Available online at www.sciencedirect.com

ScienceDirect

Procedía Social and Behavioral Sciences 2 (2010) 7763-7764

Sixth International Conference on Sensitivity Analysis of Model Output

Preliminary results for the global sensitivity analysis of SALTIRSOIL model outputs

Fernando Viscontia*, José Miguel de Pazb, José Luis Rubioa, Juan Sáncheza

aCentro de Investigaciones sobre Desertificación-CIDE (CSIC, UVEG, GV), Camí de la Marjal s/n, 46470 Albal (Valencia) Spain bInstituto Valenciano de Investigaciones Agrarias-IVIA (GV), Crta. Moncada-Náquera km 4.5, 46113 Moncada (Valencia) Spain

Abstract

SALTIRSOIL is a model for the prediction of soil salinity, sodicity and alkalinity in irrigated well-drained lands. These three characteristics are respectively assessed through the electrical conductivity and the sodium adsorption ratio of the soil saturation extract (ECse and SARse), and the pH of the soil saturated paste (pHsp). A global sensitivity analysis (GSA) was carried out to ascertain what input variables are more influential on these three outputs. The standardised regression coefficients of the linear regression analyses were used to calculate sensitivity measures. The irrigation water quality represented by ECiw and SARiw is the most influential factor on salinity and sodicity calculation, i.e. ECse and SARse respectively, while the carbon dioxide partial pressure so is on alkalinity (pHsp). Next there are the variables featuring the soil water balance: rainfall, average annual basal crop coefficient and reference evapotranspiration.

Keywords: SALTIRSOIL model; global sensitivity analysis; soil; salinity; sodicity; alkalinity

1. Main text

Soil salinisation is one of the main desertification processes decreasing agricultural productivity in lands under arid, semi-arid and dry-subhumid climates. Identification of areas under risk of salinisation is an important task for soil and water conservation purposes. This identification can be made through measurement or modelling provided the models are validated. SALTIRSOIL (SALTs in IRrigation SOILs) (Visconti et al., 2010) is a new model aimed at predicting soil salinity, sodicity and alkalinity in irrigated well-drained lands. The main factors affecting soil salinity, sodicity and alkalinity in such lands can be arranged in four classes: climate, soil, crop and irrigation. SALTIRSOIL uses basic data already available or that can be easily obtained by means of regular land surveys. The standards to assess soil salinity, sodicity and alkalinity are respectively, the following characteristics of the soil saturation extract: electrical conductivity at 25°C (ECse), sodium adsorption ratio (SARse), and pH of the soil saturated paste (pHsp).

The aim of this work is to present the preliminary global sensitivity analysis (GSA) carried out to find what input variables are more influential on SALTIRSOIL outputs, i.e. ECse, SARse and pHsp.

The GSA was done according to a Factors' Prioritisation Setting (Saltelli et al., 2004), with the 16 + 3 input variables shown in table 1. A Monte Carlo experiment with 250 trial sets for the evaluation of SALTIRSOIL was

* Corresponding author. Tel.: +34-961-220-540; fax: +34-961-270-967. E-mail address: fernando.visconti@uv.es.

1877-0428 © 2010 Published by Elsevier Ltd. doi:10.1016/j.sbspro.2010.05.217

Fernando Visconti et al. / Procedia Social and Behavioral Sciences 2 (2010) 7763-7764

devised. Two-hundred and fifty random values were independently calculated for each variable according to normal marginal distributions featured by the means and standard deviations shown in table 1. Each model run took 2.2 s. A linear regression analysis (LRA) for each one of the outputs ECse, SARse, and pHsp was tried. According to the results of the LRAs the importance of the input variables could be calculated on basis the standardised regression coefficients (SRCs). The SRCs were squared, divided by the sum of squares and multiplied by 100 to obtain a percent measure of sensitivity.

Table 1. Statistical summary of the variables used for the GSA of SALTIRSOIL.

Class Variable (abbreviation) / units Mean St. D. Max. Min.

