1 SIS Meeting 17-19 Oct 2016 AgriCLASS Agriculture CLimate Advisory ServiceS Phil Beavis - Telespazio VEGA UK Indicators, Models and System Design Michael Sanderson - UK Met Office Climate Data and Weather Implemented by
2 Input (ECVs) YEAR Julian Day T max ( C) T min ( C) Rain (mm) Series of daily weather vector: (Tx, Tn, PP, ) for each location of interest (or grid point). T is surface air temperature at 2m above ground level. Sources: Direct observation from weather station Gridded downscaled data (E.g. E-OBS + WorldClim) CMIP5 model data Synthetic data (e.g. calibrated obs data, Weather Generator)
3 Bio-climatic Indicator (Annual) Period Tn Tx T m Average maximum temperature of hottest months Jul-Oct * Degree day accumulation (base 8.99 C) in winter Nov-Feb * * Degree day accumulation (base 8.99 C) in spring Mar-Jun * * Degree day accumulation (base 30 C) of maximum daily temperature Jul-Oct * ET0 reference evapotranspiration (mm day-1) Jul-Oct * * Hydro-climatic balance - ET0 reference minus Precipitation accumulation Jul-Oct * * * Mean temperature in winter Nov-Feb * Mean temperature in spring Mar-Jun * Mean temperature in summer Jul-Oct * Minimum temperature (average) of winter Nov-Feb * Minimum temperature (average) of spring Mar-Jun * Mortality estimated by 2*10-6 * etemp0.2539 Jul-Oct * * Number of frost (< 0 C) days in winter Nov-Feb * Number of frost (< 0 C) days in spring Mar-Jun * Number of hot (> 30 C) days in summer Jul-Oct * Precipitation accumulation Jul-Oct * Spring flight: day of year of 379 degree day accumulation (base 8.99 C) Nov-Feb * * Standardized Precipitation-Evapotranspiration Index (SPEI) Sep-Jul * * * Minimum temperatures between January and March Jan-Mar * Maximum temperatures between March and April Mar-Apr * Maximum temperatures between June and July Jun-Jul * Mean temperatures from March to April Mar-Apr * Day of Year of Degree Day Accumulation N (base 10 C) Jan-Dec * * Huglin Index (base 10 C) Apr-Sep * * P P
Case Study 1 Beech Forest Based on Dendrochronology. Tree Ring Width Index (RWI) can be predicted from Standardised Precipitation- Evapotranspiration Index (SPEI). SPEI is computed annually for the 11 months up to July, using: daily maximum temperatures, ºC daily minimum temperatures, ºC precipitation, mm latitude of the site, in degrees Under dryer conditions (SPEI<SPEI up ): RWI = A d x SPEI(11,Jul) + B d When climate conditions are wetter (SPEI>SPEI up ): RWI = A w x SPEI(11,Jul) + B w 4
5 Case study 2 - Olives Olive Fruit Fly Infestation Index: START_INF (%) Jul Aug AVG_INF (%) Jul - Oct DAM_INF (%) Sep - Oct Steps: 1. Generalized Linear Model (GLM) for the effects of each bioclimatic index 2. Principal Component Analysis (PCA) reduces number of indices, removes co-linearity 3. Principal Component Regression (PCR) - GLM implements results of PCA Bioclimatic Indices PC1 PC2 T_med_Spr -0.32-0.48 T_med_Win -0.37 0.30 Fr_days_Spr 0.31 Fr_days_Win 0.33-0.34 T_min_Spr -0.34-0.26 T_min_Win -0.36 0.32 DD 8.99 _Win -0.34 0.19 DD 8.99 _Spr -0.27-0.60 Doy_Spr 0.34 Eigenvalue 6.0519 1.4198 Stand. Dev. 2.46 1.19 Variance (%) 67.24 15.77 Cum. Var. (%) 67.24 83.01
6 Case study 3 - Vines Vine productivity is driven by a series of phenological stages: Stage Baggiolini index Eichhorn & Lorenz index BBCH index Budbreak C 5 07 Flowering I 23 65 Veraison M 36 85 Maturity N 38 89 Day of Year for each stage can be predicted from bioclimatic indicator: Sum of Degree Days (SDD) Huglin Index (HI)
7 Generalised Methodology 1. Climate: Generate multiple realisations of future daily weather data (ECV) for each point of interest (site or grid point), for each CMIP5 model and each emissions scenario. 2. Concept: Extract seasonal weather features as bio-climatic (Tier-1) indicators. 3. Proof of Concept: Develop and run correlation models to predict cropspecific (Tier-2) indicators. DS5 Synthetic daily weather for each site, model, epoch (LARS format) DPT5 Compute Bioclimatic Indicators DS6, DS7 Annual Bioclimatic Indicators for each site, model, epoch DPT8 Run Crop Model DS8 Predicted Crop Indicators for each site, model, epoch
8 Architecture Temp max Temp min CMIP Climate Data Weather Obs Crop Obs Tree Ring Infestation Rainfall Phenology Compute Platform SPEI Frost Days Degree Days Bioclimatic Indicators Crop Impact Indicators Tree Ring Infestation Phenology Huglin index Winkler index etc.. Web Portal
