Blowouts are aeolian features consisting of an erosional depression and an associated downwind depositional lobe or apron. They play a critical role in diagnosing landscape changes, acting as an initial source of sand in reactivating dune fields as well as supplying sediment to downwind features (Barchyn and Hugenholtz, 2013). Blowouts are found in coastal, semi-arid, sandy grassland, and desert landscapes (Hugenholtz and Wolfe, 2006; Hesp and Walker, 2012) and are found across the world, including Europe (van Boxel et al., 1997; Käyhkö, 2007; Gonzlez-Villanueva et al., 2013; Smyth et al., 2013; García-Romero et al., 2019), North America (Fox et al., 2012; Abhar et al., 2015; Garès and Pease, 2015), Africa (Lancaster, 1986), and Asia (Sun et al., 2016; Kang et al., 2017; Luo et al., 2019a). Blowout morphology varies, with scour hollows being classified as saucer, bowl, or trough-shaped and the eroded sediment deposited immediately downwind as either a large lobe or a thinly spread layer of sediment called an apron (Smyth et al., 2020).

Blowouts develop when a stabilised dune’s surface is disturbed, exposing the underlying sand to wind erosion. A positive feedback mechanism is then initiated with the reduction in surface roughness facilitating an increase in wind velocity, allowing for further aeolian deflation to occur (Hesp and Walker, 2012). Additional erosion and subsequent destabilisation of the surface, including any anchorage vegetation, then leads to further reductions in surface roughness, and an increasing rate of deflation. Over time, a small area of exposed sand may erode to form a large hollow, supplying sediment to the downwind environment (Luo et al., 2019a). Once initiated, the positive feedback mechanism is self-sustaining and can continue even if conditions are no longer conducive to blowout initialisation (Käyhkö, 2007). Subsequent morphologic development can be restricted by physical characteristics such as the size of the original stabilised dune, a layer of calcrete, or an armoured surface (Hesp, 2002). Seasonal changes, such as the height of the water table, surface moisture levels, or the magnitude and direction of prevailing winds, can also limit blowout development (Hugenholtz and Wolfe, 2006; Davidson-Arnott et al., 2008; Hesp and Walker, 2012). Blowout stabilization requires primary succession of pioneer flora to re-establish vegetation cover, mitigating blowout development by increasing surface roughness and decreasing bed shear stress (Schwarz et al., 2018).

The initial exposure of underlying sediment to wind erosion can occur through natural processes as well as anthropogenic activity. Natural factors which lead to blowout formation include coastal wave erosion, storm events, vegetation cover change, and increased wind speeds (Barchyn and Hugenholtz, 2013; Smyth et al., 2013; Hesp et al., 2016). Biotic factors associated with the grazing pressures from wild fauna, such as rabbits and bison, have also been found to influence blowout formation (van Boxel et al., 1997; Drees and Olff, 2001; Fox et al., 2012). Natural processes may occur seasonally, with reduced efficacy in months with more precipitation or a weaker wind regime (Gonzlez-Villanueva et al., 2013; Abhar et al., 2015). In addition, anthropogenic activities can cause blowout initiation through direct and indirect actions. Off-road vehicle disturbance, pathway trampling, and poor land-use practices can remove anchorage vegetation facilitating blowout genesis (Hesp et al., 2010; Acosta et al., 2013; Houser et al., 2013; Jewell et al., 2014). Pastoral farming poses a dual threat of increased grazing pressures as well as high-intensity trampling involved with moving livestock between paddocks (Blanco et al., 2008; van der Hagen et al., 2020). Indirectly, infrastructure projects and urbanisation may also divert wind patterns, changing the spatiality of sand transport mechanisms and influencing the formation of deflation landforms (García-Romero et al., 2019; Wernette et al., 2020). Abiotic and biotic factors rarely work independently (Corenblit et al., 2015). As such, blowout initiation may be exacerbated by a number of processes at any given study location.

