2. Contents Background: climate and agriculture Future climate and GCMs Downscaling methods Disaggregation CCAFS-T1 / CIAT-DAPA data inventory CCAFS climate data strategy
3. Climate and agriculture Information on climate is critical for agriculture, because: 1. Agriculture is a niche-based activity 2. Abiotic factors (i.e. climate, soils) and their interactions are main drivers Location Performance Adaptive responses Management practices 3. Weather and climate predictability is fairly limited 4. Each system is an specific case, so is its future…
4. Climate and agriculture Agriculture demands: Multiple variables Very high spatial resolution Mid-high temporal (i.e. monthly, daily) resolution Accurate weather forecasts and climate projections High certainty Both for present and future
8. GCMs: How do we predict the future? GCMs are the only means we have to predict future climates… ~24 exist up to now All different… so we can expect issues
18. So we need downscaling Even the most precise GCM is too coarse (~100km) To increase resolution, uniformise, provide high resolution and contextualised data Different methods exist… from interpolation to neural networks and RCMs DELTA (empirical-statistical) DELTA-VAR (empirical-statistical) DELTA-STATION (empirical-statistical) RCMs (dynamical) …
19. Why do we need higher resolution data? Temperature Ethiopia Rainfall
20. The delta methodHay et al. 2007 Use anomalies and discard baselines in GCMs Climate baseline: WorldClim Used in the majority of studies Takes original GCM timeseries Calculates averages over a baseline and future periods (i.e. 2020s, 2050s) Compute anomalies Spline interpolation of anomalies Sum anomalies to WorldClim
22. Delta-VARMitchell et al. 2005 AKA pattern scaling Climate baseline: CRU Provided by Tyndall Centre (UK) Use captured variability in GCMs (MAGICC)and anomalies Run a new GCM pattern at a higher resolution (CLIMGEN) Calculate averages over specific periods using the GCM scaled time-series
23. Delta-StationSaenz-Romero et al. 2009 Most similar to original methods in WorldClim Climate baseline: weather stations Calculate anomalies over specific periods (i.e. 2020s, 2050s) in coarse GCM cells “Update” weather station values using GCM cell anomalies within a neighborhood (400 km) Inverse distance weighted Use thin plate smoothing splines with LAT,LON,ALT as covariates for interpolation
24. RCMs: PRECISGiorgi 1990 RCMs (Giorgi 1990) Climate baseline: GCM boundary conditions Develop complex numerical models to simulate climate behaviour “Nest” the RCM into a coarse resolution model (GCM) and apply equations to re-model processes in a limited geographic domain Resolution varies between 25-50km Takes several months to process Requires a new validation (on top of the GCM validation)
25. Disaggregation Similar to the delta method, but does not use interpolation Climate baseline: CRU, WorldClim Calculate anomalies over periods in GCM cells Sum anomalies to climate baseline
33. A quickcomparison 1 PRECIS run (10 year) = 2 weeks 1 interpolation (37 steps) x 15 periods = 1 week x 1 GCM x 7 periods x 1 scenario x 20 GCMs 30 weeks x 3 scenarios ÷ 2 processes 210 weeks ÷ 3 servers ÷ 4 processes = 5 weeks ÷ 4 servers x 20 GCM s Hypothetically.. = 26 weeks x 3 scenarios = 6 months!! = 300 weeks = 6 years!!
34. Capabilities and limitations Our in-house capacity: Four 8-core processing servers in a blade array under Windows (empirical downscaling) Three 16-core processing servers in a blade array under Linux (PRECIS) ~80TB storage Publishing data is a lengthy process and requires massive storage and network capacity (esp. 1km global datasets)
39. What’s nextCCAFS climate data strategy Improve baseline data and metadata (incl. uncertainties) Gather and process AR5 projections Downscale with desired methods Evaluate (against weather stations) and assess uncertainties Publish all datasets (original and downscaled) and results Use the AMKN platform to link climate data, and modelling outputs
40. In summary CIAT and CCAFS data to be one single product (other datasets are being added) Downscaling is inevitable, so we are aiming to report caveats on the methods Continuous improvements are being done Strong focus on uncertainty analysis and improvement of baseline data Reports and publications to be pursued… grounding with climate science