Analysis of Plant Height, Crop Cover, and Biomass of Forage Maize Grown on Reclaimed Land Using Unmanned Aerial Vehicle Technology

Abstract

Unmanned aerial vehicle (UAV) and sensor technologies are rapidly developing and being usefully utilized for spatial information-based agricultural management and smart agriculture. Until now, there have been many difficulties in obtaining production information in a timely manner for large-scale agriculture on reclaimed land. However, smart agriculture that utilizes sensors, information technology, and UAV technology and can efficiently manage a large amount of farmland with a small number of people is expected to become more common in the near future. In this study, we evaluated the productivity of forage maize grown on reclaimed land using UAV and sensor-based technologies. This study compared the plant height, vegetation cover ratio, fresh biomass, and dry biomass of maize grown on general farmland and reclaimed land in South Korea. A biomass model was constructed based on plant height, cover ratio, and volume-based biomass using UAV-based images and Farm-Map, and related estimates were obtained. The fresh biomass was estimated with a very precise model (R² =0.97, root mean square error [RMSE]=3.18 t/ha, normalized RMSE [nRMSE]=8.08%). The estimated dry biomass had a coefficient of determination of 0.86, an RMSE of 1.51 t/ha, and an nRMSE of 12.61%. The average plant height distribution for each field lot was about 0.91 m for reclaimed land and about 1.89 m for general farmland, which was analyzed to be a difference of about 48%. The average proportion of the maize fraction in each field lot was approximately 65% in reclaimed land and 94% in general farmland, showing a difference of about 29%. The average fresh biomass of each reclaimed land field lot was 10 t/ha, which was about 36% lower than that of general farmland (28.1 t/ha). The average dry biomass in each field lot was about 4.22 t/ha in reclaimed land and about 8 t/ha in general farmland, with the reclaimed land having approximately 53% of the dry biomass of the general farmland. Based on these results, UAV and sensor-based images confirmed that it is possible to accurately analyze agricultural information and crop growth conditions in a large area. It is expected that the technology and methods used in this study will be useful for implementing field-smart agriculture in large reclaimed areas.

1. Introduction

The Korean government has been encouraging the diversification of cultivated crops and land use away from the single-use cultivation of crops such as rice. To solve the problem of continuously increasing imports in the agri-food sector, policies are being prepared to enhance food self-sufficiency and strengthen export competitiveness. The Korean government is trying to meet these goals by encouraging various uses of reclaimed land. For example, they have promoted studies on soil improvement and salt removal, which are thought to have the greatest impact on field crop cultivation on reclaimed land (Seo et al., 2009). Lee (2006) suggested that five factors, namely soil, soil salinity, permeability and drainage capacity of the soil, irrigation water security, and water supply, should be considered when introducing field crops on reclaimed land. However, it has been reported that forage crops can grow if only three conditions are satisfied: appropriate salinity, permeability, and drainage capacity of the soil. In the 2019 land use plan, reclaimed land was classified into 11 production complexes. In particular, the grain production complex consists of 12,556 ha (41.1%) of developed land and 7,736 ha (45.1%) of land under construction (MAFRA, 2019)

Despite the efforts of the Korean government, reclaimed soils continue to have high salinity and high groundwater levels, and the problem of poor drainage due to fine soil particles has not been resolved (Lee et al., 2014). In addition, reclaimed land has a higher soil acidity (pH), a lower organic matter content (0.5%), and a higher Na content than general agricultural land, and it is characterized by dispersed soil particles, which results in poor soil physical properties (Hwang et al., 2012). Most reclaimed soil is poorly drained during the rainy season, and it becomes muddy due to the water inflow and excessive moisture. Conversely, if there is continuous sun, the soil becomes hard as the soil dries, making it difficult for plants to grow. Therefore, it is desirable for reclaimed land to be used as paddy fields for a certain period and then be converted for field crop cultivation after some decontamination has occurred (Shao et al., 2012).

Forage maize is favored by livestock farmers because their livestock feeds well on it, and the grains are rich in nutrients. Thus, it is necessary to further expand forage maize cultivation to provide a stable supply of high-quality forage and to increase the income of livestock farmers. However, the cultivated area of forage maize for feed in South Korea is only about 13,000 hectares as of 2019. This area is mainly located in Gyeonggi and Chungcheong, where there are many dairy and livestock farms. To produce more high-quality forage maize for feed, it is necessary to understand the pros and cons of various growing environments. However, research on the degrees of differences in the growth characteristics and yield of forage maize depending on the cultivation environment has been limited. Therefore, to solve this problem, it is essential to clearly analyze the environment of forage crops that grow on reclaimed land.

