Scientific Investigations Report 2006–5209
U.S. GEOLOGICAL SURVEY
Scientific Investigations Report 2006–5209
It is of interest to know whether the data collected during the 3 years of this study and the resulting insights into water quality generally are applicable, or whether 2002–04 were in some way extraordinary in a historical context. Equally important, because there was a small fish die-off in 2003, was to investigate whether, in the context of 15 years of data, certain conditions distinguish the years when fish die-offs occurred. In addition to the die-off that occurred in 2003, severe die-offs occurred in 1996 and 1997, and a smaller die-off occurred in 1995. The comparison of the numbers of dead fish collected during die-offs is compromised by the lack of reliable estimates of the effort expended in collecting the fish, so these comparisons should not be considered reliable quantitative measures of the relative severity of the die-offs in the mid-1990s. Nonetheless, based on the number of dead fish collected, the most severe die-offs occurred in 1996 and 1997, and the 1995 die-off was more severe than the 2003 die-off [472 dead suckers were collected in 1995; 4,453 in 1996; 2,335 in 1997 (Perkins and others, 2000)], and 113 in 2003 (B.J. Adams, U.S. Geological Survey, written comm., 2006). It is possible, however, that the low number of fish collected in 2003 reflects the devastating loss to the population caused by the mid-1990s die-offs, and that from a water-quality perspective the 2003 die-off was as severe an event as occurred in the mid-1990s. The die-off in 1995 differed from the other three in that it did not begin until September 1, whereas each of the other three began in July [July 16 in 1996, July 23 in 1997, and July 22 in 2003 (Perkins and others, 2000)]. Because of the need to limit the scope of the analyses, a focus was placed on the July/August time period, and the 1995 fish die-off was not considered.
The datasets that are long enough and of adequate quality for such an investigation are limited, but this discussion makes use of four. The first is the water-quality dataset collected by the Klamath Tribes since 1990. This dataset consists of water samples collected every 2 weeks and analyzed for nutrient and chlorophyll a concentrations, and accompanying water-quality profiles of dissolved oxygen, pH, temperature, and specific conductance. The second dataset is a weather dataset (wind speed and direction, and air temperature ) obtained from the National Climatic Data Center (NCDC, 2006) collected since 1990 at the Klamath Falls airport. Both of these datasets are used to look at a 15-year length of record. The third dataset is a wind dataset collected by the Bureau of Reclamation (1997) at an Agrimet site near Agency Lake; the length of record is only the 5 years since 2000, but the comparison with the NCDC data for that 5-year period is informative. The last dataset is the discharge collected at the Williamson River gaging station maintained by the USGS , 10.3 mi upstream of where the Williamson River discharges into the lake (fig. 1; USGS site identification No. 11502500; U.S. Geological Survey [no date]).
One of the distinguishing characteristics of 2003 was water temperature. The median study-area water temperatures peaked at the end of July in 2003 at 24.9°C, which was higher than the highest temperatures observed in 2002 (23.3°C) or 2004 (23.5°C). High water temperatures can contribute to an LDOE by accelerating oxygen demanding processes, and can contribute to a fish die-off by facilitating the spread of disease. Historic water-temperature records are not available, but water temperature is largely a function of air temperature, and so it is logical to look for years that stand out in the air-temperature record. The median air temperature for the months of July and August were determined from hourly values in the NCDC dataset, and subjected to an ANOVA (table 8). Fish die-offs in 1996, 1997, and 2003 began in mid- to late July, although the majority of dead suckers that were collected in 1996 and 1997 were collected later—between August 26 and September 12 in 1996, and between August 20 and September 8 in 1997 (Perkins and others, 2000)—so both July and August temperatures were used in this analysis. The years 1994, 1996, and 2003, although statistically indistinguishable from each other in this analysis, as a group had the highest median air temperature in July of the entire 15-year record. In August, the years 1992, 1996, 1997, 1998, and 2001 ranked as the hottest group of years, while 2003 was cooler. It seems appropriate to conclude that 3 of the 4 fish die-off years were among the warmest years in July and August of the 15-year record.
An ANOVA also was done on the wind-speed data collected at the Klamath Falls Airport (table 9). Because of the way wind speeds are distributed throughout the day, peaking for a relatively brief period but being much lower over the rest of the day, it was difficult to see important differences by looking at the entire distributions, which were dominated by the lowest wind speeds. Under the assumption that it is the peak wind speeds that are most important, only wind speeds that exceeded 50 percent of the distribution for the entire month were extracted for this analysis. In July, the years 1996 and 1997 ranked as the lowest wind speed group of the 15-year dataset; 2003 ranked slightly higher. In August, all 3 die-off years ranked higher than they did in July; 1996 and 2003 ranked in the second lowest group, but 1997 ranked in the middle of the 15-year record. Low July and August wind speed, therefore, seems to be a distinguishing characteristic of fish die-off years.
