Predicted production decline (click at target location in Marcellus first):
When arriving at the site, one should initially see this:
It is a map of EUR (P50) in units of Bcf/1000 ft. of lateral (for every 1000 ft. of the horizontal portion of the well, one would expect to achieve <i>X</i> Bcf, where <i>X</i> is the value corresponding to the color scale). P50 is the most probably outcome, whereas P10 is a reasonable best-case scenario and P90 a reasonable worst-case scenario (there is an 80% chance that actual EUR falls between P90 and P10). The map can be changed to display P10 or P90 using the associated drop down menu. One can also view maps of the decline curve parameters and their variances with the same drop down.
Overlying the map are points representing the wells used in the analysis (not all wells in the Marcellus had usable production data; only about a third of all wells were used). The points are initially colored by the natural logarithm of their initial production rate, as determined by decline curve fitting. The variable displayed at the points can be changed with its associated drop down menu.
When the map is displaying EUR values, the color scale is for ranges of values. One may want to know the specific EUR value at a target location. This can be achieved by simply clicking at that target site, and waiting a moment for a graph to be generated looing like this:
This graph shows the predicted decline in production rate over time at the target site. The specific values of EUR P10, P50 and P90 are displayed, again in units of Bcf/1000 ft. of lateral. The production rates themselves are also normalized by 1000 ft. of lateral (y-axis). Time is in months.
The underlying approach used to achieve the results presented in this web app is purely data-driven. Predictions of EUR and production decline are only based on historical production data. The method does not make any assumptions about geological conditions or the physical behavior of the reservoir, beyond what is assumed in Duong's decline curve model. Also, the co-kriging approach used here gives very rational predictions of production at undrilled sites; the approach weights the data by distance of wells from the target site. So the prediction at the target site is influenced more by closer wells and less so by wells further away. More importantly, the co-kriging approach tells us how confident it is in it's predictions; low variance means high confidence and vice versa. The variance in the prediction reflects the variance in the data, so if surrounding wells have very different production, the P10 and P90 at the target site will be further apart.
The underlying approach used to achieve the results presented in this web app is purely data-driven. Predictions of EUR and production decline are only based on historical production data. The method does not explicitly consider geology or reservoir physics. Furthermore, the method does not consider variations in completions or operations from well to well. Therefore, future wells may take advantage of advances in technology, completion design, and well operation that would boost production well above that surrounding, older wells. More generally, the predictions made by this web app are only as good as the data available. Other limitations are discussed in:Xi and Morgan (2019)