You are on page 1of 9

Agricultural and Forest Meteorology 128 (2005) 5765 www.elsevier.

com/locate/agrformet

Calculating daily mean air temperatures by different methods: implications from a non-linear algorithm$
Albert Weiss*, Cynthia J. Hays
School of Natural Resources, University of Nebraska, Lincoln, NE 68583-0728, USA Received 6 February 2004; accepted 30 August 2004

Abstract Daily mean air temperature is used as an independent variable in algorithms describing many biological applications. These algorithms are usually of a non-linear nature. The questions addressed in this study were: is there a difference in the daily mean air temperature calculated by different methods and what is the impact of the various calculation methods on a non-linear algorithm? The empirical coefcient in the non-linear algorithm used in this study was determined from a daily mean air temperature based on the mean of 24 hourly mean temperature values. Daily mean air temperature was calculated by ve methods: mean hourly (Hourly); three equally spaced hourly mean observations weighted with the last observation (Weighted); 3 h mean temperatures (Mean 3 hour); the algorithm used in the CERES family of crop simulation models (CERES), and the mean of the maximum and minimum daily temperatures (Max/Min). It was assumed that the Hourly method best represented the daily mean air temperature and the other methods were compared to it. Two forms of air temperature were used in a non-linear algorithm; a sequential approach where the algorithm was run as many times as the number of individual temperature values used in each method, the results then averaged; and a single daily mean air temperature value. This non-linear algorithm was evaluated over a wide range of locations, ranging in elevation from 2.4 to 1252 m and annual precipitation from 108 to 1820 mm. There was little difference in daily mean air temperatures between the different methods. However, there were large differences in responses from the non-linear algorithm when using any sequential approach when compared to the single daily mean temperature values. The Mean 3 hour method worked well in all locations. The CERES method worked well except for two locations characterized by high mean annual maximum and minimum temperatures. These results do not mean that the sequential approaches are inappropriate, just that the temperature method used to determine empirical coefcients in the nonlinear algorithm must be consistently used in all applications. These results are a guide to different methods used to calculate daily mean air temperature and the range of possible results when used in a non-linear algorithm. Although a specic example was used in this study, the results are relevant to any non-linear algorithm containing empirically determined coefcients. # 2004 Elsevier B.V. All rights reserved.
Keywords: Daily mean air temperature; Non-linear algorithm

$ A contribution of the University of Nebraska Agricultural Research Division, Lincoln, NE. Journal Series No. 14458. This research was supported in part by funds provided through the Hatch Act. * Corresponding author. Tel.: +1 402 472 6761; fax: +1 402 472 6614. E-mail address: aweiss1@unl.edu (A. Weiss).

0168-1923/$ see front matter # 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.agrformet.2004.08.008

58

A. Weiss, C.J. Hays / Agricultural and Forest Meteorology 128 (2005) 5765

1. Introduction Daily mean air temperature is used as a driving variable in many simulations of physical or biological processes and is calculated by various methods. Hartzell (1919) used a planimeter to determine the hourly mean temperature from thermograph charts, summed the hourly means and divided by 24 to obtain the daily mean air temperature for data collected at Fredonia, NY for the year 1916. Regarding the best method to calculate daily mean air temperature, Hartzell (1919) concluded that if daily mean air temperature was used for . . . temperature coefcients, for the study of thermal inuence in botany and zoology, thermograph averages alone should be used . . .. He continued to state that daily mean air temperature calculated as the mean of the maximum and minimum temperatures was not suitable for these purposes, because the daily mean air temperatures calculated by the two methods did not always agree. Even though measurement and temperature sensor technologies have changed dramatically since this study, the question of the best method to calculate daily mean air temperature is still very relevant with respect to the role of temperature as a driving variable in the simulation of physical or biological processes. At least two other methods have been used to calculate the daily mean air temperature (aside from the mean of the 24 hourly mean values and the mean of the maximum and minimum temperatures). A weighted mean of hourly mean temperature at threexed observation times (e.g., 07:00, 14:00, 21:00 local standard time) is determined by the sum of these temperature values, where the last temperature observation is weighted twice, the sum then divided by four, Landsberg (1958). The other method employs the mean of equally spaced observations, e.g., the mean of eight values, which represents the mean temperature for a 3-h period, can also be used to determine the daily mean air temperature. A form of this method is used in the CERES family of crop simulation models, Jones and Kiniry (1986). Harris and Pedersen (1995) compared daily mean air temperature calculated by summing individual 20-min temperature observations and then dividing by 72 to the mean of the maximum and minimum temperatures at Calgary, Canada over a 24-month period. They found the largest discrepancies (up to

