Coffee refers to one of the leading agricultural products in the Brazilian commodities agenda with effective participation in the Brazilian gross domestic product (GDP), and the country is the largest coffee producer and exporter in the world (https://www.conab.gov.br/). However, the coffee culture demands high production costs, so it is necessary to implement viable techniques to optimize resources to ensure increased crop yield and profitability (Chemura et al., 2017).

Precision agriculture, which in the context of coffee growing is called Precision Coffee Growing, can be defined as the set of techniques and technologies capable of assisting the coffee grower in the management of the crop to maximize profitability and provide higher yield and final product quality (Ferraz et al., 2017). Therefore, technologies for investigation, mapping, and intelligent decision-making about the crops are sought to enhance the field activities (Balastreire, 2000).

It stands out the application of remote sensors coupled to different platforms, aiming to acquire information from the surface of crops through the response of electromagnetic energy interaction (Jensen, 2011). The remotely piloted aircraft system (RPAS) constitutes a typical sub-orbital platform application, with an on-board remote sensor, and makes it possible to obtain products of different resolutions, allows obtaining information capable of anticipating and predicting trends of plant behaviour in response to influences of various agronomic and environmental factors (Tsouros et al., 2019).

Leaf attributes of macro and micronutrients are essential for the plant to reach its stage of maturation and guarantee yield, since each mineral element plays a specific role in the plant’s metabolism, ensuring structural and physiological quality (Matiello et al., 2010). Consequently, studies on the attributes of the leaves of the plants are essential since, due to their sensitivity, they promote changes in the development of the plants due to their assimilation and use in important biological functions, affecting several important physiological processes (such as photosynthesis and cellular respiration) that therefore influence the growth and production of cultures.

Such information directly affects the crop yield, resulting from the action of the harvest, a relevant phase in the development of a crop, as results are determined and, consequently, the financial return about profitability (Pereira & Fachini, 2014). However, since the fruits act as drains on the plant, harvesting can also be considered a stress factor. Falling leaves are one of the main damages caused to the coffee plant by the fruit harvesting process (Bártholo & Guimarães, 1997), and it should be studied together with other attributes (genetics, pests and diseases and environmental ones) that can affect future yield values.

In remote sensing, the use of vegetation indices, which refer to mathematical combinations between the different bands of the electromagnetic spectrum, highlights important issues in the agricultural context, especially when using multispectral bands that detail aspects not visible to the human eye, and may bring different views and decision-making on the behavior of the crop, mainly on final yield information (Bento et al., 2022).

In the literature for orbital sensors, it is possible to find some coffee crops studies that correlate vegetation indices to yield. Jiang et al. (2008) found that a more vigorous canopy tends to present a higher potential for crop yield, which is captured by vegetation indices to estimate coffee yield. Brunsell et al. (2009) conducted a study to evaluate the feasibility of using MODIS data to monitor coffee yield. Rosa et al. (2010) used MODIS low spectral resolution sensor images and agrometeorological data to estimate yield and found a positive correlation for Normalized Difference Vegetation Index (NDVI) when capturing plant leafiness. Bernardes et al. (2012) correlated the Enhanced Vegetation Index (EVI) and NDVI vegetation indices derived from the MODIS sensor in estimating yield. Although it is possible to find this kind of study using satellite images, it was not possible to find studies using vegetation indices from RPAS images to coffee yield. In this way, aligning the study of yield and predicting its estimate by applying vegetation indices is fundamental since it optimizes the time spent on field measurements and it is also possible with the use of RPAS.

Thus, the objective of this work was to verify and evaluate the area occupied by coffee plants pre-harvest, post-harvest, and difference area pre and post- manual harvesting of the fruits; demonstrate the correlation between data on chemical attributes of leaves, yield, vegetation indices, and areas occupied by coffee plants; and estimate yield based on the variable with the best statistical indicator.

Figure 2 shows the percentage of difference CPA by the loss leaves, fruits, and branches of the harvest for each sample point considering the difference of pre-harvest CPA and post-harvest CPA and the zoom demonstration for three sample points with different scales of percentage. In this study, the difference area values varied from 7.038% for point 58 to 48.758% for point 53.

Since it is described in the literature that the coffee area (CPA) operates in the yield of the following harvests, Fig. 3 shows the distribution of the yield in the study area, with the highest yield value at point 21 of 139.204 bags ha-1 and the lowest yield value at point 47 of 48.927 bags ha-1. The distribution was not uniform in the study area. However, it stands out that the northwest area presented higher yield values and refers to the part of the land with the highest elevation of approximately 982,065 m of altitude. The southern area showed lower values of yield with an approximate height of 948,537 m of altitude.

Table 2 shows the values of descriptive statistics for yield, chemical foliar attributes, vegetation indices, and CPA (pre-harvest CPA, post-harvest CPA, and difference in CPA) for the study area’s sample points. The values of the chemical foliar attributes, vegetation indices, and CPA varied along with the sampling points, as evidenced by the coefficient of variation (CV). Regarding the vegetations indices, NDRE (Index of the Standardized Difference - Red Edge) showed a high CV value of 22.53%, and a minor variation was observed for the NDVI with a CV of 2.16%.

