Outburst floods from glacial Lake Missoula largely explain erosion of the Channeled Scabland, a system of overfit, basaltic channels in Washington, northwestern United States. However, it is challenging to explain Missoula flood routing into Moses Coulee due to its topographic isolation from flood routes. To clarify flood pathways into Moses Coulee, we mapped channels that delineate a radial-anastomosing network connecting to Moses Coulee. Channels consist of coulee-like features eroded mainly in basalt. Channels climb adverse slopes and cross divides, and the network is draped with eskers and recessional moraines. These channel geometries and glacial landform associations suggest a tunnel channel network. Large channel dimensions and network anastomosis indicate formation in subglacial floods. The network connects to Moses Coulee with sufficient cross-sectional area to convey megafloods, hinting that subglacial floods may have been a significant source of Moses Coulee meltwater, in addition to possible diverted Missoula floods.The Channeled Scabland, northwestern United States, is a paradigmatic example of outburst flooding that provides insight into megaflood landscapes on Earth and Mars (Gallagher and Bahia, 2021). The scabland landscape, developed in Columbia River Basalts (CRBs), consists of anastomosing bedrock channels containing rock basins, cataracts, and huge gravel bars (Baker, 2009). Current interpretations of the Channeled Scabland attribute formation to megafloods (Q > 106 m3 s–1) from glacial Lake Missoula (GLM; O'Connor et al., 2020).Water routing into Moses Coulee, a prominent flood tract of the Channeled Scabland, is enigmatic due to a lack of connectivity to other scabland routes (Bretz, 1923). Moses Coulee heads gradually from the terminal position of the Okanogan lobe of the Cordilleran ice sheet (CIS) on the Waterville Plateau (Fig. 1), suggestive of a connection to Okanogan lobe hydrology (Freeman, 1933). However, the Okanogan lobe may have been an insufficient meltwater source to account for inferred megafloods in Moses Coulee, which thus may have required contributions from a GLM source (Hanson, 1970), routed by overspill or ice-marginal diversion (Waitt, 2021). Geologic evidence needed to distinguish these hypotheses is elusive. Here, we investigated water routing into Moses Coulee by mapping channel networks and glacial landforms.The Okanogan lobe glaciated the Waterville Plateau during marine isotope stage (MIS) 2, generating glacial landforms (Fig. 2). There is circumstantial evidence for pre–MIS 2 glaciation in the area (Flint, 1935), but no pre–MIS 2 glacial deposits on the Waterville Plateau have been found. The Withrow moraine marks the MIS 2 limit of the Okanogan lobe, north of which are eskers, recessional moraines, and other glacial deposits (Kovanen and Slaymaker, 2004; Hanson, 1970). Varve records and exposure dating bracket the timing of ice cover on the Waterville Plateau to a maximum of 2000–3000 yr (Atwater, 1986) between 17 and 14 ka (Gombiner, 2022). During this time, Okanogan ice impounded glacial Lake Columbia (GLC) and contributed to rerouting of Missoula floods.Okanogan lobe landforms hint at surge-type behavior, enabled by an integrated hydrologic system (Kovanen and Slaymaker, 2004). Despite inferences that subglacial floods from the CIS contributed to Channeled Scabland erosion (Shaw et al., 1999), evidence for subglacial floods beneath the Okanogan lobe has not been closely examined.We mapped channels, eskers, and moraines on the former bed of the Okanogan lobe (Figs. 1 and 2) by tracing features on high-resolution digital topography and satellite imagery (U.S. Geological Survey [USGS] 3D Elevation Program 10-m-resolution digital elevation model and Google Earth terrain-enhanced imagery). This work built on prior landform mapping in the area by identifying previously unrecognized features that we compiled into georeferenced landform inventories. We field-checked portions of the channel network to verify channel boundaries and to inspect bedrock surfaces and sediment deposits within and near channels.Mapping revealed a radial-anastomosing channel network eroded in basalt and sediment, with a minor part in crystalline