Prairie Drought Project Intermediate Report

1. Objectives:

1. Reconstruct changes in salinity of three prairie lakes at 1-3 year intervals over the past 800 years.

2. Calibrate inferred lakewater salinity with common indices of drought severity using 50-100 years of local instrumental climate and agricultural data.

3. Quantify the frequency, intensity and duration of drought during the past 800 years in each prairie province.

4. Determine relationships among paleoclimatic reconstructions in each province to estimate the geographic extent of past droughts.

5. Document how drought occurrence impacts water quality.

6. Develop predictive models to estimate drought occurrence during the next 5-50 years.

7. Incorporate future drought probabilities into actuarial calculations for determination of crop insurance premiums.

8. Publish a summary monograph on the history and future of droughts in western Canada.

| top |

2. Progress to date:

In Nov 1997, we were awarded a three year strategic grant to develop novel paleoecological and mathematical techniques to predict the intensity, duration and frequency of droughts on the Canadian Prairies. This interim report summarizes the first 2 years of this project.

Surveys to identify climatically-sensitive lakes (total 35) were conducted in March-July 1998. Short cores were obtained at each lake, sectioned in 1 cm intervals, and analyzed for fossil diatoms, evidence of historical changes in salinity, and congruence with known climatic events (e.g., droughts of 1930s, 1980s). Based on these analyses, Humboldt Lake SK (52^o08.5'N, 105^o06.48'W) and Chauvin Lake AB (52^o41.41'N, 110^o06.02'W) were identified as climatically-sensitive sites. Two-meter cores encompassing ~1000 years were obtained from each lake in July 1998 and sectioned in 2.5 mm intervals. Poor diatom preservation in all 10 MB lakes necessitated sampling of an additional 28 lakes in June 1999. As a result of this effort, we obtained a core from the final prairie site - Nora Lake, MB (50^o28.30'N, 99^o56.21'W). In addition to our three main sites in AB, MB, and SK, we have developed collaborations with scientists at Queen's University that will allow us to add drought predictions at Oro Lake SK (49^o47'N, 105^o21'W) and Moon Lake ND (46^o51'27"N, 98^o09'30"W).

Radiometric analysis (²¹⁰Pb) of Humboldt and Chauvin lakes (Flett Research Ltd, MB) indicate that we have high quality cores that exhibit high temporal resolution (~2 yr). ¹⁴C-AMS analyses of buried wood are in progress (AECL, ON). Analysis of fossil pigments is complete, and preservation was uniformly excellent. Diatom analyses are complete for Humboldt Lake and Chauvin Lake, and are scheduled for Jan 2000-June 2000 for our MB site, as originally proposed. Drought prediction modelling is ahead of schedule because we have used pre-existing paleoclimate analyses of Moon Lake ND to develop our time series procedures.

Space limitations do not allow us to describe all our new data and modelling in any detail. Briefly, though, our drought prediction analyses clearly demonstrate that modern risk assessment models do not adequately capture the true frequency, duration and intensity of past droughts, and suggest that insurance models underestimate future drought risks, especially considering the accelerating effects of global warming on drought occurrence. Specifically, our analysis of Moon Lake paleoclimate records demonstrates that the probability of a drought as or more severe than that of 1988 is 42% by the year 2030 AD. In 1988, drought-related losses in North America exceeded $39 billion US dollars. Preliminary analysis of our SK site suggests that similar drought risks may apply to other prairie regions. Below we present details of our North Dakota analysis and compare these with on-going SK and AB analyses.

Drought measurement and fossil calibration: As originally proposed, our first drought workshop (11 Sept 1998; see below) was held to identify issues important to the user sector. Based on round-table discussions, we were asked to identify what type of drought was reconstructed from paleoclimatic records (agricultural, hydrologic, precipitation, fire, etc.), to demonstrate that fossil records were statistically related to long-term historical data, and to quantify regional similarities among drought reconstructions. To address the first two questions, we used new analyses of pre-existing paleoclimatic data from Moon Lake to investigate the relationship between fossil records and long-term historic data.

