Extracting biophysical information from EO data
Biophysical parameters can provide information on the state, health and growth of surface vegetation such as Forests, croplands and grasslands. Monitoring crop health and production is necessary for food security planning on regional, national and global scales but can also be used to support farmers on a local scale as well as for the agro-inputs industry. Monitoring of forests and grassland is necessary for management of these biomes as well as for locating land use change and ecosystem change risk factors. Satellite measurements are key for making these observations over large scales.
The European Space Agency (ESA) Sentinel satellites provide observations of the land surface at high spatial resolutions. The optical, microwave and synthetic aperture radar sensors allow us to monitor land surface vegetation at up to 10 m resolution. These observations complement high temporal resolution satellites, like MODIS (which also has a long historical record); and commercial high-density satellites like planet labs, which produce very frequent, low accuracy, observations; among others.
Assimila works closely with University College London to use the latest Data Assimilation techniques to combine satellite observations of the land surface with radiative transfer and other physical models to provide best estimates of biophysical parameters. Data assimilation allows us to optimally combine multiple types of satellite observations together, with prior knowledge of the vegetation system and physically based models to monitor the land surface. Importantly Data assimilation also provides uncertainties on optimized parameters.
Assimila has worked on several projects with partners across Europe to develop these tools including for ESA and on European Union Horizon 2020 projects as well as applications projects working with partners in the UK and China. Applications crop modelling and yield prediction services, as well as a yield gap dashboard for individual farm planning and management.
Assimila is also working to incorporate biophysical parameter monitoring with crop modelling and weather forecasting for a comprehensive crop monitoring and modelling system.
Assimila Analysis Ready Data (ARD)
A typical workflow for any Earth Observation (EO) data user involves i) discovering a dataset, ii) downloading the corresponding image(s), iii) create a spectral and/or spatial subsets, and iv) perform any pre-processing needed. After all these steps, data is ready to be analysed. This is a major barrier to fully exploit EO data by any user, the Committee on Earth Observation Satellites (CEOS) has identified these limitations as a threat to major global and regional initiatives that rely on the successful use of EO data.
CEOS Analysis Ready Data for Land (CARD4L) are satellite data that have been processed to a minimum set of requirements and organized into a form that allows immediate analysis with a minimum of additional used effort and interoperability both through time and with other datasets. The CEOS CARD4L specification for optical surface reflectance applies to: Data collected with multispectral sensors operating in the VIS/NIR/SWIR wavelengths with a ground sample distance and resolution in the order 10- 100m however the specification is not inherently limited to this resolution.
Assimila as developed a method for generating coarse resolution optical surface reflectance ARD at 500m using the Moderate Resolution Imaging Spectroradiometer (MODIS) on board the Terra and Aqua platforms and the Ocean and Land Colour Instrument (OLCI) instrument on board Sentinel-3. The Assimila ARD fulfill all the CARD4L requirements, it was generated to provide a merged dataset using MODIS and OLCI data that can capture the land surface behavior and that is ready for any user to be accessed via Jupyter Notebooks.
True- and false-colour composites of BRDF isotropic parameters for MODIS-like bands 1,4,3 (RGB) and 7,2,1 (RGB) using MODIS and OLCI observations. Upper panel shows for Julian day 89 (March 30rd, 2018) and lower panel Julian day 273 (September 30th, 2018), left hand side images are the true colour, right hand side false colour.
Applying Earth Observation to assess UK land use change
The UK has to make annual submissions on its Greenhouse Gas (GHG) Inventory. GHG inventories estimate emissions and removals from each sector of the economy . In the Land use, land-use change, and forestry (LULUCF) sector, activity data are often areas or changes in area of land use categories over time (i.e. expressed as hectares per year). Assimila developed for the UK Department for Business, Energy & Industrial Strategy (BEIS) a method to estimate activity data for the LULUCF sector using coarse resolution (>250m) Earth Observation data for a long time series going back to 1990 and capable of being extended in future years.
A test area covering almost the whole UK (only excluding the far Eastern part), MODIS 500m and AVHRR 5km spatial resolution data were used to derive annual Land Cover / Land Use (LC/LU) products for 2011 and 2012.
Using completely independent reference data to validate the 500m LC/LU product derived using MODIS data an overall classification accuracy of 96.57% and 96.58% for 2011 and 2012 respectively was achieved with all individual accuracies better than 94%. Change detection overall accuracy was 95.66%. Hence the methodology using MODIS 500m data is considered feasible to track LC/LU change over time at this coarse resolution.
|True colour composites of BRDF isotropic parameters for MODIS bands 1,4,3 (RGB). On the left-hand side for Julian day 113 (April 23rd, 2011), on the right-hand side for Julian day 233 (August 21st, 2011).|
|Figure 6 Assimila LC/LU product derived using MODIS data for 2011 and 2012.|