**Qian Zhang**

Monitoring Data Analyst

University of Maryland Center for Environmental Science Integration and Application Network

*Email:* qzhang@umces.edu

*Biography:*

Qian Zhang has joined UMCES as a monitoring data analyst at the USEPA Chesapeake Bay Program (CBP) Office. Within the CBP's Scientific, Technical Assessment and Reporting (STAR) team, his main role is to work with colleagues to develop statistical methods of data analysis that uses the extensive data available through the CBP partnership to better understand factors that drive water-quality patterns. Qian obtained his Ph.D. in Environmental Engineering from the Johns Hopkins University (JHU), in addition to two master degrees from JHU (one in environmental engineering and the other in statistics). Under the advisement of Professor Bill Ball (JHU/CRC), Qian's doctoral research focuses on quantifying nutrient and sediment export from the major tributaries of Chesapeake Bay through statistical analysis and synthesis of river water-quality monitoring data sets. In general, he is interested in computationally-intensive modeling and synthesis of large-scale data sets for watersheds and estuaries, including but not limited to: (1) statistical analysis of water-quality monitoring data of various systems, (2) improvement of statistical methods for quantification of riverine fluxes, trends, and uncertainties, and (3) exploration of linkages between riverine fluxes and trends to watershed changes (e.g., land use, input sources) and to estuarine processes (e.g., oxygen, water clarity). He is both thrilled and thankful to to stay in the Chesapeake community and to continue contributing his skills toward Chesapeake Bay management and protection.

More info...

Reduction of suspended sediment (SS), total phosphorus (TP), and total nitrogen is an important focus for Chesapeake Bay watershed management. The Susquehanna River, the bay’s largest tributary, has drawn attention because SS loads from behind Conowingo Dam (near the river’s mouth) have been rising dramatically. To better understand these changes, we evaluated histories of concentration and loading (1986-2013) using data from sites above and below Conowingo Reservoir. First, observed concentration-discharge relationships show that SS and TP concentrations at the reservoir inlet have declined under most discharges in recent decades, but without corresponding declines at the outlet, implying recently diminished reservoir trapping. Second, best estimates of mass balance suggest decreasing net deposition of SS and TP in recent decades over a wide range of discharges, with cumulative mass generally dominated by the 75th~99.5th percentile of daily Conowingo discharges. Finally, stationary models that better accommodate effects of riverflow variability also support the conclusion of diminished trapping of SS and TP under a range of discharges that includes those well below the literature-reported scour threshold. Overall, these findings suggest that decreased net deposition of SS and TP has occurred at sub-scour levels of discharge, which has significant implications for the Chesapeake Bay ecosystem.