rock. The network is elongate on the Omak plateau, becoming radial southward toward the Withrow moraine (Fig. 2). Broad trunk channels obliquely join connector channels of smaller dimensions, forming anastomosis. Trunk channels are typically 200–500 m wide and 20–30 m deep, not inclusive of sediment fills, which are on the order of meters to tens of meters in limited well-log reports (WDE, 2023). Boulder lags ornament channel fills (Fig. 3). Channels interweave around stripped bedrock residuals that are increasingly sediment-covered away from channel margins (Fig. 3). Basalt channel morphology is box-like, with stepped channel margins, reflecting the jointed and planar structure of the CRB flows. Crystalline bedrock channels are smaller in all dimensions (Fig. 2). Channels have variable widths and undulatory floors with closed rock basins (tens of meters in depth; Figs. 2 and 3). On the Waterville Plateau, channels climb adverse slopes (Fig. S4 in the Supplemental Material1) and cross a series of topographic divides (between 650 and 800 + m above sea level [asl]; Fig. 2). The network lacks clear breaks, though this apparent absence may reflect data resolution. At the downstream outlets of the anastomosing network on the Waterville Plateau, smaller channels in bedrock and sediment parallel the Withrow moraine. Some include scabland terrain in basalt (Hanson, 1970), while others contain outwash. These proglacial channels follow the surface slope—unlike the anastomosed network—and converge at the head of Moses Coulee (Fig. 2).Eskers and recessional moraines overprint portions of the channel network. Eskers (n = 485 mapped features, defined as sinuous ridges of sand and gravel) occur throughout the study area. They typically exhibit single-ridge morphology, with dimensions of hundreds of meters in length, single digits to tens of meters in width, and <20 m in height. Greater numbers of eskers occur in the central-eastern portion of the Waterville Plateau, where some terminate against moraines (Fig. 2). Many eskers occur on channel floors, aligned subparallel to channels. Others are oblique to channels and cross residuals and unchannelized interfluves.Recessional moraines (n = 404 mapped features, defined as arcuate to sinuous diamict ridges) occur throughout the study area as discontinuous segments (hundreds of meters in length) that become more continuous (kilometers in length) northward. They commonly block channels and drape across topography.The radial pattern and adverse bed slopes of the tunnel channels are best explained by subglacial flows following hydrostatic pressure gradients under the Okanogan lobe (Lelandais et al., 2016; Supplemental Material D). Additionally, recessional moraines and eskers overprinting channels are a common landform association indicative of tunnel channels (Sharpe et al., 2021).The coulee morphology and anastomosis of the tunnel channel network are like other areas of the Channeled Scabland, where anastomosis records overwhelming of drainage capacity and progressive channelization (Baker, 2009). The channels of the Omak and Waterville plateaus could be reasonably explained by an analogous process where high-magnitude flows exceeded bankfull capacity in a subglacial environment. Subglacial flows also erode upward into overlying ice, creating additional drainage capacity in ice tunnels, but the existence of the anastomosed tunnel channel network implies that subglacial drainage was not wholly accommodated by ice tunnels. We focused on the tunnel channels eroded downward into the land surface to reconstruct subglacial hydrology, aware of the limitation that additional meltwater may have drained through upward-eroding ice tunnels.Repeated, low-magnitude flows (5 × 102 to 5 × 104 m3/s; Beaud et al., 2018) may have contributed to erosion of tunnel channels on the Waterville and Omak plateaus, but such flows cannot explain the overall anastomosed network pattern, which requires flows exceeding bankfull capacity. Low-magnitude discharges are typically associated with esker deposition (Lally et al., 2023) and dendritic channel networks that terminate against moraines (Kirkham et al., 2024), while larger floods (104 to >106 m3/s; Kirkham et al., 2019) are