Patrick Filippi

Dr. Patrick Filippi

The University of Sydney
orcid

Research Focus & Metrics

Academic Career Length

11Years (2016-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
8 January 2019 in Precision Agriculture
Verified
Journal Article
Closed access
PDF file
PFPatrick Filippi
Patrick Filippi
EJEdward J. Jones
Edward J. Jones
NWNiranjan S. Wimalathunge
Niranjan S. Wimalathunge
PSPallegedara D. S. N. Somarathna
Pallegedara D. S. N. Somarathna
LPLiana Pozza
Liana Pozza
TBT.F.A. Bishop
T.F.A. Bishop

7 December 2024 in Precision Agriculture
Verified
Journal Article
Open access
PDF file
PFPatrick Filippi
Patrick Filippi
SHSi Yang Han
Si Yang Han
TBThomas F.A. Bishop
Thomas F.A. Bishop
There has been a recent surge in the number of studies that aim to model crop yield using data-driven approaches. This has largely come about due to the increasing amounts of remote sensing (e.g. satellite imagery) and precision agriculture data available (e.g. high-resolution crop yield monitor data), as well as the abundance of machine learning modelling approaches. However, there are several common issues in published studies in the field of precision agriculture (PA) that must be addressed. This includes the terminology used in relation to crop yield modelling, predicting, forecasting, and interpolating, as well as the way that models are calibrated and validated. As a typical example, many studies will take a crop yield map or several plots within a field from a single season, build a model with satellite or Unmanned Aerial Vehicle (UAV) imagery, validate using data-splitting or some kind of cross-validation (e.g. k-fold), and say that it is a ‘prediction’ or ‘forecast’ of crop yield. However, this poses a problem as the approach is not testing the forecasting ability of the model, as it is built on the same season that it is then validating with, thus giving a substantial overestimation of the value for decision-making, such as an application of fertiliser in-season. This is an all-too-common flaw in the logic construct of many published studies. Moving forward, it is essential that clear definitions and guidelines for data-driven yield modelling and validation are outlined so that there is a greater connection between the goal of the study, and the actual study outputs/outcomes. To demonstrate this, the current study uses a case study dataset from a collection of large neighbouring farms in New South Wales, Australia. The dataset includes 160 yield maps of winter wheat (Triticum aestivum) covering 26,400 hectares over a 10-year period (2014–2023). Machine learning crop yield models are built at 30 m spatial resolution with a suite of predictor data layers that relate to crop yield. This includes datasets that represent soil variation, terrain, weather, and satellite imagery of the crop. Predictions are made at both the within-field (30 m), and field resolution. Crop yield predictions are useful for an array of applications, so four different experiments were set up to reflect different scenarios. This included Experiment 1: forecasting yield mid-season (e.g. for mid-season fertilisation), Experiment 2: forecasting yield late-season (e.g. for late-season logistics/forward selling), Experiment 3: predicting yield in a previous season for a field with no yield data in a season, and Experiment 4: predicting yield in a previous season for a field with some yield data (e.g. two combine harvesters, but only one was fitted with a yield monitor). This study showcases how different model calibration and validation approaches clearly impact prediction quality, and therefore how they should be interpreted in data-driven crop yield modelling studies. This is key for ensuring that the wealth of data-driven crop yield modelling studies not only contribute to the science, but also deliver actual value to growers, industry, and governments.

7 April 2021 in Remote Sensing
Verified
Journal Article
Open access
PDF file
IMIan Marang
Ian Marang
PFPatrick Filippi
Patrick Filippi
Tim WeaverTim Weaver
Tim Weaver
BEBradley J. Evans
Bradley J. Evans
BWBrett M. Whelan
Brett M. Whelan
GRGuy Roth
Guy Roth
Hyperspectral imaging spectrometers mounted on unmanned aerial vehicle (UAV) can capture high spatial and spectral resolution to provide cotton crop nitrogen status for precision agriculture. The aim of this research was to explore machine learning use with hyperspectral datacubes over agricultural fields. Hyperspectral imagery was collected over a mature cotton crop, which had high spatial (~5.2 cm) and spectral (5 nm) resolution over the spectral range 475–925 nm that allowed discrimination of individual crop rows and field features as well as a continuous spectral range for calculating derivative spectra. The nominal reflectance and its derivatives clearly highlighted the different treatment blocks and were strongly related to N concentration in leaf and petiole samples, both in traditional vegetation indices (e.g., Vogelman 1, R2 = 0.8) and novel combinations of spectra (R2 = 0.85). The key hyperspectral bands identified were at the red-edge inflection point (695–715 nm). Satellite multispectral was compared against the UAV hyperspectral remote sensing’s performance by testing the ability of Sentinel MSI to predict N concentration using the bands in VIS-NIR spectral region. The Sentinel 2A Green band (B3; mid-point 559.8 nm) explained the same amount of variation in N as the hyperspectral data and more than the Sentinel Red Edge Point 1 (B5; mid-point 704.9 nm) with the lower 10 m resolution Green band reporting an R2 = 0.85, compared with the R2 = 0.78 of downscaled Sentinel Red Edge Point 1 at 5 m. The remaining Sentinel bands explained much lower variation (maximum was NIR at R2 = 0.48). Investigation of the red edge peak region in the first derivative showed strong promise with RIDAmid (R2 = 0.81) being the best index. The machine learning approach narrowed the range of bands required to investigate plant condition over this trial site, greatly improved processing time and reduced processing complexity. While Sentinel performed well in this comparison and would be useful in a broadacre crop production context, the impact of pixel boundaries relative to a region of interest and coarse spatial and temporal resolution impacts its utility in a research capacity.