9 System Data Flow Climate and Weather Agriculture CMIP5 Federated Databases Weather Observation Data Crop Observation Data DPT1 Ingest Climate Data (wget scripts) DPT3 Analyse Weather (monthly statistics) Ingest Weather Observation Data Ingest Crop Observaton Data DS3 Selected CMIP model variables for EU in NetCDF gridded format DS4A Climate Statistics and Change Factors (CF) for each site, model, epoch DS1 Observed Weather Data (daily variables in LARS format) DS2 Observed Crop Data (annual indicators in CSV format) DPT2 Compile Climate Data for each site and model (interpolation?) DPT4A Generate Weather (apply CF to weather obs) DPT7 Develop / Learn Crop Model Crop Model and Parameters DS4 Compiled CMIP daily weather for each site and model (LARS format) DS5 Synthetic daily weather for each site, model, epoch (LARS format) DPT5 Compute Bioclimatic Indicators DS6, DS7 Annual Bioclimatic Indicators for each site, model, epoch DPT8 Run Crop Model DS8 Predicted Crop Indicators for each site, model, epoch Presentation DPT4B Generate Weather (Replicate similar data using WG) C3S Climate Data Store DPT6 Export Bioclimatic Indicators AgriCLASS web portal (proof of concept demonstrator)
10 Climate Data & Weather Michael Sanderson UK Met Office Implemented by
12 CMIP5 ensemble Simulations of historical and future climate by ca.30 global climate models. Climate projections available under four different scenarios. Extensively assessed in scientific literature and IPCC reports. Freely available for commercial use. Not necessary to use all 30+ models. Subset of 10 models recommended: Reproduce key large scale circulation patterns for Europe Span (about) same range of possible future climates as entire ensemble.
13 Winter wind speeds and direction at 850 hpa
14 Selection of CMIP5 models Text and list Solid circles selected models (10). Grey triangles rejected models
15 Combining observed and modelled climate data All climate models contain biases. Need to combine observations with climate data to reduce these biases. Wide range of methods of varying complexity in the literature. Largest differences between methods generally seen in the tails for example, very warm or very cold temperatures. Choice to be made: Create climate change factors and apply to observations; Or, calculate correction factors and apply to model data Data needed: Daily weather variables from observations and models over a common period. Daily data from models for future period(s).
Fig. 2. Schematic of the two general types of calibration. (a) Bias correction uses raw model output and corrects it using the differences (Δ) between reference data from the model and observations. If no correction is used then this is the RAW method. (b) Change factors are calculated from raw model data and added to the observations (from Hawkins et al., Agricultural and Forest Meteorology, Volume 170, 2013, 19 31)
17 Contrast the two methods from Räisänen J & Räty O, Clim Dyn (2013).
18 Bias correction: Apply to model data Advantages Creates a continuous series of corrected model data Disadvantages Any errors in model data, e.g., persistence of wet or dry periods, warm temperatures will still be present Model data may not represent climate of location of interest accurately, esp. in complex terrain.
19 Delta-Change: Create climate change factors and apply to observations Advantages Use of climate change factors partially removes model biases Observations are real events Disadvantages Future climate data will partly resemble the observations Future series will only be as long as the observed series
20 Chosen method Apply climate change factors to observations Climate data are from global models so unlikely to represent climate at actual sites. T(fut) = Mean(Tmod) + sd(mod) / sd(ref) [O(t) Mean(Tref)] Where: T(fut) daily series of corrected temperatures for the future T(mod) raw daily series of modelled temperatures for future sd(mod) Standard deviation of future modelled temperatures sd(ref) - Standard deviation of baseline modelled temperatures O(t) Daily temperatures from observations T(ref) - raw daily series of modelled temperatures for baseline Baseline = 1991-2010, Future = 2046-2065 Apply on a monthly basis
21 Precipitation Bias correction of precipitation more difficult than for temperatures Distribution of rainfall totals non-normal. Some methods may produce negative values. P(mod) modelled precipitation P(corr) = a.p(mod) ^ b Parameter b chosen so that coeff of variation of modelled data equals coeff of variation of obs data over common time period. Parameter a then chosen so mean of transformed values equals mean of obs. hence a depends on b.
22 Weather Generator: ClimGen Produces synthetic time series of weather data of unlimited length for a location based on the statistical characteristics of observed weather Most generate series of precipitation and then infer values of other variables. Many weather generators have been created but very few are freely available and/or may be used commercially. ClimGen was recommended based on a review of weather generators. ClimGen could be used to effectively extend the series of observations.
23 Thank you climate.copernicus.eu/ copernicus.eu/ Implemented by