Improved collection of rooftop micrometeorites through optimized extraction methods: The Budel collection

The scientific value of micrometeorites collected from deep‐sea sediments or glacial deposits can be limited by poorly constrained accumulation times or severe alteration, coupled with a complex infrastructure of sampling expeditions. Collecting micrometeorites from rooftops has recently become a feasible alternative, but extraction methods have not been optimized or standardized to date. Here, we show that existing methods for the recovery of melted cosmic spherules (CSs) can be strongly improved by using a sequence of mineral separation techniques, including shape separation with an asymmetric vibrator and heavy liquid density separation with overflow centrifuges. We retrieved 1006 micrometeorites from the gutter of a barn in Budel, the Netherlands. Particle diameters are 80–515 μm, with the major mode at 130 μm and a slope exponent of −4.88. Differences in size distributions among various types of CSs indicate a multi‐source influx, with CS textures controlled by their parent body's mineralogy and orbital parameters. Repeated sampling of the rooftop after accumulation times of 959 and 333 days allows for a time‐integrated global mass flux estimate of 472 t year−1. This estimate is notably higher than previous rooftop‐based estimates but is still severely affected by micrometeorite loss from the gutter through drainage. The mass flux peaks at an equivalent particle diameter of ~200 μm. The Budel collection is the first rooftop collection to contain abundant vitreous micrometeorites and include the coarse‐grained S‐type CS class. Unmelted and I‐type micrometeorites remain difficult to extract from rooftop samples. Vitreous micrometeorites display various stages of weathering, showing that severe alteration of glass can progress at a faster rate in populated regions than previously assumed. This study demonstrates that methodological adjustments can drastically increase the scientific potential of rooftop micrometeorite collections.


INTRODUCTION
Originating from both asteroids and comets, micrometeorites ranging in size from~50 to 2000 μm are remnants of the extraterrestrial dust flux to the Earth (Genge et al., 2008;Taylor, Matrajt, et al., 2005). Depending on their size, mass, and entry velocity and angle, micrometeoroids entering the Earth's atmosphere experience varying degrees of heating, resulting in flash melting and a wide variety of quench textures, morphologies, and chemical compositions (e.g., Toppani et al., 2001). It is estimated that 90% of the cosmic dust mass flux captured by the Earth's gravity evaporates upon atmospheric entry (Love & Brownlee, 1991;Taylor et al., 1998). The remaining particles either melt and recrystallize, forming cosmic spherules (CSs), or retain their preatmospheric characteristics if temperatures remain sufficiently low, forming unmelted and scoriaceous micrometeorites (ScMMs; Taylor et al., 1998). Genge et al. (2008) used a combination of textures and chemical compositions to standardize the classification of micrometeorites. In general, CSs belong to one of three chemically differing groups: stony S-types, which are closest to chondritic in composition with depletion of volatile elements; Fe-rich I-types, which mainly consist of magnetite and wüstite; and G-types, which consist of abundant magnetite in a glassy mesostasis (Blanchard et al., 1980). S-type spherules are further subdivided by their textural and mineralogical features into coarsegrained (CG), porphyritic olivine (PO), barred olivine (BO), cryptocrystalline (CC), vitreous (V), and Ca-Al-Tirich (CAT). Unmelted micrometeorites are subdivided into fine-grained, CG, composite, refractory, and ultracarbonaceous (Dobric et al., 2009). Partially melted ScMMs, which are characterized by their irregular morphology, high porosity, and the presence of a magnetite rim, are either fine-or CG (Genge et al., 2008).
For the last decades, these various types of micrometeorites have been collected in environments where the accumulation of natural terrestrial and anthropogenic particles is limited, for example, deep-sea sediments, hot deserts, and the ice and glacial deposits of Greenland and Antarctica (Brase et al., 2021;Duprat et al., 2007;Genge et al., 2018;Goderis et al., 2020;Maurette et al., 1991;Prasad et al., 2013;Rudraswami et al., 2011;Taylor et al., 1998;van Ginneken et al., 2017;Yada et al., 2004). The micrometeorites derived from such sites often have poorly constrained accumulation times and areas, may have experienced severe alteration, or their collection requires logistically challenging sampling expeditions. The recovery of micrometeorites, especially CSs, from rooftops is a new developing branch in the field of micrometeorites studies and is rapidly establishing itself as an alternative to obtaining well-preserved CSs, although unmelted micrometeorites have yet to be recovered from rooftops (Genge, Larsen, et al., 2016;Suttle et al., 2021). The results of these studies, aided by various community projects (e.g., Blake et al., 2018;Esposito et al., 2020), have sparked the broad interest of many new citizen scientists, with growing numbers of private collections (e.g., Hasse, 2020;Larsen & Kihle, 2020;Peterson, 2020).
Typical anthropogenic contaminants in rooftop samples complicating micrometeorite extraction include industrial metallic and Fe-oxide spherules, ceramics, bitumen, and glassy spherules of various colors and compositions originating from road-surface markings, mineral wool, and fireworks (Hasse, 2020;Jonker et al., 2023;Larsen, 2017). So far, collection methods on rooftops have relied predominantly on the use of neodymium magnets to increase the proportion of micrometeorites relative to such terrestrial and anthropogenic particles. These initial magnet-based methods have proven successful, but lack the efficiency to obtain representative numbers of small micrometeorites (<150 μm) and vitreous spherules that are commonly nonmagnetic. This results in a non-negligible loss of micrometeorites and strongly biased rooftop micrometeorite collections (Genge, Larsen, et al., 2016;Suttle et al., 2021).
Here, we present a new approach to improve micrometeorite extraction from rooftop samples using state-of-the-art mineral separation techniques, which are used to isolate minerals with specific shapes, densities, and magnetic susceptibility properties from bulk rock or unconsolidated sediment samples in the field of Earth sciences (Faul & Davis, 1959;IJlst, 1973). In this study, the separation methods have been optimized for micrometeorite extraction and are successfully applied to retrieve a new rooftop micrometeorite collection. The new collection consists of 1006 micrometeorites recovered from a single rooftop gutter located in Budel, the Netherlands. All micrometeorites have been verified nondestructively using scanning electron microscope (SEM) imaging equipped with energy-dispersive X-ray spectroscopy (EDS). We demonstrate the efficiency of the newly developed methods by examining the size and density distributions of the new collection and provide insights into the characteristics of the present-day global micrometeorite flux.

