Agricultural Grid Weather Information System based on Digital Weather Forecast in Korea and its Application to Rice Blast Disease Warning International Workshop on the Content, Communication and Use of Weather and Climate Products and Services for Sustainable Agriculture May 17-22, 2009 Kyu Rang KIM, Wee Soo KANG 1, and Eun Woo PARK 1 National Institute of Meteorological Research, Korea Meteorological Administration 1 Seoul National University 0
Outline I. Conventional Ag-Met Information Services in Korea II. Introduction to Digital Weather Forecast III. Current Development of Application Services based on Digital Weather Forecast IV. Outcome and Further Development 1
I. Conventional Ag-Met Information Services 2
I. Conventional Ag-Met Information Services Currently Available AgroMet Information from KMA Chemical Spray Favorableness Index Agricultural Facility Warning Index Mostly Region-based information Higher Resolution Services are Required Due to complex terrain and land use/cover, spatial variation of AgroMet conditions is very large Farmers want their own field-specific data Automated weather stations (AWS) were installed to monitor Ag- Met variables in such highly variable fields 3
I. Conventional Ag-Met Information Services To meet the site specificity AWS in agricultural field Air Temperature Relative Humidity Soil Temperature Rainfall Solar Radiation Wind Speed/Direction Leaf Wetness Soil Humidity Installed in agricultural field as needed by farmers/extension services * KMA installed AWS (>600) generally on roof-top 4
I. Conventional Ag-Met Information Services Examples of AWS based AgroMet Services Rice Paddy (Rice Blast) = very successful Apple Orchard Farmers provide/share significant information on orchard management Pear Orchard, etc Problems in using AWS high cost difficult maintenance limited resolution no real forecasting 5
I. Conventional Ag-Met Information Services Examples of AWS based AgroMet Services Rice Paddy (Rice Blast) = very successful Apple Orchard Farmers provide/share significant information on orchard management Pear Orchard, etc Problems in using AWS high cost difficult maintenance limited resolution no real forecasting 6
I. Conventional Ag-Met Information Services Examples of AWS based AgroMet Services Rice Paddy (Rice Blast) = very successful Apple Orchard Farmers provide/share significant information on orchard management Pear Orchard, etc Problems in using AWS high cost difficult maintenance limited resolution no real forecasting 7
I. Conventional Ag-Met Information Services Examples of AWS based AgroMet Services Rice Paddy (Rice Blast) = very successful Apple Orchard Farmers provide/share significant information on orchard management Pear Orchard, etc Problems in using AWS high cost difficult maintenance limited resolution no real forecasting 8
I. Conventional Ag-Met Information Services Examples of AWS based AgroMet Services Rice Paddy (Rice Blast) = very successful Apple Orchard Farmers provide/share significant information on orchard management Pear Orchard, etc Problems in using AWS high cost difficult maintenance limited resolution no real forecasting 9
I. Conventional Ag-Met Information Services Examples of AWS based AgroMet Services Rice Paddy (Rice Blast) = very successful Apple Orchard Farmers provide/share significant information on orchard management Pear Orchard, etc Problems in using AWS high cost difficult maintenance limited resolution no real forecasting 10
I. Conventional Ag-Met Information Services Examples of AWS based AgroMet Services Rice Paddy (Rice Blast) = very successful Apple Orchard Farmers provide/share significant information on orchard management Pear Orchard, etc Problems in using AWS high cost difficult maintenance limited resolution no real forecasting 11
