Stochastic modelling of soil carbon stocks under different land uses: a case study in South Africa

• Main Author: Ros Mesa, Ignacio
• Other Authors: Rozanov, Andrei
• Format: Others
• Language: en_ZA
• Published: Stellenbosch : Stellenbosch University 2015
• Subjects: Soil carbon stocks; Carbon accounting; Soil modelling; UCTD
• Online Access: http://hdl.handle.net/10019.1/97097

Thesis (MSc)--Stellenbosch University, 2015. === ENGLISH ABSTRACT: The research was conducted in the Kwa-Zulu Natal midlands, South Africa. The vertical distribution of soil organic carbon (SOC) stocks were successfully predicted by stochastic exponential models developed for the three main land uses in the area, which are farmlands, forestry plantations and grasslands. These models, in combination with regular surface sampling, may be used for monitoring SOC dynamics in the area and mapping SOC stocks. Bulk density measurements are needed in combination with SOC content (%wt) to calculate such SOC stocks. Considering the disadvantages of bulk density sampling and measurement, an effort was made to determine if one of the commonly-used existing stochastic models could be used to successfully predict bulk densities for soils with known texture and SOC content to replace direct measurements, taking into account that different managements might affect final results. Statistica software was used to correlate the Saxton & Rawls model predictions and associated regressions with measured values for the study area. A clear distribution trend was achieved using Statistica and the correlations were fair with r2 values close to 0.5 for individual regressions and substantially higher for area averages. However, considering the depth-stratified averages and correcting for the effects of particle density changes for soils with high soil organic matter, high correlations for 2 of the 3 studied land uses were achieved (r2 values of 0.99 and 0.81 in forests and grasslands respectively). Therefore, although Saxton and Rawls (2006) predictions of bulk density may be used, it is preferable to conduct direct bulk density determinations. The proposed models to calculate the vertical distribution of SOC would substantially reduce the cost of soil carbon inventories to 1m soil depth in the study area by limiting observations to the soil surface. Triplicate 5cm-deep soil core samples would be collected at the soil surface per observation point for determination of ρb (bulk density) and Corg (organic carbon). On average, the accuracy of the normalized depth-distribution model is rather high for grasslands and forests/forest plantations (R2 = 0.98), but somewhat lower for cultivated lands (R2 = 0.96) due to mixing of the plough layer to cultivation depth. Carbon stocks to 1m depth were calculated as an integral of the normalized exponential distribution, multiplied by the value of Corg observed at the soil surface and expressed on volume basis as carbon density (Cv, kg∙m-3). The resulting stock assessment was compared to the observed values using piece-integration for sampled depth increments to give SOC stocks on an area basis (kg∙m-2). The estimated prediction error on average was 1.2 (9%) and 3.7 kg∙m-2 (21.6%) in grasslands and forests respectively, while for cultivated lands the error was 1.3 kg.m-2 (9.5%). Further improvement to reduce these errors may be achieved by introducing the soil type as variable and grouping the functions by soil type rather than land uses. The results of this work were presented at the seminar of the department of Soil Science, Stellenbosch University (Ros et al., 2014), the combined congress of the South African Soil Science, Horticulture and Agronomy societies (Rozanov et al., 2015), the First Global Soil Map conference, France (Wiese et al., 2013), the 20th International Congress of Soil Science, Korea (Wiese et al. 2014) and were submitted for publication in Geoderma special issue dedicated to digital soil mapping of soil