Results and discussion Directory UMM :Data Elmu:jurnal:A:Agricultural & Forest Meterology:Vol102Issue1April2000:

A. Soltani et al. Agricultural and Forest Meteorology 102 2000 1–12 5 differences using Kolmogorov–Smirnov test. The Kolmogorov–Smirnov statistic is simply the max- imum absolute difference between the cumulative probability distributions of the two samples in ques- tion. The smaller the calculated test statistic, the smaller are the differences between the distributions in question. Statistical comparisons were performed using SAS SAS Institute, 1989.

3. Results and discussion

3.1. Comparison of recorded historic and generated weather data The mean rainfall amounts and mean number of wet days for each month of the year and for the an- nual totals are shown in Table 2 for the observed and generated data. The mean precipitation amounts from generated data differed significantly from the values obtained from the observed data for 4, 5, 3 and 3 of the 12 months for W3–W10, respectively. Rainfall of October–December and January–March has an im- portant role for increasing stored soil water and the successful growth of the crop under rainfed condi- tions is related to this stored water. The greatest dif- ferences between generated W3–W10 and historic rainfall amount occurred in these months, specially in January and February. The discrepancy may be cru- Table 2 Comparison of average monthly historic and generated rainfall amount and number of wet days for the four base periods used to parameterise WGEN Month Rainfall amount mm Number of wet days Observed W3 W5 W7 W10 Observed W3 W5 W7 W10 1 23.0 23.1 16.7 ∗ 13.2 ∗ 15.0 ∗ 8.6 11.0 ∗ 8.4 7.6 7.9 2 21.6 7.5 ∗ 10.7 ∗ 9.1 ∗ 14.4 ∗ 8.1 7.3 6.7 6.7 7.6 3 41.2 45.6 38.7 41.6 35.4 11.8 12.9 12.2 12.1 12.0 4 50.3 70.8 ∗ 50.2 46.3 47.3 11.4 11.6 10.0 10.1 11.1 5 44.0 46.3 55.0 43.5 41.8 11.3 11.2 14.3 ∗ 11.8 10.5 6 18.9 21.5 21.7 16.3 18.3 5.6 9.0 ∗ 8.5 ∗ 6.1 6.9 7 3.7 1.7 5.5 3.5 4.0 1.4 0.6 ∗ 1.5 1.6 1.7 8 3.5 2.2 3.0 3.1 5.5 1.2 1.7 1.4 1.1 1.7 9 8.5 4.7 2.5 ∗ 2.2 ∗ 1.8 ∗ 2.0 2.7 1.5 1.4 1.0 10 23.6 17.7 11.0 ∗ 18.7 28.4 6.7 3.2 ∗ 2.9 ∗ 3.7 ∗ 6.4 11 29.3 41.9 ∗ 41.8 ∗ 31.9 26.6 7.5 15.7 ∗ 11.8 ∗ 9.3 8.9 12 27.7 13.8 ∗ 25.7 23.8 35.2 8.2 3.8 ∗ 7.5 6.8 8.7 Annual 295.3 296.8 282.5 253.2 273.7 83.8 90.7 86.7 78.3 84.4 ∗ The observed mean and the generated mean are significantly different at the 1 level. cial when applied to the crop model and discussed in latter section. The average number of wet days gen- erated for each month did not compare well with the long term averages for 6, 4 and 1 of the 12 months for W3, W5 and W7, respectively. The proper description of the occurrence of wet days by season is important, because the generation of solar radiation and maxi- mum temperature are conditioned on the occurrence of wet or dry days. Most of the differences between the recorded and the generated data that are shown in Table 2 were due to deviations of means of base peri- ods from the means of the historic period. For exam- ple, February rainfall are 7.5, 10.7, 9.1 and 14.4 mm for W3–W10 in comparison to 21.6 mm for the ob- served historic mean. For this month averages of rain- fall amounts are 8.7, 14.5, 11.6 and 17.4 mm for base periods of 3, 5, 7 and 10 years, respectively. There was a trend of reducing significant differences with lengthening of the base period used for parameter es- timation of the WGEN. t-Tests showed no significant differences between the generated data and their cor- responding base periods used for parameter estimation for precipitation and number of wet days data not shown, suggested that model assumptions are valid. The solar radiation data generated with the model using various base periods for parameter estimation are compared with the observed data in Table 3. The means were significantly different from the historic