Location and Sampling
The two samples for this study were collected under dry conditions from the gutter of a large barn in Budel, southeast Brabant (P. Beerten B.V.; 51°15 0 56.7 00 N, 5°35 0 55.8 00 E; constructed in 2006; Jonker et al., 2023), on June 16, 2021, and May 15, 2022. The owners last emptied the gutter on October 31, 2018. The respective accumulation times are thus 959 and 333 days. The roof of the barn has a total surface area of~3600 m 2 and contains two 105 m gutters. Access to part of the gutter was obstructed by silos; hence, a total of 190 m of gutter were sampled, linked to a collection area of~3220 m 2 . Within the sampled gutter section, 20 unenclosed drains are present, around which a large portion of the material had washed away. The accumulated sediment consists mostly of fine-grained eolian dust from the adjacent crop fields. In some sections of the gutter, the deposit was several centimeters thick and encrusted with a layer of organic-rich matter. Nine~5 m wide sections of the roof are covered with solar panels, creating a smooth tilted surface from which eolian dust is easily blown away. Thus, the largest accumulation of sediment occurred in the gutter underneath roof sections consisting of corrugated sheets. During the first sampling, most of the organic crust was discarded on-site using a sieve with a 4 mm mesh size. The first collected sample weighed roughly 50 kg. The second sample weighed~19.5 kg, including the organic crust, which was taken apart to analyze methodological loss. Thus, the combined total mass of the rooftop sediment in this work is~69.5 kg. Supplementary imagery regarding the methods outlined in the following sections is provided in Jonker et al. (2023).
Both samples were first cleaned to remove all organic matter. To minimize micrometeorite loss during decanting, small subsamples (~1 kg) were placed in ultrasonic baths for roughly an hour and frequently stirred to allow micrometeorites to detach from organic particles. Quarters of the subsamples were then transferred to a bowl and most of the organics and part of the clay and silt particles were removed using density-based separation techniques (agitation and stratification) similar to gold panning (Jonker et al., 2023). This process was executed a second time with~0.3 kg subsamples, 20 min of ultrasonic cleaning, and more rigorous stratification. This technique provides more control over decanting than repeatedly filling and decanting water from a large bucket (Larsen, 2019). The decanted material of the second step that was subsequently dried and sieved only contained solid grains <90 μm in size. This falls outside the target size range of this study (between 90 and 1000 μm). Smaller micrometeorites are frequently found in other environments but become increasingly hard to separate and distinguish from contaminant particles in rooftop samples. Rooftop studies, therefore, commonly focus on size fractions >100 μm (Genge, Larsen, et al., 2016;Suttle et al., 2021). The micrometeorite loss from decanting is thus substantial for sizes <90 μm, but not significant for the currently targeted size range.

Particle Separation
After cleaning, the first sample underwent various tests to obtain the highest micrometeorite yield. The sample was sieved to optimize the accuracy and effectiveness of later processing steps using mesh sizes of 90, 125, 150, 180, 200, 250, 355, 425, 500, 1000, 1600, and 2000 μm. In the size fractions above 1000 μm, no micrometeorites were encountered. The other fractions were processed on a Faultable, which is a pitched table with an asymmetric vibrator that separates particles based on sphericity (Jonker et al., 2023). The principle of the Faultable is described by Faul and Davis (1959). Although similar shape separation methods have been applied on a small scale in the past, the Faultable, originally designed for the extraction of micas, has not yet been used for micrometeorite extraction. The high sphericity, high density, and generally smooth surface of most micrometeorites, especially CSs, cause micrometeorites to concentrate in the spherical fractions. Both the spherical and subspherical fractions were collected to allow the recovery of scoriaceous and unmelted micrometeorites with irregular morphologies.
After separation using the Faultable, density separations were applied to size fractions 125-500 μm using a laboratory overflow centrifuge (LOC;IJlst, 1973;Jonker et al., 2023). Heavy liquid separation for micrometeorite extraction has rarely been applied in the past and only involved the use of funnels and a singledensity liquid, for example, by Suavet et al. (2009) and Prasad et al. (2013). Alternatively, density separation using water as a medium was performed by Suttle et al. (2021) on a rooftop sample, which involved decanting water before all particles in suspension had settled. However, this method is prone to inconsistency and will consequently remove micrometeorites with relatively low densities, which will have similar settling times as quartz. The principle of the LOC, developed at the Mineral Separation Laboratory at Vrije Universiteit Amsterdam, is based on a fast-spinning cup, in which a heavy liquid with the sample in suspension is introduced from above. Particles with a density lower than the heavy liquid will float and are flung out of the cup along with the excessive liquid. Particles with a higher density will sink and remain trapped inside the cup. The cup spins at 5000 rpm, which corresponds to about 700 times the force of gravity. This makes density separation with the LOC faster and more accurate than separation funnels which operate at 1G.
Two types of centrifuges were used in this study, LOC-50 and LOC-100, where the number indicates the maximum amount of sink in cm 3 . The heavy liquids used are diiodomethane and 1,2-dichlorobenzene with densities of 3.32 and 1.30 g mL −1 , respectively. Mixing these two liquids allows all particles with densities between 1.30 and 3.32 g cm −3 to be separated with precision within 0.02 g cm −3 . Two initial LOC-50 separations have been executed on all fractions between 125 and 500 μm with heavy liquid densities of 2.59 and 2.66 g mL −1 . This relatively small interval of just 0.07 g mL −1 allows the removal of all terrestrial quartz and most feldspar grains from the sample. These minerals with a density of 2.59-2.66 g cm −3 constitute about 90% of the total sample mass on average. Additional separations with densities >2.66 g mL −1 have been performed on fractions 125-250 μm to provide insight into the density ranges of micrometeorites. Heavy liquid densities used include 2.77, 2.90, 3.01, 3.12, and 3.30 g mL −1 .
For particles <140 μm, an additional controlled magnetic separation was performed using a neodymium magnet (48 mm diameter) and an Eclipse demagnetizer to remove residual magnetism of the particles. After Faultable separation, the 90-125 μm fraction was split in half. One portion was processed with the LOC-50 (2.66 g mL −1 and higher densities) and subsequently a neodymium magnet, the other with a Frantz magnetic separator (Jonker et al., 2023), and subsequently with the LOC-50 (2.66 g mL −1 ). These two batches yielded 99 and 64 micrometeorites, respectively, suggesting that the order of processing or the type of magnet used may affect the final yield.