organic carbon following the presentation at the 20th ICSS, Korea (Wiese et al., 2014). === AFRIKAANSE OPSOMMING: Hierdie navorsing is in die Kwa-Zulu Natalse middellande van Suid-Afrika gedoen. Die vertikale verspreiding van grondorganiese koolstof (GOK) is suksesvol voorspel deur middel van stogastiese eksponensiële modelle wat vir die drie hoof landsgebruike ontwikkel is. In kombinasie met roetine monsterneming by die grondoppervlak kan hierdie modelle suksesvol aangewend word vir die monitering van GOK dinamika in die studiegebied, sowel as kartering van GOK voorraad. Bulkdigtheidsmetings word tesame met GOK inhoud (%massa) benodig om die GOK voorraad te bereken. Weens die nadele van monsterneming vir bulkdigtheidsbepalings is ‘n poging aangewend om te bepaal of een van die mees algemeen gebruikte bestaande stogastiese modelle (Saxton & Rawls 2006) gebruik kan word om die bulkdigtheid van gronde suksesvol vanaf tekstuur en GOK inhoud te voorspel en sodoende direkte metings te vervang. Statistica sagteware is gebruik om die voorspellings met behulp van die Saxton & Rawls modelle en gevolglike regressies met gemete waardes vanuit die studiegebied te korreleer en ‘n duidelike verspreidingstendens is hierdeur opgelewer. Die korrelasies vir individuele regressies was redelik met r2 waardes naby 0.5 en merkwaardig hoër waardes vir area gemiddeldes. Hoë korrelasies is egter behaal vir 2 van die 3 bestudeerde landsgebruike (r2 waardes van 0.99 en 0.81 in bosbou en grasveld onderskeidelik) wanneer die gemiddelde dieptestratifikasies gebruik en gekorrigeer word vir die verandering in deeltjiedigtheid vir gronde met hoë grondorganiese material. Alhoewel die Saxton and Rawls (2006) voorspellings van bulkdigtheid gebruik kan word, behoort bulkdigtheidsbepalings egter verkieslik direk gedoen te word. Die voorgestelde modelle vir die bepaling van vertikale GOK verspreiding tot 1m gronddiepte sou die koste van grondkoolstof opnames in die studiegebied dramaties verlaag deur grondmetings tot die grondoppervlak te beperk. Grondmonsters sal in triplikaat per waarnemingspunt met 5cm diep silinders op die grondoppervlak geneem word vir ρb (bulkdigtheid) and Corg (organiese koolstof) bepalings. Die gemiddelde akkuraatheid van die genormaliseerde diepteverspreidingsmodel is hoog vir grasveld en woude/bosbou plantasies (R2 = 0.98), maar ietwat laer vir bewerkte landerye (R2 = 0.96) as gevolg van die vermenging van die ploeglaag tot op die diepte van bewerking. Koolstof voorraad tot 1m gronddiepte is bepaal deur middel van die integraal van die genormaliseerde eksponensiele verspreiding, vermenigvuldig met die waarde van Corg op die grondoppervlak en op ‘n volume basis uitgedruk as koolstofdigtheid (Cv, kg∙m-3). Die gevolglike voorraadopname is met gemete waardes vergelyk deur middel van ‘n stuksgewyse integrasie van die gemonsterde diepteinkremente om GOK voorraad per area (kg∙m-2) te lewer. Die gemiddelde geskatte fout van voorspelling was 1.2 (9%) en 3.7 kg∙m-2 (21.6%) in grasveld and plantasies onderskeidelik en 1.3 kg.m-2 (9.5%) in bewerkte landerye. Verdere verbetering van die modelle en ‘n verlaging in hierdie foute kan verkry word deur die grondtipe inligting as veranderlike in te bring en die funksies volgens grondtipe eerder as landsgebruik te groepeer. Resultate van hierdie werk is reeds aangebied tydens ‘n seminar by die department Grondkunde, Stellenbosch Universiteit (Ros Mesa et al., 2014), die gesamentlike kongres vir die Suid-Afrikaanse Verenigings vir Grondkunde, Hortologie, Onkruidwetenskap en Gewasproduksie (Rozanov et al. 2015), die Eerste Global Soil Map konferensie, Frankryk (Wiese et al, 2013), die 20ste Internasionale Grondkunde Kongres, Korea (Wiese et al. 2014) en is ingehandig vir publikasie in ‘n spesiale uitgawe van Geoderma wat, na aanleiding van die aanbieding by die 20ste Internasionale Grondkunde Kongres, Korea (Wiese et al., 2014), fokus op digitale grondkartering van grondorganiese koolstof.