PFG 2014 / 5, 0383–0392 Article Stuttgart, October 2014

Analysing Phenological Characteristics Extracted from Landsat NDVI Time Series to Identify Suitable Image AcquisitionDatesforCannabisMappinginAfghanistan

MATTEO MATTIUZZI,COEN BUSSINK &THOMAS BAUER, Vienna, Austria

Keywords: illicitcropmonitoring,,NDVI,phenology,seasonality,senescence

Summary: Since2009,theUnitedNationsOffice Zusammenfassung: Analyse phänologischer Ei- on Drugs and Crime (UNODC) has carried out genschaften basierend auf Landsat NDVI Zeitrei- various annual surveys of in hen für die Bestimmung geeigneter Zeitpunkte für . The mapping of cannabis fields is die Aufnahme von Satellitenbildern für die Kartie- based on very high resolution satellite imagery. The rung von Cannabis in Afghanistan. Das Büro für image acquisition dates have a strong influence on Drogen- und Verbrechensbekämpfung der Verein- whether cannabis can be visually differentiated ten Nationen (UNODC) führt seit 2009 jährlich from other crops. Expert knowledge shows that the eine Auswertung der Anbaufläche von Cannabis in idealacquisitiondatesareattheendofthegrowing Afghanistan durch. Die Kartierung von Cannabis- period (senescence) of cannabis. For future surveys feldernerfolgtmitHilfevonsehrhochauflösenden amethodwasdevelopedtopredicttheoptimumob- Fernerkundungsdaten. Der Aufnahmezeitpunkt servationdatesindependentoftemporaldifferenc- der Daten ist sehr bedeutend dafür, ob Cannabis es of the phenology. The study includes an analysis von anderen Feldfrüchten visuell eindeutig unter- ofLandsat-basedNDVItimeseriesandthecalcula- schieden werden kann. Nach Expertenansicht liegt tionofinformationaboutthesenescencestagesof dieser Zeitpunkt am Ende der Wachstumsperiode cannabis and other vegetation. The approach was (Seneszenz) von Cannabis. Fürzukünftige Aus- applied to four test sites in Afghanistan. wertungenwurdeeineMethodeentwickelt,den idealen Aufnahmezeitpunkt unabhängig von zeitli- chen Unterschieden der Phänologie im Voraus zu bestimmen. Der Ansatz umfasst eine Analyse von NDVI Zeitreihen basierend auf Landsat-Daten und die Ableitung von Informationen über die Senes- zenz-Phasen von Cannabis und anderen Pflanzen. Die Methode wurde in vier Testgebieten in Afgha- nistan angewandt.

1Introduction for alternative development and law enforce- ment.Inordertoascertaintheimpactofthese Illicit crops like cannabis, opium poppy and measures, the United Nations Office on Drugs coca bush form the basis for highly lucrative andCrimemonitorsillicitcropproduction, narcotic drug production. Illicit crop cultiva- providing technical assistance to its member tion(cannabisandopiumpoppy)inAfghan- statesonmethodology.Oneofthekeycompo- istantakesplaceinareaswhereinsecurity nents to estimate plant-based drug production andinsurgenciesareprevalentandisstrong- istheareaundercultivation,whichismostly ly linked with organized criminal networks done with the help of satellite images and ap- (UNODC 2013a).Policymeasurestotacklethe propriate statistical techniques. These estima- problem require an integrated approach of sta- tionsfacevariouschallengesandinparticular bilization measures, creating opportunities the optimum timing of the satellite imagery.

