A CENSUS OF LARGE-SCALE (≥10 PC), VELOCITY-COHERENT, DENSE FILAMENTS IN THE NORTHERN GALACTIC PLANE: AUTOMATED IDENTIFICATION USING MINIMUM SPANNING TREE

Distance is determined by three methods, listed in decreasing order of robustness: (1) type P: trigonometric parallax measurements of associated masers from the BeSSeL project (Brunthaler et al. 2011, 2009; Moscadelli et al. 2009; Kurayama et al. 2011; Xu et al. 2011; Immer et al. 2013; Wu et al. 2014; A. Sanna 2016, private communication); (2) type ML: maximum likelihood distance from Bayesian evaluation of kinematic distance using external data to place priors (Ellsworth-Bowers et al. 2013, 2015); and (3) type KN, KF: near or far kinematic distance computed using the procedure of Reid et al. (2009), with updated Galactic parameters from Reid et al. (2014).

Wherever available, we use parallax distance, then Bayesian distance, then kinematic distance. Of the 54 filaments, 9 are assigned for type P, 38 for ML, 4 for KN, and 3 for KF distances. When more than one clump in a given filament has an ML distance, the median of the clump distances is used, computed after excluding extreme values. This occurs, for example, when a filament is generally IR-dark (therefore more likely to be located at near distance), but a minority of its clumps are IR-bright, so the far distance is assigned by the ML evaluation. Kinematic distance is used in seven cases. The distance ambiguity is resolved with the IR emission/extinction (see Figure 6) and the fact that all clumps in a given filament should have a consistent distance computed by the Reid et al. (2009) code.

We evaluate the temperature of every clump from three resources, listed in order of decreasing priority to search for a match: (1) gas kinetic temperature determined from ${\mathrm{NH}}_{3}$ for a subset of the BGPS sample (Dunham et al. 2011), (2) same but for a subset of the Atacama Pathfinder EXperiment Telescope Large Area Survey of the Galaxy (ATLASGAL) sample (Wienen et al. 2012), and (3) dust color temperature determined by comparing 350 $\mu {\rm{m}}$ and 1.1 mm fluxes for a subset of the BGPS sample (Merello et al. 2015). We exclude color temperatures with large uncertainties (>100%).

When querying for ${\mathrm{NH}}_{3}$ temperature the clump must match in position and velocity, and for color temperature only position is available for matching. If a clump does not have a match, we assume a temperature of 15 K based on the average value of BGPS clumps (Dunham et al. 2011). Among the 54 filaments, 46 have at least 1 temperature match. The minimum and maximum temperature values are listed in Table 1.

Figure 2 presents histograms of selected parameters. Panels (a)–(d) show mass, length, distance, and orientation angle for the filaments. Panels (e)–(h) present normalized histograms of $l,b,z,{R}_{\mathrm{gc}}$ where the distribution of filaments is compared to an appropriate BGPS sub-sample when data are available.

Zoom In Zoom Out Reset image size

Among the 54 identified filaments, F5 stands out as an extreme: 276 pc and $6.4\times {10}^{5}$ ${M}_{\odot }$. Excluding this extreme, the other 53 filaments extend 10–93 pc over the projected sky, with a mass of the order of ${10}^{3}\mbox{--}{10}^{5}$ ${M}_{\odot }$. For all 54 filaments, the angular length lies in the range of $0\buildrel{\circ}\over{.} 17\mbox{--}1\buildrel{\circ}\over{.} 50$, with a median of $0\buildrel{\circ}\over{.} 35$. The column density ${N}_{{{\rm{H}}}_{2}}$ lies in the range of $(0.2\mbox{--}16.9)\times {10}^{22}$ ${\mathrm{cm}}^{-2}$, with a median value of $0.9\times {10}^{22}$ ${\mathrm{cm}}^{-2}$, and volume density ${n}_{{{\rm{H}}}_{2}}$ is in the range $(0.4\mbox{--}22)\times {10}^{3}$ ${\mathrm{cm}}^{-3}$, with a median value of $1.8\times {10}^{3}$ ${\mathrm{cm}}^{-3}$. The filaments are 1–2 orders of magnitude denser than the filaments identified by the "mid-IR extinction" (Ragan et al. 2014; Zucker et al. 2015; Abreu-Vicente et al. 2016), while as dense as the filaments identified from Herschel far-IR emission by Wang et al. (2015).

