ICICLE-AI/yield-estimation

Yield Estimation Transformer

A Hugging Face custom pipeline using Transformers model for county-level corn yield estimation using multi-temporal weather observations and static soil properties.

The model combines weekly weather time-series with static soil features to estimate corn yield in bushels per acre (bu/acre). It is packaged for inference using Hugging Face Transformers and has been tested for deployment through FlexServ.

The Hugging Face text-classification task is used as the FlexServ-compatible serving interface. The underlying model performs regression, and the returned score represents predicted corn yield in bu/acre.

The training_code directory composes the source code for data preparation, model training, evaluation, and inference.

Tags

Tags: Digital-Agriculture, Foundation-AI

• In-season Crop Yield Estimation
• Digital Agriculture
• Transformers
• Multi-Temporal Modeling
• Regression
• Hugging Face Transformers
• FlexServ

For guidance on what to include in Tutorials, How-To Guides, Explanation, and Reference, see Diátaxis.

License

This project is released under the MIT License. The full text is available in LICENSE.

References

USA County Level Crop Yield Dataset

This model uses the USA County Level Crop Yield Dataset.

@article{Khaki2020CNNRNN,
  author    = {Khaki, Saeed and Wang, Liang and Archontoulis, Sotirios V.},
  title     = {A CNN-RNN Framework for Crop Yield Prediction},
  journal   = {Frontiers in Plant Science},
  volume    = {10},
  pages     = {1750},
  year      = {2020},
  doi       = {10.3389/fpls.2019.01750},
  publisher = {Frontiers Media SA}
}

FlexServ

The model is packaged and validated for deployment with FlexServ.

FlexServ documentation: https://zhangwei217245.github.io/FlexServ/

Acknowledgements

This work was developed as part of the ICICLE AI Institute.

National Science Foundation (NSF) funded AI institute for Intelligent Cyberinfrastructure with Computational Learning in the Environment (ICICLE) (OAC 2112606)

Issue reporting

Contact:

For questions or support:

Sarikaa Sridhar: sridhar.86@buckeyemail.osu.edu

---

Tutorials

Overview

The Yield Estimation Transformer is a pretrained model for in-season county-level corn yield estimation. It combines temporal weekly weather observations with static soil properties and produces a scalar yield prediction in bushels per acre.

The model accepts six weekly weather variables:

• prcp
• srad
• swe
• tmax
• tmin
• vp

It also uses 66 static soil features defined in config.json.

The model supports prediction cutoffs at:

20, 24, 28, 32, 36, 40, 44, 48, 52

A cutoff determines how many weeks of weather information are available to the model. A cutoff of 52 represents full-season inference.

For deployment through FlexServ, the model uses the Hugging Face text-classification pipeline as its serving interface. This is an interface choice for inference compatibility; the underlying prediction task remains regression.

---

How-To Guides

Problem Description

The model estimates county-level corn yield from weather and soil information.

The pretrained architecture expects structured numerical inputs rather than natural-language text. To make the model deployable through FlexServ's supported pipeline tasks, the model is exposed through the Hugging Face text-classification interface.

The structured yield input is serialized as a JSON string. The custom tokenizer parses this string and converts the weather, soil, crop, and cutoff information into the tensors expected by the pretrained model.

The resulting inference path is:

JSON-formatted input string
        ↓
YieldTokenizer
        ↓
weather + soil + crop + cutoff tensors
        ↓
Yield Estimation Transformer
        ↓
scalar yield prediction
        ↓
YIELD_BU_ACRE score

The score returned by the pipeline is therefore a yield estimate in bu/acre, not a classification probability.

