Like natural soils, unfertilized agricultural fields show emissions of NO and N_{2}O. A prerequisite for such emissions is the availability of N, which is determined primarily by the mineralization of soil organic matter and N deposition. Mineralization rates depend on the history of land use (i.e. previous crop, residue management and fertilization).
To estimate the anthropogenic effect on emissions many authors have used the concept of the fertilizer induced emission (i.e. the emission from the fertilized plot minus that of the control plot expressed as a percentage of the N applied). In this concept, the emission from the control plot is assumed to be equivalent to the `natural' or background emission.
Tables 4 and 5 present the calculations of fertilizer induced emission for different fertilizer types irrespective of the length of the measurement period, climate and soil conditions. Not all measurements included control plots, hence the number of measurements for the fertilizer induced emission is smaller than that for the total N_{2}O or NO emission.
The measurements in the data set from the literature since 1995 have not changed significantly from earlier results found using a subset of the present data set except for the high mean values for nitratebased fertilizers and the high maximum values for ammonium sulphate and urea (Table 4). These results stem from one experiment in Costa Rica, with extremely high emissions. The authors of these measurements speculated that emissions in the tropics may exceed those in temperate regions as a result of the temperature and moisture conditions which are prone to high N_{2}O losses. However, the high maximum values in the data set for injected anhydrous ammonia were from measurements in a temperate climate: 7.3 percent and 6.8 percent.
The mean and median values for the fertilizer induced emission for various fertilizer types (Table 4) show important differences, indicating that the emission rates are not normally distributed.
Bouwman (1996) concluded that the length of the measurement period strongly influences the fertilizer induced N_{2}O emission. The N input from fertilizer appears to have an effect that lasts longer than the crop growing season. Figure 2 presents the relation between fertilizer N application rate and the measured N_{2}O emission for the measurements with a measurement period equal to or longer than one year present in the data set. Contrary to the results of Bouwman (1996), Figure 2 shows that there is no clear relation between N application rate and annual emission.
The data on fertilizer induced NO emissions from fertilized fields also show important differences between the mean and median values (Table 5). The median values for the fertilizer induced NO emission are lower than those for N_{2}O, with typical values of 0.10.4 percent, with the exception of the data for ammonium nitrate (AN) and ammonium sulphate. The high values for ammonium nitrate stem from one publication, while those for ammonium sulphate are from three measurements.
TABLE 5.

Hutchinson and Brams (1992) observed high NO:N_{2}O ratios in a Mediterranean climate. Harrison et al. (1995) observed the high values for calcium ammonium nitrate (CAN) at Rothamsted, the United Kingdom. However, the number of measurements is small, and the high standard deviation of the NO:N_{2}O ratio suggests that the uncertainty is high.
Figure 3 presents the relationship between N fertilizer application rate and NO emission from all the studies in the data set. It indicates that there is no clear relationship between N application rate and NO emission. This contrasts with Veldkamp and Keller (1997a) who concluded that 0.5 percent of the N applied as fertilizer is lost as NO (on the basis of a subset of the data used in this report).
The above discussion indicates that it is not possible to use simple relationships between fertilizer N application rates and emissions. The reason for the lack of correlation is that the data set includes measurements from a variety of locations with differing climate, soil, crop and management conditions. This calls for a different approach whereby the data set is summarized based on the factors that regulate N_{2}O and NO emissions.
The factors selected for the data summary include: climate, crop type, fertilizer type, application rate, mode and timing of application, soil organic C and N content, soil pH, soil texture and drainage, measurement technique, frequency of measurements and length of the measurement period. The analysis did not include organic soils.
Table 6 presents the classification used for climate. The description of the differences in soil conditions uses functional groupings based on: soil texture, drainage, soil organic C and N content, and soil reaction (pH) (Table 16, Annex 2). Crops were grouped into five crop types: grass, grassclover mixtures, wetland rice, legumes and other crops. Table 16 also shows the groupings for fertilizer types, application rate, and mode and for timing of fertilizer application (excluding measurements with grazing and fertilizer type CAN with grazing, and organic soil texture).
