Modeling seasonal chlorophyll change in satsuma mandarin using hyperspectral data with partial least squares regression (PLSR) machine learning algorithm

In recent years, many concepts such as sustainable agriculture and precision agriculture have emerged in every aspect of our lives. Agricultural practices that are ecological, economical and provide savings in every aspect are preferred. Working models created with remote sensing and computer science also enable fast and non-destructive production processes in the field of agriculture. Protecting plant health throughout the production season is an important criterion for all agricultural inputs. For this reason, chlorophyll content, which is one of the important indicators of plant health, is also seen as a basic parameter in research. In remote sensing studies; In spectral reflectance values, being sensitive close to 700 nm and in the range of 530-630 nm guides the interpretation of chlorophyll content and therefore plant health and the execution of agricultural activities. Machine learning algorithms, which have increased their popularity in recent years, provide significant advantages in analyzing intensive data sets quickly and reliably and creating prediction models. As a matter of fact, it is possible to train different models using various machine learning algorithms and to make current and future predictions of an important parameter for plants such as chlorophyll. One of these machine learning algorithms is partial least squares regression (PLSR) method. In this study, hyperspectral reflectance measurements were performed with ASD FieldSpec Handheld Spectroradiometer and chlorophyll measurements were performed with SPAD-502Plus chlorophyll measurement device on leaf samples taken every month for 12 months from Satsuma mandarin orchard trees produced in Antalya province. The aim of the study was to determine the reflectance and chlorophyll relationship statistically by periodically questioning the spectral properties of the plant based on remote sensing and to reveal the success of plant spectral properties in chlorophyll estimation with PLSR method. As a result of the findings, significant success was achieved in estimating the dependent variable of chlorophyll (SPAD) with the independent variables of TGI (Triangular Greenness Index), VARI (Visible Atmospheric Resistant Index), GRVI (Green Red Vegetation Index), HNDVI (Hyperspectral Normalized Difference Vegetation Index), CIG (Chlorophyll Index – Green) and LCI (Leaf Chlorophyll Index) calculated from spectral measurements. Among the estimation models performed for each month in Satsuma mandarin, it was concluded that the period that gave the highest R2 value (RMSE: 1.88; MAPE: 4.06) and best predicted the dependent variable of SPAD was May, which also coincides with the flowering and fruit set period. It was determined that the May model provided the highest success in the dependent variable of SPAD by a factor of 2 at a rate of 83.57% and the highest accuracy was reached with an R² value of 0.84.