The mass–length relation is best-fitted as $\mathrm{lg}L=0.41\,\times \mathrm{lg}M-0.19$. This suggests that the filaments are not N small unrelated filaments that appear to be accidentally connected. Suppose the sub-units have a typical length l and a typical mass m. The total length of the filament would then be $L=N\times l$, with mass $M=N\times m$, resulting in $L\sim M$ and not $L\sim {M}^{0.41}$ as observed here. The mass–length relation implies $M\sim {L}^{2.4}$, i.e., a fractal dimension of 2.4. This is comparable with the observed GRS molecular cloud catalog (although of whole clouds) in the Milky Way (Roman-Duval et al. 2010) and with numerical simulations of supersonic gas turbulence (Federrath et al. 2009). The mass–length scaling of filaments, from large to small-scales, will be discussed in detail in a forthcoming paper (A. Burkert et al. 2016, in preparation).

The orientation of the filaments is clearly not random. The distribution of the orientation angles θ is close to Gaussian ($K=-0.3$), showing a weak but significant concentration toward 0^◦. The vertical distribution z of the filaments is surprisingly not symmetric with respect to the Galactic plane ($S=-0.7$). It skews toward the negative values, with a mean at a positive value of z = 11.5 pc. This is likely an observational bias (see Section 5.2). Nevertheless, z concentrates toward small values, with 50% of the (27/54) filaments located within $| z| \leqslant 20$ pc and 70% (38/54) within $| z| \leqslant 30$ pc. Toward higher vertical positions the number of filaments decreases much faster than a Gaussian function (K = 2.8). Three parameters have a Gaussian-like distribution: the Galactocentric radius ${R}_{\mathrm{gc}}$ (K = 0.0), the clump velocity dispersion $\sigma (v)$ ($K=-0.01$), and the flux-weighted LSR velocity ${v}_{\mathrm{wt}}$ ($K=-0.3$). Notably, the mean velocity gradient along the filament is small (0.43 ±  0.31 km s⁻¹ pc⁻¹), but in broad agreement with simulations (Duarte-Cabral & Dobbs 2016).

Across the Galaxy, in general, the distribution of filaments follows the number density of BGPS sources, i.e., it is more likely to find a filament where the BGPS sources are crowed. Specifically, the probability density¹¹ of filaments (${P}_{\mathrm{Filament}}$) largely agrees with the probability density of BGPS sources: ${P}_{\mathrm{Filament}}\sim {P}_{\mathrm{BGPS}}$, as binned in $l,b,z,{R}_{\mathrm{gc}}$ (Figures 2(a)–(d)). This is expected because we used the BGPS spectroscopic catalog as input, which is a homogeneous sub-sample of the BGPS catalog. However, there are interesting exceptions. First, a significantly lower filament probability (${P}_{\mathrm{Filament}}\lesssim \,0.5\times {P}_{\mathrm{BGPS}}$) is seen toward $l\sim 30^\circ$, the Scutum tangent. The same discrepancy is seen toward $b\sim 0\buildrel{\circ}\over{.} 2$, $z\sim 20$ pc, and ${R}_{\mathrm{gc}}\sim 5\,\mathrm{kpc}$. Second, on the other hand, a much higher ${P}_{\mathrm{Filament}}$ is found in the inner Galaxy of $b\sim 10^\circ$, toward a Galactocentric radius of ${R}_{\mathrm{gc}}\sim 4$ and 8 kpc, and most significantly, toward a zero Galactic vertical scale height, z. Third, the averaged b and z for filaments are closer to zero than those of the BGPS sources. In another word, the filaments show a more symmetric distribution (than BGPS sources) with respect to the physical Galactic mid-plane.

Are filaments closer to the mid-plane more likely to align with the mid-plane, i.e., θ approaches 0°? The data do not indicate so (Figure 3): the Pearson's product-moment correlation coefficient between $| z|$ and $| \theta |$ is ${C}_{\mathrm{Pearson}}=+0.26$ or −0.35 for the filaments and bones, respectively, far from a linear correlation (+1 or −1).

Zoom In Zoom Out Reset image size

The statistical trends observed in these filaments provide excellent targets for quantitative tests with future theoretical calculations and numerical simulations.

Our MST algorithm finds some filaments previously identified by other methods. Our search field (Section 2) partially overlaps with previous searches by Ragan et al. (2014), Wang et al. (2015), Zucker et al. (2015), and Li et al. (2016).

