SYSTEMATIC LAVA FLOW IDENTIFICATION

During Leg 183, we encountered clear examples of aa, pahoehoe, slab pahoehoe, and rubbly pahoehoe lava flows (Coffin, Frey, Wallace, et al., 1999). Table T2 details the characteristics that were used on board the ship to define each lava type. However, a more systematic methodology for interpreting these data is desirable in many cases. The procedure described in this chapter can readily be incorporated into a spreadsheet program to provide a very simple "expert system" for lava flow identification. The concept is to list the key macroscopic attributes that allow one to distinguish the different lava flow types. Each attribute is given a weighting factor depending on its importance in identifying that specific lava type. Then for each lava flow, the presence, absence, or nondetection of that attribute is noted. The sum of the observed attributes gives a quick estimate of how closely that lava flow matches a given lava type. Ideally, one and only one lava type will have a high sum, providing a definitive identification of the type of lava flow. In the absence of sufficient data to make a definitive identification, the process should provide some estimate of the level of confidence in assigning a type to the flow and eliminate as many lava types as possible.

The attributes and weights used in this paper are shown in Table T3. Characteristics that are listed as "required" in Table T2 for a specific lava type are given a weight of 10. Characteristics that are "common" are given a weight of 1. Similarly, characteristics that are commonly absent and required to be absent were given weights of -1 and -10, respectively. The negative weights are what allow one to rule out a given lava type for the flow being studied.

Upon examining Table T3, one immediately can see that this scheme needs to be improved upon by adding more diagnostic characteristics for the transitional lava types. In particular, it is important to note that the total number of distinguishing characteristics for slab pahoehoe is only about half of that for the other three lava types. This means that we have a shortage of attributes with which to identify slab pahoehoe. This can lead to problems. For example, if the only observation we have is that a lava flow has a autobrecciated flow top, this has 19% of the attributes of an aa flow, 17% of the attributes of a rubbly pahoehoe flow, and 34% of the available attributes of slab pahoehoe.

One additional step is required before this technique can be used. In reality, not all the important observations can be made. For example, the base of a lava flow might not be exposed or recovered. Thus a level of confidence in the presence (or absence) of each attribute is required. Table T4 summarizes the values used here: confident detection and nondetection are assigned a value of 1 and -1, respectively, whereas a value of 0.5 or -0.5 is used when the detection or nondetection is questionable (e.g., because of extreme weathering of the core). A value of zero is used when the feature is not observable at all.

The weighting factor of each attribute is multiplied by the confidence in determining the presence (or absence) of that attribute. Note that this scheme provides a positive value when an attribute the lava should not have is determined to be absent. For example, the tentative nondetection of an attribute a flow must not possess results in -0.5 x -10 = 5. These products are summed, then normalized by the maximum possible total, to provide a score (in percent) for that flow being a specific lava type. This entire procedure was incorporated into a spreadsheet program that is available in the volume supplementary material (see the "Supplementary Materials" contents list).

Even in this preliminary state, this system is able to adequately distinguish the four different lava types and provides good results when applied to four actual Hawaiian flows (see Tables T5, T6; also see "Appendix A"). Table T5 shows that the an ideal pahoehoe flow shares significant similarity to a rubbly pahoehoe flow but is clearly distinguished from aa and slab pahoehoe flows. The similarity between pahoehoe and rubbly pahoehoe is primarily because both flow types are commonly inflated. Aa flows share some characteristics with slab pahoehoe flows but are very different from both pahoehoe and rubbly pahoehoe flows. An ideal slab pahoehoe flow is clearly distinguished from both pahoehoe and rubbly pahoehoe but is difficult to distinguish from an aa flow. This is not just because of the shortage of identifying characteristics for slab pahoehoe flows; it is also a reflection of the great similarity in the way both lava types form. An ideal rubbly pahoehoe flow is clearly separated from aa flows and is significantly different from pahoehoe. However, the ideal rubbly pahoehoe flow also scores highly as a slab pahoehoe flow. Again, this reflects inherent similarities between rubbly and slab pahoehoe as well as our limited list of characteristics with which to identify slab pahoehoe.