Satellite imagery methods are the most efficient for understanding the cultivation environments of large areas such as reclaimed lands. However, the images available throughout the year are very limited. In particular, few images can be used for diagnosing crops due to the influence of clouds in the summer season, when crops are most cultivated. In addition, there are many difficulties in investigating and analyzing the environment of crops on reclaimed land, as these areas are quite wide and are adjacent to the sea, and there are very few resting areas. Therefore, unmanned aerial vehicle (UAV), sensor, and spatial information convergence technology can be usefully utilized to solve these problems. UAVs are being used in various ways in agricultural research as an alternative to solving these problems (Kim and Lee, 2017). However, there are very few studies conducted on crops grown on reclaimed land. Fortunately, UAV and sensor technologies are advancing very rapidly and have been applied and utilized in similar studies (Zhao et al., 2020).Agricultural production sites have been rapidly decreasing the number of agricultural workers while expanding the scale of management. To this end, the implementation of smart agriculture techniques will allow a few workers to properly manage a large number of agricultural fields. Until now, there have been many difficulties in identifying and managing agricultural production information in a timely manner for large-scale areas of reclaimed land. UAVs have been developed to be able to observe large areas precisely at low altitudes with technological advances (Park et al., 2015), and it is possible to obtain adequate information with minimum observational flying time; thus, the application of smart management techniques using this technology on reclaimed land is promising. UAV and sensor technology are expected to play important roles in the spatial evaluation of crop growth on reclaimed land and the links between smart agriculture and precision agriculture. If image processing technology is used in conjunction with UAV and sensor technology, smart farming in the reclaimed field will be able to proceed quickly (Syeda et al., 2021).

The purpose of this study was to develop and evaluate a maize plant height (PH), cover ratio, fresh biomass(FB), and dry biomass(DB) estimation model based on field survey data for reclaimed land that is difficult to observe by a convergence of UAV and sensor-based spatial information utilization technology.

2. Materials and Methods

2.1. Study Area

This study was conducted on forage maize grown on reclaimed land in the Busa district (36°11′22″ N, 126°33′32″ E), which is located on the border between Boryeong-si and Seocheon-gun in Chungnam, in South Korea (Fig. 1). The reclaimed area of Busa is 1,234 hectares, and the developed paddy area is 825 hectares, accounting for 67% of the total area. As shown in Fig. 1, Lake Busa is located in the center of the reclaimed land, and paddy rice is mainly cultivated in the southern part of the reclaimed land due to its high salinity. In 2020, about 63 hectares of maize was grown for feed in the eco-friendly livestock and seed complex located in the northeast area of the reclaimed land (Fig. 1). The northern part of the reclaimed land was created by reclaiming the existing islands and coastal areas, and it is characterized by poor drainage and high salinity because it is difficult to supply this area with water from the Ungcheon stream.

Fig. 1. The location of the study area. The target area is the reclaimed land that is located in the Busa district on the border between Boryeong-si and Seocheon-gun in Chungnam, South Korea (indicated by red lines).

2.2. Experimental Design

The purpose of this study was to identify the growth characteristics of forage maize using remote sensing (RS) technology during a minimum field survey. As shown in Fig. 2, the maize growth survey was conducted on reclaimed land (R1) and general farmland (G1–G4) by selecting one field lot for each land type. Maize was planted at an interval of 75 cm and a row distance of 17 cm at a plant density of 70,000 per ha.

Fig. 2. Reclaimed land (R1) and general farmland (G1–G4) cultivation areas were studied to compare maize growth characteristics according to the cultivation environment.

As shown in Fig. 3, for the sampling field lot, one field lot was selected from reclaimed land (R1) and general farmland (G1), and a growth survey was performed. For the growth survey, a total of 16 points were selected in a sampling area of 1 × 1m, eight points for each sample site (Points A1 to A16 on the left of Fig. 3). Sample point selection criteria were arbitrarily selected so that the maize height was evenly distributed from the low to the high point to express the diversity of maize height when judged in the field. The growth survey was carried out in the order of PH measurement, sampling, coordinate measurement, FW measurement, and DW measurement for each sample point. For PH measurement, the height from the ground to the tassel was measured using a leveling staff (BAR-STF; SOKKIA, South Korea) for 5 to 6 maize included in the sampling point(Fig. 3a).After the PH measurement, the subject was sampled to measure the FW. The sampled points were coordinated using the Real-Time Kinematic Global Positioning System (V30; HI-TARGET, China) to match the UAV image (Fig. 3b). After measurement, the maize was sampled and classified for each point to measure the FW (Fig. 3c). The maize DW was obtained by measuring the weight after drying for 72 hours at 70°C using a laboratory drying oven (TG-100-ADCT; Nippon Medical, Japan) (Fig. 3d).

