| Land use statistics serve as important inputs to studies focused on human impacts on the agricultural, ecological, and environmental systems. However, data sources for land use are very limited, especially those that are at a fine spatial resolution---i.e. below national or subnational (state/province) level---and across a large geographic area. Using aggregate level data such as state totals or their averages tends to mask the fine-scale heterogeneity and may lead to potential bias in impact assessment. This lack of fine-scale land use data puts a constraint on achievable research topics and adds uncertainty to the quality of research conclusions; it may also impede decision-makers from implementing policies in a cost-effective manner.;We create a quasi-maximum likelihood framework to predict fine-scale land use allocation and provide data to land related studies. The framework combines aggregate level land use statistics and land attributes information at a fine-scale resolution of 5 by 5 arc-minute (which we call the pixel level), and estimates land use at the pixel level. For empirical demonstrations, we set up two models, one focuses on maize as the crop of interest, the other focused on maize, soybeans, and wheat simultaneously. Both models analyze a circa 2000 cropland allocation for the Americas (North, South, and Central).;We employ in-sample and out-of-sample validation exercises to verify our prediction results, and compare our estimation results with other available sources of predicted data. In addition, we establish two Monte Carlo simulation designs to test the finite sample performance of our framework, including one case incorporating spatial clustering. Results show that our framework is capable of producing reliable land allocation estimates.;Based on our estimation framework and empirical analyses, we create the Fine-scale Land Allocation Tool (FLAT) user interface to publicize our model and facilitate visualization of the estimation results. The tool is free to all, and the user can download the datasets and models, modify the framework, and plot their own estimates. We hope FLAT can serve as a pilot program and support the idea of open source data sharing platforms to promote research development. |
