Resilience and pastoralism: satellite-based decision support system for pastures

畜牧业占肯尼亚土地面积的37.5%，占国内生产总值的12%，占农业部门的40%，但由于过度放牧，畜牧业容易受到频繁干旱和退化的影响。在这种情况下，这项由苏塞克斯大学领导的工作将评估新的地球观测（EO）数据集的潜力（例如，哨兵1和2）提供近实时监测和预测畜牧业有用和可获得的生物量。将采用与当地利益攸关方共同参与的方法绘制牧场地区的可达性地图，以确定与土地保有权、保护要求、迁移模式、水和其他社会文化因素有关的问题。监测有用的生物量将依赖于绘制主要植物功能类型（PFT）地图，然后跟踪这些PFT的生物量动态、植物健康和物候周期。PFT在牧场包括一年生和多年生草和落叶和万年青灌木。将利用Sentinel光谱中的光谱和时间差异，使用光谱混合分析或机器学习技术来实现这种分类。将从以下方面监测植被动态：i）将Sentinel-1雷达后向散射和干涉测量高度与地面生物量观测相关联的生物量动态; ii）使用植被和水指数（例如，通过对各种指标的Logistic函数拟合，建立了牧草的预测模型，并根据EO和气候数据的实时更新，定期对模型进行更新。物候周期、生物量和植物健康的变化将与气象变量和其他地理空间数据集（如土壤质地、牲畜）有关。预测将依赖于时间序列分析，如自回归移动平均法、高斯过程与季节气候预测相结合

Livestock accounts for 37.5% of Kenya's land area, 12% of its GDP and 40% of itsagricultural sector, but is susceptible to frequent droughts and degradation due toovergrazing. In this context, this Sussex University-led work will assess the potential ofnew earth observation (EO) datasets (e.g., Sentinel 1 and 2) to deliver near real-timemonitoring and prediction of useful and accessible biomass for pastoralism.The accessibility of pasture areas will be mapped using a participatory approach withlocal stakeholders to identify issues related to land tenure, conservation requirements,migration patterns, water, and other socio-cultural factors.Monitoring of useful biomass will rely on mapping major plant functional types (PFTs), andthen tracking biomass dynamics, plant health, and phenological cycles of these PFTs. PFTsin pastures include annual and perennial grasses and deciduous and evergreen shrubs. Thisclassification will be achieved using spectral mixture analysis or machine learningtechniques, taking advantage of spectral and temporal differences in Sentinel's spectra.Pasture dynamics will be monitored from i) biomass dynamics relating Sentinel-1 radarbackscatter and interferometric height with ground biomass observations; ii) vegetationhealth and stress using vegetation and water indices (e.g., NDVI, NDWI), as well asmonitoring plant photosynthetic activity using the 3 red-edge Sentinel-2 bands; iii)phenology through fitting logistic functions on various indices.Finally, predictive models of pastures will be developed which will be updated regularlywith the real-time update of EO and climate data. Changes to phenological cycles, biomassand plant health will be related to meteorological variables and other geospatial datasets(e.g. soil texture, livestock). Forecasting will rely on time-series analyses such asautoregressive moving average methods, Gaussian Processes in combination with seasonalclimate forecast