Correction: Sesnie et al. In-Situ and Remote Sensing Platforms for Mapping Fine-Fuels and Fuel-Types in Sonoran Semi-Desert Grasslands. Remote Sens. 2018, 10, 1358

• Text Correction

3. Results

In the original article [1], there was a coding error that resulted in an error in the values reported in the first and second paragraph of Results. In Section 3.1. In-Situ Biomass Results and Section 3.2. Remotely Sensed Fine-Fuel Biomass Results, reporting the average grams per meter squared on sample quadrats and overall plot averages in kilograms per hectare should have been as shown below. The corrected values are increased by a factor of two to four. Specifically, in the first sentence of the first paragraph of Results Section 3.1. in the original article, we reported the values “0.0 g/m² to 185.4 g/m² and averaged 25.8 g/m² ± 25.5”. The correct values should be “0.0 g/m² to 741.6 g/m² and averaged 103.1 g/m² ± 102.0”. In the same paragraph of the original article, we reported the value “mean squared error (RMSE) of 0.97 g/m²”. The correct value should be stated as “mean squared error (RMSE) of 3.88 g/m²”. In the third sentence of the second paragraph of Section 3.1. of the original article, we reported fine-fuel values “51.0 kg/ha to 1671.8 kg/ha”. The correct values should be shown as “104.3 kg/ha to 3339.6 kg/ha”. In the same sentence of the original article, we reported fine-fuel biomass values “averaged 685.5 ± 327.0 kg/ha”. The corrected values should be stated as “averaged 1363.0 ± 649.2 kg/ha”. In Results Section 3.2. of the original article, we reported an “(RMSE = 201.9 kg/ha)” in the third sentence of the second paragraph. The correct value should be “(RMSE = 402.8 kg/ha)”. In Results Section 3.2. of the original article we reported a value “(RMSE = 220 kg/ha, Table 4)” in the fourth sentence of the second paragraph. The corrected value should be shown as “(RMSE = 440 kg/ha, Table 4)”. We greatly apologize for any inconvenience caused. These changes do not affect the overall interpretation of results or conclusions of the article. The original article has been updated.

3.1. In-Situ Biomass Results

Herbaceous biomass clipped from n = 431 quadrats on 20 plots ranged from 0.0 g/m² to 741.6 g/m² and averaged 103.1 g/m² ± 102.0. Using 10-fold cross validation as our RFE training control methods for feature selection, six of the 10 variables used for predicting fine-fuel biomass were selected as optimal using minimum RMSE values (Figure 3). We used the selected optimal predictors to further tune the number of variables tried at each node for RF models. Two predictors tried were found to more substantially decrease OOB error and were used in the final RF model with six selected predictors. The final model resulted in 84% of the variation explained and a root mean squared error (RMSE) of 3.88 g/m².

Predicted biomass for non-destructively sampled plots ranged from 104.3 kg/ha to 3339.6 kg/ha and averaged 1363.0 ± 649.2 kg/ha, which was within the range of values previously reported by Marsett et al. [30] for Sonoran and Chihuahuan rangelands with a burning and grazing history.

3.2. Remotely Sensed Fine-Fuel Biomass Results

Therefore, three separate fine-fuel biomass models were developed for each sensor type, first using spectral predictors, second with spectral and terrain predictors, and lastly a combined model that included vegetation cover variables. The combined model performed substantially better than either of the two previous sets of predictors for both sensor types (Table 4). The optimal number of predictor variables selected with RFE decreased substantially from 37 to 19 for the WV3 model while the amount of variance explained increased from 51.1% to 65.0% (RMSE = 403.8 kg/ha). Similarly, the number of OLI model predictors was decreased from 43 to 10 while increasing the variance explained from 47.1% to 57.6% (RMSE = 440 kg/ha, Table 4). Further discussion of results considers only comparisons and model outcomes from the two sensors using all variables combined.

4. Discussion

The same calculation mistake also affected the average fine fuel estimated per quadrat in the Discussion Section 4. These errors were the result of the same coding error identified for the Results Section 3 and corrected values are increased by a factor of two. In the original Discussion Section 4 and last sentence of paragraph six, we reported that “our plots averaged 68.6 g/m²”. The corrected value should be stated as “our plots averaged 136.3 g/m²”. We greatly apologize for any inconvenience caused. These changes do not affect the overall conclusions of the article. The original article has been updated.

Field measurements showed an average of <1 g/m² and <10% cover on plots suggesting that trace amounts of herbaceous plant biomass are insufficiently estimated at the spatial and spectral resolution of WV2 or, by extension, WV3 imagery [84]. In contrast, our plots averaged 136.3 g/m² biomass and 67% cover by herbaceous plants.

