IOP Conference Series: Earth and Environmental Science PAPER OPEN ACCESS Graphical approach to assess the soil fertility evaluation model validity for rice (case study: southern area of Merapi Mountain, Indonesia) To cite this article: E A Julianto et al 2018 IOP Conf. Ser.: Earth Environ. Sci. 129 012012 View the article online for updates and enhancements. This content was downloaded from IP address 148.251.232.83 on 12/10/2018 at 15:45
Graphical approach to assess the soil fertility evaluation model validity for rice (case study: southern area of Merapi Mountain, Indonesia) E A Julianto¹,2, W A Suntoro 3, W S Dewi 3 and Partoyo 2 ¹Doctoral Program of Agricultural Science, Graduate School, Sebelas Maret University, Surakarta Jl. Ir. Sutami No. 36A Surakarta, Indonesia 2 Dept. Of Agrotechnology, Universitas Pembangunan Nasional Veteran Yogyakarta, Jl. SWK 104 Lingkar Utara Condong Catur Yogyakarta, Indonesia 3 Dept. of Soil Science, Fac. of Agriculture, Sebelas Maret University, Jl. Ir. Sutami No. 36A Surakarta, Indonesia Abstract. Climate change has been reported to exacerbate land resources degradation including soil fertility decline. The appropriate validity use on soil fertility evaluation could reduce the risk of climate change effect on plant cultivation. This study aims to assess the validity of a Soil Fertility Evaluation Model using a graphical approach. The models evaluated were the Indonesian Soil Research Center (PPT) version model, the FAO Unesco version model, and the Kyuma version model. Each model was then correlated with rice production (dry grain weight/gkp). The goodness of fit of each model can be tested to evaluate the quality and validity of a model, as well as the regression coefficient (R²). This research used the Eviews 9 programme by a graphical approach. The results obtained three curves, namely actual, fitted, and residual curves. If the actual and fitted curves are widely apart or irregular, this means that the quality of the model is not good, or there are many other factors that are still not included in the model (large residual) and conversely. Indeed, if the actual and fitted curves show exactly the same shape, it means that all factors have already been included in the model. Modification of the standard soil fertility evaluation models can improve the quality and validity of a model. 1. Introduction Climate change is known to cause diminishing land resources including soil fertility deterioration. Soil as the source of life, is the most vital and valuable natural resource which is not renewable quickly. Soil fertility is a dynamic natural property that can change the impact of natural and human-derived factors [1]. Soil, being the natural medium for plant growth has a direct impact on yield and quality of crops growing on it. Measurement of the fertility of an agricultural soil informs a lot of about the potential productivity. Fortunately, producers can control fertility by managing the plant s nutritional status [2], and good management on fertility requires a good fertility evaluation model. Speaking of models, one imagines something intricate, complex, and difficult, and thus this research is an attempt to bridge those knowledge gap. Any input from a farming operation is aimed at stepping up the welfare of the farming community. Choosing an evaluation model highly correlated with crop productivity is a rational step for efficiency and effectiveness in planning soil fertility management. This is an important effort to ensure that a set of recommendations on fertilizing from the research results can be restricted in terms of its area in the form of a map based on the homogeneity of the Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. Published under licence by Ltd 1
biophysical and socio-economic characteristics of a farming operation [3]. The soil fertility evaluation models evaluated were the PPT (Indonesian Soil Research Center) version model, the FAO Unesco version model, and the Kyuma version model. Each model was correlated with wet field rice production (dry grain weight). There are several indicators that can be used/seen to infer that the quality of a model is good. In the language of statistics, it is to test a model s goodness of fit. After fitting a regression model, it is important to determine whether all the necessary model assumptions are valid before performing inference. If there are any violations, subsequent inferential procedures may be invalid resulting in faulty conclusions. Therefore, it is crucial to perform appropriate model diagnostics. In constructing our regression models, we assumed that the response y to the explanatory variables were linear in the parameters and that the errors were independent and identically distributed normal [4]. Model diagnostic procedures to analyze the quality of a model can be seen from the regression coefficient (R²) and a graphical approach. The objective of this research is therefore to assess the validity of a soil fertility evaluation model using a graphical approach in the southern area of Merapi Mountain, Indonesia. 