Of the 9 cold and dense prominent filaments presented by Wang et al. (2015), 7 are in our search field, 3 of which are identified by MST (F7, F25, and F41 correspond to G11, G24, and G49, respectively). Others are not identified because of a lack of dense BGPS sources (G29, G47, and G64) or too large of a disruption in velocity space due to active star formation (G28).

Among the 10 "bone" candidates presented by Zucker et al. (2015), 6 are in our search field, 2 of which have dense BGPS sources: BC011.13–0.12 and BC024.95–0.17. The former corresponds to F7 (the Snake), and the latter is not identified by MST because of too large of a velocity disruption. In addition, our MST filament F28 is visible in their Figure 13 and seems to fulfill all their criteria, but was not identified by Zucker et al. (2015).

Among the 7 giant molecular filaments presented by Ragan et al. (2014), 6 are partially covered in our field ("partially" because most of those filaments extend beyond our coverage in $| b|$). F36 is a small dense part of GMF38.1-32.4a, but note that Ragan et al. (2014) used a kinematic distance of 3.3–3.7 kpc, a factor of 2 larger than the parallax distance of 1.56 kpc. F19 and F38 fall in the positional coverage of GMF20.0-17.9 and GMF41.0-41.3, respectively, but outside the velocity ranges.

Li et al. (2016) identified 20 filaments longer than 10 pc in our searching field using ATLASGAL continuum data, 6 of which partially match with our filaments. F7, F25, and F32 are part of G011.046-0.069, G023.985+0.479, and G030.315-0.154, respectively. The IR-dark part of F35 matches with G033.685-0.020. F33 and F37 match with G032.060+0.082 and G037.410-0.070, respectively. In the cases of F7, F25, and F32, Li et al. (2016) cataloged a longer angular extent than our filaments. However, the extra extent is not coherent in velocity.

In addition, F33 is the dense part of the "massive molecular filament" G32.02+0.06 presented by Battersby & Bally (2014) in a case study. F13 is part of the IRDC G14.225-0.506 (Busquet et al. 2013). F36 contains IRDC G034.43+00.24 (Kurayama et al. 2011; Sakai et al. 2015), and the full extent of the filament has also been reported recently by Xu et al. (2016).

F31 runs across a well-studied IRDC G28.34+0.06, also known as the "Dragon" nebula (Wang et al. 2011, 2012; Wang 2015). The IRDC is the IR-dark and submillimeter-bright arc, bent toward the bottom of the panel in Figure 6. The MST filament F31 is a new filament that runs across the IRDC at P1, where a protocluster is forming (Wang et al. 2011, 2012; Zhang et al. 2015). Interestingly, the clump-scale magnetic fields (Wang et al. 2012) are aligned with F31. At scales of the order of 10 pc, magnetic fields may be shaped by gravity, while on smaller scales (within 1 pc, or clump-scale), the magnetic fields control the formation of a secondary filament, as interpreted in Wang et al. (2012). The secondary filament is a small part of F31. Dust polarization observations of these filaments are needed to further investigate the role of magnetic fields on the formation and evolution of these filaments.

In summary, our MST method successfully finds previously known filaments where the criteria are satisfied. An important difference between the MST identified filaments and others is that the former contains dense clumps over the whole extent, while this is not the case for previously "by-eye" identified large filaments ("gaps" in velocity space are allowed). The MST method also finds filaments embedded in a crowded PPV space, which are difficult to isolate with human eyes (e.g., F31).

It is noteworthy that, because of the filamentary and hierarchical nature of the ISM, one can find an arbitrary number of filaments in the same data set using different criteria. Therefore, when presenting a filament sample, it is equally important to explicitly list the criteria used to define filaments. For the same reason, when comparing different samples of filaments one has to note the difference in criteria, otherwise the comparison is misleading.

The 54 filaments form the first comprehensive sample of large-scale velocity-coherent gas structures in the northern Galactic plane covered by the BGPS spectroscopic survey. The homogeneous sample allows us to investigate statistical trends (Section 4.3) for the first time. With length in the range of 10–276 pc and average column density above 10²¹ ${\mathrm{cm}}^{-2}$, these filaments are among the densest and largest structures observed in the Galaxy, and provide excellent tracers for Galactic structure and kinematics (e.g., Englmaier & Gerhard 1999; Dame et al. 2001; Dobbs et al. 2012; Reid et al. 2014; Smith et al. 2016; Vallée 2016).

