Uncertainty reduction of soil moisture simulations based on combined improvement approach of model structure and parametrization

Soil moisture has gained importance in recent years mainly for water management purposes. The main aim of this thesis was to improve soil moisture simulations based on the improvement of the model structure and parameterization. The model was used as a tool to assess the efficiency of irrigation management in two surface irrigated maize fields in northern Italy. This research work included both field and laboratory experiments together with modeling activities. The first part of PhD activities was dedicated to the improvement of soil moisture simulations through the implementation of modes to solve the infiltration process. FORTRAN 90- VADOSONE PROGRAM was developed within which: Philip’s equation, modified SCS-Curve number method, Green Ampt method and Ross 2003 model as a numerical solution of Richards equation, were implemented. These methods aim at quantifying the infiltration in the vadose zone that have been tested by several researchers under different assumptions (ponded or not ponded conditions, steady or unsteady rainfall, layered or non-layered soils). For Green-Ampt the selected algorithm follows the method of Mein and Larson (1973) to estimate the infiltration under unsteady rainfall. Ross 2003 model, a non-iterative fast solution for Richards equation, was modified and included within the subset of infiltration module. The developed VADOSONE PROGRAM was included in a further step within FEST-WB distributed hydrological model. Within Ross 2003 model different soil water retention curve models were implemented Brooks and Corey (1964), Van Genuchten-Mualem (1976) and modified Van Genuchten (Vogel et al., 2001). Sensitivity analysis of input parameters of these equations was carried out in order to understand which are the most important parameters for model calibration. The calibration and validation of soil moisture simulations based on the different implemented infiltration models were carried out for Livraga’s study site for 2012 as a calibration year and 2014 for the validation. Results showed that all tested methods when properly calibrated gave quite similar performances and sufficient accuracy. The advantage of ROSS solution is that it allows following the variation of soil moisture along the soil profile that was not possible with the other analytical methods. From the sensitivity analysis we concluded that soil hydraulic parameters and in particular the saturated hydraulic conductivity have a great importance for the accuracy of the results of soil moisture simulations. So we decided to investigate more deeply the accuracy of implemented parameter retrieval methods. Adding to that, when we carried out the calibration of the VADOSONE PROGRAM for 2012 cropping season, we found that achieving a good calibration with the same set of soil hydraulic parameters during the entire cropping cycle was not possible since part of the simulation period was under or overestimated that is due probably to the temporal dynamic of these parameters. In order to confirm or deny this hypothesis, we decided to assess these parameters temporal as well as spatial variability during a cropping cycle. Noting that only few studies have addressed this variability. The second part of this PhD was about testing different methods of soil hydraulic properties measurements and estimations, as well as assessing their spatial and temporal variability due to a possible effect of the crop growth or agricultural practices on these properties. For the determination of the saturated hydraulic conductivity parameter, several field campaigns within which field measurements were performed using double ring infiltrometer and Guelph permeameter, also intact and disturbed soil samples were collected in order to measure the bulk density, organic matter content and particle size distribution. Laboratory measurements of saturated hydraulic conductivity were carried out using KSAT-UMS device following falling and constant head methods. We compared different methods for the KSAT determination using in-situ and laboratory measurements (falling head and evaporation method) and assessed the applicability of HYPRES and ROSETTA pedotransfer functions (PTFs ) for the KSAT estimation. The results showed that the Guelph permeameter underestimated the KSAT in comparison to the double ring and falling head measurements. The ROSETTA and HYPRES PTFs showed quite similar performances while their easiness and convenient use make them potential alternative techniques for the KSAT estimation in comparison to on-field and laboratory measurements. The KSAT is sensitive to the selected determination method that yielded different results when implemented as input for hydrological model simulations. For the determination of soil water retention and conductivity curves, the evaporation method experiments using HYPROP-UMS device were performed for each soil sample. Fitting procedure of the results of these experiments was carried out following Brooks and Corey (BC) and Van Genuchten (VG) models. Several pedotransfer functions were tested: seven of them that fit the BC and four for VG using CalcPTF (Guber et al., 2010). Results of these tested PTFs were compared to evaporation method results. Together with these tested regressional PTFs, ROSETTA (Schaap et al. 2001) that is based on ANN (artificial neural network) was also used to predict soil parameters. A multimodal approach was applied based on simple averaging of outputs of simulations based on both direct and indirect determined soil water retention and conductivity curves allowed an improvement of soil moisture simulations results. The third step was dedicated to assess the temporal and spatial variability of soil hydraulic properties over a cropping cycle and to quantify their effect on soil water movement simulations. Monitoring of soil properties was carried out during three measuring campaigns during