7 8C) between these two methods during the winter months. Cesaraccio et al. (2001) and Raworth (1994) used different algorithms to determine thermal time (growing degree days) on a daily basis from values of maximum and minimum temperatures. Raworth (1994) found that observed hourly temperature data provided a better estimate of thermal time than the mathematical functions t to minimum and maximum temperature data. The algorithm developed by Cesaraccio et al. (2001) was found to be superior over a wide range of California climates when compared to other published thermal time algorithms. Reicosky et al. (1989) evaluated ve algorithms used in simulation models that generate hourly air temperature values from daily maximum and minimum temperature observations. They noted that although there were sometimes large differences between observed and simulated hourly temperatures, the ramications of these differences in simulated processes were not determined by Reicosky et al. (1989) as this was not their main objective. Sadler and Schroll (1997) had a similar objective to Reicosky et al. (1989), to evaluate the same algorithms used in this earlier study plus a newly developed algorithm. This newly developed algorithm was not restricted by the assumption that the shape of the daily pattern ts a predened curve, such as a sine wave. This algorithm performed as well as or better than the ve algorithms in about 50% of the cases but required the development of a cumulative distribution function of normalized temperature for a year at each location. An important extension to these previous efforts is to evaluate the use of observed air temperatures at different time scales (e.g., maximum and minimum, hourly mean, the mean of eight 3 h means, three individual hourly mean observations, and a daily value) in a non-linear process. Many simulated plant related processes are of a non-linear nature, e.g., photosynthesis and respiration. While different results from using different methods to calculate daily mean air temperature might be small on a daily basis, in a non-linear process, these small differences can be magnied if they are accumulated over any length of time. An analogous situation occurred when McMaster and Wilhelm (1997) used two accepted protocols to calculate accumulated thermal time with the same data set. Large monthly differences in accumulated thermal time (in some cases exceeding 80%)

A. Weiss, C.J. Hays / Agricultural and Forest Meteorology 128 (2005) 5765

59

were observed between the two methods. Differences in accumulated thermal time over the growing season varied starting in September with no difference, a 23% difference in March, and with a 9% difference in July at physiological maturity. In developing new applications, where daily mean air temperature is a driving variable, it is important to clearly dene how this variable was calculated in order to avoid any ambiguities in the calculation, which may impact the results. When the application involves a nonlinear algorithm, how is a single daily mean output value determined with multiple input values of a parameter (e.g., maximum and minimum temperatures)? Are multiple input values used in the algorithm and the resulting outputs used to calculate a daily mean output value or is a daily mean input value used to calculate a daily mean output value? The objectives of this effort were two-fold: (1) to evaluate different methods of calculating daily mean air temperature from observed values at different time scales and (2) to investigate the responses of a non-linear algorithm using these different methods of calculating daily mean air temperature and using multiple input values. A consequence of the second objective will provide an indication of the range of possible results when using non-linear algorithms over contrasting climates in the United States.

Fig. 1. Locations of the sites used in this study.

