re areas on the surface of figures 6 and 7 showing retention values that 

 There a ^^p^^g j_i-,2e ^ and there are areas wherein there are no real data values. 

 jQgi^a > point on the high F surface of the 1/3 bar model where OM = 12 and 



The 



predicted retention is approximately 33 percent by volume. This is 



- I 



- l-^' ly^the volume of the total pore space. We do not want to promote the con- 

 ,j-Q,xima _^'^^^^tj^Qn percent is equal to 1/3 bar percent for the above array of inde- 



^''ables. The tables presented in the Appendix show dashed lines instead 

 j^,nt vari ^^^^^^^ vvhere the model is not deemed applicable. 



■■-all^'i same responses in water retention at 1/3 bar are found at 15 bars 

 '^'^^ ^''Q^g' ^ajor difference is the lack of sharp change in retention in the middle 

 ^i^'^'^'b ranc^e for high OM values and low F values. This indicates that the distinct 

 ^j. tiiese values noted in the 1/3 bar model may be due to a structural binding 



I ithi 



.K in;:i- organic particles to form voids that drain at tensions greater than 1/3 

 ^ '\-iut^less than 15 bars. This effect is diminished as percent F increases in the 



'-'bar model. This may indicate that higher fines either mask the effect, or change 

 ' s'tructural interaction of OM and lithic material. 



\l 15 bars tension, organic matter is again the independent variable that most 



in 



-1 pnces water retention. At all OM levels, retention increases with increasing 

 values of both Pb and F. 



Model Performance 



The two retention models were developed from soil samples of a very localized 

 ii-oa within the Batholith. How applicable these models might have been to alternative 

 areas was unknown, but some insight was provided by an evaluation of model performance 

 ^-^r two additional sets of soil samples from new and geographically diverse areas 

 of the Batholith. Univariate data descriptions for the new data sets appear in 

 tables 2 and 3. 



Evaluation of the scale and form of the original models was made possible by 



first adjusting these models to the new data sets by least squares, in a simple 



linear model forced through the origin: adjusted retention = b (model retention, M) ; 



y = new retention values; and, b = ZM.Y./IM.^. 

 1 111 



Tiic b-value simply shrinks or stretches the original model to fit the new data set 

 \>ithout changing the specified form and provides a basis for judging adequacy of the 

 original model scale. A b-value of 1.000 represents a perfect scaling match between 

 model and data set and as may be seen in table 4, the b's ranged from 0.909 to 1.065 

 for the new data sets; i.e., the original model scales ranged from about 9-percent liigh 

 to h-percent low. In a larger array of clieck data sets, we might reasonably expect an 

 even larger range of corrections, perhaps ±15 percent. Users of prediction tables 5 

 and 6, Appendix, should be aware of the magnitude of this potential scaling error for 

 t:ie ■Kaajustad models, but users should also realize that the models can be adjusted to 

 ■^oil data for any-^ area as they were above for the check data sets, thus removing scal- 

 ing bias of the models for new areas. 



Some idea of the applicability of the interaction form for the new data sets 

 IS provided by an evaluation of departures of actual retention values from the corres- 

 ponding ones given by the models adjusted to each new data set (fig. 8, table 4). 



Again, the authors advise that application of the original models be limited 

 to areas having similar soils and due consideration be given to the need for scaling 

 to the new soils (p. 9, par. 7). Extrapolation to soils having widely differing 

 P'i>sical properties should onlv be attempted after laboratory confirmation of both 

 t'>e form and scale of the models. 



9 



