.)

AC

Well name

21

ACCEPTED MANUSCRIPT

Property

Equation

Reference

𝐼𝑝 = 𝑉𝑝 × 𝜌

P-impedance

Hampson and Russell, 2006

∅𝑁𝐷 =

Ø𝑁 + Ø𝐷 2

Bertozzi et al., 1981

Volume of Shale

𝑉𝑠ℎ =

0.5 × 𝐼𝐺𝑅 1.5 − 𝐼𝐺𝑅

Stieber, 1970

Water saturation

𝑆𝑤 𝑛 =

IP

CR

Archie, 1942

US

Measured in core data

AC

CE

PT

ED

M

AN

Absolute permeability

𝑎 𝑅𝑤 × 𝑚 ∅ 𝑅𝑡

T

Porosity ∅

22

Al-Harouge Co.

ACCEPTED MANUSCRIPT

Reservoir property

Training/verification wells

Testing wells

12 wells

RRR02 and RRR33

Volume of shale

8 wells

Water saturation

12 wells

Absolute permeability

2 wells

RRR59 and RRR30 RRR08, RRR42 and RRR71 RRR10

AC

CE

PT

ED

M

AN

US

CR

IP

T

Porosity

23

ACCEPTED MANUSCRIPT

RRR 21 12 22 17 0 4 2 16 100 58

RRR 28 0.02 20 10 0 1.6 0.8 19 61 40

RRR 31 15 25 20 0 5.3 2.6 16 92 54

US AN M 24

RRR 33 0.03 19 9.5 0 12 6 17 96 56

T

RRR 20 1 25 12 0 2.5 1.2 15 100 57

IP

RRR 10 12 22 17 0 7.4 3.7 17 85 51

CR

RRR 09 16 20 18 1.8 5.2 3.5 24 59 41

ED

Sw

RRR 04 13 20 16 1 3.6 2.3 12 100 56

PT

Vshale

Min Max Avg Min Max Avg Min Max Avg

RRR 02 1.9 19 10 0 2.7 1.35 24 62 43

CE

Total porosity

Value

AC

Property

RRR 42 0.8 22 11.4 0 9 4.5 10 40 25

RRR 47 4.9 20 12 0 10 5 56 100 78

RRR 71 8 24 16 0 11 5.5 24 98 61

ACCEPTED MANUSCRIPT

RRR01 RRR02 RRR04 RRR07

Training + Verification Training + Testing Training + Verification Training + Verification

71% 69% 93% 92%

Farrud Formation Top Bottom (ft) (ft) 5959 6042 6002 6086 5897 6098 5922 6011

RRR08 RRR09 RRR10 RRR20 RRR21 RRR26 RRR30 RRR31 RRR33 RRR37 RRR40 RRR42 RRR47 RRR59 RRR71

Testing Training + Verification Training + Testing Training + Verification Training + Verification Training + Verification Testing Training + Verification Training + Testing Training + Verification Training + Verification Training + Testing Verification Testing Testing

78% 95% 97% 97% 93% 82% 75% 80% 91% 97% 88% 90% 87% 96% 75%

5903 5884 5849 5965 5896 5747 5770 5765 5992 4740 6870 5930 5888 5892 5905

M

ED

PT CE AC

25

US

CR

IP

T

Well tie correlation

Well application(s)

AN

Well

5998 5994 5951 6050 6000 5985 5995 5950 6081 7500 7610 6014 5979 6050 6012

ACCEPTED MANUSCRIPT

Quadrature trace Filter 25/30-35/40 Integrate Derivative col. inversion_ Zp sqrt Raw seismic Amplitude weighted cosine phase Filter 35/40-45/50 Filter 45/50-55/60 Second deriv. instant. amplitude Second derivative cosine instant. Dominant frequency Average frequency Constant (W0) Number of attributes

