Automated Analysis of Gridded Geologic Map Data*

By

Susan M. Schrader¹ and Robert S. Balch¹

Search and Discovery Article #40198 (2006)

Posted July 3, 2006

*Adapted from oral presentation at AAPG Annual Convention, Houston, Texas, April 9-12, 2006, and associated abstract and extended abstract

Click to view presentation in PDF format.

¹New Mexico Tech Petroleum Recovery Research Center, Socorro, NM ([email protected])

Abstract 

With the advent of exploration tools such as expert systems and neural networks, the geophysical, formation and production knowledge required for making decisions about a potential prospect often requires digitized or gridded map data and automated map data analysis. In order to analyze this input data, algorithms need to be developed that can look at numerically gridded data and interpret them in the same way an experienced geologist interprets a feature map. As part of the development of expert exploration systems for two New Mexico formations, a series of algorithms were developed to interpret data from cores, well tests and production reports. The simplest of these algorithms include search algorithms for linking each prospect location to the nearest map gridpoint, finding the producing well closest to the prospect, and finding the nearest location with high predicted production. Another set of algorithms use formation tops to compute the relative dip between each prospect and the nearest producing well, and search for the nearest downdip source rock. More involved algorithms were designed to search for an updip sand pinchout near the prospect, evaluate the consistency of a feature such as formation thickness in a region around the prospect, and determine if the paleostructure in the region supports trap potential. The results of these algorithms provide valuable input for the expert system tools as well as providing a fast method to scan large amounts of gridded numerical data for values that can be linked to production potential.

Selected Figures 

Background 

In recent years, there has been increasing interest in using computing tools to aid in exploration and production decisions. Some of these tools use artificial intelligence techniques to analyze log, production, and geophysical data in a matter similar to a human expert. An example of this is the Fuzzy Expert Exploration Tool (FEE Tool) developed to evaluate potential drilling prospects in the Lower Brushy Canyon formation of the Delaware Basin and the Devonian Carbonates in Southeast New Mexico. 

The FEE Tool expert systems utilize a knowledge base of rules initially derived by working with experts on those formations. Many of these rules require numerical inputs. While the system allows such inputs to be provided by the user, there are default inputs for all of the rules stored in the system’s answer base. These numerical inputs can be divided into two types of data: primary data and derived data. In this context, primary data is a term for measured data that are interpolated over the grid and used as inputs, while derived data are inputs derived from measured and interpolated data. An example of primary data used by the Delaware and Devonian FEE Tools is total organic carbon (TOC). A set of knowledge base rules uses TOC as an input, with a high value of TOC at a prospect location serving to enhance the prospect. Measured values of TOC at a number of locations in the region were available and were interpolated over the region to provide an input for the rule set at any location the user selected. A simple example of derived data is the distance from each prospective drill site to the nearest producing well, calculated by selecting the minimum distance between each possible drill site and existing producing wells. A more involved example of derived data involves determining if an updip sand pinchout exists at each of the possible drill sites. In creating the derived data used by the expert system tools, methods and algorithms were developed to analyze mapped data.

Regional Grids 

The two regions studied in this work were overlapping areas in southeastern New Mexico (Figure 1). The Delaware Basin region was gridded into 60,478 1320ft x 1320 ft squares corresponding to 40-acre spacing. Each square was referenced by its center point. The grid was generated using UTM coordinates, although lat-long and TSR coordinates systems are also available to users. The Devonian region is a larger rectangular region that was gridded into 64,347 2640 ft x 2640 ft squares representing 160-acre spacing. An internal set of UTM coordinates were generated from the lat-long coordinates by Rockware.

Simple Derived Data 

Both the Delaware and the Devonian FEE Tools contain a set of rules that evaluate a prospective drill site by measuring the distance to the nearest producing well. Data containing the location of the producing wells in the region were collected, and a sample was held aside for testing. In the case of the Delaware, a cutoff date was selected, beyond which wells were not used in generating answerbase data. In the case of the Devonian, a random set of wells was kept out of the production data for the answerbase. The distance to the nearest production was computed for each gridpoint by computing the distance to each producing well using the standard Euclidean distance and the UTM coordinates for each point, then selecting the minimum distance. This was coded in Matlab. 

Along with distance to actual production, a second distance computed for each gridpoint was the distance to the nearest region of high predicted production. For each region, predicted production was computed using a neural network (Balch et al., 1999; Schrader et al., 2005), Each gridpoint has a value of predicted production in units of predicted barrels of oil (or oil equivalent if gas production was significant) per month (PBOEPM). To calculate the distance to nearest high predicted production, the first step was to define five linguistic variables related to the values of PBOEPM. The predicted values could be classified as very low, low, average, high or very high. The distance code from the previous step was then modified to do the following:

1. If the predicted production at the gridpoint was high or very high, the distance to high predicted production is 0.

2. If the predicted production at the gridpoint was average, the distances between the gridpoint and all gridpoints where predicted production was very high were computed, and the minimum of these distances was selected as the distance to high predicted production.

3. If the predicted production at the gridpoint was low, the distances between the gridpoint and all gridpoints where the predicted production was either very high or high were computed, and the minimum of these distances was selected as the distance to high predicted production.

