SPE-230340-MS

Data-Driven Water Injection Planning: Leveraging DBSCAN and K-Means for

Digital Petrophysics and Reservoir Stress Characterization

Sarthak Chauhan and Shaktim Dutta, SLB, Mumbai, India; Arvind Kumar, SLB, Abu Dhabi, United Arab Emirates;
Dario Ernesto Solohaga, SLB, London, United Kingdom; Tety Benedicta Wydiabhakti, SLB, Abu Dhabi, United Arab
Emirates; Saket Shrivastava, Vaibhav Deshpande, and Velerian Sanjao Lopes, SLB, Mumbai, India; Bhaben Deori,
SLB, Houston, United States of America

Copyright 2025, Society of Petroleum Engineers DOI 10.2118/230340-MS

This paper was prepared for presentation at the SPE Caspian Technical Conference and Exhibition scheduled to be held in Baku, Azerbaijan, 25 – 27 November 2025.

This paper was selected for presentation by an SPE program committee following review of information contained in an abstract submitted by the author(s). Contents
of the paper have not been reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not necessarily reflect
any position of the Society of Petroleum Engineers, its officers, or members. Electronic reproduction, distribution, or storage of any part of this paper without the written
consent of the Society of Petroleum Engineers is prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300 words; illustrations may
not be copied. The abstract must contain conspicuous acknowledgment of SPE copyright.

Abstract
Water injection programs play a crucial role in secondary and tertiary recovery stages of a field. The
success of water injection strategy relies heavily on efficient injection of water into the desired zone
and hence relying heavily on accurate reservoir injectivity determination. These abstract aims to address
the complexities during injectivity modeling of the heterogenous sand layers. The primary objective is
to enhance reservoir characterization in such geological contexts by estimating the layer injectivity for
monitoring water injection. The proposed method involves a fusion of advanced petrophysical logs. Use of
various machine learning (ML) methods have emerged recently in context of petrophysical modeling like
facies characterization. In this paper, we have discussed various ML clustering techniques like K-means
clustering and density-based clustering methods for facies classification. Density-based spatial clustering
of applications with noise (DBSCAN) method has been used to perform the facies classification, required
as a multi-layer model for pressure transient analysis in order to estimate the permeability of each facies.
A comprehensive workflow is developed, integrating basic and specialized petrophysical logs, including
lithology logs, nuclear magnetic resonance logs, and petrophysical elemental analysis results. This synergy
is further enriched by the incorporation of borehole image logs. Artificial neural networks are strategically
employed for rock typing and facies model creation. This process facilitates the identification of rock types
or hydraulic flow units for reservoir quality (RQ), further combined with the completion quality (CQ), forms
the basis for defining multiple layers in reservoir models.
Another key parameter to have during planning a water injection strategy is the knowledge of insitu
stresses. This determines the safe injection pressure limits to avoid flow of injected water to unwanted
reservoir zone. A stress test workflow has been demonstrated using various stress analysis techniques such
as G-function and SQRT plot techniques to estimate the minimum horizontal stress, commonly known as
the fracture gradient in a vertical well scenario.
The abstract presents a fusion of petrophysical analysis with water injection fall-off transient
methodologies, introducing a sophisticated multi-layer reservoir modeling technique. The innovative aspect

Downloaded from [Link] by Schlumberger Oilfield UK Plc user on 26 December 2025

 2 SPE-230340-MS

lies in the seamless integration of advanced petrophysical logs, and various machine learning based
facies modeling, providing a comprehensive approach to reservoir characterization. The methodology
offers a nuanced understanding of permeability, skin, and permeability anisotropy in thin laminated or
heterogeneous sands, contributing significantly to improved reservoir model quality and vertical resolution.
Estimation of safe water injection pressure complements the overall water injection strategy.

Introduction
Injectivity is one of the key parameters in the planning process of water injection projects for mature
reservoir management plans. Water is injected into both water zone and oil zone for distinct reasons. Water
injection into the water zone below the overlying oil zone leads to supplement the reservoir energy in the
water drive system. On the other hand, water injection into oil zone leads to water flooding and displacement
of oil towards the producer wells in a particular pattern of the injection program. Various analytical methods
originated in 90s, towards solving for single phase low compressible fluid flow as well as two phase flow
of immiscible fluids. Injectivity test became important to understand and plan the water injection plan in
ahead of the operations for increasing the efficiency of reservoir management. Injectivity refers to the ease
of fluid flow through the defined reservoir intervals (Eltoum et al., 2021), which is directly correlated to the
conductivity of formation given by product of permeability and thickness. In more specific terms related
to water injection, the water injectivity is given by water injection rate over differential pressure between
a producer and injector (Satter et al., 2016). An incompatibility between injected water and formation
water can lead to reduction in the water injection. Civan (2007) defined injectivity ratio, α(t) as the ratio
between the instantaneous (II) and the initial injectivity indices (IIo). Bedrikovetsky (2001) illustrated the
concept of impedance index in details, as the reciprocal of injectivity index. Abbaszadeh (1989) discussed
an approximate solution of injectivity in a single layer reservoir for vertical wells. Banerjee (1988) discussed
the injection fall off solutions in both homogenous and heterogenous reservoirs. With the increasing interest
in horizontal wells, the injectivity solutions were developed. Peres (2003) showcased an analytical solution
for injectivity estimation for highly deviated and horizontal wells. Raghavan (1974) considered the response
of sand zone thickness on the overall pressure response. Barreto (2011) showcased the analytical methods
of interpreting water injection tests for multilayered oil-bearing sand zones penetrated by a vertical well. He
also showcased the usage of partial differential equation systems for modeling the dual phase displacement
of oil by injected water. Sommerauer (2007) showcased the water injection modeling across the multilayer
reservoir zones that were fractured. This was implemented further for the pressure support during the
secondary recovery campaign in the champion southeast oilfield (Sommerauer et al., 2007). Elkatatny
(2014) demonstrated a water re-injection simulator to study the single layer and multi-layer injection results.
He considered both the injectivity or conductivity and minimum insitu stress as parameters in the simulator.
Fakhruzan (2017) explored the intricacies of injection rate control across multilayer reservoirs through the
innovative use of injection mandrel. Abdelmoneim (2018) showcased the impact of injecting low salinity
water on the overall sweep efficiency in the multilayer reservoirs.

