Piers Holden
As a Senior Scientist in the Solutions Enablement group at Planet Labs, Piers works at the intersection of earth observation (EO) and machine learning to solve problems in the agricultural monitoring domain. His research interests include the development of EO-ML data infrastructure, and applications that leverage EO time series or the fusion of multiple EO modalities.
Sessions
The advent of modern satellite constellations–combined with significant advances in computational technology and resources–enables the systematic mapping of agricultural areas in high spatiotemporal resolution and through this provides a foundational building-block for the study and monitoring of soil health. In this context, parts of the Earth Observation and Machine Learning community have placed a growing focus on the development of capabilities to remotely identify crop types growing on agricultural cropland. These capabilities can support soil health monitoring efforts by providing systematic insights into crop rotations and cropping practices that affect soil health (e.g. cover crops). End users of these capabilities desire high accuracy across large areas with diverse agro-environmental conditions, as early in the season as possible.
Whilst recent efforts have typically shown strong performance on datasets that express limited spatial and/or temporal variability, the community is yet to explore performance across an adequate number of years at the pan-European scale to assess robustness both to the spatial variations between environmental & political zones and to the full range of variations in interannual weather patterns. To do so, there is still a need for datasets that provide sufficient spatial and temporal depth.
In Europe, the public release of historical LPIS and GSAA datasets by EU member states at either regional or national scales has made high quality crop type ground-truth data more available than ever before. The RapidCrops dataset combines these datasets across a range of countries and multiple years to provide a deep spatio-temporal stack of crop type ground-truth data; enabling assessments of generalization across both space and time. The dataset leverages harmonization standards developed under the EuroCrops initiative to standardize data from different countries. It adopts & extends the fiboa (https://fiboa.org) field boundaries data standard to make parcel boundaries and crop type labels available in an open, interoperable format. Finally, the dataset provides additional usability metrics for each parcel to help users identify and access the data most appropriate to their context, quickly.
Boermarke-Zeijen is a 350-year old farmers collective in the North of the Netherlands. The farmer collective
owns 1200 hectares of land and consists of five dairy farms, four arable farms, two mixed farms, and one chicken
farm. The farmers are developing economically and ecologically sustainable farming practices to strengthen soil
fertility and biodiversity at company and regional scale. The soils are sandy with peat patches, which is
representative of the soils of the Netherlands. The farmers producing agricultural crops face subsoil compaction
and soil crusting, leading to decreased infiltration capacity. Some of the dairy farms face phosphate deficiencies.
As part of AI4SoilHealth project, Planet Labs is teaming up with the farmer collective and the regional water
board Noorderzijlvest, to correlate in-situ measurements with Planet’s satellite data products to demonstrate its
value for monitoring and decision making. The farmers have a specific interest in monitoring nutrient fluxes, soil
moisture, land surface temperature and biomass conditions, specifically for the sustainable application of
fertilizer.
The in-situ monitoring network collects location-specific, real-time information about the weather outlook, the
level of surface water and groundwater, the moisture content of the topsoil and the water quality. Legacy data
includes in-situ samples and soil profiles covering a range of soil parameters. A field sampling campaign is
planned in April 2025 to collect the AI4SoilHealth baseline indicators (texture, SOC, pH, CEC, nutrients, density)
and additional indicators (e.g. enzymatic activity, eDNA, Macrofauna, NIR-spectroscopy).
We have used Planet’s high spatial and temporal resolution satellite data [Planet Product] to monitor and
characterize the landscape of Boermarke Zijen, by looking at Vegetation health [PlanetScope, Crop Biomass];
Water balance [Soil Water Content]; Environmental stresses [Land Surface Temperature]; Soil health [Tanager,
Sentinel, PlanetScope] and Above ground carbon [Forest Carbon, Crop Biomass]. With the recent launch of
Planet’s Hyperspectral Mission Tanager, the pilot at Boermarken Zijen will also facilitate investigations into the
feasibility of soil property estimation using spaceborne hyperspectral sensors. Using our satellite data products
in combination with the in-situ sample data, will evaluate the water regime and the potential and effects of
year-round greenery to address some of the challenges that the farmers at Boermarke Zijen are facing.