4. If the predicted production at the gridpoint was very low, the distances between the gridpoint and all gridpoints where the predicted production was very high, high, or average were computed, and the minimum of these distances was selected as the distance to high predicted production.

Dip, Sand Pinchout, and Related Algorithms 

A set of rules in the Delaware FEE Tool knowledgebase considers dip between the gridpoint and the nearest producing well. The dip angle is shown in Figure 2. The angle for each point was calculated in Excel. 

The elevation differences between formation tops and the gross formation thickness are used in other calculations as well. An important production indicator for Delaware prospects is the existence of an updip sand pinchout. An algorithm was developed to search in the neighborhood of each gridpoint and to determine if an updip sand pinchout exists. The existence of such an updip pinchout enhances the user’s prospect. 

To search for the sand pinchout, the eight neighboring points were defined as shown in Figure 3. A Matlab code collected the eight neighboring points, and for each point, calculated the dip between the point and the central gridpoint. The gross formation thickness was collected for each point updip of the central point. If there was a thinning of the formation updip of the point in question, the gridpoint was flagged as having an updip pinchout. On the other hand, if the updip points had a formation thickness significantly higher than the gridpoint, it had a negative effect on the gridpoints prospect. 

To consider migration, an algorithm was developed to look for quality source rock downdip from the gridpoint. The total organic carbon measurements, which were also used as primary data, were used in a modified distance algorithm. This algorithm was designed to calculate the distance to the nearest downdip regions with high source rock content.

Consistency Algorithms 

Two sets of rules in the Delaware FEE Tool’s knowledge base suggested by experts involve the consistency of primary data in a neighborhood of the prospect. The consistency of the formation thickness and the consistency of the predicted production values from the neural network are the basis of these rule sets. If the formation thickness is consistent in a neighborhood of the prospect, it enhances the prospect. If there are variations in thickness, it could indicate the edge of a structure. Similarly, if the predicted production (based on a neural network using geophysical data as input) is consistent, it enhances the prospect. 

To determine consistency, a small neighborhood was defined as shown in Figure 3. A larger neighborhood was defined as containing all gridpoints within 3960 ft of the central gridpoint. For each of the neighborhoods, the mean (μ) of the thickness was computed, along with the standard deviation (σ). Intervals were built around the mean using the standard deviation and used to determine the consistency of the variable at the center gridpoint. 

For instance, if the value of the variable at the center gridpoint is outside of the interval (μ-σ, μ+σ) then it is flagged as inconsistent. The same approach was used with the predicted production value.

Paleostructure 

The Devonian FEE Tool knowledge base contained a rule set requiring an input related to paleo thickness. For the Devonian Carbonates, the paleo thickness was defined as the difference between the Abo formation top and the Mississippian formation top (Broadhead et al., 2004). The goal of creating the variable was to recognize locations where there was a significant thinning of the paleo thickness with respect to the surrounding area.

This was done using the following steps:

1. The paleo thickness was calculated from the gridded Abo and Mississippian formation top data.

2. A Matlab code was created that compared the paleo thickness value at each point to the mean paleo thickness in the surrounding area, (defined as x +/- 26,400, y +/- 26,400).

The code then compared the difference between the mean and the value at the point.

3. The result of this code, termed paleo difference, was exported to an Excel spreadsheet, where the paleo output was computed. A significantly positive value of this would indicate a thinning at a point compared to the surrounding area.

4. To compute paleo output for each gridpoint, first the Abo and Woodford thicknesses were compared. If either of these were zero, the paleo output was zero. If not, the paleo output value was computed based on the value of the paleo difference.

5. This was based on the 5th, 6th, 7th, 8th, and 9th deciles of the paleo difference values. If the paleo difference was between the 5th and 6th decile, it was set to 1, between the 6th and 7th, if was set to 2, etc.

6. Once this was computed, it was mapped and evaluated at the test well sets. As with the other derived data, statistical tests were done to help fine-tune the knowledgebase rules.

FEE Tool Results 

The FEE Tools produce a quality estimate for each of the gridpoints in the region. This quality estimate is a number between 0 and 1. Values close to one indicate that the gridpoint is a good prospect for oil and/or gas production. Figure 4 is a map of the Delaware FEE Tool quality estimates. The map includes producing wells (dots) and dry holes (squares). The tools have been tested and are successful in aiding exploration in the region.

References 

Balch, R.S., Stubbs, B.S., Weiss, W.W., and Wo, S., 1999, Using artificial intelligence to correlate multiple seismic attributes to reservoir properties: paper SPE 56733 presented at the 1999 SPE Annual Technical Conference, Houston, Oct. 3–6.

Broadhead, R.F., Jianhua, Z., and Raatz, W.D., 2004, Play analysis of major oil reservoirs in the New Mexico part of the Permian Basin: Enhanced production through advanced technologies: New Mexico Bureau of Geology and Mineral Resources, Open-file Report 479, CD-ROM (2004).

Schrader, S.M., Balch, R.S., and Ruan, T., 2005, Using neural networks to estimate monthly production: A case study for the Devonian carbonates, Southeast New Mexico: paper SPE 94089, presented at the 2005 SPE Production and Operations Symposium, Oklahoma City, April 17-19.