📄 agriculture

Time-Series Accumulation Forecasting of Organophosphate Pesticides in Agricultural Soils and Vegetables of Delta Central District, Nigeria

This study develops a hybrid ARIMA-LSTM time-series model to forecast the accumulation of organophosphate pesticides in Delta Central District, Nigeria, demonstrating that while business-as-usual practices could lead to a 93% increase in soil residues by 2035, a 50% reduction in pesticide use starting in 2025 would significantly mitigate this risk.

Original authors: Ibigoni C. Howard, Kngsley E. Okpara, Chioma C. Howard

Published 2026-07-07

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Ibigoni C. Howard, Kngsley E. Okpara, Chioma C. Howard

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the soil in the Delta Central District of Nigeria as a giant, slow-moving bathtub. Every year, farmers pour in a bucket of "pesticide water" to protect their crops. Usually, the drain (nature's ability to break down chemicals) removes some of it, but not all. Over time, the water level in the tub rises.

This research paper is like a weather forecast for that bathtub, but instead of rain, it's predicting how much pesticide will build up in the soil and vegetables over the next decade.

Here is the story of the study, broken down simply:

1. The Problem: We Only Have a Snapshot

Until now, scientists have mostly taken "snapshots" of the soil. They check how dirty the water is today, but they can't tell you if the tub will overflow next year or in ten years. They also couldn't easily say, "If we stop pouring 50% of the water, how fast will the tub empty?"

The researchers wanted to build a time machine (a forecasting model) to see the future of these chemicals, specifically 14 different types of organophosphate pesticides (OPPs).

2. The Ingredients: Mixing Old Data with New Math

To build their time machine, the team needed a long history of data, but they only had a detailed measurement from 2023.

The Anchor: They used the 2023 data as a solid anchor point.
The Stretch: To fill in the years before 2023 (from 2000 to 2022), they used a "recipe." They looked at how much pesticide Nigeria bought and used overall (from FAO records) and scaled it up or down to match the 2023 measurements. It's like knowing the size of a cake today and using the flour sales from the last 20 years to guess how big the cakes were back then.

3. The Tools: The "Smart" Forecasters

The researchers tried several different "crystal balls" to predict the future:

The Linear Thinker (ARIMA): This model assumes the future looks just like the past, just a little bigger. It's good at straight lines but bad at curves.
The Group Thinker (VAR): This model looks at how the amount of pesticide bought relates to the amount found in the soil.
The Deep Learner (LSTM): This is an Artificial Intelligence (AI) brain. It's great at remembering long-term patterns and spotting complex, wiggly curves that the other models miss.
The Super-Team (Hybrid ARIMA-LSTM): This was the winner. They combined the Linear Thinker and the Deep Learner. The Linear Thinker handled the steady, predictable rise, while the AI handled the messy, unpredictable bumps.

The Result: The Super-Team was the most accurate. It predicted the future with very little error (only about 5% off), beating all the other models.

4. The Crystal Ball: What Happens by 2035?

The team ran four different "what-if" scenarios to see what the bathtub would look like in 2035:

Scenario A: Business as Usual (The "Keep Pouring" Plan)
If farmers keep using pesticides at the current growing rate, the soil will get 93% dirtier by 2035. The chemical level will jump from 31 ng/g to nearly 60 ng/g. By 2031, it will likely cross a safety line set by the European Union (50 ng/g), meaning the soil is becoming dangerously toxic. One specific chemical, Pirimiphos-methyl, is the main culprit, acting like a heavy stone that sinks to the bottom and stays there.
Scenario B: The 30% Cut (The "Lighten Up" Plan)
If they reduce usage by 30%, the water level stops rising as fast, but it doesn't go down. The soil stays dirty, just slightly less so. It's like turning the faucet down a bit, but the tub is still filling up.
Scenario C: The 50% Cut (The "Turn the Tap" Plan)
If they cut usage by 50%, the magic happens. The drain finally beats the faucet. The soil starts to clean itself, dropping from 31 ng/g down to 19.9 ng/g. This is the minimum effort needed to actually reverse the pollution.
Scenario D: The Phase-Out (The "Stop Pouring" Plan)
If they stop using these pesticides entirely by 2035, the soil cleans up almost completely, returning to near-natural levels (5.6 ng/g). However, the "heavy stone" chemical (Pirimiphos-methyl) takes the longest to wash away because it sticks to the soil longer than the others.

5. The Takeaway

This study proves that we don't have to wait for a disaster to happen. By using a smart mix of old math and modern AI, we can predict exactly how much pollution will build up.

The main lesson is clear: Small cuts in pesticide use won't fix the problem; only big cuts (50% or more) will actually clean the soil. If we do nothing, the soil in this region of Nigeria will become significantly more toxic by the mid-2030s.

The researchers say this method can be copied for other farming areas in Africa to help governments make smarter decisions before the "bathtub" overflows.