Fig. 3. Sampling points for reclaimed land (R1) and general farmland (G1) forage maize growing areas. (a) Method of measuring the maize plant height. (b) Examining the location of the measurement point using GPS-RTK. (c) A sample taken to measure fresh weight. (d) Measuring dry weight of the maize sample in the laboratory.

2.3. UAV Image Data Acquisition

After sowing, forage maize takes about 90 to 120 days until harvest, although the growing period varies depending on the growing region and method. Most of the forage maize grown in Boryeong and Seocheon is sown in early June and harvested in mid-September. The planting date of the two plots for which the growth survey was conducted was June 5, 2020, and the harvest was carried out on September 17, 2020.

The UAV images were acquired using a fixed-wing eBee Plus (Sensefly, Cheseaux-sur-Lausanne, Switzerland). The UAV-mounted sensor used an RGB S.O.D.A.(Sensefly, Cheseaux-sur-Lausanne, Switzerland) camera (Fig. 4).Table 1 shows the specifications of the UAV and RGB S.O.D.A. used for image acquisition.

Table 1. Specification of equipment for image acquisition

Fig. 4. From image acquisition to digital surface model(DSM) and RGB orthographic image creation process.

Data acquisition of UAV and field was carried out 13–14 September 2020 during the representative four (R4) stage of the maize growth stage and 6–7 October 2020 imagery for topographic map production after the maize had been harvested. The field survey was conducted on the same days as the UAV imagery acquisition before harvest.

2.4. Farm-Map

Farm-Map is an electronic map that displays the actual usage data for farmland, which is acquired and processed using satellite and aerial images of South Korean farmland. Farm-Map divides farmland usage into seven categories (paddy fields, fields, orchards, facilities, ginseng, bare soil, and fallow land). Farm-Map does not require a lot of effort and time for image processing (Lee et al., 2021). In this study, Farm-Map was used to obtain information on the classification of reclaimed agricultural land, specifically to determine the maize cultivation area. It was used in conjunction with UAV images obtained from the field.

2.5. Biomass Measurement

In situ biomass was determined in two ways: FB and DB. The FB of maize grown on farmland was obtained as shown in Eq. (1) by measuring the FW after sampling the maize.

\(\begin{aligned}F B=\frac{F W}{A}\end{aligned}\)       (1)

where FW is fresh weight (g), A is the unit area of the sampling section (m²), and FB is fresh biomass(g/m²). The resulting FB measurements were converted to tons per hectare.

Sampled maize was brought to the laboratory and used for DW measurement using a dry oven. The drying temperature was set to 70°C. The samples were dried for 72 hours and then their weight was measured. DB was calculated using Eq. (2) with the measured DW.

\(\begin{aligned}D B=\frac{D W}{A}\end{aligned}\)       (2)

where DW is dry weight (g) and DB is dry biomass. The resulting DB measurements were converted to tons per hectare.

2.6. Biomass Model

2.6.1. Estimation of Plant Height

The relative height of each pixel was found during the mosaicing process using the structure from motion (SfM) algorithm. The SfM algorithm estimates the relative height of each feature by calculating the displacement based on the location of the same feature in each image.

Digital surface models(DSMs) were used to estimate the maize height by analyzing the changes over time that were determined using the SfM algorithm. The differences in the DSM values between the time of maize cultivation and the time after maize cultivation (Fig. 5) were used to obtain a crop surface model (CSM), as shown in Eq. (3).

Fig. 5. Crop surface model (CSM) extraction process. (a) UAV image acquired during the maize cultivation period. (b) UAV image after maize harvest. (c) CSM extraction using digital surface models (DSMs) before and after harvest.

CSM = DSM_{growth stage} – DSM_harvest       (3)

where CSM is the crop surface model, DSM_{growth stage} is a digital surface model of each maize growth stage, and DSM_harvest is a digital surface model after harvest.

