Integrated Nutrient Management for Lowland Rice: A PRISMA-Guided Meta-Analysis of Published Field Trials

Integrated Nutrient Management for Lowland Rice: A PRISMA-Guided Meta-Analysis of Published Field Trials is an M.Sc agronomy dissertation that asks a deceptively simple question and answers it with properly documented statistics: does integrated nutrient management (INM) — combining the state's inorganic fertilizer schedule with organic sources such as farmyard manure, green leaf manure or biofertilizers — produce a real, repeatable grain-yield advantage in lowland rice?

Field trials across India have reported INM effects for decades, but individual studies disagree. Reported gains range from negligible to double-digit percentages, trial designs and fertilizer doses vary, and no single small trial can settle the question. Instead of running yet another one-season experiment, this project treats the published literature itself as the dataset. Following the PRISMA 2020 framework, it searches, screens and codes published rice trials from Tamil Nadu and comparable Indian lowland systems, extracts grain-yield outcomes for INM versus inorganic-only treatments, and re-analyses them as effect sizes in a random-effects meta-analysis using the R package metafor.

The deliverable is more than one pooled number. The dissertation reports heterogeneity statistics (Q, I², τ²), pre-planned subgroup analyses by soil order, season and manure type, funnel-plot and Egger-test checks for publication bias, an adapted risk-of-bias assessment of every included trial, and a forest plot an examiner can interrogate line by line. Every step is scripted in R, so the complete analysis re-runs from the raw extraction sheet in a single command — which is exactly what makes a review defensible in a viva.

Farmers, extension officers and even syllabus writers repeat the claim that adding organic manures to the inorganic fertilizer schedule raises rice yields — but the strength of that claim is unknown to the people who repeat it. The supporting trials are scattered across decades of journals and theses, each with its own doses, soils and seasons; reported gains conflict; and many papers omit the variance data needed to judge how trustworthy each estimate is. A student or officer who picks one convincing paper is unknowingly cherry-picking, and the possibility that negative trials simply went unpublished is never checked.

There is a need for a transparent, reproducible synthesis that states the question in advance, searches systematically, applies inclusion criteria that are written down before any result is seen, pools the grain-yield evidence with statistics appropriate for heterogeneous field trials, tests whether the result depends on soil, season or manure type, and grades how confident one can be in the pooled answer. This project delivers exactly that synthesis for lowland rice, and in doing so demonstrates the review methodology that PG agricultural research increasingly demands.

Abstract

Integrated nutrient management is central to Indian fertilizer policy: the recommendation is to supplement the inorganic schedule for lowland rice with organic manures so that yields hold while soil health improves. The published evidence, however, is scattered across journals, doses, seasons and soil types, and individual trials often lack the replication to be conclusive. This dissertation conducts a PRISMA 2020-compliant systematic review and meta-analysis of published INM-versus-inorganic field trials in lowland rice, pooling grain-yield outcomes with a random-effects model and reporting the strength of evidence honestly, including its weaknesses.

Introduction

Rice is the staple crop of Tamil Nadu, grown through the Samba and shorter Navarai seasons under transplanted lowland conditions. The state's fertilizer recommendation for the crop is well established, and agricultural universities and KVKs promote INM combinations — farmyard manure, Sesbania green manure, Azospirillum and phosphobacteria biofertilizers, paddy straw compost — layered on that schedule. Students read claims in textbooks; the underlying trials live in journals, theses and ICAR institute bulletins, each reporting one site, one season and one dose.

Existing evidence and the gap

Three weaknesses recur in the literature. First, single trials report widely different effect sizes, and a reader who cites one favourable paper is cherry-picking without knowing it. Second, many reports omit variance information or report only CV, so later authors quote means without uncertainty. Third, small or non-significant results are less likely to be written up at all, so the visible literature is probably biased optimistic. No openly reproducible synthesis of these trials, written to current reporting standards, exists for a student to build on — most available syntheses are narrative and do not pool numbers.

Proposed study

• Protocol first: a written review protocol with pre-declared search strings, databases (Google Scholar, AGRIS, ICAR and TNAU repositories), inclusion criteria and extraction fields, registered as a file before screening begins.
• Screening: duplicate records removed; titles and abstracts screened in Rayyan; full texts assessed against criteria; every exclusion logged for the PRISMA flow diagram.
• Extraction: trial metadata (site, soil order, season, variety duration, doses) and outcomes (mean grain yield, variance, replication) for INM and inorganic-only arms, double-entered in Excel and cross-checked.
• Analysis: mean difference in grain yield (t/ha) per trial, pooled under a random-effects model in metafor; heterogeneity (Q, I², τ²); subgroups and meta-regression; funnel plots and Egger's test; risk-of-bias coding visualised with robvis; GRADE-style certainty statement.
• Deliverables: reproducible R scripts, PRISMA flow diagram, forest and funnel plots, and a dissertation structured to PG guidelines.

