Development of a deep neural network model to estimate solar radiation using temperature and precipitation

온도와 강수를 이용하여 일별 일사량을 추정하기 위한 심층 신경망 모델 개발

Abstract

Solar radiation is an important variable for estimation of energy balance and water cycle in natural and agricultural ecosystems. A deep neural network (DNN) model has been developed in order to estimate the daily global solar radiation. Temperature and precipitation, which would have wider availability from weather stations than other variables such as sunshine duration, were used as inputs to the DNN model. Five-fold cross-validation was applied to train and test the DNN models. Meteorological data at 15 weather stations were collected for a long term period, e.g., > 30 years in Korea. The DNN model obtained from the cross-validation had relatively small value of RMSE ($3.75MJ\;m^{-2}\;d^{-1}$) for estimates of the daily solar radiation at the weather station in Suwon. The DNN model explained about 68% of variation in observed solar radiation at the Suwon weather station. It was found that the measurements of solar radiation in 1985 and 1998 were considerably low for a small period of time compared with sunshine duration. This suggested that assessment of the quality for the observation data for solar radiation would be needed in further studies. When data for those years were excluded from the data analysis, the DNN model had slightly greater degree of agreement statistics. For example, the values of $R^2$ and RMSE were 0.72 and $3.55MJ\;m^{-2}\;d^{-1}$, respectively. Our results indicate that a DNN would be useful for the development a solar radiation estimation model using temperature and precipitation, which are usually available for downscaled scenario data for future climate conditions. Thus, such a DNN model would be useful for the impact assessment of climate change on crop production where solar radiation is used as a required input variable to a crop model.

일사량은 자연 생태계와 농업 생태계에서 에너지 수지와 물 순환을 추정하는데 중요한 변수이다. 일별 일사량을 추정하기 위해 심층 신경망(DNN) 모델이 개발되었다. 일조시간 등의 변수보다 기상 관측소에서의 가용성이 더 높은 온도와 강수량이 심층 신경망 모델의 입력 자료로 사용되었다. five-fold crossvalidation 을 사용하여 심층 신경망을 훈련시키고 검증하였다. 국내 15 개의 기상 관측소에서 30 년 이상 장기간의 기상 자료가 수집되었다. Cross-validation을 통해 얻어진 심층 신경망 모델은 수원 지역 기상 관측소의 일별 일사량 추정치에 대해 비교적 작은 RMSE($3.75MJ\;m^{-2}\;d^{-1}$) 값을 가졌다. 심층 신경망 모델은 수원 지역 기상 관측소의 일사량의 변위의 약 68%를 설명했다. 1985 년과 1998 년의 일사량 관측값은 일조시간에 비해 상당히 낮은 값이 관측되었다. 이는 후속 연구에서 일사량 관측 데이터의 품질 평가가 필요할 것임을 시사했다. 해당 연도의 데이터를 분석에서 제외했을 때, 심층 신경망 모델의 추정값은 통계적 수치가 약간 높게 나타났다. 예를 들어, $R^2$ 와 RMSE 의 값은 각각 0.72 와 $3.55MJ\;m^{-2}\;d^{-1}$ 이었다. 심층 신경망 모델은 기온과 강수량을 통해 일사량을 추정하는데 유용하며, 이는 미래 기후 시나리오 자료에 대해서 활용할 수 있을 것이다. 따라서, 공간에 대한 제약이 완화된 심층 신경망 모델은 작물 모델의 입력 자료로 일사량이 필요한 작물 생산성에 대한 기후 변화 영향 평가에 유용하게 활용될 수 있을 것이다.

Fig. 1. Weather stations where meteorological data were collected. Weather stations marked by a blue circle were for train data. Weather station marked by a red star is for test data.

Fig. 2. Rectified Linear Units (ReLU) function. X and Y indicate independent and dependent variables, respectively.

Fig. 3. Box plots of daily solar radiation observed at the weather station located in Suwon from 1982-2017.

Fig. 4. Distributions of daily solar radiation and sunshine duration observed at the Suwon weather station in (A) 1985, (B) 1998, (C) 1997 and (D) 2015.

Fig. 5. Comparison of root mean square error (RMSE) for solar radiation estimates using five-fold crossvalidation.

Fig. 6. The scatter plot of measured and estimated solar radiation. The measurement data were obtained from the Suwon weather station from 1982-2017. The deep neural network model was used to estimate daily solar radiation.

Fig. 7. Box plots of errors for daily solar radiation estimates using the deep neural network model by a range of daily solar radiation observation at the Suwon weather station.

Fig. 8. Year-by-year comparison of R² values for daily solar radiation estimates using the deep neural network model at the Suwon weather station.

Fig. 9. Scatter plots of daily solar radiation measurements and estimates. The measurements were obtained at the Suwon weather station in (A) 1997, (B) 2015, (C) 1985 and (D) 1998. The daily solar radiation was estimated using the deep neural network model.

Fig. 10. The scatter plot of measured and estimated solar radiation. The measurement data were obtained from the Suwon weather station from 1982-2017 except for 1985 and 1998 where the maximum values of solar radiation were relatively lower than others. The deep neural network model was used to estimate daily solar radiation.