ABSTRACT This study presents a novel approach to crop mapping using remotely sensed satellite images. It addresses the significant classification modelling challenges, including (1) the requirements for extensive labelled data and (2) the complex optimisation problem for selection of appropriate temporal windows in the absence of prior knowledge of cultivation calendars. We compare the lightweight Dynamic Time Warping (DTW) classification method with the heavily supervised Convolutional Neural Network ‐ Long Short‐Term Memory (CNN‐LSTM) using high‐resolution multispectral optical satellite imagery (3 m/pixel). Our approach integrates effective practical preprocessing steps, including data augmentation and a data‐driven optimisation strategy for the temporal window, even in the presence of numerous crop classes. Our findings demonstrate that DTW, despite its lower data demands, can match the performance of CNN‐LSTM through our effective preprocessing steps while significantly improving runtime. These results demonstrate that both CNN‐LSTM and DTW can achieve deployment‐level accuracy and underscore the potential of DTW as a viable alternative to more resource‐intensive models. The results also prove the effectiveness of temporal windowing for improving runtime and accuracy of a crop classification study, even with no prior knowledge of planting timeframes.

Hyperspectral remote sensing of advanced plant traits and physiological status are explored for leaf nitrogen (N) and phosphorus (P) monitoring in the context of sustainable forestry. Previous studies on agricultural species have demonstrated that plant biochemical and biophysical constituents derived via radiative transfer models (RTMs), and other parameters, such as solar-induced chlorophyll fluorescence (SIF), provided an improved prediction of leaf N compared to traditional methods based on chlorophyll indices. In this study, we assessed the transferability of such methods to assess needle N and P in coniferous canopies, where highly heterogeneous tree-crown structures dominate. Our study across three years showed that RTM-based functional traits (chlorophyll a+b (Ca+b), carotenoids (Car), anthocyanins (Anth), and Leaf Area Index (LAI)) along with SIF could provide moderate prediction accuracy for N (Ca+b, Car, Anth, SIF: R2=0.4 to 0.72) and P (Ca+b, LAI, SIF: R2=0.4 to 0.73). Furthermore, this work highlighted that SIF contributed to needle P and N assessment to different degrees.

Wheat (Triticum spp.) is widely grown and receives large amounts of nitrogen fertiliser. Grain protein content (GPC) is key to its nutritional and economic value and is influenced by genetic, management and environmental factors. Accurate pre-harvest estimation of GPC patterns could improve N efficiency and profit. Some plant traits with physiological links to GPC have been identified, but assessment of traits retrievable from satellite imagery could advance GPC estimation at commercial scales. By radiative transfer model inversion of Sentinel-2 imagery, we retrieved chlorophyll (Ca+b), leaf dry matter (Cm), leaf water content (Cw) and leaf area index (LAI) for > 6,000 hectares of commercial wheat across two years. Farmers supplied ~90,000 GPC data points collected during harvest. Using a gradient boosted machine, we predicted GPC (R2= 0.86, RMSE = 0.56) and found that Cwexplained most GPC variability under stress conditions but that contributions to GPC were distributed among plant traits in milder conditions.

Integrating far-red solar-induced chlorophyll fluorescence (SIF760) and leaf biochemical constituents (primarily leaf chlorophyll content (Ca+b)) has recently been demonstrated to improve the estimation of leaf nitrogen (N) concentration from airborne and spaceborne hyperspectral imagery in homogenous and heterogeneous crop canopies. The advent of sub-nanometer resolution imagers capable of detecting narrow solar Fraunhofer lines (FLs) has enabled a novel opportunity to investigate the prospect of leaf N estimation using individual FLs in addition to SIF760and Ca+btraits. This study seeks to determine whether incorporating distinct FL depth derived from sub-nanometer airborne hyperspectral imagery could improve leaf N estimates. A sub-nanometer hyperspectral imager with ≤0.2 nm full-width at half-maximum (FWHM) resolution was flown in tandem with a narrow-band hyperspectral imager with 5.8 nm FWHM over a winter wheat field. Plots were fertilized with variable concentrations of nitrogen to enable nutrient variability. Regression models utilizing Gaussian process regression (GPR) were built with different permutations of SIF, Ca+band depths of individual FLs for estimating leaf N concentration. Laboratory-determined leaf N estimates were obtained by destructive sampling. Results show that GPR models incorporating the depth of distinct Fraunhofer lines as predictor variables performed better than the benchmark model constructed using Ca+band SIF760alone. The best leaf N-estimation model built with FLs from the red and far-red regions (Ca+b, FL682.97 nm, FL757.002 nm) yielded an R2of 0.71, outperforming the standard approach used in previous works (Ca+b, SIF760) (R2= 0.56).