Because short-term wind reversals were observed to cause oxygen concentration to decrease in the study area, an attempt was made to identify years in which wind reversals might have been relatively numerous compared to other years, thus increasing the average traveltime in the trench. To do this, the vector wind data from the NCDC were projected onto a set of axes rotated 60° clockwise, such that one axis was aligned along the length of the lake. The resulting “along-lake” component of the wind is positive for a northwest wind. An ANOVA was performed on this component of the wind (table 10). The analysis shows clearly the reversals in August of 2003, as evidenced by a slightly negative along-lake component, but these reversals occurred after the peak of the July LDOE that precipitated the fish die-off. August 1997 also had a slightly negative median along-lake component, indicating that there were several wind reversals in August of that year, when the wind speed was moderate. Three die-off years—1996, 1997, and 2003—rank in the middle of the 15-year record in the magnitude of the July along-lake component, indicating that the number of reversals in July of those years was not exceptional compared to the entire 15-year record. Fish die-off years were not, therefore, characterized by a higher-than-average number of wind reversals in July or August.
There were indications that the NCDC wind dataset was not ideally suited to highly quantitative analysis. There was clear evidence, for example, of “binning” in the NCDC wind dataset—values always fell on discrete values rather than being continuous. In addition, wind speeds of zero were numerous, so it is likely that all values below a threshold became zero. As a means of assessing the quality of the data and the confidence one should have in the ANOVA analysis, an analysis of only 2000–04 data was compared for consistency to the same analysis for those 5 years of wind data collected at the Agency Lake Agrimet site (table 11). Among other discrepancies, the year 2003 ranks in the lowest wind speed group in the NCDC data, but ranks second highest in the Agrimet data. The relative ranking among the 2002, 2003, and 2004 wind datasets collected at the profiling buoys in this study are consistent with the Agrimet data but not the NCDC data, the 2003 July wind data being the highest of the three. The significance of this comparison is that the NCDC wind data probably should be interpreted qualitatively. It is appropriate, for example, to conclude that wind speed in the years 1996, 1997, and 2003 ranks toward the low end of the entire 15-year period, but the quantitative differences between any pair of years is not reliable, and the characterization of 2003 as a very low wind speed year should be viewed skeptically because it is inconsistent with data collected over the 3 years of this study, and with the 5 years of Agrimet data.
An ANOVA similar to that used to rank the wind and air temperature datasets was used to rank the years of the 15-year water quality dataset collected by the Klamath Tribes (fig. 31). Only data from seven lake sites that were consistent through 1990–2004 were used in the analysis. The variability between sites and within a 2-month time frame largely swamps interyear variability in chlorophyll a and dissolved oxygen, making the distributions nearly indistinguishable from each on a yearly basis. This type of aggregate analysis over several sites and the entire July–August time frame averages out important distinctions, and no characteristics emerge to distinguish the fish die-off years of 1995, 1996, 1997, and 2003. Attempts to relate the Klamath Tribes water-quality variables to NCDC wind speed over the 15 years of record available in both datasets yielded nonsignificant relations as well (fig. 32), in spite of the fact that significant correlations were obtained by others on a shorter period of record (Kann and Welch, 2005).
In contrast to chlorophyll a and the other nutrients, large multiyear trends in the distribution of the July–August ammonia concentration have been observed over the 15-year record; these multiyear trends strongly suggest that climatic factors are involved. In order to test this, correlations were calculated among two climate variables, July–August NCDC wind speed and the October–May cumulative discharge in the Williamson River, May–June and July–August median values of ammonia and chlorophyll a concentration, and the median of the difference between the water column maximum and minimum dissolved oxygen (denoted ∆ dissolved oxygen) (table 12). July–August median chlorophyll a concentration and ∆ dissolved oxygen were not correlated with either climate variable, indicating that year-to-year variability in these variables is not climate-driven. Nor are year-to-year changes in chlorophyll a concentration, ammonia concentration, or ∆ dissolved oxygen correlated with each other. July–August median ammonia concentrations were, however, highly correlated with both climate variables, confirming that the interannual variability in ammonia probably is climate related. Because the two climate variables—July–August wind speeds and discharge from the Williamson River for the previous October–May—are themselves correlated, it could be difficult to sort out which, if either of these, represents a causal relation. Because, however, the July–August wind speed median also was correlated with the ammonia concentration in the previous months of May and June, it is more likely that the correlation between ammonia and Williamson River discharge represents some kind of causal relation. A straightforward explanation for this relation would be that higher discharge from the Williamson River during the winter and spring delivers more organic material that acts as a source of ammonia when it decays in the summer. Although the climate connection to ammonia appears solid, ammonia concentration alone is not a predictor of fish die-off years. Concentrations in 1997 were among the highest on record, but other years had similarly high concentrations. Concentrations in 1996 were in the midrange, and concentrations in 2003 were among the lowest of the record, comparable to the early 1990s.
