Integration of Optical and Microwave Satellite Data for Monitoring Vegetation Status in Sorghum Fields

1. Introduction

Monitoring crop conditions is essential for optimizing agricultural practices and reducing water wastes. The increasing impacts of climate change, with more and more frequent extreme weather events and persistent droughts [1,2], has strong effect on crop productivity [3], health and yields [4]. Recently, these risks have been expanding in several regions within the Mediterranean basin [5]. Therefore, an effective management of soil and water resources can be considered as a key factor to achieve sustainable food production [6]. In this context, monitoring vegetation from space can greatly enhance our understanding of Earth's surface processes as hydrological cycle and crop productivity. With the rapid advancements in remote sensing technology, estimation of several natural parameters with high spatiotemporal resolution is becoming more and more feasible. Remote sensing techniques support observations of the most important surface characteristics according to different sensors: the optical ones, such as visible-near-infrared [7], hyperspectral [8], and thermal infrared sensing [9], have limited penetration capabilities, making impossible to obtain soil information in presence of dense vegetation layers and of the entire vegetation structure in case of well-developed plants. Nevertheless, the signal in these wavelengths (λ) is very sensitive to vegetation pigments and photosynthetic activity [10]. Furthermore, the use of these sensors is hampered by weather conditions like the presence of clouds [11], rainfalls and solar illumination. Microwave sensors, such as synthetic aperture radar (SAR), have instead a deeper penetration and better capability to observe objects in absence of solar light and cloud cover. In this wavelength range, the dielectric properties of vegetation are affected both by water content and geometric characteristics of plants, including the size of leaves, branches, and stems [12]. These elements influence scattering and absorption mechanisms, which vary according to frequency, incidence angle of observation, and polarization [13,14]. Moreover, the electromagnetic response of soil under vegetation greatly influences the backscattering (σ⁰), due its high sensitivity to soil moisture content (SMC) and surface roughness [15]. The dielectric properties govern the interactions of backscatter with the medium through various phenomena such as absorption, scattering, and penetration The penetration of microwave signal increases at lower frequency, thus is higher in L- and P-band [16,17].

The launch of S-1 and S-2 missions by the European Space Agency (ESA) has significantly increased the possibility of monitoring crop conditions and soil moisture from space [18,19]. S-1, equipped with a C-band SAR, offers all-weather, day-and-night imaging capabilities that are crucial for consistent monitoring of agricultural fields. S-2 provides high-resolution optical imagery by using 13 spectral bands, allowing detailed assessment of vegetation health and vigor by most widely used vegetation indexes [20].

In this paper, a very important index, i.e. the Photochemical Reflectance Index (PRI), has been taken into consideration because it is widely considered as an indicator of photosynthetic efficiency, productivity and related to water-stress index [21]. PRI is obtained from the normalization between the difference of reflectivity at 531 nm and a reference wavelength, typically at 550 or 570 nm and it is used to assess Radiation Use Efficiency (RUE) at leaf, canopy, and ecosystem levels [22,23,24,25]. The consistency of the RUE–PRI relationship suggests a functional convergence in the components affecting carbon uptake efficiencies [26]. Then, complementing NDVI-like indices with PRI can improve the precise monitoring and management of agricultural ecosystems [27]. PRI can be measured in-situ by using portable sensors; however, due to the bands needed for its computation, PRI can be derived from MODIS but not from S-2.

Due to the peculiar characteristics of optical and microwave satellite data, the integration of these two types of measurements enables a comprehensive approach for assessing crop health, water status, and soil conditions, by combining different sensitivities, thus obtaining more in-depth information on the canopy [28]. The experimental results have been confirmed and evaluated by using well assessed electromagnetic models such as the Water Cloud Model (WCM) which was used for estimating the vegetation water content. Although the WCM is not conceived recently, it is still used at present in in remote sensing studies to estimate the water content in the canopy, the aboveground biomass and the soil moisture [29,30,31,32,33], since it offers a good tradeoff between accuracy and computational cost. In this study, WCM has been reappraised as proposed in [15], by coupling it with the Integral Equation Model (IEM) [34], Oh [35] and the Dobson model [36].

In this paper, the vegetation status and health in two sorghum fields close to Florence, Italy has been evaluated by exploiting on-site measurements of PRI and further integration with the PWC obtained from S-1 data. The analysis and data collection have been performed during summer 2022 [37] and in 2023 [38]. Additionally, we used a series of well-known S-2 derived indices, that are sensitive to plant health conditions to confirm their sensitivity to vegetation water status.

