Classification of Rice Growth Phases Using the K-Nearest Neighbor Algorithm in the Irrigation area of Seulimeum Sub District, Aceh, Indonesia

Rice (Oryza sativa L.) is the main food commodity for most of Indonesia’s population. The existence of irrigation areas dramatically affects the water demand in each phase of rice growth. Due to a lack of remote sensing knowledge, classifying the rice growth phase is done manually. This study aims to classify rice growth phases using the K-Nearest Neighbor Algorithm in the irrigation area of the Seulimuem Sub-District. This research uses Google Earth satellite imagery in November 2021 through three stages: multi-resolution segmentation, guided classification, and accuracy testing. The results showed that the paddy and non-paddy fields are 1,425.56 ha and 6,581.70 ha, respectively. The classification results of rice plant growth phases consist of 4 phases, namely the inundation phase, the vegetative phase, the generative phase, and the fallow phase. Based on the accuracy assessment results in November 2021, the overall accuracy ranged from 90.65% to 92.92%, with a kappa index value of 0.85 to 0.87, categorizing it as “almost perfect.”


Introduction
Rice plants are very important to maintain food security to prevent the problem of rice shortages in Indonesia.The Krueng Aceh irrigation dam in Seulimeum Sub District, Aceh Besar, is crucial for rice farmers in supplying and allocating water to the rice fields.The existence of irrigation areas dramatically affects the water demand in each phase of rice growth.The irrigation area plays a vital role in the sustainability of rice plants as it can fulfill the plants' water needs according to their requirements, thus potentially increasing the productivity of the yielded rice.
Monitoring rice plant areas is crucial to balancing a region's food availability and demand [1].Tracking the growth and estimating rice production can assist the government in formulating strategic policies to maintain and enhance rice production in an area [2].Considering Indonesia's vast territory, comprising numerous islands and many remote islands, using satellite imagery from multiple consecutive recording dates (multi-temporal data series), monitoring the extent of rice fields and their productivity can be carried out more accurately and precisely [3].
The application of remote sensing technology has been used to study rice plants, utilizing satellite imagery through Google Earth satellite images.The reason for selecting Google Earth imagery is that it has high spatial resolution quality, extensive coverage, and is freely available.Google Earth images are 1297 (2024) 012006 IOP Publishing doi:10.1088/1755-1315/1297/1/012006 2 provided by Digital Globe with sufficiently high spatial resolution, offering advantages for monitoring purposes [4].
The K-nearest neighbor (KNN) algorithm is a classification method that employs supervised learning techniques [5].The KNN algorithm can automatically classify based on the spectral characteristics specific to different growth phases of rice fields in satellite images.Rice fields, when observed through satellite images, exhibit distinct characteristics compared to other land cover types on the Earth's surface, such as forests or settlements.It is due to the fluctuations in rice fields, resulting in changes in the greenness level during the growth of rice plants [6].A lack of knowledge about remote sensing technology leads to the manual classification of rice plants.Remote sensing can be utilized without the need for manual data analysis or feature assessment; the growth phases of rice plants can be classified automatically, providing helpful information and expediting the mapping of rice plant growth phases in required locations.This study aims to determine the classification results and the accuracy of the classification of rice plant growth phases using the K-Nearest Neighbor algorithm in the irrigation area of Seulimeum Sub District, Aceh, Indonesia.

Materials
The tools and materials utilized in this research encompass a Laptop, eCognition Essentials 1.3 software, ArcGIS 10.8 software, QGIS 3.30 software, Microsoft Office software, Google Earth Pro, Global Positioning System (GPS), writing tools, a camera, High-resolution Google Earth satellite image of Seulimeum Sub District in the year 2021, administrative boundary map of Aceh Besar District, shapefile of Aceh Besar District administrative boundary, and shapefile of Subdistrict administrative boundary.

Methods
This research was carried out from March -July 2023.Data processing was conducted at the Remote Sensing and Cartography Laboratory at Syiah Kuala University.This research uses Google Earth satellite imagery in November 2021 through three stages: multiresolution segmentation, guided classification, and accuracy testing.

