c o p e r n i c u s l a n d m o n i t o r i n g s e r v i c e · 2019. 7. 25. · accuracy of...

Copernicus EU

Copernicus EU www.copernicus.eu

Copernicus EU

C o p e r n i c u s L a n d M o n i t o r i n g S e r v i c e

Grassland Harvest Estimation (Bavaria, Germany)

LandMonitoring

G R A S S L A N D H A R V E S T E S T I M A T I O N

• Grassland is one of the main (indirect) protein sources in Bavaria (Southern Germany), important for producing milk and meat.

• One third of the agricultural land of Bavaria is grassland, mostlyin the Alpine Foreland.

• Cattle and dairy farming is one of the most important sources of agricultural income.

• Grassland is also a growing energy source for bio-gas production.

• Even slight changes in grassland harvest have strong impacts on the milk/meat and energy market and related prices.

2

Introduction

LandMonitoring

• Variation of 3-6 cuts per year forintensively used grassland areas

• Current estimations based on few surveyson harvest intensity and amount.

• Harvest estimation of individual parcelscan deviate by 100% from real conditions

• Aim: Overcoming the existing information gap, in a spatially explicit manner, contri-buting to federal reporting obligations

• The grassland harvest in Bavaria can currently only be estimated roughly, on county level.

• No weighing system or reporting obligation exists.

3

The Challenge

Source: LfL


LandMonitoring

• The Bavarian agricultural administration wants to:a) know the number of grassland harvests per season

b) use / improve a model for grassland productivity

• The case study has been implemented by:a) Landesamt für Landwirtschaft LFL (state authority for agricultural

research) providing test areas for verification / validation

b) CAU (University of Kiel) for modelling grassland productivity

c) Service providers (GAF AG / e-GEOS) for radar remote sensing data processing

4

Introduction of use case


LandMonitoring

5

Available Input Data

• COPERNICUS can provide information on Grassland through:

– the High Resolution Layer (HRL) “Grassland” (http://land.copernicus.eu/pan-european/high-resolution-layers/grassland/view): currently HRL 2012; large improvements expected from HRL 2015

– The Copernicus Local Land component products: Natura2000, Riparian Zones, Urban Atlas – comprising VHR grassland polygons, amongst others

– Satellite Radar Data from Sentinel-1 and Copernicus Contributing Missions (here: COSMO-SkyMed) – to analayse dates and quantity of harvests

• IACS – ‘Integrated Administrative and Control System’: administrative information system on farm and field level from the agricultural administrations for EU subsidies and environmental measures


http://land.copernicus.eu/pan-european/high-resolution-layers/grassland/view

LandMonitoring

6

Change detection approaches based on SAR data:

1) Change detection by comparing pairs of images (of the same sensor) in a time series:

➢ Fixed geometries of the sensors allow four data acquisitons every few days

→ image-to-image comparison, depending on sensor e.g. at Days 1, 3, 4, 8, 9, 12, etc.

(click Video)

2) Improvement/adjusting of timelines by combining and comparing the detected changes of different times series➢ Variable geometries of the sensors allow for up to four data acquisitions per day

→ improving the data acquisition cycles, as needed by the respective applications

(click Video )


CLMS_Submodule_F_1403_025_AR_EN.mp4

CLMS_Submodule_F_Cosmo4.avi

LandMonitoring

7

Grassland harvest as seen by SAR:


LandMonitoring

8

Grassland harvest as seen by SAR:


LandMonitoring


Detecting Grassland cut dates

LandMonitoring

Demonstration site ‚Bruckmühl‘ in southern Bavaria – parcel-based changes in May:

10

© GAF AG 2016

Ratio: 1.5 – 9.5.Ratio: 2.5 – 10.5.Ratio: 9.5 – 17.5.Ratio: 10.5 – 14.5.Change: 1.5 - 9.5.Change: 2.5 - 10.5.Change: 9.5 - 17.5.Change: 10.5 - 14.5.

IACS parcels/field boundaries

Ratio: change of signalfrom image at date 1 to image at date 2


LandMonitoring

• Demo ‚Bruckmühl‘

– Retrieved harevest dates of observed parcels: 1st - 17th May 2015– Data: COSMO-SkyMed, 3 pairs of data, 2 observation geometries

11© GAF AG

Time of harvestMay 2015:

1. – 2.5.


1. – 2.5.

2. – 9.5.


1. – 2.5.

2. – 9.5.

9. – 10.5.


1. – 2.5.

2. – 9.5.

9. – 10.5.

10. – 14.5.


1. – 2.5.

2. – 9.5.

9. – 10.5.

10. – 14.5.

14. – 17.5.


1. – 2.5.

2. – 9.5.

9. – 10.5.

10. – 14.5.

14. – 17.5.

unknown

IACS Parcels (reference parcels)


LandMonitoring

Harvest Intensity


Source: LfL

LandMonitoring

13

S127.4.-9.5.

S14.5.-16.5.

CSK1.5-9.5.

CSK9.5.-17.5.

"Cuts" correctly captured 153 317 155 156"No Cuts" correctlycaptured

147 2 157 136

"Cuts" not captured 16 13 9 16"Cuts" erroneouslycaptured

24 7 19 32

Detection Rate 88,2% 92,3% 91,8% 85,9%

Accuracy of Grassland harvest detection, May 2015


∑ 340 Test areas

LandMonitoring

14

• Determining quantity and quality of grassland growth phasesbetween the cuts through harvest models

• Using a combination of weather data, agrometeorologicaldata and crop type information

• Alternative Approach (still a research topic): directmeasurement of biomass between cuts by radar remote sensing

Modelling of grass/biomass growth in-between cuts


LandMonitoring

15


LandMonitoring

16


• Measured vs. modelled dry matter development of grassland

Dry Matter production (g/m²)

c o p e r n i c u s l a n d m o n i t o r i n g s e r v i c e · 2019. 7. 25. · accuracy of...

Documents