Layer permeability and insitu stress characterization

Injection fall-off pressure transient analysis methods have been used to estimate the minimum horizontal
stress, breakdown pressure and other parameters such as instantaneous shut-in pressure. Boughrara (2007)
showcased the practical approaches of interpreting the injection fall-off tests performed in horizontal wells.
Geomechanical modeling is key to design the injection fall-off operations since it provides an insight into the
expected insitu stress systems. It also helps to design the toolstring with sufficient injection packer pressure
differential rating. Various authors have showcased the usage of geomechanical modeling in estimating the
insitu stresses in both isotropic and anisotropic formations. Patro (2018) showcased the use of integrated
stress analysis using acoustic measurements to determine the minimum and maximum horizontal stresses

Downloaded from [Link] by Schlumberger Oilfield UK Plc user on 26 December 2025

 SPE-230340-MS 3

in the anisotropic formation intervals. Kumar (2023) illustrated an integrated approach taking acoustic
geomechanics and mini frac techniques to estimate the safe injection pressure for CO2 storage in CCS
projects. Wydiabhakti (2023) showed a workflow integrating the petrophysics, acoustics, image logs and
stress analysis workflows for stress estimation in CCS projects. Chatterjee (2018) showed geomechanical
analysis integrating the acoustic measurements with other petrophysical logs in unconventional formations.
Kumar (2023) showcased a novel insitu stress determination for safe injection pressure in CCS formations.
Figure 1 above showcases the layer permeability characterization workflow. In this method, a
petrophysical evaluation of both the basic and advanced open hole logs is performed. This is followed by the
facies classification workflow using various methods based on the number of logs, level of heterogeneity,
type of geology etc. Nowadays the facies modeling is being performed using various available machine
learning and data analytics methods. In this paper, we have analysed K-means clustering method and density
based spatial clustering of applications with noise (DBSCAN) methods to perform facies classification.

Figure 1—a)Layer permeability characterization workflow- injection pressure < fracture

gradient b) Stress Testing (Micro Frac) workflow – injection pressure > fracture gradient

During an injection fall-off test, injection fluid (water based) is injected into the isolated reservoir interval
and then the injection is stopped for a falloff. It is to note that in this case, the injection pressure is
always below the fracture gradient. After the facies classification is performed, a multi-layer base reservoir
model is prepared, which works as an input to analyse the injection fall-off pressure transient analysis
(PTA). Injection fall-off analysis is performed to estimate horizontal permeability, vertical permeability,
skin of reservoir zone being tested. During a multi-layer PTA, individual reservoir layers are analysed in
conjunction with the total reservoir interval during the PTA modeling and regression stage. As a result, we
derive the multi-layer reservoir model with each layer having its own set of horizontal permeability, vertical
permeability, and skin. This helps in performing a relative evaluation of the injectivity of individual layers
within the reservoir zone and decide upon the active perforation interval.
During a stress test (micro fracturing) operation, the injection is carried out at pressures above the fracture
gradient. This causes a formation breakdown, followed by injection fall-off cycles. In this case, small
hydraulic fractures are created, propagated followed by fall-off. The injection fall-off cycle is analysed
using the standard plots like G-function and square root plot techniques. Instantaneous shut-in pressure

Downloaded from [Link] by Schlumberger Oilfield UK Plc user on 26 December 2025

 4 SPE-230340-MS

and minimum horizontal stress (fracture gradient in most cases) are estimated from the analysis, which are
used further to calibrate the one-dimensional mechanical earth model (MEM) stresses, which were earlier
derived from the open hole logs and drilling data.

Facies Modeling
Multilayer facies modeling approaches investigate the key differences in rock quality over a definite
reservoir interval by combining various lithology parameters such as permeability, porosity, sand contents,
volume of shale etc. Kumar (2023) showcased the use of artificial neural network in generating the rock
types / facies models across the heterogenous sandstone reservoirs, before going for modeling and regression
in the pressure transient analysis. This led to determination of reservoir parameters such as horizontal
permeability, vertical permeability, skin across each facies defined in the reservoir interval. This is key
to the water injection, EOR and water coning modeling projects, where layered properties are required to
effectively model the fluid flow. Kumar (2019) performed multilayer vertical interference test analysis,
where rock types were created based on image logs and petrophysical analysis. Kumar (2010) showcased
the usage of artificial neural network in permeability modeling from the openhole log data. Shi (2017)
used clustering-based technique to identify the log facies models for permeability estimations. Lu (2000)
demonstrated use of a stochastic method to characterize the heterogenous layered sand units. Duci (2013)
performed a multilayer facies modeling techniques to characterize the thin layers of the reservoir units.
Wang (2013) showcased the seismic inversion-based technique for facies modeling and geostatistics. Verma
(2021) devised a novel methodology for multi-well facies modeling using supervised and unsupervised
ML techniques. Mad Sahad (2022) used deep learning models to characterize the channel reservoirs, using
supervised learning.
Various well-log measurements provide invaluable insights into lithology, aiding in classifying
and understanding geological formations. This paper explores vital petrophysical parameters used for
petrophysical facies, encompassing aspects such as porosity, density, resistivity, and the presence of specific
minerals. Effective reservoir evaluation relies heavily on porosity measurements, with Neutron Porosity and
Total Porosity being pivotal as Neutron Porosity estimates hydrogen content, aiding in the identification
of porous formations. Bulk Density plays a central role in lithological assessments by offering insights
into the bulk composition of the formation. High-density readings often correlate with lithologies rich in
minerals like shale. Figure 2 below, showcases the open hole petrophysical logs that were used as input for
the clustering algorithms for facies classifications.
Additionally, the Volume of Shale is estimated through various logs, including Gamma Ray, Resistivity,
and Neutron Porosity, collectively contributing to a comprehensive understanding of shale content.
Resistivity is a key parameter for lithology classification, helping distinguish between formations with
different electrical conductivity. The determination of Water Saturation is vital for assessing fluid content
within the formation, impacting lithological interpretations, and identifying hydrocarbon-bearing zones.
P-wave (Vp) and S-wave (Vs) velocities provide insights into the elastic properties of rocks. Vp aids in
distinguishing lithologies based on compressional wave velocities, while Vs contributes to understanding
shear wave responses. These parameters play a crucial role in characterizing the mechanical behavior of
formations.