Getting Started

The repository contains the files required for standalone Hugging Face and FlexServ inference:

.
├── README.md
├── LICENSE
├── component-info.yaml
├── training_code/
├── config.json
├── configuration_yield.py
├── model.safetensors
├── modeling_yield.py
├── requirements.txt
├── sample_input_weekly.json
├── tokenization_yield.py
├── tokenizer_config.json
└── yield_transformer.py

A complete inference example is provided in:

sample_input_weekly.json

Because the repository provides custom model configuration, tokenizer, and architecture code, Hugging Face loading requires:

trust_remote_code=True

Usage

Local Hugging Face Inference

Load the model through the Hugging Face text-classification pipeline:

import json
from transformers import pipeline

pipe = pipeline(
    "text-classification",
    model="ICICLE-AI/yield-estimation",
    tokenizer="ICICLE-AI/yield-estimation",
    trust_remote_code=True,
)

with open("sample_input_weekly.json") as f:
    sample = json.load(f)

prediction = pipe(json.dumps(sample))

print(prediction)

Example output:

[
    {
        "label": "YIELD_BU_ACRE",
        "score": 165.1769561767578
    }
]

The score is the predicted corn yield in bushels per acre.

Input Format

The structured input contains:

{
  "crop": "corn",
  "weather_format": "weekly",
  "cutoff": 52,
  "weather": {
    "prcp": ["52 weekly values"],
    "srad": ["52 weekly values"],
    "swe": ["52 weekly values"],
    "tmax": ["52 weekly values"],
    "tmin": ["52 weekly values"],
    "vp": ["52 weekly values"]
  },
  "soil": {
    "bdod_mean_0-5cm": 0.0,
    "...": "remaining soil features"
  }
}

The complete set of 66 soil variables and their expected ordering are stored in config.json.

The tokenizer:

1. parses the JSON-formatted string,
2. validates the expected input fields,
3. constructs the weather, soil, crop, and cutoff tensors.

The Hugging Face pipeline then passes these tensors to the pretrained model for inference.

FlexServ Inference

The model has been tested for inference through FlexServ using:

Task: text-classification
Model: ICICLE-AI/yield-estimation

FlexServ's inputs field expects a string. Therefore, the structured yield input must be supplied as a JSON-formatted string, rather than directly as a nested JSON object.

Conceptually, a FlexServ request has the following form:

{
  "task": "text-classification",
  "inputs": "{\"crop\":\"corn\",\"weather_format\":\"weekly\",\"cutoff\":52,\"weather\":{...},\"soil\":{...}}",
  "parameters": {},
  "model": "ICICLE-AI/yield-estimation"
}

A successful response has the form:

[
  {
    "label": "YIELD_BU_ACRE",
    "score": 165.1769561767578
  }
]

The returned score is the estimated yield in bu/acre.

Validation

The packaged model can be validated locally against the included sample:

python - <<'PY'
import json
from transformers import pipeline

with open("sample_input_weekly.json") as f:
    sample = json.load(f)

pipe = pipeline(
    "text-classification",
    model=".",
    tokenizer=".",
    trust_remote_code=True,
)

print(pipe(json.dumps(sample)))
PY

Expected output for the included sample is approximately:

[{'label': 'YIELD_BU_ACRE', 'score': 165.1769561767578}]

Training own model

Installation

Clone the model repository:

git clone https://huggingface.co/ICICLE-AI/yield-estimation
cd yield-estimation/training_code

Create and activate a Python environment:

conda create -n yield_hf python=3.10
conda activate yield_hf

Install the required dependencies:

pip install -r requirements.txt

Prerequisites

• Python 3.10+
• PyTorch
• Hugging Face Transformers
• Dependencies listed in requirements.txt

For GPU training, a CUDA-compatible PyTorch environment is recommended.

The included SLURM script can be used on a compatible HPC system.

Problem Description

The objective is to estimate county-level corn yield from weather and soil information.

Each sample contains two primary sources of information:

1. multi-temporal weather observations;
2. static soil properties.

The weather input is represented as:

[K, W]

where:

• K is the number of temporal observations;
• W = 6 is the number of weather variables.

The static soil input is represented as:

[S]

where:

S = 66

The model learns a regression function of the form:

weather + soil + crop information
                ↓
        transformer model
                ↓
       predicted corn yield

The predicted value is expressed in bushels per acre (bu/acre).

The model supports multi-temporal estimation by evaluating the weather sequence at different seasonal cutoffs.