Table 3 shows the Pearson correlation coefficients (R) between the study variables. The yield showed a correlation with the NDVI. The chemical foliar attribute nitrogen showed no correlation with any other chemical foliar attributes or studied vegetation indices. Normalized Difference Vegetation Index Green (GNDVI) showed correlation with NDRE; Green-Red Ratio Index (GRRI) with Modified Photochemical Reflectance Index (MPRI); Normalized Different Index (NDI) with MPRI and Green-Red Ratio Index (GRRI); and no vegetation indices showed correlation with chemical foliar attributes.

Regarding the chemical foliar attributes, P showed no correlation with any indices and chemical foliar attributes; K showed a correlation with Ca, Mg and S; Ca showed a correlation with K, Mg and S; Mg showed a correlation with K, Ca and S; and S showed a correlation with K, Ca and Mg.

The values of CPA, considering pre-harvest CPA, post-harvest CPA, and difference in CPA, did not correlate with any chemical foliar attributes and vegetation indices. For the study areas only, we detected a correlation between pre-harvest CPA and post-harvest CPA.

Fig. 4 shows the scatter plot of the results. In this study, the yield correlated by vegetation indices and difference CPA by harvest generated determination coefficients (R²) as 0.501 (NDVI), 0.018 (GNDVI), 0.002 (NDRE), 0.018 (Green Optimal Soil Adjusted Vegetation Index - GOSAVI), 0.016 (MTCI), 0.076 (MPRI), 0.073 (GRRI) and 0.071 (NDI) and 0.008 (difference CPA) with only the NDVI being satisfactory for estimating this variable. This demonstrates the relationship between the set of values, a linear estimation equation, determination coefficient (R²) and correlation coefficient (R). Fig. 5 shows the NDVI distribution in the study area, since this vegetation index was the only one to present a correlation with the analyzed yield data, allowing the estimation of the prediction equation.

As shown in Fig. 2, falling leaves promote a drop-in yield in the following year by reducing the number of leaves and reducing the photosynthetically active radiation intercepted by the canopy of these plants, reflecting the biennial effects of production (Oliveira et al., 2007). The very productive plants tend to become weaker due to the drainage of nutrients from the leaves to the fruit in a year of high yield, leading to physiological disorders in the plant, which promotes leaf fall, reduced photosynthesis, reserves directed to the growth of the leaves and, therefore, low yield in the following year (Bártholo & Guimarães, 1997).

The fall in yield occurs because the high demand for nutrients, both mineral and organic, by the fruits reduces the growth of the new branching, where flowering and fruiting will take place for the next harvest. Therefore, the maintenance of the coffee plant’s leaf area at harvest is a determining factor for the subsequent harvest, and therefore attention is required for this production stage of the coffee. The leaf area is of paramount importance for the plant because of the leaf area’s more productive potential and crop yield (Manfron et al., 2003). The leaves directly contribute to the supply of photoassimilates and energy required for plant maintenance (Cunha et al., 2004).

The limits values presented by the chemical foliar attributes (Table 2) are close to the limits and critical ranges expressed in literature for healthy plant conditions (Malavolta et al., 1997; Mills & Jones Junior, 1996). The most significant variation was observed for the difference in CPA with a CV of 39.67%, demonstrating that the area values fluctuated considerably in the study area, evidenced by manual harvesting, which, due to human activity, suffers various interferences for the sampling points. For NDRE this more significant variation presented is justified by the presence of the red edge transition band and near-infrared in its equation, which consequently has greater sensitivity and faster perception of the transition from healthy to initially stressed plants (Rodriguez et al., 2006).

The CPA values did not correlate with the yield data (Table 3). Silva et al. (2010) found that manual harvesting promoted more significant falling leaves in places of higher yield, but this fact was not verified in this study since the point of most excellent falling leaves does not refer to the highest yield and the lowest falling leaves to the lowest yield, as described by Pearson’s correlation coefficients (R) (Table 3).

It is noteworthy that the harvest promotes removing fruits and, consequently, leaves and branches. In this way, the aerial images capture all the canopy’s integration, justifying the low correlation between the data to the difference CPA and the yield data.

Based on vegetation indices, only the NDVI had a significant correlation with the yield at Pearson’s correlation rate of 0.708 (Table 3). The low correlations of the other vegetation indices with the chemical foliar attributes explain the stress presented by the coffee plants. Plants under stress conditions decrease the leaves’ chlorophyll concentration, affecting light absorption in the visible, red-edge, and near-infrared region, reflected in the vegetation indices values (Marin et al., 2019). This low correlation between the other vegetation indices and yield (except for NDVI) was justified by combining spectral bands in the equations involving indices, which were not enough to describe the changes presented by yield in the field.

Although yield did not show a trend with falling leaves data, an intense depletion of coffee plants was observed in this study due to harvesting, resulting in reduced vegetation indices values and area occupied by plants compared to pre and post-harvest periods. Understanding the behaviour of falling leaves caused by fruit harvesting directly influences the yield and profitability of the crop and should therefore be the focus of research.

In conclusion, it was observed a reduction in the area occupied by coffee plants due to the manual harvesting of fruits. No significant correlations were detected between leaf chemical attributes, yield data, and the area occupied by pre and post-harvest coffee trees. Only the Normalized Difference Vegetation Index (NDVI) presented adequate correlation and determination coefficients to describe a linear equation to estimate coffee yield.