Background Information

What is a tree-ring reconstruction?

A tree-ring reconstruction is a time series of tree ring-width data that have been calibrated with an instrumental or gaged record of a hydrologic or climatic variable (e.g. annual streamflow, precipitation, snow water equivalent, drought index). The reconstruction, based on a statistical model that describes the relationship between tree growth and the gage record, extends that record back hundreds of years into the past (see figure below).

Tree-ring reconstruction of annual mean streamflow for the Salinas River at Paso Robles, CA, for the period 1409 to 2003. The green line shows annual values, and the black line is a 5-year running mean.

Tree growth is generally controlled by climate conditions during the year prior to and including the growing season. In California and the West, variations in tree growth often closely reflect the amount of soil moisture at the onset of the growing season, which is controlled by variations in precipitation, and, to some degree, temperature, humidity, and wind. Since measures of streamflow, seasonal snowpack, and drought integrate both precipitation and temperature over the course of the previous seasons (similar to tree rings), they are often closely correlated with tree growth.

Trees that provide the best information about hydroclimatic variability are those particularly sensitive to variations in moisture. These include species such as blue oak, ponderosa pine, Douglas-fir, and western juniper, usually growing at lower elevations in sparse stands on dry and rocky sites where soil moisture storage is minimal. Trees growing in these types of sites are also less likely to be subject to non-climatic disturbances (such as fires and insect infestation) and the effects of competition from nearby trees. In addition, the oldest trees of these species tend to be found on these sites. Giant sequoia are an exception to these tendencies, as they grow in relatively moist and dense forests in the high Sierra foothills, yet still record very useful information about hydroclimatic variability. (At higher elevations, approaching treeline, trees tend to be limited not by moisture availability but by the warmth and length of the growing season. So these trees can be used to reconstruct temperature variability.)

Trees used in hydroclimatic reconstructions are not necessarily located in the same watershed as the instumental or gage records. The atmospheric flows of moisture which influence both tree growth and streamflow are regional, crossing watershed divides, so trees in one basin may capture a significant portion of the variability in streamflow in another basin. For example, reconstructions of streamflow for the Sacramento River basin are improved when tree-ring chronologies from Oregon are added to the pool of model predictors.

Tree-ring reconstructions of hydroclimatic variables are developed from tree-ring chronologies. A tree-ring chronology is a time-series of annual values derived from the ring-width measurements of 10 or more trees of the same species at a single site.

To create a tree-ring chronology, cores from the sampled trees at each site are cross-dated (i.e., patterns of narrow and wide rings are matched from tree to tree) to account for missing or false rings, so that every annual ring is absolutely dated to the correct year. Then all rings are measured to the nearest thousandth of a millimeter using a computer-assisted measuring device. After growth-related trends unrelated to climate are statistically removed, the ring-width values from all sampled trees for each year are averaged to create a time series of annual ring-width indices. The complete series of ring-width indices from a site is called a tree-ring chronology.

Once a gaged record of interest is selected for reconstruction, a set of tree-ring chronologies from the region near the gage is calibrated with the gage record to form a reconstruction model. A statistical technique called multiple linear regression is commonly used. The reconstruction is evaluated by comparing the observed gage values with the reconstructed values by assessing the amount of variance in the gage record that is explained by the reconstruction. The reconstruction model is then validated by (1) testing it on a portion of the gage data that was withheld from the calibration process, or (2) testing the ability of the chronologies used in the model to estimate streamflow in different subsets of the data, or in an iterative process using bootstrapped datasets, such as a linear neural network. For a much more detailed description of the reconstruction process and of how reconstructions are evaluated, see the Blue River Case Study.

The Salinas River at Paso Robles reconstruction model (green) and the gaged record (blue) over their common time period (1940-2003).

Water managers have long used instrumental records of climate and gaged records of streamflow to assess the natural variability of the system they are managing, determining the frequency of drought events of varying severity and establishing a drought of record to use as the worst-case scenario in contingency planning. However, instrumental and gaged records are usually only 30 to 100 years long and are unlikely to capture the full expression of potential natural variability. Tree-ring reconstructions, by providing a much longer window into the past (300-1000+ years), more completely describe the potential natural variability of the system, including severe drought events. Many tree-ring reconstructions have indicated that droughts longer and more intense than those in the instrumental record have occurred in past centuries. This additional information on long-term hydroclimatic variability can guide water resource planning to better meet the challenges of potential future conditions. Water managers using tree-ring reconstructions will not be surprised by events, like the 1988-94 drought, that exceed the bounds of their prior operational experience of their system.