in 58, 17, 0 and 17 of the time for W3–W10, 6 A. Soltani et al. Agricultural and Forest Meteorology 102 2000 1–12 Table 3 Comparison of average monthly historic and generated solar radi- ation for the four base periods used to parameterise of WGEN Month Solar radiation MJ m − 2 d − 1 Observed W3 W5 W7 W10 1 8.0 7.7 8.4 8.3 8.3 2 9.9 9.2 ∗ 9.5 9.5 9.5 3 13.4 13.6 13.5 13.6 13.3 4 18.7 17.3 ∗ 19.1 18.7 18.4 5 23.5 24.1 ∗ 23.1 23.4 23.9 6 26.9 26.5 26.8 27.0 27.1 7 27.9 27.7 27.8 28.1 28.8 8 25.2 24.5 ∗ 24.5 ∗ 24.8 25.4 9 20.5 21.0 ∗ 20.5 20.4 21.0 ∗ 10 14.7 14.9 14.9 14.5 14.4 11 10.1 8.3 ∗ 8.7 ∗ 9.7 9.8 12 7.8 9.0 ∗ 7.9 7.8 7.3 ∗ Annual 17.2 17.0 17.0 17.2 17.2 ∗ The observed mean and the generated mean are significantly different at the 1 level. respectively. The differences here were also due to deviations of means of base periods from the his- toric means. There were not significant differences between the generated data and their corresponding base periods for solar radiation data not shown. Table 4 shows a comparison between the historic and generated data W3–W10 for maximum temper- ature, minimum temperature, number of days with maximum temperature greater than 35 ◦ C and num- ber of days with minimum temperature less than 0 ◦ C by month and annually. For maximum temperature, 35 of the time significant differences were observed between the generated data of W3, W5 and W7 and the historic data. For W10 only 1 of the 12 months has a significant difference. For minimum tempera- ture significant differences were detected between the recorded and the generated data in 5, 3, 4 and 2 of the 12 months, respectively. There was a tendency for lower significant differences with increasing of year number used to parameter estimation of the WGEN. The greatest differences between generated and actual number of days per month with a maximum tempera- ture greater than 35 ◦ C and the number of days with a minimum temperature less than 0 ◦ C were occurred in the warmest and coldest months of the year, respec- tively, specially when WGEN’s required parameters were calculated from 3 years of recent actual data. The total annual number of days with a maximum temperature greater than 35 ◦ C were significantly dif- ferent from the historic data for W7 and W10. The differences here between the generated data and their corresponding base periods data were not significant data not shown. The CDFs of the daily precipitation, solar radiation, maximum temperature and minimum temperature ob- tained from the generated and the recorded data were constructed and evaluated data not shown. Statistical comparison using Kolmogorov–Smirnov test showed that all CDFs of generated maximum temperature and minimum temperature data and 50 of CDFs of gen- erated daily rainfall data differed significantly from the historic data Table 5. There were not significant dif- ferences between simulated and actual solar radiation distributions in virtually all cases. The greater num- ber of years used to estimate the required parameters of WGEN did not affect on the number of significant differences. Briefly, the limited number of significant differences between the means of generated data of W3–W10 and their corresponding means of base periods showed that the generated values for all climatic variables were very similar to that actual data used for parameter esti- mation for all base periods tested. Thereby, the model is capable of representing the characteristics that ex- isted in the observed data. However, the analysis also indicated that 3–10 years were not enough to ade- quately represent the 30 years of recorded data, be- cause of the deviations of means of base periods from the means of the historic. With increasing the num- ber of years used for parameter estimation of WGEN from 3 to 10, significant differences were 38, 26, 17 and 13 for