For the density fraction <2.59 g cm −3 , microscopic identification of micrometeorites was complicated by an overwhelming number of terrestrial and anthropogenic particles, including residual organics and bitumen. As a result, only three micrometeorites were identified within this fraction, including one ScMM and a PO-and V-type spherule. In the fractions between 2.66 and 2.90 g cm −3 , 24 micrometeorites were found, mostly PO-and V-type spherules, accounting for 4.1% of micrometeorites >2.66 g cm −3 . For this reason, and because of the absence of positively identified unmelted micrometeorites, it was decided to focus primarily on denser CSs in the second sample.
The second sample was processed with the final optimized methodology (Figure 1). The sample was cleaned using the same methods as the first sample and subsequently sieved with a 90 and 1000 μm mesh. The <90 μm fraction contains nearly half of the sample mass after cleaning and is thus not viable to process by LOC and Faultable. It was, therefore, decided to discard this fraction in this study. The 90-1000 μm fraction was directly subjected to density separation using the larger LOC-100 with a preset density of 2.90 g mL −1 (typical for heavy mineral separation) and subsequent processing on the LOC-50 using densities of 3.01, 3.12, and 3.30 g mL −1 . Within the first sample,~5% of the micrometeorites were found in density fractions <2.90 g cm −3 . Thus, the use of a higher liquid density of 2.90 g mL −1 instead of 2.66 g mL −1 would have resulted in a~5% loss of micrometeorites in addition to other methodological losses due to, for example, sieving, washing, and microscopy. However, the use of the larger LOC-100 significantly reduced the sample processing time. Magnetic extraction on the <2.90 g cm −3 fraction could be used to recover most of the spherules that would otherwise be lost but was not performed in this study. Density fractions >2.90 g cm −3 were sieved, and the (sub) spherical portions were extracted using the Faultable. The mass remaining for microscopic assessment using these methods is generally in the order of 0.01% of the initial gutter sample mass after cleaning, in this case~4 gr. The sample mass reduction with each step is shown in Figure 1. For size fractions below 180 μm, a final magnetic extraction using a neodymium magnet was used to further reduce the sample mass. The application of magnetic extraction reduces the fraction mass by roughly a factor of 10. The total processing time using the final methodology involves 1 day of sample collection, a few days of sample cleaning, and roughly 2 days of particle separation (sieving, LOC, Faultable, and magnetic extraction). The time required for microscopy and SEM analyses strongly depends on the number of CSs present in the sample. Omitting the magnetic separation applied to the smallest size fractions is possible but would make microscopy considerably more time-consuming.
The organic crust that was separated from the second sample on site was cleaned and magnetically separated using a neodymium magnet. Within the magnetic fraction, two micrometeorites were identified with diameters of 108 and 126 μm. CS loss from discarding the organic crust is thus presumably small (<1%). However, this organic crust may act as a barrier preventing new falls to enter the deposit within the gutter, enhancing loss through drainage.

Microscopy, Verification, and Shape Analysis
Particles were studied using an Olympus stereomicroscope with KL 1500 HAL dual optical glass fiber lighting. Potential micrometeorites were identified based on a combination of textural and morphological criteria, including shape (spherical to subspherical with large protrusions), color (black, gray, white, brown, green, yellowish, and colorless), transparency (opaque to translucent or transparent), luster (vitreous, pearly, metallic, and dull), and surface texture (smooth to irregular with protruding crystals or magnetite dendrites). The presence of beads and vesicles was also used during particle selection. The particles were transferred to a carbon sticker for nondestructive analysis with a JEOL Neoscope II JCM-6000 benchtop SEM equipped with standardless energy-dispersive X-ray spectroscopy (EDS) at the VU. Images were taken using backscattered electron (BSE) imaging. Chemical analyses were done under lowvacuum conditions with an accelerating voltage of 15 kV, a high probe current, and a stage working distance of 19 mm. When applicable, EDS analyses were performed mostly on large and representative surface areas to obtain an average semiquantitative chemical composition representative of the entire particle. Spot analyses were employed in some cases, for example, when single crystals or platinum group nuggets (PGNs) were exposed. Additionally, 115 micrometeorites were mounted in epoxy resin, polished, and carbon-coated to study the interior with the SEM under high-vacuum conditions.
Verification of an extraterrestrial origin of a particle is provided by a combination of textural and morphological characteristics and a near-chondritic composition shown by EDS. Suttle et al. (2021) showed that exterior EDS data are inaccurate but sufficient to complement BSE images (BEI) for positive micrometeorite identification.
Major element abundances used for verification of an extraterrestrial origin are generally close to chondritic with O (30-50 wt%), Mg~Si (10-30 wt%), Al~Ca (<6 wt%), Fe (0-30 wt%), and depletion in volatile elements (Brownlee et al., 1997;Genge et al., 2008). However, some surface measurements may significantly deviate from these ranges, especially in the case of, for example, weathered spherules, wetting events, or magnetite rims, with surface compositions in a few cases varying up to an order of magnitude from CI chondrite values (Lodders, 2003). Other minor elements included in the analyses are Na, P, S, K, Ti, Cr, Mn, and Ni, and typically have (bulk) abundances <1 wt%. Cu and Zn were also analyzed as these elements sometimes indicate a terrestrial origin of contaminant particles, but their abundances are usually below detection limit in CSs. Since some terrestrial and anthropogenic contaminant particles mimic chondritic compositions, it is important to consider both the textural and chemical properties of potential micrometeorites. Particles with textures characteristic of CSs but with nonchondritic surface compositions were also considered, but most of these were later rejected due to a lack of definitive evidence. However, sectioning of some of these spherules showed the presence of relict minerals, providing evidence of an extraterrestrial origin.
In the case of some fully vitreous micrometeorites that lack crystalline domains or metallic beads, textural characteristics are virtually absent. In these cases, the verification relies solely on the near-chondritic composition and additional observations made with the optical microscope including particle color, transparency, and luster. Glassy spherules with nonchondritic compositions were not further considered in this study, since these would require additional trace element or isotopic analyses to be distinguished from glassy particles of anthropogenic origin.