associated with anastomosing channels in bedrock (e.g., Lewis et al., 2006), like those on the Waterville and Omak plateaus. Further, low-magnitude flows erode slowly, on time scales of ~7500 yr, to generate channels of comparable dimensions to those mapped here (Beaud et al., 2018); this time scale is longer than the <3000 yr during which the Okanogan lobe covered the Waterville Plateau during MIS 2. Additionally, depositional records of low-magnitude flows (eskers) frequently occur on unchannelized surfaces on the Waterville and Omak plateaus (Fig. 2), implying that esker-forming flows did not erode bedrock channels.These morphologies and landform associations indicate that subglacial floods eroded the tunnel channel network. The subglacial floods drained at the ice margin into moraine-parallel channels that connected to Moses Coulee (Fig. 2). Scabland in these proglacial channels (Hanson, 1970) suggests conveyance of large outburst floods into Moses Coulee, which plausibly originated as subglacial floods.The tunnel channel network is likely a palimpsest eroded by flows of varying magnitude, potentially over multiple glaciations. Sedimentary evidence for multiple glaciations in the Okanogan Valley (Lesemann et al., 2013) suggests pre–MIS 2 Okanogan lobes, but pre–MIS 2 deposits of the Okanogan lobe have not been found, so it is unclear when tunnel channel erosion initiated. The absence of breaks in the network suggests simultaneous operation of large sectors, like the Mansfield Channels upstream of Moses Coulee (Hanson, 1970). However, given the hydraulic variability of an active ice lobe, localized operation and/or erosion of smaller subsectors likely occurred.Connectivity between tunnel channels on the Waterville Plateau and Moses Coulee suggests that subglacial floods drained into Moses Coulee, contributing to coulee erosion and flood deposition. Here, we tested this hypothesis against two prevailing models that route Missoula floods into Moses Coulee in front of the Okanogan lobe: one via back-flooding of Foster Valley and spillover across a 653 m asl divide (Fig. 2, spillover point 1; Waitt, 2021), and the other via NE to SW ice-marginal diversion across the Waterville Plateau (Fig. 2, spillover point 2; O'Connor et al., 2020). An implicit prediction of these models is channel development along flow paths. Analogous ice-marginal diversion has created a distinct channel pattern near Northrup Canyon (ice-marginal spillover channels on Fig. 2). However, no equivalent channel pattern occurs along the proposed diversion path into Moses Coulee, an absence possibly explained by glacial overprinting (Bretz et al., 1956). While faint channelization exists along the proposed diversion paths, the diversion models cannot explain the complete anastomosed tunnel channel network.Both models that route Missoula floods into Moses Coulee require specific landscape configurations that are incompatible with some geologic evidence and are mutually contradictory (Supplemental Material A). One model requires that upper Grand Coulee was not yet eroded, that the head of Foster Coulee was not yet eroded to its modern elevation, and that the Okanogan lobe selectively blocked the Columbia valley west of the mouth of Foster Creek (Fig. S1; Waitt, 2021). The other model requires that upper Grand Coulee was not yet eroded, that the head of Foster Coulee was eroded to its modern elevation, and that the Okanogan lobe had advanced onto the Waterville Plateau, diverting flow along its margin (Fig. S2; O'Connor et al., 2020). Whether these landscape configurations existed at the time of Moses Coulee floods remains unclear. Some sedimentary records indicate that Grand Coulee breached prior to the last glaciation (Atwater, 1986), contradicting required landscape configurations within both models. Subglacial floods through the tunnel channel network into Moses Coulee are consistent with reconstructed landscape configurations.The mouth of Moses Coulee preserves sedimentary evidence for four MIS 2 floods (Fig. 1; Waitt, 2021), dated between 17.4 ± 0.8 and 15.5 ± 0.8 ka (Gombiner, 2022). While these floods have been attributed to a GLM source, connectivity between the tunnel channels and Moses Coulee