2. Material and methods Hourly mean (local standard time) and daily maximum and minimum air temperatures from seven

locations in the contiguous U.S. (Astoria, OR; Bishop, CA; Brorson, MT; Caribou, ME; Del Rio, TX; Mead, NE; and Tampa, FL) representing a wide range of climate conditions (Fig. 1 and Table 1), were used to calculate daily mean air temperatures. The hourly mean data obtained from the National Climatic Data Center (NCDC) were calculated based on a hourly period that ended during the last 10 min of the hour, this ending time varied from station to station. For this study, the time of observation was noted as the following hour (e.g., the hourly mean temperature was calculated from measurements taken from 9:51 to 10:50 and this time of observation was changed from 10:50 to 11:00). The High Plains Regional Climate Center (HPRCC) hourly mean data was calculated based on a hourly period that began 1 min after the hour (e.g., the hourly mean temperature was calculated from measurements taken from 10:01 to 11:00 and the time of observation was noted as 11:00). The time of observation of the data from NCDC was changed in order to be consistent with

Table 1 Location, length of record, latitude, longitude, elevation, mean annual daily minimum and maximum temperatures for the length of record and mean annual total precipitation Location and length of record Astoria, OR (19972002)a Bishop, CA (19972002)a Brorson, MT (19952002)b Caribou, ME (19972002)a Del Rio, TX (19972002)a Mead, NE (19932002)b Tampa, FL (19972002)a Latitude Longitude Elevation (m) 2.7 1252.4 691.0 191.1 312.7 366.0 2.4 Mean annual daily minimum temperature (8C) 7.1 3.0 0.5 0.6 15.5 3.8 18.5 Mean annual daily maximum temperature (8C) 14.8 23.6 12.6 9.9 28.0 16.6 27.8 Mean annual total precipitation (mm) 1820.4 108.3 373.9 868.0 403.4 665.5 1219.9

468090 N 378220 N 478470 N 468520 N 298220 N 418090 N 278580 N

1238530 W 1188220 W 1048150 W 688020 W 1008550 W 968290 W 828320 W

Superscripts denote source of data. a National Climatic Data Center (NCDC) http://www.ncdc.noaa.gov/oa/ncdc.html. b High Plains Regional Climate Center (HPRCC) http://www.hprcc.unl.edu/.

60

A. Weiss, C.J. Hays / Agricultural and Forest Meteorology 128 (2005) 5765

the time of observation of the data from HPRCC. The mean annual daily minimum temperatures ranged from 0.6 to 18.5 8C, while the mean annual daily maximum temperatures ranged from 9.9 to 28.0 8C. The location with the largest difference between mean annual daily maximum and minimum temperatures was Bishop, CA (20.6 8C), the smallest Astoria, OR (7.7 8C). These locations also had the precipitation extremes, 108.3 and 1820.4 mm, respectively. The ve methods used to calculate daily mean air were: temperature T  Hourly: P24 i1 Ti T (1) 24 where i is the hour, Ti is the hourly mean air temperature for hour i (8C).  Weighted: T07:00 T14:00 2T21:00 (2) T 4 where T07:00 is the hourly mean air temperature for 07:00 local time (8C), T14:00 is the hourly mean air temperature for 14:00 local time (8C), T21:00 is the hourly mean air temperature for 21:00 local time (8C).  Mean 3 hour: P8 T3i (3) T i 1 8 where i is every 3 h (03:00, 06:00, . . ., 21:00, 24:00 local time), T3 is the 3 h mean temperature for hour i (8C).  CERES (Jones and Kiniry, 1986) P8 Tci T i 1 T (4) 8 where i is 18, Tci = Tmin + tmfaci(Tmax Tmin) (8C), Tmin is the daily minimum temperature (8C), Tmax is the daily maximum temperature (8C), tmfaci = 0.931 + 0.114i 0.0703i2 + 0.0053i3; for i = 18.  Max/Min: Tmin Tmax (5) T 2 where Tmin is the daily minimum temperature (8C) and Tmax is the daily maximum temperature (8C).