I. Conventional Ag-Met Information Services Examples of AWS based AgroMet Services Rice Paddy (Rice Blast) = very successful Apple Orchard Farmers provide/share significant information on orchard management Pear Orchard, etc Problems in using AWS high cost difficult maintenance limited resolution no real forecasting 12
I. Conventional Ag-Met Information Services Examples of AWS based AgroMet Services Rice Paddy (Rice Blast) = very successful Apple Orchard Farmers provide/share significant information on orchard management Pear Orchard, etc Problems in using AWS high cost difficult maintenance limited resolution no real forecasting 13
I. Conventional Ag-Met Information Services Examples of AWS based AgroMet Services Rice Paddy (Rice Blast) = very successful Apple Orchard Farmers provide/share significant information on orchard management Pear Orchard, etc Problems in using AWS high cost difficult maintenance limited resolution no real forecasting 14
I. Conventional Ag-Met Information Services Examples of AWS based AgroMet Services Rice Paddy (Rice Blast) = very successful Apple Orchard Farmers provide/share significant information on orchard management Pear Orchard, etc Problems in using AWS high cost difficult maintenance limited resolution no real forecasting 15
I. Conventional Ag-Met Information Services Examples of AWS based AgroMet Services Rice Paddy (Rice Blast) = very successful Apple Orchard Farmers provide/share significant information on orchard management Pear Orchard, etc Problems in using AWS high cost difficult maintenance limited resolution no real forecasting 16
I. Conventional Ag-Met Information Services Examples of AWS based AgroMet Services Rice Paddy (Rice Blast) = very successful Apple Orchard Farmers provide/share significant information on orchard management Pear Orchard, etc Problems in using AWS high cost difficult maintenance limited resolution no real forecasting 17
I. Conventional Ag-Met Information Services Examples of AWS based AgroMet Services Rice Paddy (Rice Blast) = very successful Apple Orchard Farmers provide/share significant information on orchard management Pear Orchard, etc Problems in using AWS high cost difficult maintenance limited resolution no real forecasting Digital Forecast can provide detailed forecast at higher resolution 18
II. Digital Weather Forecast Data from Digital Weather Forecast Digital Weather forecast Components Operational since Oct. 2008 5km, 3hourly forecast Qualitative forecast 48 hours of forecast length 16 layers of forecast data Horizontal domain grid size: 149(E-W) * 253(N-S) = 37,697 AIR TEMPERATURE (T3H) MINIMUM TEMP (TMN) MAXIMUM TEMP (TMX) PROBABILITY OF PRECIPITATION (POP) SKY CONDITION (SKY) WIND DIRECTION (WDD) WIND SPEED (WDS) SIG WAVE HEIGHT (WAV) RELATIVE HUMIDITY (REH) ACC. PPTN IN 12-HOUR (R12) TYPE OF PRECIPITATION (PTY) ACC. SNOW IN 12-HOUR (S12) 19
1. Procedures for Agricultural Digital Forecast A. Data Extraction from Digital Forecast for Agricultural Models Extract three hourly 5km grid data Temporal interpolate to produce hourly 5km grid data Apply additional models to estimate leaf wetness, essential for disease forecast B. Implementation of Application (Plant Disease Development) Models Implementation and optimization of disease forecast and application models Map-based Internet interface for disease forecast and information C. Evaluation of the system Accuracy assessment between AWS data- vs. digital forecast-based disease forecast 20
2. System Overview Input Data & Processing Information Delivery System Disease Forecasting Model Output Data 21
