LEAF AREA INDEX AND CANOPY CHLOROPHYLL CONTENT ESTIMATION OF ARABLE CROPS FROM SENTINEL-2 WITH GAUSSIAN PROCESS REGRESSION: A MULTI-SITE, YEAR AND CROP VALIDATION

Spatio-temporal estimation of crop bio-parameters (BioPar) is required for agroecosystem management and monitoring. BioPar such as Canopy Chlorophyll Content (CCC) and Leaf Area Index (LAI) contribute to assess plant physiological status and health at leaf and canopy level. Remote sensing techniques are instrumental in spatially explicit CCC and LAI retrieval of arable crops across different scales. Machine Learning (ML) techniques, especially Gaussian processes regression (GPR), has outperformed traditional approaches based on Vegetation Index in BioPar estimation. However, being ML model based on data driven approach it is necessary to thoroughly evaluate the performance of GPR across different sites, seasons, and crop types to assess the exportability of the models. This study aimed to develop a transferable GPR algorithm using a large dataset collected over several years (2018-2022), on different locations (5 sites) and with different canopy conditions by sampling 10 different arable crops. The study objectives included developing a robust GPR algorithm for LAI and CCC estimation from Sentinel-2 data, validating GPR against independent datasets, and comparing results with other methods and available products. The study utilized 301 (209 crop + 92 soil spectral) CCC and 301 LAI observations for GPR model training. Validation on independent datasets (698 LAI and 364 CCC) revealed the reliability of GPR estimation, compared to Sentinel-2 Level 2 Prototype Processor (SL2P) estimates. LAI and CCC estimation metrics varied across datasets achieving coherent and similar performance between the two method (GPR and SL2P). In general, SL2P model better fits the overall data with slightly higher R2 values with respect to GPR especially for LAI parameter. GRP estimates provided better results when accuracy analysis is performed by crops showing lower RMSE (Root Mean Square Error) and MAE (Mean Absolute Error). GPR outperforms SL2P for mais and wheat in particular for CCC parameter. These results showed the potential of GPR in BioPar estimation, especially when a robust training set was used. BioPar estimation using Sentinel 2 data provided high-quality quasi-weekly information, essential for smart crop management and early warnings in decision support systems.