Repository Structure

The final repository is organized as follows:

.
├── README.md
├── requirements.txt
├── training.slurm
│
├── checkpoints/
│   ├── config.json
│   ├── metrics.json
│   └── model.safetensors
│
├── config/
│   ├── __init__.py
│   └── config.py
│
├── data/
│   ├── __init__.py
│   ├── dataset.py
│   └── preprocessing.py
│
├── examples/
│   └── sample_input_weekly.json
│
├── hf/
│   ├── __init__.py
│   ├── auto.py
│   ├── configuration_yield.py
│   └── modeling_yield.py
│
├── models/
│   ├── __init__.py
│   └── unimodal_ws_crossattn.py
│
├── scripts/
│   ├── __init__.py
│   ├── prepare_cornbelt.py
│   ├── train_hf.py
│   ├── evaluate_hf.py
│   └── inference_hf.py
│
└── training/
    ├── __init__.py
    └── engine.py

The major components are:

• data/ — dataset loading and preprocessing
• models/ — core neural network architecture
• training/ — training and evaluation utilities
• hf/ — Hugging Face AutoClass-compatible regression wrapper used by the training repository
• scripts/ — data preparation, training, evaluation, and inference entry points
• checkpoints/ — final trained checkpoint and configuration
• examples/ — example structured model input
• training.slurm — example HPC training job

Data Preparation

The USA County Level Crop Yield public dataset is used for training, validation and testing. The data preparation workflow is implemented in:

scripts/prepare_cornbelt.py

After preparation, the expected dataset structure is:

data/
└── cornbelt/
    ├── train.h5
    ├── val.h5
    └── test.h5

The model uses the following six weather variables:

prcp
srad
swe
tmax
tmin
vp

The 66 soil variables used by the final checkpoint are recorded in the model configuration.

Training

The primary training entry point is:

scripts/train_hf.py

The final model uses multi-cutoff training with:

20,24,28,32,36,40,44,48,52

An example training command is:

python scripts/train_hf.py \
  --train_file data/cornbelt/train.h5 \
  --val_file data/cornbelt/val.h5 \
  --test_file data/cornbelt/test.h5 \
  --weather_vars prcp,srad,swe,tmax,tmin,vp \
  --soil_vars bdod_mean_0-5cm,bdod_mean_5-15cm,bdod_mean_15-30cm,bdod_mean_30-60cm,bdod_mean_60-100cm,bdod_mean_100-200cm,cec_mean_0-5cm,cec_mean_5-15cm,cec_mean_15-30cm,cec_mean_30-60cm,cec_mean_60-100cm,cec_mean_100-200cm,cfvo_mean_0-5cm,cfvo_mean_5-15cm,cfvo_mean_15-30cm,cfvo_mean_30-60cm,cfvo_mean_60-100cm,cfvo_mean_100-200cm,clay_mean_0-5cm,clay_mean_5-15cm,clay_mean_15-30cm,clay_mean_30-60cm,clay_mean_60-100cm,clay_mean_100-200cm,nitrogen_mean_0-5cm,nitrogen_mean_5-15cm,nitrogen_mean_15-30cm,nitrogen_mean_30-60cm,nitrogen_mean_60-100cm,nitrogen_mean_100-200cm,ocd_mean_0-5cm,ocd_mean_5-15cm,ocd_mean_15-30cm,ocd_mean_30-60cm,ocd_mean_60-100cm,ocd_mean_100-200cm,ocs_mean_0-5cm,ocs_mean_5-15cm,ocs_mean_15-30cm,ocs_mean_30-60cm,ocs_mean_60-100cm,ocs_mean_100-200cm,phh2o_mean_0-5cm,phh2o_mean_5-15cm,phh2o_mean_15-30cm,phh2o_mean_30-60cm,phh2o_mean_60-100cm,phh2o_mean_100-200cm,sand_mean_0-5cm,sand_mean_5-15cm,sand_mean_15-30cm,sand_mean_30-60cm,sand_mean_60-100cm,sand_mean_100-200cm,silt_mean_0-5cm,silt_mean_5-15cm,silt_mean_15-30cm,silt_mean_30-60cm,silt_mean_60-100cm,silt_mean_100-200cm,soc_mean_0-5cm,soc_mean_5-15cm,soc_mean_15-30cm,soc_mean_30-60cm,soc_mean_60-100cm,soc_mean_100-200cm \
  --crop corn \
  --time_agg weekly \
  --train_cutoffs 20,24,28,32,36,40,44,48,52 \
  --eval_cutoffs 20,24,28,32,36,40,44,48,52 \
  --epochs 30 \
  --lr 3e-5 \
  --batch_size 32 \
  --out_dir checkpoints

The final checkpoint is stored in:

checkpoints/

The checkpoint includes:

config.json
model.safetensors
metrics.json

Training on Your Own Data

The training pipeline can also be used to train a new yield estimation model on a compatible dataset.