3. Data Storage System - Stores weather data transferred from KMA -Stores interpolated weather data created from JPS sub-system - Provides the data to the JPS and WSS sub-systems 22
4. Job Process System - Executes three jobs hourly 1. Interpolate Data - Interpolate hourly mesh weather data from the Digital Weather Forecast 2. Run Application Models - Calculate the rice blast infection model 3. Render Maps - Converts the forecasting data to map images 23
5. Web Service System - Interacts with users - Presents the data as maps through web map interface in a web site Web Map Interface - Supports panning, zoom-in and zoom-out of the maps - Shows disease forecasting map layer overlaid on layers of digital elevation and district maps 24
6. Input Data Processing Digital Forecast (12) MINIMUM TEMP. (TMN) Input for Disease Forecast AIR TEMPERATURE (T3H) Step 1 Decode Digital Forecast MAXIMUM TEMP. (TMX) WIND DIRECTION (WDD) SIG. WAVE HEIGHT (WAV) TYPE OF PPTN. (PTY) SKY CONDITION (SKY) REL. HUMIDITY (REH) PROBABILITY OF PPTN. (POP) ACC. PPTN. IN 12-HOUR (R12) WIND SPEED (WDS) ACC. SNOW IN 12-HOUR (S12) Step 2 Hourly Interpolate T3H, REH, POP, R12 - Hourly interpolation of 48 hour digital forecast data for T3H and REH - Hourly estimate of rainfall from POP and R12 - Leaf wetness estimation from T3H, RH, and WDS Step 3 Save 5km x 5km Grid Weather Data 25
6-1. Input Processing Hourly Interpolation Air Temp, RH Digital Forecast Air Temp (3-hourly) +1h +4h +7h Linear Interpolation Interpolated Air Temp (hourly) +1h +2h +3h +4h +5h +6h +7h 26
6-2. Input Processing Hourly Interpolation Probability of Precipitation Probability of precipitation for one hour: P 1 Digital Forecast Probability of Precipitation (3-hourly) +1h +4h +7h Hourly Estimation 1 P P 3 P 1 3 1 1 1 P 3 1 1 P 1 1 P 3 3 3 Probability of precipitation for three hours: P 3 Estimated Probability of Precipitation (hourly) +1h +2h +3h +4h +5h +6h +7h 27
6-3. Input Processing Hourly Interpolation Amount of Precipitation Cumulative Amount of Precipitation (12-hourly) + Estimated Probability of Precipitation (hourly) Precipitation Forecast for the 12 hour period +1h +4h +7h +10h +13h +1h +4h +7h +10h +13h Rainfall Estimation Using the Hourly Probability of Precipitation as Weights Estimated Precipitation (hourly) +1h +4h +7h +10h +13h The 12-hourly Precipitation is Distributed to Each Hour 28
6-4. Input Processing Hourly Interpolation Leaf Wetness (Simple RH) RH 95% Hourly Interpolated Relative Humidity + Hourly Estimated Precipitation +4h +7h +10h +13h +16h +4h +7h +10h +13h +16h Rainfall 0.1mm Hourly Estimation Hourly Estimated Leaf Wetness Period RH: Relative Humidity Rain: Rainfall Wet: Leaf Wetness Period IF Rain 0.1mm OR RH > 95% THEN Wet = 1 ELSE Wet = 0 ENDIF Leaf Wetness 1 hour +4h +7h +10h +13h +16h 29
Hourly Interpolated Air Temperature Hourly Interpolated Relative Humidity Hourly Data (19:00-09:00, w/o rainfall) DPD 2 C DPD < 2 C 6-5. Input Processing Hourly Interpolation Leaf Wetness CART (Classification and Regression Tree) model DPT 16.3 C DPT < 16.3 C Hourly Dew Point Temperature (DPT) WS 0.6 m/s WS < 0.6 m/s DPD < 1.4 C DPD 1.4 C Hourly Dew Point Depression (DPD) + Hourly Interpolated Wind Speed (WS) Group 1 No Dew Group 2 Dew Group 3 No Dew Group 4 No Dew Group 5 Dew (Yun et al., 1998) CART Hourly Estimated Leaf Wetness Period +4h +7h +10h +13h +16h Leaf Wetness 1 hour 30
