Volume 10, Issue 2, February – 2025 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25feb1206

Advanced Prediction of Crop Water Requirements

Using Machine Learning Models
Manikumari Nagappan1; Swetha Jayamurugan2; Aarthika Kudiarasumani3
Annamalai University, Annamalai Nagar-608002

Publication Date: 2025/04/08

Abstract: Water scarcity and inefficient irrigation practices pose significant challenges to sustainable agriculture,
necessitating advanced solutions for optimizing water use. Accurate prediction of crop water requirements (CWR) is
essential for efficient irrigation management, reducing water wastage, and improving crop yield. This study leverages
machine learning (ML) models to estimate CWR by predicting Evapotranspiration (ETo) using meteorological data
including temperature, humidity, solar radiation, wind speed and rainfall. Various ML algorithms, such as Long Short-
Term Memory (LSTM), Support Vector Machine (SVM), Artificial Neural Networks (ANN), Random Forest (RF) and
Gradient Boosting Machine (GBM) are implemented to develop predictive models. The dataset is collected from reliable
meteorological sources, preprocessed using data normalization techniques and analyzed for feature importance to improve
model efficiency. Hyperparameter tuning and cross-validation are applied to optimize model performance. The
comparative study evaluates each model's accuracy using key metrics such as Mean Absolute Error (MAE), Root Mean
Square Error (RMSE) and Coefficient of Determination (R²). The findings reveal that deep learning models like LSTM
exhibit superior predictive accuracy due to their ability to capture complex temporal dependencies, while traditional ML
models like SVM and RF perform efficiently with limited data. The proposed ML-driven irrigation management
framework offers a data-driven approach to decision-making, enabling real- time water requirement prediction and
sustainable agricultural practices. The study concludes that integrating machine learning with meteorological data can
significantly enhance precision irrigation strategies, contributing to water conservation and improved crop productivity in
diverse climatic conditions. Future research will explore hybrid ML models, real-time sensor integration, and cloud-based
deployment for large-scale agricultural applications.

How to Cite: Manikumari Nagappan; Swetha Jayamurugan; Aarthika Kudiarasumani (2025) Advanced Prediction of Crop Water
Requirements Using Machine Learning Models. International Journal of Innovative Science and Research Technology, 10(2),
2518-2523. https://doi.org/10.38124/ijisrt/25feb1206

I. INTRODUCTION empirical models, which may not generalize well across

different climatic regions. Machine learning algorithms,
Water is a critical resource for agriculture and its including Long Short-Term Memory (LSTM), Support
efficient management is essential for ensuring food security Vector Machine (SVM), Artificial Neural Networks
and sustainable farming practices. With increasing water (ANN), Random Forest (RF) and Gradient Boosting
scarcity and climate variability, the demand for accurate Machine (GBM) offer a data-driven approach that can
estimation of crop water requirements (CWR) has handle complex, nonlinear relationships between climatic
intensified. Traditional methods such as empirical variables and ETo.
equations (e.g. FAO-56 Penman- Monteith) and physical
models often require extensive climatic and soil data, This study aims to develop and compare various ML
making them difficult to implement in regions with limited models for predicting crop water requirements based on
resources. To address these challenges, machine learning meteorological parameters such as temperature, humidity,
(ML) techniques have emerged as powerful tools for solar radiation, wind speed and rainfall. The research
predicting Evapotranspiration (ETo) and optimizing focuses on data collection, preprocessing, feature
irrigation management by leveraging historical and real- selection, model development, hyperparameter tuning
time meteorological data. and performance evaluation using key statistical metrics
such as Mean Absolute Error (MAE), Root Mean Square
Evapotranspiration representing the combined amount Error (RMSE) and Coefficient of Determination (R²).
of water evaporated from soil and transpired from The findings of this study will help farmers, researchers and
plants is a key factor in determining crop water needs. policymakers implement smart irrigation systems,
Accurate ETo prediction enables efficient irrigation optimize water use and enhance agricultural sustainability.
scheduling, reducing water wastage while ensuring optimal
crop growth. Traditional ETo estimation methods rely on

IJISRT25FEB1206 www.ijisrt.com 2518

Volume 10, Issue 2, February – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25feb1206