Downloaded from [Link] by Schlumberger Oilfield UK Plc user on 26 December 2025

 SPE-230340-MS 5

Figure 2—Petrophysical logs used as the input variable for the clustering algorithm

It is essential to characterize subsurface formations, and the clustering technique is an advantageous

instrument for identifying patterns and similarities in the borehole petrophysical data. The context of the
problem in discussion determines labels. When attempting to predict a feature based on the others, that
feature is usually the label. Based on information about those rocks, the label is used when attempting to
anticipate the sort of rock that is present in the lithology (for example, basalt or granite). The label is used
when attempting to anticipate whether a rock is mafic or felsic based on indications and other information.
Because the label is not stated in the study and there are no explicit instructions, unsupervised learning
takes place, in which the algorithm investigates the structure of the data and identifies relationships. The
clustering process aims to segregate groups with similar attributes and assign them into clusters. In order
to group comparable data points rather than have them fall into different clusters (a group of related data
points depending on their relation to surrounding data points), clustering involves splitting the unlabeled
data or data points into various clusters. By identifying clusters, one can gain insight into the heterogeneity
of subsurface formations, aiding in reservoir characterization and optimizing drilling strategies, well
placement, and resource recovery. The synergy between clustering techniques and petrophysical data is
evident in their combined application. Kumar (2021) utilized facies modeling to interpret the deep transient
testing pressure rate transient data. Kumar (2024) further used facies modeling to prepare the single well
reservoir model before simultaneous inversion of permeability.
K-Means Clustering method. The most common type of clustering is centroid-based clustering. Although
swift and efficient, it is sensitive to the initial parameters it delivers. These algorithms divide data points
according to multiple data centroids. A cluster develops from each data point proportional to its squared
distance from the centroid aiming to minimize intra-cluster variance and maximize inter-cluster variance.
The most popular kind of clustering is this one. K-means clustering is applied to petrophysical logs,
facilitating the identification of distinct lithological units and fluid content zones. Given a dataset X with n

Downloaded from [Link] by Schlumberger Oilfield UK Plc user on 26 December 2025

 6 SPE-230340-MS

data points xi in d-dimensional space, K-means aims to partition the data into K clusters. Mathematically,
each data point should be assigned to the cluster whose nearest centroid is given by equation 1 below:
(1)
where Ci is the cluster assignment for xi and tj is the centroid of cluster j. Update the centroid of each cluster
as give by equation 2:

(2)

where Cj is the set of data points assigned to cluster j.

The algorithm converges, and each data point belongs to a specific cluster. K-means performs. This
is the basis of the heterogeneous rock analysis (HRA) discussed by Kumar et al. (2023). This method
fuses core data with logs, especially in unconventional reservoirs. At its core, HRA segregates rocks
based on their intrinsic texture and composition attributes, as deciphered by log data. Its main goal is to
distinguish the rock's material attributes. Meanwhile, density-based clustering groups data into areas with
high concentrations of data points flanked by areas with low concentrations. The algorithm identifies and
classifies clusters based on how dense data points are distributed. The remarkable aspect of this is that
the clusters can take any shape and is not limited by expected conditions. These clustering techniques do
not attempt to assign outliers to clusters; therefore, they are ignored. A priori knowledge of the number of
clusters is required for classification in HRA and set was defined for the evaluation and comparison with
DBSCAN model, figure 3 shows the final classification results for the same using the petrophysical logs
that when through Principal Components Analysis (PCA) to rearrange the input data onto independent axes

Figure 3—Multiple facies were identified using K-Means technique implemented across the water-bearing injection interval.

Density-Based Spatial Clustering of Applications with Noise (DBSCAN). Density-Based Spatial

Clustering of Applications with Noise (DBSCAN) involves the delineation of petrophysical logs and

Downloaded from [Link] by Schlumberger Oilfield UK Plc user on 26 December 2025

 SPE-230340-MS 7

hydraulic flow units (HFUs) based on the density of elements in the dataset based on the unsupervised
pattern recognition algorithm. This concept is employed to categorize the reservoir into distinct
petrophysical facies. DBSCAN classification relies on data variability rather than the pre-established notions
of class representation, as the rock classes can be extrapolated to non-cored wells or segments through
unsupervised classification.
The reliability of identifying the rock classes stems from their reliance on multiple data patterns,
highlighting the random noise and poor borehole quality, as Handwerger et al. (2011). DBSCAN is quite
significant in finding clusters of arbitrary shapes, handling noise, and not requiring the number of clusters
to be specified beforehand, as shown below in a workflow diagram (figure 4).

Figure 4—Multiple facies DBSCAN workflow.

