Hip – Historical data – machine learning and uncertainty
Description
The scaled-down model calculations provide a nationwide overview of the depth of ground water in high spatial resolution (10x10 meters) for a typical winter situation and a typical summer situation based on data for the period 1/1 1990 – 31/12 2019. The depth is indicated in centimeters.
For each situation, an uncertainty band has been calculated covering an 80 % confidence interval.
The model calculations can be used as a basis for assessing the uncertainty of the 10 m machine learning models when comparing results for winter-q50, winter-q10 and winter-q90 (ditto for summer).
Specification for scaled-down modelling:
The depth of ground water is calculated in 10 m spatial resolution.
The two models are trained against 75 % of data and 25 % of data is used for validation.
A means absolute error is used in the training process.
The parameters of the machine learning model were optimised with an automated trial-and-error method.
The same parameter set is used for both models, winter and summer.
The model calculations will be useful to use as a starting point and basic data for assessments where the spatial variation of the off-road groundwater is in focus e.g. planning of urban development, LAR, planning of infrastructure, risk assessment of construction, design of drains and drainage systems, etc.
Risk assessment for soil and groundwater pollution, agriculture and other land use, etc.
Uncertainty specification:
Uncertainty of depth maps to ground-level groundwater scaled down from 100 to 10 m grid represents an 80 % confidence interval of the most likely value.
Instead of a mean absolute error, as used in training the summer/winter models, uncertainty models use a Quantile objective function.
To represent an 80 % confidence interval, 10 and a 90 % percentile respectively are used.
Each layer of uncertainty has its own model, which provides 4 models in total, q10 summer, q10 winter, q90 summer and q90 winter.
All models use it together parameter sets, as used in the summer/winter models.
The model calculations could usefully be used as a starting point for studies where the spatial variation of the off-road groundwater is in focus.
Resources
| Name |
Format |
Description |
Link |
|
0 |
|
https://api.dataforsyningen.dk/hip_dtg_10m_100m?service=WMS&request=GetCapabilities&token= |