© 2014 E. Schweizerbart'sche Verlagsbuchhandlung, Stuttgart, Germany www.schweizerbart.de DOI: 10.1127/1432-8364/2014/0231 1432-8364/14/0231 $ 2.50 384 Photogrammetrie • Fernerkundung • Geoinformation 5/2014

This is especially difficult for monitoring can- tection through remote sensing using spectral nabisinAfghanistanduetothepossiblecon- differences detected from aerial platforms or fusion with other crops. hyperspectral images. DAUGHTRY & WALTHALL UpuntilnowUNODChasconductedfour (1998) and AZARIA etal.(2012)identifiedspe- annual cannabis surveys in Afghanistan to cific spectral ranges that are optimum for dif- determine the area under cultivation (2009 ferentiating cannabis from other crops that – 2012). In addition UNODC has been con- areattheirbestwhenthecannabisfieldshave ductingannualopiumpoppysurveysinAf- densecanopies.Althoughthereisalotofin- ghanistansince1994andsince1999with formalknowledgeonthebehaviourofcanna- satellite imagery. For estimating the area, bisplantsandhowtheyareaffectedbytheen- UNODC used satellite imagery that were ac- vironment and cultivation practices, as far as quired for locations based on a spatial sam- theauthorscouldverify,thereisnoscientific plingapproach.Theveryhighresolutionsat- literature on monitoring the phenology of can- ellite images (VHR) obtained on a single ac- nabiswithremotesensingtimeseries. quisition date were used to identify cannabis Phenology is defined as the study of the fields(Fig.1)andtocalculatetheareaunder timing of recurring plant life cycle events that cultivation, which was extrapolated to a larg- are driven by environmental factors (MORI- er cannabis risk area (UNODC 2013b). Reliable SETTE etal.2009).Thedocumentationofthe information on the growing cycle of cannabis timing of phenological events often relies on plants (phenology) is essential for the selec- visual observations in the field. In order to col- tion of the ideal acquisition time of the VHR lect data on phenology over large geographic images. Within the growing cycle that is sub- areas, measurements of reflected electromag- jecttoseasonalandregionalvariations,anat- neticradiationfromthelandsurfacecanbe temptismadetotryandfindthemostappro- used. The timing of recurring changes in re- priatetimefortheacquisitionofthesatellite flectance is defined as land surface phenology imageswheretheillicitcropcanbestbeiden- (HANES etal.2014).Inthispaper“phenology” tifiedanddistinguishedfromothervegetation refers to this term. Information on phenology bymeansofavisualinterpretation.According asderivedfromremotelysensedimagesdiffer to local UNODC experts in Afghanistan can- from conventional phenology data in that the nabis is harvested later than most other crops. information relates to the aggregated charac- Studiesonthemappingofcannabisare teristics of the vegetation within the areal unit often directed towards the early detection measured by satellite sensors. Phenological of plants in order to guide eradication cam- changescanbemonitoredbymeansofremote paigns, e.g. by mapping of potential high-risk sensingasplantschangeinappearanceand areas to reduce the search area (LISITA et al. structure during their growth cycle. 2013).Therearevariousunpublishedstudies Theanalysisasdescribedinthispaperis butonlyafewscientificpublicationsonde- based on the normalized difference vegetation

