Pendekatan Pertanian Cerdas dan Presisi untuk Pemilihan Mesin Pengolahan Tanah Berdasarkan Variabilitas Spasial

Abstract

Penerapan prinsip-prinsip pertanian presisi pada pertanian skala besar seperti perkebunan membutuhkan dukungan teknologi dan manajemen operasi mesin untuk menangani kegiatan pertanian dan mengelola variabilitas spasial lahan. Mesin pengolah tanah merupakan komponen input utama dan berperan signifikan terhadap produktivitas dan keberlangsungan kegiatan pertanian serta kelestarian lingkungan. Efektivitas dan efisiensi mesin pengolahan tanah disesuaikan dengan kebutuhan aktual yang dapat ditentukan dari luas, variabilitas spasial fisik dan mekanik tanah serta faktor lingkungan. Kadar air tanah diketahui berperan penting terhadap dinamika fisik dan mekanik tanah. Namun, variabilitas spasial yang tinggi menyebabkan kadar air tanah pada lahan pertanian skala besar sulit terwakili oleh pengambilan beberapa sampel tanah. Sedangkan pengambilan sample tanah dalam jumlah besar cenderung tidak efisien, baik ditinjau dari waktu maupun biaya yang dibutuhkan. Adopsi teknologi sistem informasi geografis, penginderaan jauh dan sistem navigasi salah satu alternatif terbaik untuk menangkap variabilitas spasial lahan pertanian. Integrasi teknologi tersebut dengan manajemen pengetahuan kondisi lahan dan faktor lingkungan dapat mendasari pengambilan keputusan dalam perencanaan dan pengolahan lahan pertanian intensif. Penelitian ini bertujuan merancang metode pemilihan mesin pengolahan tanah berdasarkan variabilitas spasial lahan perkebunan tebu yang belum ditanami menggunakan pendekatan pertanian cerdas dan presisi. Perancangan tersebut memerlukan tahapan-tahapan studi, meliputi: (1) pengembangan model perhitungan kadar air tanah dan perhitungan tingkat kemampuan kerja tanah (soil workability), secara cepat dan mendekati waktu nyata berdasarkan data radar Sentinel-1A; (2) pengembangan model perhitungan tahanan penetrasi berdasar informasi dekat waktu nyata kadar air tanah (yang diperoleh dari tahap 1) dan fraksi tanah; (3) perhitungan dan pemetaan distribusi spasial draft spesifik pengolahan tanah berdasar tahanan penetrasi dan indeks plastisititas; dan (4) pengembangan metode seleksi mesin pengolah tanah berdasarkan distribusi spasial informasi dekat waktu nyata draft spesifik pengolahan tanah, kedalaman olah dan ukuran implemen. Data sampel dan pengukuran dikumpulkan dari perkebunan tebu di Kabupaten Kediri dan Sidoarjo, Jawa Timur Indonesia. Penelitian membuktikan terdapat korelasi yang erat antara nilai hamburan radar Sentinel-1A dengan kadar air tanah permukaan (kedalaman 15 cm); dan kadar air tanah dengan tahanan penetrasi tanah (kedalaman 15 cm) pada lahan perkebunan tebu yang belum ditanami. Pendekatan machine learningrandom forest dan analisis regresi berganda telah digunakan untuk mengembangkan model perhitungan kadar air dan tahanan penetrasi tanah; dan diuji terhadap data testing. Indikator kinerja model regresi berganda dan model machine learning-random forest (RMSE, MAPE, dan akurasi) telah dihitung dan diuji dengan hasil RMSE = 0,250 dan 0,270, MAPE = 18,79% dan 17,62%, akurasi = 81,21 % dan 82,38%. Penentuan kadar air tanah menggunakan metode machine learning-random forest menunjukkan kinerja yang lebih baik dibandingkan iv pemodelan regresi berganda. Penelitian juga membuktikan bahwa dinamika tahanan penetrasi tanah dipengaruhi oleh dinamika kadar air tanah dan tekstur tanah, khususnya fraksi liat dan debu. Metode machine learning-random forest menunjukkan kinerja terbaik dibandingkan dengan metode regresi berganda dalam menghitung tahanan penetrasi tanah. Korelasi antara tahanan penetrasi tanah dan kadar air tanah, fraksi liat dan debu menggunakan kedua metode telah dihitung dan diuji dengan kinerja yang baik, yang ditunjukkan oleh RMSE = 0,235 dan 0,208, MAPE = 18,32% dan 16,08%, akurasi = 81,68% dan 83,92%. Distribusi spasial kadar air tanah, tahanan penetrasi, kemampuan kerja tanah dan draft spesifik pengolahan tanah dihitung dan divisualisasikan menggunakan peta spasial. Informasi variabilitas lahan, dalam hal ini draft spesifik pengolahan tanah sebagai fungsi dari kadar air dan tahanan penetrasi tanah digunakan dalam pemilihan mesin yang tepat untuk pengolahan lahan tebu berdasarkan prinsip-prinsip pertanian presisi.