The learning process revolves around three elements: distance metrics, epsilon neighbourhoods and
Density Reachability, which are the main outlines of finding potential neighbours. Choosing the right
distance metric (Euclidean distance, Manhattan distance or Cosine similarity) is crucial for accurately
capturing the data's spatial relationships. Working with well logs the "reach" of an element, encompassing
potential neighbours which are confined by the area within a radius of Epsilon from the element, and if the
element satisfies the minimum neighbours within the radius and lies within the Epsilon neighbourhood of the
neighbouring element the density reachability is met, and allows DBSCAN to grow clusters by connecting
elements and their accessible neighbours, forming densely occupied regions.
Initially, the approach utilized exploratory data analysis to visualize and do a thorough statistical analysis
of the petrophysical logs. Figure 5 shows the initial cross-plot distribution generated using the logs; the plot
disturbs the data among itself (univariate distribution) and with others (multivariate distribution); this helps
in checking the most frequent values in data and the heterogeneity of the rocks. The probability distribution
function within the plots shows the probabilities associated with each log, helping to understand the chances
of different occurring events.

Downloaded from [Link] by Schlumberger Oilfield UK Plc user on 26 December 2025

 8 SPE-230340-MS

Figure 5—Petrophysical logs cross plot with the univariant (diagonal) and multivariant distribution. The univariant distribution
is shown as the probability density function, and the multivariant distribution is shown as a cross plot with other logs).

Feature scaling transformation and feature engineering were performed to improve the algorithm's
performance and convergence speed. This also prevents domination by logs with larger values. This step
enhances the data visualization by highlighting genuine variance and reducing noise, as seen in below in
figure 6.

Downloaded from [Link] by Schlumberger Oilfield UK Plc user on 26 December 2025

 SPE-230340-MS 9

Figure 6—Feature scaled Petrophysical log cross-plot with the univariant (diagonal) and
multivariant distribution. The univariant distribution is shown as the probability density
function, and the multivariant distribution is shown as a cross plot with other logs).

Spearman's correlation coefficient is a statistical measure of the strength of a monotonic relationship

between paired data. In a sample, it is denoted by rs and is by design constrained within positive and negative
1, and the closer rs is to the stronger the monotonic relationship. The calculation of Spearman's correlation
coefficient and subsequent significance testing of it requires the following data assumptions to hold:

• interval or ratio level or ordinal.

• monotonically related

In Figure 2, the log view plot displays the petrophysical logs that will be used as input clusters. The
upper masked heatmap in figure 7 shows how well the variable logs correlate with each other. The high
correlations can easily be noticed in the case of petrophysical logs because the core is setting the facies

Downloaded from [Link] by Schlumberger Oilfield UK Plc user on 26 December 2025

 10 SPE-230340-MS

apart. A positive correlation is seen with gamma-ray log and clay volume, neutron density and porosity, a
bulk volume of water and water saturation, and a high negative correlation is with density log and porosity,
gamma-ray log and permeability, clay volume and permeability, resistivity, and a bulk volume of water.

Figure 7—Correlation map between the different petrophysical logs

The silhouette plot is an algorithm used to assess whether the number of clusters is suitable. It
demonstrates the degree to which individual data points reflect their assigned classes in relation to other
classes. Density-Based Spatial Clustering of Applications with Noise has two crucial parameters, epsilon
radius and number of points forming a cluster, that require silhouette score to get an optimized solution.
A silhouette value of +1 suggests a perfect class assignment, whereas a silhouette value of -1 implies a
complete mismatch, although the latter is practically non-existent. If negative excursions are observed,
it means certain points align better with a neighbouring cluster than the one they are currently part of.
This suggests the necessity of introducing another cluster during computation. Figure 8 and figure 9 below
showcase the multilayer facies model post DBSCAN modeling.

Downloaded from [Link] by Schlumberger Oilfield UK Plc user on 26 December 2025

 SPE-230340-MS 11

Figure 8—Multiple facies were identified using the DBSCAN technique implemented crossplot.

Figure 9—Multiple facies were identified using the DBSCAN

technique implemented across the water-bearing injection interval.

Downloaded from [Link] by Schlumberger Oilfield UK Plc user on 26 December 2025

 12 SPE-230340-MS

Discussion of results from DBSCAN method

K-means perform poorly as the geological facies have varied cluster densities because they assign equal
importance to all points within a cluster, leading to suboptimal results in datasets with unevenly distributed
data points. While DBSCAN easily identifies clusters with arbitrary shapes and sizes, making it more
suitable for the geological setting as they exhibit complex geometries, Unlike K-means, DBSCAN is less
sensitive to initial parameter choices (such as the initial placement of cluster centroids, which causes
the model to convergence to local minima), reducing the impact of random initialization on the final
classification results. Additionally, it is noted that the outliers to the nearest cluster were potentially
misclassified by K-means as part of a cluster instead of being recognized as noise. On the other hand,
DBSCAN distinguishes outliers and noise points, effectively handles the petrophysical datasets with varying
densities and noise without assuming all points belong to a cluster. K-means demands a priori knowledge
of the number of clusters, which can be challenging in practice, especially when the true number of clusters
is unknown or variable. As DBSCAN can recognize and adjust to the innate structure of the data, it does
not need to know how many clusters to use beforehand. DBSCAN doesn't assume clusters to be spherical,
which is a limitation in K-means.

Injection Fall-Off Pressure Transient Analysis (PTA)

Injection fall-off operation was carried out in a vertical well across the fully perforated 70ft of formation
interval, and bottom hole injection pressure was recorded using a downhole pressure gauge. Figure 10a
below showcases the injection fall-off pressure versus injection flow rates. Figure 10b showcases the 10-
layer reservoir model system used for injection fall-off pressure transient analysis in industry standard PTA
software for modeling and regression analysis in log-log derivative model match step.