In order to match the areas of the sampling points, the CSM resampled them using the nearest neighbor technique and regenerated them as an image with a spatial resolution of 1 m. The sixteen points were measured using GPS-RTK and used to extract the CSM value of the pixel corresponding to the location. The extracted CSM value and the PH accuracy were evaluated, and error correction was performed using regression analysis.

PH = a × CSM + b       (4)

where PH is plant height, and a and b are regression coefficients.

The generated regression equation was applied to the CSM image and used to obtain the estimated PH.

2.6.2. Calculation of Maize Coverage

To calculate the maize coverage, the area of the maize-growing area is required. Area calculation was performed using Otsu’s method (Otsu, 1979) to separate maize from surrounding backgrounds such as soil and weeds. In this study, this method created a pixel histogram with a bimodal shape consisting of maize pixels and background pixels. Based on the calculated threshold, the image pixels were classified into two groups as shown in Eq. (5) and (6).

σ_w² = ασ(t)₁² + βσ(t)₂²       (5)

σ_i² = αβ(μ₁(t) – μ₂(t))²       (6)

where σ_w² is within-class variance, which is a weighted sum of the variances of the two classes; α and β are weighted, which are the probabilities of the two classes separated by the threshold (t); σ₁² and σ₂² are the variances of the two classes; σ_i² is inter-class variance; and μ_i is the mean of class i

In this study, the maize vegetation area was estimated by automatically calculating the threshold value between the maize and non-maize pixels using the histogram of the estimated PH image and Otsu’s method (Fig. 6).

Fig. 6. The process of classifying maize cultivation areas by applying Otsu’s method to the images acquired at the research site. CSM: crop surface model, RGB: red green blue.

Farm-Map data was used to calculate the area of each field lot, The maize fraction was calculated as in Eq. (7) using the calculated maize coverage area and the corresponding field lot area.

\(\begin{aligned}M F=\frac{\text { Maize coverage area }}{\text { Area of field lot }}\end{aligned}\)       (7)

where MF is the maize fraction, Area of field lot is the area of each field lot based on Farm-Map, and Maize coverage area is the maize coverage area calculated by Otsu’s method.

2.6.3. Volume-based Biomass Model

The biomass estimation model used a simple regression model comparing volume and biomass. The volume was calculated using the unit area of points A1 to A16 in Fig. 3 and the estimated PH of the maize cultivation area. The biomass estimation model is shown in Eq. (8).

Biomass = a × Volume + b       (8)

where a and b are the regression coefficients.

The biomass estimation model was applied to obtain the FB and DB, and it was applied in the same way for both reclaimed and general land.

2.7. Accuracy Analysis

To evaluate the accuracy of the biomass model, the root mean square error (RMSE) was used. The RMSE is useful for comparing prediction methods on the same data set, but it has the disadvantage of being scale-dependent. Each variable used in this study had different scales (i.e., units and measurement ranges). Therefore, the normalized RMSE (nRMSE) was used to ensure the accuracy of the PH and biomass models and allow for mutual comparison. The RMSE and nRMSE were calculated using Eq. (9) and (10).

\(\begin{aligned}R M S E=\sqrt{\sum_{i=1}^{n} \frac{\left(\widehat{y}_{i}-y_{i}\right)^{2}}{n}}\end{aligned}\)       (9)

\(\begin{aligned}n R M S E=\frac{\sqrt{\sum_{i=1}^{n} \frac{\left(\widehat{y}_{i}-y_{i}\right)^{2}}{n}}}{y_{\max }-y_{\min }}\end{aligned}\)       (10)

where \(\begin{aligned}\widehat{y}_{i}\end{aligned}\) is the predicted value, y_i is an observed value, n is the number of observations, y_max is the maximum of the observations, and y_min is the minimum of the observations.

2.8. Data Processing Workflow and Method

Fig. 7 shows the workflow and process of the study. The research process mainly consisted of data correction and image acquisition, image pre-processing, image processing, and maize cultivation area estimation to calculate the PH, maize cover rate, and maize biomass estimation.

Fig. 7. Schematic procedure of the unmanned aerial vehicle (UAV) image acquisition, Farm-Map, field survey, pre-processing, plant height (PH), maize cover rate, and biomass estimations.

All image processing and visualization in this study were performed using ArcGIS Pro software (Esri, Redlands, CA, USA) version 2.5.