The NCDC data were missing some air temperature observations; the missing data could range from an hour to several days. No attempt was made to estimate these missing data and a daily mean air temperature was not calculated for any days that had missing data. Mean bias errors (MBE) and root mean square errors (RMSE) were calculated with respect to the daily mean air temperature based on the Hourly method using the following equation: P m T T (6) MBE n (P )0:5 2 m T T RMSE (7) n is the daily mean air temperature calculated where T m is the daily mean air from the Hourly method (8C), T temperature calculated from the other methods (C8), and n is the number of observations. Any days with missing air temperatures were omitted from the MBE and RMSE calculation. The RMSE was calculated monthly and for the length of record. While many non-linear algorithms of plant processes could be used, the non-linear algorithm (Streck et al., 2003) describing winter wheat development rate (Rdev, d1) was selected. This algorithm was selected because it has both positive and negative slopes providing a very demanding test for the objectives of this research effort, whereas other non-linear algorithms may have either have positive or negative slopes. It is obvious that winter wheat is not grown at all the locations in this study. However, the objective of this study was to evaluate a temperature driven non-linear algorithm, not the feasibility of growing winter wheat in different geographic locations. The use of the same algorithm allows for the evaluation of the different non-linear responses from the different methods of calculating daily mean air temperature. The following relationships describe the non-linear algorithm for winter wheat phenological development: f T 2T Tn a Topt Tn a T Tn 2a Topt Tn 2a if T < Tn or T > Tx ; (8) (9)

if Tn  T  Tx f T 0;

A. Weiss, C.J. Hays / Agricultural and Forest Meteorology 128 (2005) 5765

61

ln 2 ln Tx Tn =Topt Tn

(10) (11)

Rdev Rmax f T

where T is the temperature (8C), Tn is the minimum cardinal temperature (8 8C for July and August and 0 8C for all other months), Tx is the maximum cardinal temperature (35 8C for July and August and 30 8C for all other months), Topt is the optimum cardinal temperature (24 8C for July and August and 19 8C for all other months), Rmax is the maximum development rate (0.0381 d1 for July and August and 0.0241 d1 for all other months). Rmax was empirically determined based on the comparison of eld observations of plant development relatively near Mead, NE for a wheat cultivar and a daily mean air temperature calculated from 24 hourly mean air temperatures (Streck et al., 2003). The value of Rmax will change if determined by the different temperature methods or if a different cultivar was selected. The change in Rmax associated with the different single value temperature methods would probably be small. However, the values of Rmax associated with the different sequential methods will probably be quite different then those of the single value temperature methods. For comparison purposes in this study, the original value found in Streck et al. (2003) was used. The development rate was calculated for each day and accumulated for three different periods representing early season plant development, mid-season plant development and fall sowing (starting on April 1, July 1 and September 1, respectively). The development stage equals the accumulated development rate and has a value of 0 at emergence, 1 at anthesis, and 2 at physiological maturity. The rst approach to the calculation of the daily development rate (Eqs. (8)(11)) used the single value of the daily mean temperature determined from Eqs. (1)(5). The second approach took into account the non-linear nature of the daily development rate by using the individual temperatures from Eqs. (1)(5) in a sequential process in Eqs. (8)(11) (which will be referred to as sequential) and then averaged the results of Eq. (11). For example, using the sequential Hourly method, Eqs. (8)(11) were ran 24 times and the daily development rate was the average of Eq. (11). When the air temperature data were missing, the

development rate was estimated as the mean of the development rate 5 days prior and 5 days after the missing data. The reason for using 5 days in comparison to 1 days were to smooth out any abrupt temperature changes that might have occurred. The development stage based on the Hourly method was compared to the development stages calculated using the other approaches. The day of year when the development stage based on the Hourly method equalled 0.75 (DOY24) was compared to the day of year when the other approaches reached 0.75 (DOY). The development stage of 0.75 was selected instead of 1.0, since for some locations a development stage of 1.0 was not achieved. This comparison was made by subtracting DOY24 from DOY (i.e., a positive value means that the time for the development stage to reach 0.75 was longer than the time using the Hourly method). The mean of the differences for each period (early or mid-season plant development and fall sowing) were calculated.

