Estimation of key biophysical parameters related to crop stress through new remote sensors and multi-crop in situ data

Resumen

Evidence suggests that human-induced greenhouse gases emissions have altered our climate at a relatively rapid rate, rising the global temperatures and inducing drastic changes in precipitation patterns to water-limited environments and agricultural areas, restricting crop yield, production rates and food availability. Biophysical parameters, such as leaf water content (LWC), leaf area index (LAI) or leaf chlorophyll content (LCC), are considered important indicators of health, growth and productivity of crops. As they define the status of the vegetation, they provide important inputs to models quantifying the exchange of energy and matter between the land surface and the atmosphere. Also, knowledge of their spatial and temporal distribution is highly useful for regional or global-scale applications related to crop monitoring, weather prediction and climate change studies. The direct field measurements of biophysical parameters require continuous updates and can be extremely time-consuming and expensive, therefore, an alternative estimation methodology is necessary. Remote sensing from satellite and airborne sensors has become a commonly used technique for monitoring agricultural areas due to its ability to acquire synoptic information at different times and spatial scales. For an optimal agricultural monitoring by remote sensing, the spatial resolution should be at least 20 m and, preferably, 10 m, and a temporal resolution of less than a week, in order to follow-up acute changes in the crop condition and provide a timely response in management practices. In this context, the Sentinel-2 (S2) missions from the European Space Agency (ESA) Copernicus program respond to such operational requirements. S2 is a constellation of satellites, with currently the Sentinel-2A (S2A) and Sentinel-2B (S2B) satellites in orbit. Together, they provide a 5-day nominal revisit, at the Equator, of the Earth’s land surfaces with a 10, 20 and 60 m of pixel size. S2A and S2B carry on-board a virtually identical sensor, the Multi-Spectral Imager (MSI), covering a spectral range from 443 to 2190 nm through 13 bands located in the visible (VIS, 440 – 690 nm), the near-infrared (NIR, 750 – 1300 nm) and the shortwave-infrared (SWIR, 1300 – 2500 nm) spectral regions. With the narrow band configurations specifically located for vegetation monitoring, the S2 missions improve the temporal, spatial and spectral resolution of remote sensing data, compared to other multi-spectral missions, such as Landsat, and offers great opportunities for agricultural monitoring. The mission’s main objective is providing quality information for agricultural and forestry practices and, hence, helping management and food security applications. In addition, ESA has incorporated a user-friendly Biophysical Processor toolbox within the SNAP (Sentinel Application Platform) program, for the straightforward delivering of biophysical parameter products, such as LAI and canopy chlorophyll content (CCC). These parameters are automatic products, associated with a quality indicator, produced through an artificial neural network (ANN) which has been trained with simulated spectra generated from well-known radiative transfer models (RTMs), i.e., physically-based models that describe the absorption and scattering of light throughout the leaf, canopy and atmosphere. Only eight bands are used (B3, B4, B5, B6, B7, B8a, B11 and B12) for the biophysical parameter products estimation. This way, the values of biophysical parameters can be obtained in any study area with available S2 images, being very useful in operational agronomic studies. On the other hand, hyperspectral sensors are becoming more and more relevant, which will be available by satellites such as the future EnMAP (Environmental Mapping and Analysis Program) mission or the recently launched PRISMA (PRecursore IperSpettrale della Missione Applicativa) satellite, or through airplanes punctually over the corresponding study area, such as AVIRIS (Airborne Visible/Infrared Imaging Spectrometer) and HyMap (Hyperspectral Mapper) sensors. This type of sensors allows to identify and to discriminate with great precision the surface, thanks to its high spectral resolution, allowing the detection of anomalies with precision. The future FLEX (Fluorescence Explorer) mission should also be specially highlighted, of which ESA is currently carrying out scientific development. The main objective of FLEX mission is to observe the vegetation functioning from space, based on the emitted fluorescence signal. The fluorescence measurement provides direct information of the photosynthesis process, constituting a novel tool for the rapid detection of vegetation stress, before damage is irreversible. In this context, all the methodologies developed to estimate biophysical parameters that are physiological state indicators are, therefore, fundamental to understand the behaviour of terrestrial vegetation at the global scale. In general, there are three approaches for estimating biophysical parameters from remotely sensed data, i.e., (1) empirical retrieval methods, which consist of relating the biophysical parameter of interest against spectral data by means of simple relations (e.g., vegetation indices—VIs), (2) statistical methods, which define complex regression functions according to information from remote sensing