IoT-Driven Tree-Specific Soil Nutrient Management for Cashew – Comprehensive Study Notes

ABSTRACT / STUDY OVERVIEW

  • Focus crop: Cashew (Anacardium occidentale) – introduced to India by Portuguese in the 16th century; highly valued export commodity.
  • Region of interest: Cuddalore District, Tamil Nadu, India – farmers motivated by export market and local suitability.
  • Goal: Develop an IoT-driven, tree-specific soil-nutrient-management system that predicts precise fertilizer doses (N, P, K) and intercrop suggestions for individual cashew trees.
  • Core technologies
    • Hardware: Arduino UNO, RS-485 module, soil NPK sensor, OLED display.
    • Software/AI: Long Short-Term Memory–based Recurrent Neural Network (LSTM-RNN) trained on soil data + tree metadata.
  • Outcomes
    • Prediction accuracy ≈98.7%\approx 98.7\% (beats Random Forest, Linear Classifier, Linear Random Classifier, Decision Tree Classifier).
    • Rapid nutrient reading: 60 s\text{60 s} via sensor versus 10–20 days\text{10–20 days} in laboratory.
    • Recommendations include: fertilizer ratio & quantity, appropriate intercrops, nearest fertilizer shop.

INTRODUCTION: INDIAN AGRICULTURE & CASHEW CONTEXT

  • Indian agriculture transitioning from traditional to technology-enabled practices; vulnerable to climate change & farmer distress.
  • Major farmer issues
    • Over/under use of fertilizer without soil testing.
    • Reliance on outdated techniques; timing errors; resultant yield loss & soil degradation.
  • Cashew importance
    • Third-largest plantation foreign-exchange earner after tea & coffee.
    • Nuts rich in fat, protein, carbs, minerals, vitamins; rising demand in snacks & confectionery.
    • Forecasted expansion due to preference for low-calorie snack bars.

CASHEW CULTIVATION DETAILS (TAMIL NADU)

  • Typical planting densities
    • Standard: 200 plants ha−1200\ \text{plants ha}^{-1}.
    • High-density (5 × 4 m): 500 plants ha−1500\ \text{plants ha}^{-1} with initial fertilizer 225:75:75 kg NPK ha−1225:75:75\ \text{kg NPK ha}^{-1}.
  • Fertilizer schedule after Year 5: 500:200:300 g NPK tree−1 yr−1500:200:300\ \text{g NPK tree}^{-1}\,\text{yr}^{-1}.
  • Key growing districts: Cuddalore, Ariyalur, Pudukottai, Tirunelveli, Villupuram, Theni.
  • Market hubs: Jayankondam, Vridhachalam, Panruti; nut grades include White/Pieces, Splits, Butts, 320, 240, 180.
Tamil Nadu Released Varieties (extract)
  • VRI-1 (M10/4): 7.20 kg tree−1, W320, yellowapple7.20\ \text{kg tree}^{-1},\ W320,\ yellow apple.
  • VRI-4 (M26/2): 16.60 kg tree−1, W240, orange–redapple16.60\ \text{kg tree}^{-1},\ W240,\ orange–red apple.
  • VRI(CW) H1 hybrid: 16.50 kg tree−1, W21016.50\ \text{kg tree}^{-1},\ W210.
Recommended Manure & Fertilizer per Tree (Year-wise)
  • Year 1: 10 kg FYM, 70 g N, 40 g P, 60 g K10\ \text{kg FYM},\ 70\,\text{g N},\ 40\,\text{g P},\ 60\,\text{g K}.
  • Year 2: 20 kg FYM, 140 g N, 80 g P, 120 g K20\ \text{kg FYM},\ 140\,\text{g N},\ 80\,\text{g P},\ 120\,\text{g K}.
  • Year 5 onwards: 50 kg FYM, 500:200:300 g NPK50\ \text{kg FYM},\ 500:200:300\ \text{g NPK}.

