Modeling the Spatial Distribution of Acequia in Northern New Mexico using GeoAI based models

Keywords: acequias, geospatial artificial intelligence, predictive models, deep learning, new mexico
Abstract Type: Paper Abstract

Abstract

Acequias — community-based, gravity-fed irrigation ditches —are vital to food production and the provision of other ecosystems in drylands around the world. As symbols of history, culture, and querencia —sense of place, through practice —they are also critical to the survival of traditional ways of life in many New Mexico communities. However, acequias are increasingly under threat from climate change, urbanization, and other pressures. To ensure the sustainability of acequia systems, a better understanding is needed of the spatial distribution of acequias, but there is currently no timely and cost-effective for obtaining this information. To address this problem, integrate remote sensing, GIS, and geospatial artificial intelligence (GeoAI; e.g., Convolutional Neural Networks, Maxent, Random Forest) to model the distribution of active and inactive acequias in the upper Mora River watershed of northern New Mexico, an environmentally diverse area in which water issues have recently been exacerbated by the 2022 Calf Canyon / Hermits Peak Fire. We implemented the models using 250 acequia presence points and 5,000 acequia absence points and explanatory environmental variables representing topography (e.g., curvature), hydrology (e.g., flow accumulation), vegetation (normalized difference vegetation index), and land use (e.g., proximity to farmland). We assessed model utility with respect to three characteristics: ease of implementation, realism, and performance (overall accuracy, area under the receiver operating characteristic). In this paper, we present the results of this work, including prediction maps, statistics on variable importance, and model utility. In general, we found that GeoAI is a timely and cost-effective way of mapping acequias.