W3–W10, respectively. Thus if adequate representation is required, a longer base period for parameter estimation would be required. Larsen and Pense 1982 showed that 17 years were not enough to adequately represent the 80 years of recorded data at Columbia. Therefore, a long base period, perhaps equal to that of historic period may be necessary. On the other hand, statistically significant differences presented in Tables 2–5, with exception of rainfall amount, are small. It is interesting to see whether these differences are important from the biological view of point to cause different responses when applied to crop models. Meinke et al. 1995 showed while the gener- ated and the recorded data had significant differences, crop models responses were the same using the gen- A. Soltani et al. Agricultural and Forest Meteorology 102 2000 1–12 7 Table 4 Comparison of average monthly historic and generated maximum temperature, minimum temperature, number of days with maximum temperature greater than 35 o C and number of days with minimum temperature less than 0 o C for the four base periods used to parameterise of WGEN Month Maximum temperature ◦ C Minimum temperature ◦ C Observed W3 W5 W7 W10 Observed W3 W5 W7 W10 1 1.9 2.4 2.2 1.3 2.1 − 6.6 − 5.2 ∗ − 6.0 − 7.6 − 7.0 2 3.9 5.1 4.6 3.4 4.1 − 4.5 − 3.4 − 4.2 − 5.1 − 4.4 3 9.9 10.6 9.3 9.8 9.1 0.3 0.3 − 0.3 − 0.1 − 0.3 4 16.9 17.3 17.2 17.8 ∗ 17.5 6.0 6.8 ∗ 6.5 6.6 6.6 5 22.2 21.9 21.1 ∗ 22.2 22.3 10.6 10.5 10.0 10.4 10.3 6 28.4 27.5 ∗ 27.8 28.5 28.3 15.2 15.2 15.4 15.7 15.4 7 32.7 32.2 32.8 33.3 33.4 19.5 19.5 19.8 20.3 ∗ 20.3 ∗ 8 32.2 33.6 ∗ 33.1 ∗ 33.0 ∗ 32.7 19.1 20.5 ∗ 20.2 ∗ 20.0 ∗ 19.7 9 28.1 28.7 28.6 29.1 ∗ 29.0 ∗ 14.4 15.2 ∗ 15.1 ∗ 15.0 ∗ 14.8 10 20.1 21.1 ∗ 21.5 ∗ 21.2 ∗ 20.7 8.1 8.6 8.9 ∗ 8.7 ∗ 8.8 ∗ 11 11.9 11.0 11.6 11.9 11.9 2.2 2.3 2.5 2.6 2.4 12 4.5 2.9 ∗ 3.0 ∗ 4.5 4.5 − 3.0 − 4.2 ∗ − 3.6 − 2.6 − 2.8 Annual 17.7 17.9 17.7 18.0 17.9 6.8 7.2 7.0 7.0 7.0 Days 35 ◦ C Days 0 ◦ C 1 28.5 26.9 ∗ 27.8 28.8 28.6 2 22.9 24.2 25.0 24.7 23.4 3 12.0 14.5 16.4 ∗ 15.4 16.2 ∗ 4 1.0 0.4 ∗ 0.9 0.9 1.0 5 0.1 6 0.6 0.9 0.6 1.0 1.2 7 7.0 4.2 ∗ 5.8 8.5 8.7 8 4.9 8.2 ∗ 8.5 ∗ 8.0 ∗ 7.5 ∗ 9 0.1 2.0 ∗ 1.7 ∗ 2.1 ∗ 1.4 ∗ 10 0.2 0.1 0.1 11 7.1 8.7 8.0 8.0 8.2 12 22.4 26.5 ∗ 25.5 ∗ 23.1 22.7 Annual 12.6 15.3 16.6 19.7 ∗ 18.8 ∗ 94.1 101.2 103.6 101.0 100.2 ∗ The observed mean and the generated mean are significantly different at the 1 level. erated and recorded data and significant differences were rare. We examined this through analyses of out- puts of a chickpea crop model applied to generated and actual data. 3.2. Sensivity test using a chickpea crop model Under rainfed conditions of NW Iran insufficient plant available soil water commonly overwrites any effects of variation in temperature and radiation. To eliminate this large influence of water availability on crop growth and hence on sensivity test, the chickpea crop model was run under irrigated i.e. non-limiting water for plant growth and rainfed conditions to eval- uate WGEN capability for generating temperature and radiation, and rainfall, respectively. In the chickpea model the effects of non-optimal temperatures on plant growth is considered by multiplying radiation use efficiency RUE by a scalar factor that has a value of 1 between 15–24 ◦ C of Table 5 Kolmogorov–Smirnov test statistics for the comparison of the dis- tribution of actual vs generated rainfall, solar radiation, maximum temperature and minimum temperature for the four base periods used to parameterise WGEN W3 W5 W7 W10 Rainfall 0.018 ∗ 0.008 0.015 ∗ 0.007 Solar radiation 0.014 0.011 0.010 0.014 Maximum temperature 0.026 ∗ 0.032 ∗ 0.025 ∗ 0.022 ∗ Minimum temperature 0.021 ∗ 0.027 ∗ 0.024 ∗ 0.019 ∗ ∗ Significantly differing distributions from those of actual data. 8 A. Soltani et al. Agricultural and Forest Meteorology 102 2000 1–12 average daily