The aggregate analysis of the Klamath Tribes’ dataset based on July–August distributions at the sites in the northern part of the lake did not yield an obvious way of identifying the fish die-off years in the 15-year record. Another approach taken was to focus on individual sites and test the idea of an “indicator site”—a particular site that could be reliably used to indicate the onset of an LDOE in the study area. Because the water quality in the study area is largely a function of the quality of the water entering the study area through the trench between Bare Island and Eagle Ridge, the onset of an extreme event should be accompanied by the transport of a large volume of water with a low concentration of dissolved oxygen through the trench. One of the Klamath Tribes’ sampling sites is located in the area of the trench between Bare Island and Eagle Ridge; this site is known as Eagle Point, or EP. Another site located roughly between the 2003–04 UKL06 and UKL07 sites is Midnorth, or MN (fig. 1). The latter site is, like UKL07, broadly representative of the conditions in the deeper part of the study area that is preferred habitat for adult suckers and is bounded by the remnants of the trench and the entrances to Ball and Shoalwater Bays to the south. Data from these two sites were used in a principal component (PC) analysis to test the “indicator site” idea.
Seven variables were included in the PC analysis—chlorophyll a concentration, ammonia concentration, average water column dissolved oxygen concentration, minimum water column dissolved oxygen concentration, the difference between maximum and minimum water column dissolved oxygen concentration, maximum water column temperature, and the difference between maximum and minimum water column temperature. At site MN, the first three eigenvalues were able to explain 36, 28, and 12 percent of the total variance, respectively; at site ER the first three eigenvalues were able to explain 39, 31, and 14 percent of the total variance, respectively. The loadings of the first three principal components on each variable are given in table 13, and the scores on each July–August sample date are shown in figures 33 and 34. The first principal component at each site largely captures variability in average water column dissolved oxygen concentration and chlorophyll a concentration, fluctuating opposite to water temperature and ammonia, although the contribution from variability in ammonia is relatively small. The second principal component captures primarily the variability in the water column stratification, and the third principal component is dominated by the variability in ammonia concentration.
The sample date July 28, 2003, has the largest negative PC1 score, reflecting a low average water column dissolved oxygen concentration and low chlorophyll a concentration, in combination with high water temperature and relatively high ammonia concentration. In this analysis, July 2003 appears unique in the context of a 15-year record. The 1996 and 1997 points that have a relatively high PC2 score and stand out from the cloud of other data points are from late July, reflecting the exceptional stratification during that time period. This characteristic of those 2 years has been documented by others (Perkins and others, 2000; Kann and Welch, 2005) in the context of a shorter length of record; in this analysis the exceptional stratification in those years stands out in the context of a longer 15-year record.
In this analysis, the fish die-off years do not group together in obvious separation from the other years on record. This suggests that 2003 does not have a lot in common, in terms of the variables included in this analysis and their correlations, with the other 2 fish die-off years, 1996 and 1997. There are two possible reasons for this. The first is that the years really were different in important ways, and that there is more than one set of circumstances that lead to a fish die-off. The second is that the two-week interval of the sample dates resulted in the most relevant conditions of 1996 and 1997, immediately preceding the die-off, being missed.
The statistics of the distribution of dissolved oxygen, ammonia, and chlorophyll a concentrations, and water temperature for the entire July and August time period were compiled separately for sites ER and MN in tables 14 and 15, respectively. The values of these variables on the sample dates most closely associated with the onset of the fish die-offs in 1996, 1997, and 2003 also are compiled in the tables. Samples were collected on July 28, 2003, at the peak of the low dissolved oxygen event. The description of this period as characterized by very low dissolved oxygen concentrations in combination with very high ammonia concentrations, as well as high temperatures and low chlorophyll a, reappears in this analysis. At both sites, dissolved oxygen and chlorophyll a fall within the first quartile, and ammonia and temperature fall within the fourth quartile, of all sample dates for the July–August period in the 15 years of data.
Data collected on the 1996 and 1997 sampling dates are more difficult to interpret because there are no continuous datasets with which to identify when the worst conditions actually occurred. These sample dates probably missed the time period of lowest dissolved oxygen at site ER. Nonetheless, the August 20, 1996 and August 12, 1997 values at site MN generally are consistent with a low dissolved oxygen/high temperature/low chlorophyll a/high ammonia combination, with the exception that the temperature fell in the third rather than fourth quartile. Consistent among all fish die-off dates at site MN is the combination of average water column dissolved oxygen concentration and chlorophyll a concentration in the first quartile (less than 6 mg/L and less than 65 µg/L, respectively) and ammonia concentration in the fourth quartile (greater than 520 µg/L). If these same criteria are used to screen data collected at site MN during all July–August sample dates in the 15-year dataset, then only four additional dates (out of a total of 71 dates) pass the screening, from 3 non-die-off years: August 12, 1998; August 9 and 22, 2000; and August 29, 2001. It appears, therefore, that there is some validity to screening for die-off conditions based on these criteria at only site MN. These screening criteria offer no predictive capability, of course; their utility is in being able to better describe the set of conditions that appear to have a greater risk of leading to a die-off, and they point to some commonality among the die-off years.