Fig. 1: Left: cannabis fields at flowering stage, as seen on a colour infra-red (CIR) satellite image, right: as seen from helicopter in Nangarhar province in 2012 (UNODC 2013). Matteo Mattiuzzi et al., Analysing Phenological Characteristics 385 index(NDVI).TheNDVIisamathematical termine the number of seasons or seasonal pa- combinationoftheRedandtheNIR(near-in- rameters,e.g.,thebeginningortheendofthe frared) band. Annual time series of satellite- season. ZHANG et al. (2003) used MODIS and derived vegetation indices and biophysical AVHRR data for modelling the key pheno- metricsthatincorporatethereflectanceofthe logical phases: green up, maturity, senescence NIR and the red wavelengths generally cap- anddormancy.Asoftwaretooldevelopedby ture the spectral changes associated with leaf LANGE & DOKTOR (2013)canalsobeusedto growth and senescence in the large areal units determine such parameters. monitored by the sensor (HANES et al. 2014). Theobjectiveofthispaperistotestwheth- In order to better distinguish cannabis from er the cannabis phenology in different regions otherplantsandtodeterminetheidealacqui- in Afghanistan is different from the general sition dates for VHR images, focus is laid on a vegetation at the end of the growing season, more detailed analysis of the senescence. Dur- when cannabis can best be identified on VHR ing the onset of senescence, deterioration of satellite imagery. In addition, the research cell walls in the mesophyll tissue produces a looks at whether the development of cannabis distinctive decline in infrared reflectance; an in comparison to the general vegetation pro- accompanying increase in visible brightness vides a basis for the development of a forecast- maybetheresultofdeclineintheabundance ingtooltodeterminethebestdatesforVHR andeffectivenessofchlorophyllasanabsorb- imageacquisition.Thisisdonebyanalysing erofvisibleradiation(CAMPBELL &WYNNE twoparameters:i)thetimespanbetweenthe 2006).Thisleadstoasignificantdecreaseof dates when the differences of the NDVI val- NDVI values. uesofcannabisandnon-cannabisvegetation While many studies have analysed pheno- reach a maximum; and ii) the maximum num- logicaltrendsingeneral(e.g.MORISETTE et al. ber of occurrences of senescence within each 2009),onlyafewstudieshavefocussedon testsite.ThemaximumdifferenceofNDVI informationonsenescencederivedfromre- between cannabis and non-cannabis is expect- motely sensed images. JÖNSSON &EKLUNDH ed to be the date where cannabis is most eas- (2004) for example developed a method to de- ilydistinguishable.Themaximumnumberof

Fig. 2: Test sites in Afghanistan and Landsat scenes necessary to cover the areas. 386 Photogrammetrie • Fernerkundung • Geoinformation 5/2014 occurrencesofsenescenceisusedtodefine 3 Methodology a comparable phenological stage sensitive to seasonal changes between the years. 3.1 Data

Theanalysiswascarriedoutusingatime 2StudyArea series of archived medium resolution satel- liteimages.Inthiscaseallavailabledataof Cannabis in Afghanistan is mostly (almost the Landsat series (Landsat 5 TM and Land- 90%)grownasanirrigated,annualmono- sat7ETM+)areacquiredfortheperiodbe- crop,whereastheremaining10%ofthecan- tween September 2008 and July 2012. Land- nabis growing farmers cultivate cannabis in sat was chosen as it offers free available data combinationwithothercropsoronbunds witharelativelyhighspatialresolution,and (UNODC 2011). Cannabis flowers have the high- the continuity of acquisition is guaranteed est concentration of the active chemical drug forthecomingyearswiththenewLandsat components ( like tetrahydro- 8data.ThetimeseriesarebasedonLand- cannabinol)andthetimeoffloweringisde- sat5TMandLandsat7ETM+(SLC-off)and pendent on the age and variety of the plant, were downloaded from the USGS Earth Ex- changeinthephotosyntheticperiodandenvi- plorer (EARTH EXPLORER 2014). Fig. 2 shows ronmental conditions. Cannabis plants usual- the Landsat scenes necessary to cover the test lystartfloweringaftertwomonthsofgrowth sites.Intotal7differentscenelocationswere but the abundant flowering starts when dark- needed. Landsat 5 TM could only be used un- ness (night length) exceeds approximately 11 til November 2011. The temporal resolution of hours per day. The flowering stage lasts be- eachLandsatis16daysresultinginarevisit- tween 4 and 12 weeks, depending on envi- ingtimeof8dayswithbothsensors.Addi- ronmental conditions (CLARKE 1993). Outdoor tionally, in Afghanistan the overlapping areas cannabis cultivation in temperate zones like from images of neighbouring Landsat orbits Afghanistantakesplaceinthesummersea- arecoveringmorethan50%ofeachscene son (March – September)andthelifecycleof (approximately30%oneachside)leadingto cannabis in Afghanistan is 5 – 6months.From data for additional dates (LANDSAT 72012).For interviewswithAfghanfarmers,itisknown theperiodapproximately230Landsatscenes that the planting season for cannabis starts be- pertestsiteweredownloadedandprocessed. tween March and May/June. The plants are Landsat7ETM+andLandsat5TMhave harvested from September/October to No- very similar spectral bands. To produce a vember/December depending on the region NDVI the Red (band 3) and the NIR (band (UNODC 2013b). Cannabis is harvested when 4)arerequired.TheRedbandsofbothsen- parts of the plants have already started dete- sors have the identical spectral wavelength, riorating. Four test sites were selected in the the NIR is in Landsat 7 ETM+ with 0.77 μm provinces of Hilmand, Kandahar, Nangarhar to 0.90 μm slightly narrower compared to and Uruzgan that represent the major canna- 0.76 μmto0.90μminLandsat5TM.Since bisgrowingareasinAfghanistan(Fig.2).The thedifferenceissmallitcanbeassumedthat northern growing provinces were not included the impact can be neglected. sincethisregionisamountainousareawhere As reference data polygon data representing thematchingofthereferencedatasetwiththe thecannabisfields(asshowninFig.1)were Landsattimeseriesishighlychallenging.The madeavailablebyUNODC.Thepolygons test sites are mainly flat, irrigated areas sur- were delineated by UNODC staff based on rounded by desert. VHR satellite images taken during the grow- ing seasons of 2009, 2010 and 2011. Therefore, nogeometricadjustmentoftheLandsatdata was necessary as Landsat images fit with the reference data. Matteo Mattiuzzi et al., Analysing Phenological Characteristics 387