31 7. Application Model (Disease Development Model for Rice Blast) Rice Blast Forecast Model as a AgroMet Application Model - Rice is the most important staple food in Korea (ca. 1,000,000 ha) - Extension services use AWS-based rice blast warning system - Input Variables: Hourly Temperature, Leaf Wetness, Rainfall - Infection Risk (Yoshino, 1979) (Forecasted Wet Hours) - (Base Wet Hours Determined by Temp) 4hr Mean Air Temperature during the previous 5 days = 20~25 Rainfall 4mm/hr - Daily Infection Risk and Warnings Levels Observed daily infection risk hours at 16 locations during 98-03 Max. daily infection risk hours were determined as 14 hours Four levels of daily infection risk hours (R) were determined from 0, 3, and 7 hours, which are 0, 40, and 80 percentile of yearly total infection risk hours, respectively
Yearly Cumulative Infection Risk (hours) 7-1. Application Model (Disease Development Model for Rice Blast) - Warning Level Determination from Climatic Data L e v e l Daily Infection Risk (hours) Infection Probability Warning 1 R = 0 None Zero 2 0 < R < 3 Low Low 3 3 R < 7 Higher Mid 4 R 7 Highest High Daily Infection Risk (hours) Fig. Yearly cumulative frequencies of daily total hours of rice blast infection, observed at sixteen locations in Korea during 1998-2003 32
7-2. Application Model Data Requirement by the Disease Model and Forecast Hours by the Digital Forecast Disease Model Run Hours: +4h ~ +27h (24h Period) Disease Model Output: Disease Infection Warning Hours (0~24h) or Warning Levels (4) Disease Model Run Hours Forecast Hours is Less than 24 Hours -> insufficient Data for the Disease Model 33
8. Web-based Disease Forecasting System - Shows the disease warning forecasts (hours) estimated for the upcoming 24 hour period - Infection risk hours are also show as the four warning levels: Zero, Low, Intermediate, and High - Current target area: Gyeonggi province 34
9. Evaluation of the Forecasts: Preliminary Results and Plans in 2009 AgroMet AWS in Gyeonggi Province - 19 AWSs in rice paddies and upland fields are available. Leaf wetness and other weather elements (Preliminary Evaluation) Disease forecast based on Digital Weather Forecast (Evaluation Plan in 2009) - Disease development will also be monitored and compared by plant pathologists. Disease Forecast based on Digital Weather Forecast DIW-based vs. AgMet AWS-based weather data AWS ( ) h AWS ( ) h h AWS ( ) h AWS ( ) h AWS ( ) h AWS ( ) h AWS ( ) AWS ( ) h AWS ( ) AWS ( ) AWS ( ) h AWS ( ) 2 AWS ( ) h AWS ( ) h AWS ( ) h AWS ( ) AgMet AWS in Rice Paddies 35
9-1. Preliminary Evaluation Results: Air Temperature R M S E 5 4.5 4 3.5 3 2.5 2 1 14 15 19 21 23 25 26 28 29 43 44 45 46 49 1.5 0 10 20 30 40 50 Forecast Hours (hr) 51 89 90 91 RMSE between the Hourly Digital Forecast and the AWS Observation RMSE increased with Forecast Hours; Interpolated Temperature had higher RMSE than the 3-hourly Forecast 36
9-2. Preliminary Evaluation Results: Air Temperature 4.5 Mean RMSE for each station R M S E 4 3.5 3 2.5 2 R = 0.061 0 500 1000 1500 2000 2500 3000 3500 Distance between the grid center and the AWS (m) location Relationship between mean RMSE for each station and distance between the grid center and the AWS location Forecast accuracy was independent of distance and elevation difference between the AWS and the grid center => Source of Error: Variation in Vegetation and Land Use/Cover 37
9-3. Preliminary Evaluation Results: Relative Humidity R M S E 20 18 16 14 12 10 0 10 20 30 40 50 Forecast Hours (hr) 1 14 15 19 21 23 25 26 28 29 43 44 45 46 49 51 89 90 91 RMSE between the Hourly Digital Forecast and the AWS Observation RMSE increased with Forecast Hours; Interpolated RH had higher RMSE than the 3-hourly Forecast 38
9-4. Preliminary Evaluation Results: Relative Humidity 19 Mean RMSE for each station R M S E 18 17 16 15 14 13 12 R = 0.004 0 500 1000 1500 2000 2500 3000 3500 Distance between the grid center and the AWS (m) location Relationship between mean RMSE for each station and distance between the grid center and the AWS location Forecast accuracy was independent of distance and elevation difference between the AWS and the grid center => Source of Error: Variation in Vegetation and Land Use/Cover 39