3. Results In general, there was very good agreement (in terms of annual RMSE values) between the Hourly method and the other methods for the calculation of daily mean air temperature for the seven locations (Table 2). Agreement between the methods was within 1 8C for all locations except Bishop, CA. The MBEs varied from 0.52 8C for the Max/Min method for Bishop, CA to +0.52 8C using the CERES method at Tampa, FL. The location with the best agreement between the methods was Astoria, OR. As would be anticipated, there was no difference between the Hourly and Mean 3 hour methods, since the 3 h mean value is the mean of three of the mean hourly values. The other three
Table 2 MBE/RMSE values of daily mean temperature calculated using the Max/Min, Weighted, and CERES methods Location Astoria, OR Bishop, CA Brorson, MT Caribou, ME Del Rio, TX Mead, NE Tampa, FL Max/Min (8C) 0.06/0.63 0.52/1.31 0.13/0.94 0.19/0.88 0.26/0.76 0.01/0.92 0.43/0.72 Weighted (8C) 0.04/0.57 0.28/1.10 0.12/0.94 0.14/0.88 0.25/0.74 0.03/0.92 0.09/0.54 CERES (8C) 0.13/0.64 0.35/1.22 0.26/0.97 0.08/0.85 0.38/0.80 0.14/0.92 0.52/0.78

62

A. Weiss, C.J. Hays / Agricultural and Forest Meteorology 128 (2005) 5765

methods provided very similar results; no method was clearly superior to another at all locations. Only the monthly temperature RMSE values for Astoria, OR and Bishop, CA will be discussed in detail since they represent the best and worst agreement between the different methods. Astoria, OR usually has a small daily variation in temperature, while Bishop, CA has a relatively large daily variation in temperature. The other locations have daily temperature variations that are between these two locations. For Astoria, OR, monthly RMSE values for the Max/ Min and CERES methods were similar while the Weighted method showed a seasonal trend with the lowest RMSE values in JuneJuly (Table 3a). In general, for the Max/Min and Weighted method for the fall and winter months the MBE was negative (ranging from 0.21 to +0.02 8C) while for the spring and summer months the MBE was positive (ranging from 0.02 to 0.23 8C). Using the CERES method all MBE values were positive with the largest values in June

Table 3 Monthly MBE/RMSE values of daily mean temperature calculated using the Max/Min, Weighted, and CERES methods Month (a) Astoria, OR January February March April May June July August September October November December (b) Bishop, CA January February March April May June July August September October November December Max/Min (8C) 0.04/0.63 0.03/0.60 0.02/0.54 0.05/0.63 0.09/0.56 0.13/0.51 0.23/0.62 0.20/0.61 0.24/0.82 0.04/0.69 0.04/0.72 0.06/0.61 0.42/1.07 0.02/0.84 0.57/1.06 0.86/1.23 1.24/1.59 1.35/1.72 1.54/1.88 1.36/1.61 0.89/1.21 0.12/0.87 0.17/0.97 0.62/1.09 Weighted (8C) 0.02/0.67 0.12/0.66 0.11/0.62 0.02/0.49 0.06/0.51 0.05/0.43 0.02/0.41 0.02/0.43 0.18/0.60 0.21/0.65 0.07/0.67 0.02/0.57 0.28/1.02 0.20/0.87 0.06/0.92 0.58/1.11 0.78/1.22 0.91/1.25 1.06/1.28 0.88/1.21 0.55/1.06 0.19/1.00 0.41/1.15 0.36/0.94 CERES (8C) 0.02/0.63 0.04/0.61 0.06/0.54 0.13/0.64 0.16/0.57 0.21/0.52 0.30/0.64 0.29/0.64 0.34/0.86 0.05/0.70 0.03/0.73 0.00/0.61 0.60/1.16 0.18/0.87 0.37/0.96 0.69/1.07 1.02/1.41 1.13/1.53 1.13/1.68 1.12/1.40 0.67/1.05 0.10/0.86 0.37/1.03 0.80/1.22

