-
We’re moving house so this might be the last post for a while. See you on the other side. In the meantime, enjoy this bunch of papers on different applications of spatial analysis to agricultural biodiversity and associated topics.
- Addressing global hotspots of drought-related crop production losses. A crop-specific drought sensitivity metric for 17 major crops finds rainfed production losses of 10% globally under historically observed extremes, enough to feed 2 billion people, with hotspots in the US Midwest, eastern Brazil, the Mediterranean and South Asia. Sustainable irrigation expansion and crop switching could avoid 60% of those losses. No word on varietal change, but I’m sure it would help.
- GEM-Forest: A Global satellite EMbedding–based map of forests and tree crops for 2020. Using Google DeepMind’s Alpha Earth Foundation embeddings, this 10 m global dataset classifies forest, non-forest and tree crop areas with 90% accuracy, showing that lightweight classifiers on satellite embeddings 1 can rival more complex forest-monitoring pipelines built to support things like the EU Deforestation Regulation.
- Global distribution of cattle, horses, goats, sheep and buffaloes at 1 km resolution for 2000–2022 based on subnational census data and spatiotemporal machine learning. Harmonizing 55,336 census polygons across 147 countries produces annual 1 km livestock headcount and density layers for five species, letting users track shifting grazing pressure over more than two decades rather than relying on single-year snapshots. 2
- A Spatial-Econometric Analysis of Fruit Tree Diversity in Lebanon: A contextual framework for supporting smallholder farmers. How geography, socio-economic conditions and local context shape fruit tree diversity in Lebanon, helping identify where interventions could best support smallholder resilience.
- High-resolution mapping of rice cropping pattern, intensity, calendar, and ecosystem type across Southeast Asia. Joins the growing pile of fine-resolution rice products for the region, this one distinguishing not just how many harvests a field gets but which rice ecosystem (irrigated, rainfed, upland) it belongs to. Is which variety is being grown next on the agenda?
- Entrenchment of cropping patterns reinforces climate exposure for certain crops in US. County-level analysis of US cropping decisions finds that climate risk has actually risen for some crops (like rainfed soybeans) even as it fell for others, suggesting economic and policy incentives are keeping farmers planted in increasingly climate-mismatched patterns rather than nudging them toward the drought-sensitivity-driven switching the first paper above models.
- What’s an embedding? A single spot on Earth might have dozens of pixels across multiple spectral bands (visible light, infrared, radar), observed repeatedly over time. That’s a lot of raw data, and most machine learning models don’t work well with raw pixels directly: too much redundancy and noise. An embedding model (like Google DeepMind’s Alpha Earth Foundation) is trained to compress all of that into a fixed-length vector that still preserves the underlying signal. Things like “this looks like dense broadleaf canopy” or “this has the seasonal greening pattern of a palm plantation” get encoded in that vector space, even though the model never explicitly labels them that way. You can then just run a relatively simple classifier on top of the embeddings, because the hard work of “understanding” the imagery was already done upstream. Neat.
- No word on embeddings, but the Land & Carbon Lab writeup of the dataset walks through why it took so long to build: livestock census data is collected at wildly different administrative scales and intervals, and the workflow that turns that mess into consistent annual maps is as much the story as the maps themselves.