Canonical Correspondence Analysis (CCA) was used to compare temporal trends in fossil and historical data. Fossil data included diatom species composition and diatom-inferred (DI) lake water salinity (1891-1980). Climate data included the Rooy Drought Index (RDI), Bhalme-Mooley Drought Index (BMDI) and precipitation data (1893-1980). Crop insurance and agricultural (wheat) production data included drought-related indemnities (1948-1980), wheat production (acres planted, harvested, yield per acre, total production; 1889-1980), and crop failure ([harvested-planted]/planted; 1919-1980). Climate and crop data were smoothed using a 3-yr running mean to approximate fossil resolution (2.7 yr), while fossil species abundances were square-root transformed prior to analysis.

CCA indicated that three historical variables were significantly correlated to past diatom species composition (Moon - Fig. 4.1). Taken together, wheat yield per acre, total crop production and crop failure explained ~32% of past variations in diatom community composition, clearly indicating that paleoclimatic reconstructions were recording drought impacts on agricultural production. Monte Carlo testing demonstrated that each variable explained an independent and significant portion of the variance in past community composition. Linear regression also demonstrated that DI-salinity is significantly correlated with crop failure (r=0.45, P<0.05; Moon - Fig. 4.2). Finally, CCA identified a weak correlation (P=0.09) between mean BMDI (May-Oct) and paleoclimate data. Together, these analyses demonstrate that paleoclimatic reconstructions based on diatoms should accurately record the major changes in climate that influence production of economically important crops, particularly wheat.

Risk Assessment using Paleoclimate records: Having demonstrated that paleoclimate data record agriculturally-relevant droughts, we developed and applied new methods of risk assessment to better quantify the probability of future prairie droughts. Our analysis to date has been based on a 2000- yr paleoclimate record developed by our Queen's University collaborators. Because our risk assessment models were not scheduled to be developed until May 1999 (i.e., after SK core analysis), this action substantially accelerated our progress on time series protocols and analyses. Below we summarize our analytical procedure, then apply these results to the Moon Lake data.

| top |

3. Analytical details:

Plots of historical changes in DI-salinity indicated that there has been a natural decline in the salinity of Moon Lake over the past 2 millennia (Analytical Summary - Fig. 2.1). Because we were interested mainly in the temporal behaviour of extreme events (high salinity; droughts), this natural decline trend was subtracted from the reconstructed salinities in the subsequent analysis (Analytical Summary - Fig. 2.2). Once detrended in this way, positive values of DI-salinity indicated above-average salinities, whereas negative values indicated below-average salinity levels.

In order to quantify past drought behaviours and accurately assess future drought probabilities, we first had to establish our definition of a drought. Based on discussions with our user group, we defined past droughts as any event in which DI-salinity exceeded the value recorded for Moon Lake in 1988, the last regional drought event. Thus in our analysis, a drought was any event in which observed DI-salinity (s_i) exceeded the 1988 critical threshold (s_t). Each drought was considered to have a beginning time (b_i), and ending time (e_i) and a duration (d_i = e_i - b_i). These values were calculated for each drought event, allowing us to estimate the number of droughts (n) during the past 2000 years that were as or more severe than the 1988 drought (s_t). As well, our protocol allows us to estimate the average duration of such droughts, as

Probabilities of future droughts were estimated by describing the past drought behaviour using a Weibull model, then calculating the conditional probabilities of future droughts. To estimate such a probability, we first calculated the time between every two adjacent droughts (x_i = b_i+1 - e_i; i = 1, 2...n-1). These values referred to as drought inter-arrival times (x_i). When these inter-arrival times are small, droughts occur frequently, whereas if inter-arrival times are large, droughts are rare.

To describe past drought behaviour, we modelled inter-arrival times using the Weibull distribution, a standard reliability model whose probability density function is given by

...where 0>x, 0>alpha, 0>beta. In this formulation, x denotes inter-arrival time, alpha is the scale parameter and beta is the shape parameter. Recent statistical research suggests that for a high threshold salinity s_t, the beginnings and ends of droughts (b₁, e₁,..., b_n, e_n) can be described by a Poisson process, thus inter-arrival times (x_i) are approximately exponential random variables. Since the Weibull distribution contains the exponential distribution as a special case (when beta=1), the use of the Weibull distribution is justified. The maximum likelihood method was used to estimate the scale parameter, alpha, and the shape parameter, beta.