2. Research method The research was conducted on the southern slope of Mount. Merapi (Cangkringan, Pakem, Turi, Tempel, and Seyegan subdistricts) of Sleman Regency, the Province of the Special Region of Yogyakarta, Indonesia. Research stations were scattered on the southern slope of Merapi Mountain (37 experiment stations) which were planted with rice (Oryza sativa) of the Ciherang variety. The research was carried out from April 2015 to April 2016. The observation was done by taking soil fertility samples in points (37 stations) determined through stratified purpose sampling. The main basis used was a wet field land (at least half-irrigated) on land map units and its area. Each point of observation was in the fieldwork map whose geographical position (coordinates) had been determined, then traced in the field, and its sample of soil fertility was taken. The soil samples were then analyzed in the laboratory to ascertain the nutrient status covering C-org, N-total, ph H 2 O, ph KC1, P 2 O 5, K 2 O, P-Bray, Morgan K₂O, Ca, Mg, K, Na, Ca-total, Mg-total, Cation Exchange Capacity (CEC), Si, Texture, Alkali Saturation, P-retention, and P-Olsen. As for the component of crop production, the observation was carried out based on the dry weight grain. The statistical analyses used Eviews 9. Regression and correlation analyses was employed on soil fertility parameters in respect with crop production. Regression was done in single and multiple ways. By using Eviews 9, a good quality model could be seen from indicators R² and a graphical approach. A high R² did not always show a good quality model. From the graphical approach, a graphic with three curves was obtained, namely data regression curve, the actual data, and residual data. If the data regression curve and the actual data coincide with each other, the model is considered to have good quality as there are only a few external factors that affect Y. 3. Results and discussion 3.1. Pre-indicators to assess the validity of a soil fertility evaluation model The results of regression analysis of standard and modified PPT, respectively, are presented in Table 1. Table 1 shows that on the CEC log the significant prob at a reliability level of 99%, on the regression coefficient of CEC if the CEC increased by 1%, the unit (dry grain weight) of rice production would increase by 2.04%, with the assumption that the other factors remained constant (cateris paribus). The total prob. in standard PPT =0.0469 means that the reliability level=99.9531, whereas in the total prob. of modified PPT =0.0034 means that the reliability level= 99.9966. This shows that modified PPT is more reliable. 2
Table 1. Regression results of standard PPT and modified PPT. Standard PPT Modified PPT Variable Coefficient Prob. Variable Coefficient Prob. C -11277.41 0.0283 C -4.97 0.1842 LOG(CEC) 2.04 0.0006 LOG(CEC) 1.36 0.0024 LOG(CORG) 1.82 0.0009 LOG(CORG) 1.34 0.0010 LOG(KB) 1.14 0.0454 LOG(KB) 6.95 0.0871 LOG(P2O5) 3.84 0.4511 LOG(BRAY1)^2 1.03 0.0000 LOG(K2O) 4.74 0.4167 LOG(K2O) 3.42 0.4193 When compared with a higher R² ex: R²=0.82, the total prob. that affects significantly can be greater although per a particular parameter e.g. from 2.04 to 1.3, this is not applicable in Table 1 Equation as shown in Table 1 on the left is : GKP = -11277.41 + 2041.68*LOG(CEC) + 1820.78*LOG(CORG) + 1138.23*LOG(KB) + 383.96*LOG(P2O5) + 474.10*LOG(K2O) (1) Regression coefficient (R²) = 0.603 While equation in Table 1 on the right is : GKPLOG = 1.08 + 0.02*LOG(CEC)^2 + 0.09*LOG(CORG) + 0.05*LOG(KB)+0.01*LOG(BRAY1)^2+ 0.03*LOG(K2O) (2) Regression coefficient (R²) = 0.82 The results of regression analysis of standard and modified FAO, respectively, are presented in Table 2. Table 2. shows that total prob. in standard FAO=0.0469 means the reliability level=99.9531, whereas the total prob. of modified PPT means the reliability level=99.9966 so that it can be inferred that the modified PPT was more reliable. In the standard FAO, there were 5 significant parameters while in modified FAO there were 6 significant parameters. In terms of its R² the modified FAO R²=0.903 whereas in the standard FAO R²=0.82. The equation as shown in Table 2 on the left is : GKP=-16098.05+1695.53*LOG(CEC)+1663.19*LOG(CORG)+ 367.94*LOG(K)+323.07*LOG(OLSENP2O5)+513.78*LOG(P-RETENTION)+1581.70* LOG(CA1)+940.08*LOG(MG1)-98.71*LOG(K2O)+ 306.29*LOG(P2O5) (3) While equation in Table 2 on the right is : GKPLN=3.37+0.001*(LN)(CEC)^2+0.11*(LN)(CORG)^2+1.27*(LN)K+0.001*(LN)BRAY1-0.0001*(LN)RETENSIP+2.56e-05*(LN)CA1+0.0003*(LN)MG1-0.0007*(LN)MORGANK2O + 0.0004*(LN)P2O5 (4) Table 2. Regression results of standard FAO and modified FAO. Standard FAO Modified FAO Variable Coefficient Prob. Variable Coefficient Prob. C -16098.05 0.0032 C -11160.00 0.0235 LOG(CEC) 1.70 0.0007 LOG(CEC) 1.51 0.0005 LOG(CORG) 1.66 0.0001 LOG(CORG) 1.46 0.0001 LOG(K) 1.37 0.0004 LOG(K) 1.10 0.0022 LOG(OLSENP2O5) -3.23 0.3133 LOG(BRAY1) 4.05 0.0158 LOG(RETENSIP) 5.14 0.2176 LOG(RETENSIP) 1.48 0.7065 LOG(CA1) 1.58 0.0002 LOG(CA1) 1.14 0.0047 LOG(MG1) 9.40 0.0239 LOG(MG1) 7.72 0.0386 LOG(K2O) -9.87 0.8302 LOG(MORGANK2O) -1.98 0.3871 LOG(P2O5) 3.06 0.4171 LOG(P2O5) 1.19 0.7338 3
The results of regression analysis of standard and modified Kyuma, respectively, are presented in Table 3. Table 3. Regression results of standard Kyuma and modified Kyuma. Standard Kyuma Modified Kyuma Variable Coefficient Prob. Variable Coefficient Prob. C 1.98 0.5788 C 0.97 0.0012 LOG(CORG) 1.26 0.0022 LOG(CORG) 0.09 0.0005 LOG(N) 7.23 0.1956 LOG(N) 0.05 0.1167 LOG(P2O5) -1.45 0.7034 LOG(P2O5) -0.01 0.5770 LOG(BRAY1) 5.34 0.0018 LOG(BRAY1) 0.03 0.0025 LOG(CAMG) -3.43 0.3654 LOG(CAMG) -0.02 0.2680 LOG(K) 4.93 0.1674 LOG(K) 0.04 0.0650 LOG(CEC) 1.35 0.0047 LOG(CEC) 0.08 0.0032 LOG(SI) 2.53 0.0967 LOG(SI) 0.06 0.3746 LOG(PASIR) 1.26 0.1241 LOG(PASIR) 0.09 0.0390 LOG(KB) 0.05 0.0228 LOG(RETENSIP) 0.008 0.7139 Table 3 shows that total prob. in standard Kyuma=0.0087 means that the reliability level=99.9913, whereas in the total prob. of modified Kyuma 0.0062 means that the reliability level=99.9938 so that it can be inferred that modified PPT was more reliable. In terms of the high R² ex: R²=0.896 there was a more significant prob. (5 parameters) whereas in standard Kyuma there were 3 parameters. In term of its R² the modified Kyuma=0.896 whereas the standard Kyuma R²=0.84. The equation as shown in Table 3 on the left is : GKP=1984.29+1261.61*LOG(CORG)+723.08*LOG(N)-144.59*LOG(P2O5)+ 533.72*LOG(BRAY1)-343.19*LOG(CAMG)+492.60*LOG(K)+1350.66*LOG(CEC)+ 2533.54*LOG(SI) + 1258.80*LOG(PASIR)...... (5) While equation in Table 3 on the right is : GKP(LN)=3.09+0.21*(LN)CORG+1.16*(LN)N-0.0002*(LN)P2O5+0.002*(LN)BRAY1-0.01*(LN)CAMG+0.61*(LN)K+ 0.018*(LN)CEC + 0.29*(LN)SI + 0.003*(LN)PASIR + 2.15e-05*(LN)KB + 9.56e-05*(LN)P-RETENTION.... (6) 3.2. Analysis of model quality with graphical approach The results of regression analysis of standard and modified PPTis presented in Table 4. Table 4. Regression results of standard PPT and modified PPT Evaluation Model R² Prob. (F-statistics) PPT/SRC LOGARITHM (LOG 10) Standard PPT (Fig. 1 left) 0.473 0.000007 GKP LOG, LOG (CEC)^2 and LOG (BRAY1)^2 (Fig. 1 right) 0.815 0.000000 4
9,000 1.95 8,000 1.90 2,000 1,000 7,000 6,000 5,000 4,000.08.04 1.85 1.80 1.75 1.70 1.65 0.00-1,000 -.04-2,000 -.08 Figure 1. Quality of PPT model using graphical approach. Two-dimensional plots of residuals versus fitted values or predictors are traditionally used to assess lack of fit of a regression model. The general idea is that if the model is correct then the sample residuals should appear independent of the predictors, with allowance for typically negligible dependence caused by substituting estimates for parameters [5]. Table 4 shows that the high regression coefficient (0.815) is followed by simultaneous influence/less Prob (F-statistic) (greater significance) at 0.000000. Conversely, the low/small regression coefficient (0.473) is followed by simultaneous influence/greater Prob (F-statistic) (less significance) at 0.000007. this is confirmed in Figure 1 where the actual and fitted curves are distantly spaced (wide apart) or irregular, indicating that the quality of the model is not good, or there are many other factors that have not been included in the model (larger residual curves) as shown in Figure 1 on the left. Conversely, if the actual and fitted curves coincide with each other the quality of the model is good, or other factors that have not been included in the model are few (small residual curves). Indeed if the actual and fitted curves became just one curve it means that there are no other factors that have not been included in the model, as seen in Figure 1 on the right. Table 5. Regression results of standard FAO and modified FAO. Evaluation Model R² Prob. (F-statistics) FAO Standard FAO (Fig. 2 left) 0.440 0.000019 GKP2 (LN) (CEC)^2 and (LN) (CORG)^2 (Fig. 2 right) 0.903 0.000000 2,000 1,000 0-1,000 9,000 8,000 7,000 6,000 5,000.10 4,000.05.00 -.05 -.10 4.6 4.4 4.2 4.0 3.8-2,000 -.15 Figure 2. Quality of FAO model with a graphical approach. 5