through September. For the majority of months in Bishop, CA, the Max/Min method had the largest RMSE and absolute MBE values and the Weighted method the lowest, Table 3b. In contrast to the trend with the Weighted method at Astoria, OR, all methods at Bishop, CA had the highest RMSE and absolute MBE values during the period MaySeptember. There were few differences between calculating daily mean air temperature by the different methods. However, there were large differences in the number of simulated days when the development stages reached 0.75 for the three different periods beginning on April 1, July 1, and September 1; (Table 4ac). As would be expected from the previous analysis (Table 2), in general Bishop, CA and Astoria, OR represent the extremes in differences of simulated days of development stage for these three periods. Since Astoria, OR; had the lowest RMSE and absolute MBE values (Table 2) all methods worked equally well for the rst two periods (Table 4a and b). For the last period, there was a difference between the single value and sequential methods (Table 4c). Mean differences for Bishop, CA ranged from 0.7 to 1.8 days for the single value methods and from 4.8 to 39.8 days for the sequential methods during the April 1 early season plant development (Table 4a). During the July 1 midseason plant development, the mean difference in days ranged from 0.5 to 1.2 days for the single value methods and from 5.8 to never achieving 0.75 using the sequential methods (Table 4b). Mean differences ranged from 0.5 to 1.8 days using the single value methods and 11.5 to 52.5 days using the sequential value methods for the September 1 sowing (Table 4c). The sequential Max/Min method had the poorest agreement among the other methods at all locations, since Eq. (11) was sometimes evaluated as 0. The best temperature method was the Mean 3 hour method with mean differences ranged from 0.8 to 0.0 days for the three different periods (Table 4ac). The Weighted and CERES methods usually performed satisfactorily, with the magnitude of the differences depending on the period and location.

4. Discussion Reicosky et al. (1989) investigated ve different subroutines to simulate hourly mean air temperatures:

A. Weiss, C.J. Hays / Agricultural and Forest Meteorology 128 (2005) 5765

63

Table 4 Mean differences in days between the sequential temperature methods, the single daily mean temperature methods and the Hourly method in the non-linear algorithm (Eqs. (8)(11)) Method (a)a Sequential Hourly Sequential Max/Min Sequential Weighted Sequential Mean 3 hour Sequential CERES Max/Min Weighted Mean 3 hour CERES (b)b Sequential Hourly Sequential Max/Min Sequential Weighted Sequential Mean 3 hour Sequential CERES Max/Min Weighted Mean 3 hour CERES (c)c Sequential Hourly Sequential Max/Min Sequential Weighted Sequential Mean 3 hour Sequential CERES Max/Min Weighted Mean 3 hour CERES
a b

Astoria, OR (days) 1.0 2.2 0.5 0.8 0.7 0.3 0.2 0.0 0.5 0.7 1.2 0.7 0.7 0.0 1.0 0.0 0.0 1.2 1.8 5.5 2.5 1.7 2.2 0.0 0.7 0.0 0.5

Bishop, CA (days) 9.3 39.8 4.8 9.0 12.5 1.8 0.7 0.0 1.2 11.7 NV0 5.8 11.0 13.7 0.5 1.2 0.0 0.5 16.7 52.2 11.5 15.5 18.8 0.7 1.8 0.0 0.5

Brorson, MT (days) 2.6 7.1 1.0 2.2 2.2 0.4 0.4 0.0 0.6 3.9 12.4 2.1 3.6 4.4 0.0 0.0 0.0 0.0 12.9 29.97 11.8 12.1 10.0 1.1 0.8 0.0 2.1

Caribou, ME (days) 1.5 4.8 0.2 1.7 2.0 0.7 1.2 0.0 0.3 2.2 6.0 1.3 2.0 2.7 0.2 0.2 0.0 0.3 9.2 8.45 7.3 9.2 6.8 3.8 0.2 0.0 0.7

Del Rio, TX (days) 4.0 12.5 10.7 3.7 8.0 4.2 8.2 0.0 5.0 1.8 4.8 5.7 1.7 4.0 1.0 2.0 0.0 2.0 2.0 7.0 3.2 2.0 4.8 2.0 0.8 0.0 2.0

Mead, NE (days) 3.7 10.9 3.0 3.7 4.7 0.3 0.1 0.0 0.1 2.5 7.8 2.0 2.3 3.2 0.2 0.1 0.0 0.1 6.3 20.1 6.4 6.0 7.6 0.2 0.5 0.0 0.1

Tampa, FL (days) 3.8 14.5 3.2 3.7 7.7 1.5 0.3 0.0 1.7 1.5 5.0 1.5 1.5 2.7 0.8 0.2 0.0 0.8 3.3 10.8 1.3 3.3 8.8 5.0 0.7 0.8 5.5

The simulations started on April 1 and ended when the development stage reached 0.75. The simulations started on July 1 and ended when the development stage reached 0.75 or August 31. In this latter case, the superscript represents the number of years that the development stage reached 0.75 out of the total years of record. NV denotes no value, in this case there were no years that the development stage reached 0.75. c The simulations started on September 1 and ended when the development stage reached 0.75 or December 31. In this latter case, the superscript represents the number of years that the development stage reached 0.75 out of the total years of record.

