Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Long-Term Probabilistic Forecast of Vegetation Conditions Using Climate Attributes in the Four Corners Region

Erika McPhillips, Hyeongseong Lee, Xiangyu Xie, Kathy Baylis, Chris Funk, Mengyang Gu

TL;DR

This work tackles long-term vegetation forecasting by coupling climate-attribution with climate-forecasting in a two-phase framework over the Four Corners region. The authors introduce a generalized parallel partial Gaussian process (G-PPGP) to quantify NDVI’s nonlinear response to informative climate covariates (notably January–August precipitation and July–August VPD) and to propagate forecast uncertainty through the NDVI prediction. Phase II forecasts the climate covariates themselves using autoregressive structures, enabling one-year-ahead NDVI predictions at high spatial resolution with calibrated 95% predictive intervals. The approach yields superior accuracy compared with deep-learning baselines and provides interpretable, actionable forecasts for farmers and policymakers, with broad applicability to other regions and remote-sensing datasets. Overall, the two-phase G-PPGP framework enhances long-range vegetation forecasting by leveraging predictable climate signals and robust uncertainty quantification.$NDVI$, $VPD$, $Precipitation$, $G-PPGP$, $uncertainty$.$

Abstract

Weather conditions can drastically alter the state of crops and rangelands, and in turn, impact the incomes and food security of individuals worldwide. Satellite-based remote sensing offers an effective way to monitor vegetation and climate variables on regional and global scales. The annual peak Normalized Difference Vegetation Index (NDVI), derived from satellite observations, is closely associated with crop development, rangeland biomass, and vegetation growth. Although various machine learning methods have been developed to forecast NDVI over short time ranges, such as one-month-ahead predictions, long-term forecasting approaches, such as one-year-ahead predictions of vegetation conditions, are not yet available. To fill this gap, we develop a two-phase machine learning model to forecast the one-year-ahead peak NDVI over high-resolution grids, using the Four Corners region of the Southwestern United States as a testbed. In phase one, we identify informative climate attributes, including precipitation and maximum vapor pressure deficit, and develop the generalized parallel Gaussian process that captures the relationship between climate attributes and NDVI. In phase two, we forecast these climate attributes using historical data at least one year before the NDVI prediction month, which then serve as inputs to forecast the peak NDVI at each spatial grid. We developed open-source tools that outperform alternative methods for both gross NDVI and grid-based NDVI one-year forecasts, providing information that can help farmers and ranchers make actionable plans a year in advance.

Long-Term Probabilistic Forecast of Vegetation Conditions Using Climate Attributes in the Four Corners Region

TL;DR

This work tackles long-term vegetation forecasting by coupling climate-attribution with climate-forecasting in a two-phase framework over the Four Corners region. The authors introduce a generalized parallel partial Gaussian process (G-PPGP) to quantify NDVI’s nonlinear response to informative climate covariates (notably January–August precipitation and July–August VPD) and to propagate forecast uncertainty through the NDVI prediction. Phase II forecasts the climate covariates themselves using autoregressive structures, enabling one-year-ahead NDVI predictions at high spatial resolution with calibrated 95% predictive intervals. The approach yields superior accuracy compared with deep-learning baselines and provides interpretable, actionable forecasts for farmers and policymakers, with broad applicability to other regions and remote-sensing datasets. Overall, the two-phase G-PPGP framework enhances long-range vegetation forecasting by leveraging predictable climate signals and robust uncertainty quantification., , , , .$