Figure 10—a) injection fall-off pressure versus injection rate, b) 10-layer reservoir model based on DBSCAN facies modeling

Figures 11a to 11c represents the injection fall-off pressure transient analysis results using the 10-layer
reservoir model system. A decent match was observed on all the three plots as shown below.

Downloaded from [Link] by Schlumberger Oilfield UK Plc user on 26 December 2025

 SPE-230340-MS 13

Figure 11—a) log-log derivative model match, b) semi-log derivative model match c) pressure rate history match

Figure 12 below shows the individual layer contributions in terms of injection flow rate. As seen below,
layers #1, #2 and #3 shows highest water injection rates, which is also evident from the permeability-
thickness derived for these layers.

Figure 12—individual layer water injection rates for 10-layer reservoir model

Table 1 below showcases the estimated permeability from the multi-layer injection fall-off pressure
transient analysis.

Table 1—permeability-thickness product estimated for each of the 10-layer reservoir model

Name H K K.h φ φ.h % kh % φ.h

ft mD [Link] ft
Layer#1 15.00 20.00 300.00 0.22 3.30 23.01 34.09
Layer#2 12.00 42.00 504.00 0.06 0.72 38.65 7.44
Layer#3 4.00 25.00 100.00 0.05 0.20 7.67 2.07
Layer#4 3.00 23.00 69.00 0.14 0.42 5.29 4.34
Layer#5 3.00 18.00 54.00 0.04 0.12 4.14 1.24
Layer#6 5.00 15.00 75.00 0.05 0.25 5.75 2.58
Layer#7 4.00 17.00 68.00 0.15 0.60 5.21 6.20
Layer#8 3.00 8.00 24.00 0.04 0.12 1.84 1.24
Layer#9 16.00 5.00 80.00 0.20 3.20 6.13 33.06
Layer#10 5.00 6.00 30.00 0.15 0.75 2.30 7.75
All layers 70.00 18.63 1304.00 0.14 9.68 100.00 100.00

Downloaded from [Link] by Schlumberger Oilfield UK Plc user on 26 December 2025

 14 SPE-230340-MS

Stress Testing
Water injection aims to maximize hydrocarbon reservoir recovery without exceeding its fracture gradient.
Sometimes, controlled fractures are required in a formation without disrupting the barriers above or below.
Prior to commencing water injection, it is essential to have knowledge of formation properties such as
minimum horizontal stress and direction for effective planning and execution. The water injection rate
during the secondary recovery strategy should not exceed the fracture gradient, which can create unwanted
conduits into the formation interval, and reduce the overall water injection efficiency due to fluid loss across
those fractures. Typically, the insitu stresses are macro parameters that do not vary a lot with the changes
in rock facies. Hence, stress test was conducted across a zone within the 70ft reservoir sand, based on the
hole conditions. In this section, we showcase the stress testing scenario to check the insitu stresses across
the injection sand zones, in a nearby well in same field.
A customary practice to obtain this information is by conducting an interval injection test, often referred
to as stress testing. This test employs a wireline conveyed dual packer, typically carried out in an open hole.
This technology offers several advantages, including better control and operational flexibility, real-time data
monitoring and interpretation, reduced wellbore storage, and minimized hydrostatic effects. The dual packer
typically maintains an interval of 3-5 feet between two straddle packers. Once a suitable depth or zone is
identified for stress testing, the dual packer is inflated at the required depth to isolate the interval from the
annular mud. The process involves several cycles. In the first cycle, interval pressure is gradually increased
by injecting fluid (usually annular mud) using a downhole pump at a low rate. Pressure readings are recorded
by a gauge installed inside the tool, and these data are plotted against time and rate, continuously monitored
during injection. As continuous injection occurs, a small fracture is created, but injection continues further.
When the formation breakdown is achieved, indicated by a significant drop in injection pressure, injection
continues for some time to propagate the fracture deeper. After stable propagation, injection ceases, and the
fracture is allowed to close, referred to as a pressure falloff. The falloff is analysed using various techniques
such as the square root and G-Function to interpret closure pressure.
In the second cycle, injection is resumed to reopen the fracture created in the first cycle. Once the fracture
is reopened, stable propagation is maintained for some time, and injection is stopped for another falloff.
The closure pressure derived from this second falloff is compared with the closure pressure from the first
cycle. If necessary, additional injection-falloff cycles are conducted to obtain consistent and reliable closure
pressure results. Once satisfactory results are achieved, the dual packer is deflated and moved to the next
depth for another stress testing operation.
G-Function Plot: It is performed by plotting the measured pressure(P), pressure derivative with respect
to G-function (dP/dG), and G-function semilog derivative (G. dP/dG). Closure pressure is identified by
deviation of the G-function semilog derivative from the straight line. The fracture closure pressure is usually
considered to be a good estimate of the minimum principal stress in the formation.
Square Root Plot: In square root plot, fall off pressure (Y axis) is plotted against the square root of
time (X axis). The closure pressure is marked by an inflection in the pressure curve, which is generally
determined from the local peak of the first derivative of pressure with respect of square root of time.
Figure 13 shows the injection fall-off cycles during the stress test (micro-fracturing) operation in the
same well, to estimate the stresses.

Downloaded from [Link] by Schlumberger Oilfield UK Plc user on 26 December 2025

 SPE-230340-MS 15

Figure 13—injection fall-off cycles (two cycles executed during micro-fracturing operation)

Figure 14a and 14b showcases the G-function and SQRT plots for first injection fall-off cycle.

Figure 14—a) G-Function Plot b) SQRT Plot for cycle #1

Figure 15a and 15b showcases the G-function and SQRT plots for second injection fall-off cycle.

Figure 15—a) G-Function Plot b) SQRT Plot for cycle #2

Table 2 shows the results of the stress test interpretation from both the injection fall-off cycles.

