Paulo H. da Silva Santos¹˒²˒⁶, João R. Vieira Manechini²˒⁶, Carlos Kantack², Mauro Xavier²˒⁶, Samira Domingues Carlin²˒⁶, Dilermando Perecin¹, Thiago G. Marconi⁴, Leandro G. Marconi⁴, Elisson Romanel³, Steve Jackson⁵, Marcos G.A. Landell²˒⁶, Luciana R. Pinto¹˒²˒⁶.

¹ São Paulo State University (FCAV/UNESP) ² Agronomic Institute of Campinas (IAC/APTA) Advanced Center for Sugarcane Development and Research ³ University of São Paulo (EEL-USP) ⁴ Digital Phenotyping - Votuporanga/SP ⁵ University of Warwick, UK ⁶*Center for Plant Molecular Breeding (CeM²P)-UNICAMP- Campinas, São Paulo, Brazil
*

Project Overview

• Data Collection: May 2023 – August 2024
• Dataset Compiled: 2025
• Location: Serra Grande, Bahia, Brazil (14°28'25.084" S, 39° 4'39.021" W)
• Experimental Design: The augmented block design consists of 13 blocks arranged in 60 metres of rows, encompassing a total of 195 plots.

This dataset integrates traditional field inspections with High-Throughput Phenotyping (HTP) using a UAV DJI MAVIC 3E to evaluate flowering time and intensity in sugarcane across two crop seasons: Plant Cane (2023) and First Ratoon (2024).

Data Description

1. Visual Field Assessments

Manual inspections conducted to determine the onset of reproductive stages:

• DTFL: Days to Flag Leaf.
• DTF: Days to Flowering.
• Stalk Count: Total culms at maturity.
• Flowering Intensity: Quantitative assessment of blooms per plot.

2. HTP Metrics (Drone-based)

Data derived from RGB sensors and orthomosaics (129 data points):

• Vegetation Index: Excess Greenness (ExG).
• Structural Metrics: Canopy Height Model (CC), Plant Height (PH), and Plant Volume (PV).
• Indirect Observation: Days to Flowering via orthomosaic database (DTF_Ortho).
• In the data files, the acronym "av" is used to denote 'average', and "sd" is used to denote 'standard deviation' with regard to the aforementioned metric.
• [!NOTE]
  Accessing Imagery: Raw imagery (~4GB per data point) is available upon request. Orthomosaic data can be explored at iacflor.fenotipagemdigital.com. Requests for additional information regarding sensitive details of the IAC sugarcane breeding program may be considered upon reasonable request.

File Dictionary

These files support analysis for ANOVA, Heritability, Repeatability, Feature Selection, and Machine Learning (Classification/Regression).

Variable Abbreviations

Dataset Description: IAC Flowering & HTP Data

This dataset contains ground truth observations and High-Throughput Phenotyping (HTP) data for sugarcane flowering trials. The experiment follows an Augmented Block Design.

1. Experimental Design & Ground Truth

• File: IAC_Flowering_Field_Data.csv
• Structure: This file contains the raw ground truth data.
  • Block: Refers to the 13 experimental blocks.
  • Treatment: Denotes the genotypes tested.
  • Checks: Three specific genotypes (125-IAC0752, 127-IAC07395, and 128-IACSP018158) were replicated across all 13 blocks to facilitate the augmented block design. This structure applies to both field and HTP data.

2. Flowering Counts (Visual vs. CNN)

• File: IAC_FlowerigCount_byCNN.csv
• Details: This file compares manual observations with deep learning outputs.
  • Date Format: Day/Month/Year (DD/MM/YYYY).
  • Visual: Manual flower counts performed in the field (Ground Truth).
  • CNN: Automated flower counts generated by a Convolutional Neural Network (Deep Learning model).

3. HTP Metrics (CC, PH, PV, and ExG)

• Files: Raw_CC_Data.csv, Raw_PH_Data.csv, Raw_PV_Data.csv, and Raw_EXG_Data.csv.
• Column Guide:
  • Ignored Columns: id, Experiment, and vi (internal database identifiers).
  • row: Equivalent to the Block in the augmented design (1–13).
  • col: Refers to the 15 plots/genotypes per block.
  • type: The flowering response classification (EF = Early Flowering; LF = Late Flowering).
  • Metric Columns: Named by the metric acronym followed by the flight date in YYMMDD format (e.g., PH230825 = Plant Height measured on August 25, 2023).

4. Vegetation Indices & Metric Definitions

All metrics were derived from RGB (Red, Green, Blue) orthomosaics and digital surface/terrain models:

Observation: To replicate the orthomosaic processing settings and ensure consistent results, please refer to the file Orthomosaic_Metashape_Reports.zip. This archive contains a detailed PDF report for every flight conducted during the experiment, providing the full software configuration used in Agisoft Metashape.

Code Availability

All statistical analyses and machine learning scripts used in this study are available at:

https://github.com/pahenrique02/Sugarcane-Flower-Paper-TPPJ