3.2 Data Pre-Processing MODIS package available in R (MATTIUZZI et al. 2012b). In order to process accurate NDVI values, spectral data were transformed from digi- talnumberswhichisbasicallytop-of-atmos- 3.3 Analysis of the Phenology and phere radiance with bias and gain introduced the Senescence for storage optimisation to top-of-atmosphere reflectance using the R package “Remote The result of the pre-processing is a yearly Sensing” (RCORE TEAM 2012, MATTIUZZI et stackoflayersofwhichNDVIgraphscanbe al.2012a).Inasecondstepthedatahasbeen derivedhighlyreflectingphenologicalstages cropped and mosaicked to the extent of the test at certain locations (Fig. 3 right). sites (Fig. 2). This was done using utilities of Fig. 3 left shows the locations of cannabis the geospatial data abstraction library (GDAL fields in Kandahar. A VHR satellite image 2012). The NDVI values were filtered to re- (colour infrared) acquired on November 15 in ducetheimpactofnoise,cloudcover,missing 2010 is used as a background image. The yel- data(Landsat7ETM+SLC-off)andtoderive lowpolygonsindicatecannabisfieldsasvis- smooth curves at regular temporal intervals. ually interpreted by UNODC experts. At the For filtering the „Modified Whittaker Smooth- highlighted green marker the NDVI values er” (ATZBERGER &EILERS 2011) was chosen as arereadoutoftheNDVIlayerstackbasedon themethodallowsfilteringdatawithoutinfor- Landsatdataanddisplayedasagraph(Fig.3 mationonpixelquality.Thismethodisbased right). The example clearly shows two peaks on two assumptions (CHEN etal.2004):i)that withinayear.Inmostlocations,wherecan- the NDVI time-series follows an annual cycle nabiswasidentified,twocropsoccurwithina of growth and decline as the index is primar- year. On those fields the first peak can be iden- ily related to vegetation density and plant vig- tifiedattheendofMarch/beginningofApril our; and ii) that clouds and poor atmospher- (winter crop) and in August/September (sum- ic conditions depress NDVI values, requir- mercrop)asecondpeakoccurs.Thesecond ingthatsuddendropsinNDVI,whicharenot peakrepresentscannabiswhilethefirstpeak compatible with the gradual process of veg- relates to other crops. The yellow points mark etationchange,areregardedasnoiseandwill the main phenological stages: start of the sea- be removed. Filtering was carried out with the son(localminimumNDVIatthebeginning

