AGRICULTURAL CROP MAPPING USING OPTICAL AND SAR MULTI- TEMPORAL SEASONAL DATA: A CASE STUDY IN LOMBARDY REGION, ITALY

Transcription

1 AGRICULTURAL CROP MAPPING USING OPTICAL AND SAR MULTI- TEMPORAL SEASONAL DATA: A CASE STUDY IN LOMBARDY REGION, ITALY G. Fontanelli, A. Crema, R. Azar, D. Stroppiana, P. Villa, M. Boschetti Institute for Electromagnetic Sensing of the Environment (CNR IREA), Via Bassini 15, Milan (Italy)

2 INTRODUCTION The availability of information about agricultural crops (i.e. typology, phenology, productivity, health) is crucial for a proper agronomic planning and management, especially for end users such as farmers and public administration. Remotely sensed data acquired from space sensors, both optical and SAR, have long demonstrated their capability in providing such information in a timely and reliable fashion. The integration of SAR and optical sensors data allow us to take advantage by the different sensitivity of both the technologies toward environmental parameters: e.g. soil roughness and moisture, plant water content and biomass for SAR and photosynthetically related vegetation features for the optical sensors. The research has been carried out in context of the projects: Space4Agri (Development of Innovative Earth Observation based Methodologies supporting the Agricultural Sector in Lombardy; Agreement Framework between Regione Lombardia and CNR). ERMES (An Earth Observation Model based RICE information Service; EU-FP7 Collaborative Project).

3 AIM OF THE RESEARCH This work aims at: mapping of crops in Lombardy region (northern Italy) for the year 2013, by using multi-temporal, intraannual series of SAR and optical satellite data; comparing the performance: of three different supervised classification algorithms, at different temporal periods during the growing season.

4 MAIN CROPS IN LOMBARDY REGION 2013 South Lombardy Region Pavia province Framed within the Po and the Ticino river basin pasture & grazing 43% hectares cropped in farms maize 26% wheat & barley 12% rice 12% soybean 5% vegetables 2% other 0%

5 13/05/ /06/ /06/ /06/ /06/ /07/ /07/ /08/ /08/ /08/ /09/ /12/ /12/ /02/ /03/ /03/ /04/ /04/ /05/ /05/ /06/ /06/ /07/ /07/ /08/ /09/ /09/ /10/2013 SATELLITE DATASET 15 COSMO-SkyMed AND CROP (CSK) SEASONALITY HIMAGE, X-band HH pol., θ=24.13 Revisiting time=16 days SPRING 13 LANDSAT 8 OLI (L8) Level 1 WHEAT SUMMER SOYBEAN MAIZE Revisiting time=16 days (8 days thanks to two overlapping frames) RICE CSK L8 26/01/ /03/ /05/ /06/ /08/ /10/ /11/ /01/2014

6 PRE-PROCESSING OF SATELLITE DATA OCCURRENCE FILTER RESAMPLING TO 30 m VARIANCE CSK MULTI TEMPORAL FILTERING (De Grandi) GEOCODING AND RADIOMETRIC CALIBRATION CO- OCCURRENCE FILTER RESAMPLING TO 30 m CONTRAST SRTM DEM RESAMPLING TO 30 m σ HH Enhanced Vegetation Index EVI PHOTOSYNTETIC ACTIVITY Enhance the vegetation ρsignal and a reduce the NIR ρ Red atmospheric EVI = G influence, improving the sensitivity in ρhigh NIR biomass C 1 ρregions Red ρand Blue better + L vegetation monitoring capabilities L8 ATMOSPHERIC CORRECTION (ATCOR) SURFACE REFLECTANCE Normalized difference Flood Index NDFI FLOOD CONDITION Sensitive NDFI = ρ Red ρ SWIR to the presence of surface water ρ Red + ρ SWIR Red Green Ratio Index RGRI LEAF STATUS AND STRESS Indicates the relative expression of leaf spectrum RGRI = ρ Red ρcaused Green by anthocyanin to that of chlorophyll

7 TIME SERIES OF SAR AND OPTICAL FEATURES(1) VARIANCE T1-T7 VARIANCE time series CSK PREPROCESSING CONTRAST T1-T7 CONTRAST time series σ HH T1-T7 σ time series EVI T1-T8 EVI time series L8 PREPROCESSING NDFI T1-T8 NDFI time series RGRI T1-T8 RGRI time series

8 TIME SERIES OF SAR AND OPTICAL FEATURES (2) CSK L8 13/02/2013 T1= /06/2013 T2= /07/2013 T3= /07/2013 T4= /08/2013 T5= T6= T7= /05/2013 T1= /06/2013 T2= /07/2013 T3= /07/2013 T4= /08/2013 T5= /08/2013 T6= T7= T8= /08/ /09/ /09/ /09/ /10/ /12/ /02/2013 T1= /06/2013 T1= CSK+L8 T3= /07/2013 T3= T7= T8= /12/2013

9 THEMATIC LEVELS Two tematic levels of the crop maps were considered DRY SEED RICE EARLY RICE LATE RICE EARLY SUMMER CROPS LATE SUMMER CROPS EARLY MAIZE MEDIUM MAIZE LATE MAIZE FORAGE 2 FORAGE 1 SINGLE WINTER CROPS DOUBLE WINTER CROPS L2 Seeding: April Seeding: May Soybean; seeding: April Soybean; seeding: May SUMMER CROPS Seeding: April Seeding: May Seeding: June RICE MAIZE Continous crop FORAGE X Triticosecale: seeding: February WINTER CROPS Wheat; seeding: November or February X Triticosecale Sorghum DOUBLE or Maize CROPS Lolium Seeding: variable L1

10 SIARL The S.I.A.R.L. (Sistema Informativo Agricolo della Regione Lombardia) is the informative agricultural database of Regione Lombardia used for the consulting and the updating of the documents of the Lombardy farms. Classifiers Training Validation The SIARL database was used for selecting two indipendent sets of pixels for L2 (12 classes). The first set (2/3 of total) was used for training the classificators. The second set (1/3 of total) was used for assessing the accuracy of the produced maps.