3. Results

3.1. UAV-based PH Estimation

UAV-based images were used to estimate the PH based on the measurements taken at the 16 points on reclaimed land (R1) and general farmland (G1). Table 2 shows the measured PH obtained from R1 and G1.

Table 2. Statistics on the plant height (PH) measured on reclaimed land (R1) and general farmland (G1)

The obtained results showed a lower distribution of all statistics for R1 compared to those for G1. The PH of reclaimed maize was lower than that of general farmland.

The estimated PH was obtained by linear regression analysis using the CSM obtained from the UAV-based images. The regression analysis resulted in a P-value lower than 0.01 and a coefficient of determination of 0.97 (Fig. 8 and Table 3).

Table 3. Statistics on plant height (PH) estimated using reclaimed land (R1) and general farmland (G1) data

Fig. 8. The result of linear regression analysis of the plant height (PH) estimated using the crop surface model (CSM) obtained from UAV-based images and the measured values. (a) Results of linear regression analysis for the estimated PH from UAV-based images and measured values. (b) Calibrated scatterplots for estimated and measured PH values.

Although the CSM reflected the distribution of the measured values well, comparison using a 1:1 line revealed that it estimated the values to be slightly lower than the measured values. Fig. 8b shows the result of obtaining the PH by applying the CSM regression equation with data correction. The RMSE of the PH estimated by regression analysis was 0.086 m, and the nRMSE was 4.92%, suggesting that the model reflected the actual maize height very well (Table 3). Observational and estimation errors were corrected using the relational expression obtained from the regression equation.

Fig. 9 displays how the maize PH can be estimated by applying the CSM extracted from a UAV-based image of maize fields grown on reclaimed land and general farmland. PH of maize was classified into a total of five classes and was visualized using the Jenks natural breaks method to minimize the variance within each class(Jenks, 1967). On the general farmland, most PH values were uniformly above 1.23m, whereas those on reclaimed land had a very non-uniform distribution.

Fig. 9. Distribution map of the estimated plant height (PH) of maize grown in fields on reclaimed land (left) and general farmland (G1–G4).

3.2. Maize Coverage

The estimated PH distribution image map (Fig. 9) was used to extract the threshold value by applying Otsu’s method to the maize fields. The histogram generated through this method (Fig. 10) revealed the threshold to be 0.42 m. Pixels showing a PH of 0.42 m or more were defined as maize, and pixels with a PH less than 0.42 m were defined as background objects, such as soil and weeds. Only the pixels corresponding to maize were extracted. This method allowed for accurate identification of the maize cover area in each field of the reclaimed land area (Fig. 11).

Fig. 10. Threshold values extracted by applying Otsu’s method to produce a histogram of the pixels representing maize cultivation on reclaimed land and general farmland. PH: plant height.

Fig. 11. Coverage map of maize obtained by applying the plant height (PH) threshold value to reclaimed land. Maize cover on the left bank (A) and right bank (B) near the reclaimed river.

The maize cover area corresponding to each field lot was calculated based on the division of the Farm-Map and Eq. (8). The calculated maize coverage for each field lot is shown in Fig. 12. The maize coverage by field lot varied from 12.15% to 96.76% in reclaimed land and from 86.13% to 97.59% in the general farmland, showing similar distributions across the two land types.

Fig. 12. Maize coverage for each field lot of reclaimed land (left) and general farmland (G1–G4).

3.3. Volume-based Biomass Evaluation

3.3.1. Volume-based FB Calculation

FB was obtained by measuring the FW after sampling maize from reclaimed land and general farmland. Table 4 shows the statistics on the FB of maize grown on each land type. Statistical values for FB were found to be very low for the reclaimed land compared to those for the general farmland, and they had a wide range of distribution.

Table 4. Statistics on the fresh biomass (FB) measured on reclaimed land (R1) and general farmland (G1)

The volume of each farmland type was calculated using a resampled PH image with a spatial resolution of 1 m. Linear regression analysis was performed using the calculated volume and the FB of the corresponding position. As shown in Fig.13a, the slopes of the reclaimed land and general farmland showed different trends. The models for the reclaimed land and general farmland had coefficients of determination of 0.93. The results obtained from the models showed that the maize growth conditions of the reclaimed land poorly reflected the results of the general farmland (Fig. 13b).

Fig. 13. Comparison of correlation between measured fresh biomass (FB) and volume and estimated FB. (a) Results of the linear regression analysis comparing the volume of the farmland and the field-measured FB according to the land conditions (reclaimed land and general farmland). (b) Scatter plot of the estimated and measured FB.

