I INTRODUCTION AND GOALS OF THE PROJECT The HJ Andrews Experimental Forest is a major environmental research center with studies continually being conducted on topics which are dependent upon climatic variables for understanding. Reliable climate stations have been operating in the Forest for over 30 years. To date however, few comprehensive climate studies have been undertaken at the HJ Andrews. Precipitation regimes are generally understood but the Forest's complex temperature patterns up to this time have been almost completely unknown. This project seeks to produce the most accurate possible temperature maps of the HJ Andrews. The goal is to gain a detailed understanding of the spatial distributions of average monthly maximum and minimum temperatures. In order to maximize the project's scientific potential most fully, we concern ourselves specifically with the spatial representation of the temperature regimes in the absence of forest cover. II OVERVIEW OF THE HJ ANDREWS The HJ Andrews Experimental Forest is a 116-square kilometer area of forest on the west side of the central Oregon Cascades. It is located about 50 miles east of Eugene just north of the McKenzie River Valley. The site varies in elevation from about 410m at the southwest corner to over 1600m at the top of Lookout Mountain. HJ Andrews is one of many Long Term Ecological Research (LTER) sites around the country. Established in 1948, it is a major center for analysis of forest and stream ecosystems in the Pacific Northwest. Several dozen university and federal scientists use the site as a common meeting ground, working together to gain a basic understanding of ecosystems and how to apply ongoing discoveries in land management policy. The area is biologically diverse and encompasses many square kilometers of untouched old growth forest. Parts of the forest are regularly logged but restrictions on the amount of harvestable timber help to maintain the area's relatively pristine conditions necessary for scientific research. Its proximity to the Pacific Ocean, the Cascade Mountain Range, and its latitude all play a role in determining the climate of the HJ Andrews. The polar jet stream gives rise to a general westerly flow of air throughout the year in this area of the United States, especially during the winter. Forced topographic uplifting occurs when moisture-laden air from the Pacific Ocean 150 kilometers west of the HJ Andrews encounters the Cascade Mountain Range, resulting in relatively high precipitation amounts on the west side of the range (from 2000 to 3000 millimeters per year in the HJ Andrews). Being on the western slopes of the Cascade Range, the HJ Andrews is generally under a maritime influence, affected mainly by subtropical, Pacific, and Gulf of Alaska air masses. Thus, it tends to be warmer in the winter and cooler in the summer than areas east of the Cascade Range that are influenced by continental air masses from the Great Basin and the Arctic latitudes. Typically, an area of high pressure positions itself off of the western coast of the United States during summer months (bringing fair weather to Oregon). During winter months, low pressure systems commonly move through the area, bringing unsettled weather to Oregon. Hence, December and January tend to be the wettest months in the HJ Andrews, and summers are dry. Because of its varied topography, vegetation types, and elevation differences, the characteristics of the Forest's microclimates are complex. III HISTORY OF CLIMATE DATA INSTRUMENTATION AND COLLECTION IN THE HJ ANDREWS With the exception of the CS2MET climate station, for which records exist back to 1958, climate data collection in the HJ Andrews generally was begun in the early 1970s. Thus, an even 30-year period of record was used for the project. Figure 1 shows the complete data inventory for every site known to have provided useable temperature data during the past 30 years in the HJ Andrews with their respective periods of operation, and figure 2 depicts their locations in the Forest. Note that not all of the climate stations operated by the HJ Andrews have been located within or even near the Forest boundaries. RS13, RS13O, and RS14 are located 8 kilometers to the northeast of the Forest; RS19 and TSGRAS operated near the McKenzie River valley to the south; the sites where RS24 and TSQRTZ operated are far to the southwest of the Forest. Currently there are 28 functioning climate stations within the HJ Andrews, and three off-site. Six of the stations have been designated 'benchmark' sites; PRIMET, CS2MET, H15MET, VANMET, CENMET, and UPLMET. Most of these sites are equipped with a complete array of meteorological instrumentation, including thermister towers (covered sensors at 1.5, 2.5, 3.5, and 4.5 meters), anemometers, snow lysimeters, and humidity sensors. The remaining 25 sites have temperature sensors only (currently covered thermisters at roughly 1.5 meters above the ground). In the 1970s and 1980s, temperature data was recorded using thermographs and rotating charts. By the mid-1990s, all of the sites had been converted to covered thermisters and Campbell digital archiving units. Since then, the raw data has been routinely collected in the field by downloading it digitally and transferring it to a more permanent medium at Forest headquarters. The 'pre-digital' data has been digitized and archived in a fashion consistent with the more recent data. Note that the geographical distribution of climate stations in the Forest is not uniform; sites have historically been clustered in certain areas. This is mainly because many of the sites have operated as part of specific temporary research projects throughout the years, and the fact that sites naturally must be located where they are easily accessible in all seasons. V SELECTING USEABLE STATION DATA AND DATA QUALITY CONTROL The data as it was initially acquired for the project was processed and formatted according to HJ Andrews guidelines. Daily maximum, minimum, and mean temperature values were acquired. Each data point had been scanned for obvious and suspected errors and accordingly assigned a particular flag corresponding to the following matters: - temperature sensor buried in snow - temperature value estimated (by either known or unknown method) - lower or upper bounds of instrument sensitivity reached - temperature value not reproducible - temperature value questionable - temperature value missing During the compilation of datasets for this project, any temperature value which was flagged in any way was discarded, resulting in that daily value to be reported as 'missing'. In this way, only values having a high probability of being true were considered for analysis. For the benchmark sites with towers, the 1.5m value was used. If that value was missing or flagged, the 2.5m value was used; if that value was missing or flagged, the 3.5m value was used, and so on. After the temperature data was filtered, 'raw' monthly averages were computed for the periods of record. Several stations were discarded entirely at this point because of their short periods of record. Any site with less than three years (10% of the total period) of actual temperature data was not considered for analysis. The only exception to this rule was in the case of GR2V, GR4C, GR8C, GRT1, and