Optimized modular transfer learning framework integrating PROSAIL and UAV-based hyperspectral reconstruction for cotton canopy water and nitrogen content retrieval

This study proposes an Additive Wavelet Transform (AWT)-based method to fuse Multispectral UAV (MS UAV, 5 cm resolution) and Sentinel-2 satellite imagery (10–20 m resolution), generating 5 cm resolution fused images with a focus on near-infrared and shortwave infrared bands to enhance the accuracy of mango canopy water content monitoring. The fused Sentinel-2 and MS UAV data were validated and calibrated using field-collected hyperspectral data to construct vegetation indices, which were then used with five machine learning (ML) models to estimate Fuel Moisture Content (FMC), Equivalent Water Thickness (EWT), and canopy water content (CWC). The results indicate that the addition of fused Sentinel-2 data significantly improved the estimation accuracy of all parameters compared to using MS UAV data alone, with the Genetic Algorithm Backpropagation Neural Network (GABP) model performing best (R2 = 0.745, 0.859, and 0.702 for FMC, EWT, and CWC, respectively), achieving R2 improvements of 0.066, 0.179, and 0.210. Slope, canopy coverage, and human activities were identified as key factors influencing the spatial variability of FMC, EWT, and CWC, with CWC being the most sensitive to environmental changes, providing a reliable representation of mango canopy water status.

With the continued advancement of spaceborne hyperspectral sensors, hyperspectral remote sensing is evolving as an increasingly pivotal tool for high-precision global monitoring applications. Novel image spectroscopy data, e.g., the PRecursore IperSpettrale della Missione Applicativa (PRISMA) and Environmental Mapping and Analysis Program (EnMAP), can rapidly and non-invasively capture subtle spectral information of terrestrial vegetation, facilitating the precise retrieval of the required vegetation parameters. As critical vegetation traits, Leaf Mass per Area (LMA) and Equivalent Water Thickness (EWT) hold significant importance for comprehending ecosystem functionality and the physiological status of plants. To address the demand for high-precision vegetation parameter datasets, a hybrid modeling approach was proposed in this study, integrating the radiative transfer model PROSAIL and neural network models to retrieve LMA and EWT from PRISMA and EnMAP images. To achieve this objective, canopy reflectance was simulated via PROSAIL, and the optimal band combinations for LMA and EWT were selected as inputs to train neural networks. The evaluation of the hybrid inversion models over field measurements showed that the RMSE values for the LMA and EWT were 4.11 mg·cm−2 and 9.08 mg·cm−2, respectively. The hybrid models were applied to PRISMA and EnMAP images, resulting in LMA and EWT maps displaying adequate spatial consistency, along with cross-validation results showing high accuracy (RMSELMA = 5.78 mg·cm−2, RMSEEWT = 6.84 mg·cm−2). The results demonstrated the hybrid inversion model’s universality and applicability, enabling the retrieval of vegetation parameters from image spectroscopy data and offering a valuable contribution to hyperspectral remote sensing for vegetation monitoring, though the availability of field measurement data remained a significant challenge.

ContextClimate change is causing landscape shifts and locally-adapted plants are becoming increasingly maladapted. As a foundation species, Fremont cottonwood facilitates adaptation to changing climate for the whole community. Populations within this species, however, have varying adaptive responses and facilitative capacity due to genetic variation. It is important to identify these differences to inform landscape restoration and management.ObjectivesUAV hyperspectral, thermal, and lidar images might reveal genetic trait differences within a single tree species. This study tests and demonstrates: (1) UAV hyperspectral images in detecting differences among populations in canopy leaf area, water content, carbon, and nitrogen content as indicators of population-level productivity, fitness, adaptability, and biodiversity they can support, and (2) UAV hyperspectral-thermal-lidar fusion in detecting and classifying 16 populations sourced from different environments across Arizona, USA.MethodsUAV hyperspectral, thermal, and lidar images were acquired from a common garden with 16 different Fremont cottonwood populations growing together. The UAV hyperspectral image was used to calculate spectral indices for canopy leaf area (LAI), canopy water content, nitrogen, carbon, and carbon-to-nitrogen ratio (C:N). The hyperspectral indices (EVI, LAI, PRI, MSI, NDWI, NDNI, NDLI, and C:N) were also examined with the UAV thermal image-derived canopy temperature data for potential correlations. Finally, all hyperspectral bands (n = 487 bands), thermal image-derived canopy temperature, and lidar-derived maximum canopy height estimates were stacked into a single image and then classified to detect 16 different populations of Fremont cottonwood using a random forest classification.ResultsThe UAV hyperspectral indices and canopy temperature were significantly different among populations suggesting that the productivity, fitness, and adaptability of varying populations are significantly different. Many of the UAV hyperspectral indices were strongly correlated with canopy temperature. Populations with greater canopy cover, lower canopy temperature, and greater canopy height were well detected in the UAV hyperspectral-thermal-lidar fusion-based classification (producer’s accuracies of > 75%), whereas populations at low abundance were poorly classified (producer’s accuracies of < 41–65%).ConclusionsThis study demonstrates the first application of UAV hyperspectral-thermal-lidar data fusion in phenotyping. The machine learning-based classification detects various populations within a single tree species. Future studies can use similar UAV data sources, derived variables, and data fusion to detect populations that have better fitness and adaptability to changing environments. Such populations can be strategically managed to sustain healthy landscapes that support diverse communities and species.