The obtained regression equation was applied to the areas corresponding to the reclaimed land and general farmland in the UAV image, and the results were compared with the measured FB. The coefficient of determination was 0.94, the RMSE was 3.18 t/ha, and the nRMSE was 8.08%,suggesting that the model was very precise. As a result of the developed model, the FB distribution map was obtained by masking the maize cultivation area of reclaimed land and general farmland. As shown in Fig. 14, the FB distribution reflected the actual cultivation situation, as maize grown on reclaimed land was distributed at lower values and the maximum value was lower than that of maize grown on general farmland.

Fig. 14. Distribution map of fresh biomass (FB) in the maize cultivation sector of reclaimed land (left) and general farmland (right, G1–G4).

3.3.2. Volume-based DB Calculation

Table 5 shows the statistical results for the DB of maize grown on reclaimed land and general farmland. DB exhibited a slightly different trend from that of FB, as maize grown on reclaimed land showed a much lower and more varied distribution than did that grown on general farmland. The overall DB of maize grown on reclaimed land was found to be 3.33 t/ha lower than that of maize grown on general farmland.

Table 5. Statistics on the dried biomass (DB) measured on reclaimed land (R1) and general farmland (G1)

Like FB, DB exhibited different trends between the general farmland and reclaimed land models, but the obtained coefficients of determination were lower than those in the FB models. As shown in Fig. 15, the coefficient of determination was 0.84 for reclaimed land and 0.90 for general farmland. The obtained regression equation was applied to the reclaimed land and general farmland, and the results were compared with the measured DB. The coefficient of determination was 0.86, the RMSE was 1.51 t/ha, and the nRMSE was 12.61%.

Fig. 15. Comparison of correlation between measured dry biomass (DB) and volume and estimated DB. (a) Results of the linear regression analysis comparing the volume of the land and the field-measured DB according to the land conditions. (b) Scatter plot of the estimated and measured DB.

The developed model was used to obtain the DB distribution map by masking the maize cultivation sector of reclaimed land and general farmland (Fig. 16). The results showed that the DB had a higher response in the low region than did FB, whereas a large difference was observed at high DB values. General farmland fields with low DB were headland and boundary points, which were not related to maize. Although there was a difference in the distribution of DB according to the location of the fields in the general farmland, it had a uniform distribution in all fields.

Fig. 16. Distribution map of dry biomass (DB) in the maize cultivation sector of reclaimed land (left) and general farmland (right, G1–G4).

3.4. Maize Biomass Comparison

3.4.1. Pixel-based Comparative Analysis

The biomass characteristics of maize grown on reclaimed land and general farmland were compared and analyzed using the differences in the pixel distributions of the image maps for PH, FB, and DB. In the PH image map, when pixels representing a PH less than 0.42 m were removed according to Otsu’s method, the remaining number of pixels was 421,450.

Among them, the number of pixels in general farmland was 20,138 and the number in reclaimed land was 401,312. Since it is difficult to directly compare the PH between reclaimed land and general farmland due to the difference in the number of pixels, the PH and biomass images were graded, and the pixel ratio corresponding to the grade was calculated and analyzed for comparison.

Fig. 17 shows the pixel ratio according to the range of each grade. The grade was determined in 4 levels except for the level less than 0.42 m which was eliminated by Otsu’s method for PH determined in 5 steps shown in Fig. 9. On the general farmland, 2.5% of the PH values were in the range from 0.42 to 0.8 m, and 20.5% of the PH values were in this range on reclaimed land. The reclaimed land had 8.2 times more PH values in this range compared to those on the general farmland, which suggests very poor growth on the reclaimed land.

Fig. 17. Comparison of land types using the pixel ratios and the range of each of the four grades of plant height.

FB and DB images were also divided into five categories according to growth conditions using the Jenks natural breaks method: very poor, poor, normal, good, and very good. The distributions of FB (Fig. 18) and DB (Fig. 19) in general farmland and reclaimed land showed very similar trends.

Fig. 18. Comparison of fresh biomass (FB) in the two land areas based on five categories of growth conditions.

Fig. 19. Comparison of dry biomass (DB) in the two land areas based on five categories of growth conditions.

In the distributions for reclaimed land, the FB and DB values showed normal distributions centered on the “poor” category, with “normal” and “very poor” being the next highest values. On the other hand, the distribution of FB and DB in general farmland had a very low distribution in the “very poor” and “poor” categories, with a steep increase toward the “very good” category.