Measuring up to 40 m deep and 1,000 m long (Luo et al., 2019a), the blowouts in Gonghe Basin, Qinghai-Tibet Plateau (QTP) distinguish themselves from previously studied, smaller features in other parts of the world (Hugenholtz and Wolfe, 2006). In comparison to their coastal counterparts, the mega-blowouts of the QTP defy conventional models of blowout development, exceeding a size in which coastal blowouts would tend towards stabilisation (Hugenholtz and Wolfe, 2009). Yet the scarcity of detailed morphological investigations into blowouts on the QTP means that much remains unknown about these mega features. Of particular interest is the difference in volume between the depositional lobes and erosional scour hollows. An inequality in their mass balance would quantify the amount of sediment lost to the surrounding area. This is either as a fine layer extending beyond the depositional lobe or as atmospherically entrained dust.

In this paper we present morphometrics and a mass balance for a range of blowouts on the QTP. An innovative methodology which combines ground-penetrating radar (GPR) surveys, Structure-from-Motion (SfM) modeling from optical drone footage, and geospatial analysis is used to determine a mass balance between the erosional hollow and depositional lobe of three blowouts. A further two compound mega-blowouts are also assessed. Such quantifications are vital to address both the lack of detailed morphometrics for the full range of blowout features and the lack of parametrization in sediment reactivation processes. This research focuses on the Gonghe Basin on the QTP, but the concepts and methodology can be applied elsewhere.

The Gonghe Basin on the QTP in China is an important region for agriculture, pastoral farming, and power production. The semi-arid steppe is comprised of grassland, farmland, and semi-shifting dunes (Zhang et al., 2003; Yu and Jia, 2014). In recent years, the cold alpine desert region of the Gonghe Basin has experienced severe land degradation from anthropogenic activities and local climatic warming (Wang et al., 2017). Aeolian sand reactivation, predominantly from wind erosion, has substantially eroded cultivated farmlands (Dong et al., 1993; Zhang et al., 2003) and proceeds to threaten downwind infrastructure projects from the risk of sand encroachment. Of particular note, the Longyangxia Dam and Solar Power Park are located downwind from such eroded farmlands in Gonghe Basin. Attempts have been made to reduce soil degradation by enclosing grasslands with fencing; however, unprotected areas remain overgrazed and the basin continues to be cultivated (Zhang et al., 2003). The threat of further wind erosion is therefore still present. Despite this, the morphology, distributions, and impacts of wind erosion and blowout development in the Gonghe Basin remains a largely understudied phenomena (Luo et al., 2019a).

The mega-blowouts of the QTP are some of the largest on Earth (Luo et al., 2020) and are themselves part of a wider sand flow system from the Gonghe Basin to the Yellow River valley. The Gonghe Basin (Figure 1) covers an area of 13,800 km² (Liu et al., 2013) and is a northeast to southwest aligned endorheic basin flanked by the Qinghai Nanshan mountains to the north and the Wahongshan Mountains to the south. There is a remarkable coincidence between the orientation of the valley, the mountains on either side, and the winds. Winds blowing along the axis of the basin have caused widespread erosion (Zhang et al., 2003; Luo et al., 2019a). Sand deflated from the valley has accumulated at the downwind end of the valley forming a gently inclined plain that rises around 450 m over 50 km, a gradient of about 1:100, before descending very steeply into the valley of the Yellow River that is around 620 m deep. Across much of the plain the sand-covered slope is vegetated and described as an “alpine meadow” (Yu et al., 2017). The topography includes northeast to southwest elongated ridges of sand that are possibly vegetated linear dunes.