21 May 2019 in Agronomy
Verified
Journal Article
Open access
PDF file
PFPatrick Filippi
Patrick Filippi
EJEdward Jones
Edward Jones
BGBradley J. Ginns
Bradley J. Ginns
BWBrett M. Whelan
Brett M. Whelan
GRGuy Roth
Guy Roth
TBT.F.A. Bishop
T.F.A. Bishop
Subsoil alkalinity is a common issue in the alluvial cotton-growing valleys of northern New South Wales (NSW), Australia. Soil alkalinity can cause nutrient deficiencies and toxic effects, and inhibit rooting depth, which can have a detrimental impact on crop production. The depth at which a soil constraint is reached is important information for land managers, but it is difficult to measure or predict spatially. This study predicted the depth in which a pH (H2O) constraint (>9) was reached to a 1-cm vertical resolution to a 100-cm depth, on a 1070-hectare dryland cropping farm. Equal-area quadratic smoothing splines were used to resample vertical soil profile data, and a random forest (RF) model was used to produce the depth-to-soil pH constraint map. The RF model was accurate, with a Lin’s Concordance Correlation Coefficient (LCCC) of 0.63–0.66, and a Root Mean Square Error (RMSE) of 0.47–0.51 when testing with leave-one-site-out cross-validation. Approximately 77% of the farm was found to be constrained by a strongly alkaline pH greater than 9 (H2O) somewhere within the top 100 cm of the soil profile. The relationship between the predicted depth-to-soil pH constraint map and cotton and grain (wheat, canola, and chickpea) yield monitor data was analyzed for individual fields. Results showed that yield increased when a soil pH constraint was deeper in the profile, with a good relationship for wheat, canola, and chickpea, and a weaker relationship for cotton. The overall results from this study suggest that the modelling approach is valuable in identifying the depth-to-soil pH constraint, and could be adopted for other important subsoil constraints, such as sodicity. The outputs are also a promising opportunity to understand crop yield variability, which could lead to improvements in management practices.

17 September 2020 in Remote Sensing
Verified
Journal Article
Open access
PDF file
DADhahi Al-Shammari
Dhahi Al-Shammari
Ignacio FuentesIgnacio Fuentes
Ignacio Fuentes
BWBrett M. Whelan
Brett M. Whelan
PFPatrick Filippi
Patrick Filippi
TBT.F.A. Bishop
T.F.A. Bishop
A phenology-based crop type mapping approach was carried out to map cotton fields throughout the cotton-growing areas of eastern Australia. The workflow was implemented in the Google Earth Engine (GEE) platform, as it is time efficient and does not require processing in multiple platforms to complete the classification steps. A time series of Normalised Difference Vegetation Index (NDVI) imagery were generated from Landsat 8 Surface Reflectance Tier 1 (L8SR) and processed using Fourier transformation. This was used to produce the harmonised-NDVI (H-NDVI) from the original NDVI, and then phase and amplitude values were generated from the H-NDVI to visualise active cotton in the targeted fields. Random Forest (RF) models were built to classify cotton at early, mid and late growth stages to assess the ability of the model to classify cotton as the season progresses, with phase, amplitude and other individual bands as predictors. Results obtained from leave-one-season-out cross validation (LOSOCV) indicated that Overall Accuracy (OA), Kappa, Producer’s Accuracies (PA) and User’s Accuracy (UA), increased significantly when adding amplitude and phase as predictor variables to the model, than prediction using H-NDVI or raw bands only. Commission and omission errors were reduced significantly as the season progressed and more in-season imagery was available. The methodology proposed in this study can map cotton crops accurately based on the reconstruction of the unique cotton reflectance trajectory through time. This study confirms the importance of phenological metrics in improving in-season cotton fields mapping across eastern Australia. This model can be used in conjunction with other datasets to forecast yield based on the mapped crop type for improved decision making related to supply chain logistics and seasonal outlooks for production.

1 January 2018 in MethodsX
Verified
Journal Article
Open access
PFPatrick Filippi
Patrick Filippi
SCStephen R. Cattle
Stephen R. Cattle
TBT.F.A. Bishop
T.F.A. Bishop
EJEdward J. Jones
Edward J. Jones
BMBudiman Minasny
Budiman Minasny
While traditional laboratory methods of determining soil organic carbon (SOC) content are generally simple, this becomes more challenging when carbonates are present in the soil; such is commonly found in semi-arid areas. Additionally, soil inorganic carbon (SIC) content itself is difficult to determine. This study uses visible near infrared (VisNIR) spectra to predict SOC and SIC contents of samples, and the impact of including soil pH and soil total carbon (STC) data as predictor variables was evaluated. The results indicated that combining available soil pH and STC content data with VisNIR spectra dramatically improved prediction accuracy of the Cubist models. Using the full suite of predictor variables, Cubist models trained on the calibration dataset (75%) could predict the validation dataset (25%) for SOC content with a Lin's concordance correlation coefficient (LCCC) of 0.94, and an LCCC of 0.83 for SIC content. This is compared to an LCCC of 0.81 and 0.35 for SOC and SIC content, respectively, when no ancillary soil data was included with VisNIR spectra as predictor variables. These results suggest that there may be promise for using other readily available soil data in combination with VisNIR spectra to improve the predictions of different soil properties. •It can be laborious and expensive to measure soil organic and inorganic carbon content with traditional laboratory methods, and there has been recent focus on using spectroscopic techniques to overcome this.•This study demonstrates that combining ancillary soil data (pH and total carbon content) with these spectroscopic techniques can considerably improve predictions of SOC and SIC content.