Table 3 summarizes the different measurement techniques in the data set. The classes used for the length of the measurement period were: <120, 120180, 180240, 240300 and >300 days. The groupings for frequency of measurement were: more than one measurement per day, one per day, one every 23 days, one every 37 days, less than one per week. Many studies have used variable measurement frequencies, with measurements that are more intensive shortly after fertilization and lower frequencies when emission rates dropped to background levels. In such cases, we selected the highest frequency. The groupings for the NO measurement data are different because of scarcity of data (Table 16).

On the basis of the classification for the different factors, the data set was summarized by calculating for each factor class: mean, median, balanced mean and balanced median values of the measured emission of N_{2}O and NO.
Balanced mean values indicate unbalanced features of mean values. They were calculated with the residual maximum likelihood procedure. The balanced median indicates unbalanced features and extreme values present in the median. Balanced median values were calculated with the residual maximum likelihood procedure using logtransformed emissions. Logtransformation reduces the influence of extreme values. Backtransformation yields a balanced mean value for the emission.
Subsequently, the most important factors in the data set explaining logtransformed N_{2}O and NO emissions were selected. The residual maximum likelihood procedure was done with `research paper' as a random effect, using the WALD statistic to select the most important factors (P < 0.005). For the selected factors with significant influence on the N_{2}O emission residual maximum likelihood models were developed to predict emissions of N_{2}O and NO. The formulation for the emission models for N_{2}O and NO is:
with Emission expressed as N in kg/ha. Backtransformation of the result of this equation yields the emission. Table 7 presents the different class values of the model terms.
Although Table 16 presents the mean, median, balanced mean and balanced median values for the measured N_{2}O emissions for all systems for all the factors under consideration, the discussion here concentrates on the balanced median values.
The balanced median values for N application rate are nearly constant emissions below 100 kg N/ha, and emissions increase along with the N application at rates exceeding 100 kg N/ha. Mean and median values show a similar pattern, except for rates of 150 kg N/ha which show higher values than the 50100 kg N/ha class. The trend is highest at rates exceeding 250 kg N/ha.
The balanced median for Clim1 exceeds that for Clim2, reflecting the observed winter emissions which are higher for Clim1 than for Clim2. The balanced median for subtropical climates exceeds that for Clim1 by 94 percent, and that for Clim2 by 127 percent. The balanced median value for N_{2}O emission in Clim5 exceeds that for Clim1 by 37 percent and that for Clim2 by 60 percent. Results for Clim6, Clim7 and Clim8 are more uncertain than for the other climate types due to the limited number of measurements.
For soil organic C content and soil N content the balanced median values indicate increasing N_{2}O emission along with increasing C and N content. The difference in the balanced median value between soils with >36 percent soil organic C and those with <1 percent is 38 percent, while the balanced median for soils with organic C content >6 percent exceeds those with <1 percent organic C by 113 percent. For soil N content the balanced median value for soils with 0.150.3 percent N exceeds that for 0.050.15 percent N by 38 percent. The number of observations in the other classes is limited and the calculated values are less certain.
The balanced median value for fine soil texture exceeds that for coarse texture by 8 percent and that for medium soil texture by 59 percent. The balanced median value for soil drainage exceeds that from welldrained soils by 35 percent. The balanced median value is highest for soils with intermediate soil pH (5.57.3) exceeding those for soils with pH < 5.5 by 7 percent and those with pH >7.7 by 49 percent.
The highest balanced median values are for mixed fertilizers, anhydrous ammonia and OS, respectively, while those for the NO_{3}^{}based and NP fertilizers (NF and NP) are lowest. The differences between highest and lowest values are 103109 percent. The high values for anhydrous ammonia and OS and low values for NF are consistent with earlier studies.
The difference between broadcasting and incorporation is small. Broadcasting at panicle initiation (bpi) in wetland rice cultivation leads to a reduction in N_{2}O emissions compared to broadcasting 23 weeks after transplanting by 23 percent, reflecting more efficient plant N uptake. The balanced median values for split applications exceed those for single applications only slightly.