My name is Qian. I work for UMCES right now. I've been here for about one month, which I enjoyed very much. Today I think this topic today is not strange to the many of you. Over the last couple of years there has been a huge amount of information delivered by all different agencies and institutions. Including but not limited to the USGS, the Lower Susquehanna River Watershed Assessment, STAC review on that study, the most recent STAC workshop which was briefed today very briefly, and also some publications by us at Hopkins. So all of these publications at Hopkins deal with this issue partially or wholly. And today my goal is to give you an overview of the findings from those publications that are particularly relevant to this topic. I want to start with this 2013 publication by Zhang, Brady, and Ball. As a background, Susquehanna River contributes more than half of fresh water and nutrient loading and sediment loading from the Bay watershed. Near the lower part of this river is the so-called the Lower Susquehanna River Reservoir System. This system consists of three members: the Lake Clarke, Lake Aldred, and the Conowingo Reservoir. The Conowingo Reservoir is the largest member in the system, which has not been filled up effectively. But the two upstream reservoirs have been filled up for many decades. So the goal of that particular study was to quantify the trends for nutrient and sediment at sites above and below the reservoir system, with particular focus on the comparison of particulate vs. dissolved species. In terms of monitoring sites and data sets, near the reservoir system there are three sites. Above the reservoir system there's Marietta and Conestoga which have been monitored by the Susquehanna River Basin Commission. Below the reservoir system there is Conowingo at the Conowingo Dam; it is monitored by USGS. You can see the similarity between the two, between the input and output, in terms of the drainage area. So if there's nothing going on in the reservoir system, then we should expect similar or identical trends between the input and output. But I will show you very soon that is not the case. In terms of monitoring data sets, each site has been monitored in terms of data discharge data by the USGS. And then the concentration data have been made available roughly around 26 to 37 sample days per year for different species of nutrient and sediment. So there has been a tool developed recently by USGS called WRTDS, which was used to obtain daily estimates for concentration and loading. To give you a very brief background on WRTDS. Basically, it uses the daily flow and sparse concentration records, which are typically available at many monitoring sites, as the model input. And it gives us the daily concentration and loading estimates. This model predicts log concentration as a function of time, discharge, and season. And hence it is called WRTDS. This model builds one single model for each day of estimation. Thereby it has a unique set of coefficients for each day of estimation. In doing so it allows the beta coefficients to change flexibly over time, season, and discharge. And thereby it does not have to make assumptions on fixed concentration-discharge relationships. This model has been shown to offer better performance in terms of estimation in a wide range of applications than some of the previous methods. This is an example showing the estimates from the model. The true-condition loading and the flow-normalized loading. I want to draw your attention here that the true-condition loading is very variable because of interannual variability in streamflow. But we deem them as the most accurate estimates. If we want to study the downstream impact in terms of ecology, then we need to use the true-condition loadings. Otherwise, if we want to assess the management progress, then we need to remove that interannual variability, and hence we need to use the flow-normalized loading. In this presentation I will show you the results from each of them -- from both of them. Now let's take a look at the result from that 2013 study. So what have been the flow-normalized seasonal trends at Conowingo Dam? Remember this is the output from the reservoir system. Here shows the results for sediment and TP. Each line here represents one season; they have been scaled by the long-term median values. You can see in general there has been a rising trend for sediment in most seasons since around 1995. Similarly there has been rising trend in total phosphorus as well in most seasons since around mid-1990s. And this similarity has led us to think that this might be an issue with the particular phosphorus which transports with sediment. And to prove that we have separately analyzed the trend for particulate phosphorus and dissolved phosphorus. We do find that at the Conowingo Dam particulate phosphorus has been trending up since around the mid-1990s, similar to the trend for total phosphorus. By contrast, dissolved phosphorous has been generally going down over the record. Clearly this is the reason behind that total phosphorus rise which is consistent with the sediment trend. And a frequent question that we have been asked is: Are these trends due to changes in the upstream watershed or due to reservoir system? To answer that question we have plotted the trends of flow-normalized loadings for reservoir input in terms of the sum of Marietta and Conestoga Stations. You can see that for sediment, [total] phosphorus, particulate phosphorus, total nitrogen, and other species not shown here. These trends have been generally downwards over the record for most seasons. In other words, the upstream watershed has been generally seeing effective action in terms of loading reduction. And the trend observed at Conowingo can be largely related to the reservoir effects. And another question we have been asked very frequently was that: are these trends biased by storm-flow samples? We don't have consistent storm sampling at upstream and downstream locations during different storm events. For example if we have some more frequent sampling of storms in the recent period, but not the early period, that might bias our estimates of the loading rise magnitude. To address that question we have come up two additional scenarios of analysis. The blue points (or the blue line here) represent the scenario based on all samples available. And this orange one was obtained based on all samples but Hurricane Ivan, which has surprisingly extremely high concentration for sediment and phosphorus in the record. The highest [concentration values] in the record. And the purple one here indicates the scenario that we assume there was no storm-flow sample collected at all. And you can see this very consistent pattern for input (Marietta) sediment and TP. All these three lines overlap very nicely. In other words, the storm-flow sample really is not making a difference here in terms of reservoir input. And I would say the same for total phosphorus at Conowingo as well. For the sediment here if you look at the long-term trend either from the beginning to the end of the record, or from the 1995 period which we think the