Traditional methods to identify biotic-induced plant stress are time-consuming and costly. Airborne hyperspectral and thermal imagery has shown promise in identifying disease symptoms caused by pathogens in several plant species. Specifically, to detect Xylella fastidiosa (Xf) and Verticillium dahliae (Vd), previous studies have aimed to detect symptomatic and asymptomatic trees with high accuracy. Nevertheless, these studies did not explore the progressive changes in plant traits with increasing disease severity levels. In this study, we investigate the dynamics of plant traits as a function of disease severity.Moreover, we focus on the plant traits derived from hyperspectral data that contribute the most to disease detection, assessing how their role is redistributed as a function of disease severity. Finally, we evaluate the contribution of the plant traits in asymptomatic trees undetectable by visual inspection, using as a reference qPCR analysis. The findings revealed that specific traits such as the NPQI and PRIn indices and SIF and Anth were the most crucial.

Sub-nanometer hyperspectral imagers are increasingly being used to quantify solar-induced chlorophyll fluorescence (SIF) due to their ability to characterize narrow absorption features accurately. However, some limitations prevent their wide use in the operational context due to their high cost, weight, and complexity. On the other hand, more widely-used narrow-band hyperspectral imagers with 4–6 nm full width at half-maximum (FWHM) resolution could be a costeffective alternative for acquiring high-spatial-resolution hyperspectral imagery to derive SIF. Due to the large effects of the spectral resolution (SR) on the quantified fluorescence, the SIF levels derived from such airborne imagers with 4–6 nm FWHM are overestimated, requiring careful interpretation. In this study, we flew in tandem two airborne hyperspectral imagers with different spectral characteristics. These sensors' imagery was used to model the impact of SR on the SIF quantification using the Soil-Canopy Observation of Photosynthesis and Energy (SCOPE) model. A Support Vector Machine regression (SVR) model trained via SCOPE simulations was employed to quantify SIF at 1 nm SR from the original 5.8 nm FWHM resolution. The performance of the SIF quantification was evaluated theoretically with SCOPE and tested against airborne hyperspectral radiance and the derived SIF. Results showed that the estimated SIF at 1 nm SR agreed well with the reference SCOPE simulations (RMSE=0.097 mW/m2/nm/sr) and with airborne-quantified SIF (RMSE=0.094 mW/m2/nm/sr).

Oak decline is a complex syndrome that increasingly affects the survival of oak species worldwide. Spectral-based physiological plant traits (PTs) indicators have successfully quantified pigment degradation and vegetation structure. The specific response of PTs to decline diseases has been answered by using high spectral- and spatial-resolution hyperspectral and thermal sensors onboard airborne platforms in the context of oak disease detection and monitoring. However, the capacity for early detection using a limited number of spectral bands, with miniaturised sensors of lower sensitivity, is unknown, seeking outstanding operability through cost-effective platforms, which is critical to detect irreversible damage timely. We evaluate the use of multispectral and thermal imagery onboard a drone together with a 3-D radiative transfer model (RTM) approach to assess a predictive symbolic classification model of Phytophthora-infected holm and cork oak areas located in Ourique (southern Portugal). The field survey comprised more than 390 trees across disease severity classes with varying disease-incidence levels and species. The classification model showed up to 83% overall accuracy ($k=0.46$) for decline detection. The proposed model allowed us to efficiently identify the physiological state of the forest canopy so that disease progression can be detected and mapped rapidly, which is essential for fighting oak decline when silvicultural practices, such as tree removal and clearing, can still prevent the spread of dieback processes. Therefore, our study demonstrates that the tandem use of multispectral and thermal sensors, together with an RTM and AI approach, helps us predict the impact of this particularly damaging disease on oak trees.