The dimensions of each micrometeorite were determined using the measuring tool in the JEOL SEM software (ver. 2.4) by manually drawing lines representing the width and length. For each micrometeorite, a BEI was taken with a scale bar of 20, 50, 100, or 200 μm according to its size. Anticipating the high number of micrometeorites that are, and will be, recovered by the aforementioned extraction method, we developed a routine for shape analyses in MATLAB (2020) with the aid of the Image Processing Toolbox. The scale bar and micrometeorite were automatically detected in binarized images and particle shape parameters including width, length, and equivalent diameter through an ellipse fit were computed in μm. For successful shape analysis, the micrometeorite should not overlap with the scale bar in the BEIs, and the micrometeorite and surrounding field of view should be free of any dust particles and fibers. The width and length that were automatically derived are in good agreement with those obtained manually using the SEM software. When extreme misfits resulting from dirt are excluded, the percentage point deviation between automatically and manually obtained average diameters is characterized by a normal distribution with an average of 2.0 AE 5.3% (2σ).
Stepwise sample mass reduction using the final proposed methodology for the extraction of micrometeorites from rooftop samples. Separation methods are shown below; the discarded particles are mentioned above. The dark gray area indicates the approximate percentage of the sample remaining after each step (~19.5 kg total sample weight). Heavy liquid densities (2.90 g cm −3 herein) may be customized. Magnetic extraction using neodymium magnets, herein only applied to <180 μm size fractions, is optional but makes microscopy considerably less time-consuming. Magnetic extraction reduces the fraction mass by roughly a factor of 10. The final steps of the methods include microscopy, verification by SEM, and shape analyses using the MATLAB code.
The average minor overestimation of the automatically derived values is most pronounced in glassy particles, which are prone to charging effects that interfere with BSE imaging, resulting in images that are slightly stretched in the vertical direction. The automated approach is a less timeconsuming, more objective, and optimized way to derive shape properties. Shape analyses were done on exterior BSE images to obtain representative whole particle data, but the code also works for sectioned particle BSE images and optical true color images (edited with a dark background and white scale bar). This approach potentially also allows for shape analyses of multiple micrometeorites within a single image. The MATLAB code is available as a supplement in Jonker et al. (2023).

Classification, Abundances, and Relict Phases
During this study, over 1400 particles were analyzed, of which 1006 were identified as micrometeorites and classified using the classification system of Genge et al. (2008). The first sample (~50 kg; 959 days accumulation) yielded 728 micrometeorites, of which 714 are CSs (98.1%) and 14 are scoriaceous micrometeorites (ScMMs; 1.9%). The second sample (~19.5 kg; 333 days accumulation) yielded 278 micrometeorites, of which 266 are CSs (95.7%) and 12 are ScMMs (4.3%). ScMMs were provisionally classified based on a combination of textural and morphological features, including an irregular shape, high porosity, and the presence of a magnetite rim. No unmelted micrometeorites have been confirmed, but some non-sectioned micrometeorites with irregular morphologies may have been misidentified as ScMM or CG CS. Similarly, no G-type spherules were identified among the sectioned spherules, but these could be present among the non-sectioned spherules. I-type and CAT spherules are equally absent, as in most other rooftop micrometeorite collections (e.g., Genge, Larsen, et al., 2016;Hasse, 2020;Peterson, 2020;Suttle et al., 2021). Combined, the entire collection consists of 980 CSs (97.4%) and 26 ScMMs (2.6%). Six micrometeorites (0.6%) were found to contain a PGN. This anomalously low number compared to the~3% reported by Rudraswami et al. (2014) and Suttle et al. (2021) for Stype spherules could be explained by the comparatively low resolution of BSE images of the benchtop SEM, complicating submicron PGN identification.
The subdivision of CSs is primarily based on the classification scheme by Genge et al. (2008) but mostly relies on particle exterior BEIs. CG CSs, described as having >25% relict anhydrous minerals (Genge et al., 2008), are often characterized by an irregular morphology and commonly a magnetite rim, often with a snowflake-like habit (Figure 2c). The silicate melts exposed at the surface usually have comparatively low Mg/Fe ratios, caused by the preferential melting of Fe-rich phases at low temperatures (Toppani et al., 2001). Externally, CG spherules can be differentiated from ScMMs by their lower porosity and from PO spherules by their lack of protruding crystals. CG spherules were found in a previous rooftop study (supplementary data in Suttle et al., 2021), but have not yet been acknowledged as a distinct class among rooftop collections. They should be identified as such since they are valuable for relict mineral and isotopic studies. Their CG precursor is subordinate to the fine-grained precursor in the total cosmic dust flux (Taylor, Matrajt, et al., 2011), making the CG-type an important part of any unbiased micrometeorite collection to study the diversity of precursor materials.
The 115 polished sections were used to evaluate our provisional exterior-based classification. An additional rb-(relict-bearing) prefix was used for sectioned PO spherules (n = 28, 75.7% of total sectioned PO). Another 19 micrometeorites were reclassified, mostly PO to CG, and slight changes between CC textural groups. External identification of CG spherules remains difficult, but only one of the five sectioned CG spherules was reclassified as an rb-PO, while 10 PO were reclassified as CG. As such, the number of CSs classified as CG in the Budel collection most likely underestimates their true abundance.