Multiresolution Segmentation
This study uses high-resolution satellite images, specifically Google Earth satellite images that have undergone segmentation using the multiresolution segmentation method.Image segmentation involves applying a multiresolution segmentation algorithm based on three parameters: scale, shape, and compactness [7].

Supervised Classification
Supervised classification is a method used to transform multispectral image data into spatial classes that possess thematic information [8].The supervised classification process is carried out using eCognition Essentials 1.3.The algorithm utilized in this research is the K-Nearest Neighbor algorithm.The guided classification stage starts with creating training areas, which involves identifying selected sample areas on the segmented image objects representing each class [9].

Calculating the Classified Area
The display menu has an option "Create Report" that aids in calculating the classified area in various units.This research employs hectares as the unit based on the classified objects, displaying the area results in the software interface of eCognition Essentials 1.3 and exporting the area results to .csv and HTML files.

Accuracy Test
Accuracy testing is the final step in the classification process, and this phase holds significant importance in remote sensing data processing.Accuracy testing plays a crucial role in evaluating the reliability of

Results of Image Classification Analysis
This study employed a supervised classification method by selecting the K-Nearest Neighbor algorithm, which facilitates the automated classification process using high-resolution imagery.Supervised classification is necessary to transform multispectral satellite image data into spatial classes [12].The results of image classification analysis in the Krueng Aceh Irrigation Area within the Seulimeum Sub District yielded an area of 1,425.56ha for paddy fields.In comparison, the non-paddy fields covered an area of 6,581.70 ha.The measurement results for paddy and non-paddy fields are presented in Table 2.In the Seulimeum Sub District, the following rice field phases have been identified: The flooding phase covered 136.97 ha, the vegetative stage covered 550.43 ha, the generative phase encompassed 55.43 ha, the harvesting phase reached 314.81 ha, resulting in a total paddy field area of 1,425.56ha (Table 3).

Results of Image Classification Accuracy Assessment Using K-Nearest Neighbor Algorithm
Based on the analysis of image classification accuracy, this assessment was carried out using overall accuracy and the kappa coefficient.The data obtained from the classified image analysis using the K-Nearest Neighbor method was validated through accuracy tests.For this validation process, k=1 was used, resulting in an overall classification accuracy of 92.92% for paddy and non-paddy classes in the Seulimeum Sub District, with a kappa value of 0.85.Additionally, the accuracy of the rice growth phase classes in the Seulimeum Sub District was 90.65%, with a kappa value of 0.87.The accuracy test results fall within the "almost perfect" category.These results indicate that the K-Nearest Neighbor algorithm can identify paddy fields based on rice growth phases.The accuracy test outcomes are presented in Table 4 and Table 5.The Kappa coefficient considers the disparity between the unity of classification outcomes and the probability of random congruence compared to reference data [13].Kappa is also valuable in assessing the agreement between model predictions and existing reality.Interpretation of Kappa Coefficient Values is shown in Table 6.

Figure 1 .
Figure 1.Segmentation in the Krueng Aceh Irrigation Area, Seulimeum Sub District, with a scale of 300, shape of 0.1, and compactness of 0.5.

Table 2 .
Results of Paddy and Non-Paddy Field Areas in the Krueng Aceh Irrigation Area, Seulimeum Sub District.

Table 3 .
Results of Rice Growth Phase Areas in the Irrigation Area in Seulimeum Sub District.

Table 4 .
Overall Accuracy and Kappa Accuracy Values for Paddy and Non-Paddy Classes in the Krueng Aceh Irrigation Area, Seulimeum Sub District.

Table 5 .
Overall Accuracy and Kappa Accuracy Values for Rice Growth Phases in the Krueng Aceh Irrigation Area in the Seulimeum Sub District.

Table 6 .
[14]rpretation of Kappa Coefficient Values[14]The classification results of rice growth phases during the rainy planting season in November revealed four phases: the flooding phase, vegetative phase, generative phase, and harvesting phase, distributed across the Seulimeum Sub District.Based on the accuracy assessment results in November 2021, the overall accuracy ranged from 90.65% to 92.92%, with kappa index values ranging from 0.85 to 0.87, categorizing the results as "almost perfect".