Prepare the dataset in the HDF5 format expected by YieldDataset and provide separate training, validation, and test files.

The weather and soil variables supplied to the training command must correspond to the variables available in the prepared dataset.

A general training command is:

python scripts/train_hf.py \
  --train_file <path/to/train.h5> \
  --val_file <path/to/val.h5> \
  --test_file <path/to/test.h5> \
  --weather_vars <comma-separated-weather-variables> \
  --soil_vars <comma-separated-soil-variables> \
  --crop <crop-name> \
  --time_agg weekly \
  --train_cutoffs <comma-separated-training-cutoffs> \
  --eval_cutoffs <comma-separated-evaluation-cutoffs> \
  --epochs <number-of-epochs> \
  --lr <learning-rate> \
  --batch_size <batch-size> \
  --out_dir <output-directory>

SLURM Training

An example SLURM job is provided in:

training.slurm

Submit it using:

sbatch training.slurm

Evaluation

The trained checkpoint can be evaluated using:

scripts/evaluate_hf.py

For the final multi-cutoff model:

python scripts/evaluate_hf.py \
  --hf_model_dir checkpoints \
  --test_file data/cornbelt/test.h5 \
  --cutoffs 20,24,28,32,36,40,44,48,52 \
  --batch_size 64 \
  --output_csv checkpoints/test_predictions.csv \
  --metrics_json checkpoints/test_metrics.json

Evaluation is performed independently at the configured seasonal cutoffs.

The evaluation process:

1. loads the final trained checkpoint;
2. loads the test dataset;
3. applies the normalization statistics stored in the checkpoint configuration;
4. performs inference at the requested cutoffs;
5. computes evaluation metrics;
6. save predictions and metrics to disk.

Inference

Inference using the trained checkpoint is implemented in:

scripts/inference_hf.py

An example structured input is provided in:

examples/sample_input_weekly.json

The sample follows the general structure:

{
  "crop": "corn",
  "weather_format": "weekly",
  "cutoff": 52,
  "weather": {
    "prcp": [],
    "srad": [],
    "swe": [],
    "tmax": [],
    "tmin": [],
    "vp": []
  },
  "soil": {
    "bdod_mean_0-5cm": 0.0
  }
}

The complete sample file contains the required weather sequence and soil variables.

Run single-sample inference with:

python scripts/inference_hf.py \
  --hf_model_dir checkpoints \
  --single_sample_json examples/sample_input_weekly.json \
  --cutoff 52 \
  --output_csv inference_prediction.csv

The output contains the predicted yield for the requested cutoff.

For example:

sample_idx,cutoff,y_pred
0,52,<predicted_yield>

---

Explanation

Features

• Transformer-Based Yield Estimation: Uses a transformer architecture to model temporal weather information for corn yield prediction.
• Weather and Soil Integration: Combines six weekly weather variables with 66 static soil properties.
• Multi-Temporal Inference: Supports yield estimation at multiple seasonal cutoffs from week 20 through week 52.
• Automatic Preprocessing: The custom tokenizer converts JSON-formatted structured inputs into the tensors expected by the pretrained model.
• Automatic Normalization: Weather and soil features are normalized using statistics stored with the model configuration.
• Regression Output: Produces a scalar corn yield estimate in bushels per acre.
• Hugging Face Integration: Uses the standard Transformers pipeline interface with repository-provided model and tokenizer code.
• FlexServ Deployment: Uses the supported text-classification task to expose the regression model as a FlexServ inference service.
• CPU and GPU Support: Supports PyTorch inference on CPU and compatible CUDA GPUs.
• Safetensors Weights: Model weights are distributed using the Safetensors format.