The balanced median values are highest for leguminous crops, followed by other crops, grass, wetland rice and grassclover mixtures. The balanced median value for other crops exceeds that for grass by 121 percent, and that for rice by more than a factor of 7. The balanced median value for leguminous crops exceeds that for the other crops class by 37 percent. Lower emissions for grass than for other crops may result from more efficient N uptake by grass as a result of longer growing periods, particularly in temperate climates.
The differences in balanced median values between the micrometeorological technique and open and closed chamber methods are 46 and 23 percent, respectively, while the balanced median value for the gradient technique is highest and that for the open chamber technique lowest. In most cases there is an increase of measured N_{2}O along with the length of the measurement period. The balanced median value for measurements covering more than 300 days exceeds that for periods of 240300, 180240 and <180 days by 54, 72 and 172 percent, respectively. Many measurements cover at most a half year (<180 days), and yield only part of the annual emission and probably only part of the fertilizer effect. In general, highfrequency measurements show the lowest balanced median values, with the highest values at intermediate frequency (1 measurement in 23 days).
In summary, the results of the data summary agree with the literature. Regarding the factors determining the N availability, the results show that: (i) there is a strong increase of N_{2}O emissions along with N application rates; (ii) warm climates show higher N_{2}O emissions than temperate climates; and (iii) fertile soils with high organic C and N contents show higher emissions than less fertile soils.
Regarding factors influencing the N_{2}O:N_{2} ratio, fine soil texture, restricted drainage, and neutral to slightly acidic soil reaction are conditions that favour N_{2}O production and emission. Furthermore, emissions from grasslands are lower than for crops. With respect to measurement techniques, the results indicate that longer measurement periods yield more of the fertilization effect on N_{2}O emissions. This finding confirms the conclusions of an earlier study. Finally, intensive measurements (>1 per day) yield lower emissions than less intensive measurements, which is in agreement with other studies.
The following factors had a significant influence (P < 0.005): N application rate per fertilizer type, climate type, soil organic C, soil texture, drainage, pH, crop type, length of experiment and frequency of measurements. The most important factors are the interactions between the N application rate and the fertilizer type and the crop type.
All the factors with significant influence on emissions were included in a REML model (see equation in the section on data handling). The values of the model terms are presented in Table 7. To calculate the full annual N_{2}O emission with this model the class >300 days for length of the measurement period was used, while emission estimates based on frequencies of more than one measurement per day were considered to be more reliable than measurements with lower frequencies.
TABLE 7.

Table 17 (Annex 2) presents the calculated values for the mean, median, balanced mean and balanced median for the measured emissions. The balanced median values show a consistent increase in emissions with increasing N application rate, which is in line with generally observed trends in the literature. The balanced median values are highest for Clim3, and the balanced median value for Clim5 is 10 percent higher than that for Clim1, while it is 9 percent lower than that for Clim2. The balanced median value for Clim3 exceeds that for Clim1 by more than a factor of 5, and those for Clim2 and Clim5 by more than a factor of 4.
The balanced median value for soils with a soil organic C content of >3 percent exceeds that for soil with <3 percent by a factor of 5, which is in agreement with the results for N_{2}O emissions. Contrary to the results for N_{2}O for soil texture, the balanced median value for NO emission for coarse soil texture exceeds that for medium texture by 160 percent and that for fine texture by 148 percent.
Except for the median values, all values for well drained soils exceed those for poorly drained soils, a reflection of the fact that high NO emission requires aerobic conditions. Soil pH has a marked effect on NO emissions, with higher balanced median values for low pH soils (pH<5.5) compared with soils with pH >5.5; a finding in agreement with the literature.
The fertilizer types that are well represented in the data set include the NH_{4}^{+}yielding fertilizers, AN, NO_{3}^{}  yielding fertilizers, urea and urine. In this group of fertilizer types the highest balanced median values are for AN, exceeding those for AF, NF and UU by 75, 17 and 26 percent, respectively.
The balanced median value for broadcasting exceeds that for incorporation by more than a factor of 5 and that for application in solution by 124 percent. For timing of fertilizer application, comparisons between the different classes are difficult due to uncertainty in the values for split application schemes.
The balanced median values for other crops exceed those for grass and leguminous crops by 85 and 176 percent, respectively. The data for wetland rice are too scarce for drawing any conclusions.