instead implies a subglacial source.We estimated subglacial discharges into Moses Coulee between 1.3 × 105 m3/s and 5.2 × 106 m3/s. Calculations combined cross-sectional areas of subglacial channels entering the Moses Coulee basin (104,000 m2) with a range of modeled velocities for pressurized subglacial flows (Clarke, 2003), percentage operation of the network, and thicknesses of sediment fill currently in channels (Supplemental Material B; Table S3). These estimates are minima that do not account for discharge through ice tunnels melted into overlying ice. Some estimates exceeded megaflood scale, consistent with the geomorphic inference that tunnel channels conveyed large floods and with reconstructed megaflood discharges in Moses Coulee (O'Connor et al., 2020).Megaflood discharges through the tunnel channels would require storage and release of water under the Okanogan lobe, consistent with geomorphic evidence upstream. Bedrock tunnel channels along the 300 km length of the Okanogan Valley, including Omak Trench and Soap Lake, indicate subglacial meltwater drainage, and they hydraulically connect the Okanogan lobe to a regional tunnel valley network (Lesemann and Brennand, 2009). Coarse MIS 2 glaciofluvial sediment in the Okanogan Valley has been interpreted as originating in subglacial flows based on thickness, coarseness, and position in overdeepened troughs (Vanderburgh and Roberts, 1996; Eyles et al., 1991), suggesting subglacial floods in the Okanogan Valley. However, identifying records of source reservoirs is challenging due to a lack of diagnostic subglacial criteria for glacial lake deposits (Livingstone et al., 2012) and the low preservation potential of such subglacial records. Considering these limitations, we explored mechanisms for water storage and release in the Okanogan Valley.In the Okanogan Valley, ice advance over lakes could have formed “catch lakes” (Rudoy, 1998) during ice expansion. Once covered by ice, high local geothermal heat flux could have enhanced meltwater production (Lesemann and Brennand, 2009), and the overdeepened valley would have favored development of a hydraulic potential low and formation of a subglacial reservoir (Livingstone et al., 2013). Initiation of subglacial reservoir drainage would have required a trigger, such as a change in thermal conditions (Skidmore and Sharp, 1999) or an abrupt input of water. Such water inputs into the Okanogan Valley could have resulted from drainage of supraglacial lakes, of ice-marginal lakes, or of subglacial lakes farther up the ice sheet. Hydraulic pressure from these lakes may have initiated drainage cascades into the Okanogan Valley, which operated as part of a water storage and transfer network, like bedrock tunnel channels in formerly glaciated areas of West Antarctica (Kirkham et al., 2019).Moses Coulee has been enigmatic due to its lack of clear connectivity to the main Channeled Scabland flood routes. We resolved this enigma by identifying an anastomosing network of subglacial channels on Waterville and Omak plateaus, Washington State, northwestern United States, which records subglacial meltwater floods beneath the Okanogan lobe and into Moses Coulee. These subglacial channels are analogous to tunnel channels in other glaciated areas. Water sources to supply these channels could have included drainage of supraglacial lakes and linked ice-marginal and subglacial reservoirs. During MIS 2 glaciation, meltwater was routed into Moses Coulee via subglacial anastomosed channels. Prevailing Missoula flood spillover hypotheses remain theoretically viable but depend upon unique configurations of Grand Coulee and the Okanogan lobe, and they are difficult to reconcile with the observed channel pattern. These conclusions emphasize the importance of Cordilleran ice sheet (sub-)glacial hydrology to the formation of the Channeled Scabland (Shaw et al., 1999).We thank Brian Atwater, Nick Zentner, Skye Cooley, and John Stone for helpful discussions and encouragement, and we thank Vic Baker and two anonymous reviewers for constructive feedback. We dedicate this article to the memory of John Shaw. J. Gombiner thanks the Geological Society of America Graduate Student Research fund for analytical support.

