Enhancing the Scale and Relevance of Seasonal Climate Forecasts - Advancing knowledge of scales Space scales Weather within climate Methods for information creation Pure dynamical systems Model output statistics Empirical predictors, lead-time issues Issues for practical improvement N. Ward - IRI acknowledgments to colleagues at IRI and partners this presentation with L. Sun and A. Robertson Climate Prediction and Agriculture: Advances and Challenges, WMO, Geneva, May 11 th, 2005
Collaborative Work in Regions
Skill of Model Hindcasts Using Observed SST
Part 1: Advances in Understanding of Predictability at Smaller Spatial and Temporal scales (a) Space Scales
Example of driving a Regional Climate Model with output from a Global Climate Model. Surface Wind at One Time Step DYNAMICAL DOWNSCALING
RSM Precipitation Forecast from Jan for Feb-Mar-Apr (Avg of 10 ensemble) Precipitation Observed (mm) Precipitation Forecast (mm) Correlation 0.79
Regional models can represent influence on local climate from detailed landscape – e.g. elevation, land cover type …
Even in this situation, how to estimate predictability at the field scale? Quantifying decline in skill at smaller scales: General: Barnston et al NE Brazil example: Sun et al
Leading pattern of small-scale rainfall anomalies over Ceara (a)Observed (b)Regional Model Hypothesis: Local physiography induces systematic variability features
RSMRSM RSMRSM RSMRSM Contingency tables for 3 subregions of Ceara State at local scales (FMA ) OBS Coast BNA B 532 N343 A235 RSMRSM RSMRSM RSMRSM Contingency tables for 3 subregions of Ceara State at local scales (FMA ) OBS Coast BNA B 532 N343 A235 RSMRSM RSMRSM RSMRSM Contingency tables for 3 subregions of Ceara State at local scales (FMA ) OBS Coast BNA B 532 N343 A235
Statistical Downscaling Results for Sri Lanka, Verification Map shows correlation skill (shading) along with contours of elevation
Statistical Downscaling Results for Senegal, Verification Map shows correlation skill (red positive) for Seasonal rainfall (upper) And NDVI (lower)
Large-scale predictability does cascade into predictability at smaller spatial scales There is need to represent the local physiographic forcing to best estimate the small scale seasonal climate
Part 1: Advances in Understanding of Predictability at Smaller Spatial and Temporal scales (b) Weather within Climate
Predictability of weather statistics through the season ……
Predictability of the interannual variability of weather statistics over Ceara, NE Brazil Blue = Observed, Pink dash = Predicted by RSM (no statistical correction) Number of Dry spells Longest Dry spell Number of Days without rain
Model Simulation vs. Observation Seasonal Rainfall total R=0.84 Drought Index R=0.74 Flooding Index R=0.84 Weather Index R=0.69
Rainfall states (S Georgia / N Florida, USA) HMM rainfall parameters “learned” from the data Rainfall occurrence probability Average rainfall amount on wet days (from parameters of mixed exponential distribution) Illustration of concepts in statistical downscaling to weather series: (From a study using the Hidden Markov Model approach)
Estimated state sequence March to August seasonality, sub-seasonal and interannual variability
Estimated state sequence for March-May rainfall in Kenya MarchAprilMay - “dry” state (#3, yellow) tends to occur in March - “wet” states (#1, green), (#2, blue) tend to occur in April–May To get rainfall sequence: P(R t | S t )
Predictability of seasonal means does cascade into predictability of weather statistics through the season Rainfall onset involves the specific timing of a set of weather events. The limit of forecasting the specific timing of weather events is about 2 weeks However, it is reasonable to think that information about the likelihood of a set of weather events over a certain time- period could be provided in situations where there is strong SST forcing on the large-scale circulation Furthermore, the possibility for projecting forward information about large-scale intraseasonal structures is open to further analysis
Part 2: Tools for Prediction
Precipitation Forecast FMA 2004, using persisted SST Note: not the raw model output - already an element of statistical transformation of model output
EOF 1 of 850mb Oct-Dec zonal wind from GCM (ECHAM4) GCM was driven with observed SST To be used as predictor for observed 20kmx20km rainfall over Sri Lanka Statistical Transformation/Downscaling Methods can be applied to the output of all categories of dynamical prediction systems
Statistical Downscaling Results for Sri Lanka, Verification Map shows correlation skill (shading) along with contours of elevation
Statistical Downscaling to NDVI Using a GCM with Sept SST to predict December vegetation (about 25km resolution) across East Africa Spatial variations in skill may reflect -variations in climate predictability -variations in climate-NDVI coupling Hypotheses to explore using RCMs. Time series of area-average predicted NDVI over NE Kenya (r=0.76) Units are correlation skill Contours are elevation Corrected high resolution NDVI provide by USGS
Climate Predictability Tool (CPT)
Example of Reservoir Inflow in Ceara, NE Brazil Probabilistic forecasts based on 2 SST indices in July of previous year Model trained on data Annual Inflow Forecast Year
Part 3: Some Further Key Issues for Practical Improvement in SI Prediction Systems Lead-time (SST development) Land surface (initial conditions, interaction) Presence of Low-frequency Climate
UKMO model, results published early 1990s Early example of 2-tier GCM forecast experiments using persisted SSTA – Sahel Seasonal Rainfall Total Sensitivity of skill to SST development from April to June
Example of Reservoir Inflow in Ceara, NE Brazil Probabilistic forecasts based on 2 SST indices in July of previous year Model trained on data Annual Inflow Forecast Year
Exploring Enhancement of Predictability from Global Initial Soil Moisture Conditions
The NCEP RSM Land Module
Enhancing the Scale and Relevance of Seasonal Climate Forecasts - Advancing knowledge of scales Space scales Weather within climate Methods for information creation Pure dynamical systems Model output statistics Empirical predictors, lead-time issues Issues for practical improvement N. Ward - IRI acknowledgments to colleagues at IRI and partners this presentation with L. Sun and A. Robertson Climate Prediction and Agriculture: Advances and Challenges, WMO, Geneva, May 11 th, 2005
Reservoir Management Tool Input: Probability Seasonal Forecasts and Reservoir System Properties Output: Properties of Reservoir operation With and without Seasonal forecasts