Presentation by Guillaume Lacombe, Somphasith Douangsavanh, Richard Vogel, Matthew McCartney, Yann Chemin, Lisa Rebelo, Touleelor Sotoukee at the International conference “Sustainability in the Water-Energy-Food Nexus” 19-20 May 2014, Bonn, Germany
Stunning ➥8448380779▻ Call Girls In Panchshil Enclave Delhi NCR
Risk management along the Mekong tributaries
1. Uniting agriculture and nature for poverty reduction
to predict flow metrics for water resource and risk
management along the Mekong tributaries
Guillaume Lacombe, Somphasith Douangsavanh, Richard Vogel, Matthew McCartney,
Yann Chemin, Lisa Rebelo, Touleelor Sotoukee
Simple power-law models
International conference “Sustainability in the Water-Energy-Food Nexus” 19-20 May 2014, Bonn, Germany
2. Uniting agriculture and nature for poverty reduction
Introduction
• Increased vulnerability to stream-flow
variability,
• Prediction models are:
– complex (use by practitioners is limited),
– mainstream-focused (away from poorest
populations),
– Physically-based (assumed physical
processes, high data requirement).
One dot = 1 village
3. Uniting agriculture and nature for poverty reduction
Objectives
• To define simple relationships to
predict flow metrics from
catchments characteristics in the
Lower Mekong River
• To assess the effect of land-
cover (forest, paddy, wetlands)
on the flow metrics and
downstream water resources
4. Uniting agriculture and nature for poverty reduction
• Multivariate power-law equation to predict flow (Q) from
catchment characteristics (Xi)
m
mXXXQ 210
21exp
• Logarithmic transformation solved by weighted least
square regression (multiple linear regressions)
)ln(...)ln()ln()ln( 22110 mm XXXQ
Method
5. Uniting agriculture and nature for poverty reduction
Method
• Daily flow metrics
– 11 flow percentiles, annual mean, min
and max
– Data from Mekong River Commission
(MRC): 65 gauging stations with 1 to 41
years of daily record
• Catchment characteristics
– rainfall, geomorphology, geography,
soil, land-cover (forest, paddy and
wetlands)
– Data from Aphrodite, HydroSHEDS,
MRC
6. Uniting agriculture and nature for poverty reduction
Method
• Selection of variables:
– Combined use of “best- subsets” and
“step-wise” regressions
– P-value < 0.05
– Constraints:
• Homoscedasticity of residual,
• Independence of variables,
• Outliers removed (Cook D)
• Leave-one-out cross-validations
to maximize the prediction R-
squared
14. Conclusions
• Highly-predictive & simple tools to
assess high and low flows in
ungauged areas
– water resources planning, flood risks
assessment, hydropower potential, storage
design
• A range of applications
– Assessment of effect of paddy area
expansion on downstream low-flow
– Prediction of climate change impact on
basin water yields
Notes de l'éditeur
This work was produced as part of a WLE-funded project which aims to assess the role of ecosystems in flow regulation in the Mekong and Volta Basins (attenuating floods and mitigating droughts).
Because of population increase and more water needs, people live closers to river to have an easier access to water resources. As a consequence, people are more exposed to flood risks and but also to water shortage because of the increased water demand. Adaptation measures are hampered by uncertainties in forecasted flows. A number of hydrological models have been developed for the Mekong Basin to predict stream-flow variability. But they remain under-used because of their complexity and lack of transparency. Only modelling experts can use them. In addition most of the models aimed at predicting flow along the Mekong Mainstream, precluding accurate assessments in headwater catchments where populations are exposed to floods and droughts.
To provide simple mathematical tools to predict various flow metrics from easily obtained catchment characteristics. To assess how different types of ecosystems alter downstream flow and thus, water resources.
We used power-law equations, already used in many parts of the world to assess Q from several catchment characteristics X1. The logarithmic transformation of this equation is a linear relationship that can be solver by multiple linear regressions. We will now see how to determine which catchments characteristics have the greatest explanatory power and to determine the corresponding coefficients of the equation.