中文翻译：

---

奥卡诺根波瓣隧道渠道和冰下洪水流入美国西北部的渠道斯卡布兰摩西古力 (Moses Coulee)


米苏拉冰川湖爆发的洪水在很大程度上解释了水道Scabland的侵蚀，这是美国西北部华盛顿的一个过度拟合的玄武岩水道系统。然而，由于米苏拉的地形与洪水路线隔离，解释米苏拉洪水路线进入摩西古力是一项挑战。为了阐明进入摩西古力的洪水路径，我们绘制了描绘连接摩西古力的径向吻合网络的河道。河道由主要在玄武岩中侵蚀的古力状特征组成。河道攀爬不利的斜坡并跨越分水岭，网络布满了埃斯克和后退的冰碛。这些河道几何形状和冰川地貌关联表明存在隧道河道网络。大的河道尺寸和网络吻合表明冰下洪水的形成。该网络与摩西古力（Moses Coulee）相连，具有足够的横截面积来输送特大洪水，这表明除了可能转移的米苏拉洪水外，冰下洪水可能是摩西古力融水的重要来源。美国西北部的渠道斯卡布兰是一个典型的例子爆发洪水的例子，让我们深入了解地球和火星上的特大洪水景观（Gallagher 和 Bahia，2021）。哥伦比亚河玄武岩（CRB）中发育的贫瘠地貌由包含岩石盆地、瀑布和巨大砾石坝的网状基岩河道组成（Baker，2009）。目前对 Channeled Scabland 的解释将其形成归因于米苏拉冰川湖的特大洪水 (Q > 106 m3 s–1)（GLM；O'Connor 等人，2020）。水流入 Moses Coulee，这是 Channeled Scabland 的一个重要洪泛区，由于缺乏与其他 scabland 路线的连通性而显得神秘（Bretz，1923）。 摩西·古力 (Moses Coulee) 逐渐从沃特维尔高原上科迪勒拉冰盖 (CIS) 奥卡诺根冰瓣的终端位置出发（图 1），这表明与奥卡诺根冰瓣水文学有关（Freeman，1933）。然而，奥卡诺根波瓣可能不足以解释摩西古力推断的特大洪水，因此可能需要 GLM 源的贡献（Hanson，1970），通过溢出或冰边转移（Waitt，2021） 。区分这些假设所需的地质证据是难以捉摸的。在这里，我们通过绘制河道网络和冰川地貌来研究流入摩西古力的水流。奥卡诺根叶在海洋同位素阶段 (MIS) 2 期间对沃特维尔高原进行了冰川作用，形成了冰川地貌（图 2）。有间接证据表明该地区存在 MIS 2 之前的冰川作用（Flint，1935），但在沃特维尔高原上尚未发现 MIS 2 之前的冰川沉积物。 Withrow 冰碛标志着奥卡诺根波瓣的 MIS 2 界限，其北部是埃斯克、后退冰碛和其他冰川沉积物（Kovanen 和 Slaymaker，2004；Hanson，1970）。 Varve 记录和暴露年代测定将沃特维尔高原上的冰覆盖时间划分为 17 至 14 ka（Gombiner，2022）之间的最长 2000-3000 年（Atwater，1986）。在此期间，奥卡诺根冰蓄积了哥伦比亚冰川湖 (GLC)，并导致米苏拉洪水改道。奥卡诺根波瓣地貌暗示了由综合水文系统促成的涌浪型行为（Kovanen 和 Slaymaker，2004）。尽管推断来自独联体的冰下洪水导致了通道式Scabland侵蚀（Shaw等人，1999），但奥卡诺根叶下冰下洪水的证据尚未得到仔细研究。我们通过追踪高分辨率数字地形和卫星图像（美国地质调查局 [USGS] 3D 高程计划 10 米分辨率）的特征，绘制了奥卡诺根波瓣前河床上的河道、冰碛和冰碛（图 1 和图 2）数字高程模型和 Google Earth 地形增强图像）。