The balanced median value for open chambers exceeds that for closed chambers by 24 percent. Most measurements covered less than 120 days, with lower balanced median values than for measurements covering >120 days. This suggests that also for NO the measurement period may be important when assessing measurements from literature. However, although the data available for measurement periods of >300 days are scarce, the results suggest that the effect of N application on NO emissions is less longlasting than for N_{2}O. Highfrequency measurements (>1 measurement per day) of NO emission show lower balanced median values than measurements with lower frequencies. This pattern is in general agreement with the results for N_{2}O.
In summary, for the factors that determine N availability, the data indicate that NO emissions increase along with N application rates, and that warm climates and fertile soils with high C contents favour NO emission. For factors influencing the relative rate of NO emission, the data show that NO emissions are highest for coarse textured and neutral soils. Longer measurement periods yield higher emissions than short measurement periods although this effect is less certain than for N_{2}O due to the smaller number of measurement data. As with N_{2}O, highfrequency measurements yield lower emission estimates than lower frequencies. All these findings agree with the general understanding of the controls of NO fluxes.
The factors that exert a significant influence on NO emissions are: N application rate per fertilizer type, soil organic C content and drainage. The influence of climate is not significant. NO emissions appear to be much more concentrated in the crop growing season than N_{2}O. During the growing season, climatic conditions differ less between climate types than during other periods such as winter, spring and autumn.
Rather than presenting the complete data set, this section summarizes the data for the major fertilizer types for upland and flooded systems.
The analysis of the literature found only a few measurements for anhydrous ammonia, an indication that NH_{3} volatilization from this fertilizer is low. This is probably related to the mode of application (usually injection), and volatilization losses may occur when the injector does not penetrate deep enough, or when the soil is either too wet or too dry. Moreover, the spacing between the lines of injection influences the NH_{3} loss, with higher losses associated with close spacing. This is related to the anhydrous ammonia concentration per unit volume of soil, which is higher with close spacing.
Few data are available for ammonium bicarbonate (ABC). An NH_{3 }volatilization rate of 21 percent was measured with the enclosure technique without forced draught. Laboratory measurements with the forced draught technique in calcareous loess soils indicate NH_{3} volatilization rates of more than 30 percent, in some cases up to 70 percent.
Measurements with the mass balance technique presented by Jarvis et al. (1989a) indicate an NH_{3} volatilization rate of 6 percent. However, these measurements refer to grazed grassland, and it is not clear whether the loss is attributable to the fertilizer alone, as part may have been the result of animal excretion in the field. The data indicate that NH_{3} volatilization rates are 02 percent at low pH, while for soils with high pH and low CEC the loss may exceed 60 percent.
Few micrometeorological measurements are available for trashcovered soils in sugar cane fields. NH_{3} volatilization rates were 02 percent for ammonium sulphate, which is much lower than observed in some experiments using other techniques. NH_{3 }volatilization rates of up to 90 percent of the applied N have been reported using forced draught systems for calcareous soils.
Forced draught techniques in laboratory measurements with both slightly acid, neutral and slightly alkaline soils resulted in NH_{3} volatilization rates of up to 6 percent, while field measurements with forced draught techniques showed much lower NH_{3} volatilization rates. Measurements with wind tunnels on slightly acid soils also showed low NH_{3} volatilization rates.
Although not an NH_{4}^{+}based fertilizer, calcium nitrate (CN) has enabled comparisons with other fertilizer types. The difference between the N applied as fertilizer and the N recovered in the crop used as a proxy for NH_{3 }volatilization loss resulted in low to negligible volatilization rates.
Both the forced draught technique and the N balance method for broadcast diammonium phosphate (DAP) show that NH_{3} volatilization rates are highly variable, ranging from 2 percent to over 50 percent. Laboratory experiments for soils with high pH reported the highest NH_{3} volatilization rates. The N balance method with high pH soils also resulted in high NH_{3} volatilization rates (up to 35 percent). The incorporation of DAP resulted in lower NH_{3} volatilization rates of about 10 percent, about half of the rate observed for broadcast DAP on alkaline soils.