Fig. 3: Left: WorldView-2satelliteimage(CIR;©DigitalGlobeInc.,allrightsreserved)andcan- nabis polygons,right:NDVIcurvewithphenologicalindicatorsatonepixellocationfor2010. 388 Photogrammetrie • Fernerkundung • Geoinformation 5/2014 of green up), maximum (highest NDVI value) selectedwhichhaveaNDVImaximumofat and end of season (local minimum NDVI at least+0.3intheperiodof40daysbeforeand the end of senescence). A senescence of 10% after the day where cannabis (mean) reaches isthepointwheretheNDVIdecreasesby10% its maximum. A NDVI value between 0.0 and of the cycle amplitude. The threshold was set +0.2 indicates a very low or no photosynthetic to 10% in order to ensure an early detection of activity. The underlying assumption is to se- the declining phase of the NDVI values. This lectthosepixelswithconsistentactivevege- valueisusedforthesubsequentanalysis.The tation within the cannabis growing period. It dashedredlinerepresentstheacquisitiondate isassumedthatwithinaregionthecomposi- oftheVHRsatelliteimageasusedforthevis- tion of planted crops is relatively stable in the ual interpretation in 2010. years.Theselectedpointscanbychancealso The best acquisition date of VHR satellite include cannabis pixels. This is because for images when cannabis can be distinguished the prediction of the senescence of cannabis in from other crops is expected to be at the date the following year no information about can- whenthemaximumdifferenceintheNDVI nabiswillbegiven.Therefore,twooutofthe occurs.Theoreticallythiscanbeatanytime three years were used to learn how the phenol- within the growing cycle but the aim is to de- ogy of cannabis behaves in comparison to the terminethebestdateattheendofthegrowing generalsenescenceofallcropsintheregion. season when cannabis can best be differenti- For the remaining year the phenological stage atedthroughvisualinterpretation.Inorderto of cannabis was predicted using the knowl- forecast this optimal date without knowing the edgederivedfromtheotheryears. location of cannabis, the phenology of all ac- BasedontheNDVIcurvesofthesummer tive summer vegetation at the test sites is fur- periodthesenescencewasderivedusingthe therinvestigated.Thebasicideaistofocuson R package “phenex” (LANGE &DOKTOR 2013). the timing of senescence of all active summer Thedateofthesenescencewasdeterminedin vegetation pixels including cannabis in the thefollowingway:Inafirststeptheampli- testregionsandtodrawconclusionsonthe tudeofthevegetationperiodwascalculated timing of NDVI differences. usingtheminimumandmaximumNDVIval- Theusageofthesenescenceforthepredic- uesfoundinthegreenupphaseofthesum- tionhasbeenselectedasitisexpectedthat mercrop.Thedateofsenescencerefersto thesenescencestageiscloselyrelatedtothe the point where the NDVI values decrease af- NDVI differences at the end of the growing terthemaximumbyagiventhresholdofthe cycle.Theperiodbetweenthemaximumnum- NDVI amplitude (see Fig. 3 right), in this case berofoccurrencesofearlysenescence(active 10% as mentioned before. summer vegetation, which can include canna- The distance between the date of the maxi- bissites)andmaximumdifferenceinNDVI mum senescence intensity of active summer (between cannabis and non-cannabis) is es- vegetation(seereddashedlineinFig.4)and timatedbasedontwoofthethreeavailable thedateofthemaximumNDVIdifference years, e.g. 2009 and 2010, using the average (black solid line) was calculated. The mean numberofdaysoftheseyears.Thistimepe- difference of the chosen reference years was riod is then added to the date of the maximum thenaddedtothemaximumsenescenceinten- number of occurrences of early senescence sity of active summer vegetation in the pre- (active summer vegetation) in the year of the diction.Thenumberofoccurrencesofsenes- forecast,inthiscase2011.Themaximum cenceisusedasanindicatorifthephenology number of occurrences of senescence is used in the test regions has changed from one year to determine the general land surface phenol- to another. For the subsequent analysis only ogy and training years are used to determine reliable cannabis pixels were taken into ac- how the NDVI of cannabis behaves compared count,theselectionwasdonebyusingonly totherestofthevegetation. those pixels that are fully contained by a can- ThetimeofthemaximumNDVIwascalcu- nabis reference polygon. latedforthesummerperiodonly.Withineach geographic region 5000 random points were Matteo Mattiuzzi et al., Analysing Phenological Characteristics 389