这项工作建立在该地区先前的地形测绘基础上，通过识别我们编译到地理参考地形清单中的以前未识别的特征。我们对河道网络的部分区域进行了现场检查，以验证河道边界，并检查河道内部和附近的基岩表面和沉积物沉积情况。测绘显示，径向吻合的河道网络受到玄武岩和沉积物的侵蚀，其中一小部分受到结晶岩的侵蚀。该网络在奥马克高原上呈拉长状，呈放射状向南延伸至威罗冰碛（图 2）。宽大的主干通道倾斜地连接较小尺寸的连接器通道，形成吻合。干河河道通常宽 200-500 m，深 20-30 m，不包括沉积物填充物，在有限的测井报告中，沉积物填充物的深度为数米到数十米（WDE，2023）。巨石滞后装饰通道填充（图 3）。河道在剥离的基岩残余物周围交织在一起，这些残余物越来越多地被沉积物覆盖，远离河道边缘（图3）。玄武岩河道形态呈盒状，河道边缘呈阶梯状，反映了CRB流的节理和平面结构。结晶基岩河道的所有尺寸都较小（图 2）。河道宽度不一，底板起伏不定，有封闭的岩石盆地（深数十米；图2和图3）。在沃特维尔高原，河道沿着逆坡爬行（图 1）。 补充材料 1 中的 S4) 并跨越一系列地形分界线（海拔 650 至 800 + 米之间 [asl]；图 2）。网络缺乏明显的中断，尽管这种明显的缺失可能反映了数据分辨率。在沃特维尔高原网状网络的下游出口处，基岩和沉积物中的较小通道与威罗冰碛平行。有些包括玄武岩中的赤裸地带地形（Hanson，1970），而另一些则包含冲刷。与网状网络不同，这些前冰川河道沿着地表斜坡分布，并在摩西古力的顶端汇聚（图 2）。埃斯克和后退冰碛覆盖了河道网络的部分区域。 Eskers（n = 485 个映射特征，定义为沙子和砾石的蜿蜒山脊）遍布整个研究区域。它们通常表现出单脊形态，长度为数百米，宽度为个位数到数十米，高度<20 m。沃特维尔高原的中东部地区出现了大量的埃斯克，其中一些的终点是冰碛（图 2）。许多eskers出现在通道地板上，与通道近平行排列。其他的则倾斜于河道、交叉残差和未河道化的河间道。后退冰碛（n = 404 个映射特征，定义为弓形到蜿蜒的硬质山脊）遍布整个研究区域，作为不连续的段（长度数百米）变得更加连续（公里）长度）向北。它们通常会堵塞河道并覆盖地形。隧道河道的径向模式和不利的河床坡度可以通过奥卡诺根叶下静水压力梯度下的冰下流动得到最好的解释（Lelandais 等人，2016 年；补充材料 D）。 此外，后退冰碛和埃斯克叠印河道是指示隧道河道的常见地貌组合（Sharpe et al., 2021）。隧道河道网络的古力形态和吻合与通道Scabland的其他区域一样，其中吻合记录压倒性的排水能力和渐进式渠道化（Baker，2009）。奥马克和沃特维尔高原的河道可以用类似的过程来合理解释，在冰下环境中，高强度的水流超过了河岸的满容量。冰下水流也会向上侵蚀到上覆的冰中，从而在冰隧道中产生额外的排水能力，但吻合隧道渠道网络的存在意味着冰下排水并不能完全由冰隧道容纳。我们关注向下侵蚀到陆地表面的隧道通道，以重建冰下水文，意识到额外的融水可能通过向上侵蚀的冰隧道排出的局限性。重复的低强度流动（5 × 102 至 5 × 104 m3/ s; Beaud et al., 2018）可能导致了沃特维尔和奥马克高原上隧道渠道的侵蚀，但这种流量无法解释整体的网状网络模式，这需要流量超过满岸容量。