temperature but declines linearly to 0 at 0 and 39 ◦ C. The effect of higher temperature on hastening crop phenology and leaf senescence is also incorporated. As indicated in Section 3.1, there were statistically significant differences between the gen- erated maximum temperature, minimum temperature, number of days with maximum temperature greater than 35 ◦ C and number of days with minimum tem- perature less than 0 ◦ C and the recorded ones during April–August months which coincides with growing season of chickpea. It should be answered to this question whether these differences have significant influences on plant growth. To do this, during model simulations the number of days to maturity, number of days with average temperature between 15 and 24 ◦ C optimal days, number of days with average temperature less than 15 ◦ C sub-optimal days and number of days with average temperature greater than 24 ◦ C supra-optimal days were saved and compared Table 6. None of the base periods used had any influence on the simulated number of days to phys- iological maturity for all three sowing dates under Table 6 Comparison of means of days to maturity DTM, days with average temperature less than 15 ◦ C DLOPT, and days with average temperature between 15 and 24 ◦ C DOPT and days with average temperature greater than 24 ◦ C DUOPT simulated for three sowing dates under irrigated and rainfed conditions using the chickpea model with actual weather data and weather data generated with WGEN Irrigated Rainfed DTM DLOPT DOPT DUOPT DTM DLOPT DOPT DUOPT 26 March Actual 120.8 40.4 55.4 25.0 110.1 40.4 54.0 15.7 W3 119.8 40.6 53.4 25.8 109.6 40.6 52.0 17.0 W5 120.8 42.9 49.5 ∗ 28.4 ∗ 110.6 42.9 48.6 ∗ 19.1 ∗ W7 118.7 39.1 50.2 ∗ 29.5 ∗ 108.0 39.1 49.3 ∗ 19.6 ∗ W10 119.4 39.6 50.7 29.1 ∗ 108.6 39.6 49.6 ∗ 19.4 ∗ 10 April Actual 111.5 26.0 55.9 29.6 100.0 26.0 54.0 19.9 W3 110.9 26.1 53.5 31.3 99.9 26.1 52.2 21.5 W5 111.6 28.8 48.8 ∗ 33.9 ∗ 100.6 28.8 48.2 ∗ 23.6 ∗ W7 109.9 25.2 49.6 ∗ 35.1 ∗ 98.3 25.2 48.6 ∗ 24.5 ∗ W10 110.3 25.3 50.4 ∗ 34.6 ∗ 98.6 25.3 49.5 23.8 ∗ 25 April Actual 104.1 13.9 55.0 35.2 91.6 13.9 52.1 25.6 W3 103.8 14.5 50.2 ∗ 39.1 ∗ 91.9 14.5 49.7 27.6 W5 104.5 17.2 ∗ 46.0 ∗ 41.4 ∗ 92.8 17.2 45.3 ∗ 30.3 ∗ W7 103.0 14.4 46.1 ∗ 42.5 ∗ 90.5 14.4 45.1 ∗ 31.0 ∗ W10 103.0 13.9 47.8 ∗ 41.5 ∗ 90.6 13.9 46.4 ∗ 30.3 ∗ ∗ Statistically significant difference. irrigated and rainfed conditions. Significant differ- ences were observed between the generated data W3–W10 and historic data for the number of opti- mal and supra-optimal days. In this respect, the lowest significant differences were observed for W3. Model output using recorded data showed that with delay in sowing date, percentage of sub-optimal and optimal days decreased but percentage of supra-optimal days increased. With the generated data of W3–W10 the same response was observed. Note that for 25 April planting date, each supra-optimal day has a more negative effect, because average daily temperatures are increased. In NW Iran rainfed chickpea is sown early spring and grows mainly on stored moisture which is pro- gressively depleted with crop growth. The crop expe- riences drought stress from late vegetative stage until maturity. The intensity of drought stress varies from year to year, depending on the amount and distribution of rainfall and on spring and early summer tempera- tures. This phenomenon was evaluated by examining the days to beginning of terminal drought stress from A. Soltani et al. Agricultural and Forest Meteorology 102 2000 1–12 9 Table 7 Comparison of means of days to terminal drought stress DTB, days with fraction of transpirable soil water less than 0.37DL37, ratio of DL37 to days to maturity FDL37 and crop evapotranspi- ration CET, mm simulated for three sowing dates under rainfed conditions using the chickpea model with actual weather