Downloaded from [Link] by Schlumberger Oilfield UK Plc user on 26 December 2025

 16 SPE-230340-MS

Table 2—stress testing results from both the cycles

Full Name Short Name 1st cycle 2nd cycle Unit

Closure time tc 0.0596249 0.059625 hr

Closure G-value Gc 0.0323148 0.032315

Closure pressure Pc 5049.36 5026.64 psia

Instantaneous Shut-In
ISIP 5064.32 5034.14 psia
Pressure

Fracture efficiency η 0.0159005 0.015901

G-function slope G-slope 485.603 241.382 psi

Results and Discussions

Water injection is one of the key methods of secondary and tertiary recovery techniques. However, an
efficient planning and modeling of water injection programs should be performed for increased efficiency.
Injectivity play a vital role in such planning and modeling operations and hence it should be estimated
with highest accuracy possible. In this paper, we demonstrated the use of machine learning based clustering
methods for multiple facies classification, integrating various petrophysical logs. A Density-Based Spatial
Clustering of Applications with Noise (DBSCAN) was used to classify the reservoir interval into a total
of 10 facies models. Post that a multi-layer injection fall-off pressure transient analysis was performed to
estimate the permeability for each of these 10 layers within the reservoir unit. Multi-layer PTA results are key
to the water injection and coning model projects due to their higher vertical resolution of the permeability
results. Stress tests are instrumental in estimating the minimum horizontal stresses to determine the safe
injection pressure. This must be done in conjunction with the injection fall-off PTA results for an overall
characterization of reservoir units.

Conclusions
In this paper, we have discussed the techniques of classification of reservoir facies / layers based on ML
techniques for better estimation of injectivity and other reservoir parameters, which can further be used in
injectivity modeling. Injectivity is one of the key parameters in determining the water injection efficiency
during the secondary recovery strategy. In case of homogenous thick sand units, the overall injectivity
remains similar across the sand thickness. However, in heterogenous sand intervals, multiple rock types /
lithofacies can be identified with each rock type or facies layer having its own relative injectivity index due
to varying rock quality within the reservoir interval. In these cases, a multi-layer injection fall-off pressure
transient analysis is recommended to estimate the injectivity of each rock type. This could help in proper
and efficient planning and simulation of the completion design or perforation design. In this paper, we have
showcased the use of DBSCAN method to perform facies classification in a heterogeneous sand interval,
where water injection can be planned during the secondary recovery stage of the field. Basic logs like gamma
ray logs have an impact on the overall simplistic layer classification; nonetheless, the combination of all
the petrophysical logs, as displayed in figure 2, served as the real foundation for this classification. The
correlation map can be used to identify significant features or logs that can also be used to improve facies
classification. DBSCAN aids in the identification of clusters based on high density values of clusters on the
multidimensional scale, considering multiple logs. DBSCAN has advantages over the K-means clustering
technique, since the latter requires a priori knowledge of number of clusters.
Based on facies model, a multi-layer injection falloff pressure transient analysis is performed to estimate
the individual layer's properties, which could be used to update a static geomodel and perform injection
forecasting based on different perforation design. In injection planning, maximum water injection rate is one
of the safety factors to be considered, so that the injection pressure should not exceed the fracture gradient.

Downloaded from [Link] by Schlumberger Oilfield UK Plc user on 26 December 2025

 SPE-230340-MS 17

Hence, estimation of fracture gradient (minimum horizontal stress) is important in such plannings. A stress
test (micro frac) operation is conducted where the injection pressure crosses the formation breakdown
pressure (above the fracture gradient), and the micro fracture is propagated followed by a falloff. These
injection falloff cycles are interpreted to estimate the minimum horizontal stresses (fracture gradient), which
can be used in injection rate design.

Acknowledgements
The authors extend their gratitude to SLB management for granting permission to conduct this research and
publish the results. Their leadership, support, and encouragement were instrumental in the automation of
the current workflows.
The authors would like to thank Equinor AS, the former Volve license partners ExxonMobil Exploration
& Production Norway AS and Bayerngas Norge AS, or their successors or permitted assignees, for
permission to use the Volve data set, and to all who have contributed to work using these data sets. Readers
are directed to [Link] for more information about the Volve data set (Equinor, 2018, Volve data
village dataset) Equinor and license terms of use. In this paper, the dataset has been used purely for research
purposes only, to explore different facies analysis workflows that could be used together in the injection fall-
off pressure transient analysis workflow. The authors would like to express their gratitude for Equinor and
other Volve data license partners for providing the open-source data set for this research work. The authors
have used the datasets only as researchers of subject topics discussed in the paper, with no commercial use
of the dataset.