The rapid, accurate estimation of leaf nitrogen content (LNC) and plant nitrogen content (PNC) in cotton in a non-destructive way is of great significance to the nutrient management of cotton fields. The RGB images of cotton fields in Shihezi (China) were obtained by using a low-cost unmanned aerial vehicle (UAV) with a visible-light digital camera. Combined with the data of LNC and PNC in different growth stages, the correlation between N content and visible light vegetation indices (VIs) was analyzed, and then the Random Forest (RF), Support Vector Machine (SVM), Back Propagation Neural Network (BP), and stepwise multiple linear regression (SMLR) were used to develop N content estimation models at different growth stages. The accuracy of the estimation model was assessed by coefficient of determination (R2), root mean squared error (RMSE), and relative root mean square error (rRMSE), so as to determine the optimal estimated growth stage and the best model. The results showed that the correlation between VIs and LNC was stronger than that between PNC, and the estimation accuracy of different models decreased continuously with the development of growth stages, with higher estimation accuracy in the peak squaring stage. Among the four algorithms, the best accuracy (R2 = 0.9001, RMSE = 1.2309, rRMSE = 2.46% for model establishment, and R2 = 0.8782, RMSE = 1.3877, rRMSE = 2.82% for model validation) was obtained when applying RF at the peak squaring stage. The LNC model for whole growth stages could be used in the later growth stage due to its higher accuracy. The results of this study showed that there is a potential for using an affordable and non-destructive UAV-based digital system to produce predicted LNC content maps that are representative of the current field nitrogen status.

Retrieval of crop biophysical-biochemical variables from airborne AVIRIS-NG data using hybrid inversion of PROSAIL-D

The spectral characteristics of and the interaction between leaves and light were analysed based on the optical absorption coefficients of foliar water and biochemical components. The equations for calculating the radiative-equivalent water thickness (REWT) of leaves and canopy were presented based on the difference in reflectance at 945 and 975 nm. Because of the direct reflection on leaf surface and the leaf internal scattering, the REWT derived from the Beer–Lambert principle was different from the leaf or canopy equivalent water thickness (EWT). Two independent datasets at canopy or leaf scales were designed to calibrate and validate the relationships between EWT and REWT. The results show that (1) the leaf or canopy REWT can be calculated from the reflectance difference between 945 and 975 nm; (2) the leaf REWT was 3.3 times larger than the EWT with a significant determination coefficient (R 2) of 0.80 for our dataset and 0.86 for the Leaf Optical Properties Experiment (LOPEX'93) dataset; (3) the canopy REWT was 1.4 times larger than the EWT with a significant R 2 of 0.56 for the winter wheat canopy spectral dataset in 2002, and 0.61 for the 2004 dataset. Therefore, the leaf or canopy EWT can be detected by calculating REWT from the difference in reflectance at 945 and 975 nm. Furthermore, because the relationship between REWT and EWT reflected the interaction of light with leaves or canopy, the multiple scattering optical pathlength in the near-infrared (NIR) bands can also be calculated by the ratio of REWT to EWT.