4 Results region in the following figures. In the lower part of each graph the intensity of senescence The main outcome of this study is shown by is shown over time for active summer vegeta-

Fig. 4: Differences in NDVI and timing of senescence by region for 2009, 2010 and 2011 (lines: black solid = date of maximum NDVI difference, blue dashed = predicted maximum NDVI difference). 390 Photogrammetrie • Fernerkundung • Geoinformation 5/2014 tion (red) and cannabis (green). In most cases bythesurveyteamifthedateofmaximum itcanbeseenthatcannabisreachesitsmaxi- difference in NDVI corresponds with the date mumnumberofsenescenceoccurrenceslater whencannabiscanbebestinterpretedinthe than the active summer vegetation. This con- VHR satellite images. This information will firmstheexperienceofUNODCinterpret- be essential for optimising the programming ersthatcannabisisharvestedlaterthanother of the VHR satellite images. crops. The upper part shows the difference in The method presented here to find the opti- NDVI (subtraction of non-cannabis from can- mumacquisitiondateshowsthatinmostcases nabis)andtheboldblackverticallineindi- the predictions based on the timing of senes- catesthedateofthemaximumdifference.The cenceoftheactivesummervegetationstrong- NDVI difference in Kandahar 2010 reaches its ly correspond with the optimum dates calcu- maximum earlier than the maximum number latedfromtheactualNDVIvalues.Differenc- of occurrences of senescence. Therefore, the es between predicted and actual dates range optimum date has been shifted to the maxi- from1to21days.Althoughforoperational mum number of occurrences of senescence of purposesthesetimerangesareacceptable,it theactivesummervegetation. should be looked at more closely whether this Thedashedbluelinesinthegraphsrepre- method of prediction yields better results than sent the predicted maximum NDVI difference othermethods,e.g.usingthestartoftheveg- (predictedidealacquisitiondate).Thecalcula- etation activity (green up) or simply using the tionisbasedonthetwootheryears.Compar- mean date of maximum NDVI difference. ingthiswiththedateoftheactualmaximum Alimitationisthelackofanindependent NDVI difference (black line) gives informa- datasettoverifyifthenumberofmaximum tion about the validity of the assumption for occurrences of senescence reliably reflects the this study. The differences between the two shifts in the timing of phenology. A deeper dates in 2011 is 18 days for Hilmand, 17 days investigation, for example on test sites with forKandahar,7daysforUruzganand9days ground based weather stations, could explain for Nangarhar. A time range can be derived, factors that influence phenological stages. taking into account the difference of the pre- A strong limitation for developing and test- viousyears.ForHilmandthistimerangecan ing the method was that the predictions are then be defined between 4 to 20 days, for Kan- basedontwoobservationsonly.Itwouldre- daharbetween3to21days,forUruzganbe- quirelongertimeseriestoobtainmorerobust tween1to10daysandforNangarharbetween results, e.g. by including subsequent cannabis 7to13days. surveys. When this research was conducted anothercannabissurveywasfinalised,which cangiveadditionaldatatoenhancetheanaly- 5 Conclusions and Discussion sis. Attention should be paid to the selection of This research shows the usefulness of using cannabis pixels, as the median cannabis field Landsat time series to determine differences size is only slightly larger than a Landsat pix- in crop phenology for the monitoring of can- el thereby reducing the number of usable can- nabisinAfghanistan.Thegraphsdepicting nabisreferencepixeldrastically.Forthisrea- the development of the vegetation consistent- son,manypixelshadtobeleftoutfromthe ly show that cannabis tends to reach the stage analysis because of probable mixed signa- of senescence later than other active sum- tures.Inthenearfuture,imagesfromSenti- mer vegetation. It is not known whether this nel-2(launchplannedin2015)couldbeused depends on the type of vegetation, e.g. crop for this purpose offering a spatial resolution of types, management of the crops or the eco- 10m.AnotherbenefitofSentinel-2isthevery physiological circumstances in Afghanistan. high temporal resolution (5 days) which would Theidentifiedpatternsupportstheperception lead to a more accurate computation of pheno- of the experts that undertake the identification logical phases. of cannabis fields in the VHR satellite images, One problem that this study faced was also howeveritshouldbesystematicallymonitored the lack of information on the cultivation of Matteo Mattiuzzi et al., Analysing Phenological Characteristics 391 other crops. Information on crops growing in Distinctive Cannabis. – Ronin Publishing, the same season would help to understand the Berkeley, CA, USA. difference between cannabis and crops with DAUGHTRY,C.S.T.