11 CLASSIFICATION ALGORITHMS Maximum Likelihood Classificator MLC Spectral Angle Mapper SMA Euclidean Minimum Distance EMD SAR TEMPORAL THEMATIC & OPTICAL LEVELS SERIES FEATURES CSK L8 L8 CSK+L8 VARIANCE MLC T1-T7 L1,L2 CONTRAST T1-T8 L1,L2 EVI T1,T3,T8 L1,L2 EVI RGRI RGRI SAM T1-T7 L1,L2 σ T1-T8 L1,L2 NDFI T1,T3,T8 L1,L2 NDFI EMD T1-T7 L1,L2 T1-T8 L1,L2 T1,T3,T8 L1,L2 σ SAM EMD VARIANCE CONTRAST σ VARIANCE CONTRAST σ EVI RGRI NDFI EVI RGRI NDFI EVI RGRI NDFI σ EVI RGRI NDFI σ For SAR sensor the requirement of Gaussian distributed data is satisfied after speckle filtering.

12 CLASSIFICATION RESULTS According to the approach described in the previous section, crop maps were produced through the classification of each combination of thematic level (L1 and L2), supervised algorithm (MLC, EMD, SAM) and different input time series derived by single L8 or CSK sensors and their combination. CSK+L8;T1 End of June SIARL CSK+L8;T3 End of July CSK+L8; T8 Half December Level 2 crop maps of 2013 of a subset of the study area derived by MLC

13 RESULTS ACCURACY 100% 90% Overall Accuracy (CSK) MLC_L1 100% 90% Overall Accuracy (L8) MLC_L1 80% MLC_L2 80% MLC_L2 70% 60% EMD_L1 EMD_L2 70% 60% EMD_L1 EMD_L2 40% SAM_L1 40% SAM_L1 30% 20% T1 T2 T3 T4 T5 T6 T7 SAM_L2 30% 20% T1 T2 T3 T4 T5 T6 T7 T8 SAM_L2 100% 90% 80% 70% 60% 40% 30% 20% Overall Accuracy (L8+CSK) T1 T2 T3 T4 T5 T6 T7 T8 MLC_L1 The MLC use of shows optical the MLC_L2 best data performance permits a better for EMD_L1 both classification the thematic EMD_L2 levels accuracy and for both all the different the SAM_L1 thematic input levels sensors. and SAM_L2 for the three classificators

14 PERFORMANCE OF MLC: INTEGRATION OF L8 + CSK Commission Error (L2) CSK+L8 Commission Error (L1) CSK+L8 80% 70% 60% Maize E. Maize M. Maize L. Rice D. Rice E. 80% 70% 60% Maize Rice 40% 30% 20% 10% 0% T1 T3 T8 Rice L. Winter S. Winter D. Forage 1 Forage 2 Summer E. Summer L. 40% 30% 20% 10% 0% T1 T3 T8 Winter Double Forage Summer Omission Error (L2) CSK+L8 Omission Error (L1) CSK+L8 80% 70% 60% 40% 30% 20% 10% Maize E. Maize M. Maize L. Rice D. Rice E. Rice L. Winter S. Winter D. Forage 1 Forage 2 80% 70% 60% 40% 30% 20% 10% Maize Rice Winter Double Forage Summer 0% T1 T3 T8 Summer E. Summer L. 0% T1 T3 T8

15 RICE MAPPING USING COSMO SKYMED DRY SEED RICE EARLY RICE LATE RICE RICE RICE EARLY SUMMER CROPS LATE SUMMER CROPS EARLY MAIZE MEDIUM MAIZE LATE MAIZE FORAGE 2 FORAGE 1 SINGLE WINTER CROPS DOUBLE WINTER CROPS SUMMER CROPS MAIZE FORAGE WINTER CROPS DOUBLE CROPS OTHER L2 L1 L0

16 PERFORMANCE OF RICE MAPPING USING CSK 100% 90% Overall Accuracy (CSK) 80% 70% MLC_L1 MLC_L0 60% T1 T2 T3 T4 T5 T6 T7 40% Commission error (L0) CSK 40% Omission error (L0) CSK 30% 20% 10% Rice Other 30% 20% 10% Rice Other 0% T1 T2 T3 T4 T5 T6 T7 0% T1 T2 T3 T4 T5 T6 T7

17 CONCLUSION A mapping project carried out using both optical and SAR data on an agricultural area in northern Italy where described. Results show that the classification accuracy obtained using only multispectral optical data is higher than the one reached using only SAR as input. The performance in the classification carried out using only SAR data have been partially invalidate by: - steep look angle (24.13 ) that limits the sensitivity toward vegetation characteristics - single polarization - worst temporal resolution than optical data. Integrating both optical and SAR multitemporal features provides some advantages in terms of a more reliable crop map, especially during an early temporal stage scenario. Among the supervised algorithms tested, Maximum Likelihood shows the best overall accuracy performances at each thematic level, time step and using both optical and SAR input data. A more sophisticated approaches able to synthetize the temporal variability of the SAR signal should be further investigated.

18 THANK YOU