The Gonghe Basin experiences average annual temperatures of 2.4—4.1°C and an annual average precipitation of 314.3—414.8 mm (Liu et al., 2013). Approximately 80% of this rainfall occurs between June and September (Yu and Jia, 2014; Yu et al., 2017). Vegetation cover during this wet season increases to a maximum of 20% but remains below 5% during the rest of the year (Wang et al., 2018). The cold alpine desert region in the northeastern QTP is an unusual aeolian environment due to its low temperature and low air pressure (Lu et al., 2005). Sitting at 3750 m above sea level and at the intersection of the westerlies wind belt, Asian, and Indian monsoon, the fragile ecosystem is extremely sensitive to global climate change (Qiang et al., 2016; Luo et al., 2019b). Strong northerly and northwest winds of up to 40 ms⁻¹ occur from February to May (Qiang et al., 2013) with average annual speeds ranging from 2.1–2.7 ms⁻¹ (Yan and Shi, 2004; Yu and Jia, 2014).

The Longyangxia Power Park is located within the Basin and is comprised of a hydroelectric dam and a solar power farm. It was the largest photovoltaic power station in the world until 2017 (Gautam et al., 2017) and generates an average of 5.94 GW h hydroelectric power and 0.48 GW h photovoltaic power a year (An et al., 2015). Aeolian sediment transport in the basin poses three main threats to the Longyangxia Power Park, including the burial of solar panels, obscuration of solar panels by dust, and the in-filling of the Longyangxia Dam Reservoir.

Five blowouts were surveyed in the Gonghe Basin (Table 1), located approximately 50 km south-west of Qabqa, Gonghe County, Qinghai Province (Figure 1). Blowouts A, B, and C are located within grazed pastures. Vehicular disturbance and animal activity in the area contribute heavily to blowout initiation. Access to the gated pastures intensifies trampling pressures at egresses and channels vehicles through corridors between paddocks. Additionally, small depressions in the landscape are used by cattle, either for shelter of dust bathing, causing similar fauna pressures on the landscape to those described in Canadian prairies by Fox et al. (2012). Downwind, the size of the blowouts increases until the sand cover increases to form a field of transverse and barchanoid dunes that migrate into the valley of the Yellow River. The larger, more complex blowouts D and E are close to the interface between the blowouts and the active sand dunes and appear to be developed within vegetation-stabilised parabolic dunes. Blowouts B, D, and E were previously studied by Luo et al. (2019a), who looked at their development and growth from 1967 to 2015 with historical satellite imagery, and monitored the expansion of their erosional hollows with annual surveys from 2015 to 2018 as well as drone images (Luo et al., 2020). Here we present updated morphometrics for these blowouts using finer-resolution DEM reconstructions and imaging of the internal structures within depositional lobes. The five blowouts exemplify the range of features found in the Gonghe Basin and are discussed in order of ascending complexity.

This study employs an innovative combination of surface and subsurface imaging. SfM optical drone mapping was combined with GPR surveys to provide a high-resolution, non-invasive approach that leaves the fragile landscape undisturbed. We extend the technical work of Luo et al. (2020) to provide geomorphometry and geospatial analysis of the Gonghe Basin using a consistent suit of SfM data. The fieldwork for this study was conducted in July 2018.

Two Phantom four drones were used to survey the full extent of the blowouts. Images were first collected at nadir angles with additional images then captured inside the scour hollows to ensure lateral coverage. A Digital Elevation Model (DEM) and ortho-mosaic were created for each blowout feature using the established SfM approach within Agisoft Photoscan Professional Edition (Scarelli et al., 2016; Lin et al., 2019). Firstly, the optical drone images were aligned and a dense point cloud calculated. Outlying points within the dense clouds were removed manually and a mesh built from the cleaned dataset. This mesh was converted into a DEM and an orthorectified true colour composite texture of the optical drone images was created. Ground control points were recorded using a differential GPS, allowing the SfM data to be precisely georeferenced. Flight details and DEM reconstruction metrics are listed in the Supplementary Material (Supplementary Table S4). The three-dimensional models of the study areas were then used to determine the erosional hollow boundary and lobe extent for each blowout feature where it could be discerned. The elevation of the erosional hollow boundary was sampled at 0.2 m intervals and an original surface interpolated using ordinary kriging. This surface was differenced from the DEM to determine the volume of the erosional hollow.