data (e.g., artificial neural network—ANN) and (3) physically-based retrieval methods, which typically refers to the inversion of RTMs. There are numerous scientific contributions related to the biophysical parameters estimation through remote sensing, but most of these studies focus on a single or a small number of crop types. The challenge arises when these remote sensing techniques are applied in a general context, i.e. for a high diversity in crop types. This is important as agricultural areas are often composed of multi-crop types but also because the retrieval algorithms should be robust on a global scale. This Thesis attempts to achieve techniques to assess the general character of three important vegetation health indicators for crop monitoring: canopy water content (CWC), LAI and CCC. The methodologies finally proposed aim to present a physical basis for applied crop monitoring and produce accurate results for a wide range of crop types. The starting point of this Thesis has been the proposal of a methodology defined for the estimation of CWC. For this purpose, the SPARC03 (Spectra bARax Campaign) dataset has been used. This field data was obtained in July 2003 in Albacete (Spain) and is composed of CWC destructive values of six different crop types, as well as the spectral information obtained from the hyperspectral HyMap airborne sensor. This sensor has a wavelength range between 430 nm and 2490 nm and, for this campaign specifically, images with a spatial resolution of 5 m were obtained. Using the multi-crop SPARC03 dataset, commonly used water content index formulations were analysed and validated for the variety of crops, indicating possibilities for improvement. Instead of using specific band combinations, influenced by other parameters such as chlorophyll content and LAI, a more physically based approach was employed. After a study of the HyMap sensor spectra and the modelled spectra simulated with PROSAIL, it was observed that, with simulations assuming no water content in the leaves, the spectrum presented a straight-line shape in the range of water absorption between 800 and 1200 nm. The slope and the magnitude of this reflectance line depended mainly on LAI. This line was used as a reference to define the Water Absorption Area Index (WAAI), which consists in determining the difference between the area under the null water content reference line and the area under the measured reflectance in the range 911 – 1271 nm, where the influence of water content is maximal. The WAAI is an area index essentially developed for high spectral resolution data. However, since hardly any of the currently operational satellite sensors are equipped with such a high spectral resolution, the so-called Depth Water Index (DWI) is proposed as a possible guide for the configuration of future optical superspectral sensors. The DWI is a simplified four-band index based on the spectral depths produced by the water absorption at 970 and 1200 nm and two reference bands. Hence, in this study two new CWC estimation indices were defined, by means of which the CWC of multi-crop areas can be estimated, obtaining more promising results (R2 > 0.7) than the conventional indices (R2 < 0.6). Secondly, in the same context of improving remote biophysical parameter retrieval of important crop growth parameters, a simple and operative methodology for the estimation of the LAI was developed. The LAI is fundamental both for its function as an indicator of the plant physiological state and for its essential role in scaling the leaf-based measurable parameters water and chlorophyll content at canopy level. LAI can be distinguished in two types. There is the LAIgreen, representing the leaves which are photosynthetically active, being the most common type of LAI and the one studied in this Thesis, and, on the other hand, there is the LAIbrown, representing the leaf area normalized which is senescent and losing photosynthetic function. Two large field campaigns were conducted that resulted in two independent datasets, composed of LAI in situ values ranging from 0 to 4.5 m2/m2, obtained at test sites in Valencia (Spain) and Foggia (Italy). The Valencia dataset is composed of LAI in situ values of 13 different crop types from the Huerta de Valencia and Foggia’s dataset is composed of information from 3 crop types. Simultaneous satellite data from S2 overpasses was analysed for the two test sites. The first analysis consisted in applying the indices commonly used in the literature to estimate LAI for the Valencia dataset, obtaining statistics that can be improved given a saturation process of the method for high LAI values (≥ 3). Subsequently, an analysis was performed to verify better band combinations for the estimation of LAI in multi-crop zones, minimizing the effects of the saturation process. A physically meaningful combination of bands was chosen for the normalized index (R865 - R705)/(R865 + R705), with the 705 nm band located in the red-edge region, a spectral area which balances the influence of strong chlorophyll absorption and minimal scattering at moderate-high LAI values, and the 865 nm band located in the NIR region, as a reference band. Improved statistics were obtained by applying this index for a partial dataset, with a linear adjustment. The new index proposed, the Sentinel-2 LAIgreen Index (SeLI), avoids saturation of the LAI estimation, allowing an improved LAI retrieval for different crop types. The third study of this Thesis