GRVC, which are located in strategic areas and all had at least 2.6 years (8.8%) of data. Table 1 shows the resulting percentages of daily data for the 30-year period 1971-2000 after all flagged data was filtered out. The table also indicates which sites were immediately discarded. Note that in all cases, with the exception of RS18, the areas in which the discarded sites were located were already well represented by higher percentage sites, so that their removal likely did not affect the accuracy of our results. VI TEMPORAL CORRECTIONS TO DATASETS There were a wide range of actual periods of record among the various sites, from just under 3 years to over 28 years. Certainly the computed average monthly temperatures for a low percentage site that operated for, say, five years are not the same as they would be had that site operated for the full 30 years. How to correct the values, then? In order to determine the most efficient and accurate method for correcting the lower percentage sites, several tests were conducted. First, any site which had data for at least 75% of the 30-year period (22.5 years or more) were considered as 'complete' datasets, to which all other sites would be compared. As table 1 shows, there were 12 of these sites (PRIMET and CS2MET were not used in this analysis due to suspected data problems likely due to long-term instrument miscalibration). Next, correlation coefficients between every lower percentage site and these higher percentage sites were computed. The top five and lowest correlated sites out of this group are listed in table 2 along with the lower percentage site to be corrected. Note that all correlation coefficients are quite high, even the lowest ones. This is due to the fact that the HJ Andrews is a relatively small area in which to perform such correlation comparisons, and general trends are similar among all sites. The question became, "is it more accurate to correct the lower percentage site with the highest correlated site, the average of the top two correlated sites, the average of the top three correlated sites..." and so forth, down to the average of the top five correlated sites. A series of tests were conducted on the high percentage sites to answer this question. Consider maximum temperatures from RS02, which provided reliable data for 93.9% (28.2 years) of the 30-year period. This is as complete a real dataset that exists for the HJ Andrews. First we divided RS02's maximum temperature dataset into periods from 1 to 28 years long over the entire 30-year period, in effect simulating what RS02's dataset would be had it operated for every conceivable length of time during any point from 1971-2000. (For comparison's sake, we considered 1-year periods even though we would not be correcting any site with less than about a 3-year period). We then took each of those simulated datasets (there were 28 + 27 + 26 + ... + 4 + 3 + 2 = 405 of them) and 'corrected' them with the highest correlated site (in this case RS07), the average of the top two highest correlated sites (RS07 and RS89), the average of the top three highest, and so on, down to the average of the top five highest correlated sites. To calculate the correction factors we considered only periods of overlap between the sites. For example, for RS02's 3-year set from 1975-1977, we compared RS07's monthly maximum temperatures during that exact period to its long term averages, known to be reliable since RS07 is another long-term site. The differences between them were computed for each of those 36 months; this difference was then added to the monthly averages for RS02 during that period. RS02's 'new' set of 12 monthly averages was determined using this 'corrected' running set of 36 average maximum temperatures. Finally, the absolute value of the difference between these simulated temperatures and the actual long-term averages was calculated, and compared graphically (figure 3 for all of the 12 high percentage maximum temperatures). In calculating the averages of the top 2,3,4,and 5 highest correlated sites, the same procedure was followed, only the averages of the differences for each of the 36 months was figured, then added to RS02's 36 monthly values. One can see immediately from figure 3 the effect of trying to correct very short periods of record. Had RS02 operated for only one year, the correction factors applied to it would have been wildly unreliable, with a range from .25 degrees to near 4.5 degrees, no matter whether it was corrected with the single highest correlated site, the average of the top two, top three, top four, or top five. Figure 3 shows as well that after about an 8-year period, the method of correction makes little difference in the accuracy of the procedure. Most importantly, note that correcting with the single highest correlated site gives the best results, especially for periods of record shorter than about 8 years. It is for this reason that the single highest correlated site was used to correct each low percentage site for the project. The same procedure was followed for the minimum temperature correction comparisons and their results were similar (figure 4). Though the differences between correction calculations are small, using the highest correlated site was judged most appropriate. It is worth noting that there are also insignificant differences in the correction results between using higher correlated sites and the lowest correlated sites (figures 5 and 6). The differences are more marked than those among the highest correlated combinations, but note that the differences are still usually less than .5 degrees. A list of the low percentage sites that were corrected and the sites used to correct them is shown on table 3. VII CANOPY CORRECTIONS TO DATASETS The second correction made to the temperature datasets concerned the presence of forest canopies over the climate stations. It is a well known and obvious fact that solar insolation plays a major role in determining a site's maximum temperatures. Almost all of the sites in the HJ Andrews are in forests, and many of them in very deep forests which attenuate most of the sun's radiation even during summer. Thus, order to attain an idea of spatial distrubution of temperatures in the absence of forest, this correction was of utmost importance. First, fisheye photographs were taken at each of the 34 sites in the Forest used in the study (except for GR2V, which will be discussed later). These photographs were taken so as to simulate as closely as possible what the temperature sensor 'sees' of it's surroundings. Care was taken to minimize the problematic effects of the sun's glare. Analysis of the resulting hemispherical images gave a monthly summary of the solar attenuation taking place at each site due to its tree canopy and topographic shading (see table 4). Next, pairs of relatively open and closed sites were chosen for analysis to determine a mathematical function between solar attenuation and the difference in monthly maximum temperatures. In order for a pair to be a suitable candidate, both of its sites must be geographically close together, at similar elevations and aspects, and the differences in their monthly solar attenuation must be 50% or greater. Table 5 summarizes the pairs of sites deemed suitable for this analysis.