The sectioned micrometeorites display a variety of relict anhydrous minerals. Representative chemical compositions of various relicts are given in Table 1. Common relict phases include forsterite, Fe-rich olivine with compositions up to Fa 30 , orthopyroxene with compositions En 84-96 , chromite, magnetite, and Fe-Ni  , top left). c, d) A CG spherule dominated by a single large Fe-rich olivine relict and a smaller potassium-rich feldspar relict (d, bottom right); the exterior has an irregular morphology and consists of an Fe-rich melt with abundant magnetite with a snowflake-like habit. e, f) rb-PO with several forsterite relicts, of which one contains small refractory inclusions rich in Al, Ca, and Ti (f, cropped detail); the surface shows the remains of a highly diluted magnetite rim and a chain of beads around the particle's perimeter, indicating rapid spin during atmospheric entry. g, h) rb-PO displaying a large dark spinel relict with a corona reaction rim and a cumulate texture with small forsterite relicts clustered at the bottom by inertial forces from atmospheric deceleration. i, j) PO with large skeletal olivine crystals protruding at the surface from a glassy matrix. k, l) BO with parallel olivine lamellae and characteristic dendritic magnetite at the surface. m, n) Microcrystalline CC with both barred and randomly oriented textural domains. o, p) Normal CC with two beads on opposite sides. q, r) Turtleback CC with several distinct crystalline domains. s, t) Weathered vitreous spherule with small unaltered crystalline protrusions, indicating the initial perimeter of the particle. BSE, backscattered electron; ScMM, scoriaceous micrometeorites; CG, coarse-grained; PO, porphyritic olivine; BO, barred olivine; CC, cryptocrystalline. droplets (kamacite and taenite). At least eight of the sectioned PO spherules display cumulate textures formed by relict forsterite grains, which accumulated at the leading face of the particle during atmospheric deceleration and served as nuclei for olivine phenocryst growth (Figure 2g, h; . Rare relict minerals that have been found include plagioclase (GMM451), refractory phases rich in Al, Ca, and Ti poikilitically enclosed in relict forsterite (GMM80; Figure 2e,f), and a large Mg-Al spinel (GMM268; Figure 2g,h). One CG spherule (GMM409; Figure 2c,d) was found to contain an extraordinary~10 μm potassium-rich feldspar with~13 wt% K 2 O (Table 1), surrounded by a magnetite-rich mesostasis and accompanied by a large Fe-rich olivine relict (Fo 87 ). To our knowledge, this is the first feldspar relict of its kind to be reported in a CS. It is likely associated with an achondritic precursor. Additionally, this micrometeorite shows microtails at its surface ( Figure 2c) similar to those observed on THMM465 (Hasse, 2020;Suttle et al., 2021).

Size Distribution
The 1006 micrometeorites of the Budel collection range in size from 80 to 515 μm and reflect an asymmetrical bimodal distribution with the dominant peak at~130 μm, a distinct shoulder peak around 180 μm, and minor peaks at 240 and~300 μm (Figure 4). Plotting PO, BO, and CC textural classes individually shows that the main peak is around 110 μm for PO, with a minor shoulder peak at 180 μm, and at~170 μm for BO. The CC-type is bimodal with the major mode at~130 μm, coinciding with the total collection peak, and a minor mode between 180 and 200 μm. Although this minor mode roughly coincides with the BO peak, it appears to be unrelated to the classification system using the microcrystalline CC group as an intermediate between BO and CC normal (Suttle & Folco, 2020), since all CC textural groups show secondary peaks around 180-200 μm (Figure 4). Another noteworthy observation is the difference in average diameter of turtleback CC between the first and second sample, being 200 AE 44 μm (n = 23, 1σ) and 147 AE 63 μm (n = 22, 1σ), respectively, resulting in the strongly bimodal distribution (Figure 4). Vitreous spherules appear to follow a similar trend, with average diameters of the first and second sample of 194 AE 69 μm (n = 29, 1σ) and 152 AE 37 μm (n = 7, 1σ), respectively. As such, the~180-200 μm shoulder peak is supported by all major textural classes. The 240 μm shoulder peak is only supported by PO and BO, while the 300 μm peak is caused primarily by PO. Whether these minor peaks are significant remains a matter of debate.
A power law trend line was fitted through the cumulative size distribution of the entire collection between 180 and 450 μm ( Figure 5). The 180 μm lower bound was chosen since most of the <180 μm size fractions have been processed using a magnet and thus cannot be regarded equally unbiased as the >180 μm fractions. The 450 μm upper bound is defined by the low count of larger micrometeorites. The power law trend line has a slope of −4.88 with an R 2 of 0.987. For comparison, we also fitted a trend line between 100 and 340 μm, the size range used by Suttle et al. (2021) for their Meppen rooftop collection, which provided a slope of −3.63 with an R 2 of 0.902.

Density Distribution and Time-Integrated Flux Estimate
A density distribution was constructed using linear interpolation of the density data for micrometeorites from the first sample (n = 585), ranging in size from 90 to 250 μm and having densities >2.66 g cm −3 (Figure 6). This provides only a general indication of density   (Taylor et al., 2000), regarded as one of the least biased collections, is shown for reference.
variation among subtypes within the Budel collection. ScMMs have the lowest densities with most particles ranging between~2.9 and~3.1 g cm −3 . A decrease in porosity leads to a density increase from ScMM to BO. BO-type spherules form the densest particles with 90% being >3.30 g cm −3 . A subsequent Fe-depletion and increase in vesicularity and abundance of glass from CC to V result in an overall decrease in density. No vitreous micrometeorites have been found with a density >3.30 g cm −3 . Data on the density fractions from which the individual particles were recovered are provided in the supplementary material (Jonker et al., 2023).
The mass of each micrometeorite has been estimated using the volume, deduced from the average diameter, and the average density of the textural class it belongs to, ranging between 3.04 and 3.29 g cm −3 . The average density for all types combined is~3.2 g cm −3 . These values may be an overestimation of the true values since particles with densities <2.66 g cm −3 are not included. Alternatively, the use of 3.3 g cm −3 as a maximum density in the calculations, with many BO spherules expected to have densities >3.3 g cm −3 , creates an underestimation of the true average value of spherules in the Budel collection. As such, our average density values

Separation Effectiveness
The new separation methods are inadequate to segregate industrial metallic and Fe-oxide spherules from I-type micrometeorites. The industrial particles are highly spherical, strongly magnetic, and have various densities due to a large range in vesicularity. They cannot be excluded from the sample using our Faultable or LOC techniques. As a result of this, no I-type CSs were identified in this study. We hypothesize that highprecision magnetic separation using Frantz magnets may be used to separate industrial from cosmic I-type spherules. Alternatively, I-type spherules, which have densities around 5.0 AE 0.5 g cm −3 (Feng et al., 2005), may be separated from industrial spherules using very highdensity liquids. The first sample contained abundant aluminum spherules after processing. Their size range is similar to CSs (~100 to 1000 μm) but their distinct silvery color, morphology, and metallic luster did not complicate microscopy. They differ from stratospheric aluminum oxide spherules mainly in terms of size (Brownlee, Ferry, et al., 1976). The origin of the aluminum spherules is most likely linked to the grinding and welding of aluminum frames during the installation of solar panels on the roof during the accumulation time of the first sample. Bitumen particles from asphalt have relatively low densities and thus only form an issue for density fractions <2.59 g cm −3 where CS abundances are already believed to be limited (Kohout et al., 2014). However, many porous scoriaceous and unmelted micrometeorites are expected to have densities <2.59 g cm −3 (Kohout et al., 2014). Although we managed to find several scoriaceous and potentially a few unmelted micrometeorites in denser fractions, many more will have been lost by discarding low-density fractions or were not identified during microscopy due to their resemblance to bitumen particles.