Power-law models were calibrated using flow data and catchments characteristics from a set of 65 gauged catchments in the Lower Mekong Basin. 14 flow characteristics were calculated in each catchment, from daily time series of flow recorded over many years. They capture the magnitude of high, median and low flows. 17 different catchment characteristics were computed for each catchment, mainly from 3 data sources: the DEM Hydrosheds, the gridded rainfall database Aphrodite and the Mekong River Commission. In particular, the percentage of surface area covered by 3 land-cover types were included as potential explanatory variables.
Selection of best set of explanatory variables among the 17 guided by combined use of two different selection algorithms.
A variable was considered to have explanatory power when her p-value is below 0.05. Overall performance of the model measured by leave-one-out cross validations, summarized by the prediction r-square.
Table shows result of MRA and provide values of coefficients associated to each explanatory variable that should be used in power-law model to predict each of the 14 flow metrics listed in the first column.
This example illustrates how to use the table to predict the fifth percentile of flow. It shows that only 3 catchment characteristics have been selected as explanatory variables among the 17. Other variables have a p-value greater than 0.05, meaning that there is significant chance that they have no explanatory power.
Annual rainfall is an explanatory variable in all models with associated coefficients exhibiting the lowest variability between models (variation coefficient < 10%). Values are much greater than unity (average=2.59) indicating the highly non-linear response of streamflow to precipitation. The rainfall coefficient associated to the model predicting mean annual flow (2.543) corresponds to the rainfall elasticity of streamflow. These elasticity coefficients can help assessing the impact of projected changes in rainfall on future changes in the studied stream-flow metrics.
Among the 3 land-use types that we tested, only 2 of them significantly explain downstream flow. Wetlands were found not to have any significant influence on downstream hydrology. The surface ratio of paddy rice is negatively correlated to four low-flow variables (0.60, 0.70, 0.80 and 0.95). One possible explanation is the ability of paddy fields to reduce groundwater recharge due to the impermeable soil layer below the rice root zone, which contributes to the maintenance of ponded water in the bunded rice fields and increased evapotranspiration (Bouman et al. 2007). However, removing paddy variable from the explanatory variables reduce explanatory power by less than 1 point on average (0.60: 0.55%, 0.70: 0.57%, 0.80: 1.17%, 0.95: 6.36%). The coefficient related to forest is more questionable and would require further research for a scientific-based explanation.
The drainage area is an explanatory variable for mean annual flow and high-flow variables (Max, 0.10, 0.20, 0.30 and Mean). The coefficients for this variable are slightly lower than 1, depicting a slight tendency for reduction in runoff depth as catchment size increases. This is in agreement with Pilgrim et al. (1982) who observed a tendency of increased seepage in larger catchments. In contrast, low-flow variables (0.40, 0.50, 0.60, 0.70, 0.80, 0.90, 0.95 and Min) are better explained by the catchment perimeter rather than the catchment area. The perimeter provides information related to the shape of the catchment. For a given catchment area, a greater perimeter implies a longer time for water to reach the catchment outlet, thus explaining the positive correlation with low flow variables.
Prediction R-squared values are excellent (>90%) for most flow variables. But generally lower for low-flow metrics. The better prediction of high flow, compared to low flow, indicates that the explanatory variables tested in this analysis (mainly geomorphological and climate characteristics) do not correspond to the catchment characteristics that predominantly control low flows. Similar contrast between the predictive power of high-flow and low-flow models has been observed under various hydrological conditions (Thomas and Benson 1975), suggesting that more efforts are needed to generate catchment characteristics suitable for multivariate low flow predictions.
Possible differences between observed (Qobs) and predicted (Qpred) flow in specific catchments indicate that the model results should be treated with caution. 30% and 50% of the catchments, respectively, have ANE greater than 40%. These errors do not necessarily result from model deficiencies and could reflect inaccuracies in the original flow values used in the model parameterization. Therefore, ANE provides an overestimation of prediction errors