The forced draught measurements reviewed in the data set yielded NH_{3} volatilization rates from broadcast monoammonium phosphate (MAP) ranging between 2 and 35 percent on soils ranging from slightly acid to alkaline. The N balance method showed lower NH_{3} losses (8 percent) for broadcast MAP than for DAP on high pH soils. Incorporated MAP showed NH_{3} volatilization rates of 2 percent.
Measurements of NH_{3} losses from urea fertilizer show a coherent pattern. Results from closed systems, forced draught, micrometeorological and wind tunnel measurements indicate that NH_{3} volatilization rates range from 1520 percent of the applied nitrogen for broadcast fertilizer. For incorporated urea, the NH_{3} volatilization rates are between 5 and 15 percent.
Volatilization rates of NH_{3} from broadcast ureaammonium nitrate (UAN) measured with micrometeorological techniques were close to 15 percent, with a range of 8 to 18 percent. However, soil pH values were not reported. NH_{3} volatilization rates measured with the forced draught technique ranged from negligible amounts to almost 45 percent in a field experiment with broadcast UAN on a heavy textured vertisol. High NH_{3} volatilization rates were also observed with the N difference method on neutral to slightly acidic soils.
The data from micrometeorological methods show that NH_{3} volatilization rates from ABC may reach 40 percent. The values obtained for Chinese paddy fields were lower, due to the low pH of the floodwater. However, where NH_{3} volatilization rates were low, the total loss of N from ABC was still high as a result of denitrification. The method of application did not influence markedly the NH_{3} volatilization in the measurements. For ABC, there were only measurements based on micrometeorological techniques.
The one micrometeorological measurement available for broadcast ammonium sulphate indicates that up to 40 percent of ammonium sulphateN may volatilize. Volatilization of NH_{3} from ammonium sulphate broadcast at panicle initiation may amount to 10 percent, and about 5 percent from incorporated ammonium sulphate applied at transplanting. The seasonal loss of NH_{3} from ammonium sulphate incorporated at transplanting and broadcast at panicle initiation was about 5 percent in the Philippines. Enclosurebased measurements show a similar pattern, with NH_{3} losses varying between negligible amounts and 2030 percent for broadcast ammonium sulphate, and between insignificant loss and 610 percent for incorporated ammonium sulphate. The measurements with open and closed bottles show high NH_{3} volatilization rates from floodwater with a pH of 910, and markedly lower from floodwater with a pH <9.5.
Micrometeorological techniques used in flooded rice fields fertilized with urea applied at transplanting of rice show a clear difference in NH_{3} volatilization rate between broadcast urea (26 percent; range 056 percent) and incorporated urea (21 percent; range 043 percent). Urea applied at panicle initiation gives lower volatilization rates of 7 percent (115 percent). Forced draught techniques show NH_{3} volatilization rates of 17 percent for broadcast urea, and higher values of 20 percent for incorporated urea. ^{15}N techniques give somewhat higher NH_{3} volatilization rates of 28 percent (560 percent). Incorporation of urea in the puddled soil before permanent flooding results in NH_{3} losses of close to 10 percent (016 percent).
The analysis of the complete set of literature data to assess relationships between the various regulating factors and NH_{3} volatilization rates made use of Genstat 5 release 4.1 (PC/Windows NT).
The regulating factors considered consisted of: type of measurement (field or laboratory), measurement technique, soil pH, CEC, and organic carbon content, temperature during measurements, fertilizer type, method of application, N application rate, and type of crop. The analysis did not consider factors for which data were scant. Nor did it consider factors related (indirectly) to weather conditions, such as rainfall during the period of measurements, wind speed, algal growth and associated floodwater pH. This is because it is not possible to use weather conditions for making predictions. Finally, the analysis also excluded studies concerning the use of chemicals such as algicides, urease and nitrification inhibitors. Hence, the analysis involved 1 667 individual measurements (out of a total of 1 900), from 148 different studies.
