Fitting Soil Respiration Models
2023-05-03
- Compare on-the-ground NDVI measurements with satellite NDVI measurements.
- Compare NDVI measurements with respiration.
- Fit and test models using NYBG lawn data.
- Fit and test models using NYBG and Lamont lawn data.
Compare Satellite NDVI with Hand-held NDVI
- Planet NDVI measurements are higher than Sentinel measurements.
- Satellite NDVI does not correlate well with hand-held measurements.
- Both satellites give much higher NDVI readings for unmanaged lawns than hand-held - possibly this is due to tree canopy around these sites?
Compare NDVI measurements with Respiration
- Hand-held measurements show positive correlation between NDVI and respiration.
- Satellite measurements show no correlation - even a negative correlation thanks to the higher ndvi measurements for unmanaged lawns.
Planet NDVI vs. Respiration further broken down by type and collar:
I also looked at EVI data from Planet, but found similar results to the NDVI data:
- From Sentinel, I also derived NDMI (Normalized Difference Moisture Index), which measures the amount of moisture present in vegetation.
- The formula is (NIR - SWIR) / (NIR + SWIR)
Fit and Test Models Using NYBG Lawn Data
- I used daily mean values of UL and ML data to fit the models.
- I then generated predictions of daily mean values for each lawn type
using the satellite data and compared predicted vs. measured values:
- Soil Temperature = LST
- Soil Moisture = NDMI
- Handheld NDVI = NDVI
Model 1: Soil temperature + NDVI
- R2 = .996, p < .000
- Both variables significant.
##
## Call:
## lm(formula = log(mean_res) ~ mean_temp + mean_hhndvi, data = mean_lawns)
##
## Residuals:
## 18 19 20 21 24 25
## 0.046532 0.016384 -0.027887 0.005516 -0.017125 -0.023420
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -1.50284 0.19336 -7.772 0.004431 **
## mean_temp 0.07064 0.01184 5.965 0.009427 **
## mean_hhndvi 3.09975 0.17042 18.188 0.000363 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.03689 on 3 degrees of freedom
## (19 observations deleted due to missingness)
## Multiple R-squared: 0.9977, Adjusted R-squared: 0.9962
## F-statistic: 661.9 on 2 and 3 DF, p-value: 0.0001075
Model 2: Soil temperature + Soil moisture
- R2 = .567, p = .001
- Both variables significant.
##
## Call:
## lm(formula = log(mean_res) ~ mean_temp + mean_moist, data = mean_lawns)
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.5849 -0.2384 -0.1063 0.3018 0.5873
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -1.71564 1.05582 -1.625 0.12647
## mean_temp 0.12658 0.04702 2.692 0.01753 *
## mean_moist 0.03449 0.01140 3.025 0.00908 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.3738 on 14 degrees of freedom
## (8 observations deleted due to missingness)
## Multiple R-squared: 0.6216, Adjusted R-squared: 0.5675
## F-statistic: 11.5 on 2 and 14 DF, p-value: 0.001111
Model 3: Soil temperature + Soil moisture + NDVI
- R2 = .996, p = .002
- Moisture variable not significant.
##
## Call:
## lm(formula = log(mean_res) ~ mean_temp + mean_moist + mean_hhndvi,
## data = mean_lawns)
##
## Residuals:
## 18 19 20 21 24 25
## 0.034179 0.021840 -0.012753 -0.023873 -0.006619 -0.012774
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -1.385393 0.217695 -6.364 0.02381 *
## mean_temp 0.060874 0.014678 4.147 0.05352 .
## mean_moist -0.002988 0.002775 -1.077 0.39416
## mean_hhndvi 3.393742 0.319532 10.621 0.00875 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.03595 on 2 degrees of freedom
## (19 observations deleted due to missingness)
## Multiple R-squared: 0.9986, Adjusted R-squared: 0.9964
## F-statistic: 465.2 on 3 and 2 DF, p-value: 0.002146
Fit and Test Models Using NYBG and Lamont Lawn Data
- I tried incorporating Lamont data to get more variation in on-the-ground NDVI data for fitting the models.
- We do not have satellite data for Lamont, so can only predict respiration for NYBG lawns.
- These new parameters did not improve accuracy of predictions.
Model 1: Soil Temp/LST + NDVI
- R2 = .925, p < .00
- Both variables significant.
##
## Call:
## lm(formula = log(mean_res) ~ mean_temp + mean_hhndvi, data = df3)
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.257375 -0.035719 -0.003331 0.031631 0.202479
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.002893 0.250974 -0.012 0.9910
## mean_temp 0.019302 0.008957 2.155 0.0491 *
## mean_hhndvi 2.668114 0.191058 13.965 1.31e-09 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.1315 on 14 degrees of freedom
## (63 observations deleted due to missingness)
## Multiple R-squared: 0.9345, Adjusted R-squared: 0.9252
## F-statistic: 99.94 on 2 and 14 DF, p-value: 5.147e-09
Model 2: Soil Temp/LST + Soil Moisture/NDMI
- R2 = .532, p < .000
- Both variables significant.
##
## Call:
## lm(formula = log(mean_res) ~ mean_temp + mean_moist, data = df3)
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.60095 -0.21678 -0.01482 0.18407 0.57641
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -1.345912 0.635271 -2.119 0.0415 *
## mean_temp 0.121697 0.025555 4.762 3.49e-05 ***
## mean_moist 0.028122 0.005848 4.809 3.03e-05 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.3326 on 34 degrees of freedom
## (43 observations deleted due to missingness)
## Multiple R-squared: 0.5576, Adjusted R-squared: 0.5316
## F-statistic: 21.43 on 2 and 34 DF, p-value: 9.522e-07
Model 3: Soil Temp/LST + Soil Moisture/NDMI + NDVI
- R2=.938, p < .000
- Moisture variable not significant.
##
## Call:
## lm(formula = log(mean_res) ~ mean_temp + mean_moist + mean_hhndvi,
## data = df3)
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.229701 -0.056080 -0.003251 0.070470 0.211391
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.540433 0.440875 -1.226 0.259903
## mean_temp 0.039773 0.020598 1.931 0.094792 .
## mean_moist 0.003980 0.004948 0.804 0.447691
## mean_hhndvi 2.611731 0.349674 7.469 0.000141 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.1392 on 7 degrees of freedom
## (69 observations deleted due to missingness)
## Multiple R-squared: 0.9566, Adjusted R-squared: 0.9379
## F-statistic: 51.37 on 3 and 7 DF, p-value: 3.912e-05
Summary
- Our on-the-ground NDVI measurements have a strong positive correlation with respiration at both NYBG and Lamont.
- However, our satellite measurements at 10m and 3m do not have the same correlation with respiration.
- Using satellite data to measure NDVI at a single collar location is not reliable. The presence of tree canopy and/or roads nearby interferes with capturing the NDVI of the target area, leading to under- and over-prediction of respiration.