9-5. Preliminary Evaluation Results: Rainfall Probability of Detection Proportion of Correct P O D P C 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 10 20 30 40 50 (hr) 1 14 15 19 21 23 25 26 28 29 43 44 45 46 49 51 89 90 91 14 Probability of False Detection False Alarm Rate 15 19 21 23 25 26 28 29 43 44 45 46 49 F A R 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 10 20 30 40 50 (hr) 51 0 89 0 10 20 30 40 50 0 10 20 30 40 50 90 (hr) 91 (hr) Forecast Hours 40 P O F D 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 1 14 15 19 21 23 25 26 28 29 43 44 45 46 49 51 89 90 91 1 14 15 19 21 23 25 26 28 29 43 44 45 46 49 51 89 90 91
Probability of Detection Proportion of Correct P O D 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 10 20 30 40 50 (hr) 1 14 15 19 21 23 25 26 28 29 43 44 45 46 49 51 89 90 91 14 F A R 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 9-6. Preliminary Evaluation Results: Leaf Wetness (Simple RH) P C 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 Probability of False Detection False Alarm Rate 15 19 21 23 25 26 28 29 43 44 45 46 49 0 10 20 30 40 50 (hr) 51 0 89 0 10 20 30 40 50 0 10 20 30 40 50 90 (hr) 91 (hr) Forecast 41 Hours P O F D 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 1 14 15 19 21 23 25 26 28 29 43 44 45 46 49 51 89 90 91 14 15 19 21 23 25 26 28 29 43 44 45 46 49 51 89 90 91
Probability of Detection Proportion of Correct 9-7. Preliminary Evaluation Results: Leaf Wetness (CART) P O D P C 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 10 20 30 40 50 (hr) 1 14 15 19 21 Probability of False Detection False Alarm Rate 23 25 26 28 29 43 44 45 46 49 51 89 90 91 14 15 19 21 23 25 26 28 29 43 44 45 46 49 F A R 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 10 20 30 40 50 (hr) 51 0 0 10 20 30 40 50 89 0 10 20 30 40 50 90 (hr) 91 (hr) Forecast 42 Hours P O F D 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 1 14 15 19 21 23 25 26 28 29 43 44 45 46 49 51 89 90 91 14 15 19 21 23 25 26 28 29 43 44 45 46 49 51 89 90 91
9-8. Preliminary Evaluation Results: Summary Source of errors for Air Temp. and RH are attributed to variations in vegetation and land use/cover of the surrounding area Amount of hourly precipitation had high False Alarm Rate of 0.8 (overestimation due to the limit of estimating hourly probability of rain) Simple RH leaf wetness model overestimates wetness (Proportion of Correct 50%, False Alarm Rate 60%) CART leaf wetness model underestimates wetness (Proportion of Correct 60%, False Alarm Rate 35%) ** Rice blast warning is highly sensitive to leaf wetness Precipitation and leaf wetness models need improvement. Detailed variations in vegetation and land use/cover should be more precisely considered for agricultural meteorological services. 43
IV. Outcome and Further Development Providing weather-driven disease risk information, which can also be used in everyday agriculture Supporting scientific decision on disease and pest control, which will lead to environment-friendly agriculture Pioneering the application of Digital Weather Forecast to various fields in agricultural meteorology More AgroMet friendly NWP models are needed to Resolve land use differences at higher resolution (~100m) Incorporate AgroMet models easily Assimilate AWS, satellite and radar data 44
IV. Outcome and Further Development Co-developers for Agricultural Digital Forecasting System Wee Soo KANG and Eun Woo PARK Seoul National University 45