University of Twente Student Theses

Login

Drought Severity : real-time evaluation of drought severity by means of Artificial Neural Networks and damage functions

Beltman, Mark (2020) Drought Severity : real-time evaluation of drought severity by means of Artificial Neural Networks and damage functions.

[img] PDF
11MB
Abstract:As the climate changes and thereby the climatic extremes intensify, droughts occur more frequently. This holds also true for the Vechtstromen region in the Netherlands. To minimize the socio-economic drought impacts to the Vechtstromen region, adequate and effective crisis management is required. Yet, a lack of quick and reliable information regarding the socio-economic drought severity, limits the effectiveness of the crisis response in mitigating societal impacts. Instead, crisis management is based upon solely hydrological drought indicators, like precipitation deficits and surface water levels, that are far from linearly related with the water use impacts. To improve drought management in the Vechtstromen region, a quick and easy real-time evaluation of the socio-economic drought severity is, therefore, desirable. Recently two tools have been developed that enable to evaluate the socioeconomic impacts of hydrological conditions quick and easily: the “Waterwijzer” Agriculture and the “Waterwijzer” Nature. Applying these tools to evaluate drought severity in real-time is, however, limited by a lack of groundwater data. Only point measurements are available, while real-time spatial groundwater patterns are required. From a literature study it was found that Artificial Neural Networks (ANNs) are likely the best way to interpolate the point measurements into spatial groundwater patterns with sufficient accuracy and speed. This research, therefore, aims to operationalize the socio-economic drought severity in real time, by using Artificial Neural Networks to obtain daily spatial groundwater data as an input for drought impact models. For this it has been studied if and how accurate ANNs can interpolate groundwater depths and if this accuracy is sufficient for drought severity evaluation. To study the ability of ANNs to accurately interpolate groundwater depths, two experiments have been setup: one in which the Vechtstromen region is interpolated by a single ANN and one in which two regional ANNs are used. This because the water systems of the northern and the southern region function differently. The northern region is predominantly a surface water controlled system, while the southern region is a free draining system. All three ANNs have been optimized individually by finding the optimal combination of input variables and number of hidden neurons. Their interpolation accuracy has subsequently been determined by testing the ANNs for an independent dataset that consisted of locations that were not used during model training and validation. From these experiments it is found that ANNs provide spatial groundwater depths with higher accuracy than the currently available alternatives that require longer calculation times. This conclusion holds true regardless of the type of hydrological system the interpolation relates to. The second major finding was that, although ANNs can cope with different types of hydrological systems separately, ANNs are not well able to distinguish between different functioning systems in a single ANN. Yet, despite this limitation also the single ANN, trained to interpolate the full Vechtstromen region by one model, outperformed the traditional methods. With these promising interpolation results, all elements to evaluate drought severity are in place. In the second research step, it has been studied if combining these elements results in sufficiently reliable severity evaluations, with a special focus on the effects of the uncertainty in the groundwater data to the severity evaluation. For this the socio-economic severity of 2019’s drought in the Vechtstromen catchment area has been evaluated (in a code green, yellow or red) at 72 drought sensitive locations. These evaluations have been performed for both the upper and the lower confidence limits of the groundwater depth predictions, to see how the uncertainty affects the severity evaluation. This study revealed that for none of the locations the difference between the upper and lower confidence limit was more than one colour code. Even more, at 58 locations the colour code evaluation was consistent. For five locations, located at the eastern Twente moraine, the plausibility of the severity evaluation is, however, questioned as here the ANN provides too shallow groundwater depths. Yet, these plausibility issues are not expected to affect the difference in severity evaluation between the upper and lower confidence level. Therefore, despite these plausibility issues, it is concluded that the groundwater depth predictions are sufficiently accurate to reliably evaluate socio-economic severity. With some minor improvements to the ANN for the eastern Twente moraine, the severity evaluation as presented in this report forms a solid basis to improve drought management. Nonetheless, there are also opportunities for further optimizations. Firstly, the informative strength of the severity evaluation to the drought management decision making process, can be enhanced when the severity evaluation links more closely to the qualitative drought severity definition, that is formulated in the first phase of this research project. For this more knowledge is required on the operationalisation of the qualitative drought severity definition in quantitative severity limits. Also for nature there needs to be found a way to separate the natural drought impacts from the human induced drought impacts. A second opportunity lies in providing drought severity predictions instead of evaluations. This will enable water managers to proactively mitigate drought severity. To enable severity predictions it is possible to combine the presented drought severity evaluation method with temporal groundwater depth predictions. The latter can be effectively done by ANNs. All in all, it can be concluded that the combination of ANNs and damage models holds a lot of potential to evaluate, or even predict, drought severity quickly and easily. Water managers are, therefore, advised to further develop and explore the application of ANNs to operationalise drought severity. This will help them to manage droughts more effectively by putting more focus on their core responsibility: facilitating water use.
Item Type:Essay (Master)
Faculty:ET: Engineering Technology
Subject:56 civil engineering
Programme:Civil Engineering and Management MSc (60026)
Link to this item:http://purl.utwente.nl/essays/86324
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page