Maceral study to determine the desorbed gas content in the Upper Cretaceous coals of the Landázuri area, Middle Magdalena Valley Basin, Colombia

15 For several decades, some coal petrographic properties have been proposed as important 16 parameters in the methane gas sorption processes. In this contribution, the petrographic variables 17 (Vitrinite Ratio, Inertinite Ratio, the petrographic indexes (Gelification Index, Groundwater Index, 18 Tissue Preservation Index, Vegetation Index, Vitrinite/Inertinite ratio, and the Vitrinite 19 Reflectance were evaluated according to the maceral preservation and were related with the 20 desorbed gas content. Twenty-five coal seams obtained from the drill cores of two wells in the 21 Landázuri Area-Valle Medio del Magdalena basin were analyzed. The coal samples were grouped 22 according to gas content using principal component analysis (PCA). The petrographic results were 23 analyzed by linear regression and multiple regression. The Medium Volatile Bituminous to Low 24 Volatile Bituminous coals from Landázuri 1 are twice as high in gas content that the High Volatile 25 Bituminous A to Medium Volatile Bituminous coals from Landázuri 2. The volume percentage 26 and the preservation degree macerals are related closely to the gas content and the pore's size 27 involved in the sorption process. The Inertinite is the maceral group related with the highest gas 28 content groups in Landázuri (600 SCF-Standard Cubic Feet/ton, 300 SCF/ton), while the other 29 groups show the correspondence with the vitrinite macerals. The syngenetic and diagenetic origin 30 of the pyrite contributes microporosity to the desorption process, while the pyrite epigenetic by its 31 size reduces it. The petrographic indexes reveal that the Upper Cretaceous coals were developed in 32 swampy or lacustrine continental basins- limnic facies.

The absorbed gas by the coal matrix was calculated from the residual gas in 193 pulverized samples (60 mesh) in a container for gas mixture or in a hermetically 194 sealed ball mill connected to a manometer. The measurements obtained were 195 processed using the TerraGas software (https://terrasolid.com/product/tgas/), 196 which performs pressure, temperature, and gas volume corrections, measured 197 from the sample weight, canister volume, volume of inert material introduced, and 198 density (ISO Standard 17892-2:2014). The cumulative gas desorption curve points 199 out the intervals of lost, measured and residual gas by plotting the cumulative gas 200 content vs time (ft/ton Vs √hours) (Fig. 3) To avoid the analyst's subjectivity when classifying coal seams groups, cluster 228 analysis was used, based on K-Means Clustering (Hartigan and Wong 1979). This 229 procedure attempts to discriminate groups of relatively homogeneous cases, based 230 on certain selected properties. In our particular case, the following variables were 231 used: RV, RI, GI, GWI, TPI, VI, V/I. Pseudocode written in R was used, which 232 allows a large number of cases to be considered. However, it is often the case that 233 the algorithm requires the user to specify the number of clusters. For this, the 234 initial number of the clusters was obtained from the Elbow Method (Lloyd 1982). 235 In the algorithm, the initial centroids of the clusters can be specified, and the 236 8 cluster centroids are updated iteratively. The algorithm used also allows automatic 237 selection of the number of clusters based on Bayesian information criteria 238 (Schwartz 1978). It can also store cluster labels for each measurement, distance 239 information, and the centers of the final clusters. The relative size of the statistics 240 provides information about the contribution of each variable to the separation of 241 the clusters. Once the classification is performed; the results are presented in the 242 form of dendograms. The R correlation factor is analyzed as a linear regression 243 product from the petrographic variables concerning the gas content in the defined 244 groups. Then, for each group, a Pearson correlation coefficient analysis was 245 carried out (Kendall and Stuart 1973), which allowed us to determine different 246 relationships between variables pairs. The most significant variables from this 247 independent study were compared statistically with each petrographic variable. To the southeast of the De Armas syncline (see Fig. 1 (Fig. 5c, 5f, 5k, 5l, 5n, 6c, 6f, 6h). Gelinite is of a homogeneous 277 aspect and is occasionally fractured (Fig. 5e, 6d), Vitrodetrinite is observed as 278 angular fragments and is concentrated in the edges of the Vitrinite macerals ( Fig.  279 5j, 6e). 280

290
The mineral matter comprises clay filling, replacing macerals ( Fig. 5a, 5d, 6a, 6c) 291 and pyrite, with values between 3.60 to 30.00%, obtaining the highest values for 292 Landázuri-1 (see Tables 1 and 2). In the coal seams analyzed, the pyrite was 293 characterized according to morphology into framboidal, granular, massive, and 294 crystalline. It was observed replacing macerals and filling voids and fractures 295 ( Fig. 5b, 5g, 5h, 5i, 6b, 6j). The Elbow plots shown in Figures 8 A and B allows us to discriminate seams 303 groups for each well. Three seams' groups from the Landázuri 1 well show that 304 the quadratic distance between the elements that comprise them is less than 305 20,000. While Landázuri 2 well has a distance between two and three seams' 306 groups in the range between 30,000 and 50,000. Although, for both wells with 4 307 and 5 groups, the distance is minimized the samples number within each was 308 taken into account. The Landázuri 1 well consists of fourteen core samples, while 309 the Landázuri 2 well comprises 11 samples. Accepting groups of 4 or 5 would 310 imply that they would be formed by 1 or 2 samples and, in that case, the 311 correlations between the variables would not make any geological sense, since 312 any straight line could be adjusted by linear regression of two measurements. 313 Thus, we decided to keep the 3 groups for Landázuri 1 and 2 groups for Landázuri 314 2 (Fig. 7). The dendograms produced using PCA (Principal Component Analysis) 315 analysis suggest that the samples integrate each group according to the measured 316 gas content (Fig. 8). 317 318