&WALTHALL,C.L.,1998:Spectral similar growing cycles. A crop by crop analy- discriminationofCannabissativaL.leavesand – sis could also help to detect which crops are canopies. Remote Sensing of Environment 64: 192–201. causing problems in the prediction. EARTH EXPLORER, 2014: http://earthexplorer.usgs. Future research can focus on what can gov (2.7.2014). belearnedfromtheNDVIdifferenceplots GDAL, 2012: Geospatial Data Abstraction Library: (Fig.4,upperparts).Thedatesderivedfrom Version1.9.1.OpenSourceGeospatialFounda- thisgraphshouldbeinterpretedasanop- tion, Land surface phenology. – http://www. timum within a time range and not as a sin- gdal.org (2.7.2014). gle date. Experts with better knowledge of HANES,J.M.,LIANG,L.&MORISETTE, J.T., 2014: thecroptypesintheregionscouldinterpret Land surface phenology. – HANES, J.M. (ed.): graphs from previous years. Similar NDVI Biophysical Applications of Satellite Remote – difference patterns between the years could Sensing. 230 p., Springer, Berlin. JÖNSSON,P.&EKLUNDH,L., 2004: TIMESAT – A indicate that similar crops were grown and program for analysing time-series of satellite therefore represent a better prediction. Anoth- sensor data. – Computers and Geosciences 30: er aspect is to focus on a more detailed analy- 833–845. sisoftheregionalsenescencepatternsinorder LANDSAT 7,2012:Landsat7ScienceDataUsers todeterminethepointintimewithintheyear Handbook. – http://landsathandbook.gsfc.nasa. whenareliablepredictionoftheidealacquisi- gov (2.7.2014). tion date is feasible. LANGE,M.&DOKTOR, D., 2013: phenex: Auxiliary functionsforphenologicaldataanalysis.Rpack- age version 1.0-3. – http://CRAN.R-project.org/ Disclaimer package=phenex (2.7.2014). LISITA,A.,SANO,E.E.&DURIEUX,L.,2013:Identi- fying potential areas of planta- Theviewsexpressedhereinarethoseofthe tions using object-based image analysis of authors and do not necessarily reflect the SPOT-5 satellite data. – International Journal of viewsoftheUnitedNations. Remote Sensing 34: 5409–5428. MATTIUZZI,M.,CHEMIN,Y.,LABORTE,A.&HIJMANS, R.J.,2012a:RemoteSensing.Rpackageversion References 0.4.9/r260. – http://R-Forge.R-project.org/ projects/remotesensing (2.7.2014). MATTIUZZI,M.,VERBESSELT,J.,HENGL,T.,KLISCH, ATZBERGER,C.&EILERS,P.H.C.,2011:Evaluating A., EVANS,B.&LOBO,A., theeffectivenessofsmoothingalgorithmsinthe 2012b: MODIS down- absence of ground reference measurements. – loadandprocessingpackage.Processingfunc- International Journal of Remote Sensing 32: tionalities for (multi-temporal) MODIS grid – 3689–3709. data. R package version 0.5-3/r276. AZARIA, I., GOLDSCHLEGER,N.&BEN-DOR,E.,2012: http://R-Forge.R-project.org/projects/modis IdentificationofCannabisplantationsusinghy- (2.7.2014). MORISETTE, J.T., R ICHARDSON,A.D.,KNAPP,A.K., perspectral technology. – Israel Journal of Plant GRAHAM,E.A.,ABNATZOGLOU,J.,WILSON,B.E., Sciences 60 (1): 77–83. BRESHEARS,D.D.,HENEBRY,G.M.,HANES,J.M.& CAMPBELL,J.B.&WYNNE,R.H.,2006:Introduction LIANG,L.,2009: to remote sensing. – 5th edition, 667 p., Guilford Tracking the rhythm of the sea- Press,NewYork,NY,USA. sons in the face of global change: phenological CHEN,J.,JÖNSSON,P.,TAMURA,M.,GU,Z.,MATSU- research in the 21st century. – Frontiers in Ecol- – SHITA,B.&EKLUNDH,L., 2004: A simple method ogyandtheEnvironment7 (5): 253 260. RCORE TEAM for reconstructing a high-quality NDVI time- ,2012:R:Alanguageandenviron- series data set based on the Savitzky-Golay fil- mentforstatisticalcomputing.RFoundationfor ter. – Remote Sensing of Environment 91: 332– Statistical Computing, Vienna, Austria. ISBN – 344. 3-900051-07-0. http://www.R-project.org CLARKE,R.C.,1993:MarijuanaBotany:AnAd- (2.7.2014). vanced Study: The Propagation and Breeding of 392 Photogrammetrie • Fernerkundung • Geoinformation 5/2014