SOIL-NUTRIENT MANAGEMENT PRINCIPLES

  • Cashew, though hardy, removes nutrients annually (30-y tree: 2.85 kg N, 0.75 kg P, 1.265 kg K2.85\,\text{kg N},\ 0.75\,\text{kg P},\ 1.265\,\text{kg K}) → requires replenishment.
  • Fertilizer timing
    • Prefer post-monsoon (Sep–Oct) single dose when soil moisture high → roots active (flushing & early flowering).
    • Alternative: split pre-monsoon & post-monsoon doses.
  • Application method
    • Sandy/laterite heavy-rain areas: circular trench 25 cm w×15 cm d25\,\text{cm w} \times 15\,\text{cm d} at 1.5 m1.5\,\text{m} radius.
    • Red-loam/low-rain: trench ≈50 cm w\approx 50\,\text{cm w}; radius increases yearly (0.5 m → 1.5 m by Year 4).
  • Organic matter: 10–15 kg10–15\,\text{kg} compost or FYM per plant mandatory.

PROBLEM STATEMENT

  • Cashew often cultivated on marginal, sloping “wastelands” with minimal inputs; farmers lack soil diagnostics.
  • Imbalanced fertilization → declining soil fertility, lower yield, disease susceptibility.
  • Need site-specific, tree-specific nutrient advice leveraging IoT + AI to reduce guesswork and resource waste.

LITERATURE SNAPSHOT

  • Nutrient constraints in acidic soils: low pH, Al/Mn toxicity, low base saturation, P fixation, low Ca, Mg, K.
  • Yield improvement via
    • Biofertilizers: Azospirillum (N) & Phosphobacteria (P).
    • Poultry manure + NPK combinations (e.g.
      500 g N+125 g P+10 kg poultry manure tree−1500\,\text{g N} + 125\,\text{g P} + 10\,\text{kg poultry manure}\,\text{tree}^{-1} under rainfed, normal density).
    • Foliar sprays: Urea 2–4%2–4\%, DAP 1%1\%, ZnSO4 4%\text{ZnSO}_4\ 4\%, Cu0.3–0.6%\text{Cu} 0.3–0.6\%.
  • Soil-NPK measurement techniques: laboratory wet chemistry, ICP, fluorescence spectroscopy, ion-selective electrodes, optical sensing, Extreme Learning Machine classification, etc.
  • Prior ML/DL for fertilizer recommendation: Random Forest, SVM, KNN, Fuzzy logic, XGBoost, IDCNN, DRQN; accuracies up to 99.7%99.7\% (groundnut case).

METHODS & SYSTEM ARCHITECTURE

Hardware Stack
  • Arduino UNO microcontroller.
  • NPK soil sensor connected via RS-485 interface; provides real-time mg kg−1\text{mg kg}^{-1} or ppm values for N, P, KN,\ P,\ K.
  • OLED display for field read-outs.
Sampling Strategy
  • 500 individual cashew trees sampled across three villages
    • Kumalankulam (red-loam)
    • M. Puthur (red-loam)
    • Nochikadu (alluvial/sandy)
  • Metadata per sample: Sample ID, GPS/location, tree age, trunk diameter.
Dataset Construction
  • Inputs
    • Sensor-measured N,P,KN, P, K.
    • Soil type label (sandy, red-loam, alluvial, etc.).
    • Soil pH.
    • Tree age (years).
  • Targets
    • Recommended fertilizer formulation (e.g.
      Urea, SSP, MOP quantities).
    • Recommended quantity (g tree-1 or kg ha-1).
  • Additional sources: Government Soil Health Cards, ICAR cashew fertilizer guides.
Model Design
  • LSTM-RNN architecture
    • 2 stacked LSTM layers → dense output layer.
    • Loss: Mean-Squared Error for regression; Softmax for categorical fertilizer-type prediction.
    • Trained–test split applied; evaluation via Accuracy, Precision, Recall, F1, MAE, RMSE, R2\text{MAE},\ \text{RMSE},\ R^{2}.
  • Baseline models for comparison: Random Forest, Linear Classifier, Linear Random Classifier, Decision Tree.
Intercropping Logic Module
  • Avoid pest host plants.
  • Suggested profitable intercrops (based on soil & climate):
    • Legumes: groundnut (yield benefit ≈ ₹16 188 ha⁻¹), blackgram (highest cost-benefit), cowpea, horse gram, beans.
    • Tuber crops: tapioca, yam.
    • Spices: turmeric, ginger.
    • Medicinal: Aloe vera (high profitability).

