Monitoring of grass cutting with Sentinel-1 time series pilot results and vision for operational service based on big data tools and cloud computing Kaupo Voormansik 21st MARS Conference November 25, 2015 - Thessaloniki Tartu Observatory s SAR team Applied research for grassland parameters retrieval since 2011, 3 campaigns, 5 research papers. Main research partner Main user partner 1
The problem According to risk analysis 0.83% of grasslands in Estonia violate the cutting requirement. In Estonia there are 471 000 ha grasslands, the violation results in 483 470 of false payments. Infeasible to do total checks with inspectors. If we had tools to check it, the money could be saved and re-distributed from sofa-farmers to actual active farmers. Vision for the solution Copernicus Sentinel-1 time series to detect cutting/grazing in the parcels. Parcel borders from paying agency databases. S1 time series SAR satellite data analysis Parcel borders Updated GIS data with cutting info Paying agency GIS database 2
27/11/2015 Why Sentinel-1 data? Synthetic Aperture Radar, weather independence continuous regular monitoring. Free and open data policy input data available at zero cost. Image source: European Space Agency, 2015 NASA MODIS Terra image, July 31, 2015 3
27/11/2015 Sentinel-1 SAR image, July 31, 2015 MAY 2017 Sentinel-1 data takes at Northern Europe (60ᵒ latitude) S1B (A) S1B (A) S1B (A) S1B (A) S1B (A) 8 4
MAY 2017 Sentinel-1 data takes at Central Europe (48ᵒ latitude) S1B (A) S1B (A) 9 What we did the experiment setup Sentinel-1 time series IW mode, VV/VH pol. 20 m by 5 m resol. 3 different acq. geometries. May to August 2015. Total of 32 images. Careful field surveys Wet and dry biomass. Grass height and cutting status. Soil moisture. In situ photographs. Weekly monitoring. Transect method, 10 data points per field. 5
What we did the data processing Studying backscatter, interferometric and polarimetric SAR parameters. Taking account sensor specific noise effects and meteorological data. Applying the accumulated experience of past 5 years and two PhD theses. 2015 experiment results Very encouraging results. We can tell the grassfield cutting status and likely cutting dates with high reliability. Can t discover the details yet, will be published in peer-reviewed journals in 2016-2017. 6
27/11/2015 Pilot results delivered to Estonian paying agency Pilot results delivered to Estonian paying agency 7
Side-product of the research. Detection of the ploughed and non-ploughed fields in spring. Estimation of ploughing date. In all weather conditions. Sentinel-1 vs high resolution SAR data (e.g. COSMO SkyMED, TerraSAR-X) Main limitation of S1: only the spatial res. works on parcels with minimum diameter of 70 m. High res. SAR data can monitor parcels with minimum diameter of 20 m or less, but: The grass cutting detection reliability of S1 is even higher thanks to the temporal resolution dense time series! Cost and coverage advantage of S1 data. 8
OK, Sentinel-1 and -2 loads of nice and useful data but how to handle this? Future of EO Data processing Challenge by the amount of data for (a) data transfer and (b) processing power. Sentinel-1 EU covering data volume 390 TB/year (depending on latitude 150-250 repeated coverages per year). Sentinel-2 EU covering data volume 610 TB/year (depending on latitude 100-180 repeated coverages per year). Infeasible to download all this to every user, who needs the info in the data. 9
EO Community Platforms Server side in cloud User side (desktop, notebook, tablet, smart phone) * Web-based or other user interface S1, S2 and other satellite data EO and GIS software Powerful computing resource (several multicore CPUs, GPU-based computing, memory [128+ GB], large disk space) Other GIS vector and raster data * Webserver or other IS * Transfer of light-weight data mainly (text, final raster or vector map, remote desktop) Source: European Space Agency s Ground- Segment Evolution Strategy Document 2015 19 Paying agency subsidy claims check service architecture concept EO Community platform Paying agency s info systems and databases EO and GIS software Powerful computing resource (several multicore CPUs, GPUbased computing, memory [128+ GB], large disk space) S1, S2 and other satellite data Other GIS vector and raster data Parcel borders Updated GIS data with checks results Specific software architecture in each country 10
Announcement Establishment of TO spin-off company, start-up KappaZeta OÜ. Concentrating on paying agency checks applications and SAR-based precision agriculture technology. Contact: http://www.kappazeta.ee and info@kappazeta.ee KappaZeta OÜ Free limited area grass-cutting and ploughed fields detection demos. About 2015 [if you have the field inspector archive data]. During vegetative season of 2016. Needed input: parcel database to carry out the checks. 11
Future developments Starting to use Sentinel-2 multispectral data in 2016. Higher res. - can do grass cutting detection on parcels with min. diameter of 30 m (cloud-free conditions). Further improving the reliability with S1+S2 hybrid method. Applying meteo-data for improvements. Conclusion (1) Copernicus Sentinel data offers the possibility to improve subsidy checks and save tens to hundreds of millions euros EU tax-payer money every year. Reducing the need for field surveys and making the checks more objective. Objective data source to solve disputes. Sentinel-1 based pilot in Summer 2015 very encouraging results. 12
Conclusion (2) Copernicus S1 and S2 taking EO to the Big Data era. Traditional data processing and transfer methods are infeasible. ESA s vision: EO Community platforms, processing next to data. Establishment of TO spin-off company KappaZeta, first products: grass cutting detection and mapping of ploughed fields each spring under all weather conditions. Opportunity for demo processing in your country. List of publications from our group K. Voormansik, T. Jagdhuber, A. Olesk, I. Hajnsek and K. P. Papahtanassiou, "Towards a detection of grassland cutting practices with dual polarimetric TerraSAR-X data," International Journal of Remote Sensing, vol. 34, no. 22, pp. 8081-8103, 2013. K. Voormansik, T. Jagdhuber, I. Hajnsek and K. P. Papathanassiou, "Improving Semi-natural Grassland Administration with TerraSAR-X," in Proceedings of the 17th GeoCAP Annual Conference, edited by: D. Fasbender, K. Taşdemir, Ph. Loudjani, V. Angileri, C. Lucau, P. Milenov, W. Devos, R. De Kok, S. Lemajic, A. Tarko and P. Pizziol, pp. 26-32, Tallinn, 2011. K. Zalite, O. Antropov, J. Praks, K. Voormansik, M. Noorma, Towards detecting mowing of agricultural grasslands from multi-temporal COSMO-SkyMed data, 2014 IEEE International Geoscience and Remote Sensing Symposium (IGARSS), pp. 5076 5079, 2014. K. Voormansik, T. Jagdhuber, K. Zalite, M. Noorma, I. Hajnsek, Obervations of Cutting Practices in Agricultural Grasslands using Polarimetric SAR, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2015 [accepted] K. Zalite, O. Antropov, J. Praks, K. Voormansik, M. Noorma X-band Repeat-Pass Interferometry for Monitoring of Grasslands, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2015 [accepted] 13
27/11/2015 Thank you! Questions? kaupo.voormansik@ut.ee 14