1. Introduction
Crop nutritional stress is usually assessed considering N deficiency on the base of plant analysis methods with no direct measurements or in consideration of soil properties information. If not considered, soil variability can prevent a correct diagnosis of the nutritional status of crops. The influence of soil properties variations in relation to crop nitrogen status assessment performed with proximal or remote sensing technique is a topic that has not been much studied in literature. As reported by [
1], several studies have demonstrated separately the potential advantages of crop spatial variability analysis with soil-based and plant-based methods in order to drive variable N fertilization, but only few studies assessed the interaction between the two, hence the possibility to combine this information in a decision process [
1,
2].
Cereals cultivation requires sustainability in terms of production and environmental impacts. Nitrogen is one of the most important nutrients for plants. Soil N availability varies spatially and temporally within a field due plant, soil–atmosphere interaction almost guided by the variations of soil properties and land morphology. Crop N demand changes during the season depending on growing conditions related to the presence of limiting factors. This requirement is met by soil N supply capacity and N fertilizations. N mineral fertilizers are globally the most used substances for fertilization, with a world supply of ammonia (NH
3) of ~180 million tons [
3]. Unfortunately, the large quantities used do not correspond to high levels of nitrogen use efficiency (NUE) [
4,
5,
6,
7]. Sustainable agriculture requires optimizing N management during the crop season, in order to achieve profitability and a healthy environment [
8,
9,
10]. N supply from soil and fertilizations and N accumulation in crops are very dynamic processes that should be considered through an integrated approach [
11,
12] and a spatial relationship analysis [
13]. The uncertainty in both plant N demand and soil N supply due to seasonality and growth potential often leads to non-optimal nitrogen management strategies [
14].
One of the most widely used crop N status assessment methods relies on the estimation of the nitrogen nutrition index (NNI), defined as the ratio of the actual N concentration in the dry above-ground biomass (Na%) to a critical N concentration (Nc%), which is in relation to the specific above-ground biomass (W) expressed in t ha
−1 [
14]. NNI can also be derived considering crop N content (kg N ha
−1), instead of concentration (%). In this case, NNI is computed using the ratio between actual plant nitrogen uptake (PNUa) and critical plant nitrogen uptake (PNUc), both expressed in kg N ha
−1 [
15,
16]. This is because the amount of N in the plant, expressed as a concentration or quantity, and biomass accumulation, have a relationship based on the theory of N dilution during plant growth [
17]. The critical nitrogen dilution curve derived from this theory [
18] determines the minimum N concentration (Nc) or N content (PNUc) [
16] in the crop, for a specific moment, which allows for a normal growth with no biomass accumulation reduction. Other authors identified N dilution curves as a function of development stages [
19,
20] or Leaf Area Index (LAI) [
16,
21,
22,
23] instead of using W. The use of LAI to obtain NNI estimation is indeed a promising approach, because it can be performed with non-destructive measures of W using in field or remote indirect methods [
24,
25]. Besides the field estimation of NNI, this indicator can also be calculated by empirical relations with remote sensing (RS) vegetation indices (VIs) or biophysical variables (BV) using proximal or satellite data. RS data can be used to derive both Na% and Nc% (via their relationships with W) or PNUa and PNUc. Many authors suggest that the latter option is preferable, because the relationship between RS data and total quantities of nitrogen present in the canopy is more robust [
14,
26].
From the literature, two approaches have been used to estimate NNI from satellite data: (1) direct methods (DM) and (2) indirect methods (IM). DM rely on the possibility to estimate NNI by “direct” relationships with Vis. According to [
27], we can use a (i) “mechanistic approach” (DM_1) that requires first to estimate Na% and W to calculate NNI or (ii) a “semi–empirical approach” (DM_2) based on the direct definition of a parametric regression between VIs and NNI values that are usually estimated with in situ data. Many authors have successfully used DM of NNI from remote sensing using multispectral [
23,
28,
29,
30,
31] or hyperspectral data [
32,
33] for optimizing the timing and the rate of N fertilizer applications. The study of [
34] tested a DM_1 approach in an operational workflow for producing NNI maps based on the combined use of high-resolution satellite images and ground-based estimates of crop parameter using smart apps.
IM are based on a different paradigm: RS data are firstly used to retrieve biophysical variables from RS, such as leaf area index (LAI) and canopy chlorophyll content (CCC). Then, from these estimates, a relation with PNUa and W is adopted to derive PNUc according to a crop-specific dilution curve. Once PNUa and PNUc are estimated, NNI is calculated. IM approaches have the advantage to exploit RS data to estimate optically related biophysical variables (BV), hence to be a physically-based solution. BV estimation can be performed using (i) parametric regression methods using Vis [
15,
35,
36,
37], (ii) non-parametric machine learning algorithm (e.g., neural network, SVR, PLSR etc.) or by inverting radiative transfer models (RTM) adopting look up table approaches [
38] or hybrid methods [
39,
40,
41]. For a detailed review of the different available approaches, see [
42]. Recently, the European Space Agency (ESA) Sentinel Application Platform (SNAP) toolbox, reached an operational stage in providing biophysical variables (LAI, fAPAR. fcover, CCC) through a biophysical processor tool based on neural networks [
43]. This greatly facilitates the operational adoption of RS products in agricultural management decision support systems.
Under this framework, the objectives of this study were to investigate the robustness and representativeness of nutritional status assessment through the computation of NNI from Sentinel-2 for maize and durum wheat in relation to different nitrogen fertilization levels and soil properties. A field experiment for two consecutive crop seasons was set up to interpret the potential and limits of NNI in identify nitrogen deficiency.