UNODC,2011:Afghanistan– Cannabis Survey 2010. Addresses of the Authors: – United Nations Office on Drugs and Crime, MATTEO MATTIUZZI & THOMAS BAUER,Universityof Vienna, Austria. Natural Resources and Life Sciences, Vienna UNODC,2013a:Afghanistan– Opium survey 2013. (BOKU), Institute of Surveying, Remote Sensing – United Nations Office on Drugs and Crime, and Land Information, Peter-Jordan-Straße 82, Vienna, Austria. A-1190Vienna,Tel.:+43-1-47654-5100,Fax:+43- UNODC,2013b:Afghanistan– Survey of Commer- 1-47654-5142, e-mail: {matteo.mattiuzzi}{t. cial Cannabis Cultivation and Production 2012. bauer}@boku.ac.at – United Nations Office on Drugs and Crime, Vienna, Austria. COEN BUSSINK,UnitedNationsOfficeonDrugsand ZHANG,X.,FRIEDL,M.A.,SCHAAF,C.B.,STRAHLER, Crime, Statistics and Surveys Section, Trends A.H., HODGES,J.C.F.,GAO,F.,BRADLEY,R.C.& MonitoringandAnalysisProgramme,POBox500, HUETE,A.,2003:Monitoring vegetation phenol- A-1400 Vienna, Tel.: +43-1-26060-4165, Fax: +43- ogy using MODIS. – Remote Sensing of Envi- 1-26060-74165, e-mail: [email protected] ronment 84: 471-475.

Manuskript eingereicht: März 2014 Angenommen: Juli 2014