Using SMAP-data and Ensemble Kalman filtering to improve local soil moisture estimates in the Netherlands.
In January 2015 I started an internship at hydrologic. This internship is part of my graduation and I will be writing my final thesis during this period. If everything goes as planned I should be finished mid-July of this year.
I first met HydroLogic on a symposium of the Dutch waterboards on the use of remote sensing data by the waterboards. We investigated a possible internship and a few weeks later I had my first interview in Delft. I was immediately excited when I heard that Hydrologic was described as a hydro-informatics company and the broad range of models and computer-related things they do. I was always very much interested in these kind of things and felt that HydroLogic and I would make a good match. And indeed: we found a perfect topic for me to work on. It includes large amounts of data, remote sensing, some mathematics, a lot of modelling and most important huge amounts of water!
In 2012 the WaterSense project in the Hunze area in the Netherlands was completed. In this project a reliable groundwater and surface water model was developed for prediction of the soil moisture in the area. The area is about 300 km2 and the predictions have to be on plot scale. For this purpose a highly detailed (25 x 25 m) SimGro model was created and tested.
SimGro is a hydrologic model that can be used to simulate situations when shallow ground water levels occur. It includes a simple surface water model and can be combined with MODFLOW for groundwater calculations. The SWAP or MetaSWAP top part of the model takes the vegetation-atmosphere interactions into account.
Comparison of the soil moisture estimates to the actual soil moisture at over 50 locations showed that the model performed poorly. The model does not cover the spatial variability well and the errors made in soil moisture calculation at each of the grid cells are too big. A call for improvement!
The use of more remotely sensed data as input can improve the model performance. It can also be used to validate model results spatially without the need for many ground measurements. Furthermore it can be applied in a data assimilation process to update and improve the model output. The Soil Moisture Active Passive (SMAP) mission, which was recently launched and will provide soil moisture data on a 3 x 3 km, 10 x 10 km and 40 x 40 km grid, opens up a new path of approach for the soil moisture prediction in the Netherlands.
The research I am working on during my intern at HydroLogic aims to answer the question:
Can data assimilation of SMAP-data be used to improve small scale soil moisture estimates in the Netherlands?
As data assimilation technique I make use of the Ensemble Kalman filter. The data that will be assimilated are not the actual SMAP data. This research focuses on the use of remotely sensed data in three ways:
- Radar based rainfall data will be input for the model. In this way a more spatially varied forcing is created.
- Actual evaporation calculations based on remote sensing data can be used to validate the evaporation calculated by the model.
- Soil moisture data based on the SMAP-mission used for data assimilation. Note that this is not actual remotely sensed data, but it serves the same purpose.
It is expected that so the spatial variability will be enhanced and that the additional methods of checking lead to new insights and make it possible to improve the model before the output. Finally it is aimed at to improve by the means of data assimilation and comparison between the three different products of SMAP will be made.
Dave de Koning, TU Delft