References
Abbaszadeh, M., Maghsood, M., & Kamal, M. (1989). Pressure-Transient Testing of Water-Injection Wells. SPE Reservoir
Engineering, 4, 115–124. [Link]
Abdelmoneim, S.S., & Nasr-El-Din, H.A. (2018). Effect of Low Salinity Waterflooding on Sweep Efficiency in
Multilayered Clay-Rich Sandstone Reservoirs. Paper presented at the SPE Improved Oil Recovery Conference, Tulsa,
Oklahoma, USA, April 2018. [Link]
Banerjee, R., Thompson, L.G., & Reynolds, A.C. (1998). Injection/Falloff Testing in Heterogeneous Reservoirs. SPE
Reservoir Evaluation & Engineering, 1, 519–527. [Link]
Barreto, A., Peres, A., & Pires, A. (2011). Water Injectivity Tests on Multilayered Oil Reservoirs. Paper presented at the
Brasil Offshore, Macaé, Brazil, June 2011. [Link]
Bedrikovetsky, P., Marchesin, D., Shecaira, F., Souza, A.L., Milanez, P.V., Rezende, E. (2001). Characterization of deep
bed filtration system from laboratory pressure drop measurements. Journal of Petroleum Science and Engineering,
32(2), 167–177. [Link]
Boughrara, A.A., & Reynolds, A.C. (2007). Practical Analysis of Injection/Falloff Data of Horizontal Wells. Paper
presented at the SPE Annual Technical Conference and Exhibition, Anaheim, California, U.S.A., November 2007.
[Link]
Chatterjee, C., Reyes, A., Lujan, V., Majumdar, C., Prasad, K., & Kumar, A. (2018). Geomechanics in Unconventional
Reservoir: A New Approach to Drilling Integrating Advanced Acoustic Measurements and Formation Evaluation.
Paper presented at the SPE Asia Pacific Oil and Gas Conference and Exhibition, Brisbane, Australia, October 2018.
[Link]
Chen, C., Han, X., Zhang, W., Zhang, Y., & Zhou, F. (2021). A New Artificial Intelligence Method to Predict Water
Flooding Performance in Layered Reservoir. Paper presented at the International Petroleum Technology Conference,
Virtual, March 2021. [Link]
Civan, F. (2007). Chapter 19 - Injectivity of the Waterflooding Wells. In Reservoir Formation Damage (Second Edition)
(pp. 775–813). Gulf Professional Publishing. [Link]
Duci, G., Galli, M.T., & Pirrone, M. (2013). Revamping the Potentiality of Thin Layered Reservoirs: A Multi-Lithological
Facies Classification. Paper presented at the Offshore Mediterranean Conference and Exhibition, Ravenna, Italy,
March 2013.
Elkatatny, S.M., Farid, A., Mohamed, I., Abou-Sayed, O.A., & Block, G.I. (2014). An Advanced, Integrated Simulator
for Management of Produced Water Re-Injection in Multilayer Vertical or Horizontal Wells. Paper presented at the
48th U.S. Rock Mechanics/Geomechanics Symposium, Minneapolis, Minnesota, June 2014.

Downloaded from [Link] by Schlumberger Oilfield UK Plc user on 26 December 2025

 18 SPE-230340-MS

Eltoum, H., Yang, YL. & Hou, JR. The effect of nanoparticles on reservoir wettability alteration: a critical review. Pet.
Sci. 18, 136–153 (2021). [Link]
Equinor, 2018, Volve data village dataset: released under a license based on CC BY 4.0, [Link] "Volve
Field Data Village Download - Data 2008-2016." Volve Field Data Village Download Data 2008-2016, https://
[Link]/en/what-we-do/digitalisation-in-our-dna/[Link].
Fakhruzan, A., Apriadi, F., Septian, M., Akbar, M., & Wicaksono, U. (2017). Injection Flow Rate Control in Multiplayer
Packer Well Using Bridge Eccentric Injection Mandrel. Paper presented at the SPE/IATMI Asia Pacific Oil & Gas
Conference and Exhibition, Jakarta, Indonesia, October 2017. [Link]
Guan, C., Zhang, J., & Li, X. (2022). Quantitative Research on Development Status of Heterogeneous Reservoirs in
Offshore Oilfields. Paper presented at the International Petroleum Technology Conference, Riyadh, Saudi Arabia,
February 2022. [Link]
Handwerger, D., Keller, J., & Vaughn, K. (2011). Improved Petrophysical Core Measurements on Tight Shale Reservoirs
Using Retort and Crushed Samples. Proceedings - SPE Annual Technical Conference and Exhibition, 6. https://
[Link]/10.2118/147456-MS.
Kumar, A. (2012). Artificial Neural Network as a Tool for Reservoir Characterization and its Application in the Petroleum
Engineering. Paper presented at the Offshore Technology Conference, Houston, Texas, USA, April 2012. https://
[Link]/10.4043/22967-MS.
Kumar, A., Balikram, A., & Ojha, K. (2021). Deep transient testing methodology: An integrated approach to redefine the
real-time reservoir complexities and well deliverability. Journal of Petroleum Science and Engineering, 207, 109179.
[Link]
Kumar, A., Batshas, S., Nagaraj, G., Nahar, S., Singh, S., & Chatterjee, C. (2019). Advanced Vertical Interference Test
Modeling by Considering Petrophysical and Image Based Rock Typing for a Better Kv Estimation in a Heterogeneous
Reservoir-A New Workflow. Paper presented at the International Petroleum Technology Conference, Beijing, China,
March 2019. [Link]
Kumar, A., Gidwani, A., Singh, S., Wydiabhakti, T., & Mishra, S. (2019). Utilizing Advanced Logs for Flow Unit
Classification in Vertical Interference Test Modeling. Paper presented at the SPE Oil and Gas India Conference and
Exhibition, Mumbai, India, April 2019. [Link]
Kumar, A., Liang, L., & Ojha, K. (2024). Simultaneous inversion of permeability, skin, and boundary from pressure
transient test data in three-dimensional single well reservoir model. Petroleum Research, 2024. [Link]
[Link].2024.01.004.
Kumar, A., Ojha, K., Wydiabhakti, T. B., & Nahar, S. (2019). Utilizing Petrophysical Facies Model in Multi-Layer Pressure
Transient Analysis: A Novel Approach. Paper presented at the SPE/IATMI Asia Pacific Oil & Gas Conference and
Exhibition, Bali, Indonesia, October 2019. [Link]
Kumar, A., Rajput, S., Nukala, P. V. S. T., Wydiabhakti, T. B., Trevisan, E. A., & Ojha, K. (2023). Artificial Neural
Network Based Facies Modeling for Multi-Layer Injection Fall-Off Transient Analysis. Paper presented at the SPE/
AAPG/SEG Latin America Unconventional Resources Technology Conference, Buenos Aires, Argentina, December
2023. [Link]
Kumar, A., Rajput, S., Nukala, P.V.S.T., Wydiabhakti, T.B., Trevisan, E.A., & Ojha, K. (2023). Artificial Neural Network
Based Facies Modeling for Multi-Layer Injection Fall-Off Transient Analysis. Paper presented at the SPE/AAPG/SEG
Latin America Unconventional Resources Technology Conference, Buenos Aires, Argentina, December 2023. https://
[Link]/10.15530/urtec-2023-3963711.
Kumar, A., Wydiabhakti, T.B., Singh, A., Hati, S., Das, B., & Thorat, A. (2023). A Novel Integration of Advanced Acoustic
Geomechanics and Reservoir Micro Fracturing Techniques for Refined Stress Characterization in Carbon Capture and
Storage Formations. Paper presented at the Asia Pacific Unconventional Resources Symposium, Brisbane, Australia,
November 2023. [Link]
Kumar, A., Wydiabhakti, T.B., Singh, A., Sharma, A., Lopes, V.S., Deshpande, V., Rajput, S., Nukala, P.V.S.T., Das,
B., Deori, B., Basu, I., & Ojha, K. (2023). An Integrated Novel Workflow for In-Situ Stress Characterization of
CCS Reservoirs. Paper presented at the SPE/AAPG/SEG Latin America Unconventional Resources Technology
Conference, Buenos Aires, Argentina, December 2023. [Link]
Liu, Y., Han, X., Chen, C., Wang, H., & Liu, H. (2020). Water Flooding Performance Prediction in Layered Heavy Oil
Reservoir. Paper presented at the Offshore Technology Conference Asia, Kuala Lumpur, Malaysia, November 2020.
[Link]
Lu, X., Sui, J., Zhao, H., & Yang, H. (2000). Stochastic Modeling Technique for Heterogeneous Multi-layer Sandstone
Reservoir. Paper presented at the International Oil and Gas Conference and Exhibition in China, Beijing, China,
November 2000. [Link]

