Smallholder farms dominate in many parts of the world, particularly Sub-Saharan Africa.These systems are characterized by small, heterogeneous, and often indistinct field patterns, requiring a specialized methodology to map agricultural land cover. Using a variety of sites in South Africa, we present a new approach to mapping agricultural fields, based on efficient extraction of a vast set of simple, highly correlated, and interdependent features, followed by a random forest classifier. We achieved similar high performance across agricultural types, including the spectrally indistinct smallholder fields as well as the more easily distinguishable commercial fields, and demonstrated the ability to generalize performance across large geographic areas. In sensitivity analyses, we determined multi-temporal information provided greater gains in performance than the addition of multi-spectral bands available in DigitalGlobe Worldview-2 imagery.

Land cover maps increasingly underlie research into socioeconomic and environmental patterns and processes, including global change. It is known that map errors impact our understanding of these phenomena, but quantifying these impacts is difficult because many areas lack adequate reference data. We used a highly accurate, high-resolution map of South African cropland to assess (1) the magnitude of error in several current generation land cover maps, and (2) how these errors propagate in downstream studies. We first quantified pixel-wise errors in the cropland classes of four widely used land cover maps at resolutions ranging from 1 to 100 km, and then calculated errors in several representative "downstream" (map-based) analyses, including assessments of vegetative carbon stocks, evapotranspiration, crop production, and household food security. We also evaluated maps' spatial accuracy based on how precisely they could be used to locate specific landscape features. We found that cropland maps can have substantial biases and poor accuracy at all resolutions (e.g., at 1 km resolution, up to $ 45% underestimates of cropland (bias) and nearly 50% mean absolute error (MAE, describing accuracy); at 100 km, up to 15% underestimates and nearly 20% MAE). National-scale maps derived from higherresolution imagery were most accurate, followed by multi-map fusion products.Constraining mapped values to match survey statistics may be effective at minimizing bias (provided the statistics are accurate). Errors in downstream analyses could be substantially amplified or muted, depending on the values ascribed to croplandadjacent covers (e.g., with forest as adjacent cover, carbon map error was 200%-500% greater than in input cropland maps, but $ 40% less for sparse cover types).The average locational error was 6 km (600%). These findings provide deeper insight into the causes and potential consequences of land cover map error, and suggest several recommendations for land cover map users. K E Y W O R D Sagent-based model, agriculture, bias, carbon, crop yield, evapotranspiration, land cover, remote sensing

Crop model-specific biases are a key uncertainty affecting our understanding of climate change impacts to agriculture. There is increasing research focus on intermodel variation, but comparisons between mechanistic (MMs) and empirical models (EMs) are rare despite both being used widely in this field. We combined MMs and EMs to project future (2055) changes in the potential distribution (suitability) and productivity of maize and spring wheat in South Africa under 18 downscaled climate scenarios (9 models run under 2 emissions scenarios). EMs projected larger yield losses or smaller gains than MMs. The EMs' median-projected maize and wheat yield changes were -3.6% and 6.2%, respectively, compared to 6.5% and 15.2% for the MM. The EM projected a 10% reduction in the potential maize growing area, where the MM projected a 9% gain. Both models showed increases in the potential spring wheat production region (EM = 48%, MM = 20%), but these results were more equivocal because both models (particularly the EM) substantially overestimated the extent of current suitability. The substantial water-use efficiency gains simulated by the MMs under elevated CO2 accounted for much of the EM-MM difference, but EMs may have more accurately represented crop temperature sensitivities. Our results align with earlier studies showing that EMs may show larger climate change losses than MMs. Crop forecasting efforts should expand to include EM-MM comparisons to provide a fuller picture of crop-climate response uncertainties.

Highlights• DIYlandcover crowdsources the generation of landcover data, using human pattern recognition skill to create accurate maps with rich geometric detail.• It incorporates representative sampling and worker-specific accuracy assessment protocols, and connects to a large online job market. This design addresses three problems with crowdsourced mapping: representativity; data reliability; product delivery speed.• In a trial case, South African cropland was mapped with 91% accuracy by novice workers. A scaling up analysis found that an Africa-wide cropland map could potentially be developed using this software for $2-3 million within 1.2-3.8 years. AbstractAccurate landcover maps are fundamental to understanding socio-economic and environmental patterns and processes, but existing datasets contain substantial errors. Crowdsourcing map creation may substantially improve accuracy, particularly for discrete cover types, but the quality and representa- * Corresponding author

No abstract