Additionally, the high concentration of organics in the sample required extensive cleaning. Although this has resulted in a minimal loss of CSs >90 μm in size, the loss of small unmelted micrometeorites would have been significant. Future attempts to find unmelted particles in rooftop sediments may resolve this issue by using hydrogen peroxide dissolution on small samples with low organic contents. This could make sample cleaning with water redundant.
Vitreous micrometeorites have relatively low densities compared to other types of CSs. Their densities range upward from 2.66 g cm −3 in the Budel collection, although less dense particles are also expected to occur. Despite an occasionally high vesicularity, their chondritic composition makes them comparatively denser than most anthropogenic, silica-dominated glassy spherules. Therefore, most contaminant glassy spherules have densities <2.66 and ≪ 2.90 g cm −3 and can be efficiently removed from the sample using LOC separation.
Roof conditions are an important factor to consider when searching for suitable sampling sites. Ideally, roofing does not add contamination to the sample, or the contribution should at least be known. It is important to identify the local major contributing natural and anthropogenic contaminants prior to sample processing to customize and optimize the extraction methods. We also recommend selecting pitched roofs with gutters rather than flat roofs, since these concentrate particles from a large collection area more effectively and allow the recovery of the entire sample from a roof, including very recent falls. Consequently, accumulation times can be determined more accurately.

Cosmic Spherule Abundances
Owing to the optimized methodology, the Budel collection is the first rooftop micrometeorite collection to contain abundant vitreous spherules (3.7%; Figure 3), although they are still severely underrepresented compared to the 17% abundance in the Antarctic South Pole water well (SPWW) reference collection (Taylor et al., 2000). Vitreous spherules remain difficult to separate and identify in rooftop samples due to their low to absent magnetic susceptibility and the many types of similar-looking anthropogenic glassy spherules. The proportion of ScMMs in the Budel collection (2.6%) is equal to that in the Meppen rooftop collection  and is similar to the Antarctic SPWW collection. The BO/CC abundance ratio of the Budel and Meppen collections are comparable, being 0.89 and 1.04, respectively, but lower than the Norwegian rooftop collection (1.45; Genge, Larsen, et al., 2016) and SPWW collection (3.42; Taylor et al., 2000). Genge, Larsen, et al. (2016) observed a difference between Antarctic collections (the blue ice-derived Cap Prudhomme and morainederived Larkman Nunatak site, both of which have high terrestrial ages), which have BO/CC ratios of~0.9, and collections with younger terrestrial ages, including the Antarctic SPWW collection and the Norwegian rooftop collection, which show much higher ratios. They hypothesized that this might be caused by gravitational perturbations resulting in temporal variations in the entry speeds of cosmic dust (Kortenkamp et al., 2001). Although this is a tempting idea, the Budel and Meppen collections contradict this statement with BO/CC ratios similar to the older Antarctic collections. Additionally, within any collection, these BO/CC ratios are strongly affected by methodological biases, the subjectivity of the classification, and whether the classification is interior or exterior-based, and should therefore be interpreted with care. Ratios between PO, V, CC, and BO are equally speculative since the abundances of PO-and V-type spherules, especially in rooftop collections, are strongly dependent on the methodology, the targeted grain size, and particle selection criteria. This is, for example, shown by the comparatively low PO abundance in earlier rooftop collections (Figure 3), where the recovery efficiency decreases for smaller micrometeorites, causing a predominant loss of PO (Figure 4).

Weathering of Rooftop Micrometeorites
In addition to emissions by heavy industry and transportation (including road traffic, public transport, shipping, and aviation), the Dutch agricultural industry releases vast amounts of reactive nitrogen into the environment. Upon contact with water, either on the ground or in the atmosphere, this forms nitric acid, resulting in widespread soil acidification. This is often referred to as the Dutch nitrogen crisis (van Damme et al., 2021). Since the adjacent crop fields are the major contributor to the sediment accumulated in the gutter, the gutter sample is expected to be equally acidified. These acidic conditions enhance chemical weathering. Although micrometeorites of the Budel collection likely have not been exposed to this environment for more than 2.63 terrestrial years (959 days, the longest sampling interval), we frequently observe micrometeorites, especially glassy, that display severe alteration by chemical weathering.
The changes in appearance from a smooth to scalloped shape caused by weathering of glass in micrometeorites have been reported before by Taylor et al. (2007) and van  but have only been shown in polished sections. Hence, the external appearance of such particles remains poorly documented, exacerbating collection bias during particle selection of rooftop samples. The vitreous spherules of the Budel collection range from fully pristine to severely altered and thus provide means to document the progressive stages of weathering ( Figure 8). Weathering of initially pristine glass (Figure 8a) usually starts as small pits and cracks scattered across the surface (Figure 8b). These gradually propagate inward and eventually cover the entire surface, leaving intact more resistant CC domains (Figure 8c,d). The newly-formed weathering rind is usually yellow to dark brown in color and chemically altered with strong depletion in magnesium and enrichment of other elements including potassium and phosphorus, mirroring Antarctic weathering (Kurat et al., 1994). Positive identification using exterior EDS is thus severely complicated in the case of highly weathered spherules and often requires the presence of small unaltered crystalline domains (Figure 8e,f). In a highly progressed stage, agitation may remove the weathering rind, revealing again a chemically unaltered but etched surface (Figure 8g) in which a new cycle of weathering may initiate (Figure 8h). GMM259 (Figure 8h) still retains small clumps of its earlier surface weathering rind, showing that the etched surface is formed by weathering in the terrestrial environment and not the result of frothing by degassing during atmospheric entry as suggested by Rudraswami et al. (2012). Additional optical color images documenting surface characteristics of weathered vitreous micrometeorites are available in the supplementary data (Jonker et al., 2023).