Table 8 presents the various factor groupings and their classifications. For crop type, the analysis assumed that in upland systems fertilizer is applied at seeding, hence the soil surface is bare. Field and laboratory experiments with nonflooded bare soil were therefore included in this group. The classification of pH values in Table 8 derived from the fact that careful analysis of the data suggested a nonlinear relation between soil pH and NH_{3} volatilization rates. Moreover, such a classification would make the relationships compatible with the classifications used in the global data bases of soil properties. The analysis did not consider floodwater pH (an important factor in flooded systems) as it is not possible to use this factor in extrapolations.
TABLE 8.

The soil texture classification in Table 8 consists of three broad groupings: coarse (including sand, loamy sand, sandy loam, loam, silty loam and silt), medium (sandy clay loam, clay loam and silty clay loam) and fine (sandy clay, silty clay and clay).
The next step was to make a summary of the data by determining straightforward mean values (M) for the NH_{3} volatilization rate for each class of all the regulating factors (Table 8). Because values from one source are probably not independent, each data source received an equal weighting to calculate these means. Weighting has no systematic influence on the result in the case of independent values. However, by weighting, only 148 degrees of freedom remain instead of the 1 667 of the full data set. The weight representation (WR) given for each factor class (Table 8) depends on the number of studies reporting NH_{3} volatilization rates for this factor class and the number of NH_{3} volatilization rates reported in each study. As the data set includes results of 148 different studies, the maximum value of WR is 148. Where, for example, a factor class occurs with nine others in only one study the weight representation is 0.1.
Next, all NH_{3} volatilization rates were log transformed. This reduced the influence of outliers, particularly the extremely high NH_{3} loss rates in the data set. The residual distribution of the logtransformed NH_{3} volatilization rates is closer to a normal distribution, and backtransformation results in an estimate of the median value for the NH_{3} volatilization rate estimated rather than the mean. Effects or differences between factor classes were shown and studied by determining balanced weighted medians (BM) for the NH_{3} loss rate for each factor class, eliminating the influence of the other factors considered. Table 8 presents the backtransformed values for the balanced weighted medians. The analysis did not study the effects of specific combinations of different factors on NH_{3} volatilization rates (interaction effects) because: (i) analysing the data set for all such combinations is difficult given the number of factors and classes; and, (ii) a priori knowledge of such combinations was not available.
The next stage was to develop a linear regression model (summary model) for logtransformed weighted values of NH_{3} volatilization rates, and to use it to calculate global NH_{3} volatilization losses. For predicting NH_{3} volatilization rates, it is advisable to perform the regression on the basis of the data used in the extrapolation. However, for laboratory measurements this is not possible. Furthermore, the resolution of 0.5°x0.5° in the maps used gives a generalized representation of environmental and management conditions on the landscape scale. Such a resolution is therefore not suitable for use in combination with local field measurements.
The results for the means and balanced weighted medians indicate that there is a clear difference between measurements carried out in the field and laboratory studies (Table 8). Various factors may account for this, including the measurement technique (generally forced draught enclosures aimed at determining the maximum NH_{3} loss), though also the environmental conditions in the laboratory may favour NH_{3} losses. Finally, in most cases (except in greenhouse studies) the soils in the enclosures were uncropped, which may also favour NH_{3} loss.
The measurement technique used to determine NH_{3} losses is also very important. The mean and balanced median values were high for ^{15}Nbased measurements and indirect open measurement (ioc). The mean value for the forced draught technique (cfd) is higher than that for micrometeorological techniques (m), while the balanced medians show the reverse order. The estimates for both techniques derive from a large number of observations in the data base, providing a much firmer basis than the data available for the other techniques.
The influence of the type of crop is less important than the location or measurement technique used (Table 8). The mean for grass is 20 percent lower than that for upland crops, and 10 percent lower than that for flooded systems. The mean values confirm the expectation that NH_{3} volatilization rates are generally lower in grasslands than in croplands, but the balanced medians show almost no difference.
However, the effect of the type of fertilizer applied on the NH_{3} loss is, as expected from the literature review, very important. The mean values and balanced medians are in broad agreement with expert judgements. One exception is anhydrous ammonia where the mode of application is not accounted for. The effect of the type of fertilizer applied on the NH_{3} volatilization is, as expected, very important. Differences between fertilizer types occur both in the mean and balanced median, with the highest values for AN applied to grazing land, then for manure, and urea, and the lowest values for CN and anhydrous ammonia.
