Temporal identification of poppy fields on high resolution satellite imagery Marc Rußwurm marc.russwurm@tum.de Michał Krupiński mkrupinski@cbk.waw.pl Stanisław Lewiński stlewinski@cbk.waw.pl
Introduction Centrum Badań Kosmicznych PAN 5 months Erasmus+ SMP Technische Universität München Master Geodesy and Geoinformation studies focus on Remote Sensing and Photogrammetry Marc Rußwurm ESA Project "Multi-sensor satellite and aerial data fusion for illicit crops detection In cooperation with Planetek Testfields of poppy (mak) Species: Morfeusz Lazur Mieszko Borowski Bialy 2012, 2013, 2014, 2016
Phenology of poppy Germination Vegetative stage Reproductive stage Senescence stage Lazur poppy 2016 12. May 23. May 7. Jun 14. Jun 22. Jun 28. Jun 06. Jul 18. Jul growth emergence flowering harvest time
Growth - Measurements measure growth with NDVI NDVI: broadly available (red, nir) sensitive for chlorophyll and red edge robust to shadows Assumptions: 1. growth is expressed in NDVI 2. environmental conditions are the same (or similar)
Testfields and sensors Fields of poppies 2012, 2013, 2014, 2016 length ~60-300m width: 40-80m Landsat 8 NDVI 180x80m sensor number of products footprint repeat cycle [days] atm. cor. algorithm Landsat 7 ETM+ 36 30m 16 LEDAPS Landsat 8 OLI 25 30m 16 LaSRC Sentinel 2A MSI 10 10m 10 Sen2cor RapidEye 2 5m 1 ReSe: ATCOR
Preprocessing field geometry database of observations Landsat 7 Landsat 8 RapidEye Sentinel 2 buffer -15m 1. select only clear pixels 2. calculate NDVI Field id name kind satellite day of acquisition NDVI_mea n NDVI_std n_pixel Datatype <int> <String> <String> <String> <Date> <float> <float> <int>
Observations poppy by species low morphine high morphine (2012,2013) (2013) (2014, 2016) Combination of 5 fields in 4 years NDVI Jan Feb Mar Apr May Jun Jul Aug Sept Oct Nov Dec day of year
Observations poppy by sensor No systematic influence by sensors NDVI Jan Feb Mar Apr May Jun Jul Aug Sept Oct Nov Dec day of year
Observations poppy by year NDVI seeding flowering harvest day of Jan Feb Mar Apr May Jun Jul Aug Sept Oct Nov Dec year
Observations poppy by year 7. Jun 22. Jun NDVI 28. Jun 12. May 18. July seeding flowering harvest day of Jan Feb Mar Apr May Jun Jul Aug Sept Oct Nov Dec year
Model Double Logistic Function (Eklundh & Jönsson, 2015) (Evrendilek et al., 2008) (Zhang et al., 2003) NDVI day of year
Identification 2012 2013 2014 2016 Training dataset Years 2012, 2014, 2016 3 poppy fields, 31 measurements Test dataset Year 2013 2 poppy fields 21 measurements + other crops
RMS as distance poppy model vs corn measurements
Map of RMS NDVI Time Stack
Map of RMS
Conclusions Using: freely available sensors very little spectral information (red/nir) We were able to: detect poppies in the test area using temporal features model the growth of a crop with good predictability detect different growth behavior of other crops Next step: potential in combination with Multi/Hyperspectral approaches
Temporal identification of poppy fields on high resolution satellite imagery Eklundh, L., & Jonsson, P. (2012). TIMESAT 3. 2 with Parallel Processing Software Manual. Lund University. Evrendilek, Fatih, and Onder Gulbeyaz. "Deriving vegetation dynamics of natural terrestrial ecosystems from MODIS NDVI/EVI data over Turkey."Sensors 8.9 (2008): 5270-5302. Zhang, X., Friedl, M. A., Schaaf, C. B., Strahler, A. H., Hodges, J. C., Gao, F., Huete, A. (2003). Monitoring vegetation phenology using MODIS. Remote sensing of environment, 84(3), 471-475. Marc Rußwurm marc.russwurm@tum.de Michał Krupiński mkrupinski@cbk.waw.pl Stanisław Lewiński stlewinski@cbk.waw.pl
Growth - Factors Approximation: in the same observation area same environmental conditions also in consecutive years observation influences are minor