Yield estimation of citrus trees with unmanned aerial vehicle and terrestrial hyperspectral data

Antalya province has an important place in the agricultural sector both in our country and in the world in terms of meteorological conditions and soil characteristics as well as its contrubition to the tourism sector. Many different varieties of citrus fruits, which are one of the important product groups in the agricultural sector in terms of domestic and foreign markets, are grown in Antalya. In this study, it was aimed to estimate the yield of different citrus tree varieties (Washington, Valencia, Clementine, Satsuma), which are also expressed as citrus fruits, with different remote sensing techniques. The study was carried out in the period of fruit ripening of citrus trees covering the years 2021-2022 in the Kayaburnu land of the West Mediterranean Agricultural Research Institute (BATEM) located in the Serik District of Antalya Province. The thesis work consists of the basic stages of data acquisition, preprocessing of unmanned aerial vehicle (UAV) and light detection and ranging (LiDAR) data, creation of databases for all datasets including terrestrial measurement data, statistical analyses, yield estimation and evaluation. Within the scope of the study, the tree crown structure, tree crown area and tree heights obtained from the data obtained from UAV and UAV-LiDAR systems were determined in the yield estimation of different citrus varieties in the fruit ripening period. As a result of the statistical analyses performed between the tree dendrometric components obtained from these two different remote sensing systems, high correlations were determined, and in the light of the findings obtained, it was decided to use the tree dendrometric components calculated from the UAV in the yield estimation of citrus trees. In another stage of the study, simple ratio (SR) index and normalized difference vegetation index (NDVI) from the images taken from the UAV mounted multispectral (MS) camera and SR, NDVI, hyperspectral normalized difference vegetation index (HNDVI) and normalized pigment chlorophyll index (NPCI) from the terrestrial hyperspectral data measured from the spectroradiometer device were calculated. In addition, trunk heights and leaf chlorophyll values of citrus trees were measured during the field studies, and the actual yield values of all citrus trees were obtained. Within the scope of the statistical analyses carried out for all data sets obtained in the study, descriptive statistics values, skewness-kurtosis coefficients, histogram graphics, normality tests and correlation analyses were performed for all variables. In addition, in the estimation of the yield of citrus trees in the fruit ripening period, 4 different yield estimation models were created by using 9 variables to explain the yield. In the study, it was decided to use the partial least squares regression (PLSR) in the estimation of the yield as a result of the evaluation considering the statistical analyses performed between all the variables. According to the results of the PLSR analysis applied for each model, it was determined that the model with all independent variables in all trial areas had the highest percentage of explanation of the yield dependent variable. In addition, the findings obtained from the research showed that some plant indices and some tree dendrometric components obtained from the UAV data are important parameters in explaining the yield in citrus orchards in the fruit ripening period. According to the model comparison results obtained with this thesis study, the highest determination coefficient (R2) was determined in the Satsuma mandarin trial area (R2=0.841). The coefficients of determination obtained in other trial areas were Washington orange (R2=0.719), Valencia orange (R2=0.685), and Clementine mandarin (R2=0.302), respectively.