The completeness of our filament sample largely depends on the data we use. The BGPS continuum catalog is 98% complete at the 0.4 Jy level (Rosolowsky et al. 2010). The spectroscopic catalog (Shirley et al. 2013) contains 50% of the sources in the survey coverage with dense gas lines detected. As our identification used the spectroscopic catalog, the results are biased to dense clumps. This is evident in the high averaged column density and high linear mass density. However, we emphasize that this is indeed our goal—we are interested in the most prominent dense filaments.

On the other hand, given the location of our Sun in the Galaxy, even homogeneous surveys like BGPS or ATLASGAL are biased to structures (a) closer to the Sun and (b) on the same side with respect to the mid-plane as the Sun (Figure 2; Schuller et al. 2009, but see discussion in Rosolowsky et al. 2010). As mentioned in Section 4.3, the large filaments show less bias than BGPS sources in z, but the distribution of z is indeed not perfectly symmetric. Should we mirror the distribution of $z\gt 0$ to $z\lt 0$, the total number of filaments would increase to 74. That is, 27% of the filaments may be missed due to this effect.

Although it has a much more improved completeness compared to previous methods, the MST approach cannot find all large filaments in our Galaxy. One of the main strengths of this method is the repeatability compared to manual approaches.

Most of the filaments are associated with major spiral arms (Figure 4), consistent with the observations by Wang et al. (2015). Many of them concentrate along the longitude–velocity tracks of the Scutum, Sagittarius, and Norma arms, and a few are associated with the Local arm, the Perseus arm, and one is associated with the Outer arm.¹² Only a small fraction (11/54, or 20%) of the filaments are not within ±5 km s⁻¹ of any arm structure, and are analogs of "spurs" observed in other galaxies.

How many large filaments exist in our Galaxy? Using our method, we have identified 48 filaments in the contiguous coverage of $7\buildrel{\circ}\over{.} 5\leqslant l\leqslant 90\buildrel{\circ}\over{.} 5$. It is reasonable to estimate a similar number of filaments in the fourth quadrant. In the outer Galaxy, the BGPS survey is targeted to several star formation regions, therefore the 6 identified filaments provide an extremely lower limit. Taking all these into account, and correcting for the bias as discussed in Section 5.2, we estimate about 200 velocity-coherent filaments longer than 10 pc and with a global column density above 10²¹ ${\mathrm{cm}}^{-2}$, as the filaments presented in this study.

How many filaments lie in the center of spiral arms and thus sketch out the "bones" of the Milky Way? Goodman et al. (2014) argued that the long and skinny IRDC "Nessie" lies in the center of the Scutum–Centaurus spiral arm in the (l, v) space, and within $z=\pm 20$ pc from the physical mid-plane. Following Goodman et al. (2014) and Zucker et al. (2015), our criteria for a "bone," on top of our large-filaments criteria (1–5) (Section 3), are:

• (6)  
  Lies in the very center of the physical Galactic mid-plane, with $| z| \leqslant 20$ pc.
• (7)  
  Runs almost parallel to arms in the projected sky, with $| \theta | \leqslant 30^\circ$.
• (8)  
  The flux-weighted LSR velocity ${v}_{\mathrm{wt}}$ is within ±5 km s⁻¹ from spiral arms.

However, the exact structure and position of the spiral arms in our Galaxy are not well-established. Diverse models have been derived from a variety of data ranging from atomic, molecular, ionized gas to stars and pulsars (e.g., Hou & Han 2014; Reid & Honma 2014; Vallée 2015, 2016). Here we have adopted the spiral segments derived from maser parallaxes (Reid et al. 2014; M. Reid et al. 2015, private communication), which have constrained distances. In Figure 4 we superpose the filaments on the spiral segments. Among the 54 filaments, 27 fulfill criteria (1–6), 21 fulfill criteria (1–7), and 13 of them also fulfill criterion (8). These 13 filaments (F2, F3, F7, F10, F13–F15, F18, F28, F29, F37, F38, and F48) are "bones" according to our definition. Our criteria for a bone are more strict than Zucker et al. (2015) in terms of velocity coherence and mean column density. When compared to other filaments, bones do not stand out in mass, length (Figures 2(a) and (b)), column/volume density, or temperature. All 13 bones are located in the first quadrant (which is not surprising given our search field; see Section 2), making 27% of the 48 filaments in the same region of blind BGPS survey. Obviously, owning to disagreement among the many spiral arm models, adopting a different model will lead to different "bones." But the fraction of bones in filaments should not change dramatically in a reasonable spiral model.