Feasibility

• Technical: the only instruments are a laptop, free software (R, Zotero, Rayyan) and library access; the statistics are standard and examinable.
• Economic: zero cost — no land, seed, chemicals or lab consumables are needed.
• Operational: a one-person, 20-week schedule with clear weekly milestones fits the Semester 4 project slot; the main risk (too few usable trials) is handled by widening the search to all-India lowland rice and pre-declaring that fallback in the protocol.

In scope

• Lowland (transplanted puddled) rice trials published in journals, ICAR institute bulletins, university theses and repository reports from Tamil Nadu and comparable Indian locations.
• The contrast of an integrated nutrient treatment (inorganic schedule + one or more organic sources) against the inorganic recommendation alone within the same trial.
• Grain yield as the primary outcome (t/ha at 14% moisture or converted); secondary descriptive notes on straw yield and soil organic carbon where reported.
• Statistical synthesis with variance-based weighting, subgroup and sensitivity analyses, all scripted in R.

Out of scope

• No new field trial: there is no land, seed or laboratory work; the dissertation is a secondary-data study by design.
• Economic optimisation of INM packages (cost–benefit analysis) — noted as future work.
• Crops other than rice, and rice systems without an inorganic-only control (e.g. fully organic comparisons).
• Re-analysis of raw plot-level data; the synthesis uses published group means and variances.
• Soil microbiology assays or greenhouse-gas measurements that some INM papers touch on.

Research design: a systematic review with quantitative meta-analysis — a secondary-data study governed by a written protocol, modelled on PRISMA 2020 and the Cochrane Handbook's general logic, adapted for agronomic field trials.

Inclusion criteria (pre-declared): replicated trials (≥3 replications) with an explicit INM-versus-inorganic-only contrast; grain yield reported with sufficient variance detail (SD, SE or CV) or recoverable from published statistics; any Tamil Nadu site or comparable Indian lowland location; any publication year but full bibliographic capture. Analysis plan: mean difference as the effect size; inverse-variance weights under a random-effects (DerSimonian–Laird and restricted-maximum-likelihood comparison) model; α = 0.05 fixed in the protocol; a trial with multiple INM arms contributes its arms jointly through a multilevel model so that the shared control is not double-counted as independent evidence.

The study is a pipeline with one-way data flow: whatever enters a stage is versioned, and every transformation downstream is scripted, so the whole chain re-runs from the extraction sheet.

flowchart TD
  A["Search strings run in Google Scholar / AGRIS / ICAR-TNAU repositories"] --> B["Zotero library: records + PDFs, one folder per phase"]
  B --> C["De-duplication and Rayyan title-abstract screening"]
  C --> D{"Full-text eligibility against pre-declared criteria"}
  D -->|excluded, reason logged| E["PRISMA 2020 flow diagram counts"]
  D -->|included| F["Double-entered Excel extraction sheet"]
  F --> G["R: unit harmonisation and SD recovery, QC checks"]
  G --> H["Effect sizes: mean difference in grain yield per trial"]
  H --> I["Random-effects meta-analysis in metafor"]
  I --> J["Heterogeneity: Q, I-squared, tau-squared"]
  I --> K["Subgroups + meta-regression: soil, season, manure, dose"]
  I --> L["Funnel plot + Egger test; robvis risk-of-bias plots"]
  J --> M["Forest plots, certainty table, dissertation figures"]
  K --> M
  L --> M

Three design decisions matter. First, the protocol precedes the data: criteria and thresholds are written and approved before screening, which is what separates a systematic review from a convenient one. Second, the extraction sheet is double-entered — two passes over every paper with a reconciliation column — because transcription errors, not statistics, are the most common defect in student meta-analyses. Third, R scripts, not point-and-click statistics, produce every number: the same command regenerates every table and figure after any correction, and the run log becomes an appendix. The multilevel modelling choice for multi-arm trials is visible in the diagram as a property of the analysis engine, not an afterthought, so dependence between arms sharing one control is handled where it belongs.

• Extend the synthesis to soil organic carbon and nutrient-use-efficiency outcomes, which most INM papers also report.
• Add a cost–benefit layer so the yield advantage can be discussed in rupees per hectare, using published input prices.
• Repeat the pipeline for pulse or millet crops — the scripts are crop-agnostic by design.
• Network meta-analysis comparing specific INM packages (FYM vs green manure vs biofertilizer) against each other, not only against the inorganic control.
• Deposit the final dataset and code in a public repository (Zenodo) with a DOI, as current reporting standards encourage.