GPR surveys have been successfully conducted to investigate the internal structure of aeolian features (Hugenholtz et al., 2008; Bristow et al., 2010), but relatively few studies have combined the technique with remote sensing to analyse blowout features (Neal and Roberts, 2001; González-Villanueva et al., 2011; Jewell et al., 2017). The low conductivity of sand and large-scale structures contained within aeolian features lend themselves well to GPR surveys (Bristow, 2009) which use the transmission of high-frequency electromagnetic fields to image subsurface structures. The GPR profiles within this paper were collected using 200 MHz antennas, a step size of 0.2 m, and an antenna separation of 1 m. CMP surveys indicated a velocity of 0.08 mns⁻¹, that of damp sand. This corresponds well with field observations as short rainfall events occurred frequently throughout the duration of the fieldwork. Similar to the drone mapping, a differential GPS was used to precisely co-locate the GPR survey transects. Topographic corrections were reconstructed from the SfM-determined DEMs. Where a depositional lobe was evident, its depth was traced and extracted at the same resolution as the step size (0.2 m). Ordinary kriging was then used to interpolate a basal surface of the lobe, which was subtracted from the DEM to determine lobe volume.

The uncertainty of the volumetric calculations was determined by the GPR parameters. The signal had a wavelength of 0.4 m, and thus a resolution one quarter that of the wavelength, 0.1 m. The volumetric uncertainty (m³) was therefore calculated by multiplying the area of the lobe footprint (m²) by ±0.05 m.

Interpretation of the GPR profiles follows the radar stratigraphy and radar facies methodology (Gawthorpe et al., 1993; Neal and Roberts, 2001; Jol and Bristow, 2003), where reflections on the GPR profile are interpreted as stratigraphic horizons and reflection terminations mark breaks in deposition or truncation due to erosion. The top of the soil horizon at the base of the lobe forms a continuous high-amplitude sub-horizontal reflection at Blowout A (Figures 2, 3).

At Blowout B the basal reflection has a very low-angle dip downwind parallel to the blowout lobe and a relatively steep slope to the west perpendicular to the lobe (Figures 4, 5), because the lobe runs along the flank of a vegetated linear dune. The GPR profile along the axis of the lobe at Blowout A shows inclined reflections that dip in the downwind direction (Figure 2). These are interpreted as former positions of the lobe surface and record the extension of the lobe downwind. Changes in the dip of the reflections suggest that the downwind margin of the lobe steepened and then flattened out. The dip of the reflections shows that the downwind margin did not reach the angle of repose. Truncation of the reflections within the lobe indicate periods of erosion, when the lobe was reshaped by changes in the wind. Reflections on the GPR profiles across the lobe are dominated by convex reflections that are interpreted to come from former positions of the lobe surface (Figure 3).

The convex morphology mirrors the lobe surface at the time of survey. However, the convex reflections are not symmetrical; some are truncated and others are only preserved on the east side of the lobe. Truncation of the reflections marks erosion, most notably at the upwind end of the lobe where a central scour has eroded into the upwind end of the lobe. Truncation of reflections within the lobe are interpreted as reactivation surfaces where the lobe was reshaped by a change in the wind. The asymmetric development of the lobe with local advance towards the west indicates a change in wind direction.

The presence of downlap and truncation surfaces within the lobes indicates reshaping of the lobe surface due to changes in the wind strength and direction similar to reactivation surfaces identified in sand dunes (Kocurek, 1996) as well as GPR profiles across sand dunes (e.g., Bristow et al., 2005). The abundance of truncation and downlap terminations within the lobe strata indicates extensive reworking of the lobe, with episodes of erosion as well as periods of deposition.