The evidence for a correspondence between low wind speeds and fish die-off years is compelling; all 3 years—1996, 1997, and 2003—stand out in the historical record has having lower-than-average July wind speeds. The connection that has been made between low wind speeds and fish die-off years is increased density stratification and the low dissolved oxygen in the lower water column that often accompanies it. Yet, low wind speeds did not correlate well with high dissolved oxygen stratification on either a daily basis within a single year or on the basis of a 2-month average over 15 years of data, nor were all fish die-off years characterized by exceptional dissolved oxygen stratification. Furthermore, strong dissolved oxygen stratification in the water column was not directly the cause of the recent, more well-documented, fish die-offs. This was noted by Perkins and others (2000) regarding fish die-offs in 1996 and 1997, when the great majority of dead suckers were counted after stratification had collapsed. It also was apparent in the difference between 2002 and 2003 in this study. Although 2003 probably was a lower-than-average wind speed year, it was not a year characterized by an exceptional degree of dissolved oxygen stratification, particularly in July, when the most severe LDOE occurred. Indeed, one of the things that made the event exceptional was the fact that stratification was weak and dissolved oxygen was low throughout the water column. In 2002, which also was a lower-than-average wind speed year, the water column was more consistently stratified in dissolved oxygen and to a greater degree than in 2003, and for the most part endangered suckers persisted in the study area and, apparently, survived.
An extended period of dissolved oxygen stratification in the early to mid-summer could increase the risk of a fish die-off several weeks later. For example, low dissolved oxygen in the lower water column could present chronic stress to fish that weakens them and makes them more susceptible to adverse conditions later in the summer. Low winds also may play a role in bloom dynamics, in that even moderate stratification tends to provide favorable conditions for AFA, thus leading to higher biomass and more dramatic bloom declines. From what we can determine from the available data, these possibilities seem applicable to the 1996 and 1997 fish die-offs, but the fact remains that in 2003, when the die-off began at the end of July, it was not preceded by a period of several weeks of exceptional dissolved oxygen stratification.
If there is a common set of conditions that link all the fish kill years, it seems possible, if not likely, that the correspondence between low wind speeds and fish die-off years has a different significance. One possibility is that interannual variability in wind stress influences interannual variability in water quality through the wind-driven circulation. Oxygen production and consumption varies around the lake, in part as a consequence of the changes in bathymetry that water encounters along a flow path. Because the wind-driven circulation is responsive to wind stress, lower wind speeds slow down the overall wind-driven circulation relative to higher wind speeds and increase residence time in different areas of the lake, providing, in general, more time for both oxygen demanding and oxygen producing processes to act in different areas of the lake. Understanding how the nonconservative processes interact along a flow path as water circulates around the lake faster or slower in response to wind stress is difficult using data from a fixed-site sampling strategy, and likely will require the use of a computer model.
In addition to the role that low winds may play in the development of an LDOE, it is clear that any severe LDOE, and consequently any fish die-off, will be coincident with a dramatic decline in the AFA bloom. Not only do the senescing cells provide water column oxygen demand, but the shutdown in photosynthetic production also is significant, as water column and sediment oxygen demanding processes continue to consume oxygen. Because, however, the trigger for a dramatic bloom decline is not well understood, we are still unable to determine the root cause of a severe LDOE. An important area of future research will be to try to better understand what causes the AFA bloom to decline, especially when it does so precipitously. Although the cause of AFA bloom declines remains undetermined, we have been able to determine some characteristics of a severe LDOE that all fish die-off years have in common with each other, but with few other non-die-off years. Water coming through the trench and into the northern part of the lake during these events was characterized by exceptionally low water-column averaged dissolved oxygen and chlorophyll a in combination with exceptionally high ammonia, indicating that nearly all AFA biomass had already been converted to inorganic form, in the process consuming nearly all available oxygen. An additional characteristic seems to be high temperature, which can accelerate the respiratory and decay processes that consume oxygen.
U.S.
Department of the Interior | U.S. Geological
Survey
Persistent URL: https://pubs.water.usgs.gov/sir20065209
Page Contact Information: Publications Team
Page Last Modified: Thursday, 01-Dec-2016 19:13:03 EST
