Projet de recherche S0/00/070 (Action de recherche S0)
Context and objectives
The framework of this research is to precisely monitor and model economically important plant production processes, in-casu in peach (Prunus persica L.) orchards by means of hyperspectral multi-angle reflectance data. The specific objectives were (i) using radiative transfer methods at the leaf and canopy levels to achieve quantitative estimates of leaf biochemistry from hyperspectral imagery collected over peach orchards, (ii) studying the potential for N estimation using hyperspectral data from model-retrieved leaf chlorophyll and dry matter in peach orchard canopies, (iii) optimizing newly developed processing methods for the large quantity of hyperspectral measurement data using multivariate techniques, thereby developing innovative generic spectral indices and detection algorithms for growth process anomalies, (iv) studying fundamentally the interaction between solar energy and living plant material by means of hyperspectral reflectance measurements, and finally (v) defining digitally measurable in situ parameters necessary to model growth processes, plus first-concept approaches for instrument construction. The general objective is directly tied in with the development of the Optical SpectroMeter Unit (OSMU) hyperspectral space sensor, on-board the Multi-Sensor Micro Satellite Imager (MSMI). This micro-satellite is planned in collaboration with Stellenbosch University, South Africa, and will have on-board modeling capabilities, specifically intended for hyperspectral process modeling.
Methodology
Ground measurements were conducted in Spain in collaboration with P. Zarco-Tejada and F. Morales from the Spanish National Research Council (IAS-CSIC). The specific objectives were fulfilled as: (i) The FLIGHT 3D canopy model was used in connection with PROSPECT leaf model. A more simple approach based on a turbid medium radiative transfer model (SAILH) was tested when targeting crowns to minimize shadows and direct soil reflectance. (ii) Leaf optical measurements of reflectance and transmittance were inverted through the PROSPECT leaf model to estimate leaf chlorophyll and dry matter. Relationship between these pigments and total C, and leaf N were evaluated; The ACRM model was used to investigate leaf-level relationships of leaf chlorophyll and dry matter with N hold at the canopy level. (iii) Multivariate data reduction techniques were used to identify the most indicative wavelength variables to detect biochemical constituents and biophysical parameters. (iv) These techniques were also used to develop new indices to effectively detect and quantify anomalies in the normal plant production process on different observation scales. (v) Field observation and in situ measurements were conducted to extend models incorporating both spectral data as well as in-situ measurements using standard regression techniques.
Results
A well-defined hyperspectral multi-layer data set was built for the study site in Spain, consisting of hyperspectral measurements at various levels (leaf, canopy, airborne), as well as derived and developed indices. Indices and model inversions were assessed in order to account for structural and BRDF effects on the estimation of chlorophyll concentration with both ACRM and FLIGHT models. Leaf chlorophyll (a+b) concentration was successfully estimated from in situ measured leaf reflectance (R2=0.81; RMSE=8.57 g/cm2) and airborne hyperspectral imagery (R2=0.49; RMSE=5.49 g/cm2) over the peach orchard using model inversions of PROSPECT and ACRM. Ground-based measurements have characterized the on-site peach (Prunus persica L.) orchard in terms of chlorophyll, dry matter, water content, and leaf-area-index. Chlorosis did appear as a result of iron deficiency. This responded in a decrease of chlorophyll as found in the biochemical measurements as well as in existing and developed vegetation indices and spectral profiles. The best suited vegetation indices were highly correlated (R2>0.65) to the measured chlorophyll concentrations. Multivariate data reduction techniques fulfilled the expectations: the visible part of the spectrum, mostly dominated by the amount of pigments (chlorophyll, carotenoids), was the most discriminative spectral region (505 - 740 nm). The multi-angle campaign enabled the assessment of vegetation indices as function of the viewing geometry used for each image collection.