a cropping cycle for a Maize field located in Secugnago. Results of this study showed that soil surface is subjected to temporal variation during the cropping cycle due to: drying/wetting cycles, fertilization, roots development, irrigation, tillage and harvest. The saturated hydraulic conductivity was a sensitive parameter to temporal changes. This parameter is required as input for many ecological, environmental and agricultural models, so this variation should be taken into consideration. At farm scale, the comparison of spatial and temporal variations of measured soil properties showed that temporal changes are much important. Results illustrated the importance of considering time and spatial variability of soil properties while modeling soil water movement. Considering these parameters as static in time yielded inaccurate results. Though soil hydraulic properties estimation is expensive and time consuming, better understanding of these parameters dynamics during cropping cycles would improve soil moisture predictions especially for precision agriculture applications. The last part of these PhD activities was devoted to the use of VADOSONE-PROGRAM and FEST-WB as tools to assess the efficiency of previous irrigation management practices through a calculated stress index. Simulations were performed for Livraga’s study site for previous cropping season of 2012 and Secugnago site for the year of 2015. For the second study site we considered stress threshold and surplus threshold as variable in time since the temporal variability of soil properties has been proven previously. Simulations were carried out on the ROSS (2003) solution and modified-SCS-CN infiltration models. It has been proven that depending on the selected infiltration model and if the temporal variation of stress and surplus thresholds were taken into consideration or neglected impacted the evaluation results. The implementation of Ross model as a numerical solution of Richards equation allowed us to get an idea about the water status within the soil profile. The numerical solutions of Richards equation are not usually used within decision tools due to their limitations (Time consuming and mainly presenting convergence problems). Implemented within the VADOSONE PROGRAM, Ross solution allowed us to assess water stress conditions at different depths. Such an evaluation is important if the simulations of soil moisture are implemented for precision agriculture practices. The developed VADOSONE-PROGRAM can be implemented also a decision support tool for irrigation scheduling for future projects. The spatial variability of soil moisture and stress index was assessed through FEST-WB simulations. Results proved that the use of distributed models provided more information about the spatial variability of water depletion and stress in the field thus the use of local simulations can lead to inadequate decisions.

Il contenuto d’acqua del suolo è una variabile fondamentale per la corretta gestione delle risorse idriche, specialmente per applicazioni nel settore agricolo. L’obiettivo principale di questa tesi è stato quello di migliorare la simulazione del contenuto d’acqua del suolo investigando sia la struttura dei modelli matematici di infiltrazione, sia i metodi per la definizione dei parametri dei modelli stessi. Il lavoro ha compreso sia campagne di misure in sito che esperimenti in laboratorio oltre allo sviluppo di modelli matematici. Nella prima parte della tesi è stato sviluppato il modello matematico VADOSONE che implementa diverse equazioni per la modellazione dell’infiltrazione dell’acqua nel suolo: l’equazione di Philip, il metodo SCS-CN modificato, l’equazione di Green Ampt e la soluzione dell’equazione di Richard proposta da Ross (2003). Quest’ultimo metodo, in particolare, è una soluzione non iterativa molto efficiente e stabile che permette di simulare il profilo di umidità del suolo a diverse profondità. Dopo una analisi di sensitività dei diversi metodi implementati, i valori dei parametri più significativi, prima fra tutti la permeabilità del suolo a saturazione, sono stati calibrati usando le misure di umidità del suolo alla stazione di Livraga. I risultati hanno mostrato come con un unico set dei parametri non fosse possibile ricostruire correttamente l’andamento dell’umidità del suolo per tutto il periodo, facendo ipotizzare l’esistenza di una variabilità temporale delle caratteristiche del suolo. La seconda parte della tesi è stata dedicata alla sperimentazione di diversi metodi, in sito ed in laboratorio, per la valutazione dei parametri del suolo e valutare la loro variabilità spaziale e temporale. In questa analisi sono state considerate anche alcune funzioni di pedotrasferimento disponibili in letteratura. I risultati hanno mostrato come alcuni metodi di misura forniscano errori sistematici nella valutazione dei parametri, a conferma di quanto noto in letteratura, e come uno schema di simulazione multi-modello sia efficace nella riduzione dell’incertezza della simulazione dell’umidità del suolo. Nella terza parte è stata analizzata la variabilità spaziale e temporale dei parametri del suolo ed il loro effetto sulla simulazione dell’umidità del suolo. I risultati hanno mostrato come il suolo superficiale sia soggetto a una significativa variabilità dovuta sia alle pratiche agricole che allo sviluppo delle radici e della componente biotica e come non tenere in considerazione tale variabilità possa introdurre una elevata incertezza nella simulazione dell’umidità del suolo. Nell’ultima parte della tesi, gli strumenti modellistici sviluppati sono stati usati sia alla scala locale che quella spazialmente distribuita, per la valutazione dell’efficacia dell’irrigazione nel mantenimento di uno stato ottimale delle colture. I risultati hanno mostrato come l’approccio spazialmente distribuito fornisca una informazione molto più dettagliata sulla possibile insorgenza di condizioni di stress per la vegetazione e sia quindi più idonea alla gestione dell’irrigazione.