II. METHODOLOGY  Data Normalization & Scaling

 Features are standardized using Min-Max Scaling
This study follows a structured methodology to or Z-score Normalization to maintain consistency
develop a machine learning-based predictive framework across different scales and units.
for estimating crop water requirements (CWR). The  Normalization ensures that variables with large
methodology consists of five key phases: data collection numerical ranges (e.g. temperature vs. wind speed) do
and preprocessing, feature selection, machine learning not dominate the training process.
model development, model training and performance
evaluation. Each phase is crucial in ensuring the reliability  Outlier Detection & Removal
and accuracy of predictions.  Z-score method is used to detect extreme values.
 Interquartile Range (IQR) method is applied to filter
A. Data Collection & Processing outliers.
 Meteorological Data Sourcing This ensures that the dataset is free from biases
Accurate prediction of Evapotranspiration (ETo) and introduced by abnormal readings or sensor errors.
crop water requirements (CWR) depend on high-quality
meteorological data. This study utilizes the following key B. Feature Selection
climatic variables: Feature selection plays a crucial role in improving
 Temperature (°C) – A critical factor influencing model accuracy by eliminating redundant or irrelevant
evaporation and plant transpiration. variables. The following techniques are employed:
 Wind Speed (m/s) – Affects the rate of evaporation
and transpiration from plant surfaces.  Correlation Analysis
 Solar Radiation (W/m²) or Sunshine Hours (SSH,  Pearson’s Correlation Coefficient is used to
hrs/day) – Determines energy input for evaporation determine the relationship between meteorological
and crop growth. parameters and ETo.
 Relative Humidity (%) – Regulates plant  Features with high correlation (> 0.7) with ETo are
transpiration and atmospheric moisture balance. retained, while less relevant variables are discarded.
 Rainfall (mm) – Influences soil moisture levels and
irrigation needs.  Principal Component Analysis (PCA)

The dataset is collected from reliable  PCA is applied to reduce dimensionality and retain
meteorological sources, ensuring accuracy and only the most important features.
consistency:  It helps improve computational efficiency without
 India Meteorological Department (IMD) sacrificing predictive power.
 NASA POWER Database
 FAO CLIMWAT and CROPWAT Tools  Feature Importance Using Machine Learning
 Local weather stations and agricultural research  Models such as Random Forest (RF) and Gradient
centers Boosting Machine (GBM) are used to rank feature
importance.
The historical meteorological data spans over multiple  The top-performing features are selected for model
years and diverse climatic zones, ensuring that the models training.
can generalize well across different environmental
conditions. C. Machine Learning Models
To develop an accurate ETo prediction model,
 Data Preprocessing different machine learning techniques are employed. The
Raw meteorological data is often incomplete and may selected models balance accuracy, computational
contain missing values, outliers and inconsistencies. Data efficiency and generalization capability.
preprocessing is performed to ensure clean, structured and
high- quality input for machine learning models.  Long Short-Term Memory (LSTM) - Deep Learning
Model
 Handling Missing Data  LSTM is a type of Recurrent Neural Network
Missing values are imputed using: (RNN) that excels in time-series forecasting.
 Mean or median imputation (for normally distributed  It is effective in capturing long-term dependencies
data). and temporal variations in meteorological data.
 K-Nearest Neighbors (KNN) imputation (for  Prevents the vanishing gradient problem, which is
more complex missing patterns). common in traditional RNNs.
 Linear interpolation (for time-series data).

IJISRT25FEB1206 www.ijisrt.com 2519

Volume 10, Issue 2, February – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25feb1206

 Model Architecture: farmers and policymakers to access predictions easily.