Here are several specific applications of tree-ring-based streamflow reconstructions by water providers:

• To test the effectiveness of drought response actions in reducing demand to match supply in severe drought years
• For scenarios, using sequences of flow from reconstructions, to evaluate the results of management decisions
• To generate flow sequences that will be used to evaluate alternatives for coordinated reservoir operations under drought conditions
• As input into a system model to determine whether a given level of demand could be met in all years of the reconstruction, including various sequences of drought years
• As qualitative guidance, for considering the flow variability in the 20th century gaged record in a much longer context

In the reconstructions

Because the reconstruction models explain most (typically between 60-80%), but not all, of the variance in the gage record, there are uncertainties in the reconstructions. Estimates of uncertainty can be described by confidence intervals around the reconstruction. These confidence intervals describe the range of uncertainty (usually at a 95% level) that can be expected in the estimates. Narrow confidence intervals represent a stable reconstruction model. There are several way to estimate confidence intervals. Two of these are the use of bootstrapped series generated in the iterative model-fitting process of the linear neural network, and the use of the root mean squared error (RMSE) from the regression equation.

Other sources of uncertainty include the inevitable decline in tree sample size back in time, and extreme high and low values in the reconstruction which are extrapolated (i.e., the tree-ring values are outside the range in the calibration period) rather than interpolated.

In the gaged records

Gaged and instrumental records are generally presumed to represent physical reality, and the tree-ring reconstructions on which they are based thus offer a close approximation of that reality. But it is worth noting that gaged and instrumental records do have some measurement error associated with them, and, more critically, may contain spurious trends and features caused by human manipulation of the watershed.

In California and across the West, most streams have seen their natural flow regimes altered through water diversions and impoundments, as well as land-use changes in the watershed. As a result, the gaged streamflow record may not accurately represent the natural long-term trends and year-to-year variability in water supply in that watershed. Water management agencies have attempted to account for these alterations, using records of diversions and changes in watershed storage to correct the "raw" streamflow records. The resulting records are called "undepleted", "naturalized", "unimpaired", or "virgin" flow records to distinguish them from uncorrected gage records. Because they more accurately represent natural water supply variation than the raw gaged records, undepleted flow records, when they exist, are used as the basis for tree-ring streamflow reconstructions. The example below, from the Fraser River in Colorado, demonstrates the typical discrepancies between the gaged (uncorrected) and undepleted (corrected) flow records in a highly manipulated watershed.

The gaged USGS streamflow record (light blue) for the Fraser River near Winter Park, Colorado, is compared with the undepleted flow record (dark blue) derived by Denver Water from the gaged record to account for diversions, which began in earnest in 1936. The gaged record poorly represents the trends in streamflow on short and long time scales.

Additional caveats for using reconstructions

• It is tempting to combine the reconstructed and gage records into a single record to which new gaged values (for 2003, 2004, etc.) could be easily appended. The problem with doing this is that the statistical properties of the reconstructed and gage streamflow series are not the same. Tree-ring chronologies do not explain all the variance in the gage record. Because of this, the estimates of flow produced by the reconstruction model do not represent the full range of values in the gage record, although dry extremes are typically more closely replicated than wet extremes. Consequently, the variance, standard deviation, range, and max/min values are different. Any attempt to blend the two records needs to acknowledge these important differences and the limitations they impose.
• Although tree-ring reconstructions can be used to guide expectations for the future, they cannot be used to predict future droughts. Furthermore, the variability in past climate may not be an analog to future variability, given the human impacts on climate in the last century. Future conditions will be influenced by climate forcings, including changes in land use and elevated inputs of greenhouse gases, as well as by natural climate variability. Future climate will reflect natural climate variabilty, including the long-term range of variability described in the tree-ring reconstructions, but superimposed over this natural variability will be the anthropogenic impacts on climate.

http://lwf.ncdc.noaa.gov/paleo/streamflow/ca/background.html Downloaded Saturday, 30-Aug-2008 02:39:41 EDT Last Updated Friday, 01-Dec-2006 13:36:02 EST by paleo@noaa.gov Please see the Paleoclimatology Contact Page or the NCDC Contact Page if you have questions or comments.