The novelty of this research work consists of the simultaneous use of SAR and multispectral data for estimating PWC and PRI. In particular, SAR data have been used as inputs of electromagnetic models as WCM and IEM for retrieving PWC, which in turn was correlated with PRI. Some indices derived from S-2 have been also related to PRI. Another important aspect was the use of morning and afternoon S-1 passes for exploiting the SAR sensitivity to the diurnal variations of PWC and PRI.

The results of this study are expected to contribute to improve agricultural management practices and enhance the understanding of remote sensing applications in monitoring crop water status and health.

2. Materials and Methods

2.1. Test Area

During summers of 2022 and 2023, measurement campaigns were carried out in two sorghum fields (Figure 1 and Figure 2). Both fields were flat, rainfed, and located close to Florence, in Italy with at the following coordinates: latitude 43.829625°, longitude 11.151517°; and 43.827674°, longitude 11.149736° [39]. Measurements were performed from 2^nd of July to 1^st of September in 2022 and from 2^nd of July to 2^nd of October in 2023.

2.2. Satellite Data

S-1 is a constellation of two polar-orbiting satellites, Sentinel-1A (S-1A) and Sentinel-1B (S-1B), sharing the same orbital plane and carrying a SAR in C-band (5.4 GHz). Unfortunately, S-1B experienced an anomaly rendering it unable to deliver radar data since the end of 2021, which reduced the revisiting frequency of most of the Earth's surface to 12 days from the original 6 days [40]. The orbits can be of ascending or descending type. Ascending orbits acquire images in the afternoon, while descending orbits in the morning. The characteristics of satellite data have been reported in Table 1.

S-2 has a temporal resolution of 5 days, overpassing the area at 10:13 UTC (12:13 UTC). S-2 has two identical satellites which operate simultaneously with a phase difference of 180° in a sun-synchronous orbit. The platform provides a total of 13 spectral bands with a spatial resolution ranging between 10 and 60 m. To understand the health status of the sorghum crops in the considered periods, the evolution of the following four S-2 derived indices were analyzed: NDRE (Normalized Difference Red Edge Index), RGR (Red Green Ratio), PSRI (Plant Senescence Reflectance Index) and SIPI (Structure Insensitive Pigment Index) (Table 2). These indices have been selected among many available indices as the best related to PRI, showing high determination coefficients (R²>0.5, 0.6), as it can be observed in the diagrams and tables of the section 3 Results.

Furthermore, the median value of the pixels in each field was chosen as the estimator, as for S-1 data.

Both S-1 and S-2 data were gathered from Google Earth Engine (GEE). GEE is a cloud computing platform designed for processing satellite imagery and other Earth observation data, with the capability to handle large datasets by distributing computations across multiple servers worldwide [45]. Users can interact with the platform via an interactive development environment (IDE), using Python and JavaScript libraries [46]. Only L2-A, S-2 cloud-free scenes were processed. Scenes from S1 and S2 acquired in (almost) the same day were considered as couple and described in Table 3. In these dates the plants reached 70% of their maximum height development in the field had a height corresponding at least to the 70% of the full height in order to exclude low PWC values because when plants are small and just started their photosynthetic activity, a low value of PWC can be associated to high PRI, especially on hot and sunny days. In the same way, for S-2 data, when the contribution of the soil to the reflectance signal was significant due to low vegetation fractional cover, we decided to exclude the points where the plants height were lower than 70% [47,48] compared to the full height.

Integrating the PRI in situ and combining this information with the PWC, obtained from S-1 data, and a series of well-known indices from S-2, which are sensitive to various plant conditions, a more precise and comprehensive investigation of crop health and water status over the two-year study period was achieved.

2.3. In-Situ Measurements

SMC was measured using the SPECTRUM-350 TDR probe [49], and vegetation PRI was assessed through the PlantPen PRI210^® [50], during the satellite passes. In situ measurements started 20 minutes before and continued until 20 minutes after the satellite acquisitions. More in detail, PRI values were obtained by taking three measurements for leaf on four different leaves from a sample of ten sorghum plants. SMC values were obtained by taking three measurements per sample across ten distinct areas of the field. Given the relatively small size of the fields, a single value of PRI and SMC was associated with each satellite pass. To reduce the influence of potential outliers on ground and remote sensing measurements, the values of σ⁰, reflectance in the S-2 bands, and PRI over the fields were represented by the median, while SMC values were obtained using the mean, due to the strong spatial homogeneity of the measurements across different field zones. Consequently, each satellite pass was associated with a value of σ⁰ and reflectance corresponding to the in-situ measurements. Additionally, the soil roughness under bare soil conditions or when vegetation is very small was measured, using a needle profilometer to measure correlation length (Lc) and the standard deviation of surface height (s) through an image processing script implemented in Matlab. Moreover, PWC was obtained by measuring the fresh weight of a sample consisting of three plants in the field, then drying in an oven at 80°C for 24 hours. The plant density in the fields was about 33 plants per m² in 2022 and 40 per m² in 2023. The PWC was measured once for each pair of consecutive descending-ascending satellite passes, since the water variation over 36 hours is negligible [51] and comparable to the intrinsic uncertainties of the instruments.

