Tea Growth Monitoring And Tea Yield Prediction By Using NDVI And Climatic Data

Tea is one of the most significant cash crops and plays an important role in economic development and poverty reduction.Northwest Vietnam region is featured by the mountain’s terrain categories with the temperature,water and light being suitable condition of tea production.Laichau is one of the main teas growing areas in the Northwest region of Vietnam where Tanuyen District is the key tea producing area.It has 2854 ha of tea plants,accounting for two-thirds of the tea-producing area of the whole province.Tea is a crop that does not compete with food crops;tea plantations cover bare land,bare hills,and prevent erosion.At the same time,tea is a cash crop that improves the quality of life for people in the Tanuyen district of Laichau Province.On the other hand,tea is an optimal choice in the extreme weather conditions of Tanuyen Laichau,Vietnam.Crop growth monitoring is one of the most important areas of basic research in agricultural science,which contributes to agricultural classification,yield estimation,agricultural management,and assess environment and climatic change on crop growth.Forecasting crop yields before harvest can help local businesses and governments can have contingency plans,such as importing when the quantity of product is not enough,or exporting when the product is in surplus.In addition,Timely and accuracy information in crop yield forecasting can provide insight about development condition of crop,make early warning.This information is significant for reducing risk in agricultural,improve yield agriculture and make timely adjustments.This paper focuses on the monitoring of tea growth,the relationship between tea yield and climate change,and the forecast of tea yield in Tanuyen District,Laizhou Province,Vietnam.The main research contents are as follows:(1)The research study focused on extract tea region by using Sentinel 2 image.There are two Sentinel 2 remote sensing images were selected on February 16,2018 and November 30,2018 to classify land cover types.Spectral,NDVI,slope,and phenology features are selected as the classification features basis,and the characteristics of six types of ground objects,including tea,cropland,forest,settlement,water body and barren land,are used as the input of the area of interest,and the maximum likelihood(ML)supervised classifier is used for LULC classification.The results showed that tea was mainly distributed in Tan uyen,Phuc Khoa and Trung Dong areas,with scattered distribution in the remaining areas.This study evaluates the accuracy of the classified images with ground truth data and Google Earth data.Create a reference dataset based on high-resolution Google Maps imagery and generate an accuracy assessment report.The overall classification accuracy was95%and the Kappa coefficient was 94%.Among them,tea,cropland,water body and forests have the highest accuracy,reaching 97%.The producer accuracy of the settlement area is85%.The possible reason is that the settlement areas in the study area are mostly scattered in small mountains and valleys,and some pixels are mixed with barren land,which affects the classification accuracy.The experimental results show that Sentinel 2 and maximum likelihood classification have great potential in crop classification,and the results are relatively accurate.(2)This study proposes a method for monitoring tea growth status and analyzing influencing factors based on NDVI trend analysis and meteorological factor correlation analysis.By extracting the NDVI of tea in Tanuyen tea area from 2009 to 2018,the monthly average and the yearly average NDVI were calculated to represent the growth status of tea in dormancy and growing season.In this paper,linear regression was used to estimate NDVI trends,their slopes were calculated,temporal trends were assessed using Mann-Kendall analysis.The Pearson and partial correlation coefficients between NDVI and meteorological parameters were calculated to identify the main meteorological drivers affecting NDVI changes.By calculating the lag correlation coefficient between NDVI and the climate variables of the previous 2 months,the previous month and the current month,the tea growth was detected the lag time between tea growth and order mean temperature,minimum temperature,maximum temperature,precipitation,solar radiation.The results showed that the variation of tea NDVI in different regions was relatively small.The maximum NDVI of tea in November was 0.67 while this value reach up only 0.4 to below 0.5 in January to March,but there were significant differences in NDVI values among the gardens.For 10-years,the NDVI of tea in Tanuyen increased slowly over the 10 years,with a slope of0.0048/a and 0.0075/a in growing season.From 2009 to 2018,the coefficient of determination(~2)between NDVI and climate variables in the growing season in the study area was 0.63(p<0.0001).The order of the influence degree of each variable on tea growth is:Tmin≥Tmean>Precipitation>SL>Tmax.During the growing season,the correlations of tea NDVI to almost all climatic factors with a lag of 1 month were higher than those of the current month.(3)This study proposes a tea yield prediction method based on machine learning and meteorological variables.The experiment used multiple linear regression(MLR),support vector machine(SVM)and random forest(RF)methods to predict tea yield,and compared the model performance by calculating RMSE,~2 and MAE.The experimental results show that SVM and RF are more effective in predicting tea yield.The coefficients of determination between the actual and estimated yields for the MLR,SVM,and RF models were 0.57,0.66,and 0.73,respectively.Compared with MLR,the prediction accuracy of SVM and RF models is higher.The MAE metric in SVM and RF is around 200kg/ha,while this metric for MLR is between 250kg/ha and 370kg/ha.The three algorithms have the highest accuracy in 2015,with an error of less than 3%.The percent error of tea yield(PETY)of the RF algorithm is the best(≤10%),and the root mean square error is smaller(≤80kg/ha).In terms of variables,the effect of NDVI and Tmin on the prediction of tea yield was greater than that of other variables,while the effect of SL variable was relatively small.