and therefore the calculation could differ according to available and used soil ... The simple regression model could be accurate enough using just actual soil.
European Scientific Journal December 2013 /SPECIAL/ edition vol.3 ISSN: 1857 – 7881 (Print) e - ISSN 1857- 7431
SIMPLE REGRESSION MODELS FOR PREDICTING SOIL HYDROLYTIC ACIDITY
Zdenko Loncaric, PhD Vlado Kovacevic, PhD Domagoj Rastija, PhD Krunoslav Karalic, PhD Brigita Popovic, PhD Vladimir Ivezic, PhD Zoran Semialjac, B.Sc Josip Juraj Strossmayer University of Osijek, Faculty of Agriculture in Osijek, Croatia
Abstract Soil acidity is global factor limiting soil fertility of about 40% of the cultivable land which are acid. The common liming recommendations are based on different soil properties and therefore the calculation could differ according to available and used soil data. The aim of this paper was to determine the suitability of simple regression models for prediction soil hydrolytic acidity for precise liming recommendation using just actual (pHH2O) and exchangeable soil pH (pHKCl) and humus content as basic soil data. These agrochemical analyses were done on basic set of 2600 soil samples and on validation set of 375 soil samples. The simple regression model could be accurate enough using just actual soil pH only for soils with lower humus content. Model accuracy increases including more soil data in prediction model, starting from adding soil exchangeable pH and then including humus data. Because of possible high soil sample variance, the best simple models are model including both actual and exchangeable soil pH, and humus, but with different regression equation for each range of soil pH or/and for each range of humus content. These kinds of models are sensitive to soil cation exchange capacity, humus content, texture and soil acidity, indicating that model adjustment to soil types could result in increasing model accuracy. The model error correlate to humus content and soil acidity, and the lowest model error were about 14% in average for soil pHKCl 4-5, and 16% for soil pHKCl 4 and 3-4%, ME was 22,9 and 25,5%, respectively), but correlation of predicted and measured hydrolytic acidity was very significant (Graph 1, Figure C). Table 5. Regression parameters (Y = Hydrolitic acidity = Intercept + AX1 + SX2 + HX3) after validation and decrease of model error (ME) based on the Y = I + AX relation Model Intercept pHH2O pHKCl Humus (%) ME ME decrease equation (A) (S) (H) decrease (%) A375 20,356 -2,64 0 0 S375 18,156 -2,87 0,12 8,9 AS375 19,193 -0,61 -2,32 0,15 11,2 ASH375 16,690 -0,47 -2,83 +1,786 0,45 33,3
However, further ME decreasing and model improvement, were made by splitting validation set into 4 groups according to soil pHKCl (Table 6). Model errors in all 4 groups were decreased using ASH375pH model comparing to ASH375 model. ME decreasing was 11,1% up to 44,2% (Table 6). This approach reduced ME for 375 samples in validation set on 16,7 % with lowest error with pHKCl 4-5 (ME 14,2%) and pHKCl < 4 (16,7%), and correlation of predicted and measured Hy was higher than for model ASH375 (Graph 1, Figure D). Table 6. Regression parameters of ASH375pH model after splitting samples into 5 groups according to soil pH Model equation Intercept pHH2O pHKCl Humus (%) ME ME decrease pHKCl range (A) (S) (H) decrease (%) < 4,0 57,583 0,493 -15,232 +3,158 0,40 21,1 4,0-5,0 15,975 0,082 -3,366 +1,621 0,08 11,1 5,0-5,5 9,281 1,485 -2,969 +0,120 0,46 44,2 > 5,5 15,067 -0,885 -1,424 +0,095 0,23 30,7
The ASH375pH model sensitivity on humus content is quite high and pH sensitive, since humus content difference of 4 % (1,01% vs. 5,01%) resulted in increasing predicted hydrolytic acidity for example 12,63 cmol/kg (20,97 – 8,34), presuming no changes in soil pH and if soil was very acid (pHKCl < 4): 57,583 + 0,493 × 4,57 (A) – 15,232 × 3,59 (S) + 3,158 × 1,01 (H) = 8,34 cmol/kg 57,583 + 0,493 × 4,57 (A) – 15,232 × 3,59 (S) + 3,158 × 5,01 (H) = 20,97 cmol/kg. Simultaneously, if soil was slightly acid (pHKCl > 5), humus content difference of 4 % will resulted in increasing predicted hydrolytic acidity for example only 0,48 cmol/kg (3,23 – 2,75). 176
European Scientific Journal December 2013 /SPECIAL/ edition vol.3 ISSN: 1857 – 7881 (Print) e - ISSN 1857- 7431
9,281 +1,485 × 5,70 (A) – 2,969 × 5,09 (S) + 0,120 × 1,01 (H) = 2,75 cmol/kg 9,281 +1,485 × 5,70 (A) – 2,969 × 5,09 (S) + 0,120 × 5,01 (H) = 3,23 cmol/kg. A)
Analysed vs. predicted (ASH2600 model) values of hidrolytic acidity 9,0
y = 0,001x3 - 0,0487x2 + 0,8572x + 1,0993 R² = 0,5611, n = 375
7,0
7,0
Hy (Predicted values)
Hy (Predicted values)
8,0
y = 0,6853x + 1,4879 R² = 0,707, n = 2600
8,0
Analysed vs. predicted (ASH2600 model) values of hidrolytic acidity of validation set (375 samples)
B)
6,0 5,0 4,0 3,0
6,0 5,0 4,0 3,0 2,0
2,0
1,0
1,0 0,0
0,0 0,0
2,0
4,0 6,0 Hy (Analysed values)
8,0
10,0
Analysed vs. predicted (ASH375 model) values of hidrolytic acidity of validation set (375 samples)
C) 16,0
0,0
10,0 15,0 Hy (Analysed values)
20,0
25,0
Analysed vs. predicted (ASH375 pH model) values of hidrolytic acidity of validation set (375 samples)
D) 25,0
y = -0,0284x2 + 1,1739x R² = 0,7355, n = 375
14,0
5,0
y = 0,9686x R² = 0,7798, n = 375
Hy (Predicted values)
Hy (Predicted values)
20,0 12,0 10,0 8,0 6,0
15,0
10,0
4,0 5,0 2,0 0,0
0,0 0,0
5,0
10,0 15,0 Hy (Analysed values)
20,0
25,0
0,0
5,0
10,0 15,0 Hy (Analysed values)
20,0
25,0
Graph 1. Regression of measured (analytical) and predicted (model) hydrolytic acidity for: A) prediction model ASH2600 for 2600 samples, B) the same ASH2600 model for new validation set of 375 samples, C) corrected prediction model ASH375 after model validation, D) prediction model ASH375 corrected for four different class of soil pH
Conclusion: Basic agrochemical soil data (actual and exchangeable soil pH, and humus) could be enough for prediction of soil hydrolytic acidity using simple regression model. The simple regression model could be accurate enough using just actual soil pH only for soils with lower humus content. Model accuracy increases including more soil data in prediction model, starting from adding soil exchangeable pH and then including humus data. Because of possible high soil sample variance, the best simple models are model including actual and exchangeable soil pH, and humus, but with different regression equation for each range of soil pH or/and for each range of humus content. These kinds of models are sensitive to soil cation exchange capacity, humus content, texture and soil acidity, indicating that model adjustment to soil types could result in increasing model accuracy. The model error correlate to humus content and soil acidity, and the lowest model error were about 14% in average for soil pHKCl 4-5, and 16% for soil pHKCl