RANCANGAN SENSOR IOT UNTUK MONITORING KELEMBAPAN DAN PH TANAH SECARA REAL-TIME DI LAHAN FAKULTAS PERTANIAN UNIVERSITAS WAHIDIYAH

Abstrak

ABSTRAK

Siti Indra Rukmana. 2026. Rancangan Sensor IoT untuk Monitoring Kelembapan dan pH Tanah secara Real-Time di Lahan Fakultas Pertanian Universitas Wahidiyah.

Kata Kunci: Internet of Things, ESP32-C3, soft sensor, Jaringan Saraf Tiruan, Fast Fourier Transform, K-Fold Cross-Validation, kelembapan tanah, pH tanah.

Tanah merupakan media tumbuh utama dalam pertanian yang kesuburannya ditentukan oleh kelembapan, derajat keasaman (pH), serta unsur hara nitrogen (N), fosfor (P), dan kalium (K). Pemantauan tanah secara manual cenderung subjektif dan tidak konsisten sehingga diperlukan sistem pemantauan otomatis. Penelitian ini bertujuan merancang, menguji, dan memvalidasi sistem soft-sensor berbasis Internet of Things (IoT) untuk menduga parameter tanah secara real-time di lahan Fakultas Pertanian Universitas Wahidiyah. Metode yang digunakan adalah kuantitatif eksperimental terhadap 50 sampel tanah. Sistem dibangun menggunakan empat unit mikrokontroler ESP32-C3 (satu Access Point dan tiga klien) yang terhubung melalui jaringan WiFi lokal, dengan data tersimpan pada basis data MySQL dan ditampilkan melalui dashboard web. Pengukuran dilakukan menggunakan elektroda kustom dua batang stainless steel yang sinyalnya ditransformasikan dengan Fast Fourier Transform (FFT) menjadi 2.048 fitur, kemudian diolah oleh Jaringan Saraf Tiruan (JST) berarsitektur 2048–30–5 dengan algoritma optimasi ADAM dan fungsi aktivasi ReLU. Nilai acuan (ground truth) pH, N, P, dan K diperoleh dari soil test kit, sedangkan kadar air ditentukan secara gravimetri. Hasil penelitian menunjukkan bahwa sistem IoT berhasil dirancang dan diimplementasikan secara utuh dengan seluruh komponen berfungsi dan saling terhubung. Pada tahap pelatihan, model JST mencapai akurasi keseluruhan 98,55% dan akurasi kadar air 93,94% dengan rata-rata MSE 0,0126, namun nilai ini semata mencerminkan kecocokan model terhadap data yang telah dihafalnya. Validasi 5-Fold Cross-Validation pada data baru menghasilkan akurasi generalisasi keseluruhan sebesar 68,58%, membuktikan bahwa model tidak mengalami overfitting yang memanipulasi performa. Akurasi tinggi pada N, P, K, dan pH bersifat trivial karena rendahnya variansi data di lapangan, sehingga kadar air, dengan rentang 6%–40%, menjadi tolok ukur prediksi paling bermakna, dengan akurasi tervalidasi sebesar 46,45%. Dengan demikian, penelitian ini berhasil membuktikan bahwa sistem IoT tidak hanya baik pada aspek perancangan dan implementasi, tetapi juga menunjukkan kemampuan prediktif yang nyata melalui validasi generalisasi, meskipun akurasi pada parameter kadar air masih perlu ditingkatkan pada penelitian selanjutnya.

ABSTRACT

Siti Indra Rukmana. 2026. Design of an IoT Sensor for Real-Time Monitoring of Soil Moisture and pH on the Agricultural Land of the Faculty of Agriculture, Universitas Wahidiyah.

Keywords: Internet of Things, ESP32-C3, soft sensor, Artificial Neural Network, Fast Fourier Transform, K-Fold Cross-Validation, soil moisture, soil pH.

Soil is the main growing medium in agriculture, and its fertility is determined by moisture, acidity (pH), and the nutrients nitrogen (N), phosphorus (P), and potassium (K). Manual soil monitoring tends to be subjective and inconsistent, so an automated monitoring system is needed. This study aims to design, test, and validate an Internet of Things (IoT)-based soft-sensor system for estimating soil parameters in real time on the agricultural land of the Faculty of Agriculture, Universitas Wahidiyah. The method used was experimental quantitative research on 50 soil samples. The system was built using four ESP32-C3 microcontrollers (one Access Point and three clients) connected through a local WiFi network, with data stored in a MySQL database and displayed on a web dashboard. Measurements were carried out using custom dual stainless-steel electrodes whose signals were transformed using the Fast Fourier Transform (FFT) into 2,048 features, which were then processed by an Artificial Neural Network (ANN) with a 2048–30–5 architecture using the ADAM optimizer and the ReLU activation function. The ground-truth values of pH, N, P, and K were obtained from a soil test kit, while moisture content was determined gravimetrically. The results show that the IoT system was successfully designed and fully implemented, with all components functioning and interconnected. During training, the ANN model achieved an overall accuracy of 98.55% and a moisture accuracy of 93.94% with an average MSE of 0.0126, although this figure merely reflects the model's fit to data it had already memorized. Validating the model on unseen data through 5-Fold Cross-Validation produced an overall generalization accuracy of 68.58%, confirming that the model does not suffer from overfitting that manipulates performance. The high accuracy for N, P, K, and pH is trivial given the low variance of the field data, so moisture content, ranging from 6% to 40%, stands as the most meaningful predictive benchmark, with a validated accuracy of 46.45%. Therefore, this study demonstrates that the IoT system performs well not only in design and implementation but also shows genuine predictive capability through generalization validation, although moisture-prediction accuracy still requires improvement in future research.