低强度的泄洪通常与埃斯克沉积（Lally 等，2023）和终止于冰碛的树枝状水道网络（Kirkham 等，2024）有关，而较大的洪水（104 至 >106 m3/s；Kirkham 等） ., 2019）与基岩中的网状通道有关（例如，Lewis 等人，2006），例如沃特维尔和奥马克高原上的网状通道。 此外，低强度的水流在约 7500 年的时间尺度上缓慢侵蚀，产生与此处绘制的尺寸相当的通道（Beaud 等人，2018）；这个时间尺度比MIS 2期间奥卡诺根波瓣覆盖沃特维尔高原的<3000年更长。此外，低强度流（eskers）的沉积记录经常出现在沃特维尔和奥马克高原的非渠道化表面上（图2） ），这意味着形成埃斯克的水流没有侵蚀基岩河道。这些形态和地貌组合表明冰下洪水侵蚀了隧道河道网络。冰下洪水在冰缘处排入与摩西古力相连的冰碛平行河道（图2）。这些前冰川河道中的 Scabland（Hanson，1970）表明，大规模的爆发洪水被输送到摩西古力（Moses Coulee），这似乎起源于冰下洪水。隧道河道网络很可能是被不同程度的水流侵蚀的重写本，可能经过多次冰川作用。奥卡诺根山谷多次冰川作用的沉积证据（Lesemann 等，2013）表明，MIS 2 之前的奥卡诺根裂片，但尚未发现 MIS 2 之前的奥卡诺根裂片沉积物，因此尚不清楚隧道河道侵蚀何时开始。网络中没有中断表明大型扇区同时运行，例如摩西古力上游的曼斯菲尔德海峡（Hanson，1970）。然而，考虑到活跃冰瓣的水力变化，可能会发生局部运行和/或较小分区的侵蚀。沃特维尔高原和摩西古力上的隧道通道之间的连通性表明，冰下洪水流入摩西古力，导致古力侵蚀和洪水沉积。 在这里，我们针对两种主流模型测试了这一假设，这两种模型将米苏拉洪水引入奥卡诺根波瓣前的摩西古利：一种是通过福斯特谷的反向洪水和跨越 653 米分水岭的溢出（图 2，溢出点 1；Waitt ，2021），另一个是从东北向西南跨越沃特维尔高原的冰缘改道（图2，溢出点2；O'Connor等人，2020）。这些模型的隐含预测是沿流动路径的渠道发展。类似的冰缘分流在诺斯拉普峡谷附近形成了独特的河道模式（图 2 中的冰缘溢出河道）。然而，沿着拟议的进入摩西古力的改道路径并没有出现等效的渠道模式，这种缺失可能是由冰川叠印解释的（Bretz等人，1956）。虽然沿着拟议的导流路径存在微弱的渠道化，但导流模型无法解释完整的网状隧道渠道网络。将米苏拉洪水引入摩西古利的两种模型都需要特定的景观配置，这些配置与一些地质证据不相容并且相互矛盾（补充材料A） ）。一个模型要求大古力上游尚未被侵蚀，福斯特古力河头尚未侵蚀到现代海拔，并且奥卡诺根波瓣选择性地封锁了福斯特溪河口以西的哥伦比亚河谷（图S1；Waitt） ，2021）。另一个模型要求大古力上游尚未被侵蚀，福斯特古力河头已被侵蚀至现代海拔，并且奥卡诺根波瓣已前进至沃特维尔高原，沿其边缘转移水流（图 S2；O'）康纳等人，2020）。这些景观构造在摩西古力洪水发生时是否存在仍不清楚。 一些沉积记录表明，大古力在末次冰川作用之前破裂（Atwater，1986），这与两个模型中所需的景观配置相矛盾。通过隧道水道网络进入 Moses Coulee 的冰下洪水与重建的景观配置一致。 Moses Coulee 河口保存了四次 MIS 2 洪水的沉积证据（图 1；Waitt，2021 年），日期在 17.4 ± 0.8 和 15.5 ± 0.8 ka 之间（戈宾纳，2022）。