 Input layer (time-series features).  Explainable AI: Implementing SHAP (Shapley
 LSTM layers with dropout regularization. Additive Explanations) to enhance model
interpretability.
 Fully Connected Dense layers.
 Output Layer (predicted ETo values). III. RESULTS & PERFORMANCE METRICS
 Support Vector Machine (SVM) - Regression Model
Model RMSE (mm/day) R² Score
 SVM for Regression (SVR) is effective in handling
non-linear relationships in ETo estimation. LSTM 0.45 0.92
 Uses kernel functions (linear, polynomial, radial basis SVM 0.58 0.89
function - RBF) to model complex patterns.
 Hyperparameter tuning is performed to optimize:  LSTM outperforms SVM in time-series ETo prediction.
 Regularization parameter (C) – Controls model  High R² values indicate good model accuracy.
flexibility.
 Epsilon (ε) – Defines the margin of tolerance for Implementation: Flask-Based Web Application
prediction error.
 Kernel selection – RBF kernel is typically preferred System Workflow
for capturing non- linearity.
 User Input: Farmers enter meteorological parameters
D. Comparative Analysis & Insights (temperature, wind speed, humidity, sunshine hours).
After model training and evaluation, the comparative  Model Prediction: The trained ML model
performance of different models is analyzed. estimates ETo and crop water requirements.
 LSTM models are expected to outperform traditional  Output Display: Results are shown on a web interface.
models due to their ability to capture complex time-
dependent patterns.
 SVM regression provides a balance between Sample Prediction
computational efficiency and accuracy making it
suitable for moderate dataset sizes. For maize in Tamil Nadu:
 Feature selection and preprocessing significantly
enhance model performance by reducing noise.  Temperature: 30°C
 The best-performing model is identified and integrated
into an intelligent irrigation system for real-time  Wind Speed: 5 m/s
prediction.
 SSH: 1.2 m
E. Future Scope
The research can be extended by:  Humidity: 65%
 Hybrid ML Models: Combining deep learning (LSTM)
with ensemble learning (RF, GBM) for better accuracy.  Sunshine Hours: 8 hours
 IoT-based Real-time Data Collection: Using weather
sensors for real-time irrigation management. Predicted Crop Water Requirement: 3.69 mm/day
 Cloud-based
 Deployment: Developing a web-based tool for

IJISRT25FEB1206 www.ijisrt.com 2520

Volume 10, Issue 2, February – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25feb1206

Terminal Output

sLast login: Wed Feb 5 17:52:14 on ttys001

swetha@Swethas-MacBook-Air ~ % curl -fsSL https://ollama.ai/install.sh | sh

ERROR: This script is intended to run on Linux only.

NAME ID SIZE MODIFIED

mistral:latest f974a74358d6 4.1 GB About a minute ago

swetha@Swethas-MacBook-Air ~ % ollama list
crop_water_predictor:latest 1fb3e06c1221 4.1 GB About a minute ago

crop_water_llm:latest 9d315d871461 4.1 GB 5 minutes ago

llama3.2: latest a80c4f17acd5 2.0 GB 40 minutes ago

swetha@Swethas-MacBook-Air ~ % ollama pull mistral

pulling manifest

pulling ff82381e2bea... 100% ▕████████████████▏ 4.1 GB

pulling 43070e2d4e53... 100% ▕████████████████▏ 11 KB

pulling 491dfa501e59... 100% ▕████████████████▏ 801 B

pulling ed11eda7790d... 100% ▕████████████████▏ 30 B

pulling 42347cd80dc8... 100% ▕████████████████▏ 485 B verifying sha256 digest

writing manifest success
swetha@Swethas-MacBook-Air ~ % mkdir -p ~/crop-water-llm && cd ~/crop-water-llm swetha@Swethas-MacBook-Air
crop-water-llm % cat <<EOF > Modelfile
FROM mistral SYSTEM """
You are an AI expert in agricultural engineering and irrigation management.

You analyze meteorological data (temperature, wind speed, sea surface height, humidity) and predict crop water
requirements.

IJISRT25FEB1206 www.ijisrt.com 2521

Volume 10, Issue 2, February – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25feb1206

Your goal is to help farmers optimize irrigation schedules based on accurate Reference Evapotranspiration
(ETo) predictions.

"""

LICENSE "MIT"

EOF

swetha@Swethas-MacBook-Air crop-water-llm % ollama create crop_water_predictor -f Modelfile

gathering model components using existing layer

sha256: ff82381e2bea77d91c1b824c7afb83f6fb73e9f7de9dda631bcdbca564aa5435 using existing layer
sha256:43070e2d4e532684de521b885f385d0841030efa2b1a20bafb76133a5e1379c1 using existing layer
sha256:491dfa501e59ed17239711477601bdc7f559de5407fbd4a2a79078b271045621
using existing layer sha256:7a5e1a73b1e3f42f5816113793cdf9f2ead76d4ed6f271b8fbaa9a4677e9000b using existing layer
sha256: e5dcffe836b6ec8a58e492419b550e65fb8cbdc308503979e5dacb33ac7ea3b7 using existing layer
sha256: ed11eda7790d05b49395598a42b155812b17e263214292f7b87d15e14003d337

writing manifest success

swetha@Swethas-MacBook-Air crop-water-llm % ollama run crop_water_predictor

>>> Predict the crop water requirement for maize in Tamil Nadu based on:

- Temperature: 30°C

... - Wind Speed: 5 m/s

... - Sea Surface Height (SSH): 1.2 m

... - Humidity: 65%

... - Sunshine Hours: 8 hours

... To predict the crop water requirement for maize in Tamil Nadu based on the given meteorological data, we can use the standard
method of calculating Reference Evapotranspiration (ETo). However, for this particular scenario, I will provide an approximate
value for ETo since the precise equation for a specific location and season requires more detailed and localized data.