Table 5 shows that the higher regression coefficient(r²= 0.903) is followed by simultaneous influence Prob (F-statistic), with greater significance at 0.000000. Conversely, the low regression coefficient (R²) (0.440) is followed by simultaneous influence/more Prob (F-statistic) (less significance) at 0.000019. This is to confirm Figure 2, where the actual and fitted curves are distantly spaced (wide apart) or irregular indicating that the quality of the model is not good, or there are still many other factors that have not been included in the model (large residual curve) as shown in Figure 2 on the left. Conversely, if the actual and fitted curves coincide with each other, the quality of the model is good, or other factors that have not been included in the model are few (small residual curve); indeed, if the actual and fitted curves become just one curve, it means that there are no other factors that have not been included in the model, as shown in Figure 2 on the right. Table 6. Regression results of standard Kyuma and modified Kyuma. Evaluation Model R² Prob. (F-statistics) Kyuma OM (Fig. 3 left) GKP (LN) and without coefficient (+ (LN) P-RETENTION and (LN) KB (Fig. 3 right) 0.122 0.896 0.042645 0.000000 Table 6 shows that the high regression coefficient (R²= 0.896 is followed by simultaneous influence Prob (F-statistic) (greater significance) at 0.000000. Conversely, the low regression coefficient (R²= 0.122 is followed by simultaneous influence/more Prob (F-statistic) (less significance) at 0.042645. This is to confirm Figure 3, where the actual and fitted curves are distantly spaced (apart) or irregular indicating that the quality of the model is not good, or there are still many other factors that have not been included in the model (large residual curve) as shown in Figure 3 on the left. Conversely, if the actual and fitted curves coincide with each other, the quality of the model is good, or other factors that have not been included in the model are few (small residual curve); indeed, if the actual and fitted curves become just one curve, it means that there are no other factors that have not been included in the model, as shown in Figure 3 on the right. 3,000 2,000 1,000 9,000 8,000 7,000 6,000 5,000 4,000.08.04.00 1.95 1.90 1.85 1.80 1.75 1.70 1.65 0 -.04-1,000 -.08-2,000 -.12 Figure 3. Quality of Kyuma model with graphical approach 4. Conclusion A high regression coefficient (R²) is followed by simultaneous influence/less Prob (F-statistic) (greater significance) and there are more parameters involved. If the actual and fitted curves are distantly spaced (wide apart) or irregular, the quality of the model is not good, or there are many other factors that have not been included in the model (large residual curve). Conversely, if the actual and fitted curves coincide with each other, the quality of the model is good, or other factors that have not been included in the model are few (small residual curve); indeed, if the actual and fitted curves become just one curve, it means that there are no other factors that have not been included in the model. Modification of the standard Soil Fertility Evaluation can improve the quality of an evaluation model. 6
The resulted model by graphical approaching could be used to evaluate soil fertility, so it can reduce the risk of climate change effect on plant cultivation. Acknowledgement Dedicated to Ministry of Research, Ministry of Technology and Higher Education, Republic of Indonesia, for funding this study. References [1] Kavitha C and Sujatha M P 2015 Evaluation of soil fertility status in various agroecosystems of Thrissur District, Kerala, India. Int. J. Agric. Crop Sci. 8 pp 328 38 [2] Flynn R, Shane T B and Baker RD 2004 Sampling for Plant Tissue Analysis Guide A-123 (New Mexico: College of Agriculture and Home Economics New Mexico State University) p 92 [3] Dobermann A et al. 2003 Estimating indigenous nutrient supplies for site-specific nutrient management in irrigated rice Agronomy Journal 95 pp 924-35 [4] Model Diagnostics for Regression www.stat.columbia.edu 2017 [Online] Available: www.stat.columbia.edu/~martin/w2024/r7.pdf [Accessed: 27-08-2017] [5] Pardoe I and Cook R D Graphical Method for Assessing the Fit of the Logistic Regression Model http://citeseerx.ist.psu.edu 2002 [Online] Available: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.138.4256&rep=rep1&type=pdf [Accessed: 27-08-2017] 7