data and weather data generated with WGEN DTB DL37 FDL37 CET 26 March Actual 10.4 33.6 0.31 235 W3 10.9 33.2 0.30 257 ∗ W5 10.7 33.9 0.31 258 ∗ W7 10.0 34.1 0.32 233 W10 9.4 34.5 0.32 236 10 April Actual 7.7 34.9 0.35 220 W3 8.5 33.7 0.34 232 W5 8.5 34.3 0.34 237 ∗ W7 7.6 34.8 0.35 214 W10 7.2 35.2 0.36 216 25 April Actual 5.7 35.3 0.39 200 W3 6.6 35.0 0.38 201 W5 6.7 34.2 0.37 213 W7 5.3 35.6 0.39 195 W10 5.3 36.0 0.40 196 ∗ Statistically significant difference. flowering as an origin DTB, days with FTSW less than 0.37 DL37, ratio of DL37 to days to maturity FDL37 and crop evapotranspiration CET using the generated and actual data inputted into the chickpea model and the results are presented in Table 7. Us- ing actual data, with delay in sowing date, terminal drought stress begins earlier and the crop experiences a greater number of days with FTSW less than 0.37 and FDL37 is increased. Simulations using generated data also showed the same response to sowing date, without any significant difference. With respect to considerable differences for winter rainfall between the generated and actual data Table 2 and its influence on the stored soil water, the results of Table 6 seems questionable. However, because of the limited soil depth of the re- gion 100 cm, see Section 2.3 and its limited capacity for water storage, the differences were masked due to a higher water loss simulated using actual data. The differences could be of greater significance in other regions and climates or even in the same climate with deeper soils although they are not common. Table 8 shows simulated irrigated and rainfed yields at three sowing dates using actual and generated weather data. With actual weather data, irrigated yield declined about 60 kg ha − 1 2 of 26 March yield for each 15-day delay in sowing date from 26 March. The same response was observed in simulated yields using W3–W10 weather data. At the two planting dates, i.e. 26 March and 10 April, there was no sig- nificant difference between irrigated yields obtained using actual weather data and those obtained using W3–W10 weather data p=0.01, with exception of W3 at 26 March. However, at 25 April sowing date, simulated yields were significantly different at a prob- ability level of 0.05 from the simulated yields using actual weather data, because the growing season is moved into the summer months when the maximum differences between number of supra-optimal days using the generated and actual weather data occurred. Ranges of irrigated yields are generally similar and the F-test showed no significant difference between variances of simulated yields using actual weather data and W3–W10 weather data. The number of years used to estimate the required parameters of WGEN did not affect accuracy of the yield predictions. Under rainfed conditions, simulated grain yield using actual weather data decreased by about 120 kg ha − 1 13 of 26 March yield per each 15-day delay in sowing date Table 8. Simulated yields us- ing generated weather data also showed the same response to sowing date. Means of yields obtained using actual weather data and W3–W10 were similar with exception of W3 at 25 April. As compared to irrigated conditions delayed sowing date did not in- crease differences in yield, because water availability is an overriding factor under rainfed conditions and the differences between the generated and the historic data were not significant in this respect Table 7. Yield ranges obtained from generated data were sim- ilar to those obtained using actual data, and F-test showed no significant difference between variances, with exception of W7 sown on 26 March. Under rainfed conditions, the number of data years used to parameterize WGEN also did not affect the conver- gence between simulated yield obtained using actual weather data and generated data. The CDFs of