**1. Introduction**

Spaceborne monitoring of agricultural landscapes is predominantly performed using optical sensors and synthetic-aperture radar (SAR). The use of passive optical remote sensing in the visual, near-infrared, shortwave infrared, and thermal spectral regions for the estimation of agricultural variables is well established [1–10]. However, optical sensors are limited by cloud cover. To overcome this problem, previous studies have suggested combining observations acquired at different times by several optical sensors [11–15], but even this approach does not always produce enough cloud-free observations to monitor cloudy regions effectively. Moreover, leaf area index (LAI) estimation from optical imagery suffers from a saturation effect when the LAI is greater than 3 [16–20]. Overcoming this limitation is desirable since LAI is commonly used as a measure of crop growth, nitrogen, and fertilization status estimation [21]. The LAI is also a good proxy for vegetation vigor [22,23], and a good yield predictor [24–27].

**Citation:** Kaplan, G.; Fine, L.; Lukyanov, V.; Manivasagam, V.S.; Tanny, J.; Rozenstein, O. Normalizing the Local Incidence Angle in Sentinel-1 Imagery to Improve Leaf Area Index, Vegetation Height, and Crop Coefficient Estimations. *Land* **2021**, *10*, 680. https://doi.org/ 10.3390/land10070680

Academic Editors: Carmine Serio, Guido Masiello and Sara Venafra

Received: 29 May 2021 Accepted: 22 June 2021 Published: 28 June 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

This study proposes complementing optical remote sensing with SAR to overcome these obstacles in monitoring vegetation properties and to facilitate better agricultural practices [28]. SAR penetration of the canopy can mitigate saturation in LAI estimation [29–31]. Moreover, since SAR can penetrate clouds, it produces high-quality imagery even in adverse weather conditions [32]. In addition to the LAI, remote sensing can be used to estimate other variables such as the crop coefficient (Kc) and height. Kc-based estimation of crop water consumption is one of the most commonly used irrigation managemen<sup>t</sup> methods [33,34]. Crop height is a good predictor of the aboveground biomass [35] and is commonly used by growers as a proxy for crop development. Therefore, deriving reliable SAR-based LAI, Kc, and height estimation models can facilitate better agricultural monitoring, especially in cloudy regions.

Several studies have employed spaceborne SAR for agricultural purposes [36–40] and demonstrated that quad-polarization SAR (e.g., RADARSAT-2, TerraSAR-X/TanDEM-X) could be used for crop monitoring. However, quad-polarization SAR images currently come at a high cost that limits their use in routine monitoring of crops and in research. Since 2014, the Sentinel-1 mission, consisting of two polar-orbiting satellites, provides a dual-polarization alternative at no cost to the user. These satellites have a revisit time of six days at 30◦ latitude at the same viewing geometry and a 10 × 10 m pixel size, thus having significant potential for agricultural applications.

One of the most critical challenges in creating time series of SAR imagery is the dependence of radar backscatter on the incidence angle [41]. The incidence angle is defined by the incident radar beam and the vertical (normal) to the surface. More specifically, the local incidence angle (θ) takes into account the local relief. The backscatter is weaker in images acquired at shallow incidence angles compared to images acquired at steeper incidence angles; therefore, the same object has different and uncomparable backscatter values in images acquired with different incidence angles. Given the dependence of the backscatter's intensity on the incidence angle, previous studies have underlined the need to correct this effect [42,43]. Until now, many studies using C-band SAR imagery from Sentinel-1, RADARSAT-2, and RISAT-1 for agricultural monitoring only used a subset of the available imagery acquired from either ascending or descending orbits with a limited range of incidence angles. Accordingly, these studies discarded imagery acquired at incidence angles that fell outside certain margins (Table 1). This practice might exclude more than half of the available images from the time series. Moreover, empirical models developed based on these limited datasets are likely applicable only for the same range of incidence angles. Therefore, the practice of excluding images from the time series reduces the applicability of SAR-based models.


**Table 1.** Summary of the incidence angle range considered in past studies.


**Table 1.** *Cont.*

Several different incidence angle normalization procedures were carried out in previous studies. For example, [44] normalized their selected subset of imagery (incidence angles between 32◦ and 42◦) to 37◦ using a simplified correction method based on Lambert's law of optics. However, this method is insufficiently effective because it is relatively reliable only at the center of the image [41,65]. Two new effective methods for incidence angle normalization were proposed by [65], but environmental conditions limited the applicability of these methods, and they have been used mostly for ocean monitoring. Other methods for incidence angle normalization, such as simplified normalization [43], radiative transfer-based models, and statistical methods, can be applied only under specific ground conditions [66]. Therefore, despite past attempts to deal with the heterogeneity of the incidence angle in the SAR time series, the challenge of incidence angle normalization remains.

Therefore, the main goal of this study was to propose methods to reduce the backscatter dependence on the local incidence angle to permit the use of all available Sentinel-1 images in a single dataset without defining a range of allowed incidence angles and omitting images that extend beyond it. The second goal of this study was to use the proposed methods to accurately estimate vegetation properties (Kc, LAI, and crop height) based on incidence angle-normalized Sentinel-1 imagery.