Problem Statement

Current environmental monitoring of organophosphate pesticides (OPPs) in agricultural systems, particularly in sub-Saharan Africa, relies almost exclusively on cross-sectional assessments. These studies provide a snapshot of contamination status but lack the capacity to forecast future soil and vegetable residue trajectories under varying agricultural management scenarios. In the Delta Central District of Nigeria, where pesticide use is near year-round and multiple harvests occur annually, surface soil concentrations of persistent OPPs are expected to increase over time. However, without longitudinal prediction models, policymakers cannot quantify the risk reduction achievable through regulatory intervention or determine what concentrations will prevail in future years (e.g., 2030 or 2035) under current agricultural trajectories.

Methodology

The study developed a time-series accumulation forecasting framework for 14 specific OPP compounds in the Delta Central District. The methodology involved four key stages:

Data Reconstruction and Augmentation:
- Baseline: The study anchored its model on the 2023 measured baseline data from Tesi et al. (2025), which quantified 14 OPPs in soils and vegetables across seven Local Government Areas (LGAs). The district mean total soil concentration was 31.0 ng/g.
- Historical Synthesis: To meet the minimum data requirements for robust time-series modeling (typically 8–12 annual points), the authors synthesized a historical dataset from 2000 to 2022. This was achieved by scaling the FAO (2024) Nigeria pesticide use index against the 2023 baseline proportions, assuming relative compound composition remained constant over the period.
Model Development and Comparison:
The study systematically compared the out-of-sample predictive performance of six modeling approaches on the reconstructed time series (2000–2022), using a hold-out test set from 2020–2022:
- ARIMA/SARIMA: Univariate models capturing linear autoregressive processes.
- VAR (Vector Autoregression): A bivariate model incorporating the FAO pesticide use index as an exogenous variable to test for Granger causality.
- LSTM (Long Short-Term Memory): Both univariate and multivariate neural networks designed to capture non-linear dependencies and long-term memory.
- Hybrid ARIMA-LSTM: A two-stage architecture where an ARIMA model first captures the linear trend, and an LSTM network models the non-linear residuals.
Validation and Metrics:
Model performance was evaluated using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and R-squared ( $R^2$ ). Pairwise comparisons utilized the Diebold–Mariano (DM) test to assess statistical significance in predictive accuracy.
Scenario Analysis:
The optimal model was used to project soil OPP levels to 2035 under four regulatory scenarios:
- Business-As-Usual (BAU): Continuation of the historical 4.7% annual growth rate in pesticide use.
- 30% Restriction: A 30% reduction in the use index starting in 2025.
- 50% Restriction: A 50% reduction starting in 2025.
- Accelerated Phase-Out: Linear reduction of the use index to zero by 2035.

Key Results

Model Performance: The Hybrid ARIMA-LSTM model demonstrated superior predictive accuracy compared to all other approaches. It achieved an RMSE of 1.52, MAPE of 5.1%, and an $R^2$ of 0.87 for total soil OPPs. This significantly outperformed the standalone SARIMA model (RMSE 2.31, $R^2$ 0.72) and the multivariate LSTM (RMSE 1.67, $R^2$ 0.84).
Causality and Trends: VAR analysis confirmed a unidirectional Granger causality from the national pesticide use index to soil accumulation ( $p < 0.001$ ). The reconstructed series showed a mean annual growth rate of 5.1% for total soil OPPs between 2000 and 2022.
Scenario Projections (2035):
- BAU: Under current trends, total soil OPP levels are projected to reach 59.7 ng/g by 2035 (a 93% increase over the 2023 baseline). The 95% confidence interval suggests a >70% probability that median-contaminated sites will exceed the European Commission's indicative threshold of 50 ng/g by approximately 2031.
- 30% Restriction: This scenario slows accumulation but does not reverse it, with levels reaching ~32.8 ng/g by 2035.
- 50% Restriction: This scenario achieves net decontamination, reducing levels to 19.9 ng/g by 2035.
- Phase-Out: Full elimination of use reduces levels to near-background concentrations of 5.6 ng/g by 2035.
Compound Specificity: Pirimiphos-methyl was identified as the dominant compound (77% of total load in 2023) and the primary driver of future threshold exceedances due to its high application frequency and soil persistence.

Significance and Claims

The paper claims four principal contributions to the field of environmental forecasting in developing regions:

First Temporal Framework: It establishes the first temporal accumulation forecasting framework for OPPs in Nigerian agricultural soils, bridging the gap between static baseline monitoring and dynamic future projections.
Methodological Validation: It demonstrates that a hybrid ARIMA-LSTM approach provides significantly better predictive performance than linear or single-neural-network models for soil contaminant accumulation, validating its utility for non-linear, long-memory environmental processes.
Policy Quantification: The study provides a quantitative basis for regulatory decision-making, specifically identifying that a 30% reduction in pesticide use is insufficient to reverse accumulation, while a 50% reduction is the minimum threshold required for net decontamination.
Generalizability: The authors posit that this hybrid modeling approach is transferable to other chronically contaminated agroecosystems in sub-Saharan Africa, offering a replicable template for emerging contaminant forecasting in data-scarce tropical environments.

The study concludes that while historical data reconstruction introduces some uncertainty, the hybrid framework successfully translates national market trends into localized soil accumulation forecasts, offering a critical tool for proactive agricultural policy management in the Niger Delta.