the yields under irrigated and rain- fed conditions obtained using the actual data and generated data indicate that using actual Tabriz 10 A. Soltani et al. Agricultural and Forest Meteorology 102 2000 1–12 Table 8 Comparison of means and ranges of chickpea grain yield t ha − 1 simulated for three sowing dates under irrigated and rainfed conditions using the chickpea model with actual weather data and weather data generated with WGEN Irrigated Rainfed Mean Prt a Range PrF Mean Prt b Range PrF 26 March Actual 2.61 – 2.31–2.87 – 0.91 – 0.52–1.32 – W3 2.71 0.0038 2.26–3.21 0.3745 0.97 0.1182 0.62–1.44 0.0442 W5 2.65 0.1609 2.26–3.10 0.9286 0.98 0.0900 0.48–1.35 0.0459 W7 2.61 0.9190 2.23–2.90 0.8562 0.90 0.9392 0.54–1.39 0.0150 W10 2.68 0.0413 2.28–3.16 0.4300 0.90 0.9316 0.43–1.37 0.0782 10 April Actual 2.56 – 2.30–2.90 – 0.81 – 0.51–1.15 – W3 2.59 0.1776 2.18–2.98 0.9543 0.84 0.2034 0.53–1.41 0.0783 W5 2.57 0.6587 2.23–2.96 0.8688 0.86 0.0720 0.48–1.27 0.2098 W7 2.53 0.4105 2.14–2.83 0.6115 0.79 0.6132 0.49–1.23 0.0865 W10 2.58 0.4494 2.12–3.00 0.3496 0.78 0.4373 0.39–1.15 0.0618 25 April Actual 2.49 – 2.21–2.81 – 0.67 – 0.47–0.92 – W3 2.42 0.0413 1.98–2.78 0.8635 0.71 0.0877 0.45–1.18 0.7960 W5 2.41 0.0082 2.00–2.71 0.5906 0.74 0.0094 0.48–1.15 0.8020 W7 2.38 0.0019 2.04–2.74 0.8473 0.66 0.7783 0.40–0.99 0.6040 W10 2.41 0.0194 2.06–2.84 0.8490 0.65 0.5625 0.40–0.95 0.7743 a Prt: Probability that a significant t value would occur by chance. b PrF: Probability that a significant F value would occur by chance. weather data, irrigated yields are between 2.34 and 2.75 t ha − 1 in 80 of years data not shown. Same quantities were 2.34–2.82, 2.32–2.74, 2.28–2.72, and 2.31–2.79 t ha − 1 for W3–W10, respectively. Grain yield obtained using W7 weather data was greater than yield obtained using actual data for most of the years. Under rainfed conditions grain yields were be- tween 0.55 and 1.11 t ha − 1 in 80 of the years. Same quantities were 0.63–1.09, 0.65–1.10, 0.58–1.01, and 0.56–1.02 t ha − 1 for W3–W10, respectively. Median of yields were 2.55, 2.57, 2.53, 2.52, and 2.57 t ha − 1 for actual data and W3–W10 under irrigated condi- tions and 0.75, 0.82, 0.85, 0.78, and 0.77 for actual data and W3–W10 under rainfed conditions, respec- tively. Statistical comparison showed that with one exception W5 under rainfed conditions all CDFs of simulated yield using generated data did not differed significantly from the simulated yield using recorded data Table 9. Briefly, in comparison of grain yields simulated using the generated solar radiation, maximum tem- perature and minimum temperature and actual ones i.e. irrigated yields, significant differences were found in 50 of cases. Under rainfed conditions and using generated rainfall in addition to generated so- lar radiation, maximum temperature and minimum temperature, only in one out of 12 cases was a signif- icant difference W5 for the 25 April planting date observed in the comparison of simulated grain yields between W3–W10 against actual data. The uniformity was not due to capability of WGEN to generating weather data with adequate representation of actual data. It was resulted because considerable differences in rainfall amounts were masked by the limited soil depth typical of the region chosen for simulations. Table 9 Kolmogorov–Smirnov test statistics for the comparison of the distribution of grain yield simulated using actual weather data vs generated weather data by base period used to parameterization of WGEN W3 W5 W7 W10 Irrigated yield 0.129 0.049 0.123 0.119 Rainfed yield 0.163 0.210 ∗ 0.091 0.081 ∗ Significantly differing distributions from those of actual data. A. Soltani et al. Agricultural and Forest Meteorology 102 2000 1–12 11 The longer base period used for parameter estimation of WGEN did not result in lower differences.

4. Concluding remarks