More specifically, Sentinel-2 acquisitions were exploited to analyze spatial and temporal variability of crop status from remote sensing with the following objectives: (i) compare the performances of direct and indirect NNI estimation approaches; (ii) assess the added value of the operational S2-BV product with respect to ad hoc calculated VIs.
4. Discussion
It is recognized that seasonal N status assessment can be performed using the NNI indicator [
68], however, not much investigation is present in literature about the interaction between detected crop N deficiency and actual source of crop grow limiting factors. NNI can be used to highlight which zones of a field present crop anomalous conditions: NNI > 1 “luxury consumption” and NNI < 1 “nitrogen deficiency”. Usually this information is used to tactically modulate nitrogen fertilization with VRT machinery, providing the appropriate amount of fertilizer. However, in some cases, plant N deficiency is not related to nitrogen availability, but to other factor that must be identified and taken into account, to support the most appropriate fertilization strategy. The present study highlighted that NNI is an appropriate tool for the diagnosis of the nutritional status of the crops, however, soil properties can have an impact on crop N uptake, showing an effect independent from N rates supply. The capacity of NNI to assess crop nutritional status in relation to soil properties was confirmed for two consecutive crop seasons. In our case, measured NNI showed that the experimental plots located in an area of the field with acidic soil (FOR1) had lower NNI values than all the other plots in the other soils’ types both for maize and wheat. High N rates did not increase Na% and consequently crops biomass in the problematic soil area of the field (FOR1), as expected and as demonstrated by other studies [
16,
18,
29]. In addition to this, the high rate plots in the FOR1 class reached only half of the biomass accumulation (4 t ha
−1) as compared to the best performance of JOL1 and CDS2 soil classes (8 t ha
−1) and the two classes are statistically different considering this variable. It is well known that most plant nutrients are optimally available to plants within the soil pH range between 6.5 and 7.5 and soil classes of the field experiment belonged to this range except FOR 1. Nutrients absorption depends on elements availability in the root zone and the growth rate potential, even in the N excessive condition, is regulated by the soil N supply [
30]. N fertilization can temporarily reduce soil pH in the surface soil layer after the supply, especially at high amounts as demonstrated by [
69]. In presence of acidic soils, this effect may be even higher, causing even greater problems in nutrients absorption, with a combination of yield limiting factors. The study of [
11] underlined that nitrogen assimilation needs to be considered in the context of inter-regulation of multiple crop physiological processes (i.e., C N assimilation and allocation) and soil N availability.
Besides this consideration, overall, the estimated NNI values during the season have revealed areas with a general sub-optimal crop growth in the experimental field, which resulted in an average yield (14.18 t ha
−1) much lower than the average of all the maize fields of the farm (18.22 t ha
−1), confirming the goodness of NNI to indicate crop yield potential [
14]. NNI calculated from remote sensing during the season resulted in a dynamic diagnostic tool to assess spatial explicit crop N status and limitation in N uptake [
14]. NNI calculated from remote sensing during the season resulted in a dynamic diagnostic tool to assess spatial explicit crop N status and limitation in N uptake [
14]. Chlorophyll content estimation for crop stress status monitoring has been considered in many studies and for site-specific N management [
66,
70], but the relationship between chlorophyll content and Na% remains highly empirical [
26,
71] because it is usually influenced by variety, phenology, and study area, limiting the generalization capability of the method using only chlorophyll content information [
39,
50]. RS DM are proposed as a rapid and robust way for nutritional status assessment to be used as an alternative to direct field measurements [
29,
31]. Information on actual nitrogen content or concentration, or direct indication of nitrogen deficiency (NNI) from remote data are of great help in crop monitoring.
From the 2018 maize experiment, we found that the direct estimation of NNI using VIs or S2_BVs derived by Sentinel-2 is feasible as demonstrated by general good correlation metrics comparable to other studies. The best performance was obtained using Cab_S2 secondary variable (R
2 = 0.82) slightly better than using CCC_S2 (R
2 = 0.77) or NDRE (R
2 = 0.79), this performance is similar to what was reported by [
65]. Our results indicated also that Na% can be directly derived using MCARI with an agreement of R
2 = 0.71, confirming the conclusions of [
32]. However, biomass estimation directly from VIs resulted not reliable. So, the direct “mechanistic approach” (DM_2) was not performing as expected. These results confirmed the possibility to directly derive NNI (DM_1) with VIs and/or S2_BVs as a suitable solution for operational crop monitoring to support site-specific nitrogen fertilization. The NNI derived maps showed comparable results in terms of correlation with DM (R
2 = 0.62, RMSE = 0.21) and IM (R
2 = 0.63, RMSE = 0.27) approaches.
Indirect method (IM) derive NNI from the separate estimations of W and PNUa. The advantage of using such approach is that estimated PNUa can be used to compute ∆PNU, this is a promising quantitative approach to define nitrogen absorption deficit of a crop [
14,
15,
26]. PNUA (kg ha
−1) is a quantitative variable preferred over a variable expressed in percentage as Na because it avoids an intermediate step to convert N contents from a mass basis to an area basis to a mass basis [
72] as reported by [
14]. Experimental results demonstrated that the biophysical variable CCC estimated from S2 is well related with PNUa (R
2 = 0.78), confirming the results of previous studies [
31,
39]. S2 estimated GAI can be used to estimate W (R
2 = 0.84), since this variable is necessary to calculate PNUc according to crop specific dilution curves. The quantitative deficit or surplus values given by the indicator ∆PNU is thus suitable for adjusting N supply in site-specific fertilization management.