The recent (1992-date) record rise in the level of the Devils Lake, North Dakota, has led to a number of questions as to the nature of regional and global climate variability, and the utility of existing methods for forecasting lake levels and assessing the associated flood risk. A purpose of the work presented here was to explore the connection of the Devils Lake volumetric fluctuations to interannual and longer regional and global climate fluctuations, and to test the performance of recently proposed time series forecasting methods.
Wiche and Vecchia (1995), and Osborne (1998) provide background information on the lake, and prior forecasting and climate analysis. Key trends in hydroclimatic variables in the Devils Lake region are first identified and discussed in the context of large-scale hydroclimate variations. Hypotheses as to operative climatic mechanisms that have led to the recent rise in the lake level are developed from this analysis. Two types of long-range lake forecasts are then considered.
A forecast of lake levels for the near future (1-5 years or an inter-annual period) is developed for assessing the potential of continued flooding and associated needs for disaster relief. Second, since closed basin lakes typically exhibit long memory, procedures to estimate conditional probabilities of lake levels for extended horizons (e. g. , over a 30-year flood control project, or inter-decadal periods) given current conditions were explored.
Nonlinear time series analysis methods using the historical volumes of Devils Lake and selected climate indices as predictors were used to develop the inter-annual forecasts as conditional means of expected future volumes. A variety of time series modeling approaches were explored for the inter-decadal forecasts. Results are presented here for a linear, Bayesian autoregressive time series model that incorporates model and parameter uncertainty. We conclude that direct applications of existing time series analysis methods are not well suited for the development of long-range probabilistic forecasts of Devils Lake.
The recent trends of Devils Lake are consistent with large-scale changes seen elsewhere. However, whether these changes are part of the natural long-term variability of climate or represent a changed climate due to human influence remains inconclusive. Consequently, while we are able to relate the recent Devils Lake trend to causative hydroclimatic factors, we are unable to confidently predict the long-term future levels of the lake. Only qualitative remarks are offered to characterize the uncertainty associated with using the past as a guide to the future of Devils Lake.
Trends in Local, Regional and Hemispheric Hydroclimatic Variables Like most closed basin lakes, the Devils Lake exhibits dramatic fluctuations (Figure 1) over decadal and longer periods, that derive from climatic fluctuations. There is a limited literature (e. g. , Lall and Mann, 1995, Mann et al, 1995, Pusc, 1993, Wiche et al. , 1986) diagnosing the climatic causes of such fluctuations. The contributing drainage area of such lakes varies with climate state. The chain lakes above Devils Lake and other depressions store water during dry periods, but contribute runoff during protracted wet periods.
This change in drainage area may be a key explanation for the dramatic changes in the Lakes volume, subsequent to moderate changes in the climate signal. Notable increases occur in 1950s, 1970, 1980s, and the late 1990s. These periods are also important at the regional streamflow stations. It is useful to first look at the annual cycle of monthly changes in the Devils Lake volume (Figure 2), to motivate the search for trends in monthly precipitation, temperature, Sea Level Pressure (SLP), and cloudiness. The influence of ice cover in the winter months decreases the magnitude of changes observed.
On average, the lake volume increases in spring (April through June), due to snowmelt runoff. Decreases occur in the summer months, with the greatest decrease usually occurring in August. However, increases or decreases can occur in any calendar month. At first glance, one would suspect that the winter/spring precipitation and summer/fall temperature (and hence atmospheric circulation patterns), are most important for diagnosing the changes in the volume of the lake. Monthly trends and the base climatology of these variables are reviewed next.
The climatology of monthly precipitation amounts for selected stations (Figure 3) is shown in (Figure 4). The data for these stations was obtained from ftp://ftp. ncdc. noaa. gov/pub/data/ushcn/daily/. Note that precipitation peaks in June, and that September/October precipitation can occasionally be as large as the average June precipitation. The range can be dramatic (e. g. , 0. 5 to 12 for June at Bottineau), even though the total average amounts reflect an arid climate.
Winter precipitation is on average much smaller than the summer/fall precipitation in the region. It is consequently remarkable that the largest annual volume increase for the lake is seen in April. The subtle role of subsurface hydrologic processes and evaporation in modulating the lake inflows and hence the volume is indicated. Monthly precipitation data were used to investigate the variability in precipitation. Data were obtained from the U. S. HCN web site (ftp://ftp. ncdc. noaa. gov/pub/data/ushcn/) and the FILNET version of the data, are used.
The accompanying literature notes that the FILNET data has been adjusted for the time of observation bias, maximum and minimum temperature system (MMTS) bias, station moves and changes bias, and contains estimated values for missing/outlier data. These adjustments provide a long, serially complete data set useful. An examination of these records reveals largely consistent precipitation trends in the recent period. Month-by-month trends for the Langdon station are presented in Figure 5.
Note that no major positive trends in the winter/spring (Dec-Apr) precipitation are evident, while May through August and October/November exhibit a positive trend in the recent period. Given that the summer and fall are usually periods of lake volume decrease, these precipitation trends are significant in that they may imply a situation where the lakes annual cycle of decrease may be reversed. The increased regional wetness in the summer may also lead to a higher contributing drainage area and runoff, and to a lower potential evaporation associated with increased local humidity.
Recently, Karl et al. (1998) investigated the secular trends in both amount and intensity of precipitation in the U. S. with relatively long data sets (1910-1996). They found positive trends of heavy rainfall in the region covering the nearby upper Mississippi River basin in all seasons, except winter, where a negative trend was found. The largest trend reported for the region was for summer, consistent with the finding of Angel and Huff (1995). These analyses are consistent with our observations from the local station data.
An examination of the 1896 to 1996 monthly temperature record at the Langdon station (Figure 6) shows no unprecedented changes over the recent period of concern. January, April, May and July temperatures have reversed earlier warming trends during the 1990s. The period of decreasing and extremely low lake levels may be associated with lower summer precipitation and warmer summers. The cooler and wetter summer/fall regime is associated with the recent rise of the Devils Lake.
A comparison between the July and October precipitation trends and the local Sea Level Pressure (SLP) trends is instructive. From Figure 7, we see that for both months the increasing precipitation trend is accompanied by a decrease in the mean monthly SLP for the month. Similar trends in the local SLP are evident for June and August. This observation sets the stage for an examination of trends in the larger-scale precipitation, river flow and atmospheric circulation fields.