Barchyn and Hugenholtz (2013) suggest that the erosion in a blowout will equal the volume of sediment in the depositional apron due to the conservation of mass. However, we are not aware of any other studies of blowout deflation hollow/depositional lobe volumes with which to compare our results. As a consequence, we can only speculate that the differences observed in this study are not unique but should apply more widely. Barchyn and Hugenholtz (2013) assume that the apron is 100% efficient in trapping wind-blown sand. By analogy with sand dunes it has been shown that dunes with slipfaces are not very efficient sand traps, trapping only a small percentage of the saltation flux, and dunes without slipfaces even less so (Bristow and Lancaster, 2004). The clear implication is that sand derived from deflation hollows can accumulate locally within depositional lobes, but equally it will be distributed more widely downwind. This combination of local sand accumulation and downwind reworking can account for most of the variations in morphology observed in the blowout lobes. The other significant influence is the wind direction, which can affect the direction of sediment transport and resulting lobe development.

Truncation of reflections on the GPR profiles provide evidence for erosion and indicates breaks in deposition, as well as reworking of the sand on the surface of the blowout lobe. Changes in the accretion direction of reflections indicates changes in the lobe morphology that is attributed to changes in wind direction. GPR also images the base of the lobes and this has been used to calculate the volume of sediment within two lobes. While the results of the GPR survey in imaging the lobe strata and the base of the lobe are positive, the calculation of lobe volumes using GPR was limited by the inability of the 200 MHz GPR system to image sand layers less than 0.5 m thickness, the section that is covered by arrival of the direct air and ground waves. Higher-frequency GPR systems e.g. 500 MHz antennas with a shorter wavelength would have helped in this respect. The use of soil pits to determine lobe thickness and volume shows a much closer agreement between the volume of the scour hollow and the volume of the depositional lobe at location C. Nonetheless, the kriging interpolation of these point measurements is associated with large statistical uncertainty (Table 2).

It would be advantageous for future work to further constrain depositional volumes. There exists, however, two difficulties in this. Firstly, an invasive field campaign risks disturbing the fragile environment and contributing to further land degradation. Secondly, the attribution of widely spread fine material to a particular blowout becomes problematic further downwind. Delimiting the input of upwind features to the amount of sediment in the topsoil surrounding a blowout would require a comprehensive monitoring campaign and/or foreknowledge of sediment emissivity. We have shown that a large amount of material is lost to the surrounding downwind environment, and thus features further downwind are unlikely to be acting in a closed system. The proximity of other blowouts and their level of activity would therefore need to be carefully considered when constraining depositional volumes further. This issue notwithstanding, repeated drone surveys could be conducted to determine lobe accretion and scour deflation through time. While sandy environments pose challenges to photogrammetric methods, studies quantifying sand deposition on beaches using UAV surveys have produced DEMs with centimetre and sub-centimetre vertical accuracies (Jayson-Quashigah et al., 2019; Casella et al., 2020, 2016). Similar applications would be useful to determine volumetric changes through time. However, these studies would be limited to quantifications of topographic changes as sub-surface imaging is still required to identify the lobe base.

High resolution topographic surveys of blowouts including erosional hollows and adjacent depositional lobes were collected using SfM optical drone mapping and DGPS to create digital elevation models (DEMs) of the blowouts. In addition, ground-penetrating radar (GPR) profiles were collected along and across the depositional lobes at three of the smaller and less complex blowouts. Volumes of the erosional hollows and their adjacent depositional lobes are calculated from the DEMs, GPR data, and soil pits. The results show that although the areas of the hollows and lobes are similar, the volume of the lobes is much less than the volume of the adjacent hollows with the lobe volume amounting to around 50% of the hollow. Part of the difference in volumes can be attributed to the method used to calculate the lobe volumes. The wavelength of the GPR prevents measurement of sand layers that are thinner than the section covered by the first arrivals of the direct groundwaves and airwaves. Despite this, the results indicate that blowout lobes are not very efficient sand traps, capturing only 50% of the sand eroded from the hollow. The remaining 50% is probably distributed further downwind, as a thin sand layer captured by local vegetation, with an undetermined fraction transported further downwind as saltating sand or a “dust” plume during storm events.