3.4.2. Field-based Comparative Analysis

For comparative analysis of the field lots, the average values of PH, MF, FB, and DB for each field lot were calculated based on their divisions in Farm-Map. The PH range was 0.165–1.66 m in reclaimed land field lots and 1.64–2.36 m in general farmland field lots. The average PH for each field lot was about 0.91 m for reclaimed land and about 1.89 m for general farmland, indicating about 48% of all PHs in each land type (Fig. 20).

Fig. 20. Comparative analysis by field lot using the average values of plant height (PH), maize fraction (MF), fresh biomass (FB), and dry biomass (DB)for each field lot.

MF ranged from 12.15% to 96.76% in reclaimed land, whereas it had a very narrow range from 86.13% to 98.58% in general farmland. The average MF of the field lots on reclaimed land was about 65% and that on general farmland was about 94%, resulting in a difference of about 29%. Reclaimed land FB was low, ranging from 1.46 to 20.79 t/ha, whereas FB in general farmland was twice as high, ranging from 20.13 to 44.07 t/ha. The average FB of the reclaimed land field lots was 10 t/ha, while that of general farmland was 27 t/ha, resulting in a difference of about 36%. DB ranged from 0.62 to 8.56 t/ha in reclaimed land and from 5.83 to 12.44 t/ha in general farmland. The average DB of the field lots on reclaimed land was about 4 t/ha and that of general farmland was 8 t/ha, indicating that the reclaimed land represented about 53% of the DB on average. As such, the average values of PH, MF, FB, and DB in each field lot were all higher for general farmland than they were for reclaimed land.

4. Discussion

In previous studies conducted in Korea on the dry matter yield of maize, general farmland yielded 11.05–15.48 t/ha (Ji et al., 2009; Chung et al., 2019) and reclaimed land yielded 7.45–11.29 t/ha (Jung et al., 2011; Shin et al., 2012). The maize yield on reclaimed land obtained in the previous study was suggested to be about 70% of that of general farmland, resulting in a difference of about 20% from the results of the present study. In 2020, the initial temperature during cultivation was very low compared to that of other years, and there was about 200 mm of rainfall compared to the previous year (KMA, 2020). This may have caused maize growth to be sluggish due to the cold and productivity to be very low due to poor drainage. In the case of reclaimed land, there were many areas with poor drainage, whereas this was not the case for general farmland. This likely led to a difference in production between the two land types.

For a more precise analysis, additional research is needed on other potential cases of these trends, and the soil and climatic conditions of the reclaimed land must be further evaluated.

5. Conclusions

In order to combine UAV remote sensing technology with smart agriculture, it is necessary that each technological element, including the UAV airframe, sensing technology, observation and operation methods, data processing methods, algorithms, and information-providing technology, is up to a high standard. This is particularly important when considering the cost and effectiveness of agriculture on reclaimed land compared to other land types. In order to address these issues, this study applied UAV and sensor technology to the maize cultivation areas of the Busa reclaimed land, constructed models for plant height, the cover ratio, fresh biomass, and dry biomass to evaluate how the reclaimed land differed from general farmland and derived the related spatial distributions. The resulting estimations were confirmed to be highly consistent with the field survey results. Therefore, knowledge of the productivity of reclaimed land can provide critical information in situations where there is little scientific data.

The results of this study differed from those of previous studies because different areas of reclaimed land were examined. It is likely that differences in salinity and drainage conditions between the different areas had the greatest influence on the contrasting results. Furthermore, it is likely that differences in the weather conditions during the survey year also had a significant effect. It is expected that food self-sufficiency and security problems will intensify worldwide due to the effects of climate change and other global shifts. To address these problems, energy-saving and carbon-reducing technologies must be implemented worldwide, particularly about the movement of grains. Development and dissemination of this technology are especially important in countries where reclaimed land has been developed for use. Korea is a key example of this, as the country has been using reclaimed land for various purposes to provide a stable supply of agricultural products and stabilize consumer prices. It is, therefore, necessary to promote complex research related to the soil and water management of reclaimed lands, particularly related to the development of information and communications technology.

In conclusion, the advancement and systematization of UAV and sensor-based technology will be critical in the future applicability of this technology in agriculture on various land types. If a workflow applicable to other agricultural environments is prepared and applied, it will be possible to quickly realize smart agriculture in the field.