Downloaded from [Link] by Schlumberger Oilfield UK Plc user on 26 December 2025

 SPE-230340-MS 19

Mad Sahad, S., Tan, N.W., Sajid, M., & Jones, E.A. (2022). Enhancing Channelised Features Interpretability Using
Deep Learning Predictive Modelling. Paper presented at the SPE Asia Pacific Oil & Gas Conference and Exhibition,
Adelaide, Australia, October 2022. [Link]
Patro, R., Chatterjee, C., Kumar, A., Das, B., & Majumdar, C. (2018). Horizontal Stress Determination in Anisotropic
Formations Integrating Advanced Acoustic Measurements. Paper presented at the SPWLA 2nd Asia Pacific Technical
Symposium, Bogor, Java, Indonesia, November 2018.
Raghavan, R., Topaloglu, H.N., Cobb, W.M., & Ramey, H.J. (1974). Well-Test Analysis for Wells Producing From Two
Commingled Zones of Unequal Thickness. Journal of Petroleum Technology, 26, 1035–1043.
[Link]
Satter, A., & Iqbal, G. M. (2016). Waterflooding and waterflood surveillance. In Reservoir Engineering (pp. 289–312).
Gulf Professional Publishing. [Link]
Shi, X., Cui, Y., Guo, X., Yang, H., Chen, R., Li, T., Li, R., Wang, R., Wang, J., & Meng, L. (2017). Logging Facies
Classification and Permeability Evaluation: Multi-Resolution Graph Based Clustering. Paper presented at the SPE
Annual Technical Conference and Exhibition, San Antonio, Texas, USA, October 2017.
[Link]
Sommerauer, G., & Zerbst, C. (2007). Rapid Pressure Support for Champion SE Reservoirs by Multilayer Fractured Water
Injection. SPE Drilling & Completion, 22, 356–368. [Link]
Thambynayagam, R.K.M. (1984). Analytical Solutions for Pressure Buildup and Falloff Analysis of Water Injection
Tests of Partially Penetrating Wells: Nonunit Mobility Ratios. Paper presented at the European Petroleum Conference,
London, United Kingdom, October 1984. [Link]
Udoh T.H. Improved insight on the application of nanoparticles in enhanced oil recovery process. Scientific African (2021),
Article e00873. [Link]
Verma, S., Bhattacharya, S., Chowdhury, N.U.M.K., & Tian, M. (2021). A new workflow for multi-well lithofacies
interpretation integrating joint petrophysical inversion, unsupervised, and supervised machine learning. Paper
presented at the SEG/AAPG/SEPM First International Meeting for Applied Geoscience & Energy, Denver, Colorado,
USA and online, September 2021. [Link]
Wang, F., Li, J., & Chen, X. (2013). Seismic Inversion Integrating Facies Classification and Multi-Point Geostatistics.
Paper presented at the 2013 SEG Annual Meeting, Houston, Texas, September 2013.
Wydiabhakti, T.B., Hati, S., Das, B., Kumar, A., Thorat, A., Singh, A., & Sherratt, P.J. (2023). A Fit-For-Purpose
Workflow for In-Situ Stress Characterization in Carbon Capture and Storage Formations. Paper presented at the
Offshore Technology Conference Brasil, Rio de Janeiro, Brazil, October 2023. [Link]
Zhan, L., Kuchuk, F.J., Ma, S.M., Shahri, A.M., Zeybek, M., Ramakrishnan, T.S., Altundas, B., De Loubens, R., &
Chugunov, N. (2009). Comprehensive Reservoir Characterization Through Data Integration and Numerical Single
Well Simulation. Paper presented at the International Petroleum Technology Conference, Doha, Qatar, December 2009.
[Link]

Downloaded from [Link] by Schlumberger Oilfield UK Plc user on 26 December 2025