Zoom In Zoom Out Reset image size

Given the importance of filamentary geometry for enhancing massive clustered star formation, it is of great interest to quantify the fraction of the ISM contained in large filaments and to evaluate the star formation activities therein.

To address this question we consider only the range of $7\buildrel{\circ}\over{.} 5\leqslant l\leqslant 90\buildrel{\circ}\over{.} 5$ where the BGPS and its spectroscopic follow-up are contiguous. In this field there are 5841 BGPS v1 sources, 2893 having ${\mathrm{HCO}}^{+}$or ${{\rm{N}}}_{2}{{\rm{H}}}^{+}$(3–2) detection, which we refer to as "dense BGPS sources." We identified 48 filaments in this field, which are comprised of 512 BGPS sources. That means that 17.7% (512/2893) of dense BGPS sources, or 8.8% (512/5841) of all BGPS sources, are confined in large filaments. If we count BGPS clumps in the bones only, 6.8% of dense BGPS sources, or 3.4% of all BGPS sources, are confined in bones.

Compact 1.1 mm continuum emission of the BGPS sources outline the dense, inner part of much larger and less dense envelopes of molecular clouds (Dunham et al. 2011). Assuming a dense gas mass fraction of 10%–20% (Ragan et al. 2014; Ginsburg et al. 2015), we infer an order of 1% of the molecular ISM is confined in large filaments, and about 1/3 of this amount is confined in bones—marking spiral arm centers.

Wang et al. (2015) pointed out an apparent upper limit of 100 pc projected length for the longest filaments in their study designed to find cold and dense filaments based on Herschel far-IR emission. In this study, we use a different approach without limiting the temperature. Except for the extremely long filament F5, all other filaments are indeed shorter than 100 pc. Interestingly, this limit is also seen in Zucker et al. (2015) despite the different search method. The 100 pc limit seems to be present in filaments with a global column density above the order of ${10}^{21}\mbox{--}{10}^{22}$ ${\mathrm{cm}}^{-2}$. Relaxing the column density to a lower cut of $\lt {10}^{20}$ ${\mathrm{cm}}^{-2}$, longer filaments start to be picked up in ¹²CO/¹³CO (1–0): the 430 pc "optimistic Nessie" (Goodman et al. 2014); the 500 pc "wisp" (Li et al. 2013), and a few filaments by Ragan et al. (2014) and Abreu-Vicente et al. (2016). However, those CO filaments have much smaller aspect ratios (typically ≪10, see figures in their papers), and the low-J CO gas outlines the relatively diffuse envelopes of denser structures traced by MIR extinction. For example, one filament reported by Ragan et al. (2014) contains F36 as a small part (Section 5.1). Whether 100 pc is a true limit for dense filaments warrants further study. This provides a quantitative test case for numerical simulations (e.g., Falceta-Gonçalves et al. 2015).

So far, the 276 pc-long filament F5 (Figure 6) is the only exception of dense ($\gt {10}^{21}$ ${\mathrm{cm}}^{-2}$) filaments longer than 100 pc. Compared to the above mentioned extremely long CO filaments, F5 is at least 10 times denser in average column density, and it may also have less dense extensions, similar to the 80 pc "classic Nessie." More systematic searches and comparison to numerical simulations can resolve the true length limit of the longest filaments. We emphasize that the average column density is a crucial parameter for defining the boundary of filaments and thus the length and aspect ratio. It is also worthwhile to note that our filaments, as defined by a collection of dense BGPS clumps, form the center of larger and less dense structures.

The origin of large-scale velocity-coherent filaments is still a mystery. Numerical simulations of the multiphase ISM in galactic disks demonstrate that the cold, dense gas component tends to organize itself naturally into a filamentary network (e.g., Tasker & Tan 2009; Smith et al. 2014). Spiral arms can sweep up and compress gas, generating bones (Goodman et al. 2014). Gravitationally unstable disk regions condense into gaseous rings and arcs (Behrendt et al. 2015). In differentially rotating disks, structures like molecular cloud complexes could be sheared into elongated filaments.