虽然这些洪水被归因于 GLM 源头，但隧道通道和 Moses Coulee 之间的连通性却意味着冰下源头。我们估计流入 Moses Coulee 的冰下流量在 1.3 × 105 m3/s 和 5.2 × 106 m3/s 之间。计算将进入摩西古力盆地（104,000 m2）的冰下河道横截面积与加压冰下流的一系列模拟速度（Clarke，2003）、网络运行百分比以及当前河道中沉积物填充的厚度相结合（补充）材料B；表S3)。这些估计是最小值，没有考虑通过融化成上覆冰的冰隧道的排放。一些估计超出了特大洪水规模，这与隧道渠道输送大量洪水的地貌推断以及摩西古力重建的特大洪水排放一致（O'Connor 等人，2020）。通过隧道渠道的特大洪水排放需要在以下条件下储存和释放水：奥卡诺根叶，与上游地貌证据一致。沿奥卡诺根山谷 300 公里长的基岩隧道通道，包括奥马克海沟和肥皂湖，表明冰下融水排水，它们通过水力将奥卡诺根叶与区域隧道山谷网络连接起来（Lesemann 和 Brennand，2009）。 根据厚度、粗糙度和过深槽的位置，奥卡诺根山谷的粗 MIS 2 冰川河流沉积物被解释为起源于冰下水流（Vanderburgh 和 Roberts，1996 年；Eyles 等人，1991 年），这表明奥卡诺根山谷发生了冰下洪水。然而，由于缺乏冰湖沉积物的冰下诊断标准（Livingstone 等，2012）以及此类冰下记录的保存潜力较低，识别源水库记录具有挑战性。考虑到这些限制，我们探索了奥卡诺根山谷的水储存和释放机制。在奥卡诺根山谷，冰在冰膨胀过程中前进到湖泊上可能形成“集水湖”（Rudoy，1998）。一旦被冰覆盖，高的局部地热通量可能会增加融水的产量（Lesemann和Brennand，2009），而过度加深的山谷将有利于水力势低点的发展和冰下水库的形成（Livingstone等，2013） 。冰下水库排水的启动需要触发因素，例如热条件的变化（Skidmore 和 Sharp，1999）或突然注水。进入奥卡诺根山谷的此类水可能来自冰上湖泊、冰缘湖泊或冰盖更远的冰下湖泊的排水。这些湖泊的水压可能引发了进入奥卡诺根山谷的排水梯级，该山谷作为储水和输送网络的一部分运行，就像南极洲西部以前冰川地区的基岩隧道通道一样（Kirkham 等人，2019 年）。Moses Coulee 发现由于缺乏与主要渠道 Scabland 洪水路线的明确连接，该区域一直是个谜。 我们通过确定美国西北部华盛顿州沃特维尔和奥马克高原上的冰下河道网状网络解决了这个谜团，该网络记录了奥卡诺根波瓣下方并流入摩西古利的冰下融水洪水。这些冰下通道类似于其他冰川地区的隧道通道。供应这些渠道的水源可能包括冰上湖泊的排水以及相连的冰缘和冰下水库。 MIS 2 冰川作用期间，融水通过冰下吻合通道流入摩西古力。普遍存在的米苏拉洪水溢出假设在理论上仍然可行，但取决于大古力和奥卡诺根波瓣的独特配置，并且它们很难与观察到的河道模式相一致。这些结论强调了科迪勒拉冰盖（冰下）水文对于通道状 Scabland 形成的重要性（Shaw 等人，1999）。我们感谢 Brian Atwater、Nick Zentner、Skye Cooley 和 John Stone 的有益讨论和鼓励，我们感谢 Vic Baker 和两位匿名审稿人的建设性反馈。我们谨以此文纪念约翰·肖。 J. Gombiner 感谢美国地质学会研究生研究基金提供的分析支持。