Abstract

Weather conditions can drastically alter the state of crops and rangelands, and in turn, impact the incomes and food security of individuals worldwide. Satellite-based remote sensing offers an effective way to monitor vegetation and climate variables on regional and global scales. The annual peak Normalized Difference Vegetation Index (NDVI), derived from satellite observations, is closely associated with crop development, rangeland biomass, and vegetation growth. Although various machine learning methods have been developed to forecast NDVI over short time ranges, such as one-month-ahead predictions, long-term forecasting approaches, such as one-year-ahead predictions of vegetation conditions, are not yet available. To fill this gap, we develop a two-phase machine learning model to forecast the one-year-ahead peak NDVI over high-resolution grids, using the Four Corners region of the Southwestern United States as a testbed. In phase one, we identify informative climate attributes, including precipitation and maximum vapor pressure deficit, and develop the generalized parallel Gaussian process that captures the relationship between climate attributes and NDVI. In phase two, we forecast these climate attributes using historical data at least one year before the NDVI prediction month, which then serve as inputs to forecast the peak NDVI at each spatial grid. We developed open-source tools that outperform alternative methods for both gross NDVI and grid-based NDVI one-year forecasts, providing information that can help farmers and ranchers make actionable plans a year in advance.
Paper Structure (22 sections, 19 equations, 11 figures, 14 tables)

This paper contains 22 sections, 19 equations, 11 figures, 14 tables.

Table of Contents

  1. Introduction
  2. Materials and Methods
  3. Study Region and Data
  4. A Two-Stage Framework for Long-Term NDVI Forecasting
  5. Climate Variable Selection and Attribution Model
  6. Feature Selection
  7. Attribution Models (Phase I)
  8. Forecasting Climate Attributes and Capturing Dynamic Correlation (Phase II)
  9. Long-Term NDVI Forecasting (Phase II)
  10. Performance Metrics and Uncertainty Quantification
  11. Results
  12. Attribution Models
  13. One-Year-Ahead Forecasts of NDVI
  14. Discussion
  15. Conclusions
  16. ...and 7 more sections

Figures (11)

  • Figure S1: (a) Map of August NDVI values for the year 2020 in the Four Corners region. The longitude and latitudes are displayed on the edges of the plots. The gray horizontal and vertical lines within the plots are the inner borders of Utah, Colorado, Arizona, and New Mexico. (b) January-August average precipitation for the year 2020. (c) July-August average Max VPD for the year 2020. (d) The dark blue line represents the average January-August precipitation over the whole region, with the respective axis on the left y-axis. The green histogram is the gross NDVI (NDVI summed over all locations). The orange line is the average July-August max VPD over the whole region, with the respective axis on the right y-axis.
  • Figure S2: Illustration of the two-phase model to obtain one-year-ahead forecasts of August NDVI. (a) The prediction by the attribution model where August NDVI is predicted using G-PPGP assuming precipitation and VPD are known. The black dots are the true observed August NDVI at location "1" and the dark green line is the model prediction. The light green band is the 95% credible interval (CI). (b) The forecast model for January-August average precipitation one year ahead at location "1". The dark blue line is the PPGP prediction, and the light blue band is the 95% CI. (c) Elevation heatmap of the Four Corners region. "1" marks the sample location in the region. (d) Heatmap of the attribution model prediction at "1" given various precipitation and VPD inputs. The location of the circle indicates the true climate attribute values during 2020, and the color of the circle indicates the true August NDVI value observed during 2020. (e) The forecast for the July-August average VPD one year ahead at location "1". The dark orange line is the PPGP prediction, and the orange band is the 95% CI. (f) The forecast of August NDVI one-year-ahead at location "1", where the dark green line is the G-PPGP prediction and the light green band is the 95% CI.
  • Figure S3: Panels (a) and (c): The lag 1 autocorrelation values for each location between the years 2003-2020 for precipitation and VPD. Panels (b) and (d): Annual precipitation and VPD averages over high elevation regions ($\geq$ 2500 m), low elevation regions ($<$ 2500 m), and the whole region (low and high elevation regions).
  • Figure S4: (a) Elevation map with three locations marked as "1"-"3". Panels (b)-(d) show the heatmaps of attribution predictions of August NDVI by the G-PPGP model given various January-August precipitation and July-August VPD values at locations 1-3, respectively. The circle locations represent the recorded precipitation and VPD in 2020, and the inside color of the circles represents the true August NDVI value observed that year.
  • Figure S5: (a) The true gross NDVI, calculated as the sum of all August NDVI values over all grids for each prediction year, and the one-year-ahead forecasted gross NDVI using the best four methods. (b) The residuals between true gross NDVI and forecasted gross NDVI for the four methods.
  • ...and 6 more figures