Applying this technique to Hawaiian lava flows shows that this "expert system" does work on real lava flows. Two classic aa flows, a transitional aa flow, and a classic pahoehoe flow are shown in Table T6 and "Appendix B". The aa and pahoehoe flows are unambiguously identified. The transitional lava flow scores high on both the aa and slab pahoehoe categories, accurately reflecting the transitional aa nature of this flow. None of these Hawaiian examples could be confused with rubbly pahoehoe.

The detailed observations of all the lava units encountered during Leg 183 are described in Coffin, Frey, Wallace, et al. (1999). These observations were converted into tables, noting the presence or absence of each of the attributes listed in Table T2 for every unit containing mafic lava flows in Holes 1136A-1139A. These tables are reproduced in "Appendix C," "Appendix D," "Appendix E," and "Appendix F". Figure F2 shows examples of the sections of drill core used to confidently determine the presence of some of the attributes listed in these tables.

A spreadsheet program uses these tabulated data to compute a score (in percent) for each unit for its similarity to aa, pahoehoe, slab pahoehoe, and rubbly pahoehoe. The unit was confidently interpreted as a given flow type only if it scored above 1 (68.26%) and the score was at least 20 percentage points higher than any other lava type. The unit was tentatively identified as a lava flow of a given type if it scored >34% for that lava type and the score was >10 percentage points higher than for any other lava type. The remaining units were considered unclassifiable. However, for half of the unclassifiable units, the allowable flow types were limited by the classification process (i.e., one or more flow types scored much poorer than the rest of the flow types).

The resulting classifications of lava type (including unclassifiable) are largely consistent with those made on board the ship. The discrepancies were reviewed, and in each case, the new interpretations are actually found to be preferable. In particular, the tables in Coffin, Frey, Wallace, et al. (1999) often did not adequately convey the degree of uncertainty in the classification.

Table T7 summarizes the results. Overall, 12 of the 42 units (29%) could not be classified. This was the result of poor recovery and/or extreme alteration and weathering of those units. In fact, in many of those cases the original interpretation of those units questioned whether they represented a single lava flow (Coffin, Frey, Wallace, et al., 1999). Of the remaining 30 units, 7% were classified as slab pahoehoe, 13% as aa, 27% as pahoehoe, and 53% as rubbly pahoehoe. Whereas aa and slab pahoehoe flows were confined to Holes 1138A and 1139A, pahoehoe and rubbly pahoehoe flows were found in all four holes. As discussed in Coffin, Frey, Wallace, et al. (1999), these and other observations suggest that Holes 1138A and 1139A penetrated lavas emplaced on steeper slopes closer to the vent than Holes 1136A and 1137A.

A large enough number of lava flows were sampled during Leg 183 to see some of the relationships between the different lava types. Figure F3 plots the score of different lava types against each other. Figure F3A shows how pahoehoe and aa are very clearly anticorrelated; the best fit line has a large R², has a slope close to -1, and passes very close to the origin. This is a striking demonstration of the fact that pahoehoe and aa flows are fundamentally different and that those differences are very well characterized. Figure F3B shows that slab pahoehoe is anticorrelated with pahoehoe and is correlated with aa. However, the best-fit lines do not have as high R² as the pahoehoe vs. aa trend. As noted earlier, this is the result of a combination of the inherent similarities between slab pahoehoe flows and aa flows as well as the relatively limited number of identifying characteristics developed for slab pahoehoe flows.

Figure F3C shows a remarkable noncorrelation between rubbly pahoehoe and any of the other lava types. R² values range from 0.02 to 0.005. Several conclusions can be drawn from this. First, this noncorrelation is clear confirmation that rubbly pahoehoe is unlike any of the common Hawaiian lava types. Second, it shows that the technique here does clearly distinguish rubbly pahoehoe from other lava types—there are no units that score high as rubbly pahoehoe flows and as another flow type. This is despite the fact that the ideal rubbly pahoehoe flow was similar to an ideal slab pahoehoe (Table T5). Finally, this figure also shows that the characterization of rubbly pahoehoe is incomplete. No lava unit scored less than -10% for rubbly pahoehoe. In other words, the characteristics that define rubbly pahoehoe are well described but the features which rubbly pahoehoe flows lack have yet to be identified. Only after more rubbly pahoehoe flows have been scrutinized will it be possible to make strong statements as to what features are never found in rubbly pahoehoe flows.