Kathleen M. Baker Western Michigan University Midwest Weather Working Group October 7, 2009
RAMP # South Dakota State University
Comparison of techniques for forecasting ANN at point locations vs spatial grids Variability across regions / diseases Cyberinfrastructure options Test cases: Potato late blight ---- MI Fusarium head blight of barley --- MN/SD/ND Leaf spot of peanut --- GA, FL Focus today: MI-PLB, August 2008
KAZO GFSX MOS GUIDANCE 12/05/ UTC FHR 24| 36 48| 60 72| 84 96| | | | WED 05| THU 06| FRI 07| SAT 08| SUN 09| MON 10| TUE 11| WED 12 X/N 26| 14 29| 25 35| 22 32| 28 41| 28 36| 22 36| TMP 20| 17 26| 28 30| 24 30| 33 35| 30 30| 24 31| DPT 16| 13 18| 23 26| 19 24| 29 31| 26 24| 20 24| CLD OV| CL PC| OV PC| OV OV| OV OV| OV PC| OV OV| CL PC WND 11| 7 9| 11 11| 8 6| 13 11| 11 11| 10 13| 15 9 P12 36| 1 4| 63 21| 16 16| 61 55| 42 30| 25 32| P24 | 8| 76| 29| 64| 51| 41| 46 Q12 0| 0 0| 1 0| 0 0| 4 3| 2 0| 0 | Q24 | 0| 1| 0| 3| 2| |
Normals MayJuneJulyAugustSept. Cooler than mean MonthHiLo LatLongNoSpatial Warmer than mean MonthHiLoNoSpatialMonthHiLoLatLong ANN Schematic
The Total 5 Year Accuracy Mean of forecasts – 81% Mean of stations– 79% Median – 79% SD – 3%
How do we create these models in the most efficient way? Randomization between years Addition of climatic normals Necessary data archive How does accuracy compare to more advanced options?
“a research environment that supports integration of geographically distributed computing and information processing services to enable data-intensive collaborative science enterprises” TeraGrid a petaflop of computing capability more than 30 petabytes of online & archival data storage rapid access and retrieval over high-performance networks
Linked Environments for Atmospheric Discovery (LEAD) Portal – $11 mil Democratization of forecasting WRF outputs Hourly grids Varying resolution
ACCURACY AUG 2008 AVGMEDMINMAXSD 71%76%16%95%17% ACCURACY AUG 2008 AVGMEDMINMAXSD 72% 55%87%6%
Variability in accuracy Optimization of spatial resolution Stablization of system Limited dataset Variable selection 2m vs canopy conditions
Gridded WRF forecasts have many advantages over ANN point based model Increased accuracy at critical locations Increased accuracy at critical times Increased spatial resolution Potential to vary spatial resolution Potential to vary height of measurements Expanded variable set Run time flexibility
Compare accuracy and variability for different crops in different regions Compare ANN interpolation from points to spatially gridded forecasts Is the investment worth it?
The high quality foundation of data provided by cyberinfrastructure projects such as LEAD has the potential to truly transform agricultural decision support Cloud computing possibilities…
Proposed Organization of Crop Disease Risk Forecasting System Model workflow Early Warning Reports Modeling services Analysis Forecasts Access to local, regional and global meteorological forecasts Access to local phys-geographic data, incl. data from other CI projects Forcings monitoring of crop data Crop Observatories Validation Assemble model Validatio n data Data conversion services (e.g. projection) Event detection and alert services workflows Neural Network, Excel, GIS, …
Funding: USDA CSREES FQPA RAMP Co-PI’s: Joel Paz, University of Georgia, Ag & Biological Engineering Jeffrey Stein, South Dakota State University, Plant Biology William Kirk, Michigan State University, Plant Pathology Phillip Wharton, University of Idaho, Plant Pathology Dennis Todey, South Dakota State Univ, Ag & Biosystems Engineering Linked Environments for Atmospheric Discovery (LEAD): Kelvin Droegemeier, University of Oklahoma; Beth Plale, Suresh Marru, Felix Terkhorn, Indiana University Research Assistants: Magdalena Wisniewska; Jason Smith; Cassandra Hoch; Doug Rivet; Susan Benston, Steve Schultz