These highly weathered spherules are challenging to optically recognize and chemically verify. The Budel collection is therefore most likely biased toward lessweathered vitreous spherules. Nonetheless, extensively weathered vitreous spherules with complete weathering rinds or etched surfaces still represent almost a third of the population. Weathering is not limited to vitreous spherules but is also frequently observed in the glassy mesostasis of PO and BO spherules. In a few of these spherules, however, the glassy matrices remain intact, while faceted cavities represent former olivine crystals exposed at the surface . CC spherules are the most resistant and do not show any signs of alteration within our collection. Although physical surface alteration of CSs is in some cases substantial, polished sections of these particles show that the inward extent of the alteration is commonly limited (Figure 2t). Therefore, chemical and isotopic alteration of the interior of rooftop micrometeorites is believed to be equally limited, retaining the potential to provide pristine oxygen isotope and cosmogenic nuclide data .
Our findings on the weathering state of rooftop micrometeorites contradict those of Suttle et al. (2021). We argue that the deficit of significantly altered vitreous micrometeorites in earlier rooftop collections has at least in part been caused by the inability to recognize such particles as extraterrestrial, combined with their nonmagnetic properties. In addition to inadequate separation techniques, weathering may thus exacerbate the underrepresentation of vitreous micrometeorites in earlier rooftop collections. Suttle et al. (2021) reported a main peak in the size distribution of the Meppen rooftop collection at 160 μm. This is lower than many other, often weathering-or accumulation-biased, Antarctic or deep-sea collections, for example,~200 μm (Taylor et al., 1998), 200-250 μm (Prasad et al., 2013, 250 μm (Suavet et al., 2009;Suttle & Folco, 2020), and~400 to 500 μm (Goderis et al., 2020), and much lower than the major mode around 300 μm claimed by Larsen (2019) for rooftop micrometeorites in general. The occurrence of a 160 μm mode in the Larkman Nunatak collection (Genge et al., 2018) was used by Suttle et al. (2021) to argue that the 160 μm peak reflects the true natural mode. However, the Larkman Nunatak collection has been affected by the probable loss of smaller particles by winnowing (Genge et al., 2018). The particles of the Meppen collection were recovered by spreading out the sample on the roof and moving a magnet across at a distance of approximately 1 cm. Small particles are consequently more likely to stick to organics or remain trapped under the 2-3 mm thick layer of sediment . As a result, both~160 μm peaks are likely an overestimation of the true peak in the global flux. At 130 μm, the main peak reported herein for the Budel collection is lower, owing to the improved recovery in this size range.

Size Distribution Modes
The~130 μm peak in the Budel collection is presumably also still an overestimation in size. This is mostly related to micrometeorites becoming increasingly challenging to identify among anthropogenic contaminant particles at smaller scales. Micrometeorite loss from the gutter through drainage or winnowing will also have preferentially removed the smallest particles. We thus assume that the true natural mode in the size distribution might be less than 130 μm. Furthermore, the general scarcity of micrometeorites >500 μm in size in rooftop collections is due to the comparatively short accumulation times and the rareness of these giant micrometeorites (Rochette et al., 2008).
The~160 μm peaks observed in the Meppen, Larkman Nunatak, and Transantarctic Mountain (TAM65) collections  are unlikely to be correlated to the 180-200 μm shoulder peak of the Budel collection. Nonetheless, shoulder peaks are observed in many collections and remain a point of interest. The Budel collection shows that these shoulder peaks can be allocated to different CS textures (Figure 4). The general notion that larger particles experience a higher degree of atmospheric heating (Love & Brownlee, 1991;Toppani et al., 2001), combined with the idea that CS texture relates to the temperature of formation as PO < BO < CC (Genge et al., 2008;Rudraswami et al., 2020;Taylor et al., 2000;Taylor, Alexander, et al., 2005), implies that a similar order of peaks should occur in the size distributions of CS textures. The occurrence of the main BO peak at a diameter~40 μm larger than CC (Figure 4) contradicts these general notions and indicates that other factors (e.g., petrological and chemical properties or atmospheric entry parameters) play an important role in determining the final texture of cosmic dust particles (van Ginneken et al., 2017). Furthermore, Suttle and Folco (2020) correlated multiple peaks occurring in the size distribution to a variety of sources with unique normal distributions contributing to the total cosmic dust flux. Within this frame of reference, the distinct size distribution peaks of various CS textures and their interrelated (shoulder) peaks within the Budel collection may reflect the specific size distributions of different sources with unique compositions and orbital parameters.
The slope exponent of −4.88 obtained for the Budel collection from the cumulative size distribution between 180 and 450 μm is in close agreement with slopes reported for the Antarctic and deep-sea collections, ranging between −3.92 and −5.34. The slopes around −5.1 of the Antarctic SPWW collection are still generally regarded as the least biased (Genge et al., 2018;Goderis et al., 2020;Prasad et al., 2013;Suavet et al., 2009;Suttle & Folco, 2020;Taylor et al., 1998). The slope of −2.62 reported by Suttle et al. (2021) for the Meppen rooftop collection between 100 and 340 μm is markedly lower than the −3.63 of the Budel collection in the same size range. In this case, differences in accumulation dynamics and micrometeorite loss through weathering are assumed to be of minor significance, so the discrepancy is more likely related to the deficit of particles <150 μm in the Meppen collection resulting from sampling bias.

Micrometeorite Densities and Global Flux Estimates
Research into the densities of micrometeorites remains limited. Initial research into the physical properties of micrometeorites was mostly aimed at interplanetary dust particles (IDPs) with diameters between 5 and 15 μm (Brownlee, Horz, et al., 1976;Love et al., 1994;Nagel et al., 1976), providing little constraints for larger melted CSs and the effects of atmospheric heating. Reported bulk values derived for micrometeorites in general range from 0.6 to 5.5 g cm −3 , averaging around 2 g cm −3 (Kohout et al., 2014). Taylor, Jones, et al. (2011) used microtomography to calculate an average bulk density of 2.9 AE 1.1 g cm −3 for a collection consisting mostly of CSs and a few ScMMs. Using the same method, Kohout et al. (2014) derived an average bulk value of 3.2 AE 0.5 g cm −3 for S-type CSs, reporting uniformity among various types (BO, CC, and V). Suttle and Folco (2020) calculated the average density by approximating the volume from BEIs and measuring masses with a microbalance. They derived densities between 1.1 and 4.2 g cm −3 with a mean of 2.7 g cm −3 for S-type CSs and densities of 5.0 and 5.8 g cm −3 for two G-type spherules. The average density of 3.2 g cm −3 reported herein for CSs may be both an overestimation due to the general lack of particles with densities <2.66 g cm −3 or an underestimation due to the use of a maximum density of 3.30 g cm −3 in the calculation for the particles extracted from the >3.30 g cm −3 fractions. Nonetheless, our results for S-type CSs are in close agreement with Kohout et al. (2014), although we observe significant variation between various textural types ( Figure 6).