Our calculations (A. Burkert et al. 2016, in preparation) show that tidal effects of the Milky Way are too weak to affect the maximum length of filaments. This is consistent with our observations, where we find filament lengths do not correlate with Galactocentric radii (${C}_{\mathrm{Pearson}}=-0.07$). The maximum filament length of the order of 100 pc might be related to the timescale of ${\tau }_{\mathrm{SF}}={10}^{7}\,\mathrm{years}$ (e.g., Burkert & Hartmann 2013) on which stars form in dense molecular gas and destroy their environments. Typical turbulent velocities on large scales in galactic disks are of the order of $\sigma =10$ km s⁻¹ (Dib et al. 2006). If σ is the maximum velocity with which coherent filaments can grow and if ${\tau }_{\mathrm{SF}}$ denotes the timescale on which they are destroyed again, their length is limited by $l=\sigma \times {\tau }_{\mathrm{SF}}=100$ pc, in agreement with the observations.

By definition, the filaments presented in this study are in the form of a chain of dense clumps physically connected by less dense gas in between. For linear filaments, this geometry resembles a fragmented "cylinder" with regularly spaced clumps under the "sausage instability" of self-gravity (e.g., F9 in Figure 6). According to the framework of Chandrasekhar & Fermi (1953), an isothermal gas cylinder becomes super-critical when its linear mass density exceeds the critical value ${({\text{}}M/L)}_{\mathrm{crit}}$, and will fragment into a chain of equally spaced fragments with a spacing of ${\lambda }_{\mathrm{cl}}$, with each fragment having a mass of ${M}_{\mathrm{cl}}={({\text{}}M/L)}_{\mathrm{crit}}\times {\lambda }_{\mathrm{cl}}$. In short, the fragmentation is governed by central density and pressure (thermal plus non-thermal). This framework has been followed by many authors (e.g., Ostriker 1964; Nagasawa 1987; Bastien et al. 1991; Inutsuka & Miyama 1992; Fischera & Martin 2012). See Wang et al. (2011, 2014) for a useful deduction of the formulas. Figure 5 (left panel) plots the mean clump mass of each filament with the mean separation between clumps (mean length of the edges in the filament). The observed fragmentation is consistent with the theoretical prediction of cylindrical fragmentation, assuming a central density of $1\times {10}^{4}$ ${\mathrm{cm}}^{-3}$ and a velocity dispersion of 0.4–2.2 km s⁻¹ (magenta line). The spread of the data points around the prediction line may be due to a range of central densities and imperfect cylinder geometry. Recent numerical simulations have shown that geometric bending, which is often seen in observed filaments, can change the regularity of the spacing (Gritschneder et al. 2016), indicating that more theoretical work is required in order to understand the stability and dynamics of filaments.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Dense clumps with a typical mass of 10³ ${M}_{\odot }$ and typical size of 1 pc (Figure 5, left panel) are, in general, capable of forming a cluster of stars. Statistically, dense clumps residing within large filaments are slightly denser than clumps elsewhere (see below). In Figure 5 (right panel), we plot clump mass versus deconvolved radius (not all BGPS clumps have a valid radius, see Section 4) for three categories of BGPS clumps: I—the 1710 clumps with well determined distance from Ellsworth-Bowers et al. (2015). II—the 294 clumps in velocity-coherent structures but not large filaments. III—the 469 clumps in large filaments. Categories I, II, and III have 41.0%, 39.8%, and 46.2% clumps, respectively, satisfying the Kauffmann & Pillai (2010) threshold of forming massive stars. Thus, categories I and II are indistinguishable, while in comparison, category III is slightly more favorable for massive star formation. If we count in mass instead of number of clumps, the fractions are 79.2% for BGPS sources, 86.3% for velocity-coherent structures but not large filaments, and 91.0% for large filaments. Surprisingly, bones do not show a higher fraction compared to large filaments, either counted in number or mass. This indicates that local environment such as a velocity-coherent filament plays a role in enhancing massive star formation. Filaments, in particular, provide a preferred form of geometry to channel mass flows that can inhomogeneously feed star-forming clumps (e.g., Arzoumanian et al. 2011; Peretto et al. 2014; Zhang et al. 2015; Federrath 2016; Heigl et al. 2016). On the other hand, the Galactic environment does not seem to affect local star formation across the few hundred parsecs spread of vertical position z, consistent with previous studies (Eden et al. 2012, 2013).