The probability of having a drought as or more severe than that of 1988 by the future year y was calculated using a conditional probability analysis. In this procedure, the condition is that there has been no drought during the length of time since 1988 (y₀ = 10 yr in 1999). Thus the probability of a future drought as or more severe than that of 1988 is

where X is a random variable which gives rise to the inter-arrival times x_i's, and y is greater than y₀. Using this model, the risk of droughts can be calculated for any year y in the future.

| top |

Results from Moon Lake, ND:

Conditional probability analysis of paleoclimate data from North Dakota suggested that the probability of a drought as or more severe than that of 1988 is only 10% by the year 2005 (Analytical Summary - Fig. 4.2). However, this probability rises to 26% by 2015 AD and 42% by 2030 AD. Over the past 2000 years, there have been 30 such droughts, with an average duration of 4.8 yr (range 1-30 yr) and an average inter-arrival time of 63.4 yr (Moon - Table 2.1). This analysis was also repeated to determine the probability of more and less severe droughts (Moon - Fig. 2.2), as well as the probability of multiple year droughts ((Moon - Fig. 2.3)). Because global warming is expected to increase the frequency and severity of droughts, we also tested the Moon Lake record to determine whether it contained evidence of global warming since 1900 AD. Here, we assumed that no paleoclimate data were available between 1900 and 1999. Instead, we modelled future drought probabilities (i.e., 1999 onwards) using only data collected prior to 1900 AD. These conditional probability analyses were then compared to forecasts based on the entire 2000 year record to determine whether inclusion of the past 100 years affected future drought forecasts. Our analyses demonstrate that drought occurrence has been greater since 1900 than before, and that including the past 100 years data increased the probability of future droughts by 2% by 2030 AD.

Conditional probability analyses were also used to model flood probabilities for the Red River drainage basin, where 1997 flood losses exceeded $850 million US dollars. Moon Lake lies in the Red River drainage and would be expected to be subject to dilution (and low salinity events) due to exceptional snow pack melt and spring rainfall, the main causes of flooding. Estimates of flood probability were based on the observation that the last major freshening event occurred in 1832, and that 165 years had elapsed by 1997. Weibull models were fit to the Moon Lake time series using the 1832 freshening as our definition of a critical flood and future probabilities were calculated using our conditional probability procedures. Results indicated that the probability of a flood event the magnitude of that in 1832 was 75% by 1997 (Moon - Fig. 3.2), suggesting that our risk assessment model is valuable for both extreme flood and drought analyses.

| top |

4. Other Results:

Analyses from Humboldt Lake demonstrated that 20th century climate has not been representative of the past ~1000 yrs in Saskatchewan (Humboldt - Fig. 2.2). Preliminary climatic reconstructions showed that major droughts were common prior to 1900 AD, with the most substantial drought occurring during the 1890s. In addition, regional climate appeared to be uniformly more arid before ca. 1200 AD, although better estimation of sediment ages are still required. Comparison of Moon (Moon - Fig. 2.2) and Humboldt lake records (Humboldt - Fig. 2.2) clearly demonstrates that past climates and drought probabilities will likely vary by geographic region. Further evidence for regional climatic variability comes from our statistical characterisation of drought (Humboldt - Table 2.1) and flood (Humboldt - Table 3.1) events for Humboldt and Moon Lakes. In central SK, droughts appear to occur more frequently and last longer than those in North Dakota. However, it must be stressed that the final ¹⁴C-AMS dating analyses are presently in progress and these results are likely to change.

Analyses of Humboldt Lake pigments have generated some novel insights into how algal productivity and community composition change under a changing climate (Humboldt - Figure 4.1). Comparisons of salinity and pigment concentration have revealed that shifts in climate do not appear to affect ecosystem productivity (total algal abundance), but the composition of the algal community does appear to shift to bloom-forming taxa (greens and blue-greens) under a drier conditions (Humboldt - Figure 4.2).http://collections.ic.gc.ca/humboldt/

Like Moon lake, the correspondence between Humboldt Lake salinity and drought indices does not appear to be well-defined (Humboldt - Figure 5.1). The relationships with crop yield and failure data that were observed in Moon Lake are currently on-going for Humboldt.