The mass flux of the Budel collection peaks around 200 μm equivalent diameter (Figure 7), corresponding to a single particle mass of~13 μg. This peak is in correspondence with earlier findings of Grün et al. (1985) and Love and Brownlee (1993), who found similar peaks around 200 μm in the mass flux distribution, deduced from lunar craters and LDEF (long duration exposure facility) satellite craters. Taylor et al. (1998) also reported a~200 μm diameter peak in the mass flux for the SPWW collection. These findings thus seem to support the idea that particles from the asteroid belt larger than this critical boundary are prone to collisions and fragmentation before Poynting-Robertson light drag pulls them toward the Earth (Love & Brownlee, 1993). By extrapolating the LDEF crater analysis, Love and Brownlee (1993) derived a total cosmic dust flux of 40,000 AE 20,000 t year −1 at 1 AU.
An estimated 90% of this mass flux evaporates within the Earth's atmosphere, resulting in 1600 AE 300 t year −1 accumulating on the Earth's surface of particles 50-700 μm, as deduced from the Antarctic SPWW collection (Taylor et al., 1998), or 2700 AE 1400 t year −1 when corrected for the deficit of unmelted micrometeorites. These numbers are arguably the most representative, since they are not influenced by, for example, loss through weathering, biased collection techniques, or poor accumulation time and area constraints, but still retain some level of uncertainty Taylor & Lever, 2001). Suttle and Folco (2020) derived a very similar flux of 1555 AE 753 t year −1 for the TAM65 collection. Rojas et al. (2021) provided slightly higher numbers of 3600 AE 1000 700 t year −1 for CSs from the CONCORDIA collection (Dome C, Antarctica) in the 12-700 μm range. Global micrometeorite fluxes have also been estimated by analyzing CS contents of deep-sea sediments, but these are generally an order of magnitude lower than the Antarctic estimates, commonly between 100 and 400 t year −1 (Prasad et al., 2013), caused by significant loss through weathering.
The global time-integrated flux of 472 t year −1 deduced from the Budel collection excludes deficits caused by, for example, methodological losses and the underrepresentation of vitreous and unmelted micrometeorites, and is at least 3× lower than the Antarctic estimates of 1600 AE 300 t year −1 for CSs in a similar size range (Taylor et al., 1998). We thus suspect that roughly 70% of the spherule mass that accumulated on the roof has not been recovered. While methodological losses are difficult to quantify, we believe that those are subordinate to the losses that occurred within the gutter through drainage and possibly winnowing. However, this new rooftop flux estimate is notably higher than the flux obtained from the Meppen rooftop collection by Suttle et al. (2021) of 13.4 t year −1 , or between 43.8 and 87.7 t year −1 when corrected for loss. Furthermore, our 30% recovery efficiency is a~300-fold improvement to the recovery efficiency of~0.1% in the Norwegian rooftop collection (Genge, Larsen, et al., 2016). This demonstrates that methodological improvements can drastically increase the yield of micrometeorites from rooftop sites.
The differences in flux obtained for the first and second sample of this study, 496 and 405 t year −1 , respectively, are exacerbated by the estimated additional 5% methodological loss related to the more timeefficient simplified density separation employed for the second sample. Extensive cleaning of the solar panels at the sampling site a month before resampling might also have increased loss through drainage. Considering these factors, the overall micrometeorite flux, or loss for that matter, seems to remain constant on an annual time scale.

CONCLUSIONS
The application of shape separation with the Faultable and heavy liquid density separation with LOCs for the extraction of CSs from rooftop samples has proven highly successful with a recovery efficiency of~30%, and has provided a large collection of S-type CSs. The Budel collection is the first rooftop micrometeorite collection to report on CG CSs, which allowed the discovery of a unique potassium-rich feldspar relict. The comparatively high abundance of PO spherules in the Budel collection has been attributed to the improved recovery of micrometeorites in the smallest size fractions. Unmelted and I-type micrometeorites remain difficult to extract from rooftop samples. Therefore, Antarctic, deep-sea, and stratospheric collections remain complementary to rooftop collections. In general, our results are broadly consistent with Antarctic micrometeorite collections, indicating no recent major changes in the global cosmic dust flux.
A MATLAB code was used to obtain particle parameters including average diameter through ellipse fit. The size distribution of the Budel collection has a comparatively low major mode at 130 μm and a slope exponent of −4.88, generally consistent with other unbiased collections. Individual CS textures display a diversity of size distributions with various major and minor modes inconsistent with a simple heating model, suggesting that CS textures and size distributions are governed by the petrological and chemical properties and orbital parameters of their parental sources.
Data on the size and average density of individual micrometeorites were used to estimate global micrometeorite fluxes. The time-integrated global flux of 472 t year −1 obtained from the Budel collection is about 3× lower than Antarctic estimates based on the SPWW and TAM65 collections. Loss of material through drainage is likely the major cause of this discrepancy. Nonetheless, our new flux estimate is an order of magnitude higher than previous estimates based on the Meppen rooftop collection. This demonstrates that improvements to the sampling and separation techniques can significantly increase the recovery of micrometeorites and lower the biases of rooftop micrometeorite collections. Future studies are encouraged to further experiment with the use of Faultable and LOC separation on other sample types.
Weathering of CSs within gutters and on rooftops was previously assumed to be insignificant owing to the young terrestrial ages of most samples. However, despite the short residence times in the gutter of up to 959 days, many glassy and some PO and BO spherules of the Budel micrometeorite collection display extensive weathering. Although not as severe as for some Antarctic or deep-sea collections, weathering probably biases rooftop collections and is in part the reason for a consistent underrepresentation of vitreous CSs in rooftop collections. Nonetheless, the scientific importance of rooftop micrometeorites is expected to continue to grow as more collections become available for petrographic, geochemical, and isotopic studies.