change or the rise began, this increase appears to be consistent among the three scenarios. But we do want to highlight that there has been a very dramatic increase during the early 2000s because of the Hurricane Ivan sample. That huge 3,680 milligram per liter value. This plot reminds us that this long-term trend is very valid but if we want to know the year-by-year trend calculation magnitude. Then that might be biased by the storm-flow samples, just to be cautious around that. In the follow-up work, which was published recently by Zhang, Hirsch, and Ball, we seek to quantify the broad long term changes in reservoir net deposition. Remember we have reservoir input, we have reservoir output so we can calculate mass balance. And to better understand the uncertainties of our statistical analysis and importantly, to quantify the relative importance of the so-called scour flow events and the moderate high-flow events which we call sub-scour levels. So these scour flow events which are above 11,000 CMS or 400,000 CFS, have been a major concern. Because they can deliver a huge amount of materials to the Bay within a very short period of time. But they only represent 0.1% of the time in the history. So we want to know, for the 99.9%, how does the reservoir function? How does that reservoir function change over time? And here is the result. We have done three different levels of analysis for that particular study. This one is the temporal change in the C-Q relationships, which is based on smoothing curves which we call LOWESS. Simply we plotted the data sets, the monitoring data sets for sediment -- for total phosphorous shown here. The log concentration as a function of discharge. And then we separated the data into three different temporary intervals. The green for the earliest one and the red for the latest one. And then we fit the LOWESS curve, the LOWESS smoothing curve, which is a non-parametric smoothing technique. You can see that there is a general increase in concentration for the red line compared with the other two lines. In other words, we have higher concentration for the most recent period for a particulate discharge. And that increase in concentration has occurred under a wide range of flow conditions, including those sub-scour levels shown here [on the left of] the purple line. The purple line here is the scour threshold. By contrast, Marietta shows a general decline in concentration for the most recent period, compared with the two earlier periods. In other words, we have positive progress in terms of reducing the concentration at the Marietta site. The implication here is that this positive progress has not been propagated across the reservoir system to emerge at the dam. In other words, there has been something going on in the reservoir system. And this is a very recent new publication taking a look at the C-Q relationships at a different angle. I don't want to go into all the details but the key contribution here was to extract this beta-2 coefficients from the WTRDS model. And use that to give a more realistic and consistent representation of the C-Q relationship over time or season or discharge. Here shows the results for this beta-2 coefficient extracted from the WTRDS model, shown in color as a function of time and discharge. And the red color means a higher sensitivity, or a higher dependence of concentration on discharge. If you look at this 5000 or the 10,000 CMS level, you can see this general increase in the magnitude of these beta-2 coefficients. In other words, at the dam there has been increasing sensitivity of concentration to discharge. And I want to point out that all this Y-axis scale is below the scour level. So clearly this shows this dynamics going on in the reservoir below the scour threshold. Okay in terms of mass balance, here we have looked at the results based on the true-condition estimates from WRTDS. We calculated the input and output and then we calculated the net deposition or net scour. The way we approached this was to use output/input ratios. Recall that this output/input ratio: less than one it means net deposition, above one it means net scour. So for sediment and phosphorus it was not surprising that they are generally less than one, because the reservoir was historically effective in terms of trapping of these materials. But we can see this ratio has been going up since around mid-1990s or 2000. And this particular phosphorus you can see is approaching to 1.0; it is getting closer to that. In other words, we see reduced net deposition based on this particular plot. For total nitrogen the ratio has been stable at 1.0 because it's largely in dissolved from, which is not heavily affected by the reservoir system. Next we want to quantify the uncertainties around these ratios. To do that, we have focused on the centerline of the box plot formed by the annual median of the 365 daily O/I ratios. What we did was to run bootstrap analysis using 100 sets of representative data sets from the original record. And then we calculated the 100 centerlines or the 100 annual medians from the box plots, then we quantified the mean and the 95% confidence interval for that annual median. You can see that, when zoomed in to this finer resolution, again this increase in the ratio is apparent for sediment and total phosphorus. In other words, the reduced net deposition conclusion is robust based on this analysis. In addition, when we zoom into this fine resolution, the total nitrogen appears to also have a slightly increasing trend in the recent decade, which may indicate an increasingly important role for nitrogen. That's for that analysis. And then, are these trends in O/I ratio biased by the differential highflow sampling? We know that these two stations Marietta and Conowingo have not been sampled comparatively over time in terms of high flow events. So we have taken a very hard look at the availability of the high flow data sets. We found that during these three big events there are big difference in terms of high flow sampling. The three events are 1996 ice jam, 2004 Hurricane Ivan, and 2011 Tropical Storm Lee. You can see the hydrographs shown here for Conowingo (the outlet station). This is for sediment, but similar for total phosphorous and total nitrogen. It has been sampled quite well in terms of either the rising limb or the peak during the three events. But at Marietta, which is the inlet for the reservoir system, you can see that the peak or the rising limb has been simply missed in terms of high flow sampling. So this may bias our estimate in terms of output/input ratio analysis. So to remove that complication, we have done some sensitivity analyses in which we removed all those big events samples for both stations. And we found that with this re-run -- we found that with these equally-censored samples: Both sediment and phosphorous also again show increasing trend in terms of O/I ratio. In other words, reduced net deposition. The next big question: Is the O/I ratio trend associated with highflow only? To address that, what we did was to simply plot the ratio, similar to the previous box plot. But here we break down