Anthropogenic influences appear to have been an important factor in shaping the Humboldt Lake aquatic ecosystem in recent years. During the course of our relatively recent occupation of the Humboldt Lake watershed (the present-day town of Humboldt was established in 1904, after a major influx of settlers in 1903 and again in 1906 - for more information see The spirit and the soil: Settlement of Humboldt), we have converted native prairie to cropland, diverted sewage to the lake, stocked fish and controlled flow regimes. The sedimentary record of diatoms records an increase in the abundance of taxa that are typically associated with "enriched" or eutrophic conditions in lakes around the 1940's and onwards (Humboldt - Figure 5.2). The associated factors need to be examined, but this change appears to be caused by a human impact on the lake ecosystem, over and above any climatic signals.

Construction of the Chauvin Lake, AB and the Nora Lake, MB salinity records are now also complete. Initial inspection of the Chauvin time series has shown that this lake has salinity fluctuations similar to those observed in Humboldt and Moon Lakes (Chauvin - Fig. 1.1). However, the Nora Lake salinity time series appear to quite distinct (Nora - Fig. 1.1). The drought record from this lake indicates that this area has not experienced as high a frequency or intensity of drought when compared to the other records. Nora Lake is the farthest east and is essentially out of the Great Plains Region. Analyses of past drought and flood statistics as well as future probabilities for these last two lakes are on-going.

| top |

5. Problems Encountered:

No significant problems have been encountered. Fossil diatom preservation had been unexpectedly poor in Manitoba, necessitating additional coring. Three suitable sites were identified in 1999 and the final core was obtained from one of these - Nora Lake. This has not delayed our progress as MB analyses were not scheduled to begin until 2000 AD. A small portion of the initial slides prepared for Nora Lake had low numbers of diatoms, resulting in the recent preparation of additional material.

Dr. Gemai Chen, project mathematician and co-PI, accepted a position with the University of Manitoba in summer 2000. Gemai has continued to devote time to the project from Winnipeg, but career changes have their way of being temporarily disruptive. Additionally, Dr. Peter Leavitt has been on sabbatical at the University of Wahington throughout 2000.

| top |

6. Partnerships and Collaborations:

¹ partners participating in our drought workshops as detailed in section 8 below.

| top |

7. Training of Research Personnel:

¹Dr. Olson has accepted employment as a private statistical consultant in Madison WI, USA. His duties have been assumed by Dr. Wunsam (PhD Austrian Academy of Sciences; diatoms) and Dr. Rusak (PhD York; climatic reconstructions);
²half-time positions;
³Presently M.Sc. Paleoclimatology (UBC);
⁴PhD thesis based on data collected here.

| top |

8. Accessibility of Results to Supporting Organizations:

Close user contact has been maintained throughout this project. Particularly frequent communication has been achieved with scientists and managers from the SK Crop Insurance Corp, Environment Canada, Agriculture Canada (PFRA), SK Water Corp, and SK Agriculture and Food, reflecting our initial efforts in Saskatchewan. All members of the user sector have also received news letters, electronic correspondence, and have access to our dedicated website at http://www.uregina.ca/drought. This site is updated regularly with our activities, progress, and latest research findings.

We have presented our drought periodicity analyses to the agricultural sector at the Agri-Food Innovation Fund Conference on Sustainable Agriculture in Regina, SK (Nov 2000), the Agriculture and Greenhouse Gas/Climate Change Workshop in Saskatoon, SK (Dec 2000), and Grain World 2000 (Feb. 27-29, 2000 in Winnipeg, MB) - a conference of national and international importance. We have presented our risk assessment analyses to the 1999 annual meeting of Crop Insurance Research Managers (14-15 June 1999; Guelph, ON), a conference which also included US Department of Agriculture and National Crop Insurance Service participants. Additional meetings were held with the SK Crop Insurance Corp research group (12 Dec 1998; Melfort, SK) to harmonize risk assessment methodologies. Similar meetings are planned with AB and MB crop insurance agencies as regional data become available.