(1) WAVE in ROOTSIMU V4.0 (Hoogenboom and Huck, 1986), (2) WCALC from the crop simulation model SoyGRO V5.3 (Wilkerson et al., 1983), (3) WEATHER in GLYCIM (Acock et al., 1983), (4) TEMP (Parton and Logan, 1981), and (5) SAWTOOTH (Sanders, 1975). Reicosky et al. (1989) found that WCALC provided the best results. The hourly mean temperatures estimated by WCALC were obtained by dividing a day into three segments; midnight to sunrise +2 h, sunset to midnight, and the daylight hours. WCALC also requires maximum and

minimum temperature for the day before, the minimum temperature for the following day, the times of sunrise and sunset of the day before, and the time of sunrise of the following day. As part of this current study, the daily mean air temperature from WCALC was compared to the Hourly and CERES methods for calculating daily mean air temperature at Mead, NE. This analysis was preformed to document why the CERES method was used in this study instead of WCALC. The Mead, NE location was selected because it had a long period of

64

A. Weiss, C.J. Hays / Agricultural and Forest Meteorology 128 (2005) 5765

record (10 years) and a data set with no missing values as compared to the other locations. In this comparison, it was found that the CERES method was superior to WCALC (RMSE values of 0.92 and 1.17 8C, respectively; MBE values of 0.14 and 0.66 8C for the period of record) and did not require data from previous and future days. This study demonstrated that daily mean air temperature could be relatively accurately calculated by Eqs. (2)(5) over a wide range of climates and usually incorporated into a non-linear algorithm. However, when sequential methods were used in the non-linear algorithm there were large differences when compared to the single value methods. These differences were due to the empirically dened coefcient (Rmax), which was developed based on the Hourly method. This result indicates the importance of the consistent use of the same temperature method as associated with the non-linear algorithm. The consistent use of the same non-linear algorithm, representing phenological development in winter wheat, across different climates was necessary to ensure that the different responses were due solely to the different temperature methods employed in the non-linear model even though winter wheat is not grown in all locations. In general, the best predictions were made for Astoria, OR and the worst for Bishop, CA, regardless of the temperature method employed. No doubt these results are due to the uniformity, or lack of uniformity, of the temperature regimes at these locations. Astoria, OR had the smallest difference between mean annual daily maximum and minimum temperatures, while Bishop, CA had the greatest difference. Brorson, MT; Caribou, ME; and Mead, NE have similar RMSE values (Table 2) for the different methods of calculating daily mean air temperatures as compared to Del Rio, TX and Tampa, FL. These ve locations can also be characterized by the mean annual daily minimum and maximum temperatures, the former locations having lower values than the latter locations. The Mean 3 hour method was the best across all locations in the non-linear algorithm. Given that these eight values are a composite of the 24 hourly values, this result is not surprising. In general, the sequential methods did not agree as well as the single value methods in this non-linear algorithm, especially at

locations in which there was a large range in daily temperatures (i.e., Bishop, CA). The Weighted and CERES methods compared equally well over all the locations. The sequential Max/Min method performed poorly. An advantage of the CERES method was that it only required daily maximum and minimum temperatures compared to the Mean 3 hour and Weighted methods, which requires hourly mean temperature observations. A disadvantage of the CERES method is that it does not work as well for locations with high mean annual daily maximum and minimum air temperatures, such as Del Rio, TX and Tampa, FL. This limitation may be overcome by modifying the tmfac algorithm (Eq. (4)) to handle such situations.