2.4. Electromagnetic Models

In order to analyze and interpret the satellite data, some electromagnetic models were employed, including the WCM, the IEM, in combination with the Oh model for estimating the cross polar σ⁰ [35], and the Dobson model for calculating the dielectric permittivity of the soil. The WCM considers the canopy and underlying soil separately, allowing the estimation of PWC by accounting the vegetation attenuation effects. In our study, we measured the SMC and the variables to estimate PWC that minimize the difference between the σ⁰ measured by S-1 and simulated by the WCM by the formula:

$${\sigma }_{WCM}^0=B\times PWC\times cos{\theta }_i\left(1-{\displaystyle \frac{1}{exp(2·APWC·sec{\theta }_i)}}\right)+{\displaystyle \frac{{\sigma }_s^0}{exp(2·APWC·sec{\theta }_i)}}$$

(1)

where:

${\theta }_i$= incidence angle

${\sigma }_s^0$= simulated soil backscatter

$A,B$= free parameters

To estimate the parameters A and B, the couple (A, B) that minimizes the difference between the measured σ⁰ by S-1 and the one simulated by the WCM was determined. A and B parameters were estimated within a discrete set of values {A₁, A₂, A₃, … A_M} and {B₁, B₂, B₃, … B_N}. For each combination of elements from these sets, the absolute difference between the σ⁰ measured by S-1 and σ⁰_WCM was calculated in linear scale. This process resulted in a matrix of σ⁰ linear differences with a number of rows corresponding to the cardinality of A (M) and columns corresponding to the cardinality of B (N). The cell with the smallest value indicates the couple (A, B), that minimize the difference between measured and simulated σ⁰. The A and B extracted for this study were 0.001 and 0.244, respectively, in VH polarization and exponential autocorrelation function for soil roughness estimation. In VV polarization the estimated PWC values were less reliable due to the higher sensitivity of VV polarization to SMC than VH, as it has already been demonstrated in literature [52] and verified experimentally during the 2022 campaign [37].

3. Results

During the two summer seasons the impact of SMC on photosynthetic activity was investigated. As it can be noted in Figure 3, where the rainfall data downloaded at [53] were reported, a very dry season occurred during summer 2022 and rainfall arrived shortly before the harvest in the last days of August. In March 2022, the rains were concentrated in only 2 thunderstorms, whereas, in 2023, rainfall was concentrated in spring and a dry September was present, when the crop was in senescence phase. The observation revealed that PRI has lower peaks in summer season of 2022, also indicating reduced photosynthetic activity in response to lower SMC. This behavior is illustrated in Figure 4, where the boxplots of measured PRI and SMC were compared for 2022 and 2023. The boxplots showed for PRI (left) and SMC (right) for each field and for the entire year. PRI values tend to be higher in 2023 than in 2022 (although with higher dispersion), whereas SMC increases significantly from 2022 to 2023 (although generally reaching low values: 5-10% in 2022 and 10-15 in 2023.

To better investigate the water and health status of vegetation, the indices listed in Table 2 were associated to the PRI measurements which have been collected simultaneously to S-2 passes. Figure 5 and Figure 6 show the relationships of the optical indices obtained from S-2 with PRI values measured on ground in the two years of observation, in the top leaves of vegetation and all leaves measured, respectively. Given the almost negligible penetration of the optical wavelengths inside the vegetation, the results using only the top leaves of vegetation tend to be better than those obtained using all the leaves.

Then, the relationship between the PWC derived from S-1 data and the PRI measured in-situ was explored to better understand the vegetation water status and its impact on photosynthetic efficiency. The results of Figure 7 showed a clear relationship between PWC and PRI, provided only data collected in well-developed conditions (more or equal to 70% of the maximum plant height) or senescence stage should be taken into consideration, as already stated in section 2.2.

To go deeper into this investigation, the relationships between PWC estimated from S-1 and the S-2 indices, already indicated in Table 2, were analyzed. The Figure 8 illustrates such relationships, highlighting the effectiveness of integrating SAR and optical remote sensing data for a more complete agricultural monitoring.

