연구보고서
- 저자
- 김광형 박사
- 작성일
- 2016.01.23
- 조회
- 244
- 요약
- 목차
An early warning system for crop diseases and pests is valuable when the system provides timely forecasts that farmers and/or extensions can utilize to inform their crop management decisions. In Korea, small radio frequency-controlled, unmanned helicopters are used by farmers to replace the conventional labor-intensive and inefficient spraying system. Using helicopters is more labor- and time-efficient, and more farmers nowadays use the helicopter pesticide spraying as their primary control option. The early warning system can be utilized for the collaborative pest and disease control using the unmanned helicopters. The system can provide seasonal disease and pest risk information that can inform decisions on optimal spraying time, which diseases or pests to focus on, amounts of agrochemicals needed, and what ratio of pesticide and fungicide to mix for the spray.
At the APEC Climate Center (APCC), we aim to develop a seasonal Disease and pest Early Warning & Spray System (DEWSSystem) using the APCC multi-model ensemble (MME) seasonal forecast. To create an operational DEWSSystem specifically tailored for the collaborative disease and pest control utilizing the unmanned helicopter spray system, we first conducted a survey to get a glance of decision making processes of the collaborative control—the selection of agrochemicals, planning processes of collaborative control, problems encountered arising from spraying with helicopters, and others. Based on the survey results we set our research direction and decided to focus on two problems in the first year: 1) development of daily disease and pest forecasting models and 2) development of spatial and temporal downscaling methods of APCC MME seasonal forecast. In this study, we developed the EPIRICE Daily Risk Model by extracting and modifying core infection risk functions from EPIRICE which, in 2013, was parameterized and validated based on ground truthed disease incidence data surveyed by Rural Development Administration (RDA). The daily riskscore generated by EPIRICE Daily Risk Model was successfully translated into a realistic, quantitative disease value through a series of statistical analyses, and subsequently validated using another ground truthed disease data recorded from 1974 to 2000 in Icheon. To utilize this daily risk model, we developed a temporal downscaling method using a stochastic weather generator (WG). This WG-based temporal downscaling method was applied to downscale monthly weather data (from the APCC MME seasonal forecast) to daily weather data, which can be utilized by daily disease and pest forecasting models. An appropriate WG was selected through comparing EPIRICE model results with original observed weather data and WG-generated synthetic weather data as well as comparing statistics of the original weather data with the synthetic ones. GEM was selected as an appropriate WG for further development of the temporal downscaling method over ClimGen. For the WG-based temporal downscaling, GEM was first used to synthesize more than 1,000 reference weather data bank based on historic 30-year weather data for each specific location. Second, the best-fit weather data was selected from the reference weather data bank based on the Mahalanobis Distance between the given monthly forecasts and the reference data. Bias-correction of the selected best-fit weather data against the given monthly forecasts was also applied to improve the WG downscaling skill. Further improvement was accomplished by introducing an additional ensemble technique of choosing 125 best-fit weather data and using them for the EPIRICE model run. This process reduces the extreme effects resulting from abnormal daily variability introduced during WG synthesis. In addition, we considered the fundamental limitation of the APCC MME seasonal forecast of consisting only of temperature and precipitation data. Thus, we developed a simple way of estimating relative humidity to be used for the developed WG downscaling system. It was done by utilizing the Mahalanobis Distance and ensemble methods to select an additional variable from more than 30 years of historical weather datasetbased on temperature and precipitation as predictors. These methods will be combined to improve the WG downscaling system, which will contribute to the seasonal rice disease and pest outlook using the APCC MME seasonal forecasts.
In conclusion, the EPIRICE Daily Risk Model developed in this study is a role model for other disease forecast models. It is the first model to be used for seasonal disease outlook that uses the APCC MME seasonal forecasts. Additionally, through the EPIRICE Daily Risk Model, many spatial and temporal downscaling methods will be evaluated for APCC MME seasonal forecast since it is developed to be able to use the downscaled seasonal forecasts. The WG temporal downscaling method showed relatively reasonable performance for the EPIRICE model when used with all the techniques developed in this study. From the WG applicability test to the evaluation of available downscaling skills, a complete set of evaluation platform is required for additional evaluation for other daily agricultural models. Through this study, a daily agricultural model and a temporal downscaling method required for the use of APCC seasonal forecast are developed as an effort to utilize climate information in agricultural area. Nevertheless, in order to connect the climate information such as seasonal forecasts with the agricultural information that is useful for various decision makings of agricultural stakeholders, bi-directional approaches from both climate and agricultural sciences are needed.