Linear regression analysis 319
Relationship between measured gas volume and Vitrinite Reflectance 320 The coal rank has been defined as an important factor in the maceral composition 321 variation and, therefore, in the gas storage capacity (e.g., Crosdale et al. 1998, 322 2017; Keshavarz 2017). Other authors deduced that a higher coal rank indicates a 323 greater depth, and as the Vitrinite percentage increases the gas content is higher 324 (Bustin and Clarkson 1998, among others). In the Landázuri coals, the reflectance 325 analysis reveals that the gas content does not increase proportionally with rank, 326 nor with depth, taking into account that the highest reflectance values are not 327 found in the lowest stratigraphic position (see Fig. 9 a,c). 328 It is evident that the middle-to-high range coals-(MVB-LVB) from the Landázuri 329 1 well, duplicate in gas content the lower-range coals (HVB -MVB) from the 330 Landázuri 2 well. The gas content (< 600 SCF / ton-Landázuri 1 and <300-331 Landázuri 2) have no relationship with Vitrinite Reflectance, while the groups 332 11 with higher gas content present a correlation factor R2 = 0.74 for Landázuri 1 and 333 0.99 for Landázuri 2 ( Fig. 9  The correlation of the Inertinite percentage with the gas content suggests a 358 positive influence in the groups with higher gas content (>600 and >300 SCF/ton). 359 The remaining groups indicate a correlation with a negative trend. (See Fig. 11; 360 Table 1 to 4). 361 The IR has merit in the defined coals groups, specifically in the >600 SCF/ton gas 362 content -Landázuri 1, where the degraded Inertinite is higher than the structured 363 Inertinite. (See Fig. 11a Although an inverse relationship of the Inertinite radius is observed in the > 600 370 SCF/ton group, the pore size distribution and its role in the gas desorption process 371 in both structured and the degraded inerts remain significant (see Fig. 11c, d).  The index that reveals the degree of persistence of wet and dry conditions is the 405 GI (see Fig. 14e, f); and the GWI and the Vitrinite/Inertinite ratio (V/I) represent 406 the degree of gelation of the fabrics according to the water and pH contributions 407 (see Fig. 14c, d, i, j). These indices show the same behavior as the percentage of 408 Vitrinite concerning the gas content illustrated in Fig. 10a, b. The GI value 409 exception is the Landázuri 2 well with >300 SCF/ton content, with a very low 410 Telovitrinite and Colodetrinite content; therefore, the index acquires a high value. 411 On the other hand, the TPI index as an indicator of the degree of humification of 412 organic matter macerals (see Fig. 14a, b) and the VI index (see Fig. 14g Vitrinite> Structured Vitrinite condition ( Fig. 15 and 16). The coal facies 420 variation is not evident in the same geological formation, as the seams are from 421 the same well. 422 Pearson's analysis 423 The relationships between pairs of variables are plotted as Pearson correlograms, 424 these summarized the linear regression analysis, highlighting the inertinitic 425 macerals contribution in the group of each well with higher desorbed gas content 426 and of Vitrinite (V) and V/I radius in the coals categorized with medium to low 427 gas content (see Fig. 17). For these correlograms, only variables that have a 428 relationship were plotted using a p-value of 0.25. Additionally, the correlograms 429 allow us to corroborate other linear relationships between variables that had not 430 been considered before and to identify their influence on the response variable, as 431 is the case of gas content.  Table 6 shows that the 444 Str V variable is significant assuming a reliability level of 99%, the adjusted R2= 445 0.229 increases, while the p-value diminishes to 0.08328, which imply that the 446 two predictors' variables Str I and Str V are significant to reproduce the gas 447 volume. 448 Considering the same five predictor variables previously mentioned, in Table 7  449 we summarized the multiple linear regression results for the Landazuri-2 well. 450 According to this Table, Table 8). 458 while the medium to low gas content is directly related to the content of vitrinite 475 (micropore) emphasizing structured macerals (see Fig. 19).      Table 7. Summary of multiple linear regression for initial models of the Landazuri-2 well 819 820      Relationships between gas content (SCF/ton) and petrographic indexes (TPI, GWI, GI, VI, and V/I).
Relationships between gas content (SCF/ton) and petrographic indexes (TPI, GWI, GI, VI, and V/I)   Synthesis of the importance of maceral groups and re ectance for the measured gas content

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. GuatameetalTables.pdf