Optical sensing techniques offer an instant estimation of leaf nitrogen (N) concentration during the crop growing season. Differences in plant-moisture status, however, can obscure the detection of differences in N levels. This study presents a vegetation index that robustly measures differences in foliar N levels across a range of plant moisture levels. A controlled glasshouse study with maize (Zea mays L.) subjected to both water and N regimes was conducted in Ottawa, Canada. The purpose of the study was to identify spectral waveband(s), or indices derived from different wavebands, such as the normalised difference vegetation index (NDVI), that are capable of detecting variations in leaf N concentration in response to different water and N stresses. The experimental design includes three N rates and three water regimes in a factorial arrangement. Leaf chlorophyll content and spectral reflectance (400–1075 nm) were measured on the uppermost fully expanded leaves at the V6, V9 and V12 growth stages (6th, 9th and 12th leaves fully expanded). N concentrations of the same leaves were determined using destructive sampling. A quantitative relationship between leaf N concentration and the normalised chlorophyll index (normalised to well fertilised and well irrigated plants) was established. Leaf N concentration was also a linear function (R2 = 0.9, P &amp;lt; 0.01) of reflectance index (NDVI550, 760) at the V9 and V12 growth stages. Chlorophyll index increased with N nutrition, but decreased with water stress. Leaf reflectance at wavebands of 550 ± 5 nm and 760 ± 5 nm were able to separate water- and N-stressed plants from normal growing plants with sufficient water and N supply. Our results suggest that NDVI550, 760 and normalised chlorophyll index hold promise for the assessment of leaf N concentration at the leaf level of both normal and water-stressed maize plants.

Remote sensing of vegetation water content from equivalent water thickness using satellite imagery

Leaf water content is an important parameter in environmental monitoring. The present study investigated the relation between leaf Equivalent Water Thickness (EWT) as a parameter to estimate the leaf water content and the reflectance in 400-2,500 nm spectral range. The data used were the well-known Leaf Optical Properties Experiment 93 (LOPEX93) field collected data. Four hundred leaf samples were used, 320 of which for modelling and the remaining 80 for testing the model. Four different approaches were investigated in this study: 1) linear regression between reflectance in individual wavelength and EWT; 2) the difference of reflectance in two wavelengths and EWT; 3) ratio of reflectance in two wavelengths and EWT; and finally 4) the normalized difference of reflectance in two different wavelengths and EWT. The results showed that the band combinations such as ratio and normalized difference had higher regressions with leaf water content. In addition, the findings of this study showed that some parts of the near infrared (NIR) and short wave infrared (SWIR) of the spectrum provided higher accuracies in EWT assessment, and correlations of more than 90% were achieved. Finally, this investigation showed that a wide range of wavelengths could be used for EWT assessment task. Despite the general belief in using water absorption bands for leaf water content assessment, this study shows that water absorption bands are not necessarily productive as other wavelengths have the potential to generate better results.

[EN] In this thesis it is introduced a method to filter GNSS height time series from vertical displacements derived from load variations in the Earth's surface, obtained from the monthly geopotential coefficients from the GRACE mission. These load variations are originated mainly by the seasonal hydrological variations as it can be confirmed from the hydrological GLDAS model. This filtering method is based on spectral analysis techniques that allow the manipulation of time series in both time and frequency domains.\n\nTo test this method data from 66 IGS Core Network reference stations have been used. The origin of the GNSS data are the weekly solutions provided by the First Data Reprocessing Campaign carried out by IGS. Monthly level-2 data from GRACE Mission provided by GFZ, with a time-span of more than 11 years, have been used to generate hydrological load and equivalent water thickness monthly time series at every reference station from which GNSS data are available. In addition, soil water content time series have also been created from monthly GLDAS data.\n\nWith the periodograms of the four time series datasets a signal with frequency close to 1 cycle per year is detected in the GNSS height, load, equivalent water thickness and GLDAS time series. As this signal has the same frequency in the equivalent water thickness and GLDAS time series it is confirmed that its origin is the hydrological cycle. Therefore the variations observed in the GNSS height and load time series can be explained, at least part of it, by these hydrological variations.\n\nThe spectral filtering technique introduced in this thesis is based on the detection and isolation of the annual signal in the load time series from GRACE data. This process is done in the frequency domain by using the Discrete Fourier Transform (DFT) with the Fast Fourier Transform (FFT) algorithm. Using the linearity property of the DFT, the isolated annual signal from the load time series is subtracted from the DFT of the GNSS height time series in the frequency domain. The new time series in the frequency domain is the GNSS height without the annual signal generated by the hydrological variations in the GNSS reference station area.\n\nAfter testing the spectral filter with times series dataset from 66 reference stations, it is demonstrated that for stations in areas with important hydrological activity the GNSS height time series reduce their WRMS by 18 % on average, with some special cases where WRMS reduction is higher than 25 %. Other parameters like GNSS height vs. GRACE load correlation before and after the filtering process show a reduction of the correlation from 53.9 % to 15.0 % In areas where the hydrological variations are not clearly detected by GNSS and GRACE, like islands, coasts or deserts, there is no WRMS reduction, and other parameters do not show any improvement in the GNSS height time series after the filtering process.