RESULTS & PERFORMANCE

Sensor Versus Lab Validation
  • Example readings (NPK kg ha⁻¹)
    • Kumalankulam: Sensor N=195, P=17, K=142N=195,\ P=17,\ K=142; Lab N=210, P=19, K=156N=210,\ P=19,\ K=156.
    • Measurement time: Sensor 60 s60\,\text{s} vs. Lab 10–20 days\text{10–20 days}.
  • Average village nutrient status (sensor values)
    • Nochikadu (sandy): Nˉ=196.3, Pˉ=19.45, Kˉ=189.1\bar N=196.3,\ \bar P=19.45,\ \bar K=189.1.
    • M.Puthur (red-loam): Nˉ=143.6, Pˉ=21.1, Kˉ=177.7\bar N=143.6,\ \bar P=21.1,\ \bar K=177.7.
    • Kumalankulam (red-loam): Nˉ=153.5, Pˉ=23, Kˉ=177.4\bar N=153.5,\ \bar P=23,\ \bar K=177.4.
    • District average: Nˉ=164.47, Pˉ=21.18, Kˉ=181.4\bar N=164.47,\ \bar P=21.18,\ \bar K=181.4.
Model Metrics
  • Proposed LSTM-RNN
    • Prediction Accuracy =98.7%=98.7\%.
    • Precision, Recall, F1 all >0.95.
    • Error values lower than baselines (MAE, RMSE minimal; R2R^{2} near 1).
  • Comparative ranking: RNN > Random Forest > Decision Tree > Linear models.
Functional Outputs (System Screens)
  • Immediate display: soil N,P,KN, P, K values.
  • AI-generated advice
    • Fertilizer name + ratio (e.g. “Apply Urea 120 g120\,\text{g}, SSP 48 g48\,\text{g}, MOP 48 g48\,\text{g} around drip line”).
    • Scheduling (pre-/post-monsoon, single vs. split dose).
    • Intercrop suggestion (e.g. “Groundnut suitable; expected net return ₹16 k ha⁻¹”).
    • Nearest agro-dealer list based on GPS.

DISCUSSION / AGRONOMIC INSIGHTS

  • Cuddalore soils generally
    • Low N <164.5\,\text{kg ha}^{-1} → focus on N-rich fertilizers (Urea) for young trees.
    • PP often >22 kg ha⁻¹ (adequate–high) except parts of Virudhachalam.
    • KK medium (110–350 kg ha⁻¹) but cashew demands higher P,KP, K than NN during bearing stage.
  • Tree-specific dosing (vs. blanket) raises yield ≈ 50%50\% and reduces input waste; also minimizes environmental runoff.
  • Labor reduction: quick sensor measure + automated AI eliminates 10-20 day lab wait.

CONCLUSION

  • Developed a practical, scalable IoT + LSTM-RNN framework for precision cashew fertilization.
  • Achieved 98.57%98.57\% accuracy; sensor-driven treatment planning doubles decision speed, lifts yield, improves profitability, and preserves soil health.
  • Provides template for other perennial orchards where intra-plot variability is high.

ETHICAL & PRACTICAL IMPLICATIONS

  • Supports smallholders by democratizing agronomic expertise.
  • Reduces over-fertilization → less groundwater pollution & greenhouse-gas emissions.
  • Data privacy: farm-level nutrient data should be anonymized when aggregated for research/policy.

KEY EQUATIONS & METRICS (IN CONTEXT)

  • Planting density (high): 15 m×4 m=500 plants ha−1\frac{1}{5\,\text{m} \times 4\,\text{m}} = 500\,\text{plants ha}^{-1}.
  • Cashew nutrient removal (30 y): N=2.85 kg, P=0.75 kg, K=1.265 kgN=2.85\,\text{kg},\ P=0.75\,\text{kg},\ K=1.265\,\text{kg} per tree.
  • Accuracy formula: Accuracy=TP+TNTP+TN+FP+FN\text{Accuracy}=\frac{TP+TN}{TP+TN+FP+FN} (applied to fertilizer-type classification).

SELECTED REFERENCES (FOR FURTHER STUDY)

  • Babu et al., 2015 – Cashew response to NPK.
  • Mangalassery et al., 2021 – Nutrient norms for coastal cashew.
  • Chitdeshwari et al., 2017 – GPS/GIS soil fertility map, Cuddalore.
  • Revati Potdar et al., 2021 – Optical NPK determination review.
  • Sivasankaran et al., 2022 – 1D-CNN fertilizer recommender (groundnut).
  • Olubode et al., 2018 – Cashew production techniques.
  • ICAR-DCR portal – Soil & fertilizer guidelines for cashew.