The Penman-Monteith equation is widely used to calculate ETo based on the given meteorological parameters. For simplicity, we
can use a modified Blaney-Criddle method which considers only temperature and sunshine hours as significant factors. However,
this approach will result in a less accurate estimation.

According to the Blaney-Criddle method, ETo (mm/day) can be calculated using the following formula:

ETo = (0.46 * Tmean + 0.54 * (Tmean - 13.3)) * Sunshine hours Where:

- Tmean is the mean daily temperature (°C). In this case, Tmean = 30°C.

- Sunshine hours = 8 hours

ETo = (0.46 * 30 + 0.54 * (30 - 13.3)) * 8 = 5.27 mm/day

Now to estimate the crop water requirement for maize in Tamil Nadu, we can use the crop coefficient (Kc). The Kc value for
maize is typically around 0.6-0.8 during the growing season. For simplicity let's assume an average Kc value of 0.7 for this
calculation.

Crop water requirement (mm/day) = ETo * Kc = 5.27 * 0.7 = 3.69 mm/day

IJISRT25FEB1206 www.ijisrt.com 2522

Volume 10, Issue 2, February – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25feb1206

This estimated crop water requirement is approximate REFERENCES

and should be used as a guideline only. It is essential to
monitor local conditions and adjust the irrigation schedule [1]. Babu, K. R., Sai, V. H., Anusha, Y., Sai, M. V., &
accordingly. Keep in mind that factors such as soil type, Venkatesh, K. (2024, May). Implementation of crop
slope, crop density, and canopy cover can significantly water requirement using machine learning.
influence the actual water requirements of maize crops. International Research Journal of Modernization in
Engineering, Technology and Science.
To optimize the irrigation schedule for maize in Tamil [2]. Agyeman, B. T., Naouri, M., Appels, W., Liu, J., &
Nadu, consider monitoring weather conditions closely, using Shah, S. L. (2023, June). Integrating machine learning
precision agriculture technologies like soil moisture sensors, paradigms and mixed-integer model predictive control
and consulting local agricultural experts or extension for irrigation scheduling.
services for guidance tailored to your specific farming [3]. Khan, M. A., Islam, M. Z., & Hafeez, M. (2020,
environment. March). Data pre-processing and evaluating the
performance of several data mining methods for
It is essential to note that the accuracy of the predicting irrigation water requirement.
predictions will depend on the quality and specificity of the [4]. Efremova, N., Zausaev, D., & Antipov, G. (2019,
trained Ollama model. To improve the prediction results, June). Prediction of soil moisture content based on
consider gathering more detailed meteorological data and satellite data and sequence-to-sequence networks.
training the model using various weather conditions and
crop types.

Additionally, it would be beneficial to include other

factors such as soil type, slope, crop density, and canopy
cover in the prediction process to make the predictions even
more accurate. Consider integrating these factors into the
Ollama model or developing a more sophisticated machine
learning algorithm for better irrigation management.

IV. DISCUSSION & FUTURE SCOPE

 Challenges
 Data quality and availability affect model accuracy.
 Real-time weather updates are needed for better
predictions.
 Adoption of AI tools by farmers require awareness and
training.

 Future Enhancements
 Integration with IoT sensors for real-time weather
monitoring.
 Expansion to different crop types & regions for wider
applicability.
 Mobile app development for easier farmer access.

V. CONCLUSION

Machine learning models significantly improve ETo

prediction and irrigation management. The LSTM model
provides superior accuracy for time-series forecasting.
Deploying this system as a web-based application enables
farmers to optimize water usage, enhance crop yields and
promote sustainable agriculture. Future enhancements will
further refine predictions and usability.

IJISRT25FEB1206 www.ijisrt.com 2523