Inter-annual variation in cattle turn out dates on Irish dairy farms and the relationship with satellite derived grassland performance indices and rainfall. Stuart Green, Teagasc, Dublin Edward Dwyer, EurOcean, Portugal Fiona Cawkwell, UCC, Cork

The question we asked: When farmers turn out cattle from winter housing are they responding to the current conditions on the farm and do they adjust their behaviour accordingly? Stuart Green. IEOS 2015 Galway. 2 of 20

Previous Studies Extending the grazing season reduces costs. In a survey of Irish Dairy farmers in 2008 found the average grazing season length was 245 days. With respect to turning out grass availability and soil condition were the main factors in the timing of the decision. Extended grazing has been examined within the context of technical adoption theory. Agricultural education and off-farm employment had the most significant positive relationship with extended grazing and past participation in agri-environment schemes had the strongest negative effect. An analysis of one year (2009) of the data set presented here found that geographic region and soil status were strongly associated with length of grazing season but that neither farm size or stocking density had a relationship with grazing season length. Stuart Green. IEOS 2015 Galway. 3 of 20

We used data from the Geo- coded National Farm Survey. Data from 300+ dairy farmers recorded the date they turn out cattle for 5 years, This gives 1536 turn out events. Average TOD is Day 61 and average range is 25 days. Stuart Green. IEOS 2015 Galway. 4 of 16

Farming by calendar? 5 of 20

Watching grass grow from space Satellites can record different wavelengths of light as separate components of an image. Each pixel in the image records separately red, green, blue and NIR light. We can manipulate this to tell us about the growing vegetation. 6 of 20

Normalized Difference Vegetation Index (NDVI) = Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec 0 1 NDVI>0.01 every week 8 of 20

MODIS on the NASA Terra & Aqua satellites The whole of Ireland is captured in a single image every other day with a resolution of 250m These Images are automatically processed into different products. The selected MOD13Q1 product is a 16-day composites where, in order to overcome cloud cover, the NDVI value for only high quality, cloud free pixels are used and the compositing algorithm selects the value that best represents that 16 day period for each pixel Thus from Jan 1 st to May 15 th there are 9 images each year. The farm locations in the study were overlaid on top of the images and the corresponding NDVI score extracted so each farm has 9 NDVI scores each year (GNFS re-projected from ITM to WGS84) We acknowledge the use of data products or imagery from the Land, Atmosphere Near real-time Capability for EOS (LANCE) system operated by the NASA/GSFC/Earth Science Data and Information System (ESDIS) with funding provided by NASA/HQ. 7 of 20

The environmental conditions at turn out: Rainfall Each GNFS farm used in the analysis had daily Rainfall data, ascribed to it. The average total rain fall (in mm) from the 3 stations closest (average distance between station and farm was 7.5km). Each farm therefore had 5*365 rainfall records which were processed to give the total rainfall in each 16day period and total number of dry days in each period in each year. 9 of 20

VariableObsMeanStd. Dev. MinMaxDescription TOD Turn Out Day (Julian Day of Year) meanvi Average NDVI Jan 1- May8 totrain Total Rain Jan1-May8 (mm) totdry Total Number of Dry Days Jan1-May8 truevi Acutual NDVI at TOD trurain Total Rain 16days prior to TOD (mm) trudry Total number of Dry Days 16 days prior to TOD totr_ Total Rain Jan1st- Jan16 (mm) totr_ Total Rain Jan17-Feb1 (mm) totr_ Total Rain Feb2-Feb17 (mm) totr_ Total Rain Feb18-Mar5 (mm) totr_ Total Rain Mar6-Mar21 (mm) totr_ Total Rain Mar22-Apr6 (mm) totr_ Total Rain Apr7-Apr22 (mm) totr_ Total Rain Apr23-May8 (mm) totr_ Total Rain May9- May25 (mm) VariableObsMeanStd. Dev. MinMaxDescription ndvi_ NDVI Jan1st-Jan16 ndvi_ NDVI Jan17-Feb1 ndvi_ NDVI Feb2-Feb17 ndvi4_ NDVI Feb18-Mar5 ndvi_ NDVI Mar6-Mar21 ndvi_ NDVI Mar22-Apr6 ndvi_ NDVI Apr7-Apr22 ndvi_ NDVI Apr23-May8 ndvi_ NDVI May9-May25 dry_ No. Dry Day Jan1st-Jan16 dry_ No. Dry Day Jan17-Feb1 dry_ No. Dry Day Feb2-Feb17 dry_ No. Dry Day Feb18-Mar5 dry_ No. Dry Day Mar6-Mar21 dry_ No. Dry Day Mar22-Apr6 dry_ No. Dry Day Apr7-Apr22 dry_ No. Dry Day Apr23-May8 dry_ No. Dry Day May9-May25 10 of 20

11 of 20

Stuart Green. IEOS 2015 Galway. 12 of 16 Farm IDYearTOD NDVI_1 … NDVI12 9 TOTR1 … TOTR1 29 DRY1… …DRY129TOTRAIN TOTDR Y MEAN VITRUVI TRURai n TRUDR Y …… To ensure a balanced panel for such a short time span, only farms with 5 years of correct records were included – 199 farms