Broadcasting and application of fertilizer in liquid form have similar balanced medians. Incorporation leads to an important reduction of 50 percent in comparison with broadcasting. In rice systems the application of fertilizer before inundation (bf; i/f) and application at panicle initiation (bpi) has much lower balanced median values than application to the flooded field (b; bw). The reduction of NH_{3} volatilization that is achievable by application at panicle initiation compared to broadcasting at transplanting is 50 percent, which is in agreement with the literature review. In contrast to the application mode, the application rate exerts no clear influence on NH_{3} loss rates.
The results for soil properties agree, to varying degrees, with the expectations based on studies in the literature. The balanced medians for soil pH > 8.5 exceed those for the pH range of 5.5 to 7.3 by 61 percent, and those for pH  5.5 by 80 percent. The balanced medians for soil pH in the range of 7.3 to 8.5 exceed those for the range of 5.5 to 7.3 by 39 percent and those for pH  5.5 by 55 percent.
The effect of soil CEC is somewhat less clear than that of soil pH. The mean values for the different CEC classes show a consistent pattern, with lower NH_{3} volatilization in soils with high CEC than for low CEC. The balanced medians are 40 percent lower for CEC > 32 cmol/kg than for soils with CEC < 32 cmol/kg. However, in the balanced medians the relationship between NH_{3} volatilization rates and CEC for soils with CEC < 32 cmol/kg has disappeared.
The influence of soil organic carbon content is not clear. Both the mean and balanced median values are higher for the lowest soil organic carbon class than for the second lowest. The number of observations is small in the classes 3 and 4, and estimates are therefore less reliable than for classes 1 and 2.
The influence of soil texture is not clear, with high means and balanced medians for the NH_{3} volatilization rate for medium textured topsoils, and lower values for both fine and coarse textured soils.
The next stage was to study the combined effects of the different factors using regression analysis. From the set of factors selected for the above data summary, the regression concerned only those that had a clear influence on NH_{3} volatilization, except for weatherrelated factors and the measurement technique and location factors.
It was necessary to exclude weatherrelated factors a priori from the regression as lack of data prevents their use in extrapolations. The different measurement techniques and locations evidenced clear differences between both the means and balanced medians. However, their exclusion depended on the fact that it is not possible to use measurement techniques for predictions as the a priori knowledge for judging their accuracy is lacking, while the data set used cannot provide a sufficient basis for explaining the influence of the measurement location. Therefore, the regression yields an average value for NH_{3} volatilization rates applying to all measurement techniques and locations included in the data set.
The N application rate, soil organic carbon content and soil texture show no consistent relationship with NH_{3} volatilization rates and were therefore not used in the regression. The absence of a relationship may be due to the fact that soil organic carbon and texture are the main determinants of soil CEC. Therefore, the influence of soil C and texture on NH_{3} volatilization rates is assumed to be included in the CEC factor.
The factors selected for the regression included: crop type, fertilizer type and application mode, temperature, soil pH, and CEC. Table 9 presents the fitted parameters for the different factor classes of the resulting summary model. The factor values are logtransformed. The calculation for the NH_{3} loss rate is: exp(factor value crop type + fertilizer type + application mode + soil pH + coil CEC + climate). For example, for grass fertilized with urea by broadcasting (b) on a soil with a pH 5.5<pH<7.3, a CEC of 16<CEC<24, in a temperate climate the NH_{3} volatilization rate is: exp(0.158 + 0.666  1.305  0.933 + 0.012  0.402) = exp(2.120) = 0.120. Hence, the loss as a fraction of ureaN application is 0.120.
The variance accounted for by the model is about 30 percent. This means that individual NH_{3} volatilization values from research papers differ, on average, about 15 percent less from the means calculated by the model than from their common mean. Hence, the summary model is not suitable for calculating NH_{3} volatilization rates from measurements in individual research papers for specific sites. The model calculates median NH_{3} volatilization rates, which are of more interest when the working scale is that of landscapes rather than point measurements done under sitespecific conditions.
TABLE 9.