The spectral reflectance of leaves from several Eucalyptus species was measured over the 400–2500 nm wavelengths with a laboratory spectroradiometer. The relationship of reflectance with the gravimetric water content and equivalent water thickness (EWT) of the leaves was analysed. The results showed that EWT was strongly correlated with reflectance in several wavelength regions. No significant correlations could be obtained between reflectance and gravimetric water content. It was also possible to confirm theoretically that reflectance changes of leaves could be directly linked to changes in EWT but not to changes in gravimetric water content. Several existing reflectance indices were evaluated for estimation of leaf water content and some new indices were developed and tested. Two semi-empirical indices developed in this study, ( R 850 - R 2218)/( R 850 - R 1928) and ( R 850 - R 1788)/( R 850 - R 1928), were found to show significantly stronger correlations with EWT than all other indices tested. It was also shown that these new indices were least sensitive to the effects of radiation scatter. The indices ( R 850 - R 2218)/( R 850 - R 1928) and ( R 850 - R 1788)/( R 850 - R 1928) are therefore proposed as two new indices for the remote estimation of vegetation water content.

Remote sensing hail damage on maize crops in smallholder farms using data acquired by remotely piloted aircraft system

Leaf nitrogen concentration (LNC) is an important indicator for accurate diagnosis and quantitative evaluation of plant growth status. The objective was to apply a discrete wavelet transform (DWT) analysis in winter wheat for the estimation of LNC based on visible and near-infrared (400–1350 nm) canopy reflectance spectra. In this paper, in situ LNC data and ground-based hyperspectral canopy reflectance was measured over three years at different sites during the tillering, jointing, booting and filling stages of winter wheat. The DWT analysis was conducted on canopy original spectrum, log-transformed spectrum, first derivative spectrum and continuum removal spectrum, respectively, to obtain approximation coefficients, detail coefficients and energy values to characterize canopy spectra. The quantitative relationships between LNC and characteristic parameters were investigated and compared with models established by sensitive band reflectance and typical spectral indices. The results showed combining log-transformed spectrum and a sym8 wavelet function with partial least squares regression (PLS) based on the approximation coefficients at decomposition level 4 most accurately predicted LNC. This approach could explain 11% more variability in LNC than the best spectral index mSR705 alone, and was more stable in estimating LNC than models based on random forest regression (RF). The results indicated that narrowband reflectance spectroscopy (450–1350 nm) combined with DWT analysis and PLS regression was a promising method for rapid and nondestructive estimation of LNC for winter wheat across a range in growth stages.

In preparation for new-generation imaging spectrometer missions and the accompanying unprecedented inflow of hyperspectral data, optimized models are needed to generate vegetation traits routinely. Hybrid models, combining radiative transfer models with machine learning algorithms, are preferred, however, dealing with spectral collinearity imposes an additional challenge. In this study, we analyzed two spectral dimensionality reduction methods: principal component analysis (PCA) and band ranking (BR), embedded in a hybrid workflow for the retrieval of specific leaf area (SLA), leaf area index (LAI), canopy water content (CWC), canopy chlorophyll content (CCC), the fraction of absorbed photosynthetic active radiation (FAPAR), and fractional vegetation cover (FVC). The SCOPE model was used to simulate training data sets, which were optimized with active learning. Gaussian process regression (GPR) algorithms were trained over the simulations to obtain trait-specific models. The inclusion of PCA and BR with 20 features led to the so-called GPR-20PCA and GPR-20BR models. The 20PCA models encompassed over 99.95% cumulative variance of the full spectral data, while the GPR-20BR models were based on the 20 most sensitive bands. Validation against in situ data obtained moderate to optimal results with normalized root mean squared error (NRMSE) from 13.9% (CWC) to 22.3% (CCC) for GPR-20PCA models, and NRMSE from 19.6% (CWC) to 29.1% (SLA) for GPR-20BR models. Overall, the GPR-20PCA slightly outperformed the GPR-20BR models for all six variables. To demonstrate mapping capabilities, both models were tested on a PRecursore IperSpettrale della Missione Applicativa (PRISMA) scene, spectrally resampled to Copernicus Hyperspectral Imaging Mission for the Environment (CHIME), over an agricultural test site (Jolanda di Savoia, Italy). The two strategies obtained plausible spatial patterns, and consistency between the two models was highest for FVC and LAI (R2 = 0.91, R2 = 0.86) and lowest for SLA mapping (R2 = 0.53). From these findings, we recommend implementing GPR-20PCA models as the most efficient strategy for the retrieval of multiple crop traits from hyperspectral data streams. Hence, this workflow will support and facilitate the preparations of traits retrieval models from the next-generation operational CHIME.

A TPE based inversion of PROSAIL for estimating canopy biophysical and biochemical variables of oilseed rape