• Error in Figure/Table

In the original article, the same coding error reported for Results and Discussion fine-fuel biomass also affected Table 4, Figure 6, Figure 8a,b, Figure 15b and Figure 16. Specifically, in Figure 6 of the original article, values on the y-axis ranged from a square root herbaceous biomass of 10 kg/ha to 40 kg/ha. The corrected values for the y-axis in Figure 6 should be shown as the square root herbaceous biomass of 10 kg/ha to 60 kg/ha. In Table 4 of the original article, we reported the column five RMSE values in the first to sixth rows as 247.7, 231.4, 210.6, 257.5, 251.8, and 231.2, respectively. The correct RMSE column five values in the first to sixth rows are 495.4, 462.8, 421.2, 515.0, 503.6, and 462.4, respectively. In addition, we reported the RMSE column eight values of Table 4 in the original article for the first to sixth row as 236.6, 219.4, 201.9, 252.0, 246.6, and 220.0, respectively. The correct RMSE column eight values in the first to sixth rows should be 473.2, 438.4, 403.8, 504.0, 493.2, and 440.0, respectively. In Figure 8a,b of the original article x and y axes ranged from 0 kg/ha to 1500 kg/ha. The corrected x and y axis values in Figure 8a,b should range from 0 kg/ha to 3500 kg/ha. In the map Figure 15b legend of the original article, the values for fine-fuel biomass ranged from 129.466 kg/ha to 1489.4 kg/ha. The corrected legend for map Figure 15b shows a range of values between 241 kg/ha and 2996 kg/ha. In Figure 16 of the original article, the average fuel values reported on the y-axis ranged from 400 kg/ha to 1000 kg/ha. The corrected values should range between 800 kg/ha and 2000 kg/ha. We apologize for any inconvenience caused by the error. These changes do not impact interpretation of results or conclusions drawn from the article. All model relationships and goodness of fit statistics (R²) shown remain the same. Mapped fuel values are two times greater but show the same pattern of low to high fine-fuel accumulation across the landscape. The original article has been updated.

The corrected values are now added to the Figure axes (fine-fuel biomass) and Table 4 (RMSE) that appear below.

Figure 6. Scatter plot of the square root transformed fine-fuel biomass from plots and ratio of herbaceous to bare ground cover from plot data with a curvilinear fit (solid line) and 95th percentile confidence intervals (dashed lines).

Figure 6. Scatter plot of the square root transformed fine-fuel biomass from plots and ratio of herbaceous to bare ground cover from plot data with a curvilinear fit (solid line) and 95th percentile confidence intervals (dashed lines).

Table 4. Remote sensing based fine-fuel biomass (kg/ha) model results from Random Forest regression tree models with and without RFE predictor variable selection.

Sensor	Predictors 1	No. Variables	Var. Explained 2	RMSE	Features Selected	Var. Explained 3	RMSE
WV3	Spectral	85	47.1	495.4	37	51.1	473.2
	Spectral and terrain	90	51.1	462.8	37	56.6	438.8
	Spectral, terrain and cover	94	61.9	421.2	14	65.0	403.8
OLI	Spectral	59	43.0	515.0	43	47.1	504.0
	Spectral and terrain	67	45.5	503.6	64	48.9	493.2
	Spectral, terrain and cover	70	54.0	462.4	10	57.6	440.0

¹ Three separate WV3 and OLI biomass models were developed using different sets of predictors. ² Percentage of variance explained from Random Forest models without feature selection. ³ Amount of variance explained from optimized Random Forest models with feature selection.

Table 4. Remote sensing based fine-fuel biomass (kg/ha) model results from Random Forest regression tree models with and without RFE predictor variable selection.

Sensor	Predictors 1	No. Variables	Var. Explained 2	RMSE	Features Selected	Var. Explained 3	RMSE
WV3	Spectral	85	47.1	495.4	37	51.1	473.2
	Spectral and terrain	90	51.1	462.8	37	56.6	438.8
	Spectral, terrain and cover	94	61.9	421.2	14	65.0	403.8
OLI	Spectral	59	43.0	515.0	43	47.1	504.0
	Spectral and terrain	67	45.5	503.6	64	48.9	493.2
	Spectral, terrain and cover	70	54.0	462.4	10	57.6	440.0

¹ Three separate WV3 and OLI biomass models were developed using different sets of predictors. ² Percentage of variance explained from Random Forest models without feature selection. ³ Amount of variance explained from optimized Random Forest models with feature selection.

Figure 8. Scatter plots comparing (A) predicted and observed plot biomass (kg/ha) from field plots and WV3 (r² = 0.70, solid line) and OLI (r² = 0.74, dashed line) models and (B) predicted biomass (fine-fuel) for plot locations by each sensor type and plot (r² = 0.73, solid line). The dashed line in (B) is forced though the origin.

Figure 8. Scatter plots comparing (A) predicted and observed plot biomass (kg/ha) from field plots and WV3 (r² = 0.70, solid line) and OLI (r² = 0.74, dashed line) models and (B) predicted biomass (fine-fuel) for plot locations by each sensor type and plot (r² = 0.73, solid line). The dashed line in (B) is forced though the origin.

Figure 15. Modeled (A) land cover types from WV3 imagery and (B) corresponding fine-fuel biomass within the study areas in 2015. Grasslands highly invaded by the non-native grass species E. lehmaninana overlapped with areas of high fine-fuel biomass accumulation. Grasslands with a greater number of native grass species were associated with lower fine-fuel accumulations.

Figure 15. Modeled (A) land cover types from WV3 imagery and (B) corresponding fine-fuel biomass within the study areas in 2015. Grasslands highly invaded by the non-native grass species E. lehmaninana overlapped with areas of high fine-fuel biomass accumulation. Grasslands with a greater number of native grass species were associated with lower fine-fuel accumulations.

Figure 16. Fine-fuels and fuel-type data compared within principal fire management units (n = 59) on the Buenos Aires National Wildlife Refuge (BANWR) showed a strong positive relationship between non-native grass cover and average fine-fuel biomass.

Figure 16. Fine-fuels and fuel-type data compared within principal fire management units (n = 59) on the Buenos Aires National Wildlife Refuge (BANWR) showed a strong positive relationship between non-native grass cover and average fine-fuel biomass.