Climate Rainfall amount (R) / mm year-1 450 120 719 88

Reference evapotranspiration amount (ET0) / mm year-1 1200 150 1537 780

Frequency of rainfall (fR) / day year-1 70 20 121 18

Soil Clay content (clay) / g (100g)-1 36 11 70 0

Sand content (sand) / g (100g)-1 25 7 42 6

Stone content (stone) / g (100g)-1 15 5 28 3

Calcium carbonate content (ECC) / g (100g)-1 50 12 85 18

Soil Organic Matter content (SOM) / g (100g)-1 2.0 0.8 4.5 0.2

Gypsum content (gypsum) / g (100g)-1 0.40 0.15 0.76 0.01

Carbon dioxide in saturated paste (log pCO2) -3.00 0.20 -2.42 -3.57

Root depth (RD) / cm 100 10 130 70

Crop Average annual basal crop coefficient (Kcb) 0.8 0.2 1.25 0.19

Percent of shaded soil (SS) 74 9 100 50

Irrigation Irrigation water amount (I) / mm year-1 700 100 1001 443

Frequency of irrigation (fI) / day year-1 40 10 71 10

Percent of wetted soil (WS) 70 9 98 44

Electrical conductivity (ECiw) / dS m-1 4.0 0.74 6.1 2.1

Sodium adsorption ratio (SARiw) / (mmol L-1)1/2 6.6 2.3 11.5 0.8

pHiw 7.76 0.31 8.63 6.83

The coefficients of determination (R2) obtained in the LRAs were 0.86, 0.90 and 0.97 for ECse, SARse and pHsp respectively. According to these high R2 SALTIRSOIL can be regarded as a monotonic model for the calculation of ECse, SAR^ and pHsp. Therefore the sensitivity analysis can be based on the SRCs (Saltelli et al, 2004). The input variables can be ordered from highest to lowest influence on the output ECse as follows: R ~ ECiw > Kcb > ET0 > I ~ RD ~ WS ~ sand ~ SOM ~ ECC ~ fR > clay > fI > SS > gypsum > logpCO2 ~ stone. ECiw and R explain 38% and 36% respectively of the variance of the output ECse, then Kcb explains 16%, and the rest of variables explain the remainder variance (10%) starting from ET0 (2.8%). The input variables can be ordered according to their influence on SARse as follows: SARiw > R ~ Kcb > ET0 > I ~ RD ~ sand ~ WS ~ SOM > fR ~ ECC > sand > fl

> SS ~ logpCO2 ~ stone > gypsum. SARiw explains 55% of the variance of the output SARse, next the same input variables as with ECse but with lower percents of explained variance: R (18%), Kcb (13%) and ET0 (8%). Finally in the case of pHsp the input variables can be ordered as follows: logpCO2 > R ~ Kcb > ET0 > RD ~ fR ~ sand ~ I ~ WS

> gypsum > sOm > stone ~ fI ~ pHiw > clay ~ SS ~ ECC. Carbon dioxide pressure (log pCO2) explains 94% of the variance of the output pHsp. Next there are the same variables as with ECse and SARse but with even lower percents: R (2.8%), Kcb (2.0%) and ET0 (0.4%). The pHiw and ECC have practically no influence on the pHsp: less than 0.02%. The most influential input variables on soil salinity and sodicity calculation are, on the one hand the salinity and sodicity of irrigation water (ECiw and SARiw), and on the other the variables featuring the soil water balance: rainfall (R), average annual basal crop coefficient (Kcb), and reference evapotranspiration (ET0).

The preliminary GSA of SALTIRSOIL model has provided the relative importance of the input variables on the outputs ECse, SARse and pHsp.

2. References

Saltelli A., Tarantola S., Campolongo F., Ratto M. 2004. Sensitivity Analysis in Practice. John Wiley & Sons. London.

Visconti F., de Paz J.M., Rubio J.L., Sánchez J. 2010. Development of SALTIRSOIL: a simulation model for the mid to long term prediction of soil salinity in irrigated well-drained lands. Agricultural Water Management (under review)..