that into five different flow classes based on the ranking at the Conowingo Dam. We found that for each of these different flow classes, ranging from the very low flows all the way to the highest flows, there appears to be an increasing trend in the O/I ratio, or reduced net deposition. In other words, this loss of trapping performance is not limited to scour flows only. It has occurred under a wide range of flow conditions. We can also look at the flow class -- which has contributed the most to the mass delivery across the dam? We calculated the relative contribution by each of these five flow classes. We found that it is Q4, which contains sub-scour high flows, that contributes the most in terms of reservoir input for flow volume, sediment, TP and TN loading. Similarly, they have contributed the most in terms of [reservoir output for] flow, total phosphorus, and total nitrogen loading. But for sediment loading at the reservoir output, it is the second highest, which may be due to the reason that the sediment estimation during the several big events have been overestimated by the WRTDS model. The final piece of analysis here is on stationary-model analysis, where we tried to tease out the effects of changing concentration regression surface from WTRDS. This is a typical output from WRTDS showing the estimated concentration for sediment at Conowingo as a function of discharge and time. You can see that the color indicates the log concentration here. And the color of the concentration estimates are not the same between 1990, 2000 and 2010. In other words, these concentration estimates have been changing over time. So any interannual comparison of the true-condition estimates can be complicated by two things. One is the change in the regression surface itself, the other is the change in the annual history of flow within the year of choice. So we want to isolate this part A effect. What we did was to develop three stationary models using these three different annual surfaces. What we did was to simply repeat this 1990 surface over the entire period of record, to come up this stationary model regression surface. Similarly, we did that for 2000 and then we did that for 2010. With these three different stationary surfaces, we applied the same actual discharge at Conowingo Dam to get the stationary model estimates. Because we used the same flow discharge history, so any difference among our estimates reflect the difference in part B (or the change in reservoir function). Next I want to show the result on this. We have many different results shown from this study, but here is just one concrete summary on them. What I show here is the prediction of cumulative sediment net deposition in the reservoir system for three different years of hydrology. A wet year, an average year, and a dry year. And you can see here, the red line indicates this most recent reservoir surface, or the 2010 reservoir condition. And the green line is the 1990 reservoir condition. Comparing these three model results you can see the red line is always below the other two lines. In other words, for the three different years of hydrology, we see reduced net deposition for those years. In addition, the uncertainty from the bootstrap runs indicates that this red envelope is entirely below the blue or the green [not red] envelopes. In other words, this is statistically significant, for the results. So finally what is the significance of the Susquehanna trend in terms of the NTCBW trend? And this is a publication that we made in 2015, quantifying the trend from the nine RIM sites on a seasonal basis. We found that -- here we define NTCBW as sum of the nine RIM rivers -- in terms of the annual average for this study period from 1979 to 2012, the Susquehanna has contributed about 62% of flow and a similar level of total nitrogen. This is not surprising because total nitrogen is mainly in dissolved form; it is not heavily trapped by the reservoir system. For phosphorus and sediment you can see slightly below 50%, because of the historical retention within the reservoir system. The interesting thing here is that in terms of NTCBW total rise in the final decade in our evaluation period, we found that in terms of particular species the Susquehanna has contributed to about 92% of sediment rise; Bay wide. But only 68% total phosphorus rise; Bay wide. And this is probably due to that there has been increasing trend in total phosphorus in other part of Bay watershed as well. In fact, we have separately analyzed the trend for the non-Susquehanna part of the Bay watershed, where we found a similar contrast between dissolved species and particulate species, with generally downward trend for dissolved species but increasing trend for particulate species. So with that I want to come to the conclusions. With these statistical analyses of the long term records we have found declined reservoir input for both dissolved and particulate species. Increased reservoir output for particulate species at Conowingo Dam. In addition, decrease in net deposition of sediment and TP has occurred under a wide range of flow conditions, including sub- scour levels. Mass of delivery across Conowingo Dam has been dominated by moderately high flows. Our conclusions are supported by a series of uncertainty and sensitivity analyses. In term of future research, we recommend further monitoring and modeling of the reservoir system in terms of loading, nutrient biogeochemistry, its impact on downstream water quality, and also consideration of all these different issues under a wide range of flow conditions. I think many of you are familiar with this: The largest reservoir in the lower Susquehanna River is no longer an effective trap for sediment and phosphorus. It means that the key assumptions behind the TMDL developed in 2010 on the parts of this reservoir system are no longer valid. So the partnership needs to improve the representation of the reservoir system in the Phase 6 Watershed Model using all pieces of available evidence and information. And the information I have is that in June 2017 the Phase 6 Model will be released, with the incorporated Conowingo new representation. In addition, the partnership needs to evaluate the options to allocate the targets among the different jurisdictions to offset the loadings from the Conowingo. And the EPA will release the final Phase III WIP targets in December next year. So with that I would like to make several acknowledgments. My advisor Bill Ball at Hopkins and Bob Hirsch from USGS. Several collaborators: Damian Brady, Walter Boynton, and Doug Moyer. And also Joel, Gary, and Ken for their comments on the various work. The funding agencies and the USGS and SRBC who collected the data sets. And these publications are available online; I have provided links in the presentation, which you can download from the STAR meeting webpage. With that I would like to thank you for your attention and welcome your questions. [ Applause ]Normal.dotm05319518212Microsoft Macintosh Word015142falseTitle1US EPAfalse21365falsefalse15.0000Qian Zhangdtaillie7@gmail.com22017-01-09T14:00:00Z2017-01-09T14:00:00Z

Coming Soon