5. Conclusion In this study, the single value temperature methods generally performed well in the non-linear algorithm. The reason for this result is that the empirical coefcient used in the non-linear algorithm was developed using a single value temperature. These results do not mean that the sequential methods are inappropriate. They just mean that the temperature method used to determine the empirical coefcients must be consistently used in all applications and these methods should be thoroughly described in any publication. One could argue that the sequential methods make more sense in a non-linear algorithm, but the application of any of these sequential methods must be balanced by the necessary accuracy of the results and the availability of the input data. These results are a guide to different methods used to calculate daily mean air temperature and the range of possible results when used in a non-linear algorithm. Although a specic example was used in this study, the results apply to any non-linear algorithm containing empirically determined coefcients.

Acknowledgements Drs. G.S. McMaster and W.W. Wilhelm provided valuable comments on an earlier version of this manuscript. The responses to the questions raised by the two anonymous reviewers further helped to clarify the contents of this manuscript.

A. Weiss, C.J. Hays / Agricultural and Forest Meteorology 128 (2005) 5765

65

References
Acock, B., Reddy, V.R., Whisler, F.D., Baker, D.N., McKinion, J.M., Hodges, H.F., Boote, K.J., 1983. Response of vegetation to carbon dioxide. The soybean crop simulator GLYCIM: Model Documentation. 1982. Joint program of the U.S. Department of Energy, Carbon Dioxide Research Division, and U.S. Department of Agriculture, Agricultural Research Service, conducted at USDA-ARS Crop Simulation Unit and Department of Agronomy, Mississippi Agricultural and Forestry Experiment Station, pp. 316. Cesaraccio, C., Donatella, S., Duce, P., Snyder, R.L., 2001. An improved model for determining degree-day values from daily temperature data. Int. J. Biometeorol. 45, 161169. Harris, S.A., Pedersen, J.H., 1995. Comparison of three methods of calculating air temperature from electronic measurements. Z. Geomorph. 39, 203210. Hartzell, F.Z., 1919. Comparison of methods for computing daily mean temperatures: effect of discrepancies upon investigations of climatologists and biologists. Mon. Wea. Rev. 47, 799801. Hoogenboom, G., Huck, M.G., 1986. ROOTSIMU V4.0. A dynamic simulation of root growth, water uptake, and biomass partitioning in a soilplantatmosphere continuum: Update and documentation. Agron. and Soils Department Series 109, Alabama Agr. Exp. Stn., pp. 83.

Jones, C.A., Kiniry, J.R., 1986. CERES-Maize: A simulation Model of Maize Growth and Development. Texas A&M University Press, College Station. Landsberg, H., 1958. Physical Climatology, second ed. Gray Printing Company, DuBois, PA, 446 pp. McMaster, G.S., Wilhelm, W.W., 1997. Growing degree-days: one equation, two interpretations. Agric. For. Meteorol. 87, 291 300. Parton, W.J, Logan, J.A, 1981. A model for diurnal variation in soil and air temperature. Agric. Meteorol. 23, 205216. Raworth, D.A., 1994. Estimation of degree-days using temperature data recorded at regular intervals. Environ. Entomol. 23, 893 899. Reicosky, D.C., Winkelman, L.J., Baker, J.M., Baker, D.G., 1989. Accuracy of hourly air temperatures calculated from daily minima and maxima. Agric. For. Meteorol. 46, 193209. Sadler, E.J., Schroll, R.E., 1997. An empirical model of diurnal temperature patterns. Agron. J. 89, 542548. Sanders, C.G., 1975. Comments on the model for estimating the completion of rest for Red Haven and Elberta peach trees. Hortic. Sci. 10, 560561. Streck, N.A., Weiss, A., Xue, Q., Baenziger, P.S., 2003. Improving predictions of development stages in winter wheat: a modied Wang and Engel model. Agric. For. Meteor. 115, 139150. Wilkerson, G.G., Jones, J.W., Boote, K.J., Ingram, K.T., Mishoe, J.W., 1983. Modeling soybean growth for crop management. Trans. ASAE 26, 6373.

You might also like