0.090583 0.117688 -0.530172 25.813349 5

0.012815 -419.925873 4

M

ED

PT

AC

CE

US

0.001236 -0.039287 -

AN

Filter 55/60-65/70 Amplitude weighted phase Instantaneous phase Filter 5/10-15/20 Derivative instantaneous amplitude Instantaneous frequency Filter 15/20-25/30 Colored_inversion_Zp _Sqrt Integrated absolute amplitude Apparent polarity Amplitude weighted frequency Amplitude envelope

CR

Porosity

Properties Water saturation -0.000054 0.002297 -0.002363

T

Reservoir Shale volume -0.161368 -0.000001 -0.005122 -

IP

Attributes

26

-0.004562 -0.003471 0.011188 0.398457 6

Permeability 0.074753 -0.001255 -0.039006 -0.543898 1.510101 -0.432515 0.462885 -0.007575 -0.144950 -0.050841 0.019168 -0.624259 0.692357 0.468366 -0.108225 0.989810 -0.871219 0.341634 0.300227 0.140740 4.123433 -0.540609 0.526309 -4.773460 278.488586 24

AC

CE

PT

ED

M

AN

US

CR

IP

T

ACCEPTED MANUSCRIPT

27

ACCEPTED MANUSCRIPT Training Error (%)

Verification Error (%)

Instantaneous phase Amplitude weighted phase Derivative instant. amplitude Filter 5/10-15/20 Filter 15/20-25/30 Filter 25/30-35/40 Filter 45/50-55/60 Filter 35/40-45/50 Filter 55/60-65/70 Colored inversion_ Zp_Sqrt Deriv. colored inversion_ Zp_Sqrt Instantaneous frequency Integrated absolute amplitude Amplitude weighted frequency Amplitude envelope Raw seismic Amplitude weighted cosine phase Average frequency Dominant frequency Quadrature trace Integrate Second deriv. instantan. amplitude Apparent polarity Second deriv. cosine instantaneous

6.23 7.5 5.32 4.52 2.39 4.63 3.65 4.21 5.18 6.28 5.26 4.23 5.89 6.12 6.18 2.93 3.57 2.36 1.32 5.14 2.36 2.66 3.50 3.68

8.89 8.35 8.17 7.88 7.75 7.62 6.62 6.61 6.16 7.72 7.69 7.57 7.50 7.39 7.08 7.03 6.69 6.63 6.61 6.58 6.48 6.32 6.25 6.19

AC

CE

PT

ED

M

AN

US

CR

IP

T

Seismic attribute

28

ACCEPTED MANUSCRIPT

Vectors classified: 500 Class 1 vectors: 132. Class 2 vectors: 111. Class 3 vectors: 231. Class 4 vectors: 26.

Average match: Average match: Average match: Average match:

0.894612 0.901632 0.910402 0.867648

IP

Center 3 0.10627446 0.12701049 1.02897203 0.76398557

CR

Center 2 0.17805749 0.12853031 1.00307187 0.71737432

US

Center 1 0.11459582 0.13763399 0.96975732 0.30206078

AC

CE

PT

ED

M

AN

[edit por] [editperm] [predicted_Vshale] [predicted_watersaturation]

T

Input mode relative importance: 38.3 ( (edit por) ) 100.0 ( (editperm) ) 73.5 ( (predicted_Vshale) ) 90.7 ( (predicted_watersaturation) )

29

Center 4 0.13517642 0.17154208 1.82944405 0.80198681

ACCEPTED MANUSCRIPT Highlights

Dear respected Editor and reviewers,

IP

T

We do not find the suitable words that express our feelings towards the valuable comments we received to upgrade our manuscript. We really appreciate your time and effort and we addressed each of your comments very carefully and this was the reason for the delayed response.

CR

Thank you again and we wish you all the best. Sincerely,

US

Abdulaziz Mohamed and Hamida Mahdi

AC

CE

PT

ED

M

AN

NB. In the text file, the first edited parts of the test is marked in red while the second edited parts are marked in blue.

30

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Figure 16

Figure 17

Figure 18