Finally, an analysis was performed to investigate the variations of PWC and PRI [54]. This was achieved by using the difference between ascending and descending orbits of S-1, taking advantage of the different time of their acquisitions (i.e. ascending in the afternoon and descending in the morning). The differences in PWC $\left(\Delta PWC\right)$ and PRI $\left(\Delta PRI\right)$ have been computed considering all orbit acquisitions, removing values corresponding to measures carried out in presence of surface water or mud and thus with extremely high SMC values (Figure 9). We made sure that $\Delta PWC$ were not due to variation in plant biomass, by taking into consideration only well-developed plants (in a range of PWC between 1.85 and 6 Kg/m²). From the diagram we can note that there is a rather clear trend between ΔPRI and ΔPWC, thus confirming the ability of both optical and SAR systems in assessing the water status of crops.

4. Discussion

The use of both SAR (of S-1) and multispectral and (S-2) can provide a robust approach for monitoring crop conditions, particularly for assessing vegetation water status under varying environmental conditions. The study demonstrated that the combination of these data is able to catch the complex interactions between soil and vegetation, providing valuable insights that single-sensor approaches might overlook or fail to evaluate. The relationship between PRI, an index sensitive to changes in the photosynthetic efficiency of plants (measured using a portable device) and SMC (measured on ground with TDR probes) highlighted how plants react with respect to stress environment conditions. This finding underscores the potentiality of combining SAR and optical indices to better understand plant-water-health relationships, which is crucial for improving irrigation practices and crop management. Data from S-1 can thus be used to estimate either PWC and SMC, whereas, the optical data from S-2 provided complementary information on leaf pigments and leaf water status, enhancing overall accuracy. Given the strong relationship between PRI and S-2 indices, similar information can also be derived using both satellite datasets avoiding the need of in-situ PRI leaf measurements. This novel approach offers a reliable method for identifying the plant water status. By leveraging the complementary strengths of SAR and optical sensors, it is possible to develop more accurate models for predicting crop yield, optimizing water usage, and responding to environmental stressors in almost real time. Furthermore, the analysis of S-1 data across different orbital passes (ascending and descending) highlighted notable temporal variations in PWC revealing variations in plant status. These variations [55], highlight the sensitivity of SAR measurements to changes in PWC, emphasizing the importance of integrating multi-sensor data. The corresponding observed variations of PRI offer a powerful tool for assessing crop health, particularly in semi-arid regions [56] where water availability is a critical factor. Although PRI cannot be measured from S-2, the correlations found between PRI and some selected optical indices derived from S-2 data confirms the possibility of investigating the water status and stress of crops from satellite. Furthermore, since most agricultural fields are rainfed, these methodologies can be used to understand the local health status of agricultural lands that in recent years have seen increasing damages caused by climate change, such as a yield loss [57,58,59].

5. Conclusions

This study aimed at exploring the integration between different satellite remote sensing data, i.e. SAR and multispectral data, to better investigating the water and health status of agricultural crops. The satellite data were compared with in-situ measurements of SMC, surface roughness, plant parameters, as well as leaf PRI measurements. The use of open-source data and products, such as those provided by the European Space Agency's Copernicus platform elaborated in GEE, ensured the full reproducibility of the methodologies developed in this study. The integration of electromagnetic models such as the WCM and IEM allowed for a detailed analysis and interpretation of the obtained information. The comparison of data from very close S-1 and S-2 passes allowed the assessment of some relationships between optical indices and the daily variations of PWC, which was directly measured in the field. PRI values were associated with several S-2 indices, showing a correlation between these variables, consistent with studies reported in the literature. These comparisons have clearly highlighted the relationships between the analysed multispectral indices and the vegetation water content, and consequently, the health status of the plants. The analysis also revealed a correlation between the variations in PWC and the xanthophyll cycle, which characterizes the photosynthetic state of the plants. Given PRI's sensitivity to plant health and stress, the study demonstrated that consistent variations in PRI may correspond to changes in water content as measured by satellite sensors. The methodology adopted in this study suggests several paths to follow for future research. One possible direction is the validation of the methodology on larger scales, for instance, by using PRI values obtained from the MODIS satellite platform, although with worse spatial resolutions, in the order of 500 meters or 1 kilometer. In any case, the high correlations found between optical indices from S-2 and PRI allow the possibility of investigating the crop water status directly from satellite. An additional area of investigation could be the represented by the analysis of the problem from a radiometry perspective, as well as the possibility of improving vegetation parameterization using more complex electromagnetic models and potentially for different frequencies. Consequently, additional experimental data and analyses will be necessary to move this research forward.