consisted in the analysis of the different existing methodologies for LAI and CCC retrieval, two key parameters for the estimation and monitoring of evapotranspiration (ET) of vegetation. In particular, this study performed a comparative analysis of empirical VIs, semi-empirical approaches (CLAIR - Clevers Leaf Area Index by Reflectance model with fixed and calibrated extinction coefficient) and artificial neural network S2 products (ANN S2 products) to analyse the most optimal approach for LAI and CCC estimation, from a statistical and operational point of view, using concomitant S2 band information. The main objective of this analysis was to verify how the method used for the estimation of these input parameters influences the final result of crop growth models, here using the Penman-Monteith model, widely applied for the estimation of ET. One of the main strengths of this third study is that in order to carry out the comparative analysis of the different estimation methods, four datasets composed of in situ LAI and CCC values from different crop types and different plot sizes were used, allowing to obtain robust results. The datasets collected in Italy (Caserta and Tarquinia), Argentina (Bahía Blanca) and Spain (Valencia) covered in situ data with a LAI range of 0 to 5, and a CCC range of 0 to 5.4 g/m2. SeLI is the index that presents good statistics in each dataset (R2 > 0.71, RMSE < 0.78) for LAI parameter and for the CCC, the CIred-edge (R2 > 0.67, RMSE < 0.62 g/m2). Both indices use bands located in the red-edge region, highlighting the importance of this spectral region. The biggest problem with VIs was that the SeLI index produced saturation with LAI values > 3. On the other hand, the LAI CLAIR model estimated with fixed extinction values (α*) of 0.41 for herbaceous crops and 0.30 for tree species obtained good statistics (R2 > 0.63, RMSE < 1.47) and the CLAIR model optimizing the parameter α* (CLAIRopt) for each of the study areas only slightly improved the RMSE in the two Italian datasets (RMSE ≈ 0.70). ANN S2 products produced statistics very similar to the VIs, but without producing saturation at high LAI values. It should be mentioned that with the Valencia area dataset, all the methodologies produce improvable results (R2 < 0.5, RMSE > 0.7 for LAI; R2 < 0.4, RMSE > 1.0 g/m2 for CCC), due to the high soil influence in this area and the small size of the plots (< 1 ha). Finally, the influence of the LAI parameter on the Penman-Monteith ET model adapted to remote sensing was analysed, which derives the reference (ETo) and potential (ETc) crop ET. During recent years, there has been a consistent effort to estimate vegetation parameters from remotely sensed data, allowing to adapt the Penman-Monteith equation for direct use with remote sensing data, minimizing time and cost. To evaluate ETc throughout the season, the dataset from the Tarquinia test site with available temporal in situ LAI data of two crop types, wheat and tomato, was used. Due to the absence of independent reference ET data, the estimated ETc with the LAI in situ values were considered as the proxy of the ground-truth. Also, the results were compared with ETc estimated as the product of the reference evapotranspiration (ETo) and the crop specific coefficient (Kc) derived from FAO (Food and Agriculture Organization) table values, according to the methods commonly used in operational studies. As a result, it was obtained that the ETc values obtained with the LAI estimated with SeLI index were the closest to the truth-terrain in the case of wheat, while for tomato the best correlation was obtained with the ETc estimated with the ANN S2 LAI product. Therefore, with all the above, it can be concluded that VIs produce the best statistics for low LAI values, but the ANN S2 products are the only ones that do not produce saturation towards higher values, both for the direct estimation of the LAI and CCC parameters and the derived estimation of ET. It demonstrates the great potential of ANN S2 products for operational use in the monitoring of agricultural areas. The influence of soil can compromise the retrieval results of all the different methodologies defined in this Thesis when the fractional vegetation cover is low (FVC < 30 %). Future work should consider the soil reflectance in order to improve the general retrieval of vegetation biophysical properties. The studies presented in this Doctoral Thesis demonstrate the high value of current operational high spatial resolution S2 satellites for the monitoring of biophysical crop parameters in the context of an increasing demand to secure optimal growth and food production. The availability of high spatial and temporal resolution of multispectral band sensors allow the application of biophysical parameters for water and pigment content at an unprecedented spatial and temporal detail. Combined with growth or evapotranspiration models, these biophysical parameters allow even further the remote scaling of biophysical processes at the agricultural unit scale. This was demonstrated by integrating the S2-derived LAI in the adapted Penman-Monteith equation for evapotranspiration modelling. Future hyperspectral missions will even allow a further improved retrieval of biophysical parameters, as demonstrated in the study on the Water Absorption Area Index. Application of such spectrally detailed monitoring will further allow to improve the monitoring of crop growth and functioning at different scales.