ABSTRACT

Brazil is one of the largest producers of maize worldwide. However, this production is threatened due to low soil fertility, especially low levels of potassium (K⁺). K⁺ is one of the most important nutrients in plant metabolism, acting on enzymatic activation and also on photosynthetic processes. The identification of its deficiency by using traditional methods is difficult with regard to timely restoration of the nutrient to adequate levels. Therefore, the use of low-cost modified cameras attached to Unmanned Aerial Vehicles (UAVs) are important tools for agricultural monitoring. Nevertheless, there are no reports of studies with the purpose of evaluating the monitoring of K⁺ deficiency in maize crops using multispectral images captured from UAVs. Therefore, this study aimed at exploring the possibility of identifying K⁺ deficiency and quantifying the nutrient leaf content by using a Vegetation Index (VI). The experiment was carried out at the National Soybean Research Centre (Embrapa Soja, a branch of the Brazilian Agricultural Research Corporation). The experimental plots were constantly managed in order to obtain different conditions of K⁺ availability to plants, achieving levels that ranged from severe deficiency to an adequate nutrient level. The following treatments were established: severe potassium deficiency (SPD), moderate potassium deficiency (MPD) and adequate supply of potassium (ASP). The evaluations were performed in the Brazilian maize crop referred to as ‘safrinha, at the V7, V12 and R3 developmental stages, with image capture covering the visible and near-infrared region, using two Fujifilm IS PRO digital cameras attached to an UAV. In these development stages, leaves were collected to determine tissue K⁺ concentration. The images were radiometrically corrected with the support of calibration targets and reference values, using an Fieldspec 3 Jr. spectroradiometer. The VIs comprised the ratio among the red, green and infrared spectral bands, that is, green normalized difference vegetation index (GNDVI), normalized difference vegetation index (NDVI), ratio between infrared and green (GRVI), ratio between green and infrared (GNIR), ratio between red and infrared (RNIR) and ratio between infrared and red (RVI). Regarding all the treatments assessed, the results showed that foliar K⁺ was statistically different. The VIs were efficient only in differentiating SPD and ASP treatments at all development stages evaluated. However, none were statistically significant for MPD. The linear regressions showed a high coefficient of determination (R ²) and low root mean square error (RMSE) value; the best prediction of K⁺ concentration obtained was at V12 for regressions with these VIs: GRVI (R ² = 0.79, RMSE 4.50 g kg⁻¹) and RVI (R ² = 0.71, RMSE 4.39 g kg⁻¹). The grain yield values showed that SPD caused an average reduction of 5,645.90 kg ha⁻¹ in relation to the ASP. Considering MPD, the grain yield was 1,242.00 kg ha⁻¹ lower in comparison with ASP. In conclusion, estimating foliar K⁺ content and identifying its deficiency in maize crops based on the VIs of multispectral images from cameras attached to UAVs is possible, which ensures agility to these evaluations in a non-destructive manner, improving efficiency of K⁺ fertilization and providing farmers with a new tool.

摘要

巴西是全球最大的玉米生产国之一。但是，由于土壤肥力低，特别是钾（K ⁺）含量低，这种生产受到了威胁。ķ ⁺是植物代谢的最重要的营养素之一，作用于酶的活性，也对光合过程。关于及时将营养物恢复到适当水平，使用传统方法很难确定其缺乏。因此，使用低成本的无人飞行器（UAV）改装摄像机是农业监控的重要工具。然而，没有关于评估K ⁺监测的研究报告。利用从无人机获取的多光谱图像，玉米作物缺乏玉米。因此，本研究旨在探索通过使用植被指数（VI）识别钾^离子缺乏和定量营养叶含量的可能性。该实验是在国家大豆研究中心（巴西农业研究公司的分支机构Embrapa Soja）进行的。为了获得不同的K ⁺条件，对实验区进行了持续管理。植物的可用性，其水平从严重缺乏到足够的营养水平不等。确定了以下治疗方法：严重缺钾（SPD），中度缺钾（MPD）和充足的钾供应（ASP）。评估是在V7，V12和R3发育阶段在称为“ safrinha ”的巴西玉米作物上进行的，使用安装在无人机上的两个Fujifilm IS PRO数码相机拍摄的图像覆盖可见和近红外区域。在这些发育阶段，收集叶片以确定组织K ⁺浓度。使用Fieldspec 3 Jr.分光辐射仪在校准目标和参考值的支持下对图像进行辐射校正。VI包括红色，绿色和红外光谱带之间的比率，即绿色归一化植被指数（GNDVI），归一化植被指数（NDVI），红外与绿色之比（GRVI），绿色与红外之比（ GNIR），红色与红外的比率（RNIR）和红外与红色的比率（RVI）。对于所有评估的治疗方法，结果表明叶面K ⁺在统计上是不同的。VI仅在所评估的所有开发阶段都能有效区分SPD和ASP处理。但是，对于MPD，无统计学意义。线性回归显示较高的测定系数（R ²）和较低的均方根误差（RMSE）值；对于这些VI的回归，获得的K ⁺浓度的最佳预测是在V12：GRVI（R ²  = 0.79，RMSE 4.50 g kg ^-1）和RVI（R ²  = 0.71，RMSE 4.39 g kg ^-1）。谷物单产值表明，SPD导致平均减少5,645.90 kg ha ^-1关于ASP。考虑到MPD，与ASP相比，谷物单产低1,242.00 kg ha ^-1。综上所述，可以根据无人机上安装的摄像机的多光谱图像的VI估算玉米叶片中的K ⁺含量并确定其缺乏，这可以确保以无损方式灵活地进行这些评估，从而提高K ⁺施肥效率和为农民提供了新的工具。