Modelling Soil Erosion, Flash Flood Prediction and Evapotranspiration in Northern Vietnam
by Hong Quang Nguyen
Date of Examination:2016-02-17
Date of issue:2016-03-08
Advisor:Prof. Dr. Martin Kappas
Referee:Prof. Dr. Martin Kappas
Referee:Prof. Dr. Ralph Mitlöhner
Persistent Address:
http://dx.doi.org/10.53846/goediss-5552
Files in this item
Name:Diss_Nguyen_Hong_Quang_2016.pdf
Size:8.95Mb
Format:PDF
Description:PhD dissertation
The following license files are associated with this item:
Abstract
English
Water-induced soil erosion (WSE) is a main factor of land degradation in many parts of the world and reported as a main threat to agriculture compared to the second largest risk by wind. Some parts of Vietnam have been suffering WSE rates of over 50 t ha year−1 (t ha−1 y−1), (particularly in the North) which has negative effects on the agriculture. In addition, Vietnam is a developing country and most of the resident livelihoods are based on agriculture. However, due to a lack of information about both spatial and rates of WSE in the region, the soil erosion mitigation efforts seem to be inadequate. Furthermore, far too little attention has been paid to soil erosion modelling in the tropics in general and in the North of Vietnam in particular. In the first part of this research, surface runoff investigation and WSE evaluation were targeted employing the hydrological modelling methodology and the so-called “regionalization approach” for generating the results of calibrated watersheds to uncalibrated ones. This makes a regional scale (for the SWAT application) from watershed scale. The KINEROS2 model was also used for modelling WSE at finer event-based and watershed scale. In the results of model calibration and validation, both the SWAT and KINEROS2 presented their capabilities to generate simulated discharge matching closely to observed data. Although the mean estimated WSE rate was 4.1 t ha−1 y−1, approximately 15% of the Yen Bai province was computed at the rate of 8.5 t ha−1 y−1. Attention was given to the changes in land use/land cover (LULC) conditions (2002-2009) which have had a growth of the WSE rate from 0.2 to 3.3 t ha−1 y−1 in some areas of the province. This scenario was also found in the results of the KINEROS2 model but for the LULC conditions in 2002 and 2007.The KINEROS2 parameter sensitivity tests indicated that the model’s outputs were very sensitive to the antecedent soil moisture condition ($\theta$ant) and the hydraulic conductivity (Ksat). This reveals a need of estimates $\theta$ant for later applications of the model such as for flash flood (FF) prediction. Flash flooding is responsible for severe loss of life and property in lots of countries. Increases of the surface runoff not only speed up the erosive processes, but also intensify the FF risk. Many parts of Vietnam have been confronted with increasing FF consequences but the situation is much worse in northern Vietnam. Due to the fact that FF often occur in small streams and are linked to short, but heavy rains, much previous research has suggested methods to identify FF occurrences early in order to mitigate their impacts. Approaches with assembled and coupling hydrological models were used for the aim of FF prediction. The assembly of the SWAT, BEACH and KINEROS2 models filled up the hole of lacking $\theta$ant and defined well the boundary conditions for the KINEROS2’s runs. The model sensitivity tests played a crucial role due to its shortening the model calibration and validation processes. To implement the latter method, the results of the KINEROS2 models serve the HEC-RAS inputs such as the hydrographs, river depths, initial flow and Manning’s n coefficient. What is interesting is that I also used several rainfall sources (satellite, radar, NWP, and gauge) to test and compare their abilities of application with the aim of FF forecast. With the calibrated parameters the KINEROS2 model computed the river discharge (Q) fitting well to observed data (average NSE ≈ 0.78, R2 ≈ 0.93). The daily soil moisture calculated by BEACH was very helpful for the assembly because changing of the $\theta$ant, Ksat, and N varied the model outputs dramatically. Remarkably, KINEROS2 predicted the Q (in streams and overland) using GSM and HRM rainfalls revealed a good possibility to predict the time, magnitude and location of approaching FFs. The most surprising result is that the use of radar rainfall produced less accurate Q compared to the use of satellite precipitation. The results of coupling the KINEROS2 and HEC-RAS models provide a more in-depth analysis of FF behaviour based not only on river discharge but also on flood water level (WL) or stage, flow velocity (FV) and power (P) at river cross sections (RSs). First, the models were calibrated and validated for four rainfall events for the Q and WL with satisfactory results (mean NSE ≈ 0.85, R2 ≈ 0.91 for the Q, and mean NSE ≈ 0.82, R2 ≈ 0.90 for the stage). A comparison between the rating curves and river banks showed the stream flow was approximately two metres over the banks during the rain on 23rd June 2011 at the outlet of the Nam Kim e.g. relationships between FV and channel slope, between top width, flow area and FV were analysed in detail. The most striking result to emerge from the HEC-RAS forecasted outputs is that the predicted Q and WL agreed basically with the in situ measurements but there have been some false/missing alarms. There is also much valuable discussion on uncertainty, methodical efficiency. The last objective is focusing on modelling evapotranspiration (ET). The ET is considered to have played a crucial role in the hydrological cycle linking as well to the above-mentioned issues of WSE and flash flooding. The modelled ET data were compared to MODIS ET. The BEACH’s parameters were calibrated for the 2001-2004 periods and validated for 2005-2012 periods using observed evaporation. Although the MODIS ET was higher than the SWAT and BEACH ET, a general fine agreement between them was found based on both monthly and yearly ET. A slightly downward trend of all ET in the 2001-2012 periods has been shown in the trend analysis. However, longer investigation of trend analysis might be needed to verify this trend (40 years for example).
Keywords: Soil erosion; Flash flood prediction; Hydrological modelling; Evapotranspiration; GIS; Remote sensing; Yen Bai, Vietnam; Surface runoff