The application of precision farming principles to large-scale agriculture such as plantations requires technological support and machine operation management to handle agricultural activities and manage land spatial variability. Tillage machines are the main input component that contributes significantly to agricultural yields, the continuity of agricultural activities and environmental sustainability. The effectiveness and efficiency of tillage machines are adjusted to actual needs which can be determined from the area, land spatial variability and environmental factors. However, soil sampling to determine spatial variability of large-scale agriculture land such as sugarcane plantation is ineffective and inefficient because it requires time and cost to test samples in the laboratory. The adoption and integration of geographic information systems, remote sensing and navigation systems into knowledge management of soil physical and mechanical conditions is indispensable for better decision making, planning and management of intensive agricultural land. This study aims to design a method of selecting soil tillage machines based on the spatial variability of unplanted sugarcane farm using a smart and precision agriculture approach. The design requires a series of studies which include (1) developing a near-real-time soil water content calculation model based on Sentinel-1A data and determining soil workability levels; (2) developing a model to calculate soil penetration resistance based on soil water content obtained from Sentinel-1A data and soil fraction; (3) calculating spatially soil specific draft based on soil penetration resistance and plasticity index; and (4) developing a smart and precision agriculture approach for selection of tillage machine based on spatial variability (i.e. near real-time soil water content, penetration resistance and soil specific draft). Samples and measurement data were collected from two sugarcane plantations in the districts of Kediri and Sidoarjo, East Java, Indonesia. Samples and data were taken randomly from each selected block with a distance of 100 meter between the sampling point and or measurement point. The Study has shown that there is a close correlation between the Sentinel-1A backscattering value and the surface soil water content (at a depth of 15 cm); and soil water content with soil penetration resistance (at a depth of 15 cm) in the two unplanted sugarcane plantations. The random forest method, which is one of the artificial intelligence methods, and multiple non-linear regression methods were used and compared to develop the best soil water content and soil penetration resistance models. Calculation of soil water content using the random forest method showed better performance than the non-linear regression method. The performance indicators of multiple non-linear regression models and random forest-based models (RMSE, MAPE, and accuracy) have been calculated and tested with the results of RMSE = 0.250 and 0.270, MAPE = 18.79% and 17.62%, accuracy = 81.21 % and 82.38%, respectively. The research results also show that the dynamics of soil penetration resistance is influenced by the dynamics of soil water content and soil texture, especially clay and silt fractions. The random forest method showed the best performance compared to the non-linear multiple regression method in calculating soil penetration resistance. The correlation between soil penetration resistance and soil moisture content, clay and silt fraction using both methods has been calculated and tested with good performance, which is indicated by RMSE = 0.235 and 0.208, MAPE = 18.32% and 16.08%, accuracy = 81.68% and 83.92%, respectively. The spatial distribution of soil water content, soil penetration resistance and soil workability of Kediri and Sidoarjo sugarcane plantations can be calculated and visualized using spatial maps. The results of the calculation of land variability are used in the selection of soil tillage machines to support sugarcane cultivation based on smart and precision agriculture principles.