Applicability Of Illuminance In Farmland Evapotranspiration

Evapotranspiration is the result of the combined effect of crop transpiration and soil evaporation between crops.It is also an important water vapor exchange process in the Soil-Plant-Atmosphere Cotinuum（SPAC）system,which has an important impact on the growth and development of crops.In this study,in the process of reference crop evapotranspiration（ET₀）estimation,solar radiation measurement equipment is expensive,difficult to deploy and apply quickly,and the stability of a single machine learning method is poor.A farmland reference that combines illuminance and basic meteorological factors is proposed.The crop evapotranspiration estimation method is based on the classical formula method and machine learning method for theoretical analysis,experimental design and result verification.The results show that it is feasible and stable to estimate the reference crop evapotranspiration through the combination of illuminance and basic meteorological data.（1）In order to solve the problem of irradiance estimation based on illuminance,a correlation analysis of the factors affecting solar radiation was carried out in combination with illuminance data and basic meteorological data.The results show that irradiance is highly correlated with illuminance,air temperature,air humidity,and weather type.Among them,irradiance,air temperature,and air humidity have a certain hysteresis,and they are more sensitive to weather types.Irradiance and light intensity have the highest correlation,with Pearson and Spearman coefficients of 0.9637 and 0.9694,respectively.（2）Based on the clear sky index,a weather type classifier and irradiance regression algorithm based on Long Short-Term Memory（LSTM）is proposed,which solves the problem of using illuminance combined with basic meteorological factors to irradiate under different weather conditions.Degree estimation problem.At the same time,the estimated irradiance is used as the basis for net radiation estimation,and based on the Penman-Monteith radiation theory,a net radiation estimation algorithm based on irradiance is proposed,which solves the needs based on the Penman-Monteith（PM）formula and the Hargreaves（HS）formula The radiation item data problem.The results show that the weather type classifier based on the LSTM network separates the stable data during the radiation transmission process,and improves the estimation results of the stable data during the radiation transmission process.Compared with the unclassified basic meteorological data,the R² of the irradiance estimation model based on the weather type classifier increased by 0.019,and the R_MSE decreased by13.871 W·m^-2;the net radiation is the net shortwave radiation and the net longwave radiation in different weather As a result of the mutual game under the conditions,compared with the standard value of net radiation calculated by actual shortwave radiation,the R² and R_MSE of the net radiation estimation algorithm are 0.990 and 1.034 MJ·m^-2·d^-1,respectively.（3）Based on the PM formula and HS formula recommended for reference crop evapotranspiration estimation,the feasibility and necessity of using the estimated net radiation to the reference crop evapotranspiration estimation model is proposed.The estimated net radiation and the actual shortwave radiation calculated net radiation standard value combined with basic meteorological factors are input into the PM formula for calculation.The results show that the estimated value of net radiation can replace the standard value to estimate the reference evapotranspiration.The R² of the reference evapotranspiration estimation result based on the PM formula is 0.989,and the R_MSE is 0.341 mm·d^-1.At the same time,the estimated net radiation is applied to the HS formula instead of the original net radiation estimation term based on the temperature method.It is easy to find that the estimation accuracy of the radiation term in the HS formula has a greater impact on the estimation result of the reference crop evapotranspiration,and the illuminance-based The estimated net radiation replaces the original radiation term based on extreme temperature,which can improve the accuracy of the evapotranspiration estimation of the HS formula.Compared with the original HS formula,the reference evapotranspiration estimation results of the improved HS formula based on estimating net radiation are 0.97 and 0.835 mm·d^-1,respectively.R² increases by0.173 and R_MSE decreases by 1.343 mm·d^-1.（4）Based on the commonly used machine learning algorithms for reference crop evapotranspiration estimation,a reference evapotranspiration machine learning model based on integrated learning ideas is proposed.It integrates XGBoost,Light GBM,RFR,and SVR unit models,thereby effectively improving the stability of the machine learning algorithm in the estimation of reference crop evapotranspiration.The results show that the advantage of ensemble learning is not only shown in the model accuracy close to the optimal effect in the unit model,but also in the overall performance of the ensemble learning algorithm in the evapotranspiration regression problem is always close to or exceeding the optimal unit model effect.The integrated learning model has the best overall effect,with R² and R_MSE being 0.9674and 0.0457mm·h^-1,respectively.Second,the sunny index based on the average daily illuminance helps to improve the accuracy of the hourly reference evapotranspiration estimation.The hourly reference evapotranspiration fluctuates with the hourly illuminance,and the weather type has a greater influence on this fluctuation.Therefore,the daily average value of illuminance can reflect this phenomenon,which is used as the characteristic quantity of the estimation model.The R_MSE of the final integrated model is reduced by 0.013 mm·h^-1 at the maximum,and the optimal value is 0.0457 mm·h^-1,R² is increased by 0.0014 at the maximum,and the optimal value is 0.9674.（5）Horizontal comparison and analysis of the estimation results of reference crop evapotranspiration based on formula method and machine learning method.The results show that the feasibility and reliability of reference evapotranspiration estimation by using illuminance are relatively clear.The daily reference crop evapotranspiration estimation results using unit machine learning algorithm,integrated learning algorithm,PM formula method,and HS correction method are relatively consistent,and the lowest correlation value of the estimation results among multiple models is higher than 0.973,because the PM formula method uses reference evapotranspiration The physical background of the quantity is the main theoretical support,and the results are standard;and the reference evapotranspiration estimated based on each machine learning algorithm is supported by mathematical statistics,and the results are statistically stable.This research combines the theory of solar radiation estimation,reference crop evapotranspiration estimation theory and machine learning regression theory,through theoretical derivation and experimental analysis,and proposes a reference crop evapotranspiration estimation method based on illuminance.The illuminance measurement is convenient and widely used in the market,and the maintenance cost of the instrument is cheaper than that of the radiation sensor.The daily and hourly estimation of reference crop evapotranspiration by illuminance instead of radiation is expected to solve the problem of convenient measurement of reference crop evapotranspiration from the perspective of equipment application,and improve the wide application of reference crop evapotranspiration theory in agricultural high-efficiency water-saving scenarios.