Analysis We used a fixed effect model to analyse the panel as we are concerned with inter-annual variation not causes of variation between framers (This was confirmed by the application of a Hausmann test strongly suggesting the rejection of a random effects model) However we used a standard geospatial statistical test to look at any geographic relationship to inter-annual changes in TOD. Moran’s I test is a test of spatial autocorrelation, whether a spatial characteristic is random (a value of zero in the test) with respect to location and neighbours, perfectly dispersed (-1) or entirely dependent on location (1). For ease of interpretation these values are transformed into a Z-score with 5% significance. We calculated Moran’s I over a range of maximum distances 13 of 20

Morans’s I results A strong relationship between location and average TOD No relationship between inter-annual variation in TOD and location I.E. there are no locations in Ireland where TOD is more volatile, year on year, than other locations 14 of 20

Variable associated with Turn Out Day (Julian Day of Year) Coefficient (t) Variable associated with Turn Out Day (Julian Day of Year) Coefficient (t) Total Rain Jan1st-Jan16 (mm) (0.05) No. Dry Day Jan1st-Jan (1.26) Total Rain Jan17-Feb1 (mm) 0.025(1.71) No. Dry Day Jan17-Feb (0.60) Total Rain Feb2-Feb17 (mm) 0.008(0.37) No. Dry Day Feb2-Feb (1.33) Total Rain Feb18-Mar5 (mm) (0.78) No. Dry Day Feb18-Mar (0.10) Total Rain Mar6-Mar21 (mm) 0.009(0.38) No. Dry Day Mar6-Mar (1.88) Total Rain Mar22-Apr6 (mm) 0.109(2.76)** No. Dry Day Mar22-Apr (0.87) Total Rain Apr7-Apr22 (mm) 0.023(0.73) No. Dry Day Apr7-Apr (1.37) Total Rain Apr23-May8 (mm) (2.64)** No. Dry Day Apr23-May (1.33) Total Rain May9-May25 (mm) 0.015(0.54) No. Dry Day May9-May (0.29) NDVI Jan1st-Jan (1.14)Constant81.505(5.76)** NDVI Jan17-Feb (0.03) NDVI Feb2-Feb (1.61) NDVI Feb18-Mar (2.03)* NDVI Mar6-Mar (1.70) NDVI Mar22-Apr (1.81) NDVI Apr7-Apr (1.37) NDVI Apr23-May (1.42) NDVI May9-May (11.54)** Results of robust fixed effects model for all variables with TOD as dependent variable Observations= 995. Panel ID FARM_CODE=199.Time ID Years=5 Within R 2 =0.387 (F=9.57***).Absolute value of t-statistics in parentheses * p<0.05; ** p< of 20

Variables associated with Turn Out Day (Julian Day of Year) with year dummies Coefficient (t) Average NDVI Jan 1-May (11.74)** (10.70)** Total rain Jan1-May8 (mm)0.036(6.77)**0.015(1.96)* Total number of Dry Days Jan1-May80.245(4.54)**0.093(1.60) Actual NDVI at TOD (11.26)** (11.37)** Total rain 16days prior to TOD (mm)-0.079(4.68)**-0.082(4.84)** Total number of dry days 16 days prior TOD (2.84)**-0.518(3.19)** Year Dummy (0.703) (2.83)** (0.83) (-4.13)** Constant98.246(8.70)**83.273(6.54)** p<0.01**,0.05* Stuart Green. IEOS 2015 Galway. 16 of 20 Observations=995. Panel ID FARM_CODE=199.Time ID Years=5 Within R 2 =0.323 (F=35.59***), with Year Dummies R 2 =0.363 (F=24.91***)

If average spring NDVI increases by 0.01 (i.e. spring is a week early) then TOD is 3.6 days earlier But If NDVI at TOD is 0.01 higher then this is associated with TOD that is is 3.3 days later (note that for the “ideal” responsive farmer there should be no relationship between the NDVI at turn out and the date) For every 10mm extra rain that falls in the period before TOD the TOD is 0.8 days earlier. For every extra dry day in the 16 day period up toTOD then TOD is 0.5 days earlier. The national condition seems to dominate over the local – good (2012) and bad (2010) springs. Results 17 of 20

Only 37% of the inter-annual variation can be put down to grass and rainfall (at least as described in the data we used). National conditions are as important as local Farmers are only responding to large scale deviations from the average Stuart Green. IEOS 2015 Galway. 18 of 20

An interpretation Farmers are responding to local conditions They respond to good years and bad years but don’t respond as quickly as possible. If spring is a “week early” they gain about 3.5 days grazing but lose a possible extra 3.3 days. In poor springs when TOD are late, they are choosing to turn out despite soil condition. 19 of 20

Next… We can see that the work here can quantify by end of April how good or bad the national conditions have been for turning out and we can build on this.. Develop an in/out model for the decision based on current conditions. A full spatial/temporal model needs to be developed. Bring in previous autumn conditions from satellite into decision making model of 20 Any Questions ?