Michael Durand (School of Earth Sciences, Ohio State University, Columbus, OH, USA Byrd Polar and Climate Research Center, Columbus, OH, USA, United States)

We have made substantial progress towards characterizing SWOT discharge accuracy. In a recent working group paper, we characterized three increasingly restrictive data quality filtering groups and describe how these groupings could be used to infer discharge accuracy in the predominantly un-gauged reaches of the world. We show a preference here for accuracy statistics that separate bias and correlation, as the former is driven by systematic uncertainty, particularly in the prior, while the latter should perform well when FLPEs are working, regardless of bias. What we find is that bias is somewhat higher than originally expected, but that the Consensus algorithm has increasing skill in correlation across these groupings. Most notably, we find that 11,274 ungauged reaches globally can be expected to perform in accordance with the “mid-filter” group, based on the restrictions used to select that group. While we believe that SWOT data quality will always be a driving factor in discharge accuracy, we have moved our main area of focus on to characteristics inherent to river reaches that are associated with high and low discharge accuracy. In our investigation of typological drivers of SWOT discharge accuracy, we are exploring the characteristics as described by SWORD, near-river landcover, topography, and the presence of manmade features, such as dams. We have already seen that some of these features play a large roll in discharge accuracy. We find that urban landcover and high elevation variation (nearby extreme topography) both have a dramatic impact on discharge correlation but not bias we see in the result. An interesting finding from this work is that the impact is most notable with the mid-filter group. Using Jenks breaks to group classes of urban landcover the median Pearson correlation is 0.76 where there is less than 1% urban landcover, and that median monotonically decreases to 0.56 in (20%) in the urban cover >34% group. Similarly median Pearson correlation monotonically decreases in the mid-tier data from 0.74 (elevation std<9.88m) to 0.49 (25% reduction at elevation std>36.01m). These patterns are not apparent with the least-filtered group and are less dramatic with the highly-filtered group. It’s likely that this demonstrates the intersection of filter quality and the inherent qualities of the reach. It’s likely that the impact of urban land cover is obscured by low data quality in least-filtered data, but as filtering is made more restrictive the impact is more substantial. Interestingly, with the highly-filtered requirement, the urban impact is less dramatic, which may indicate that enforcing hydraulic consistency (one of the requirements) counters much of the error driven by urban land cover, while the elevation std medians still span 30% in the CH data. Overall, we find that given the correct SWOT data criteria and some inherent reach properties, we can infer what accuracy we can expect from most of the worlds ungauged river reaches.