Several studies have demonstrated the influence of the spectral resolution (SR) on the retrieval of solar-induced chlorophyll fluorescence (SIF) from ground-based sensors with different spectral configurations. However, research studying the implications of the SR of airborne hyperspectral imagers on the retrieved SIF is lacking, and its interpretation is critical for precision agriculture, plant stress detection and phenotyping studies. This work investigates the effects of SR on SIF performance through the Fraunhofer Line Depth (FLD) principle at the O2-A absorption feature (760.4 nm) using two airborne hyperspectral imagers with different spectral characteristics. A sub-nanometer hyperspectral imager with 0.1-0.2 nm full-width at half-maximum (FWHM) resolution and a broader-band hyperspectral imager of 5.8 nm FWHM were flown in tandem. The campaigns were conducted over a winter wheat field with randomized experimental design, with plots receiving different nitrogen rates to ensure SIF variability. Results showed a bias on the SIF levels quantified by both airborne imagers$(\text{RMSE} =3.7$mW/m2/nm/sr), but a strong relationship between both sensors at the O2-A absorption feature ($\mathrm{R}^{2}=0.84,\ \mathrm{p} < 0.001$). Results confirm the utility of hyperspectral imagers ca. 5 nm FWHM resolution for stress detection and plant phenotyping where assessing the relative variability of SIF across experimental plots is sought.

This study present in situ measurements and model simulations aimed to understand the ability of sun-induced fluorescence (SIF) and other physiological and structural hyperspectral indices as an early indicator of forest decline. Experiments were conducted over an oak forest (Quercus ilex) affected by water stress and Phytophthora infection in the southwest of Spain. The robustness of the SIF quantification through the Fraunhofer Line Depth (FLD) principle with three spectral bands F (FLD3) was assessed using high-resolution hyperspectral imagery. FluorFLIGHT, a modified version of the 3-D radiative transfer model FLIGHT was developed to enable the simulation of canopy radiance and reflectance including fluorescence effects accounting for forest structure. Fluorescence retrievals performed better than structural and physiological indices. Albeit other pigment-related vegetation indices such as CRI550515 and RNIRCRI700 were also strongly related to physiological variables. The 3D modelling approach significantly improved the relationship between Fs and SIF and enabled the quantification of SIF as a function of fractional cover, leaf area index and chlorophyll content, yielding significant relationships between Fs ground-data measurements and fluorescence quantum yield estimated with FluorFLIGHT. The methodology also demonstrated its capabilities for mapping SIF at the crown level to detect early damage assessment in oaks undergoing forest decline.

This paper attempts to provide an understanding of the potential application of Sentinel-2 imagery for the monitoring and detection of disease symptoms caused by Xylella fastidiosa (Xf) in olive trees. A time series data of 188 Sentinel-2a images collected over the last two years was used to analyse the temporal trends in areas with Xf infected olive trees in Puglia, Southern Italy. The robustness of different physiological and structural hyperspectral indices was evaluated as an early indicator of Xf symptoms. Three validation sites for Sentinel-2a products were hence established over olive orchards in the Xf-infected zone in two different years (2016 and 2017) and overflown with a hyperspectral sensor to acquire high spatial resolution images (50 cm). Disease incidence and severity levels were recorded for more than 3300 olive trees in 18 orchards. Results demonstrate the capability of temporal Sentinel-2a was able to detect and discriminate between high and medium Xf incidence, reaching the maximum differences during the summer season. Among all the vegetation indices evaluated from Sentinel-2 imagery, OSAVI showed superior performance for detecting Xfincidence trends and OSA VI1510for detecting changes in Xf severity levels.