Informations et ressources scientifiques
sur le développement des zones arides et semi-arides

Accueil du site → Doctorat → Australie → Application of high resolution data sources via data fusion processes in deriving a comprehensive drought index

Monash University (2016)

Application of high resolution data sources via data fusion processes in deriving a comprehensive drought index

Azmi, Mohammad

Titre : Application of high resolution data sources via data fusion processes in deriving a comprehensive drought index

Auteur : Azmi, Mohammad

Université de soutenance : Monash University

Grade : Doctor of Philosophy (PhD) 2016

Résumé
The main purpose is to develop a new methodology to address existing shortcomings in spatio-temporally evaluating water stress conditions by employing individual drought indices for a specific location. The proposed methodology is able to monitor water stress conditions of terrestrial ecosystems by objectively linking different aspects of the ecosystem such as water availability and vegetation conditions. The developed index, called Data Fusion-based Drought Index (DFDI), makes use of advanced statistical methods, and also considers eco-meteorological characteristics, such as landuse, land-cover, and climate of an area to determine the water stress conditions at each time step for each specific location. The capabilities of the DFDI are demonstrated comprehensively making use of data from three OzFlux tower sites in Australia, as well as satellite data to develop a spatially distributed weekly water stress evaluation framework across Victoria, Australia. To achieve this, observations from those OzFlux Tower sites were complemented with nearby synoptic stations, while the spatial data where obtained from the Australian Water Availability Project, as well as the Soil Moisture and Ocean Salinity and the MODIS-Terra satellite missions.One of the main advantages of the spatial DFDI methodology is to present explicit mathematical equations, ultimately making the process of spatial water stress monitoring at current and future time steps more user-friendly and consequently more accessible for the industry. To achieve this objective, the area is first regionalized according to the two criteria pairs of wetness/dryness and active/non-active vegetation via a K-mean clustering method using mean monthly normalized data of precipitation and the vegetation cover fraction, respectively. Then, for each sub-region of the wet/dry (active/non-active) map, a mathematical equation is developed by employing the Symbolic Regression Method in which the dependent variable is Standardized Aggregated Water Availability Index (or Standardized Aggregated Vegetation Index) and independent variables are selected amongst individual drought indices of the water content (or vegetation conditions) cluster. The final derived mathematical equations have generally shown promising functionalities and accuracies, especially in appropriately detecting the spatial trend of the extreme events. As the methodology was developed using short in situ datasets, a sensitivity analysis has been implemented to investigate the issues arising from this for the performance of DFDI. A particular focus was put on the statistical characteristics of such data, specifically how the short-term observations of the three focus sites compared against long-term model predictions. The rationale behind this approach is that while ground observations are generally sparse both in space and time, and satellite data are also often not covering sufficiently long periods, models provide the only consistent long-term reference for climate-related studies. As an alternative, short-term observations could potentially provide the same statistical characteristics as long-term observations would. Thus, the 3-year dataset from the three OzFlux towers was compared against a long-term dataset (1911-2016) from the AWAP in terms of both their stationary and dynamic statistical characteristics. The results show that observations and model displayed similar statistical features of long-term water conditions despite their different lengths, suggesting that both may be used for validation purposes.

Mots clés : Water stress monitoring Data fusion Australia

Présentation

Version intégrale

Page publiée le 1er août 2017