As originally proposed, we have presented a one-day workshop, Sustainable Agriculture in Western Canada: Planning for Droughts using the Past, on 11 Sept 1998 at the University of Regina. Preliminary results of our drought prediction analysis from North Dakota were presented to over 20 managers from agencies in our user sector. Represented agencies included: Agriculture and Agri-Food Canada; Prairie Farm Rehabilitation Administration; Environment Canada (Atmospheric Environment Branch); Heritage and Parks Canada; Geological Survey of Canada (NRC); Manitoba Hydro; SK Crop Insurance Corp; SK Agriculture and Food; SK Environment and Resource Management; SK Grazing and Pasture Technology Program; SK Water Corp; SK Research Council; and Plains Research Centre. Members of the print (Western Producer) and television media (CTV-Regina) were present and reported results of the workshop to the public. This meeting was also used to further identify user needs with respect to drought forecasting, data requirements, and future research directions. A second workshop was held on the 22 October 1999 to report the results of our Saskatchewan analyses and preliminary Alberta findings. A third and final workshop will be held in February 2001 to present the remainder of our analyses and to synthesize our findings.

We have made significant efforts to broadcast this project to the greater management and scientific communities. For example, prediction analyses were presented in invited talks to the US Climate Change Impacts Group (Nov 2000), University of Washington (Feb 2000), Duke University (Mar 1999), North Dakota State (Jan 1999), University of Quebec-Montreal (Mar 1999) and University of Manitoba (Nov 1998). In addition, we have participated in invited climate change workshops held by the US National Oceanographic and Atmospheric Administration (June 1999) and BIOCAP Foundation Canada (Mar 1999).

During this past year, our professional activities included presentations to; Quaternary Research Center, University of Washington (Seattle, Feb 2000), "Monitoring for Ecosystem Health" Conference SERM/CPRC (Regina, April 2000), the American Society of Limnology and Oceanography annual meeting at Copenhagen, DK (June 2000), the Canadian Water Resources Association annual meeting at Saskatoon (June 2000), and the Ecological Society of America annual meeting at Snowbird, Utah (Aug 2000). In previous years, we have presented at meetings held by the Society of Canadian Limnologists (invited; Edmonton, Jan 1999), American Society of Limnology and Oceanography (Santa Fe, Feb 1999), Geological Society of America (two papers; Toronto, Oct 1998), and Societas Internationalis Limnologiae (Dublin, Ireland, Aug 1998).

Because of the importance of climate change to Western Canada, our research has been the subject of intense regional interest by television, radio and print media. These reports include: 1) Print articles in the Agri-Food Innovation Report (Feb 2000), Western Producer (Oct 1998), Regina Leader Post (21 Nov 1997; 24 Mar 1999), Regina Free Press (Nov 1997), SK Agriculture and Food newsletter Agri-View (in Western Producer; June 1998), Humboldt Journal (Aug 1998), and the University of Regina Third Degree (alumni magazine - Fall 1997); 2) Television reports on CKCK TV (CTV) Regina (11 Sept 1998; 24 Mar 1999), CBC TV Regina (15 Apr 1999), and Global TV Regina (15 Apr 1998; 23 Mar 1999), and; 3) Radio interviews on CBC Winnipeg (Nov 1997), CBC Regina (Nov 1997; 10 Sept 1998; 30 March 2000), and QR Radio Calgary (Nov 1997). Several publications are currently being prepared, and all students involved with the project will be submitting theses.

| top |

9. Potential Benefits:

Accurate estimates of drought likelihoods are essential for a stable economy and sound resource management in western Canada. Knowledge of drought probabilities will reduce the uncertainty in crop insurance calculations that lead to unstable, and therefore high, insurance premiums which often persist decades after droughts. In particular, re-insurance companies will benefit from improved catastrophe prediction. Improved drought prediction will benefit farmers by providing a sound rationale for selecting agricultural practices (crop, tillage, rotation) that protect against drought impacts. Because moisture availability and crop yield are highly correlated, improved drought prediction will also aid long-range planning by grain marketing and transportation companies, resulting in lower costs to producers and ultimately the general public. Similarly, uncertainty in drought occurrence has forced Manitoba Hydro to maintain a $400 million reserve to withstand revenue losses that would accrue during a multi-year drought. Improved drought predictions will improve the efficiency of lake and reservoir management, increase economic flexibility, and ultimately reduce costs to the public. Overall, the project will provide predictive environmental technologies that will directly benefit farmers and crop insurance corporations, drought relief agencies, agricultural planning agencies, and power generation sectors.